Search

• Features are selected from the perspectives of machine learning and field application. • Stepwise FS path combining optimization with machine learning algorithms is proposed. • Best performance using existing features and corresponding feature set are revealed. • Recommendations for feature supplementation to further improve performance are given. • Feature sets are verified to be general and effective by experiments and comparisons. Fault diagnosis (FD) is vital for enhancing chiller efficiency and reliability. Feature selection (FS) is the prerequisite and key to diagnose faults. This paper addresses two intriguing questions from machine learning (ML) and field perspectives. Question-1: Based on commonly installed sensors, what is the best performance that the FD models based on ML algorithms can achieve, and what features are relevant? Question-2: Which features can enhance diagnostic performance? and to what extent? This paper designs a stepwise FS process. First, a field investigation is conducted to gather information on sensors installed in actual chillers. Based on actual field installation, feature calculation cost, and thermodynamic mechanism, three levels of initial feature libraries are created, each containing an increasing number and type of features. An FS method combining an optimization algorithm with an FD model based on ML algorithm is proposed. In the end, the insight into the best diagnostic performance achieved by ML-based models using existing sensors and the corresponding optimal feature subsets is provided, and recommendations for feature supplementation to further improve diagnostic performance are also provided. Compared with other literature-reported feature subsets, the recommended feature subsets show better generality and effectiveness on seven commonly used ML-based models. [ABSTRACT FROM AUTHOR]

The renewal of green home appliances is a crucial measure for households to save energy and reduce emissions. However, how online reviews, especially those relate to energy-saving, affect green home appliance purchase behavior (GHAPB) lacks exploration. In this paper, we investigate over 1 million online reviews on about 3,116 types of air conditioner from JD. By applying word2vec, we divide energy-saving related information in the following three types: norm information, environmental health information and price information, and construct dictionaries for each. Then, the effect value of energy-saving information is quantified from perspectives of breadth, depth and intensity through sentiment analysis. The influence of energy-saving information in online reviews on GHAPB is finally analyzed by linear regression and machine learning models. The results show that all energy-saving information has positive impact on GHAPB, and environmental health information is the most important one. In addition, the attributes of online reviews impose a greater influence on GHAPB, comparing with those of products. The in-depth exploration of energy-saving information in online reviews provides targeted recommendations for the manufacturer and the retailer to promote the adoption of green home appliances. [ABSTRACT FROM AUTHOR]

[Display omitted] • Discusses the concept of building performance simulation versus new digital trends. • Reviews Artificial Intelligence, Machine Learning, Digital Twins, Cyber-Physical Systems, Internet of Things, Data Mining. • Explores common notions about differences, similarities, strengths and limitations of each concept. • Reviews emerging work on the interface between these concepts. • New concepts enjoy traction and energize the building performance research area. In an increasingly digital world, there are fast-paced developments in fields such as Artificial Intelligence, Machine Learning, Data Mining, Digital Twins, Cyber-Physical Systems and the Internet of Things. This paper reviews and discusses how these new emerging areas relate to the traditional domain of building performance simulation. It explores the boundaries between building simulation and these other fields in order to identify conceptual differences and similarities, strengths and limitations of each of these areas. The paper critiques common notions about these new domains and how they relate to building simulation, reviewing how the field of building performance may evolve and benefit from the new developments. [ABSTRACT FROM AUTHOR]

To reduce global carbon emissions and limit climate change, controlling energy consumption in buildings is an important piece of the puzzle. Here, we specifically focus on using a demand response (DR) algorithm to limit the energy consumption of a residential building's heating system while respecting user's thermal comfort. In that domain, Reinforcement learning (RL) methods have been shown to be quite effective. One such RL method is Monte Carlo Tree Search (MCTS), which has achieved impressive success in playing board games (go, chess). A particular advantage of MCTS is that its decision tree structure naturally allows to integrate exogenous constraints (e.g., by trimming branches that violate them), while conventional RL solutions need more elaborate techniques (e.g., indirectly by adding penalties in the cost/reward function, or through a backup controller that corrects constraint-violating actions). The main aim of this paper is to study the adoption of MCTS for building control, since this (to the best of our knowledge) has remained largely unexplored. A specific property of MCTS is that it needs a simulator component that can predict subsequent system states, based on actions taken. A straightforward data-driven solution is to use black-box neural networks (NNs). We will however extend a Physics-informed Neural Network (PiNN) model to deliver multi-timestep predictions, and show the benefit it offers in terms of lower prediction errors (−32% MAE) as well as better MCTS performance (−4% energy cost, +7% thermal comfort) compared to a black-box NN. A second contribution will be to extend a vanilla MCTS version to adopt the ideas applied in AlphaZero (i.e., using learned prior and value functions and an action selection heuristic) to obtain lower computational costs while maintaining control performance. • Evaluated a PiNN for multi-timestep predictions. • Implemented and assessed advanced MCTS for residential heating DR. • The PiNN enables MCTS to obtain higher rewards compared to a Black-Box model. • Incorporating prior knowledge via NN enhances tree search efficiency. [ABSTRACT FROM AUTHOR]

Sensor-based data-driven building energy consumption prediction could play an important role in designing and operating energy-efficient buildings. However, existing approaches rely on rigid sensor networks to collect indoor physical parameter data, and the impact of sensor locations on the prediction performance is often poorly understood. To address this gap, this paper aims to study the impact of sensor deployment configurations on building energy consumption prediction, where a configuration is defined in terms of the number and locations of sensors and their flexibility (i.e., fixed or can change over time). Forty-eight (48) configurations were defined and tested. In identifying the sensor locations of the configurations, three clustering approaches were tested. Indoor physical parameter data were collected from an office building using sensors placed at different locations to test these configurations. For each configuration, four feature combinations were also defined to test how well the collected data from the sensor configurations could predict the hourly building energy consumption of three end uses (total, HVAC, and both lighting and receptacles) – alone and in combination with other data (e.g., outdoor weather data). The results showed that the sensor configurations could impact the prediction performance [measured by coefficient of variation (CV)] by 35–76%. The findings from this work could help define sensor deployment configurations for enhanced building energy consumption prediction. [ABSTRACT FROM AUTHOR]

Energy prediction in buildings plays a crucial role in effective energy management. Precise predictions are essential for achieving optimal energy consumption and distribution within the grid. This paper introduces a Long Short-Term Memory (LSTM) model designed to forecast building energy consumption using historical energy data, occupancy patterns, and weather conditions. The LSTM model provides accurate short, medium, and long-term energy predictions for residential and commercial buildings compared to existing prediction models. We compare our LSTM model with established prediction methods, including linear regression, decision trees, and random forest. Encouragingly, the proposed LSTM model emerges as the superior performer across all metrics. It demonstrates exceptional prediction accuracy, boasting the highest R 2 score of 0.97 and the most favorable mean absolute error (MAE) of 0.007. An additional advantage of our developed model is its capacity to achieve efficient energy consumption forecasts even when trained on a limited dataset. We address concerns about overfitting (variance) and underfitting (bias) through rigorous training and evaluation on real-world data. In summary, our research contributes to energy prediction by offering a robust LSTM model that outperforms alternative methods and operates with remarkable efficiency, generalizability, and reliability. • The proposed LSTM model forecasts accurate and reliable short, medium, and long-term energy consumption predictions. • The proposed LSTM model is capable of energy consumption predictions in both residential and commercial buildings. • The proposed LSTM requires a minimal amount of training data to forecast accurately and uses a variety of features including weather and occupancy data for precise energy consumption prediction. • The model achieves an average R2-score of 0.91 and average MAE of 0.013. [ABSTRACT FROM AUTHOR]

• A data-informed building energy management (DiBEM) framework was proposed. • It was demonstrated in an office/laboratory building with two-years' experimental data. • The impacts of energy efficient interventions were quantified through data-driven models. • The energy use intensity was reduced from 54.1 kWh/m2 to 42.8 kWh/m2. To support building operations in reaching ultra-low energy targets, this paper proposes a data-informed building energy management (DiBEM) framework to improve energy efficiency systematically and continuously at the operation stage. Specifically, it has two key features including data-informed energy-saving potential identification and data-driven model-based energy savings evaluation. The paper demonstrates the proposed DiBEM with a detailed case study of an office and living laboratory building located in Cambridge, Massachusetts called HouseZero. It focuses on revealing the performance of the energy-efficient interventions from two-years' building performance monitoring data, as well as evaluating energy savings from the interventions based on the data-driven approach. With Year 1 as baseline, several interventions are proposed for Year 2 including improvements to controls and operation settings, encouragement of occupants' behavior for energy savings, and hardware retrofitting. These were deployed to heating/cooling, domestic hot water, lighting, plug and other loads, and photovoltaic (PV) systems. To quantify the impacts of different interventions on energy end uses, several data-driven models are developed. These models utilize linear regression, condition model, and machine learning techniques. Consequently, the heating/cooling energy consumption that was already ultra-low in Year 1 (12.8 kWh/m2) is further reduced to 9.7 kWh/m2 in Year 2, while the indoor thermal environment is well maintained. The domestic hot water energy is reduced from 2.3 kWh/m2 to 1.2 kWh/m2. The lighting energy is only increased from 0.3 kWh/m2 in pandemic operations without occupancy in Year 1 to 0.8 kWh/m2 in partial normal operations in Year 2, while the indoor illuminance level meets occupants' requirements. Combined with other relatively constant loads and the reduction of plug and other loads due to COVID building operation restrictions, the total energy use intensity is thereby reduced from 54.1 kWh/m2 to 42.8 kWh/m2, where 5.4 kWh/m2 of energy reduction for Year 2 is estimated to be contributed by the energy-efficient interventions. PV generation is 36.1 kWh/m2, with an increase of 1.4 kWh/m2 from a new inverter. In summary, this paper demonstrates the use of DiBEM through a detailed case study and long-term monitoring data as evidence to achieve ultra-low energy operations. [ABSTRACT FROM AUTHOR]

• The paper explores the construction method for compact heterogeneous ensemble models. • The paper adopts the exhaustive search method for optimal model identification. • The paper proves the necessity for optimal model identification. • The paper analyzes the benefits and limitations of the exhaustive search method. Heterogeneous ensemble learning has received increasing attention in building energy prediction research because it provides more stable and accurate predictions than the conventional single model-based methods. Although existing studies have proved the feasibility of adopting heterogeneous ensemble learning in building energy prediction, they have not fully explored the method of constructing an optimal heterogeneous ensemble learning model based on limited base model resources. This paper attempts to identify the optimal heterogeneous ensemble learning model for building energy prediction by using the exhaustive search method. Six machine learning methods are used as the alternative base model for constructing heterogeneous ensemble learning models. All 57 heterogeneous ensemble learning models that can be built from the six base models are compared and the optimal model is identified. The hourly building energy prediction experiment is performed on an institutional building located at the University of Florida to verify the proposed exhaustive search method. Three modules representing different semesters of the test building are developed and tested to examine the predictive performance of the proposed heterogeneous ensemble learning models. The research findings demonstrate the feasibility of the proposed exhaustive search method in identifying the optimal heterogeneous ensemble learning model. The optimal heterogeneous ensemble learning model found in this study can reduce the MAPE by 4.9% − 6.6% compared with its most accurate base model in the model testing stage. The experimental results and research findings provide critical guidance for applying heterogeneous ensemble learning to data-driven building energy prediction. [ABSTRACT FROM AUTHOR]

[Display omitted] • Systematic review of hybrid system controls for mixed-mode buildings. • Meta-analysis of controls, case studies, platforms and building performance. • Optimal control most often use model predictive control or reinforcement learning. • Studies across 15 climate zones showed a medium energy saving of 28%. • 80% of case studies were simulation-based signaling need for actual implementation. Mixed-mode buildings utilise a combination of natural ventilation from operable building envelopes and mechanical systems to realise climate-friendly ventilation and cooling. The control of such hybrid systems is a promising solution to improve energy efficiency for buildings and comfort for occupants. Meanwhile, it is also a challenging task due to the complexity of the systems and coupled indoor and outdoor air flows and disturbances. This paper presents a systematic review of mixed-mode building controls. From the topical aspects of case studies, the paper analyses locations, climates, building and space types, and sensors. From a control point of view, the paper presents control algorithms based on a well-structured taxonomy, where various algorithms are classified into four categories (On-Off and PID control, rule-based control, optimal control including model predictive control and reinforcement learning, and computational intelligence including fuzzy logic and data-driven control). From an evaluation point of view, the paper illustrates evaluation platforms (simulation and real-world implementation) and results. Modelling and control tools for (co-) simulation are summarised. The evaluation results are compared based on metrics of natural ventilation and night cooling potentials, thermal comfort, indoor air quality, and energy savings. Finally, findings from the review and future research concerns are discussed. [ABSTRACT FROM AUTHOR]

[Display omitted] • The cost-of-living crisis affected energy use in winter 2022/23 in British households. • Domestic electricity consumption dropped 8.4%, gas consumption dropped 10.8%. • Reducing heating was most effective; underheating and fuel poverty are key concerns. • Energy reductions saved consumers around £29/month on their bills. • Winter 2022/23 energy bills were still £34/month higher on average than before. In October 2022 British households entered a heating season amidst exceptionally high energy prices – squeezing household incomes and increasing fuel poverty. This study analyses electricity and gas consumption in 5594 households from October 2022 to March 2023 using XGBoost counterfactual models trained on historic data. With survey data collected in early 2023 we investigate how consumption reduction correlated with energy-saving actions, household and dwelling characteristics, and indicators of underheating and fuel poverty. Our analysis showed that electricity consumption was 8.4% lower and gas consumption 10.8% lower than the previous winter (accounting for weather), saving consumers around £29/month. Despite this and a government subsidy, energy bills were still around £34/month higher than the previous winter (£158/month (median); £500/month (95th percentile)); price elasticity was −0.10 for electricity and −0.07 for gas consumption. Greatest consumption reduction correlated with largest reported changes to heating practices, in particular heating for fewer hours and turning thermostats down lower. We find evidence of greater fuel poverty and underheating among the greatest energy reducers. This paper presents novel methods for analysing energy saving using smart meter data for changes without a control group, plus novel findings related to short-term price elasticity and the energy-saving impacts of behaviour change. [ABSTRACT FROM AUTHOR]

This paper proposes a framework applicable to preliminary building design that cohesively integrates material cost estimation, embodied energy assessment, and the prediction of the thermal demand of a building. At its core is a climate-responsive machine learning model specifically developed using readily accessible early-stage design data. This model forecasts the thermal demand both in the present and in the face of likely climatic shifts. To strike a balance in energy performance goals, material cost estimation is seamlessly incorporated. Building on the gradient boosting machine, the proposed approach showed better performance than other algorithms, predicting thermal demand 25 to 37.5 times faster than conventional energy software without compromising accuracy. The key findings of the presented case studies highlight the significant influence of the Window-to-Wall ratio and floor thermal mass on thermal demand—especially in mitigating cooling demands posed by potential climate change scenarios. By enhancing interior floor types, heating demand can be reduced by nearly 49%. The study further suggests that incorporating climate change considerations into life cycle energy designs is projected to amplify thermal and cooling loads by 32% and 94%, respectively, highlighting the crucial role of early design decisions in shaping sustainable and efficient buildings. • A framework estimates cost, embodied energy, and predict thermal load in pre-design stage. • This framework integrates climate changes for energy forecasting over the life cycle. • Floor thermal mass emerges as a key predictor in temperate climates. • Varied floor materials and types and their impact on thermal demand are detailed. • Slab on the ground and WWR are found to mitigate cooling demand amid climate changes. [ABSTRACT FROM AUTHOR]

Stakeholders such as urban planners and energy policymakers use building energy performance modeling and analysis to develop strategic sustainable energy plans with the aim of reducing energy consumption and emissions from the built environment. However, inconsistent energy data and the lack of scalable building models create a gap between building energy modeling and traditional planning practices. An alternative approach is to conduct a large-scale energy usage survey, which is time-consuming. Similarly, existing studies rely on traditional machine learning or statistical approaches for calculating large-scale energy performance. This paper proposes a solution that employs a data-driven machine learning approach to predict the energy performance of urban residential buildings, using both ensemble-based machine learning and end-use demand segregation methods. The proposed methodology consists of five steps: data collection, archetype development, physics-based parametric modeling, machine learning modeling, and urban building energy performance analysis. The devised methodology is tested on the Irish residential building stock and generates a synthetic building dataset of one million buildings through the parametric modeling of 19 identified vital variables for four residential building archetypes. As a part of the machine learning modeling process, the study implemented an end-use demand segregation method, including heating, lighting, equipment, photovoltaic, and hot water, to predict the energy performance of buildings at an urban scale. Furthermore, the model's performance is enhanced by employing an ensemble-based machine learning approach, achieving 91% accuracy compared to the traditional approach's 76%. Accurate prediction of building energy performance enables stakeholders, including energy policymakers and urban planners, to make informed decisions when planning large-scale retrofit measures. • Utilization of ensemble-based machine learning modeling. • Implementation of an end-use demand segregation. • Identification of key building features for each end-use demand prediction. • Analysis of the impact of retrofit measures using future weather data. [ABSTRACT FROM AUTHOR]

[Display omitted] • A blockchain and ML-based energy management framework for multiple buildings. • Multi-types of energy (i.e., heat, cold and electricity) allocation and transmission. • Machine learning to predict day-ahead patterns of energy generation and consumption. • P2P energy trading among prosumers and consumers within the multi-energy network. • Optimal and automated energy allocation among multiple energy users through PSO. In most domestic buildings, gas and electricity are supplied by energy and utility companies through centralised energy systems. This often results in a high burden on central management systems and has adverse effects on energy prices. Blockchain-based peer-to-peer energy trading platforms can deliver strategic operation of decentralised multi-energy network among multiple domestic buildings to reduce global greenhouse gas emissions and address global climate change issues. However, prevailing blockchain-based energy trading platforms focused on system implementation for peer-to-peer electricity trading while lacking predictive control and energy scheduling optimisation. Therefore, this paper presents an integrated blockchain and machine learning-based energy management framework for multiple forms of energy (i.e., heat and electricity) allocation and transmission, among multiple domestic buildings. Machine learning is harnessed to predict day-ahead energy generation and consumption patterns of prosumers and consumers within the multi-energy network. The proposed blockchain and machine learning-based decentralised energy management framework will establish optimal and automated energy allocation among multiple energy users through peer-to-peer energy transactions. This approach focuses on energy-matching from both the supply and demand sides while encouraging direct energy trading between prosumers and consumers. The security and fairness of energy trading can also be enhanced by using smart contracts to strictly execute the energy trading and bill payment rules. A case study of 4 real-life domestic buildings is introduced to determine the economic and technical potential of the proposed framework. In comparison to prevailing approaches, a key benefit from the proposed approach is an improved computational load/failure of a single point, energy trading strategy, workload, and capital cost energy. Findings suggest that energy costs reduced between 7.60% and 25.41% for prosumer buildings and a fall of 5.40%-17.63% for consumer buildings. In practical applications, the proposed approach can involve a larger number of prosumer and consumer buildings within the community to decentralise multiple energy trading, thus significantly contributing to the reduction of greenhouse gas emissions and enhancing environmental sustainability. [ABSTRACT FROM AUTHOR]

Personalized thermal sensation models play a crucial role in ensuring occupant's thermal comfort satisfaction and improving building energy efficiency. However, an adaptive and accurate personalized model that can be easily implemented in real life is still challenging. This paper investigates the influencing factors of thermal sensation vote (TSV) and proposes a personalized regression model that only uses a single local skin temperature as the key indicator. A survey is conducted with forty subjects aging from 20 to 59 years old. The relationship among ambient temperature, skin temperatures, and subjective TSV is analyzed. The forehead temperature is recommended as the key indicator for prediction because it exhibits a strong correlation with ambient temperature and TSV, and it is easy to capture. Furthermore, the impact of individual characteristics on TSV is investigated. The proposed model effectively captures and compensates for individual differences by incorporating subjects' set point skin temperature and body fat percentage (BF%). The proposed model can be readily applied in real-life scenarios due to its minimal requirement for occupant's feedback and its higher accuracy compared to other models. Specifically, it exhibits a significantly lower Root Mean Square Error (RMSE) of 15.8 %, 9.4 %, and 65.2 % compared to the Support Vector Regression (SVR) model Zhang's model and Zhou's model. Moreover, the proposed model showcases the lowest mean absolute error among the compared models. This approach of developing a personalized regression model based on local body temperature holds promise for future international ergonomic standard development. [ABSTRACT FROM AUTHOR]

[Display omitted] • Infrared technology enables non-invasive temperature measurement. • Existing literature related to personal thermal comfort using infrared technology was reviewed. • Measurement equipment, experimental conditions, subject's information, and analytical methods were analyzed. • Accurate predictions should be translated into control commands for the equipment at different scales. • Diverse application scenarios including car cabin, sleep, and outdoor environments were discussed. Personal thermal comfort studies are crucial for balancing both human comfort and building energy savings. Thus, personal comfort models (PCMs) are expected to play a pivotal role in the development of smart buildings. Infrared technology has received significant attention from researchers because of its non-intrusive nature for collecting personal data. This systematic review examined 36 personal thermal comfort studies that employed infrared technology over the past decade. The review analyzed the (1) infrared devices, (2) experimental designs, (3) subjects, and (4) analysis methods and results. The review found that most studies were experimental and focused on exploring feasibility at a theoretical level, while the number of practical application studies was limited. The accuracy of infrared devices was identified as a primary concern for researchers, and some studies proposed methods to mitigate measurement errors. Most studies included young subjects and examined single clothing insulation and metabolic rates, limiting the generalizability of the corresponding PCMs, which were mainly constructed using machine learning. To facilitate data sharing and comparison of results and to overcome the limitations in terms of applicable populations and application scenarios, the review suggests that standardized data collection and processing should be used in future studies. Furthermore, the paper proposes the integration of PCMs and personal comfort systems, which are two hotspots in personal thermal comfort research, to achieve cooperative applications. Finally, the study discusses the future applications of infrared technology in scenarios such as vehicle cabins, sleep areas, and outdoor environments, considering their unique characteristics. [ABSTRACT FROM AUTHOR]

• Metadata inclusion and threshold optimization improve state models. • State models outperform action models in predictive performance. • Sequential state models outperform non-sequential models. • The short-term past sequence does not significantly enhance action models. • Explainable AI enables local and global comparison of state and action models. Window operation significantly impacts energy use and indoor environmental quality in buildings. Individuals behave differently, making it difficult for models trained on a specific dataset to encompass the entire spectrum of these actions. A generalizable model is essential to predict the behavior of diverse occupants in office spaces. To address this need, this paper presents a systematic approach that captures this diversity, thereby contributing to developing a model towards generalizability. The approach involves state and action modeling through a Random Forest algorithm on the ASHRAE Global Occupant Behavior Database. The data pre-processing, hyperparameter tuning, and evaluation are deeply described and applied to window action and state datasets. Our results demonstrated that including metadata in a state model and applying a G-Mean threshold moving technique can result in an F 1 -score of 0.74. This score slightly outperformed the state room-wise model, which was trained only on its own dataset and achieved an F 1 -score of 0.73. However, both models had similar accuracies of 77%. The action model did not fare as well as the state models, with an F 1 -score and accuracy score of just 0.42 and 49%, respectively. In contrast, the action model showed more explainable results for domain experts than state models. [ABSTRACT FROM AUTHOR]

• Predict thermal comfort non-invasively from skin temperature features. • Identify ROIs of face component based on 68-point face landmarks in thermal images. • Achieves 90.26% accuracy and withstands diverse head pose conditions. • Enhance occupant-centric control and better thermal comfort prediction. This paper proposes a framework for predicting thermal comfort based on the automated extraction of skin temperature features from thermal images. This aims to non-invasively collect thermal comfort information for occupant-centric control (OCC), significantly impacting energy consumption, health, and overall well-being. The proposed framework adopts 68-point face landmarks to identify regions of interest (ROIs) of face components in thermal images, and subsequently extracts skin temperature features from those ROIs to predict thermal comfort. To assess its performance, the face landmark detection performance was evaluated using various colormaps on thermal images. Furthermore, the validity of the proposed skin temperature features extraction from ROIs of face components was evaluated, determining which skin temperature features are effectively entered into machine learning models. Additionally, the reliability of the framework for predicting thermal comfort under different head pose conditions was evaluated to ensure its validity. The results revealed that the proposed framework achieved an accuracy rate of 90.26% and showed robustness even in the extreme head pose. The study's findings suggest that the proposed framework can make OCC more effective based on more accurate thermal comfort prediction using a single thermal camera device. [ABSTRACT FROM AUTHOR]

The construction industry in the United States consumes more than 50% of the global energy supply per year, suggesting that significant efforts may be needed to reduce building energy demand and its carbon footprint. During their lifespans, buildings consume embodied energy (EE) and operational energy (OE). Building professionals, therefore, conduct life cycle energy assessments (LCEA) to quantify and understand the paradox and interconnectedness between EE and OE. Traditionally, simulation-based optimization techniques were used for design space exploration to identify a building design with the fewest energy implications. However, literature shows these data-driven approaches are often error-prone, time-consuming, and computationally expensive, and they fail to provide real-time feedback to the user. Moreover, EE and OE assessment tools are disjointed and suffer from interoperability issues. These limitations restrict design space exploration, which eventually hinders the design decision-making process. Over the last few years, the increased availability of building data has made machine learning (ML) techniques a popular choice for building performance assessments. Several articles have developed prediction models to assess or optimize OE. While this work is significant, studies utilizing ML techniques for building LCEA are lacking, mainly due to the unavailability of a large-scale LCEA database. In this paper, we propose a methodology to (1) generate a simulation-based building energy dataset for different building typologies using a parametric framework, (2) utilize the synthetically generated database to develop an ML-based prediction model to predict EE and OE, and (3) test the model using a case study. The case study results show that the model achieves high prediction performance using minimal inputs available during the early design phase. The results further indicate that ML techniques can be used by building designers with no or limited LCE expertise to instantaneously estimate building LCE performance and help them select design options with minimal LCE consumption. [ABSTRACT FROM AUTHOR]

For cooling equipment management and scheduling optimization, accurate building cooling load forecasting technology is crucial. Currently, the physics-based forecasting models are too complex to achieve, and existing shallow-machine and deep learning algorithms are difficult to capture and retain sequential information from historical building cooling loads, leading to insufficient prediction accuracy. This paper considered the dependency relationship between time-series information in load data and proposed a building load prediction model based on a transformer network to improve the accuracy of building load prediction. This encoder-decoder block-based model can encode and decode all input data, capture sequence information from mapping vectors with user-defined dimensions, and learn important features through the Attention mechanism. In addition, input features were analyzed to verify the importance of each input feature, and to explaine the reasons for the impact of used features on the TRN-based model. Finally, the performance of the proposed model is evaluated using real data from an office building. Compared with other existing methods, the proposed model has the best prediction accuracy (RMSE, MAE, R2 were 0.01, 0.03, and 0.98, respectively), and maintained the best predictive stability over a longer time (uncertainty ranged from −11% to + 11%). The results show that the proposed method can support the development and optimal operation of energy-saving HVAC systems, thereby lowing power consumption. [ABSTRACT FROM AUTHOR]

Residential air-conditioning systems are subject to soft faults, which can reduce their efficiency, capacity, and life expectancy, without being detected by the occupants. Early detection and diagnosis of faults could potentially be done by using machine learning (ML) on measurable features of the air-conditioner, such as refrigerant temperatures and pressures. This paper describes a novel approach to applying ML, which is to train a fault classifier using data from one system, and apply it to different systems. If this is effective, it can significantly reduce the cost of widescale application of ML fault classification. The support vector machine (SVM) algorithm was used. An additional question for practical application is whether the number of sensors required for input feature measurement can be reduced. This work also tests the feature reduction effect, in combination with applying an ML classifier trained on one system, to classify faults in two other systems. The results are disappointing, showing that fault classification is poor when the SVM classifier is trained on a different system than it is applied to. However, with additional work, this method has potential for improvement, possibly to a point where it can be a viable approach for field deployment. [ABSTRACT FROM AUTHOR]

Energy load forecasting is critical for sustainable building development and management. Although the energy data could be collected through Internet of Things (IoT) systems, it is a big challenge to train a large-scale machine learning model due to data isolation. Since the building energy data could reveal confidential information such as user behaviors and building operations, the privacy regulations would not allow central service to collect distributed data from data owners directly. This paper designs a secure federated data analytics system for forecasting community buildings' energy data load. A novel adaptive weight federated learning algorithm is proposed to handle the system faults frequently happening during networking operations. Moreover, a new deep learning model is re-invented to improve energy load forecasting performance. The experiments of the system are performed on an actual university campus dataset, and the results show the new federated algorithm improves the load forecasting accuracy and achieves the best load forecasting result. The new deep learning model improves the forecasting accuracy by almost 10% on error reduction under the same federated learning settings. To evaluate the load forecasting model's practical usefulness, an anomaly prediction pipeline is designed through the combination of gaussian mixture model and load forecasting model, which reveals the system's effectiveness at building energy management that 92% F1 score with 97% accuracy is achieved by the best model. • Develop an adaptive federated learning algorithm for community-level building energy load forecasting. • Re-invent a new deep learning load forecasting model through a combination of existing methodologies. • Design an anomaly prediction pipeline through a combination of load forecasting models and unsupervised GMM. • Evaluate the practical usefulness of load forecasting and anomaly prediction through extensive experiments. [ABSTRACT FROM AUTHOR]

• Propose a novel machine learning-based model predictive control (MLB-MPC) framework with predictive inputs of cooling load and outdoor wet-bulb temperature. • Obtain the optimal match of prediction models and optimization algorithms for improving the performance of the MLB-MPC. • Investigate the impact of disturbances on the performance of global optimization. • Achieve an energy-saving ratio of 7.1% for the air-conditioning system when compared with the measured operation data. The dynamic optimization of key setpoints (e.g., supply water temperatures of chillers and cooling towers, or indoor temperature and humidity) can track the efficient performance point of the air-conditioning system (ACS), thus obtaining a great energy-saving effect. The model predictive control is an effective way to precisely adjust these setpoints. However, most of the existing studies focused on simple linear strategies, e.g., experience-based, rule-based, or component-based, which results in unsatisfactory energy efficiency. Aiming at this, the paper proposed a novel machine learning-based model predictive control (MLB-MPC) framework with predictive inputs of disturbances. The framework first employed the grid search and cross-validation to optimize the total energy consumption prediction models (TECPMs), which were built by data-driven models (e.g., multiple linear regression, artificial neural network, support vector regression (SVR), and random forest). And then the optimization performances of each TECPM to the controlled variables were obtained by using different optimization algorithms (e.g., genetic algorithm, particle swarm optimization (PSO), and simulated annealing). Finally, the optimal match of TECPMs and optimization algorithms was achieved by trade-off among prediction accuracies, optimization accuracies, and optimization time. The case studies demonstrated that the optimal match for the MLB-MPC is SVR and PSO as they had the highest prediction accuracy (mean absolute percentage error of 2.5%) and shortest optimization time (41 ms/per) and optimization accuracies with little difference. After adopting the optimal match, the MLB-MPC with predictive disturbing inputs achieved a great energy-saving ratio of 7.1% for the ACS, which was only less than the MLB-MPC with ideal disturbing inputs by 0.4%. This indicated that exact predictions of disturbances (e.g., cooling load and outdoor wet-bulb temperature) are important for the MLB-MPC. The proposed MLB-MPC framework would provide a method for improving the energy efficiency of ACSs. [ABSTRACT FROM AUTHOR]

• A 5th-generation district heating and cooling system in business district is studied. • Design and operation strategies are based on load analysis and load forecasting. • Load predictions are based on machine learning algorithms and historical loads. • Larger flow rate in water loop than that on source-side makes system more efficient. • Source-network-load-storage collaboration in a district energy system is explored. The fifth-generation heating and cooling system (5GDHC) is a bidirectional low temperature network, which is a good solution for decarbonization and energy-saving in building cooling and heating. Because of its complex configurations, 5GDHC requires specific advanced control strategies for different scenarios. This paper aims to investigate 5GDHC with multiple sources in the business district in hot-summer and cold-winter zone through modelica simulation. The analysis and forecasting of district loads were used to study the design and operation strategies of 5GDHC. A suitable heat pump performance curve was presented for 5GDHC after comparing several heat pumps. And the larger the total flow in the network, the more beneficial it is for users. In this study, 5GDHC for the business district performed best when it adopted the strategy of continuous operation of the ground heat exchanger, nighttime operation of the surface-water coil, and supplemental cooling of the cooling tower upon it is needed, resulting in a 16.6% reduction in CO 2 emissions compared to the strategy that all sources operated continuously. Besides, carbon emission analysis showed the importance of balancing operational cost with carbon emission. This study will help optimize the design and operation of 5GDHC in business district scenarios. [ABSTRACT FROM AUTHOR]

• Renewable heat performance demonstrated in Air Source Heat Pump field trial. • Useful statistical and machine learning models to predict instantaneous Coefficient of Performance (COP). • Estimate of performance difference between in-situ COP and manufacturer's rated datasheet values. Air Source Heat Pumps (ASHPs) are part of the solution to decarbonise the residential heating sector. The coefficient of performance (COP) is a measure of the instantaneous efficiency of a heat pump. The heat energy produced by an ASHP is deemed renewable if it meets a specified sustained COP over a period of time, e.g., a seasonal performance factor (SPF). Heat pump performance in situ often differs from laboratory test conditions. This paper explores the performance of ASHPs in a field trial of deeply retrofitted Irish houses. Analysis shows that all houses in the trial qualify as producing renewable heat but vary from the manufacturer's laboratory test performance. ASHPs with ratings of 8.5 kW (11.2 kW) underperformed against the manufacturers COP values on average by 16 (24%) at outside temperatures of 7 °C, and 3 (11%) at outside temperatures of 2 °C. The paper evaluates statistical and machine learning models, including a random forest (RF) model, to predict COPs and compare the difference between the manufacturer's laboratory COPs and the trial results. [ABSTRACT FROM AUTHOR]

• Limited generation capacity and penetration of renewables need demand response. • Energy-storing water heaters are suitable for demand response with scheduling. • Artificial intelligence potentially suited due to patterned hot water usage. • Review of scheduling techniques and machine learning for optimal scheduling. • Machine learning clear advantages over classical control strategies. A high penetration of renewable energy sources such as wind power generation and photovoltaic generation causes some problems in power systems such as the duck curve and unreliability due to environmental variability. An effective solution to this problem is Demand Response (DR). Electric Water Heaters (EWHs) are considered ideal candidates for DR due to their energy storage capability. Due to the benefits, control strategies or techniques for EWHs have received considerable academic attention. The energy sector has recently tapped into the disruptive artificial intelligence world to learn, among other related priorities, how to enhance operations, maintain energy resilience and improve consumer service. Consequently, this paper reviews the use of machine learning (ML) for optimization and scheduling of EWHs. The main contributions of this review paper are, firstly, to identify state of the art of energy optimization and scheduling of EWHs. Secondly, to review the current ML models for energy optimization and scheduling of EWHs in smart grids and smart building environment. While classical control strategies may deliver substantial improvements, optimum efficiency may not be reached. ML has demonstrated clear advantages over classical control. Based on these conclusions, recommendations for further research topics are drawn. [ABSTRACT FROM AUTHOR]

• We review sixteen models for Variable Refrigerant Flow System (VRF). • We develop three data-driven models for VRF. • Physics-based and data-driven models are compared on a real-world dataset. • The impacts of different input features on VRF efficiency are analyzed. Variable Refrigerant Flow (VRF) systems are gradually gaining popularity in small and medium-sized commercial and residential buildings owing to their high part-load performance, flexible control, and ease of installation and maintenance. Developing models of VRF systems to predict their performance are important for model-based control, fault diagnostic and detection. There are VRF models published in existing literatures, however those models were developed and validated in different datasets. As a result, the model accuracy cannot be directly compared. To fill this gap, this paper presents a comprehensive review of the existing VRF models, and summarizes the input/output parameters and mathematical formulas of 16 VRF models from literature (referred to as physics-based model). Next, we validate and compare the model accuracy of existing models using the same dataset. Additionally, we develop data-driven models using the state-of-art machine learning algorithms, and compare the model accuracy between existing physics-based models with data-driven models. We find the model proposed by Hu et al. in 2019, which regresses the VRF cooling capacity and COP as a linear combination of indoor and outdoor temperatures times a cubed polynomial function of compressor frequency, is the most accurate physics-based model, with a prediction error of 22.19% in the training dataset and 22.44% in the validation dataset. XGBoost is the most accurate data-driven model, with a prediction error of 19.29% in the training dataset and 22.02% in the validation dataset. The data-driven model is more accurate while the physics-based model is more generalizable. The findings of this study can help researchers to select the proper VRF model for building energy prediction, model-based optimization, and fault diagnostic and detection. [ABSTRACT FROM AUTHOR]

Traditional Heating, Ventilation and Air-conditioning systems operate on a fixed schedule, regardless of occupancy or external temperature. With the rise of smart buildings, building managers and owners are seeking ways to reduce energy consumption while maintaining occupant comfort. There are various environmental and personal factors that impact thermal comfort levels. In this paper, we aim to develop a machine learning-based approach for precisely forecasting the thermal comfort levels of building occupants using the readily available ASHRAE RP-884 database. The dataset was first pre-processed using a k Nearest Neighbor (kNN)-based imputation method. Then, the RRelifF algorithm was used to select highly relevant and practically retrievable input features. Based on the selected features, four ensemble models were created to assess the relationship between input features and thermal comfort levels. A decision-making rule was employed to determine the credibility of each predictor, with only credible outputs selected. The credible outputs were aggregated to produce the final predicted mean vote (PMV) using Genetic Algorithm (GA) optimized coefficients. The testing results were promising, exhibiting high accuracy and a low Root Mean Squared Error (RMSE) value of 0.157 in the prediction of PMV. [ABSTRACT FROM AUTHOR]

Energy studies of buildings are becoming more widespread as stakeholders strive to improve energy efficiency and reduce carbon emissions. As a result, there is an increased need for novel numerical techniques to automatically calibrate building energy models (BEMs) for these energy studies. In this paper, a new automated building calibration methodology was developed, which uses a surrogate (i.e. , meta) multilayer perceptron artificial neural network (ANN) to infer unknown building parameters. Typically, in the field of BEM calibration, surrogate models are used as a time-saving technique. Instead of running the building simulation software during each iteration of an optimization search, a surrogate model can approximate the output of the BEM very quickly, leading to much faster optimization of the unknown building parameters. However, we show that, once trained, the surrogate model itself can be used to find the unknown building parameters. Since the surrogate model is differentiable, gradient descent can be used to find the building parameters that minimize the error between true and predicted metered energy consumption. Previous surrogate modeling approaches for calibrating BEMs map only the unknown building parameters to the BEM's output. As a result, they are blind to the effects of other crucial variables like weather and the building's schedules. On the other hand, our surrogate ANN model accounts for these other predictors of energy consumption. With this advantage, our method was able to outperform a powerful black box optimizer when finding 14 unknown building parameter values in a controlled case study with hourly energy data. In a real metered data case study, our ANN method and the black box optimizer performed similarly on average, but our method had less variance in performance across trials. [ABSTRACT FROM AUTHOR]

Optimal planning and operation studies of modern energy systems are tied up with medium to long-term predictions of energy demand. Deep learning algorithms have recently become significantly useful in this regard because of their capacity to learn load patterns with a high degree of fluctuation and uncertainty. Yet, empirical evaluation of different configurations of deep learning models, for enhancing prediction accuracy, has been mostly ignored in the preceding literature. In this paper, different architectures of deep recurrent neural networks (DRNNs) are explored and adjusted specifically for the purpose of performing medium- and long-term predictions of energy demands. We intend to create a bespoke DRNN for heating and electricity consumption prediction with a 1-h resolution. Moreover, hyperparameter optimization, which is a time-consuming and rigorous task in deep learning algorithms due to their abundance, dependence on the particular application, and empirical nature, is studied comprehensively. In this respect, the space of possible configuration variables is thoroughly explored to identify a collection of hyperparameters that enables the DRNN model to more accurately forecast energy consumption. We also benchmark the proposed algorithm's performance against other data-driven models, namely support vector machine (SVM) and gradient boosting (GB) regression. The proposed model outperforms SVM by 5.4% and GB regression by 7.0% in terms of energy forecasting accuracy. [ABSTRACT FROM AUTHOR]

• A novel Digital Twin framework is presented to detect occupant discomfort. • The Digital Twin approach integrates BIM with sensor data & occupant feedback. • Bayesian network used to evaluate occupant comfort in non-residential buildings. • Nine ML algorithms are used for predictive maintenance. • XGB is up to 5% more accurate than other algorithms while RF is 96% faster than XGB. This study introduces a Bayesian network model to evaluate the comfort levels of occupants of two non-residential Norwegian buildings based on data collected from satisfaction surveys and building performance parameters. A Digital Twin approach is proposed to integrate building information modeling (BIM) with real-time sensor data, occupant feedback, and a probabilistic model of occupant comfort to detect and predict HVAC issues that may impact comfort. The study also uses 200000 points as historical data of various sensors to understand the previous building systems' behavior. The study also presents new methods for using BIM as a visualization platform and for predictive maintenance to identify and address problems in the HVAC system. For predictive maintenance, nine machine learning algorithms were evaluated using metrics such as ROC, accuracy, F1-score, precision, and recall, where Extreme Gradient Boosting (XGB) was the best algorithm for prediction. XGB is on average 2.5% more accurate than Multi-Layer Perceptron (MLP), and up to 5% more accurate than the other models. Random Forest is around 96% faster than XGBoost while being relatively easier to implement. The paper introduces a novel method that utilizes several standards to determine the remaining useful life of HVAC, leading to a potential increase in its lifetime by at least 10% and resulting in significant cost savings. The result shows that the most important factors that affect occupant comfort are poor air quality, lack of natural light, and uncomfortable temperature. To address the challenge of applying these methods to a wide range of buildings, the study proposes a framework using ontology graphs to integrate data from different systems, including FM, CMMS, BMS, and BIM. This study's results provide insight into the factors that influence occupant comfort, help to expedite identifying equipment malfunctions and point towards potential solutions, leading to more sustainable and energy-efficient buildings. [ABSTRACT FROM AUTHOR]

• Building performance simulation is powerful but computationally expensive. • Machine Learning (ML) surrogate models can reduce simulation times significantly. • Building energy and peak loads are predicted under different operational settings. • Extreme gradient boosting outperformed other methods. • Linear regression showed the lowest training time with competitive predictive power. Building Performance Simulation (BPS) is a powerful and widely used technique to evaluate building design and operation strategies prior to construction or retrofitting. However, BPS models often have high computational costs, which is particularly limiting for applications that require a significantly large number of simulations, such as building design optimization or uncertainty analyses. To overcome this gap, researchers have turned to surrogate modeling, where a mathematical model, such as a machine learning algorithm, is trained to mimic the performance of a BPS, allowing to test numerous building design/operation configurations at low computational costs. Past studies have applied surrogate BPS modeling to predict the impact of building design parameters on energy performance. However, few have considered building operational parameters, such as occupancy, equipment and lighting usage, and thermostat setpoints, which significantly impact energy consumption and peak loads, especially in harsh climate conditions. This paper presents a unique evaluation and comparison of machine learning algorithms as surrogates to BPS predictions of building performance (energy consumption and peak loads) under different operational settings. Results indicate that Extreme Gradient Boosting outperformed all other methods in predictive accuracy, with R2 values reaching as high as 0.99 for some models. In contrast, linear regression models were the fastest to train and easiest to interpret while still achieving competitive prediction accuracies (R2 values > 0.9). This work provides direct evidence of the machine learning surrogate models' ability to accurately predict building performance under different operational settings. It also offers unique insights into the strengths and weaknesses of white-box and black-box predictive modeling approaches and the effect of dataset size on the results. [ABSTRACT FROM AUTHOR]

• Developed and applied a feature selection algorithm to unitary equipment at part load conditions. • Feature relevancy analysis of the air and refrigerant side is performed. • Feature selection algorithm verified using Artificial Neural Network. • A subset of 33% of features are identified with the same level of accuracy as a complete dataset. • Developed guidelines for the formulation of future models using algorithms. A variable-speed air-conditioning system provides more control than single-or multi-stage units, resulting in increased energy efficiency and more precise temperature control. Current commercial and residential direct expansion (DX) equipment testing standards require the embedded control placed on variable-speed equipment to be deactivated during performance testing and instead the fan and compressor speeds are fixed. For this reason, the test and field performances differ, creating a performance as well as an energy consumption gap. This paper presents a machine learning-based approach to investigate critical features from the air and refrigerant side influencing the cooling capacity of unitary air conditioning equipment. High-fidelity experimental data obtained from 4 different state-of-the-art high-efficiency units, 1 fixed-speed 35.2(10) kW(tons) commercial and 3 variable-speed residential unitary equipment of 12.3(3.5), 14(4), and 17.6(5) kW(tons) of cooling capacity have been analyzed by utilizing a feature selection methodology, Elastic Net (EN). Influential features with higher relevance scores corresponding to the total cooling capacity are fed into a supervised Artificial Neural Network (ANN), formulated as a predictive model to evaluate the validity of features selected. Results show that the proposed method of feature selection when applied, ANN is able to predict the equipment cooling capacity with a mean absolute percent error of less than 2% for all tested units, thus approving the proposed input features. The findings may be used as recommendations for the development of a semi-empirical model for better cooling capacity and COP predictions. [ABSTRACT FROM AUTHOR]

Very little work has been done on the feasibility of Machine Learning (ML) for predicting buildings energy demand right at the design stage. This feasibility, if proven, would help to avoid the construction of inefficient buildings. This paper uses dataset from 7559 buildings, and estimates their energy consumption using nine ML models. Results show that deep neural network (DNN) is the most efficient ML model with MAE, MSD and RMSE of 0.93, 1.12 and 1.06 respectively achieved in less than 7 s despite the huge data size. Its r 2 is also the highest (0.96) which means that the DNN approach manages to explain 96% of the energy consumption in buildings and only 4% remains unexplained certainly due to the limitation of independent variables. Also, this result is not affected by building clusters nor by data from a particular climate zone. As an innovation, this study proposes a model that professionals could use in the design phase of a construction project. This model will allow them to take into account all crucial aspects of the design of an energy efficient building. The model will then serve as a decision-making tool to control and optimise the project and to anticipate energy consumption even before the building is constructed. [ABSTRACT FROM AUTHOR]

• A two-step novel method is proposed to create hourly building meta -models. • The most significant feature of this new meta -model method is sequential. • Ensemble machine learning models are recommended to create building meta -models. • Distributed lag machine learning models are fit for hourly building energy models. The meta -models have been widely used to replace computationally expensive engineering-based building energy models for model calibration, sensitivity analysis, and performance optimization of buildings. However, most meta -models are used for monthly or annual building energy analysis applying the same procedure as building energy or load prediction. A few studies thoroughly explore the characteristics of meta -models when creating hourly machine learning-based meta -models for building energy assessment, which is very different from conventional building load or energy prediction. Therefore, this paper proposes a new method of creating time series models for building energy analysis based on machine learning techniques. The main feature of this new method is to include two steps (model preselection and data folding) by taking advantage of the sequential nature during the process of constructing meta -models. A case study of office buildings is used to demonstrate the application of this new approach using the EnergyPlus program and R computational environment. The results indicate that the time-series meta -models constructed using this new method have very high accuracy with a moderate computational cost. For the best models, the R2 values are above 0.99 and the CV(RMSE) (coefficient variation of root mean square error) values are below 0.06 for both heating and cooling energy prediction. The ensemble machine learning models, including Cubist, gradient boosting machine, and stacking, are recommended to create the meta -models for hourly building energy analysis. The distributed lag models combined with these ensemble models can be used to utilize the time-series features of hourly building energy assessment. The method proposed here can be extended to the time-series energy analysis in different time scales (sub-hourly, hourly, or daily) for other energy systems (such as PV, solar thermal, and wind). [ABSTRACT FROM AUTHOR]

• Implementation and benchmarking of five advanced control methods. • Comparison of black-, gray-, and white-box MPC with RL and approximate MPC. • Introduction of qualitative KPIs to assess the applicability. • Application in a one-year simulation to evaluate adaptability. Advanced building control strategies like model predictive control and reinforcement learning can consider forecasts for weather, occupancy, and energy prices. Combined with system and domain knowledge, this makes them a promising approach to reduce buildings' energy consumption and CO 2 emissions. For this reason, model predictive control and reinforcement learning have recently gained more popularity in the scientific literature. Nevertheless, publications often lack comparability among different control algorithms. The studies in the literature mainly focus on the comparison of an advanced algorithm with a conventional alternative. At the same time, use cases and key performance indicators vary strongly. This paper extensively evaluates six advanced control algorithms based on quantitative and qualitative key performance indicators. The considered control algorithms are a state-of-the-art model-free reinforcement learning algorithm (Soft-Actor-Critic), three model predictive controllers based on white-box, gray-box, and black-box modeling, approximate model predictive control, and a well-designed rule-based controller for fair benchmarking. The controllers are applied to an exemplary multi–input–multi–output building energy system and evaluated using a one-year simulation to cover seasonal effects. The considered building energy system is an office room supplied with heat and cold by an air handling unit and a concrete core activation. We consider the violation of air temperature constraints as thermal discomfort, the yearly energy consumption, and the computational effort as quantitative key performance indicators. Compared to the well-tuned rule-based controller, all advanced controllers decrease thermal discomfort. The black-box model predictive controller achieves the highest energy savings with 8.4 %, followed by the white-box model predictive controller with 7.4 % and the gray-box controller with 7.2 %. The reinforcement learning algorithm reduces energy consumption by 7.1 % and the approximate model predictive controller by 4.8 %. Next to these quantitative key performance indicators, we introduce qualitative criteria like adaptability, interpretability, and required know-how. Furthermore, we discuss the shortcomings and potential improvements of each controller. [ABSTRACT FROM AUTHOR]

The introduction of smart building technology promises many operational and productivity benefits and enables smart grid integration. A significant barrier to deploying smart building applications is mapping the building sensor metadata to the requirements of the smart building applications. In this paper, we have studied the problem of weakly supervised machine learning to accelerate the metadata mapping process, and demonstrated that weak supervision is a promising approach, which to the best of our knowledge, has not been studied in this context. We developed a pattern-based workflow that enables subject matter experts to craft simple rules that process the idiosyncratic text found in building sensor metadata. Our method was validated using three medium-sized commercial office buildings and a public dataset. We worked closely with a team of commercial building subject matter experts to develop and validate our workflow. We compared our approach to active learning and discovered that weak supervision appears to scale better with building size. Subject matter experts overwhelmingly preferred this approach compared to manual methods, and our study with three buildings shows that weak supervision can reduce the time to annotate the metadata in building systems by a factor of 4. [ABSTRACT FROM AUTHOR]

[Display omitted] In recent years deep neural networks have been proposed as a lightweight data-driven model to capture high-dimensional, nonlinear physical processes to predict building thermal responses. However, the need of a large amount of data for the training process of deep neural networks clashes with the potential limited data availability in most existing or new buildings. Transfer learning aims to enhance the performance of a target learner exploiting knowledge from related and similar environments. This study conducted a suite of experiments that leveraged 250 data-driven models based on a synthetic dataset of a building archetype to study the influence of data availability, energy efficiency level, occupancy and climate for the transfer process of thermal dynamics. The performance of the transfer learning process was compared against a classical machine learning approach. The results suggest that building thermal dynamics can be effectively transferred under the same climatic conditions, increasing performance when dealing with different occupancy schedules, efficiency levels and low data availability. Furthermore, the paper compares the performance of both transfer learning and machine learning approaches in an online fashion, to support the implementation in real-world deployment. [ABSTRACT FROM AUTHOR]

• Potential electricity demand charges reductions between 63% and 75% during a year. • Tested with real data from an industrial, an educational, and a residential consumer. • Methodology allows demand charges savings even without utility input or information. • Methodology based on use of arithmetic models and tree-based machine learning models. This paper presents a novel methodology that empowers virtually any electricity consumer paying for peak demand charges to proactively execute demand response actions even without receiving signals or information coming from the utility, and only when necessary to effectively reduce demand charges and user inconvenience. The proposed methodology employs different arithmetic models and tree-based machine learning models to determine an efficient electricity demand threshold value before the start of a billing period. This methodology is completely model agnostic so additional models can be integrated without changing the proposed process. The threshold value produced can be used to proactively trigger peak demand shaving and other demand response actions in order to reduce demand charges. The results obtained using real data showed that regression random decision forest based models outperformed different arithmetic models and other tree-based machine learning models at determining this threshold value for an industrial, an educational with solar photovoltaic electricity generation, and a residential consumer. The results also showed that the consumers evaluated could potentially achieve between 63% to 75% of total potential demand charge reductions during a year. These results translate to US$ 159.00, US$ 23,290.00, and US$ 107,389.00 in savings for the residential, industrial, and educational consumer respectively. [ABSTRACT FROM AUTHOR]

• Operation data of chiller is collected through the Internet of Things Intelligent Agent. • A hybrid model of across working conditions fault diagnosis of chiller is proposed. • Domain knowledge is used to assist deep learning avoids directly extracting features. • The proposed model achieves outstanding effect under the across working conditions. In the real application of chiller fault diagnosis models, the various working conditions may cause the different data distribution between test data and training data, which may lead to the lower diagnosis accuracy of models. In this paper, a novel fault diagnosis method for chillers across working conditions based on the domain knowledge-assisted Deep Extreme Learning Machine (DELM) is proposed. Firstly, the actual normal and performance degradation operation data of the chiller is collected through the Internet of Things Intelligent Agent (IOTIA). The Random Forest (RF) is used to determine the importance of features selected by domain knowledge analysis, and different feature subsets are selected as the model input. Then, the model of the DELM is used for fault diagnosis of chillers across varying temperature and load rate working conditions. Finally, experiments show that the proposed model achieves outstanding effect under the designed seven experimental across working conditions, indicating that the model has good generalization ability and is suitable for fault diagnosis across working conditions of chillers. [ABSTRACT FROM AUTHOR]

• We use a machine learning (ML) approach to study energy poverty (EP) risk in Europe. • We produce an EP risk classifier with accuracies ranging from 60% to 75%. • Through ML we identify three EP predictors: income, household size, and floor area. • We suggest the presence of universal predictors complemented by contextual ones. • Increasing data collection efforts in Europe can improve EP policy research insights. In this paper we identify drivers for energy poverty in Europe using machine learning. The establishment of predictors for energy poverty valid across countries is a call made by many experts, since it could provide a basis to effectively target energy-poor households with adequate policy measures. We apply a "low income, high expenditure" framework that classifies households as being at risk of energy poverty to a dataset from a survey conducted at the household-level in 11 European countries with vastly different economies, cultures, and climates. A gradient boosting classifier is successfully trained on a set of socio-economic features hypothesized as predictors for energy poverty in this diverse set of countries. The classifier's internal model is analyzed, providing novel insights into the intricacies that underlie energy poverty. We find that besides the main driver - income - floor area and household size can be confirmed as predictors. Our results suggest the presence of universal predictors that are valid across Europe, and contextual ones that are governed by local characteristics. To facilitate advanced research into energy poverty in Europe, we recommend to increase and streamline household data collection efforts, both at the country- and EU-level. [ABSTRACT FROM AUTHOR]

[Display omitted] • A critical analysis of the properties of the most widely used residential, commercial and industrial datasets for NILM. • Comprehensive presentation of all the feature extraction and pre-processing techniques that have been applied in energy disaggregation domain. • Up-to-date detailed overview of the existing NILM approaches, focusing mainly on the latest machine and deep learning methods. • Key summaries are provided based on visualizations of NILM performance across various datasets/appliances/methods. • Assessment of the existing performance evaluation methods and suggestion of a proper comparability and evaluation framework. This paper presents a critical approach to the non-intrusive load monitoring (NILM) problem, by thoroughly reviewing the experimental framework of both legacy and state-of-the-art studies. Some of the most widely used NILM datasets are presented and their characteristics, such as sampling rate and measurements availability are presented and correlated with the performance of NILM algorithms. Feature engineering approaches are analyzed, comparing the hand-made with the automatic feature extraction process, in terms of complexity and efficiency. The eolution of the learhes through time is presented, making an effort to assess the contribution of the latest state-of-the-art deep learning models to the problem. Performance evaluation methods and evaluation metrics are demonstrated and it is attempted to define the necessary requirements for the conduction of fair evaluation across different methods and datasets. NILM limitations are highlighted and future research directions are suggested. [ABSTRACT FROM AUTHOR]

This paper focuses on the modelling of occupant behaviour in the case of a non-residential mixed-mode building on the tropical island of La Réunion. For such areas and types of buildings, occupants can operate passive solutions to achieve comfort while energy-consuming ones can offer alternatives during the hottest months. Yet, compared to other climatic zones, specific knowledge on occupant comfort and behaviour is limited, making the work of engineers difficult during the design phase. In this work, occupants' operations on hygrothermal comfort controls, such as windows and fans, were first measured and analysed. Secondly, these behaviours were modelled using two deterministic methods based on machine learning techniques and a probabilistic graphical model. A model was also implemented to estimate the number of people, using the power demand of the electrical outlets. The estimation ability of the behavioural models was evaluated and led to F1 scores greater than 0.7. A two-classifier model was proposed to estimate the level of ceiling fan use. This combined model slightly improves the F1 scores by more that 2%, which demonstrates the necessity of taking into account the links between the different controls. [ABSTRACT FROM AUTHOR]

In recent years, a large number of building energy consumption prediction models, with various intended uses, have been proposed. The majority of these models have either taken a data-driven or a physical modeling approach, with each approach having its own strengths and limitations. Towards leveraging the strengths and reducing the limitations of each approach for improved prediction performance, this paper presents a hybrid machine-learning approach for occupant-behavior-sensitive energy consumption prediction. The proposed approach is composed of three constituent models: (1) a machine-learning model that learns the impact of outdoor weather conditions from simulation-generated data, (2) a machine-learning model that learns the impact of occupant behavior from real data, and (3) an ensemble model that predicts cooling energy consumption based on the outputs of the first two models. The simulation-generated data were created through simulating a set of reference buildings in EnergyPlus. The real data were collected from an office building in Pennsylvania. The proposed hybrid model was validated on an unseen real dataset. It achieved 0.73 kWh RMSE and 9.07% CV in hourly cooling energy consumption prediction, which indicates that the proposed approach is promising. [ABSTRACT FROM AUTHOR]

[Display omitted] • Multimodal occupancy detection framework for residential buildings. • Two variants of Occ-STPN models with feature level fusion and decision level fusion. • Feature extraction and occupancy detection models for various data modalities. • Transferability of a trained occupancy detection model from one building to another. • Occupancy prediction performance evaluation on three open-sourced datasets. For several years now, smart building energy systems have been a research area of intensive activity. In light of the increasing need for sustainable buildings and energy systems, this trend motivates an increasing need for a solution to reduce carbon dioxide emissions and improve energy efficiency. This paper proposes a high-performing and transferable occupancy detection framework that combines sensor data from different data modalities, including time series environmental data (temperature, humidity, and illuminance), image data, and acoustic energy data using ensemble method. To draw out the best prediction performance in each modality, the proposed framework was developed, including various models that were designed to learn the occupancy patterns reflected in the physical data streams. To tackle the time series environmental data, we designed two variants of an occupancy detection spatiotemporal pattern network (Occ-STPN) that performs both feature level and decision level fusion, respectively. We also propose a new metric; the fading memory mean square error (FMMSE), that provides a fair evaluation and penalization of delayed occupancy predictions. Multiple open-sourced datasets, including the Electricity Consumption and Occupancy and the University of California, Irvine's (UCI) building occupancy detection dataset, along with our own real data collected from six different houses, were used to validate the algorithms' performance. The experimental results presented herein break down the performance for each sensing modality, and a detailed analysis of the performance is also discussed. [ABSTRACT FROM AUTHOR]

• Future occupancy-state and arrival-time prediction models are developed. • A performance score based on thermal comfort and energy efficiency is proposed. • kNN, DT, MLP and GRU machine learning techniques are implemented. • Occupancy-state prediction models provide superior control performance. • Using only accuracy to evaluate occupancy models might not be reliable in some cases. This paper analyzes the impact of using different types of occupancy prediction models on the performance of occupancy-based heating, ventilation, and air conditioning (HVAC) control systems in residential buildings. The future occupancy prediction problem is approached from two different viewpoints: 1) as a regression task for arrival-time prediction and 2) as a classification task for occupancy-state prediction. Four machine learning techniques, namely decision trees, k-nearest neighbors (kNN), multi-layer perceptron, and gated recurrent units, are implemented to predict future occupancy from each perspective. The performance of the occupancy models is evaluated in terms of mean absolute error (MAE), accuracy, and the ability to provide occupants' thermal comfort and to save energy. An overall performance score is proposed by making a trade-off between the energy and thermal comfort objectives using the technique for order of preference by similarity to the ideal solution method. The results demonstrate that selecting the optimal viewpoint has a higher impact on the control performance than selecting the machine learning techniques, with occupancy-state prediction models providing superior performance in most cases. It is also shown that the machine learning evaluation metrics (i.e., MAE and accuracy) provide a weak to moderate correlation with the overall performance score. Consequently, relying solely on the MAE and accuracy might fail to provide a reliable evaluation of the occupancy model performance for use in HVAC control systems. [ABSTRACT FROM AUTHOR]

[Display omitted] • Urban wind resource is a promising renewable energy source. • The characteristics of urban wind are chaotic and full of turbulence. • Conventional wind mapping methodologies are not enough to judge the characteristics of urban wind. • An effective, precise, economic, and swift technique is essential for the assessment of urban wind. • Artificial intelligence including machine learning, deep learning, and neural networks are predicted to be the ultimate choice for urban wind mapping to ensure its commercialization. Wind energy in urban areas is a prominent alternative source of renewable energy that is becoming increasingly popular. The technological development, low cost of installation, and availability of wind make it a promising and essential renewable energy sector. Wind farms in urban settings, on the other hand, have yet to be commercialized in a meaningful way. In metropolitan areas, the urban wind becomes turbulent and unpredictable to some extent due to high-rise buildings and variations in their structures, railway tracks, and roads. Hence traditional wind mapping technologies have proven ineffective in metropolitan areas, preventing its commercial use. This paper presents a thorough analysis of the most recent urban wind-resource evaluation techniques, covering multiple strategies, methodologies, software instruments, and the impact of site and building on the accurate installation of wind farms. Noise and vibrations, as well as aesthetic and shadow effects, ecological impact, high investment prices, and data availability, have all been considered, with potential solutions proposed. It has been found that in metropolitan settings, the traditional method of analyzing wind resources is hampered by the inability to collect data in a turbulent environment. As a result, this study predicts that artificial intelligence, such as machine learning, deep learning, and multiple neural networks, will dominate this industry in order to provide precise forecasts while minimizing the operational time and costs of the utilization system. [ABSTRACT FROM AUTHOR]

[Display omitted] • Most of the reviewed studies (62%) focused on group-based comfort models. • ML models, especially PCMs outperformed conventional models (PMV and adaptive). • ML-based controls can improve IAQ while reducing energy consumption and CO 2 levels. • The most frequently used algorithms were SVM, ANN and Ensemble Learning. This paper provides a systematic review on the application of Machine Learning (ML) in thermal comfort studies to highlight the latest methods and findings and provide an agenda for future studies. Reviewed studies were investigated to highlight ML applications, parameters, methods, performance and challenges. The results show that 62% of reviewed studies focused on developing group-based comfort models, while 35% focused on personal comfort models (PCMs) which account for individual differences and present high prediction accuracy. ML models could outperform PMV and adaptive models with up to 35.9% and 31% higher accuracy and PCMs could outperform PMV models with up to 74% higher accuracy. Applying ML-based control schemas reduced thermal comfort-related energy consumption in buildings up to 58.5%, while improving indoor quality up to 90% and reducing CO 2 levels up to 24%. Using physiological parameters improved the prediction accuracy of PCMs up to 97%. Future studies are recommended to further investigate PCMs, determine the optimum sample size and consider both fitting and error metrics for model evaluation. This study introduces data collection, thermal comfort indices, time scale, sample size, feature selection, model selection, and real world application as the remaining challenges in the application of ML in thermal comfort studies. [ABSTRACT FROM AUTHOR]

Non-intrusive load monitoring (NILM) is a topic that lately attracts both the academic and the industrial interest. NILM is used to reveal useful information regarding the consumption breakdown on appliance or activity level, thus can be a key solution to unlock various smart-home services and opportunities. To that end, deep learning has arisen as a prominent solution. Although most of the known solutions so far focus on a predefined number of home appliances, this paper proposes a multi-class NILM system which can detect in real-time any number of appliances and can be efficiently embedded into simple microprocessors. The key feature for the identification of the appliances is the processing of measured turn-on active power transient responses sampled at 100 Hz. The NILM system includes three stages; adaptive thresholding event detection method, convolutional neural network and k-nearest neighbors classifier. For future extensions, it is capable to automatically identify new appliances; thus, no retraining and additional modeling is required. [ABSTRACT FROM AUTHOR]

• A framework of building energy modeling for data predictive control is developed. • The framework integrates active learning strategy with feature selection technique. • The framework is demonstrated in a virtual testbed (EnergyPlus and BCVTB). • The case study shows improvement in energy modeling accuracy and DPC performance. Three gaps impede the development of cost-effective and accurate data-driven building energy modeling/models (DBEM) for energy forecasting and predictive control strategies. Gap 1: data bias is common in building operation data, but this topic is hardly studied in DBEM; Gap 2: high data dimensionality is common in DBEM, but a systematic and scalable methodology is lacking to solve the problem; Gap 3: the interactions between data bias and high dimensionality have not been systematically studied for DBEM and predictive control in buildings. In this paper, to address the three gaps mentioned above, we develop a framework that integrates active learning and feature selection for DBEM used for whole building data predictive control (or DPC, which is a branch of model predictive control). The framework provides a systematic methodology and automatic workflow that starts with raw data from building automation systems to the establishment of data-driven energy models for DPC controllers. The developed strategies and framework are evaluated in a virtual testbed based on EnergyPlus and BCVTB. Improved performance and reduced computational complexity are observed from the DBEM built with the developed framework, as well as the DPC controller based on that DBEM, indicating the effectiveness of the developed framework. [ABSTRACT FROM AUTHOR]

An effective and accurate building energy consumption prediction model is an important means to effectively use building management systems and improve energy efficiency. To cope with the development and changes in digital data, data-driven models, especially deep learning models, have been applied for the prediction of energy consumption and have achieved good accuracy. However, as a deep learning model that can process high-dimensional data, the model often lacks interpretability, which limits the further application and promotion of the model. This paper proposes three interpretable encoder and decoder models based on long short-term memory (LSTM) and self-attention. Attention based on hidden layer states and feature-based attention improves the interpretability of the deep learning models. A case study of one office building is discussed to demonstrate the proposed method and models. Firstly, the addition in future real weather information yields only a 0.54% improvement in the MAPE. The visualization of the model attention weights improves the interpretability of the model at the hidden state level and feature level. For the hidden state of different time steps, the LSTM network will focus on the hidden state of the last time step because it contains more information. The Transformer model gives almost equal attention weight to each day in the coding sequence. For the interpretable results at the feature level, daily max temperature, mean temperature, min temperature, and dew point temperature are the four most important features. The four characteristics of pressure, wind speed-related features, and holidays have the lowest average weights. [ABSTRACT FROM AUTHOR]