Search

Artificial neural network (ANN) based surrogate models have been widely used in place of high-fidelity simulation tools. Error metrics such as mean absolute error, and root mean squared error (RMSE) have been widely used as de facto criteria. However, whether the ANN-based surrogate model can adequately reproduce the interwoven relationships and nonlinear causalities between design variables and simulated outputs are often overlooked. In this regard, the authors designed a case study regarding four ANN-based surrogate models. It was found that despite all of the models having low RMSEs, the models failed to adequately predict the causal relationships between input variables and energy use. In other words, the surrogate models were not always capable of providing accurate assessments of expected energy use reduction as a result of design changes. In this paper, we present a workflow for validating whether a surrogate model can reproduce the causal relationships between inputs and outputs. [ABSTRACT FROM AUTHOR]

The 'ideal' optimal control of a radiant floor heating system (RFHS) can be achieved when a central controller is fully aware of the interrelationship between the different thermal dynamics of a boiler, radiant floor, and room. This study proposed integrated control strategies that can consider the interaction of heterogeneous RFHSs using a minimalistic approach with minimal data easily available from residential buildings. Three individual artificial neural networks were developed to choreograph the heterogeneous dynamics of the RFHSs for each household. Then, model predictive control was applied to realize the real-time integrated control of the RFHSs in the residential buildings. Approximately 14–46% of heating energy could be saved by integrated control. Moreover, the potential energy saving rate from integrated control varied depending on the degree of human intervention. [ABSTRACT FROM AUTHOR]

Accurate prediction of indoor daylight illuminance is crucial for daylight-based lighting controls. However, determining the illuminance using physics-based simulation tools requires significant amounts of information, e.g. grid of sensors, sky model, 3D geometry of a target building and surroundings, etc. In this study, the authors suggest a daylight illuminance estimation method with minimal data of two reference sensors and two prior measurements. It is shown that the daylight coefficient and sky luminance distribution can be substituted by the illuminance of the reference points and illuminance of two or more target points at past times. The method was validated on a large open space with north-facing skylight windows and showed an 11.9% mean absolute percentage error. Additionally, a reference point selection method is presented. The proposed method is practical for daylight-based lighting control applications. [ABSTRACT FROM AUTHOR]

This paper presents a real-time implementation of model predictive control (MPC) for HVAC systems in an ice-cream factory building. The target building consists of two large open spaces served by two HVAC systems. We developed four artificial neural network (ANN) models that predict the thermal states of the supply air and indoor air of the two thermal zones and prove to be accurate enough (MBE = 2.65, CVRMSE = 9.43). The control variables employed in this study are the number of operating chillers, frequency of supply-air fan inverter and outdoor-air intake ratio. The objective function minimizes total energy use, and a constraint was set to maintain average indoor air temperatures close to set points. Real-time MPC was implemented at a sampling time of 20 min from 3 August to 30 August 2021 and could save approximately 31.7% of electricity when compared to the existing simple rule-based control. [ABSTRACT FROM AUTHOR]

It is important to control the heating system by following real-time demand, while considering the dynamic changes and non-uniform distributions of indoor environments. This paper presents a model predictive control (MPC) scheme for predicting indoor air temperatures at multiple points in a large factory building that consists of large irregular spaces and heat-generating equipment. Instead of using a full-blown dynamic simulation model (e.g. EnergyPlus), the authors developed a lumped simulation model. This model can accurately predict the temperatures and is, therefore, used for the optimal on/off control of 61 unit heaters installed in the factory building. Based on the MPC, energy savings of 56.3% were realized over three weeks, and the indoor air temperatures were maintained within a comfortable range. It is highlighted in the paper that this MPC approach based on the minimalistic lumped model can accurately predict indoor thermal behaviour and save significant energy. [ABSTRACT FROM AUTHOR]

Although it is widely acknowledged that reinforcement learning (RL) can be beneficial for building control, many RL-based control actions remain unexplainable in the daily practice of facility managers. This paper reports a rule reduction framework using explainable RL to enhance the practicality of the control strategy. First, deep Q-learning was applied to explore the optimal control strategies of a parallel cooling system (ice-based thermal system + geothermal heat pump system) of an existing office building. A set of modularized and interconnected data-driven models was developed using ANNs for pretraining an artificial agent. After exploring the control strategies, the decision-making rules of the agent were reduced using a decision tree. The performance of the reduced-order rule-based control proved comparable to the complex and uninterpretable control strategy of deep Q-learning. The difference in energy savings between the two is marginal at 1.2%. [ABSTRACT FROM AUTHOR]

This paper presents an integrated lighting control that employs a daylit illuminance prediction model for a large, open-spaced factory building in which interactions between multiple luminaires and points of the workplane exist. The prediction model was developed with Radiance and consists of daylight and electric lighting prediction components. Both components showed reliable accuracy after calibration with measured illuminance (daylighting: MBE = 4.9%, CVRMSE = 24%; electric lighting: MBE = 3.7%, CVRMSE = 7.7%). An optimal policy trained by the deep deterministic policy gradient determines the dimming levels of multiple luminaire groups. The policy is developed with an artificial neural network model whose input is the current state (daylight distribution) and whose output is an action (lighting control variables). The model could provide an appropriate amount of electric lighting that meets the target illuminance and uniformity, while only 54% of the average consumption power (7,292 W) was needed compared with that of the existing rule-based control (13,629 W). [ABSTRACT FROM AUTHOR]

This study introduces a hybrid model that combines physics and machine learning (ML) models to describe the behaviour of variable refrigerant flow (VRF) systems. The standalone ML model was developed with identical data and conditions for comparison between the hybrid and ML models. A Bayesian neural network (BNN) was used for both the models, and the predictive abilities and uncertainties were investigated. For the experimental dataset, the predictive performances of both models were similar. For example, the predictive performance of the hybrid and ML models showed mean absolute error of 0.73 and 0.78 kW, respectively. However, the epistemic uncertainty of the hybrid model quantified using the BNN was 36.4% lower than that of the ML model. A parametric study showed that the hybrid model combined with the physics model can achieve better generalization performance than the ML model, yielding results that are more reliable and physically explainable. [ABSTRACT FROM AUTHOR]

This study proposes a transfer learning (TL)-based inverse modelling to identify unknown building properties. This study examines the transfer from virtual buildings to existing buildings, especially for identifying wall U-value, HVAC efficiency and lighting power density (LPD). For this purpose, synthetic data were generated from simulation results of sampled EnergyPlus models, and then we developed artificial neural network (ANN) models using this data. By adopting TL, the ANN models were transferred to the domain of existing buildings and evaluated on 61 existing buildings. As a result, the relative improvements in CVRMSE achieved by the transferred models against the models trained only with existing buildings' data were 8.85%, 10.34% and 15.73% for nominal cooling COP, wall U-value and LPD, respectively. Moreover, it is expected that the use of TL enables the developed model to be reusable for another group of buildings with improved performance and reduced training time. [ABSTRACT FROM AUTHOR]

Data-driven simulation models can be used for optimal control and decision making in building systems only if they are developed accurately and reliably. An unsolved issue pertains to the data collected from BEMS that generally contain errors owing to sensors, malfunctioning systems, and other unknown reasons. These anomalies generally occur in a multidimensional space and cannot be easily detected. Unfortunately, these anomalies can prevent the simulation model from being accurate, reliable, and scalable. This paper presents an automatic anomaly detection method using a support vector data description (SVDD). The data obtained from a compression chiller in a real office building were classified into raw and filtered datasets based on SVDD. Two artificial neural network (ANN) models were developed (ANNraw and ANNSVDD). Based on the comparison between the two ANNs, the ANNSVDD is found to be better than ANNraw in terms of the model reliability and reproducibility. It is also interesting that the model accuracy differences between the two ANN models are marginal. However, the accuracy of the two ANN models can be improved as long as they are tested against filtered data by SVDD. The improvements in accuracy signify the importance of the elimination of these anomalies. [ABSTRACT FROM AUTHOR]

A deep Q-network (DQN) was applied for model-free optimal control balancing between different HVAC systems. The DQN was coupled to a reference office building: an EnergyPlus simulation model provided by the U.S. Department of Energy. The building was air-conditioned with four air-handling units (AHUs), two electric chillers, a cooling tower, and two pumps. EnergyPlus simulation results for eleven days (July 1–11) and three subsequent days (July 12–14) were used to improve the DQN policy and test the optimal control. The optimization goal was to minimize the building's energy use while maintaining the indoor CO2 concentration below 1,000 ppm. It was revealed that the DQN—a reinforcement learning method—can improve its control policy based on prior actions, states, and rewards. The DQN lowered the total energy usage by 15.7% in comparison with the baseline operation while maintaining the indoor CO2 concentration below 1,000 ppm. Compared to model predictive control, the DQN does not require a simulation model, or a predetermined prediction horizon, thus delivering model-free optimal control. Furthermore, it was demonstrated that the DQN can find balanced control actions between different energy consumers in the building, such as chillers, pumps, and AHUs. [ABSTRACT FROM AUTHOR]

This study describes the development of the optimal control strategies of eight parallel heat pumps in an existing building. The building consists of seven floors above ground and two floors underground with a total floor area of 22,440 m2. The chilled water generated by each of the eight parallel heat pumps runs through a common primary pipe to multiple air-handling units in the building. Because only one flowmeter and two thermometers (entering and exiting) are installed in the primary pipe, the heat removal rate and efficiency of each heat pump are unknown. The existing control of the heat pumps is as follows: if the chilled water return temperature in the primary pipe becomes greater than a predetermined temperature, the controller increases the number of operating heat pumps. The heat removal rate and efficiency of each heat pump were first identified using a Gaussian process (GP) machine-learning algorithm to develop the optimal control strategy of the eight heat pumps. Two GP models, one for estimating the heat removal rate and the other for estimating the coefficient of performance (COP), were developed based on the measured data for 27 days in July at the sampling time of 15 min. After developing the GP models, the authors applied a COP-based sequencing control strategy to the eight parallel heat pumps. The new optimal control strategy is to switch on the heat pumps in order from highest to lowest COP. Compared with the existing control logic, the new optimal control can reduce energy consumption by 20.9%. [ABSTRACT FROM AUTHOR]

This article compares two modeling approaches for optimal operation of a turbo chiller installed in an office building: (1) a machine learning model developed with artificial neural network (ANN) and (2) a hybrid machine learning model developed with the ANN model and available physical knowledge of the chiller. Before developing the ANN model of the chiller, the authors used Gaussian mixture model in order to check the validity of measured data. Then, the hybrid model was developed by combining the ANN model and physics-based regression equations from the EnergyPlus engineering reference. It was found that both the ANN and hybrid ANN model are satisfactory to predict the chiller's power consumption: mean bias error (MBE) = −2.63%, coefficient of variation of the root mean square error (CVRMSE) = 8.05% by the ANN model; MBE = −3.99%, CVRMSE = 11.98% by the hybrid ANN model. However, the hybrid model requires fewer inputs (four inputs) than the ANN model (eight inputs). The energy savings of both models are similar coefficient of performance (COP) = 4.32 by the optimal operation of the ANN model; COP = 4.44 by the optimal operation of the hybrid ANN model. In addition, the hybrid ANN model can be applied where the ANN model is unable to provide accurate predictions. [ABSTRACT FROM AUTHOR]

The philosophy of NEN 2916 was utilized for the US federal government agency, GSA as part of GSA building energy performance toolkit during 2002-2007. After a long career spanning more than 40 years of teaching and research in building simulation, Professor Godfried (Fried) Augenbroe died on 14 May 2021 at the age of 73. Graph In his seminar class designed to train M.Sc. and PhD students in Building Performance and Simulation at Georgia Tech, Fried would face a dozen students every year, most of whom would be new to this sub-field that sits between architecture and engineering. [Extracted from the article]

Copyright of Building Research & Information is the property of Routledge and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)