Search

• We proposed a novel modal decomposition based ensemble learning method for GSHP energy consumption prediction. • Variational mode decomposition and empirical mode decomposition were used to decompose the energy consumption data. • The method was compared with two common used machine learning models with optimization process. • The method showed outstanding effect on improving the performance of GSHP energy consumption prediction. This study presents a case study of office buildings using modal decomposition based ensemble learning method to forecast energy consumption of ground source heat pump systems (GSHP). Conventional machine learning methods have uncertainty in practical application as there are lots of stochastic terms in the structure of the algorithm. Therefore, ensemble learning models are proposed to ameliorate the problem. In this paper, the prediction potential of modal decomposition based ensemble learning models are investigated while providing a comprehensive comparison on different single models and ensemble learning models in the building field. Results show that the proposed VMD based ensemble learning models have remarkable advantages in GSHP energy consumption prediction comparing with other machine learning methods, and the prediction performances of VMD based ensemble learning models measured by RMSE and MAE are 30–50% better than common machine learning methods. This work is enlightening and indicates that VMD based ensemble learning models fit well with energy consumption prediction, which could bring more efficient and concise solutions for GSHP energy consumption predictions. [ABSTRACT FROM AUTHOR]

This paper depicted the novel data mining based methods that consist of six models for predicting accurate future heating and cooling load demand of water source heat pump, with the objective of enhancing the prediction accuracy and the management of future load. The proposed model was developed to ease generalization to other buildings, by making use of readily available measurements of a comparatively small number of variables related to water source heat pump operation in the building environment. The six models are - tree bagger, Gaussian process regression, multiple linear regression, bagged tree, boosted tree and neural network. The input parameter comprised the prescribed period, external climate data and the diverse load conditions of water source heat pump. The output was electrical power consumption of water source heat pump. In this study, simulations were conducted in three sessions - 7-day, 14-day and 1-month from 8th July to 7th August 2016. The forecast precisions of data mining models were measured by diverse indices. The performance indices which were used in assessing the prediction performance were - mean absolute error, coefficient of correlation, coefficient of variation, root mean square error, mean square error and mean absolute percentage error. The mean absolute percentage error results for 7-day future energy demand forecasting from tree bagger, Gaussian process regression, bagged tree, boosted tree, neural network and multiple linear regression were 3.544%, 0.405%, 1.703%, 1.928%, 2.592% and 13.053%, respectively. Moreover, when the proposed data mining model performance was compared with the existing studies, the mean absolute percentage error of 2.515% was found out for the first session, 7-day. The results also showed that the six models were efficient in foreseeing the abnormal behavior and future cooling and heating load demand in the building environment. [ABSTRACT FROM AUTHOR]

Energy consumption models play an integral part in energy management and conservation, as it pertains to buildings. It can assist in evaluating building energy efficiency, in carrying out building commissioning, and in identifying and diagnosing building system faults. This review takes an in-depth look at energy-demand prediction models for buildings in that it delves into recent developments in building energy approaches used to predict energy usage. By enlisting current approaches to the modelling of buildings, methods for building energy simulations can be categorized into four level classes as follows: (i) data-driven approaches; (ii) physics-based approaches; (iii) large scale building energy forecasting approaches; and (iv) hybrid approaches. The focus of this review is to target the data-driven approach and large-scale building energy predicting-based approaches. Here the data driven approaches can be categorized by (1) artificial neural network based approaches; (2) clustering based approaches; (3) statistical and machine learning-based approaches; and (4) support vector machine based approaches. From there, the type of data-driven based approach is further grouped by (a) benchmarking models; (b) energy-mapping models; (c) energy forecasting models; and (d) energy profiling models. Large-scale building-energy prediction techniques is then categorized as follows: (1) white-box based approaches; (2) black-box based approaches, and (3) grey-box based approaches. The current study explores first-rate data-driven based approaches about building energy analysis for industrial, commercial, domestic, etc., within a rural and urban setting. This review paper is based on the necessity of identifying points of departure and research opportunities for urban and rural-level analyses of building level energy performance. A variety of issues are explored which include: energy performance metrics; end-use of different building types; multiple levels of granularity; and urban and rural scales. Each technique encompasses a variety of input information as well as varying calculations or simulation models along with furnishing contrasting outcomes that suggest a variety of usages. A thorough review of each technique is presented in this study. This review highlights strengths, shortcomings, and purpose of the methods of numerous data-mining based approaches. A comprehensive review of energy forecasting models that are specified in the literature part is also provided. [ABSTRACT FROM AUTHOR]

An intelligent fault diagnosis network for variable refrigerant flow air conditioning system is proposed in this study. The network is developed under the foundation of bayesian belief network theory, which comprises two main elements: the structure and parameters. The structure obtained by machine learning and experts’ experiences illustrates the relationships among faults and physical variables from the qualitative prospective, and its parameters (including prior probability distribution and conditional distribution) describe the uncertainty between them quantitatively. Once the structure and parameters are determined, the posterior probability distribution which can be used to complete fault diagnosis and isolation will be calculated by some algorithms. In comparison with other fault diagnosis approaches, the proposed approach can make full use of performance information. Moreover, it is more reasonable and precise to express the relationship between faults and variables rather than Boolean variables. Evaluation was conducted on a variable refrigerant flow air conditioning system, which demonstrated that this strategy is effective and efficient. [ABSTRACT FROM AUTHOR]

This paper presents a statistical training data cleaning strategy for PCA-based chiller sensor Fault Detection, Diagnosis and Data Reconstruction method. Finding and removing outliers from the original training data set, the training data quality can be improved by the presented data-cleaning strategy. This can enhance the efficiency of the fault detection and increase the accuracy of the data reconstruction. Outliers cannot be easily found in the original data set used for training the PCA model. These outliers would severely affect the projection directions of PCA's two orthogonal subspaces (PC subspace and Residual subspace). Therefore, the threshold of Q -statistic is changed by the unexpected projection subspaces so that the detection efficiency of the sensor fault is decreased. The Euclidean distance was employed as an index to detect outliers from the original training data. In order to achieve optimal training data for the sensor FDDR, the z -scores of each sample's Euclidean Distance were employed as the key to remove the outliers. A field measured data set of a screw water-cooled chiller was used to validate the presented strategy. Results demonstrate that the quality of the training data is optimized and sensor fault detection efficiency, as well as the reconstruction data accuracy, is improved when compared to the normal PCA method. [ABSTRACT FROM AUTHOR]

Highlights: [•] The more the porosity, the lower the pressure inside FADS. [•] The higher the porosity, the poorer the uniformity of air velocity distribution. [•] The effect of fiber diameter is similar with the effect of porosity. [•] Increasing air flow rate aggravates the nonuniformity of air velocity distribution. [•] Results would help guide the optimum design of FADS in penetration mode. [ABSTRACT FROM AUTHOR]

Abstract: This paper presents a self adaptive chiller sensor fault detection strategy based on Principal Component Analysis (PCA) method, namely a self-Adaptive Principal Component Analysis (APCA) method. The original data set used to train the PCA model usually contains some error samples, whose useless residual subspace information make the threshold of Q-statistic higher than the expected threshold. This leads to a low sensitivity and low fault detection efficiency at low sensor fault levels. APCA is developed to automatically remove error samples in the original data set in order to improve fault detection efficiency especially for temperature sensor faults with absolute magnitude less than 1°C. The self adaptive process of APCA has been presented and been compared with the Normal Principal Component Analysis (NPCA) method. The APCA strategy is validated by the operational data of a screw chiller system in a real electric factory. The results show that the APCA method can significantly enhance the fault detection efficiency, and the symmetry for positive and negative fault levels with same absolute magnitude becomes better than NPCA method. [Copyright &y& Elsevier]

Abstract: No existing diffuser models may be used to effectively describe the fabir air dispersion system (FADS), a new ventilation terminal made of ploymer. A porous media model was proposed to describe FADS in present work. And the corresonpongding mathematical model was developed for the airflow through FADS. The characteristics of airflow through FADS was predicted using the commercial software FLUENT with the standard k − ɛ turbulent model. The simulation results were well matched with the corresponding experimental value suggesting that it is feasible to describe FADS using the porous media model. Results showed that the air in the central zone flows forward with a decreasing velocity along the flow direction, and the air in the porous fibre eventually creeps out through its tiny-size holes at very low speed in the direction perpendicular to the porous fibre. The airflow close to the FADS is evenly distributed along the flow direction with a very low velocity. It is also concluded that the internal total pressure and dynamic pressure decrease slightly, while the internal static pressure increases slightly with a decaying increment along the flow direction of the FADS. [ABSTRACT FROM AUTHOR]

• 1-D CNN is an innovative approach utilized for fault diagnosis in VRF system. • 1-D CNN models achieves good accuracy and is capable of fault diagnosis. • Enhanced the method to an ensemble model by using parallel 1-D CNNs. • The ensemble model's performance is improved and archived state-of-the-art accuracy. Variable refrigerant flow (VRF) systems are widely-adopted air conditioning systems. When system faults occur in VRF systems, the efficiency of VRF system will drop drastically. This paper presents a single 1-D CNN model and an ensemble model with parallel 1-D CNNs for diagnosing VRF system refrigerant charge faults under heating condition. From the cleaned experiment data of a commercial VRF system, 15 features are selected as the input for the proposed model with ReliefF algorithm. After training, the diagnosis accuracy of the single 1-D CNN model and ensemble 1-D CNN models is evaluated and compared with that of BPNN model and DT model. The result shows that both single 1-D CNN and ensemble 1-D CNN model can diagnose VRF system refrigerant charge fault effectively. The fault detection is also achieved in proposed models. The average diagnosis accuracy of 9-level refrigerant charge faults of the ensemble 1-D CNN model is up to 97.4%, surpassing that of BPNN model, SVM model, DT mode and DBN model. 1-D CNN based model is utilized for VRF system fault diagnosis for the first time, which lays a foundation for the expansion of the related researches. [ABSTRACT FROM AUTHOR]

With the development in information technologies, today's building energy consumption can be well monitored by the building energy management systems. However, in most real applications there is no clear definition of abnormal building energy consumption. To overcome this limitation, this work proposes a novel deep learning based unsupervised anomaly detection framework that includes recurrent neural networks and quantile regression. Moreover, this framework is able to produce a prediction interval to detect and evaluate abnormal building energy consumption. The framework has been applied to analyze the energy data collected from three different residential houses, and anomaly detection results are evaluated by the quantile regression range. The research results can provide promising solutions for building managers to detect abnormal energy performance, and is also valuable to assess the level of anomalies and spot opportunities in energy conservation. [ABSTRACT FROM AUTHOR]

The variable refrigerant flow (VRF) system has changeable energy performance due to shifty meteorological conditions and unstable internal factors (e.g. complex control strategies, various part load ratios). It is difficult to determine the causes of drastic energy consumption variations are normal factors or faults. This study therefore proposed an energy diagnosis method for VRF systems based on data mining techniques and statistical quality control approaches. The correlation analysis is employed to select key factors and the DBSCAN method is used to remove the transient data and outliers. Besides, the system power consumption is predicted using the SVR algorithm. The EWMA control chart is used to diagnosis the system energy performance and it is comparatively analyzed with the x ¯ -R and CUSUM control charts. The reliability of the proposed method is verified by diagnosing the system energy in refrigerant undercharge and overcharge conditions, respectively. Results show that the SVR method is reliable to predict the system energy consumption with a R 2 value of 0.9974. In addition, the EWMA control chart can significantly improve the VRF energy diagnosis performance compared to the x ¯ -R and CUSUM control charts. It achieves high correctly diagnosis ratios at both refrigerant undercharge and overcharge conditions. [ABSTRACT FROM AUTHOR]

Analysis and application for real-time operational data of buildings is important for energy management. But original data inevitably contains a number of outliers and which usually lead to a significant negative impact on performance of data-based models. In order to eliminate the influence of outliers and improve the robustness of data-based models, this study employs three methods (Boxplot, local outlier factor (LOF) and PCOut) to identify the potential outlying observations in original data set. For purpose of evaluating the outlier detection performance of these three methods, four SVR-based electricity consumption prediction models, Original-SVR, BOX-SVR, LOF-SVR and PCO-SVR, are established. And the performance indexes (RE, RMSE and RSE) of the models are compared and analyzed. The results show that the accuracy of electricity consumption prediction is improved with the help of Boxplot and LOF methods for outlier detection, but PCOut method reduces the accuracy compared with the Original-SVR model. Further study indicates that these observations which repeatedly identified as outlying by Boxplot and LOF methods are the most likely to be abnormal, and when these samples are removed from training data set, the RMSE falls to 2.76 from 6.44 and the RSE falls to 0.11 from 0.58 during testing course. [ABSTRACT FROM AUTHOR]

The outdoor air-side fouling fault is almost inevitable for the practical variable refrigerant flow system. Fouling fault grows gradually and naturally causing increasingly energy penalty and performance degradation to the system. Implementing a proper and reliable fault detection and diagnosis strategy is crucial for practical systems keeping away from fouling fault and maintaining optimal operations. However, traditional model-based and data-driven methods cannot work in practical systems due to the lack of critical sensors and interpretation for model reliability, respectively. Therefore, this study proposes a data mining approach to identify and isolate fouling faults using only built-in sensors. Density-based spatial clustering of applications with noise(DBSCAN) is used for data pre-processing. The classification and regression tree(CART)-based classifier is employed for fault detection. Based on Pearson’s correlation analysis, a multiple linear regression(MLR)-based fault indicator is developed for fault isolation. A case study on a 29.8 kW cooling capacity R410A variable refrigerant flow system is used to validate the proposed strategy. Four levels of foulings are experimentally investigated under three cooling conditions. Results reveal that the proposed strategy correctly identifies 98% of fouling data using only three built-in sensor measurements. It also isolate fouling data from both normal and refrigerant charge fault data. [ABSTRACT FROM AUTHOR]

Sensor faults of air conditioning systems are harmful to optimal control strategies and system performance resulting in poor control of the indoor environment and waste of energy. In order to improve the fault detection and diagnosis (FDD) performance, this paper presents an enhanced sensor fault detection and diagnosis method based on the Satizky-Golay (SG) method and principal component analysis (PCA) method for the VRF system, namely SG-PCA method. Due to the volatility of the original data set of VRF system, the original data are smoothed using SG method at first. Then, the smoothed data are used for PCA model training and fault detection and diagnosis. In order to determine parameters of the SG method, an optimization index is proposed, which is calculated by the signal to noise ratio (SNR), the standard deviation (SD) and the self-detection efficiency. This SG-PCA method for VRF system sensor FDD is validated using field operation data of the VRF system. Various sensor faults at different fault levels are introduced. The results have showed that the SG-PCA method can significantly improve the fault detection and diagnosis performance compared to conventional PCA method. [ABSTRACT FROM AUTHOR]

• The influences of various common faults on the VRFs under different fault levels and load rates was analyzed. • A faults experienced database for VRFs was provided. • The fault mechanism was clarified through sensitivity analysis between the physical parameters and different faults. • A reference basis for selecting various fault features and key fault decoupling variables was provided when a multiple fault diagnosis model for VRFs is established. Due to improper installation, operation, and maintenance, many common faults inevitably occur in the air conditioning systems. However, the same fault could demonstrate different characteristics under different load rates and fault levels; different faults may present similar fault phenomena. This paper reveals the quantitative impact of faults on system performance of VRF system with common faults imposed in heating mode through fault experiments. The system parameters sensitivity study has been carried out to understand the internal impact mechanism related to system parameters. This provides a reference for selecting characteristic parameters and fault decoupling variables and developing interpretable, professional, and reliable fault detection and diagnosis models. The results show that the outdoor unit fouling fault can cause the heating capacity to drop by 73.2%, and the indoor unit fouling can cause the system power to drop by 65.5%. The coefficient of performance drop is as high as 48.6%. The sensitivity study result shows that the system load rate is an influential parameter; it needs to be considered when characteristic parameters and fault decoupling variables are selected. [ABSTRACT FROM AUTHOR]

This paper presents an entire sensor fault detection and diagnosis (FDD) for screw chiller system using the support vector data description (SVDD) algorithm. It has advantages of solving problems on describing non-linear and non-Gaussian distributed data. A distance-based D -statistic plot is employed to detect sensor faults. Based on the distance transformation in mathematical way, a new distance variation-based DV -contribution plot is proposed to diagnose the sensor fault. The screw chiller field measured data is utilized to train the SVDD model via a hybrid parameter tuning approach combined the grid search and the 10-fold cross validation. Six typical sensor faults are introduced for validation, i.e. positive and negative biases, positive and negative drifts, precision degradation and complete failure. Test results of both D -statistic and DV -contribution plots show that the SVDD-based method has good FDD results for the six sensor faults. Furthermore, the proposed DV -contribution plot shows more accurate fault diagnosis results compared with the principal component analysis (PCA)-based Q -contribution plot. [ABSTRACT FROM AUTHOR]

• Proposed an interpretable deep learning model using the attention-BiLSTM. • Four typical faults of VRF are well identified with an overall diagnosis accuracy of 97.5%. • Effective fault diagnosis under incomplete information is realized through the integrated model of different feature combinations. • Realize the quantification of diagnosis uncertainty through the ensemble model. • The model performs well on the unknown data set under different working conditions. Developing effective fault detection and diagnosis (FDD) model is of great significance to the energy efficiency and comfort of variable refrigerant flow (VRF) system. In recent FDD studies, the consideration of incomplete information and uncertainty, as well as the generalization of model and the interpretability of fault action mechanism have become a great concern. This paper accordingly proposes a fault diagnosis strategy based on attention-BiLSTM with ensemble feature sets. Three methods including Correlation Analysis, Information Entropy, and Gini Impurity are used for feature selection and ensemble feature sets combination. The proposed model can achieve interpretability of fault diagnosis and quantification of diagnosis uncertainties through attention mechanism and ensemble feature sets. A total of four typical faults under different working conditions are used to verify the accuracy and generalization of the proposed method. The results show that the ensemble attention-BiLSTM model has a fault diagnosis accuracy rate of 98.3% if the first two most likely outputs are considered, and when some sensors fail, the accuracy rate can still remain above 90%. In addition, a generalization verification idea is proposed, that is, the model is trained under specific operating conditions and tested on datasets with different operating conditions or fault levels. The results show that the average generalization of the optimized model reaches 86%. [ABSTRACT FROM AUTHOR]

• A combined generative network is built based on VAE and WGAN-GP. • The ensemble classifiers are embedded into generative network for FDD. • A detailed comparison between SMOTE and generative network method is discussed. Faults in building Heating, Ventilation, and Air-condition (HVAC) system create an uncomfortable indoor environment and cause energy waste. The data-driven method has been widely applied for Fault Detection and Diagnosis (FDD) in the complex building HVAC system. This method relies on the availability of many fault data which is difficult to collect. This makes it quite challenging to apply the data-driven methods for the FDD of the HVAC system. Thus, a novel data-driven FDD method that only utilizes small fault data collected from a Variable Refrigerant Flow air condition system has been proposed. Under different conditions, the fault and normal data are collected in an enthalpy difference laboratory to create small and imbalanced data. A generative network is developed by combining Wasserstein Generative Adversarial Network with Gradient Penalty and Variational Auto-Encoder. To improve the FDD classifier's accuracy and to train an end-to-end network model using small and imbalanced data, two ensemble classifiers are embedded into the generative network. The dataset includes normal and fault data have been applied to train the modified generative network, and two ensemble classifiers are used to detect and diagnose the fault, respectively. The performance indexes show that the proposed method is much better than the SMOTE-based methods in almost all training groups. Besides, the comparison between the proposed method and generative network with a single classifier indicates that the ensemble classifiers can improve the F1-score of fault detection and the accuracy of fault diagnosis. [ABSTRACT FROM AUTHOR]

Detecting the faults at the incipient stage is important for keeping chiller systems healthy and saving energy and maintenance cost. Traditional principle component analysis (PCA) and support vector data description (SVDD) methods are insensitive to two common faults, condenser fouling (CdF) and refrigerant leakage (RfL). To improve the fault detection performance, this study proposed a PCA-R-SVDD based method. Instead of principle component subspace (PCs), it develops a SVDD model in the residual subspace (Rs) using the PCA modeling residual data. The SVDD based distance based monitoring statistic was used for fault detection. The proposed method shows significant improvement comparing with the traditional methods due to the better fault data distribution and tighter monitoring statistic. It is sensitive to six common faults. At least 50% of the fault data can be correctly detected even at the least severe fault level. Centrifugal chiller experimental data from the ASHRAE Research Project 1043 (RP-1043) was used to evaluate the methods. [ABSTRACT FROM AUTHOR]

Abstract: This paper presents a novel thermoelectric radiant air-conditioning (TE-RAC) system, which employs the thermoelectric modules as radiant panels instead of conventional hydronic panels. The novel radiant system brings a lot of benefits including Freon free, convenient installation, no complex water distribution pipes, quiet and reliable operation, etc. Switching between cooling and heating modes can be easily achieved by reversing the input current. The purpose of this study is to investigate the feasibility and performance of the novel TE-RAC system, including its cooling and heating performances, condensation risk, dynamic thermal response and inertia, and cost-effectiveness. A case study of the TE-RAC system is presented to compare the TE-RAC system with conventional air-conditioning (AC) systems and radiant air-conditioning (RAC) systems in terms of cooling and heating capacities, energy consumption and operation reliability. It is found that the COP of the TE-RAC system could be comparable to conventional RAC systems and AC systems, besides many other benefits in life-cycle performances. [Copyright &y& Elsevier]

• The proposed strategy aims at system- and component-level fault types of VRF systems. • An infrequent fault type of VRF systems is studied. • The FDD strategy is based on hybrid data mining methods. • The results are explained and supported by thermodynamics theory. • The established models can accurately detect and diagnose faults of VRF systems. VRF systems are always vulnerable to kinds of faults. Fault detection and diagnosis research should not only accurately identify these faults, but also be capable of obtaining explanations and support in thermodynamic theory. In this study, a strategy is proposed for four types of VRF system faults, including system-level and component-level. The strategy is based on hybrid data mining methods and analyzes the thermodynamic interpretation of the results at the same time. The first preprocessing step eliminates the effect of noise caused by defrosting action in heating mode. We apply unsupervised principal component analysis for feature extraction to reduce the dimensions of data sets. The correlation between principal components and features are investigated. Supervised Gauss naïve Bayes is used to establish the fault detection model with an accuracy of 98.6%. Besides, infrequent fault type is often difficult to be studied because of lacking sufficient data. Therefore, RUSBoost algorithm is used to solve the unbalanced set problem, and the results show enough competitiveness in the comparison of similar algorithms and online testing. Conclusive remarks confirm the truth that the proposed strategy enjoys high versatility, accuracy, and robustness. [ABSTRACT FROM AUTHOR]

To address the problem that the energy consumption of chiller is affected by multiple factors and lacking of abnormal detection model which can be used for energy consumption of chillers, an integrated model based on Partitioning around medoids(PAM), Support Vector Classification (SVC) and 3-σ rules is proposed in this study. In the detection of abnormal energy consumption of the chillers, the observed data of a Heating Ventilation and Air Conditioning (HVAC) system from Hubei province, China are used to develop and evaluate the integrated model. PAM is applied to divide consumption patterns of the historical data. SVC is applied to recognize consumption patterns of the testing data. 3-σ rules are used for abnormal detection. The results show that PAM divide the energy consumption into four patterns, SVC multi-classifier can effectively recognize energy consumption patterns with an accuracy reaches 98.29%. Compared with the popular model, the integrated learning model applied for the detection of abnormal energy consumption reaches a twice as high as the 40% detection success rate of the 3-σ rules. Moreover, the patterns recognition component can be used to understand the people's energy consumption behavior and the anomaly detection component can be used to regulate the energy usage. [ABSTRACT FROM AUTHOR]

Abstract The refrigerant leakage would occur in a multi-split variable refrigerant flow (VRF) system after years of operation, thus contributing to the inefficient operation and even increased energy consumption. Diagnosing the refrigerant charge amount (RCA) malfunction in time is of great necessity to ensure the normal operation of VRF system and avoid superfluous energy waste. This paper proposes an efficient online weighted k-Nearest-Neighbor model (wk NN) strategy to diagnose the RCA malfunction of the VRF system, which is based on the coupled feature selection method. The minimal-redundancy-maximal-relevance (mRMR) algorithm is first applied to derive the subset of features that have the maximum correlation with the target category and the minimum redundancy among each other. The ReliefF algorithm is utilized to rank variables in descending order. Final variables importance is ascertained on the basis of the average weighs of importance of variables and contributions of variables to the model's classification error rate. Correlation analysis (CA) is implemented to verify the rationality of the selected variables. Finally, a subset of six variables is served as the input variables to establish three kinds of models. Results indicate that the coupled feature selection method outperforms the single feature selection method, and the proposed wk NN model based on the coupled feature selection method is superior to the other two models and achieves desirable diagnostic performance on both experimental and practical data. [ABSTRACT FROM AUTHOR]

Variable refrigerant flow (VRF) systems have gained much attention and been widely used in commercial and residential buildings benefitting from their competitive advantages. However, after long-term operation in a complex environment, various faults may occur in the VRF systems, resulting in failure to meet users comfort requirements and even unnecessary increase in energy consumption. This paper proposes a simulated annealing wrapped generic ensemble fault diagnosis strategy for typical faults of VRF systems, such as refrigerant charge amount (RCA) faults, valve faults, and compressor liquid return (LF) faults. The simulated annealing algorithm based on random forest (SA-RF) is first utilized to perform feature selection process on the three kinds of fault datasets to select the optimal variables that can well characterize the fault states, which can improve the modeling efficiency while reducing the data dimensionality. Then five component learners and the proposed ensemble model based on them are established adopting the optimal variables as input variables. Through visualizing the error evolution and margin of the boosting models built in the first stage of the integration process, it was found that the boosting models can effectively avoid overfitting and most samples are correctly classified with high confidence. By comparing with the five component learners, it is concluded that the boosting strategy in the first stage can improve the diagnostic performance of the models, and the weighted voting integration strategy in the second stage can further improve the diagnostic performance of the model. The final ensemble model can effectively compensate for the deficiencies of each component learners and its diagnostic accuracy for the three fault data sets is as high as 95.37%, 99.36% and 98.3%, respectively, indicating that the model can be applied to diagnose the three types of faults in VRF system at the same time, showing a high versatility. [ABSTRACT FROM AUTHOR]

• We propose a comparison study of basic data-driven fault diagnosis for VRF system. • The performance under single fault and multiple faults is tested and compared. • DNN is preferred for multiple faults, and SVM is selected for single faults. • All method has lower HR for F5 and F6 than the rest in the multiple fault diagnosis. • We came up with suggestions based on comparative analysis for subsequent research. HVAC systems occupy a large part of the building's energy consumption. The development of fault detection and diagnosis (FDD) techniques for HVAC system is becoming increasingly essential for building energy saving. The present paper proposes a comparison study on the basic data-driven methods for variable refrigerant flow (VRF) system fault diagnosis. And five widely used data-driven methods were analyzed, which were decision tree (DT), support vector machines (SVM), clustering (CL), shallow neural networks (SNN), and deep neural networks (DNN). The six common types of fault data in VRF system and three evaluation indexes were used to compare the performance of the proposed five FDD methods in single fault and multiple faults. Results indicate that the single fault diagnosis performance of all methods is better than multiple fault diagnosis. The performances of DNN, SNN, and SVM are better than CART and CL no matter single fault or multiple fault diagnosis. Among them, we believe that the SVM method is preferred for single fault diagnosis and the DNN method is preferred for multiple fault diagnosis. And the performance of method CL is the worst, especially in the case of multiple fault classification, the accuracy is only about 48%. The study results are dedicated to providing a reference for subsequent research in FDD of VRF system. Some important remarks are finally concluded in this paper. [ABSTRACT FROM AUTHOR]

• The potentials of deep reinforcement learning (DRL) techniques are investigated. • Three commonly-used DRL techniques, i.e. A3C, DDPG and RDPG are utilized for building energy consumption forecasting. • The prediction performances of six popular predictive algorithms are studied. • DDPG and RDPG can evidently enhance prediction performance while accounting for more computation time. • A3C doesn't show advantage in forecasting building energy consumption. Reliable and accurate building energy consumption prediction is becoming increasingly pivotal in building energy management. Currently, data-driven approach has shown promising performances and gained lots of research attention due to its efficiency and flexibility. As a combination of reinforcement learning and deep learning, deep reinforcement learning (DRL) techniques are expected to solve nonlinear and complex issues. However, very little is known about DRL techniques in forecasting building energy consumption. Therefore, this paper presents a case study of an office building using three commonly-used DRL techniques to forecast building energy consumption, namely Asynchronous Advantage Actor-Critic (A3C), Deep Deterministic Policy Gradient (DDPG) and Recurrent Deterministic Policy Gradient (RDPG). The objective is to investigate the potential of DRL techniques in building energy consumption prediction field. A comprehensive comparison between DRL models and common supervised models is also provided. The results demonstrate that the proposed DDPG and RDPG models have obvious advantages in forecasting building energy consumption compared to common supervised models, while accounting for more computation time for model training. Their prediction performances measured by mean absolute error (MAE) can be improved by 16%-24% for single-step ahead prediction, and 19%-32% for multi-step ahead prediction. The results also indicate that A3C performs poor prediction accuracy and shows much slower convergence speed than DDPG and RDPG. However, A3C is still the most efficient technique among these three DRL methods. The findings are enlightening and the proposed DRL methodologies can be positively extended to other prediction problems, e.g., wind speed prediction and electricity load prediction. [ABSTRACT FROM AUTHOR]