Conventional automated machine learning (AutoML) technologies fall short in preprocessing low-quality raw data and adapting to varying indoor and outdoor environments, leading to accuracy reduction in forecasting short-term building energy loads. Moreover, their predictions are not transparent because of their black box nature. Hence, the building field currently lacks an AutoML framework capable of data quality enhancement, environment self-adaptation, and model interpretation. To address this research gap, an improved AutoML-based end-to-end data-driven modeling framework is proposed. Bayesian optimization is applied by this framework to find an optimal data preprocessing process for quality improvement of raw data. It bridges the gap where conventional AutoML technologies cannot automatically handle missing data and outliers. A sliding window-based model retraining strategy is utilized to achieve environment self-adaptation, contributing to the accuracy enhancement of AutoML technologies. Moreover, a local interpretable model-agnostic explanations-based approach is developed to interpret predictions made by the improved framework. It overcomes the poor interpretability of conventional AutoML technologies. The performance of the improved framework in forecasting one-hour ahead cooling loads is evaluated using two-year operational data from a real building. It is discovered that the accuracy of the improved framework increases by 4.24%–8.79% compared with four conventional frameworks for buildings with not only high-quality but also low-quality operational data. Furthermore, it is demonstrated that the developed model interpretation approach can effectively explain the predictions of the improved framework. The improved framework offers a novel perspective on creating accurate and reliable AutoML frameworks tailored to building energy load prediction tasks and other similar tasks.

Transfer learning is an effective method to predict the energy consumption of information-poor buildings by learning transferable knowledge from operational data of information-rich buildings. However, it is not recommended to directly use the operational data without protection due to the risk of leaking occupants' privacy. To address this problem, this study proposes a federated learning-based method to learn transferable knowledge from building operational data without privacy leaking. It trains a transferable federated model based on the operational data from the buildings similar to the target building with limited data. An advanced secure aggregation algorithm is adopted in the training process to ensure that no one can infer private information from the training data. The federated model obtained is evaluated by comparing it with the standalone model without federated learning based on 13 similar office buildings from the Building Data Genome Project. The results show that the federated model outperforms the standalone model concerning the prediction accuracy and training time. On average, the federated model achieves a 25.4% decrease in CV-RMSE when the target building has limited operational data. Even if the target building has no operational data, the federated model still achieves acceptable accuracy (CV-RMSE is 22.2%). Meanwhile, the training time of the federated model is 90% less than that of the standalone model. The research insights can help develop federated learning-based methods for solving the data silos problem in building energy management. The methodology and analysis procedures are reproducible and all codes and data sets are available on Github.

Probabilistic graphical models (PGMs) can effectively deal with the problems of energy consumption and occupancy prediction, fault detection and diagnosis, reliability analysis, and optimization in energy systems. Compared with the black-box models, PGMs show advantages in model interpretability, scalability and reliability. They have great potential to realize the true artificial intelligence in energy systems of the next generation. This paper intends to provide a comprehensive review of the PGM-based approaches published in the last decades. It reveals the advantages, limitations and potential future research directions of the PGM-based approaches for energy systems. Two types of PGMs are summarized in this review, including static models (SPGMs) and dynamic models (DPGMs). SPGMs can conduct probabilistic inference based on incomplete, uncertain or even conflicting information. SPGM-based approaches are proposed to deal with various management tasks in energy systems. They show outstanding performance in fault detection and diagnosis of energy systems. DPGMs can represent a dynamic and stochastic process by describing how its state changes with time. DPGM-based approaches have high accuracy in predicting the energy consumption, occupancy and failures of energy systems. In the future, a unified framework is suggested to fuse the knowledge-driven and data-driven PGMs for achieving better performances. Universal PGM-based approaches are needed that can be adapted to various energy systems. Hybrid algorithms would outperform the basic PGMs by integrating advanced techniques such as deep learning and first-order logic.

Expert systems are effective for anomaly detection in building energy systems. However, it is usually inefficient to establish comprehensive rule bases manually for complex building energy systems. Association rule mining is available to accelerate the establishment of the rule bases due to its powerful capability of discovering rules from numerous data. This paper proposes a real-time abnormal operation pattern detection method towards building energy systems. It can benefit from both expert systems and association rule mining. Association rules are utilized to establish association rule bases of abnormal and normal operation patterns. The established rule bases are then utilized to develop an expert system for real-time detection of abnormal operation patterns. The proposed method is applied to an actual chiller plant for evaluating its performance. Results show that 15 types of known abnormal operation patterns and 11 types of unknown abnormal operation patterns are detected successfully by the proposed method.