Anomaly detection of energy consumption in building lighting and socket systems can effectively improve energy efficiency. It holds significant importance for the implementation of building energy optimization measures and the realization of energy-saving management and control. Since the energy consumption of building lighting and plug load systems is heavily influenced by the random behavior of building occupants, and given the challenges posed by noisy time-series data and difficulty in feature extraction, this study proposed an unsupervised anomaly detection method that integrates operating condition classification with deep learning, aiming to enhance the accuracy and robustness of energy consumption anomaly identification. First, the decision tree algorithm was employed to classify the energy data based on attributes such as working days vs. non-working days and working hours vs. non-working hours. Then, for each identified condition, a long short-term memory autoencoder (LSTM-AE) model was constructed to detect anomalies. This model learns to reconstruct normal data and calculates the reconstruction error. By setting differentiated thresholds, it enables energy consumption anomaly detection under unlabeled data conditions. Using 578 days of hourly lighting and socket energy consumption data from an office building located in a hot-summer and warm-winter region, the study conducted model training and hyperparameter optimization experiments. Results indicate that the number of iterations, the number of neurons, and the activation function have significant effects on the model’s performance. Energy data during working days demonstrate greater stability than those on non-working days, resulting in higher detection accuracy. The proposed method achieves average precision, recall, and F₁ of 91.23%, 90.87%, and 90.80%, respectively, across four typical operating conditions, demonstrating its effectiveness in detecting energy anomalies in building lighting and socket systems.

Short-term power load multi-step forecasting for commercial buildings plays a pivotal role in urban orderly power consumption and virtual power plant scheduling. The power load time series in commercial buildings is characterized by strong stochasticity, non-stationarity, and nonlinearity, and traditional iterative multi-step power load forecasting strategy suffers from error accumulation effects that degrade prediction accuracy, a short-term power load multi-step forecasting method based on Frequency Enhanced Channel Attention Mechanism (FECAM)-Sparrow Search Algorithm (SSA)-Informer is proposed. Based on the time-domain features output by the Informer encoder, the method uses FECAM to adaptively model the frequency dependence between feature channels, and further extractings the frequency-domain features of multi-dimensional input sequences. The decoder then integrates both time-frequency domain information to directly generate future multi-step load sequences. Furthermore, due to the lack of theoretical basis for the improved Informer hyperparameter settings, the SSA is used to optimize model hyperparameters such as learning rate, batch size, fully connected dimensions, and dropout rate. Experimental validation using annual load data from a commercial building in Guangzhou demonstrates that, compared with other deep learning models, the proposed model significantly improved prediction accuracy across varying forecast horizons (steps of 48, 96, 288, 480 and 672), exhibiting superior performance in short-term power load multi-step forecasting.

The research on the anomaly detection of the operation state of complex energy consumption system is of great significance to the safe, stable, efficient and energy-saving operation of buildings. However, due to the influence of many factors, such as the variety of external environmental factors of buildings, the uncertainty of insider behavior, the complexity of equipment operation data and so on, the detection of abnormal operation state often meets some difficulties in information feature extraction, abnormal operation state definition and so on. Based on the classification of operation parameters and by using the quantitative description of uncertainty by information entropy and the adaptability of unsupervised learning, this paper proposed a secondary clustering anomaly detection method for the operation state of complex energy consumption system based on information entropy and by integrating K-means and self-organizing mapping model. According to the actual monitoring data of the central air-conditioning system of a large office building in the hot summer and warm winter area, the external parameters were clustered for the first time through K-means, and the steps such as dividing the primary working conditions, calculating the information entropy of the operating parameters of the subsystem under each working condition, and determining the abnormal operating state by secondary clustering of self-organizing mapping model were used to realize the abnormal detection of the operating state of the complex energy consumption system of the building. The results show that the average intra class anomaly detection accuracy of the method proposed in this study is 97.45% under the operation condition of single machine and 96.70% under the operation condition of two machines. In addition, this study further discussed and analyzed the relationship between the number of classes and the relevant indicators of in class anomaly rate under different working conditions. It concludes that the in class anomaly detection rate of single and dual hosts increases with the decrease of the total number of operating states in their class. This study provides a new systematic idea and method for the abnormal operation state detection of complex energy consumption system under the background of building energy big data.

The optimization of the number of central air-conditioning cooling source units and their operating parameters is a collaborative optimization problem involving both discrete and continuous variables, which poses challenges for classical reinforcement learning algorithms. To address this problem, this paper proposed an energy-saving optimization control strategy for central air-conditioning cooling source systems based on a combination of the options-critic and actor-critic frameworks. Firstly, a hierarchical actor-critic (H-AC) algorithm was utilized to hierarchically optimize the number of units and operating parameters, with both the high-level and low-level models sharing a Q-network to evaluate state values, thereby addressing optimization challenges across multiple time scales. Secondly, the H-AC algorithm was improved in terms of agent architecture, policy, and network update mechanisms to accelerate the convergence of the agent. Finally, the proposed method was validated on the cooling source system of a research building located in a hot summer and warm winter region, using a TRNSYS simulation platform for experiments. The results demonstrate that, under conditions where the average indoor comfort time proportion is increased by 14.08, 11.23, 29.70 and 9.07 percentage points, respectively, the system energy consumption based on the improved H-AC algorithm is reduced by 32.28%, 28.55%, 28.64%, and 11.53% compared to four classical DRL algorithms. Although the system energy consumption of the improved H-AC algorithm is 0.27% higher than that of the options-critic framework, it achieves a more stable learning process and increases the average indoor comfort time proportion by 4.8%. This approach offers effective technical solutions for energy-saving optimization of central air-conditioning cold source systems in various building types, contributing to the achievement of buildings’ dual-carbon goals.