Driven by the global energy crisis and the "dual carbon" target, building energy consumption, as a core area of energy consumption, has made energy conservation and consumption reduction a key direction for industry development. Problem: Current building energy consumption monitoring largely relies on traditional manual inspections, which suffers from problems such as slow response, low detection accuracy, and insufficient diagnostic targeting, making it difficult to meet the needs of refined energy-saving management. This paper’s structure and content: First, it constructs a multi-dimensional data acquisition system for building energy consumption, integrating multi-source data such as basic building information, equipment operating parameters, and environmental sensing data; second, it designs an energy consumption anomaly detection model based on data mining algorithms, compares and analyzes the detection performance of three algorithms—Isolation Forest, LSTM, and Support Vector Machine (SVM)—selects the optimal algorithm, and optimizes its parameters. Finally, an energy-saving diagnostic scheme that integrates algorithm detection results with building energy consumption mechanisms is proposed to achieve precise location of anomaly causes and quantitative assessment of energy-saving potential. Experimental results show that the optimized STM algorithm performs best in all performance indicators, with accuracy, recall, and F1 score reaching 96.8%, 95.3%, and 96.0%, respectively. Compared with the second-best isolated forest algorithm, the accuracy is improved by 2.1 percentage points and the recall by 3.5 percentage points, verifying the effectiveness and practicality of the proposed method.
Rui Nie (Thu,) studied this question.