Comparison and analysis of different machine learning models for predicting air-conditioning cooling loads in comprehensive large public buildings across various time scales

Large-scale comprehensive public buildings feature complex functionalities and drastic fluctuations in air-conditioning cooling loads, posing significant challenges to accurate prediction. Cooling load forecasting at different temporal scales enables the fulfillment of varied energy-saving control requirements, such as real-time control, optimization of HVAC operational efficiency, and formulation of cold storage/release schedules. However, most existing studies focus on single-function buildings and single-time-scale cooling load forecasting. To explore accurate multi-time-scale cooling load prediction for large-scale comprehensive public buildings, this study employs six defferent neural networks, namely Long Short-Term Memory (LSTM), Gate Recurrent Unit (GRU), Recurrent Neural Network (RNN), Backpropagation algorithm (BP), Temporal Convolutional Network (TCN) and Extended Long Short-Term Memory (XLSTM), to develop and compare 18 machine learning models for daily, hourly, and ten-minute time scales. A comprehensive analysis of model performance is conducted based on prediction accuracy and computational efficiency. The results demonstrate that different models exhibit varying prediction performances across distinct time scales, particularly in terms of prediction accuracy and runtime. Meanwhile, an improved prediction optimization method is proposed, which reduces the Mean Absolute Percentage Error (MAPE) by 36.25 % and increases the coefficient of determination (R²) by 14.62 % compared to non-optimized approaches. Furthermore, the research presented in this paper fills a gap in the field of air-conditioning cooling load prediction for comprehensive large public buildings using various neural networks across different time scales. This is of great significance for real-time control of air-conditioning cooling, efficient utilization and storage of cooling capacity, and the realization of energy-efficient machine rooms.