A novel multi-model operation mechanism based on Gaussian classification and data decomposition for air-conditioning cooling load prediction in complex public buildings

Accurate prediction of air conditioning cooling load is crucial for the rapid response of air conditioning equipment, timely adjustment of indoor comfort conditions, and improvement of the operational efficiency of air conditioning units. In complex public buildings, due to their diverse and unique functions, the start and stop times of air conditioning equipment are inconsistent, leading to sudden changes in cooling load during certain periods. This poses significant challenges for precise cooling load prediction and real-time control of machine rooms. To promote the realization of high-efficiency machine rooms and accurately predict the cooling load of complex public buildings, a novel approach was first proposed. This method combines Gaussian classification based on building functions, data decomposition, and a multi-model operation mechanism integrating neural networks. The results demonstrate that this approach not only significantly improves the overall prediction accuracy of cooling load but also effectively reduces large prediction errors caused by sudden changes in cooling load during certain periods. The optimized model achieved mean absolute percentage error values of approximately 6.78% and 8.38% on different datasets, representing reductions of 63.61% and 45.48% respectively compared to the unoptimized neural network model. Additionally, by comparing data from different years and using different composite models, the effectiveness and accuracy of this method were further validated. This research holds significant importance for achieving real-time forecasted of cooling load in complex public buildings and enhancing the efficient regulation of air conditioning systems.