Prediction of Office Building Air Conditioning Cooling Load Based on the CEEMDAN-TPA-LSTMModel

Accurate prediction of air conditioning cooling load is fundamental to optimizing energy-efficient operation of air conditioning systems. In this paper, we propose a combined forecasting model, CEEMDAN-TPA-LSTM, based on Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN), Temporal Attention (TPA), and Long Short-Term Memory networks (LSTM). Initially, cooling load data for the summer three months were simulated using EnergyPlus software, and correlation analysis was performed between cooling load data and meteorological information to select the most relevant external features for cooling load prediction. Subsequently, the CEEMDAN algorithm was employed to decompose the original cooling load data, resulting in several Intrinsic Mode Function (IMF) sequences and residual sequences. Finally, the LSTM model with an attention mechanism was utilized to predict each signal component, and the combination of these components yielded the predicted cooling load data. The ultimate prediction results demonstrate that the CEEMDAN-TPA-LSTM combined forecasting model outperforms other models such as GRU, LSTM, and TPA-LSTM, providing robust support for energy optimization of air conditioning systems.