Cooling load prediction, Wavelet threshold denoising, Convolutional neural network, Long short-term memory, Electrical engineering. Electronics. Nuclear engineering, TK1-9971

Aiming at the problems of insufficient feature extraction, low prediction accuracy and sensitivity to noise in the cooling load prediction, a hybrid model named WTD–CNN–LSTM is proposed in this article, in order to accurately predict the cooling load. It is combined with wavelet threshold denoising (WTD), convolutional neural network (CNN) and long short-term memory network (LSTM). Firstly, WTD is selected to denoise the cooling load sequence. Then, a neural network combined with CNN and LSTM is built, which can extract complex temporal and spatial features. Finally, the performance of the model at different time granularities (TG) and time step lengths (TSL) is studied. Compared with SVR, CNN, RNN, LSTM, GRU, BiLSTM and CNN–LSTM models, the errors of proposed model at different TGs and TSLs are smaller than other models. The experimental results show that the prediction accuracy of the proposed model is higher, and the generalization ability is stronger.