Load forecasting of large CCHP system based on improved BP neural network

Gu Zhaoxiong★, Huang Zhijian, Xie Ming, Lin Mengya, Qi Jiaye and Gao Wenzhong

2021.01.20

BP neural network has problems such as slow learning speed, dimensionality disaster, easy to fall into local convergence, and unable to guarantee the best global convergence. In order to solve above problems, applies Spearman rank correlation coefficient firstly to analyse the main influencing factors of cooling load, so as to determine the input parameters of a dynamic cooling load prediction model, and then develops an improved BP neural network prediction model of the combined genetic algorithm. Finally, uses BP and GA-BP neural network models to predict the cooling load of a large district CCHP system located in Shanghai respectively. The results show that Spearman rank correlation coefficient analysis can shorten the training time, avoid dimensionality disaster and improve the prediction accuracy. Compared with the prediction results of BP neural network, GA-BP neural network avoids local convergence and obviously improves the prediction accuracy of the model.