Heat load prediction for heat supply system based on RBF neural network and time series crossover

In order to improve the energy-saving efficiency, a novel heat load prediction method based on radial basis function neural network (RBF NN) and time series crossover is proposed according to the characteristics of heat supply process. The dimension of the input vector in the RBF NN model is established with autocorrelation method. Then the horizontal and vertical prediction models are constructed using the RBF neural network, respectively. And those two prediction models are combined to produce the crossover prediction model whose crossover weight coefficients are calculated through the least-squares method. The comparison of simulation results shows that the accuracy of crossover prediction is superior to horizontal and vertical predictions. In addition, the speed of crossover prediction based on RBF neural network is proved faster than the one with back propagation neural network (BPNN).