Skip to Main Content
Forecast of annual electricity demand is very important for the market settlement and transmission pricing of power system. Therefore, a forecasting model combing back propagation (BP) neural network and three sub-swarms particle swarm optimization (THSPSO) is proposed. Some important economical factors of the year to be forecasted, such as the gross product, the population, the price index, and so on, are considered in the forecast model. On the other hand, annual electricity demands are considered as a time series. Firstly, the weights and bias of the neural network if globally optimized based on THSPSO, which has a stronger diversification than the basic PSO. Secondly, the network is trained by BP algorithm with the obtained values from THSPSO as the initial values. The case study of Liaoning Province of China indicates that the network can be trained quickly by the hybrid algorithm of THSPSO and BP, and that annual electricity demand can be forecasted by this network with high precision.