Electrical Load Forecasting Using Echo State Network and Optimizing by PSO Algorithm

ESN load forecasting model has high stability, and is able to learn fast and not easy to fall into local optimum, compared with standard recurrent neural network. In the process of constructing the typical ESN model, the choice of parameters is always empirical or random. The forecasting performance of ESN was analyzed on the basis of its key parameters. While the dynamic reserve pool has black box characteristics and in order to overcome the lack of principles of the design of constructing ESN prediction model, this paper adopts PSO algorithm to optimize four key parameters of dynamic pool model in the process of constructing ESN electricity load forecast model, and uses two PSO boundary treatment strategies in the tests with analyzing different strategies on the pros and cons. PSO-ESN model is compared with SVM, BP and typical ESN model. Simulation results demonstrate that the proposed PSO-ESN model can obtain more accurate forecasting results than the others.