Neural Network Based on Self-adaptive Differential Evolution for Ultra-Short-Term Power Load Forecasting

Ultra-short-term power load forecasting, which is a complex and nonlinear optimization problem, is an important problem in power system. Self-adaptive Differential Evolution (SaDE), whose control parameter (mutation factor F, crossover factor CR) and mutation strategy are changed gradually and adaptively according to the previous search performance, has been a widely used optimization algorithm among so many improved Differential Evolutions for its strong ability of global numerical optimization and good convergence characteristic. SaDE is employed to optimize a two-layer Neural Network (NN) for the problem of Ultra-short-term power load forecasting. The result shows that SaDE has higher accuracy comparing with Back Propagation (BP) when it is applied in Ultra-short-term power load forecasting.