A modified genetic algorithm for developing dynamic neural network model and its Application in Daily Short-Term Load Forecasting

In order to solve the problem with being easily trapped in a local optimum of back propagation neural network (BPNN) and the premature convergence based on standard genetic algorithm (SGA), a dynamic and adaptive model which combines the modified genetic algorithm (MGA) with BPNN is proposed in this paper. By introducing modified genetic operators and dynamic mutation probability measure, the MGA-BP model can be used to configure the structure of BPNN in a rational way and achieve excellent performance in terms of relative error rates. For illustration, Application example on Daily Short-Term Load Forecasting (STLF) are given to show the merits of the presented model, which is compared with the method of BP and SGA. Empirical results show that our proposed method with comparatively dynamic structure has the higher prediction accuracy and the better performance in convergence rate.