Improved PSO-BPNN algorithm for SRG modeling

Particle swarm optimization is an excellent algorithm solution for nonlinear, non-differentiable problems. It has strong global search ability, but in the process of looking for the global excellent result, it is easily turn into slow speed and precocious. BP neural network also has strong nonlinear approximation ability, but its nature of gradient descent algorithm determines that it´s easy falling into local optimum and sensitive to the initial values. In order to take the advantages of the two algorithms, an improved particle swarm optimization and BP neural network (IPSO-BPNN) algorithm is proposed. The algorithm is applied to the non-linear modeling of switched reluctance generator (SRG). The efforts suggest that the IPSO-BPNN model has strong generalization ability, it can expression the flux and torque characteristics of SRG perfectly.