A Novel Hybrid PSO-BP Algorithm for Neural Network Training

In order to search better solution in the high dimension space, the novel hybrid PSO-BP algorithm which combines the PSO mechanism with the Levenberg-Marquardt algorithm or the conjugate gradient algorithm is proposed. The main idea employs BP algorithm with numeric technology to find the local optimum, and takes the weights and biases trained as particles, and harnesses swarm motion to search the optimum. Finally, the hybrid algorithm selects some good particles from the local optimum set to predict the new samples. Simulation results show that the hybrid PSO-BP algorithm is better than the basic BP algorithm and the adaptive PSO algorithm in the stability, correct recognition rate and training time.