A Novel Hybrid Evolutionary Algorithm Based on PSO and AFSA for Feedforward Neural Network Training

Abstract

In recent years, the multilayer feedforward neural network (FNN) has been received considerable attention and have been extensively used in many fields. Levenberg-Marquardt back-propagation (LMBP) algorithm as an FNN training method has some limitations associated with overfitting, local optimum problems and slow convergence rate. In order to overcome the limitations, some people proposed particle swarm optimization (PSO) as an evolutionary algorithm to train the FNN. But PSO has disadvantages such as low precision, slow convergence in the later stage of the evolution, and parameter selection problems. In this paper, a novel hybrid evolutionary algorithm based on AFSA and PSO, also referred to as AFSA-PSO-parallel-hybrid evolutionary (APPHE) algorithm, has been used in FNN training. Compared to FNN trained by LMBP algorithm, FNN training by the novel hybrid evolutionary algorithm show satisfactory performance, converges quickly towards the optimal position, convergent accuracy, high stability and can avoid overfitting in some extent. FNN training by the novel method has been testified by using in Iris data classification and the results are much more accurate and stable than by Levenberg-Marquardt back-propagation algorithm.