A hybrid particle swarm optimization–back-propagation algorithm for feedforward neural network training

摘要

The particle swarm optimization algorithm was showed to converge rapidly during the initial stages of a global search, but around global optimum, the search process will become very slow. On the contrary, the gradient descending method can achieve faster convergent speed around global optimum, and at the same time, the convergent accuracy can be higher. So in this paper, a hybrid algorithm combining particle swarm optimization (PSO) algorithm with back-propagation (BP) algorithm, also referred to as PSO–BP algorithm, is proposed to train the weights of feedforward neural network (FNN), the hybrid algorithm can make use of not only strong global searching ability of the PSOA, but also strong local searching ability of the BP algorithm. In this paper, a novel selection strategy of the inertial weight is introduced to the PSO algorithm. In the proposed PSO–BP algorithm, we adopt a heuristic way to give a transition from particle swarm search to gradient descending search. In this paper, we also give three kind of encoding strategy of particles, and give the different problem area in which every encoding strategy is used. The experimental results show that the proposed hybrid PSO–BP algorithm is better than the Adaptive Particle swarm optimization algorithm (APSOA) and BP algorithm in convergent speed and convergent accuracy.