基于粒子群优化的神经网络训练算法研究

本文提出了基于连接结构优化的粒子群优化算法(SPSO)用于神经网络训练,该算法在训练神经网络权值的同时优化其连接结构,删除冗余连接,使神经网络获得与模式分类问题匹配的信息处理能力.经SPSO训练的神经网络应用于Iris,Ionosphere以及Breast cancer模式分类问题,能够部分消除冗余分类参数及冗余连接结构对分类性能的影响.与BP算法及遗传算法比较,该算法在提高分类误差精度的同时可加快训练收敛的速度.仿真结果表明,SPSO是有效的神经网络训练算法.

Particle Swarm Optimization Based Algorithm for Neural Network Learning

This paper proposes a structure-improving particle swarm optimization (SPSO) algorithm for training artificial neural network (ANN).The algorithm is successfully applied to pattern classification problems including Iris,ionosphere and breast cancer.By tuning the structure and connection weights of ANN simultaneously,the proposed algorithm generates optimized ANN with problem-matched capacity for processing classification information.By doing this,it also eliminates some ill effects introduced by redundant input features and the corresponding structure of ANN.Compared with BP and GA based training techniques,SPSO can improve the classification accuracy while speeding up the convergence process.Simulation results show that SPSO is a potentially robust learning algorithm and could be extended to real world applications.