一种基于文化粒子群算法的BP网络优化方法

Back Propagation Network Optimization Algorithm Based on Cultural Particle Swarm Algorithm

BP网络良好的逼近特性和泛化能力使其在模式识别、智能控制和系统决策等领域有广泛应用。但网络训练过程中存在的收敛速度慢、容易陷入局部极值等局限性限制了进一步应用。提出一种新的智能优化算法-文化粒子群算法来对BP神经网络的权值和阈值同时进行优化。算法设置了群体空间和信念空间两类独立空间,群体空间采用自适应粒子群算法完成进化,信念空间通过更新函数来进行演化。两类空间的交互通过接受操作和影响操作利用同步式传输方式完成。以Iris分类问题的BP网络模型为仿真实例,对算法的正确性和有效性进行验证。仿真结果表明,改进算法具有较快的收敛速度。

The approximation and generalization characteristics of Back Propagation（BP） network make it successfully apply to the areas of pattern recognition,intelligent control and system decision and so on.The low convergence and easy trapping into local extremum of BP network limits its further application.A new intelligent optimization method,Cultural Particle Swarm Optimization（CPSO）,was proposed to optimize the weight value and the threshold.Two kinds of spaces,population space and belief space,were set in the algorithm.The population space was evolved with adaptive PSO strategy and the belief space was with update function.The interaction between two spaces was by the acceptance operation and impact operation using synchronous transmission mode.The validity and effectiveness of the proposed algorithm was verified by the simulation Iris classification problem.The simulation result shows that the algorithm has faster convergence speed.