The prediction error is relatively large when BP neural network fits non-linear function. To solve this problem, the standard particle swarm algorithm was improved to immunization-based particle swarm optimization (IPSO); and then this algorithm was combined with BP neural network theory to achieve a non-linear function fitting algorithm of BP neural network which was optimized by IPSO algorithm. First, new fitting algorithm determines the BP neural network structure; second, IPSO optimizes the initial weights and thresholds; and finally BP neural network predicts the output of nonlinear function. Numerical experiments show that IPSO improved the fitting capabilities and fitting accuracy and reduced the fitting error of BP neural network.%为解决BP神经网络拟合非线性函数的预测结果误差较大问题,笔者将标准粒子群算法进行改进,形成基于免疫接种的粒子群算法(IPSO);然后将该算法与BP神经网络理论相结合,实现基于IPSO算法优化的BP神经网络非线性函数拟合算法.新的拟合算法首先确定BP神经网络结构,然后用IPSO算法优化初始权值和阈值,最后进行BP神经网络预测.数值实验表明,本文提出的IPSO算法提高了BP神经网络的拟合能力,减小了拟合误差,提高了拟合精度.
展开▼