基于粒子群优化的神经网络预测模型

A Neural Network Prediction Model Based on Particle Swarm Optimization

BP学习算法多采用梯度下降法调整权值,针对其易陷入局部极小、收敛速度慢和易引起振荡的固有缺陷,提出了一种改进粒子群神经网络算法.其基本思想是:首先采用改进粒子群优化算法反复优化BP神经网络模型的权值参数组合,再用BP算法对得到的网络参数进一步精确优化,最后用得到精确的最优参数组合进行预测.实验结果表明,该算法在股指预测中的预测性能明显提高.

According to the inherent defects of BP learning algorithm, we proposed an improved particle swarm neural network algorithm to improve the prediction ability of the BP neural network. We optimal the combination weights of BP neural network model parameters with the improved particle swarm optimization algorithm, then use BP algorithm to get the further accurate optimization of network parameters. The optimal parameter combination can be used to make prediction. Experimental results demonstrate the efficacy of the algorithm in the stock index prediction.