基于遗传算法优化BP神经网络的短期发电量预测

Short-term power generation prediction based on BP neural network optimized by genetic algorithm

针对传统神经网络的网络初始权值和阈值随机给出,训练结果易陷入局部极小值的问题,本文提出一种采用遗传算法(GA)对BP网络的初始权值和初始阈值进行优化的方法.通过模拟生物进化中的自然选择和遗传机理,首先将网络的初始权值和初始阈值表示为染色体基因编码,再利用遗传算子进行组合交叉、变异、选择,产生出新的染色体来完成初始参数优化,形成初始权值和初始阈值,最后经BP网络训练得到最终权值和阈值,建立用于短期发电量预测的网络.实验结果表明,与传统BP神经网络预测方法相比,该预测方法精度更高,为短期电量预测提供了一种新途径.

In order to reduce the influence of unreliable initial weights and thresholds problem,a back-propagation neural network based on genetic algorithm(GA-BP)optimization is proposed which leads to improve the prediction model performance.Genetic mechanism of crossover,mutation and selection and in nature is introduced to represent the network′s initial weight and threshold values as the chromosomes coded gene.According to the principle of survival fitness in the evolution,the genetics combination of crossover and mutation operators helps to produce new chromosomes,which achieves the network parameter optimization.Then the GA-BP prediction model is constructed via BP training to predict the shortterm electricity power generation.Experimental results show that,GA-BP prediction approach superiors to the tradition BP prediction algorithm with better accuracy.The proposed approach may also be utilized in wide other prediction domains.