改进的基于遗传优化BP神经网络的电网故障诊断

摘要

BP神经网络具有良好的自学习、自适应和泛化能力，但运算过程中容易陷入局部极小值，同时隐含层节点数的选择也影响着诊断的效果。文中根据经验公式缩小隐含层节点数范围，在小范围里寻找最优的隐含层节点数。根据遗传算法具有全局寻优的特点，用遗传算法优化BP神经网络训练的初始权值阈值，可以避免BP神经网络陷入局部极小值的问题。结合两种方法对电网进行故障诊断，实例分析表明该方法可以准确有效地诊断出电网故障位置，提高电网故障诊断的容错性。%Despite the advantage of self-learning,self-adaptation and generalization ability,BP neural network is prone to be trapped into local minimums during the operation process,and the node numbers in the hidden layer also affect the diagnosis effects. In this paper,according to the empirical formula,the range of node numbers in the hidden layer is narrowed,and the optimal number is searched in a small range. Considering its global optimization,genetic algorithm (GA)is adopted to optimize the initial weight and threshold value of the BP neural network training,which can avoid the local minimums. Using the combination of the above two methods,the grid fault can be diagnosed accurately and ef⁃fectively,and the fault tolerance is improved.