改进人工鱼群算法优化小波神经网络的变压器故障诊断

Transformer fault diagnosis based on wavelet neural network with improved artificial fish-swarm algorithm

针对油浸式变压器故障类型的复杂难辨,结合油中气体分析法,提出一种基于改进人工鱼群算法优化小波神经网络的故障诊断模型。基于经典三层小波神经网络,采用粒子化的人工鱼群算法对小波神经网络输入和输出层的权值、小波神经元的伸缩和平移系数进行修正,通过引入动态反向学习策略实时优化人工鱼分布,迭代后半程采用基于柯西分布的自适应人工鱼视野范围提高算法精度。结果表明,该改进鱼群算法优化的小波神经网络相比标准粒子群算法优化小波神经网络和标准鱼群算法优化小波神经网络,诊断速度更快,准确率更高。

Combining dissolved gas analysis,a wavelet neural network diagnosis model based on the improved artificial fish-swarm algorithm was proposed to accommodate complicated fault types of the transformer.Based on the standardized three-layer wavelet neural network structure,the input and output weighting coefficients,the stretching and translation coefficients of wavelet elements were modified by the particle artificial fish-swarm algorithm.A dynamic opposition-based learning strategy was introduced to optimize the distribution of artificial fish,and the adaptive vision of artificial fish based on Cauchy distribution was used to improve the accuracy in the latter half-progress.Experimental results showed that the proposed method applied to the transformer diagnosis well for its faster and more accurate search results than those obtained by wavelet neural network diagnosis optimized by standard particle swarm algorithm and wavelet neural network diagnosis optimized by standard artificial fish-swarm algorithm.