基于鲸鱼算法改进小波神经网络的GIS局部放电诊断方法研究

Partial Discharge Diagnosis Method of GIS Based on Whale Algorithm and Improved Wavelet Neural Network

针对封闭式气体绝缘开关装置由于生产、运输安装和运行环境等因素,引发的局部放电现象而造成的绝缘故障问题,提出了一种基于改进鲸鱼算法与小波神经网络结合的封闭式气体绝缘开关装置局部放电诊断方法。该方法利用灰度图谱与矩特征实现对局部放电信号的特征提取,基于矩特征值构建小波神经网络的输入样本集;然后使用改进鲸鱼算法对小波神经网络参数进行寻优,以解决神经网络存在的参数敏感问题;最后将优化好且训练完成的小波神经网络应用于绝缘开关装置局部放电诊断中。改进鲸鱼算法引入非线性收敛因子与自适应思想提升了算法的性能,对小波神经网络的超参数有较好的寻优效果。仿真结果表明,相比于通用参数配置的小波神经网络,改进诊断方法诊断精度提升了9.45%。

The insulation failure caused by partial discharge caused by production,transportation,installation and operation environment of enclosed gas insulated switch device is investigated.A cable partial discharge diagnosis method based on improved whale algorithm and wavelet neural network is proposed.The method uses the gray spectrum and moment feature to extract the features of PD signals,and constructs the input sample set of wavelet neural network based on the moment characteristic value.Then an improved whale optimization algorithm is used to optimize the parameters of the wavelet neural network to solve the problem of parameter sensitivity.Finally,the optimized and trained wavelet neural network is applied to partial discharge diagnosis of insulated switchgear.The improved whale optimization algorithm introduces nonlinear convergence factor and adaptive thought to improve the performance of the algorithm,and has a good optimization effect on the hyperparameters of the wavelet neural network.The simulation results show that compared with the wavelet neural network with general parameter configuration,the diagnosis accuracy of the improved method is improved by 9.45%.