为提高变压器故障诊断准确率,提出了基于混合采样技术的数据处理方法和改进鲸鱼算法优化 BP 神经网络的故障诊断模型.首先通过Tomek links和 SMOTE 对原始数据进行去噪处理及少数类数据样本合成,然后采用Cubic map混沌映射及自适应权重调整策略改进鲸鱼算法(WOA),利用改进后的鲸鱼算法优化BP神经网络参数;最后,利用混合采样前后的数据对传统 BP 神经网络、WOA-BP、CAWOA-BP 三种模型进行对比仿真试验.结果表明,使用混合采样后的数据训练模型会使模型故障诊断准确率提高,且CAWOA-BP模型的表现优于传统BP神经网络和 WOA-BP模型.
Transformer Fault Diagnosis Based on Hybrid Sampling and CAWOA-BP
In order to improve the accuracy of transformer fault diagnosis,a data processing method based on hybrid sampling technology and an optimized BP neural network fault diagnosis model by improved whale algorithm are pro-posed.Firstly,the original data are denoising and synthesis of minority data samples are performed by Tomek links and SMOTE.Then the whale algorithm(WOA)is improved by Cubic map chaotic mapping and adaptive weight adjustment strategy,and the BP neural network parameters are optimized by the improved whale algorithm.Finally,the traditional BP neural network,WOA-BP and CAWOA-BP models are compared and simulated by using the data training before and after mixed sampling.The results show that training the model using mixed sampling data can improve the accuracy of fault diagnosis model,and the CAWOA-BP model outperforms the traditional BP neural network and WOA-BP model.
万方数据
维普
