基于多元感知信息融合的电抗器故障声纹诊断研究

Research on Voiceprint Diagnosis of Reactor Faults Based on Multivariate Perception Information Fusion

为提高电抗器故障诊断准确率,提出一种基于多元感知信息融合的诊断方法。方法通过采用PSO(Particle SwarmOptimization)算法优化LSTM(LongShort-TermMemory)网络的神经元个数与dropout值改进LSTM网络,并利用改进LSTM网络对电抗器故障声纹进行分类识别,实现了电抗器故障诊断。仿真结果表明,采用PSO算法改进LSTM网络,可提高LSTM网络的收敛速度和分类识别精度;采用PSO算法改进LSTM网络模型对电抗器故障声纹进行分类识别,可有效诊断额定预紧力松动、单边夹件松动、双边夹件松动等不同类型的电抗器故障,且具有较高的准确率,平均准确率约为98%;相较于SVM、CNN、BP模型以及标准LSTM网络模型,所提PSO算法改进的LSTM网络模型对电抗器故障声纹诊断的准确率更高,具有明显优势。

To improve the accuracy of reactor fault diagnosis,a diagnostic method based on multi perception information fusion is proposed.The method improves the LSTM(Long Short Term Memory)network by using the PSO(Particle Swarm Optimization)algorithm to optimize the number of neurons and dropout values,and uses the improved LSTM network to classify and recognize the voiceprint of reactor faults,achieving reactor fault diagnosis.The simulation results show that using PSO algorithm to improve LSTM network can improve the convergence speed and classification recognition accuracy of LSTM network;The PSO algorithm is used to improve the LSTM network model for classifying and identifying reactor fault voiceprints,which can effectively diagnose different types of reactor faults such as rated pre tension looseness,unilateral clamp looseness,and bilateral clamp looseness,and has a high accuracy rate,with an average accuracy of about 98%;Compared to SVM,CNN,BP models,and standard LSTM network models,the LSTM network model improved by the PSO algorithm has a higher accuracy in voiceprint diagnosis of reac tor faults and significant advantages.