基于AM算法优化LSTM的微型断路器剩余电寿命预测  

Residual electrical life prediction of miniature circuit breaker based on AM algorithm optimized LSTM

作　　者：杨钰炜 迟长春[1] 李兴家 YANG Yuwei;CHI Changchun;LI Xingjia(School of Electrical Engineering,Shanghai Dianji University,Shanghai 201306,China)

机构地区：[1]上海电机学院电气学院,上海201306

出　　处：《上海电机学院学报》2024年第6期344-350,共7页Journal of Shanghai Dianji University

摘　　要：针对微型断路器剩余电寿命预测中特征选取单一、预测精度较低的问题,提出了基于注意力机制(AM)算法优化长短期记忆神经网络(LSTM)的预测模型。首先,通过搭建微型断路器电寿命试验平台提取特征参量;然后,采用皮尔逊相关系数(PCC)从众多特征参量中选择最优特征子集,从而有效表征电寿命退化过程;最后,将微型断路器的剩余电寿命作为预测标签,通过AM-LSTM预测模型对微型断路器的剩余电寿命进行预测。试验结果表明:该模型比GRU、LSTM模型预测效果好,有效精度达到87.78%,能够满足实际工程的需要。Aiming to address the issues of single-feature selection and low prediction accuracy in the residual life prediction of circuit breakers,a prediction model based on an attention mechanism(AM)algorithm is proposed to optimize the long short-term memory(LSTM)network.First,characteristic parameters are extracted using a circuit breaker electrical life test platform.Next,the pearson correlation coefficient(PCC)is employed to select the optimal feature subset from multiple parameters,effectively capturing the degradation process of the electrical life.Finally,the residual electrical life of the miniature circuit breaker is used as the prediction label,and the remaining life is predicted using the AM-LSTM model.Experimental results demonstrate that the proposed model outperforms both GRU and LSTM models,achieving an accuracy of 87.78%,which meets the practical engineering requirements.

关 键 词：微型断路器 特征选择 剩余电寿命 长短期记忆神经网络 

分 类 号：TM561[电气工程—电器]