基于数据增强SDAE-BiGRU的交流接触器剩余电寿命预测

Remaining Electrical Life Prediction of AC Contactor Based on Data Augmentation SDAE-BiGRU

针对目前交流接触器剩余电寿命存在单一特征预测精度低、未充分考虑开断前后的关联性和忽略了长时间序列特点的问题,该文提出基于数据增强堆叠降噪自动编码器-双向门控循环单元(stacked denoised autoencod-er-bidirection gated recurrent unit,SDAE-BiGRU)的交流接触器剩余电寿命预测方法。首先,通过交流接触器全寿命试验提取特征参量,采用近邻成分分析(neighborhood component analysis,NCA)和斯皮尔曼等级相关系数选择最优特征子集,来有效表征电寿命退化信息。然后,对最优特征子集进行数据增强,充分考虑前后状态的关联性,并利用SDAE对增强后的特征信息进行融合来降低输入维度。最后,将交流接触器剩余电寿命视为长时序问题,通过BiGRU进行时序预测。实例分析表明,该模型比循环神经网络(recurrent neural network,RNN)、长短期记忆网络(long short-term memory,LSTM)、GRU、BiGRU和SDAE-BiGRU模型预测效果好,平均有效精度达到96.68%,有效证明了时序预测模型应用在电器设备剩余寿命预测领域中的可行性。

To tackle the challenges,such as the low prediction accuracy of remaining electric life of AC contactor caused by single feature modeling,the insufficient consideration of correlation before and after opening,and disregarding the characteristics of long time series,we proposed a method for predicting the remaining electrical life of AC contactor using a data augmentation stacked denoised autoencoder-bidirectional gated recurrent unit(SDAE-BiGRU).First,the feature parameters were extracted from the AC contactor full-life test,and an optimal feature subset was selected using neigh-borhood component analysis(NCA)and Spearman rank correlation coefficient to characterize the degradation state of electrical life effectively.Then,the optimal feature subset was augmented to fully consider the correlation between the anterior and posterior states.The SDAE was employed to fuse and reconstruct the original feature information,in which the dimension was reduced and the computing speed was increased.Finally,the remaining electrical life of the AC con-tactor was treated as a long time series and predicted in time series prediction by BiGRU.The case analysis demonstrates that the model has better prediction accuracy than recurrent neural network(RNN),long short-term memory(LSTM),GRU,BiGRU and SDAE-BiGRU models,with an average effective accuracy of 96.68%.The feasibility of time series prediction model applied in the residual life prediction of electrical equipment is verified effectively.