Research On The Life Prediction Model Of Contactor Based On Dynamic Neural Network

With the continuous development of new technologies such as smart grid and distributed generation in my country,higher requirements have been put forward for low-voltage electrical appliances that undertake the role of power transmission and distribution.How to strengthen the status information feedback of electrical equipment in the full life cycle has been a hot topic in recent years.AC contactor is an automatic control electrical appliance that can realize frequent breaking at a long distance.Excavating the information of its full life status,judging its degradation status and constructing a product life prediction model,strengthening the understanding of the degradation process of AC contactors,will help to gradually promote the intelligentization of electrical appliances and improve the operational reliability of the power system.The degradation parameters are obtained from the whole life test in this paper.Based on the k-nearest neighbor mutual information(k NN-MI)degenerate feature selection method,a subset of features that are highly correlated with the life process is selected and input to the dynamic neural network to establish a contactor life prediction model.Firstly,starting from the physical structure of the AC contactor,in-depth analysis of its on-off process and failure mechanism,the degradation characteristics used to characterize the life state information of the contactor are excavated.With the aid of test platform for performance evaluation of AC conductor,the CJX2-8011 AC contactor was selected for the full life test.The three-phase contact voltage signal,the three-phase contact current signal,the coil voltage signal and the coil current signal of the tested contactor during the full life status are obtained,and seven kinds of degradation characteristic datas such as contact resistance are calculated from the signals.Secondly,in order to reduce the dimensionality of the input features in the life prediction model,and to retain the information in the degraded feature datas to a certain extent,combined with the large number of degraded features and non-linear characteristics,the k NN-MI method is used for degraded feature selection.The features that are highly correlated with the life process are efficiently selected to form the optimal strong correlation feature subset,which is used as the input parameter of the subsequent prediction model.Finally,the optimal strong correlation feature subset is preprocessed,and then input to the nonlinear autoregressive with external input(NARX)dynamic neural network to construct the life course prediction model of the AC contactor.After comparing the prediction results,the model based on the optimal strong correlation feature subset and the model based on contact resistance have better prediction results.Feature subsets fused with multiple features are more effective than any single feature in predicting.In addition,by comparing the static nonlinear input-output neural network with the same parameters,the NARX dynamic neural network considering the output delay has higher prediction accuracy.