Abstract: Accurate prediction of the residual electrical life of AC contactors is of great significance in ensuring the safety of the electrical system. Through accelerated electrical lifetime testing of AC contactors, completing electrical lifespan data of test samples were acquired.Wavelet transform method was employed to determine the starting and ending moments of arcing in the breaking phase of an AC contactor,and the arcing time, arcing energy and average arcing power in the time domain of the arcing signal were calculated accordingly. Extracting root mean square, standard deviation, skewness, and form factor feature parameters in the statistical domain, and then the fusion of the time domain and statistical domain feature parameters during arcing is selected by the method of mutual information, and the voltage standard deviation,voltage root mean square, average arcing power, and the arcing energy feature parameters are selected and used as inputs to the gate recurrent unit(GRU), temporal convolutional network(TCN), improved TCN prediction models to complete the prediction task. By comparing the fitting degree of the prediction results of the three models with the actual electric lifetime values of AC contactors and the evaluation indexes of the prediction results, it is obtained that the prediction results of the improved TCN model have a higher fitting degree with the actual electric life values, higher prediction accuracy, and the best performance among the three models.

    Key words: AC contactor; electrical lifetime; wavelet transform; improved TCN model; mutual information