The assessment and prediction method for VCB contact endurance research based on LM-BP neural network

The LM-BP neural network prediction model for contact system electrical endurance of vacuum circuit-breakers is established, which based on the electrical endurance curve and the advanced breaking current weighted cumulative method (BCWC). The influencing factors that affect the prediction of VCBs electrical wear were researched. Then the LM-BP algorithm iterative process of contact electrical wear is significantly analyzed, and the detailed process of establishing the LM-BP electrical endurance model is presented also. Furthermore, the technique and the cautions of selecting hidden layer neurons and confirming of hidden layer node number during the process of establishing the LM-BP model are specially analyzed. At last, the neural network training and prediction has been carried out based on the obtained samples. The relative error of desired output and actual output is analyzed and the corresponding results are presented also. The results show that the LM-BP neural network of VCBs contact system, discussed in the paper, can accurately predict the residual contact wear. And the method provided in the paper presents not only a new idea to the on-line condition monitoring of the electrical endurance of VCBs contact system, but also valuable for engineering application of on-line monitoring and forecasting.