Preventive Prognosis of Aluminum Electrolytic Capacitors Based on SAPSO-LSTM

Aluminum electrolytic capacitor is one of the most important components in power converters, that is also mainly responsible for breakdowns. The preventive prognosis of the capacitors is helpful to maintain the health of the whole system, avoiding structural failures and increasing the maintenance cost of the equipment. In order to explore the degradation characteristics of aluminum electrolytic capacitors, a high temperature degradation experiment of aluminum electrolytic capacitor is designed to analyze the degenerate properties. Long-short term memory (LSTM) network is proposed to predict the capacitance, aiming at solving the problem of nonlinear in the process of capacitor degradation. Then the simulated annealing particle swarm optimization (SAPSO) is applied on the optimization of hyper-parameters. In the end, experimental result shows that the prediction algorithm based on SAPSO-LSTM has the lower error compared with RNN, LSTM, GRU and PSO-LSTM. The RMSE and time of convergence of the SAPSO-LSTM are 1.3883 and 1487. 11s, which are 11.915% and 7.289% less than that of PSO-LSTM.