Early prediction of cycle life of supercapacitors based on GA-LSTM

Supercapacitors are increasingly utilized in the new energy automotive industry, favored for their long cycle life, high power density, and environmental sustainability. Accurately predicting the Remaining Useful Life (RUL) of supercapacitors is crucial for timely replacement and avoiding losses associated with unexpected failures. This paper introduces a novel cycle life prediction method for supercapacitors, employing a Genetic Algorithm-Long Short-Term Memory (GA-LSTM) approach. Initially, a Moving Average Filter (MAF) is applied to the degradation data of 20 different supercapacitors for noise reduction. Subsequently, a multi-task learning strategy is implemented, training the LSTM model multiple times to enhance its predictive accuracy. The Genetic Algorithm is then utilized to optimize the LSTM's hyperparameters, significantly refining the prediction model. By inputting only 10% of the test data, the trained model successfully predicts the supercapacitor’s RUL. To validate the method's efficacy, performance tests on eight diverse datasets were conducted. The results indicate that the Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) of this method are maintained within 0.0041 and 0.0032, respectively, showcasing a more precise RUL prediction for supercapacitors compared to traditional LSTM approaches and those without noise reduction.