State of health prediction of supercapacitors using multi-trend learning of NARX neural network

Abstract

Accurate and reliable supercapacitor’s state of health (SOH) prediction is necessary to avoid premature systems failure and ensure safe operation. Supercapacitors’ lifecycle and degradation trends vary significantly due to their wide range of operating conditions. Machine learning (ML) techniques have shown an excellent performance in prediction of SOH for many devices. However, conventional methods optimize ML’s hyperparameters for SOH prediction using a single degradation trend, leading to a relatively poor performance of the ML algorithm on a different aging trend. In this study, we proposed a novel multi-trend learning approach to predict the SOH of the supercapacitors by a non-dominated search genetic algorithm III (NSGAIII) and a non-linear autoregressive network with exogenous inputs (NARX) recurrent neural network. Seven supercapacitors were cycled with various charge-discharge currents and depths of discharge voltage to generate multiple degradation trends. The multi-trend learning approach greatly enhanced the robustness and efficiency of the NARX neural network. The proposed NSGAIII-NARX model shows a 49.14% decrease in average root mean square error compared to the conventional technique that uses a single trend for learning. The proposed method also shows good performance against other benchmarks like long short term memory recurrent neural network and artificial neural network. The same network also predicts the SOH of different supercapacitors’ dataset with average RMSE and MAPE of 0.059 and 0.655%, respectively