Prediction of lithium-ion battery remaining useful life based on hybrid data-driven method with optimized parameter

Lithium-ion battery remaining useful life (RUL) is a key parameter on battery management system. An accurate battery RUL can provide reference for replacement and remind safety risk when battery life is nearly end. There are lots of researches on battery RUL prediction. However, until now, the accuracy problem of battery RUL prediction algorithm based on little sample is still not solved. Since a single method is hard to solve the complex problem of battery RUL prediction, this paper introduces an optimized hybrid data-driven method combined with discrete grey model (DGM) and relevance vector machine (RVM). The method takes advantages of trend forecast from DGM and non-linear regression ability of RVM. The kernel parameter of RVM in the algorithm is optimized by an artificial fish swarm algorithm (AFSA). The algorithm also provides confidence interval of the prediction results, which describe the probability distribution of prediction results. Experimental data analysis with NASA battery data set shows the impact of kernel parameter on result error and the improvement on accuracy of battery RUL prediction by the parameter optimization.