Battery Capacity Trajectory Prediction with Multi-output Gaussian Process

Abstract

Battery health monitoring is critical for the safe management and sustainable maintenance of electrical equipment. The uncertainty of battery usage scenarios and the huge cost of aging experiments make it a challenge to construct accurate and general-purpose battery lifetime prediction models. In this paper, based on the multi-output Gaussian process (MOGP) with transfer learning, the battery aging data under different working conditions can be applied to accurately predict the capacity trajectory. The performance of the two dominant MOGP models, symmetric and asymmetric, in battery capacity prediction, is thoroughly analyzed, and compared with other machine learning algorithms. Two different types of batteries with different working conditions are used to verify the performance of the models. Considering the performance of the model for different aging degrees, the battery degradation is divided into two stages: early stage and late stage. The MOGPs are proved to be the best performer. The asymmetric MOGP is suitable for the rapid prediction of batteries in the late aging stage, while the symmetrical MOGP can accurately predict the change of capacity trajectory and has high robustness to batteries at different aging stages. The average mean absolute errors (MAEs) of the symmetrical MOGP with three outputs for the early prediction of different batteries are only 0.027Ah and 0.017Ah, respectively.