The battery is the primary power source of electrified vehicles (EV). Prediction of battery performances with digital models is essential for both the R&D stage and real-world operation. However, the battery model developed in the R&D stage is not suitable for all real-world conditions, and it will be good if it can be optimized online. This paper proposes an Online Double-layer System Identification (ODSI) scheme to calibrate a battery model for State-of-Health (SoH) prediction with measured data. To determine the unified settings for the base battery model, the ODSI scheme firstly conducts robust optimization in the lower layer based on offline particle swarm optimization (PSO). It then incorporates a deep convolutional neural network (DCNN) to the base model to enable knowledge transfer from offline optimization to online adaption for SoH prediction under different working conditions. By reducing the size of the learning dataset, the study indicates that the proposed scheme has high robustness of uncertainty management. Besides, the ODSI scheme saves the computation resource by avoiding training from scratch.
Keywords Intelligent Energy system; Battery digital modelling; System Identification; Particle swarm optimization; convolutional neural network