Deep neural network battery charging curve prediction incorporating external information

The utilization of the complete Li-ion battery charge curve provides access to a multitude of critical battery states, which are indispensable for evaluating the safety and dependability of battery-powered devices. Nonetheless, the diminishing health of batteries, coupled with the challenges associated with data collection from battery management systems, presents a substantial hurdle in obtaining complete charging curves. In this study, we introduce an innovative neural network architecture, demands only a segment of the charging curve as input in order to prognosticate the complete constant-current charging curve. Further, this model can be enhanced by optional external data, (e.g., ambient temperature, total battery charge), along with composite inputs, (e.g., battery temperature–voltage–capacity sequence), thereby augmenting its operational performance. This method undergoes rigorous validation across three diverse battery datasets encompassing various data inputs, charging profiles, and temperatures. These assessments reveal an exceptional level of accuracy, with an average error rate falling below 9.35 mAh for 1.1 Ah batteries in the absence of external information, and dipping below 7.37 mAh for 1.1 Ah batteries when incorporating external data. The promising outcomes derived from these validations unequivocally affirm the effectiveness of our proposed model in accurately estimating battery charge curve predictions.