Novel deep learning model developed for battery lifespan prediction

Correct prediction of lithium battery lifespan is essential for the right functioning {of electrical} tools. Nevertheless, predicting battery lifespan precisely is difficult as a result of nonlinearity of capability degradation and the uncertainty of working circumstances.

Lately, Prof. Chen Zhongwei and Assoc. Prof. Mao Zhiyu from the Dalian Institute of Chemical Physics of the Chinese language Academy of Sciences, in collaboration with Prof. Feng Jiangtao from Xi’an Jiaotong University, designed a novel deep studying mannequin, the twin stream-vision transformer with the environment friendly self-attention mechanism (DS-ViT-ESA), to foretell the present cycle life (CCL) and remaining helpful life (RUL) of the goal battery. The research was revealed in IEEE Transactions on Transportation Electrification.

The researchers developed the deep learning model utilizing a small quantity of charging cycle information. This mannequin employed a imaginative and prescient transformer construction with a dual-stream framework and an environment friendly self-attention mechanism to seize and combine hidden options throughout a number of time scales.

The mannequin was capable of precisely predict the battery’s CCL and RUL. With simply 15 charging cycle information factors, it achieved RUL and CLL prediction errors of solely 5.40% and 4.64%, respectively. Furthermore, the mannequin maintained low prediction errors even when examined on charging methods not included within the coaching dataset, demonstrating its zero-shot generalization functionality.

This battery lifespan prediction mannequin was additionally a vital part of the first-generation Battery Digital Mind, referred to as PBSRD Digit. The system built-in with this algorithm has considerably improved accuracy. At the moment, the Battery Digital Mind system serves because the core vitality administration system for large-scale industrial storage and electric vehicles, with the potential to be deployed on each cloud servers and client-side embedded gadgets.

“The battery lifespan prediction model effectively balances prediction accuracy with computational cost, thereby increasing the application value of the Battery Digital Brain for lifespan estimation,” stated Prof. Chen. “We plan to further optimize the model using techniques such as model distillation and pruning, aiming to enhance robustness and resource utilization of the system.”