SOH prediction for Lithium-Ion batteries by using historical state and future load information with an AM-seq2seq model

Abstract

Accurate state of health (SOH) prediction is essential for lithium-ion batteries from the perspectives of safety and reliability. However, most existing data-driven methods only take the historical state information of a battery (e.g., its historical SOHs) as input. Considering that the future SOH degradation trends of lithium-ion batteries are highly affected by future loads, a new SOH prediction method that takes both historical state information and future load information as inputs is developed for batteries operating under dynamic loading conditions. To integrate these two types of information, an attention-based multisource sequence-to-sequence (AM-seq2seq) model consisting of two encoders and one decoder is built. Within this structure, advanced attention layers are employed to learn the global dependencies between the target SOH predictions and the model inputs. For the purpose of the validation, two case studies are conducted under different discharge currents and different ambient temperatures, respectively. It is shown that the proposed AM-seq2seq model is capable to provide accurate long-term SOH predictions for all of the cases with different future loads and beginnings of prediction (BOPs). Moreover, it also exhibits great robustness against various historical state input and future load input lengths. As a result, the proposed AM-seq2seq model is feasible for adaptively predicting the SOHs of batteries under different future loads with limited historical SOHs.