Cost-Effective Lebesgue Sampling Long Short-Term Memory Networks for Lithium-Ion Batteries Diagnosis and Prognosis

Diagnosis and prognosis are critical for the safe operation of lithium-ion batteries in avoiding serious damage and economic loss caused by power failure, fire, or explosion. With capabilities in learning the long-term dependencies in battery capacity degradation, long short-term memory (LSTM) network is widely used in the fault diagnosis and prognosis (FDP) of lithium-ion batteries. However, FDP based on traditional LSTM network in the Riemann sampling framework has heavy demands on computation and training. To address the problem, this article proposes a cost-effective FDP based on Lebesgue Sampling LSTM (LS-LSTM) network. In the proposed method, a Lebesgue time model is developed to describe the fault growth process, which has lower complexity compared with the traditional fault state space model. Based on the Lebesgue time model, the LS-LSTM network is able to implement the FDP of lithium-ion batteries in a unified "as needed" philosophy. The Monte Carlo method is employed to manage the uncertainties of measurement and prognosis. Offline experiments and online experiments on battery capacity degradation are presented to demonstrate the effectiveness of the proposed method. Experimental results and comparison studies show that the proposed method has good performance in terms of accuracy and computation cost and can be applied to many different applications.