Joint state-of-health and remaining-useful-life prediction based on multi-level long short-term memory model prognostic framework considering cell voltage inconsistency reflected health indicators

Abstract

Battery-health prognostics for the state-of-health (SOH) and remaining-useful-life (RUL) are necessary to ensure the safety and reliability of system operation. However, the aging of an actual battery pack is related to internal electrochemical reactions in the constituent cells and the inter-cell inconsistency effect, which cannot be measured directly. Thus, reasonable feature selection and an effective prognostic framework are indispensable for achieving feasible and accurate results in actual systems. This paper proposes a prognostic framework for the joint prediction of the SOH and RUL of a battery pack. The proposed method utilizes indirect health indicators (HIs) during the charging stage. Moreover, Pearson correlation analysis and incremental capacity analysis are used to recognize the inconsistency effect and optimize the sampling interval of the HIs that can be adapted to the battery pack. A multi-level long short-term memory (LSTM) model is designed to jointly predict the SOH and RUL for a battery pack. The first-level LSTM model is applied to short-term SOH estimation using the optimized HIs. For the RUL prediction, the second-level LSTM is established iteratively to extend the degradation pattern from the current step to end of life using the SOH estimation result and previous output. The robustness of the extracted features and predictive capability of the multi-level LSTM model are demonstrated by accelerated aging tests for series battery packs. The experimental results confirm that the proposed prognostic framework can provide robust, reliable, and accurate SOH and RUL predictions.