Multimonth-ahead data-driven remaining useful life prognostics of frequency regulation-battery energy storage systems in cell voltage inconsistency

Lithium-ion batteries (LiBs) have become increasingly popular, which are constructed as energy storage units for various systems including battery energy storage systems (BESSs) and electric (excursion) vehicles, owing to their high energy density, small size and environmental friendliness. In an electric grid that is upgraded as a smart grid with a high penetration of renewable energy resources, BESSs consisting of a large number of LiBs arranged in series and constructed in parallel are being increasingly used to improve the stability, reliability and safety of that grid. However, the battery performance strongly depends on the aging state and degradation tendency, which eventually harm the safety and reliability of the systems containing inconsistent battery cells producing cell voltage inconsistency to deteriorate over time to cause the instability during charging and discharging. Most of recent research in the literature is developed for single individual LiBs to be evaluated with complete charging/discharging cycles for their lifetime, which is not suitable for the estimation of a lot(s) of LiBs connected together to construct a battery module(s)/pack(s) in a (frequency regulation (FR))-BESS(s) and monitored with limited measurement data in the system. FR-BESSs operate their LiBs with incomplete (partial) charging/discharging cycles. Thus, this study proposes a newly developed multimonth-ahead data-driven remaining useful life (RUL) prognostics approach for FR-BESSs in cell voltage inconsistency, which focuses on degradation feature engineering, feature modeling and feature forecasting of major inconsistent LiBs in a monotonic increasing trend of cell voltage inconsistency. In the approach, from targeted inconsistent LiBs, the tendency of cell voltage inconsistency, in a compact monotonic increasing health indicator, by further identified main inconsistent LiBs is forecasted, so that an alert can be issued and a battery module(s) containing the main inconsistent LiBs can be considered for maintenance/replacement for predictive maintenance with more than enough time. Experimental results obtained in this study have demonstrated the effectiveness of the proposed approach; an early detection rate of at least 40% in cell voltage inconsistency exceeding a prespecified cell voltage imbalance threshold to an actual in-service FR-BESS has been obtained.