Novel cell screening and prognosing based on neurocomputing-based multiday-ahead time-series forecasting for predictive maintenance of battery modules in frequency regulation-energy storage systems

Energy storage systems (ESSs) by a large number of lithium-ion batteries arranged in series and/or in parallel for their energy storage unit have increasingly become important. This is because, for example, an electrical grid upgraded as a smart grid with a widespread use of renewables and electric vehicles needs to be stabilized under grid requirements for grid safety, stability and reliability. In a frequency regulation (FR)-ESS, (severe) cell voltage imbalance associated with battery performance strongly depending on the aging state and degradation tendency needs to be prevented such that potential safety hazards can be precluded. This research presents a novel battery cell screening and prognosing methodology based on neurocomputing-based multiday-ahead time-series forecasting for predictive maintenance (PdM) of battery modules constituting battery racks of an FR-ESS. Where, battery cell screening can more precisely quantify relative deterioration, relating to cell voltage imbalance, of lithium-ion battery cells, allowing the traceability in terms of cell abnormalities to be quantified and visualized for battery cell outliers inside battery modules. Moreover, from targeted battery cell outliers, battery cell prognosing can predict the tendency of cell voltage inconsistency produced by the main inconsistent battery cells identified from the battery cell outliers so that an alert may be issued and the main inconsistent cells may be considered for maintenance/replacement in PdM. The presented methodology is a preliminary implementation, which has been experimentally validated by an on-site, in-service FR-ESS. Its effectiveness has been confirmed, as reported in this research.