State-Space Refined Kalman Filter Model for Battery State-of-Health Estimation in Energy Storage Power Stations

Abstract

Numerous issues, including partial cycling, current pulses, thermal gradients, and module anomalies, affect battery state-of-health estimation in grid energy storage systems. In this research, develop a state-space refined Kalman filter model that interprets resistance increase and capacity loss as slowly changing states instead of post-processed indications. Temperature-normalized ageing drift, innovation-governed covariance inflation, and a station-level consistency requirement are combined with a second-order equivalent-circuit representation. Twelve integrated sensors gathered temperature data and over 214 operation days of 18,640 charge-discharge fragments from nine-six lithium iron phosphate cells arranged into eight 52 V modules. The state-of-health mean absolute errors were lowered to 1.18%, 1.91%, and 2.78%, respectively, by using a dual Kalman filter and a recursive least squares baseline in place of the traditional extended Kalman filter. The variance in capacity estimation for operation at 0.2C to 1.0C has decreased by 31.7%, and the 95th percentile error has decreased to 2.64%. Additionally, the estimator can identify the high-temperature rack's accelerated degradation 17 days ahead of the baseline alert rule. The findings suggest that, without the use of black-box features, explicit state-space modification can improve the model's accuracy, stability, and engineering interpretability.