Model-Based vs Data-Driven Battery State Estimation in PV Systems under Different Operating Conditions

Abstract

The precise determination of battery state of charge (SOC), state of health (SOH) and state of power (SOP) is required to ensure the safe and reliable operation of photovoltaic (PV)-battery energy storage systems that act under extremely dynamic charging and discharging conditions. In the given paper, a narrow comparative study of three classical estimation methods, including open-circuit-voltage (OCV)-based estimation, linear Kalman filter (KF) and extended Kalman filter (EKF) will be conducted and evaluated with the help of a common 24-hour ultra-challenging PV and load current profile as the reflection of a realistic microgrid functioning. The profile consists of high-frequency bidirectional variations on current, half-cycle reversible charge, constant power discharge substrate, and effects of temperature variation.All estimators are checked by means of a high-fidelity electro-thermal battery model which includes capacity degradation and inner resistance increasing with aging and temperature variations. The accuracy of the SOC tracking, SOH estimation error, and SOP prediction potentialities of every method are evaluated.The findings indicate that the OCV-based approach has serious limitations when subjected to dynamic loading, which results in almost linear SOC estimates and excessively conservative SOP predictions because of the effect of polarization of voltages. The KF enhances the performance of SOC tracking over that of OCV estimation but has observable deviations at elevated SOC and towards endof-discharge due to the linearization assumptions of the model. The EKF also improves the accuracy of the estimation by considering the battery nonlinearity which results in better SOC tracking and enhanced SOP prediction at the cost of higher estimation sensitivity to fast changing parameters.The paper emphasizes that though the OCV tools are only applicable when the system is in a steady-state, model-based filtering techniques, especially the EKF, offer reliable real-time predictive estimation of a PV-battery system under extreme operating conditions, and therefore are better placed when it comes to the practical use of energy management.