Data-Driven Prognostics of LiFePO4 Battery Remaining Useful Life in Residential Microgrids: A Hybrid Approach Integrating Recursive Least Squares and Long Short-Term Memory

Abstract

The growing use of residential photovoltaic systems is increasing the deployment of Lithium Iron Phosphate (LiFePO₄) battery energy storage systems. However, dynamic load demands and frequent cycling accelerate battery degradation. Accurate state-of-health estimation and remaining useful life prediction are critical for operational reliability. This study introduces a hybrid framework that integrates recursive least squares with a long short-term memory network. A first-order Thevenin equivalent circuit model is used to extract physical parameters such as ohmic resistance, polarization resistance, and polarization capacitance. The recursive least squares algorithm, with a forgetting factor of 0.999, adaptively tracks resistance-based State of Health, detecting an increase in ohmic resistance from 26.5 to 37.04 milliohms after 348 cycles. This approach achieves a root-mean-square error of 1.0554% and a coefficient of determination of 0.8354. These parameters are then used to train the long short-term memory network to model degradation trends. The network predicts remaining useful life with a root mean square error of 0.8947% and a coefficient of determination of 0.9553, forecasting lifespans of 511, 953, and 1,310 days for end-of-life thresholds of 80%, 70%, and 60% State of Health, respectively.