Degradation-Aware Management of Second-Life Lithium-Ion Banks Using Probabilistic SOH Derating in Isolated Microgrids

Second-life lithium-ion banks in isolated microgrids combine heterogeneous, partially degraded cells with harsh tropical operation, yet conventional battery management reacts to voltage limits only once they are reached. This work proposes a degradation-aware battery management system in which the probabilistic state-of-health (SOH) estimate of each cell, obtained by Gaussian process regression, sets a proactive current-derating factor through an uncertainty-penalized effective SOH (μ−2σ), so that estimation uncertainty becomes a control variable. The estimator was validated on four public datasets spanning four chemistry/format combinations, including six LG INR21700-M50T cells with a measured 50-month electric-vehicle-to-storage transition (out-of-fold RMSE 0.28% SOH; 0.69–2.29% on NASA, CALCE, and RWTH benchmarks). An independent 36-cell pack dataset confirms the premise: 14 cells fall below 3.0 V while the pack voltage still reads nominal. At the cycle level, the derating removes every envelope violation of unprotected operation while extracting 3.5% more energy than a safe per-cell cutoff. Over 1500 simulated days of an off-grid photovoltaic microgrid in the Colombian non-interconnected zones, with a bank built from the six measured second-life cells, the strategy preserves 6.5 SOH percentage points at zero loss of supply, rising to 11.8 pp at 35 °C and 13.9 pp for a pre-aged bank, where only the managed bank avoids end of life. The control law is algebraic (∼2 μs per evaluation), computationally compatible with low-cost embedded implementation.