Accurate and timely lithium-ion battery discharge capacity prediction is vital for ensuring the safety, reliability, and performance of electric vehicles and grid energy storage systems. However, the aging mechanism of batteries is highly dynamic and complex, posing significant challenges for effective modeling. This paper presents a novel interpretable machine-learning approach that stands out by predicting discharge capacity under variable operating conditions using minimal input descriptors. Three nature-inspired hybrid machine learning algorithms were developed: Quantum-inspired Particle Swarm Optimization-Adaboost (QIPSO-ADB), Harris Hawks Optimization- Extreme Gradient Boosting Machine (HHO-GBM), and Sparrow Search Algorithm-Light Gradient Boosting Machine (SSA-LGBM). We demonstrate that discharge capacity across cycles can be accurately predicted using only temperature and cycle number, thus simplifying model inputs while maintaining high accuracy. Two model series were evaluated: Combo1 (C1), incorporating all input descriptors, and Combo2 (C2), using only temperature and cycle number. All three hybrid models demonstrated strong predictive performance under variable conditions. Notably, the HHO-GBM-C1 model achieved the highest prediction accuracy, with a mean absolute error (MAE) of 0.0816 and a correlation coefficient of 95.2 % during the testing phase. For the reduced-descriptor series, HHO-GBM-C2 achieved a low MAE of 0.1438 and a correlation coefficient of 85.99 %. Validation was performed using multiple samples from eVTOL and MIT public datasets, confirming the robustness and generalizability of the models across both variable and fixed operating conditions. These findings provide strategic insights for optimizing battery performance, contributing significantly to the development of reliable electric vehicles and sustainable energy storage solutions.
Discussion(0)
No comments yet. Be the first to comment.