Machine learning in energy storage optimization for carbon neutrality: A review

The global drive toward carbon neutrality by 2050 has amplified the demand for optimized energy storage systems to address the intermittency of renewable energy sources. This review examines machine learning (ML) applications in energy storage optimization across electrochemical, mechanical, thermal, and emerging storage technologies, focusing on bridging renewable variability and grid stability through intelligent strategies. Deep learning models demonstrate higher performance in grid stability prediction, with artificial neural network (ANN) models achieving 97.27% accuracy compared to traditional support vector machine at 92.77% and XGBoost at 92.98%. In thermal energy storage applications, MLP models achieve exceptional R2 values of 0.9994 for liquid fraction predictions and RMSE of 7.18 MW for heat demand forecasting. Digital twins demonstrate substantial operational improvements, achieving up to 38% enhancement in comprehensive performance indices and 5% operational cost reductions through intelligent thermal management and predictive analytics. ML-optimized systems further deliver significant environmental benefits, including 41.5% carbon emission reductions and 42% operational cost savings across different system configurations. For photovoltaic integration, ML optimization enhances effective capacity utilization by 24.8% annually through optimized dispatch scheduling. Deep neural networks maintain 97% of optimal profit benchmarks while reducing computational time from over 30 h to 1.78 s in thermal energy storage optimization. These findings highlight ML's transformative potential in achieving carbon neutrality through intelligent energy management, enhanced system efficiency, and accelerated renewable energy integration. Future research should focus on uncertainty quantification, physics-informed neural networks, and standardized data frameworks.