Development of a hybrid approach for real-time SOC estimation in renewable energy systems

Nayeemuddin Mohammed, Ala A. Hussein, Hiren Mewada

Abstract Lithium-ion batteries rely on the state of charge (SOC) as a key indicator of remaining energy and overall battery condition, making it essential for efficient operation in energy storage systems. However, SOC cannot be measured directly, and its estimation is often affected by nonlinear battery behavior, temperature variations, and model uncertainties, which limit prediction accuracy. In this study, a gated recurrent unit-convolutional neural network (GRU-CNN) and slime mold algorithm-optimized support vector regression (SMA-SVR) were employed to determine the SOC of lithium-ion batteries in an innovative manner. The suggested model uses the measurements of time, voltage, current, and surface temperature. Thus, overcoming the drawbacks of the old approaches to SOC regression, which frequently rely on past data and are prone to compounding errors. The experimental findings indicate that the GRU-CNN model outperforms the SMA-SVR model. It achieved an R 2 of 0.999 in both training and testing sets, close to 1, and mean absolute error values of 0.0006 and 0.0007. In addition, the SMA-SVR had higher prediction errors, with an R 2 score of 0.965. The strong correlation among SOC, temperature, and voltage also confirms the GRU-CNN model’s strength. It has been shown that this method is not only very effective at predicting the SOC of new lithium-ion cells but also highly accurate for old cells. This is an effective and scalable solution to battery management systems in both electric cars and grid-scale energy storage. Finally, the study opens the prospect of improving battery performance and reliability with highly innovative predictive analytics.