Assessment of battery energy storage uses in microgrid operation under varied demand uncertainties through a machine learning based stochastic optimization approach

Abstract The sustainability of Microgrid (MG) relies on improving reliability through effective energy management strategies. However, the presence of volatile Renewable Energy Sources (RES) and uncertain energy demand pose serious challenges. Although integrating Battery Energy Storage Systems (BES) is a promising solution, the utilization of BES under uncertain operating conditions remains underexplored. Given this, the present study aims to investigate BES operational behavior while minimizing the overall MG operation cost under varied operating conditions. Also, it evaluates the impact of prediction accuracy on overall MG operation costs and DER operation. Here, a Machine Learning (ML) -based Gaussian Process Regression (GPR) is utilized for predicting the volatile RES power output while the uncertain energy demand is estimated with Monte Carlo Simulation (MCS).The problem is formulated as a cost-minimization problem and solved using the physics-based Artificial Electric Field Algorithm (AEFA) at different MG operating conditions. Results show that MG operation cost varies approximately by 5%, with demand variations around the mean, and BES energy contribution is 1.3 times higher for the case with additional thermal load. Moreover, analysis confirms that the BES depreciation cost is faster with a lower initial State of Charge (SOC), and the proposed strategy ensures a higher SOC by day’s end. Additionally, GPR’s applicability is validated through comparison with the Gated Recurrent Unit (GRU), revealing lower BES depreciation and MG operation costs due to higher prediction accuracy. Further, a comparative analysis with 2m Point Estimate Method is done to highlight the superiority and robustness of the proposed approach.