Real-Time PV Power Smoothing Using Energy Storage System: A Deep Learning Approach

High solar photovoltaic (PV) penetration poses significant ramping issues for utility-operated power plants due to the intermittency of PV output caused by cloud cover. Addressing these fluctuations is pivotal for ensuring grid stability, operational efficiency, and improved power quality. To this end, this paper proposes a deep learning based real-time PV power smoothing framework using battery energy storage. The method leverages K-means clustering to extract worst-case PV profiles from minute-level data per week. By solving an optimization framework over multiple scenarios for diverse PV profiles, battery sizes, and ramp rates, a comprehensive dataset is generated that overcomes the limitations of prior methods, which typically rely on fixed parameters and larger time intervals. Subsequently, a long short-term memory (LSTM)-based deep learning model is trained offline on this dataset to effectively capture temporal dependencies and battery constraints, delivering an optimal, scalable, and computationally efficient solution for dynamic operations. Simulation results demonstrate that the proposed hybrid optimization with an LSTM-based model performs better in terms of accuracy compared to alternative deep learning methods.