Abstract

The increasing complexity and decentralization of modern power systems demand advanced decision-making techniques to ensure efficient and reliable operations. Deep Reinforcement Learning (DRL) offers a promising approach for optimizing smart grid performance by enabling adaptive, real-time control in dynamic environments. This paper presents a DRL-based framework for optimal energy dispatch, load balancing, and renewable integration in smart grids. The proposed system models the grid as a Markov Decision Process, using actor–critic algorithms to learn optimal control strategies from simulated and real-world operational data. Experimental results on a test grid with high renewable penetration demonstrate improvements in energy efficiency, peak load reduction, and system stability compared to traditional optimization methods. The findings indicate that DRL can significantly enhance grid resilience, reduce operational costs, and support the transition toward sustainable, low-carbon energy systems in the context of Industry 4.0.