<p>This paper presents a novel reinforcement learning-based optimization framework utilizing a Deep Q-Network (DQN) for decision-making in dynamic power system environments. Unlike traditional methods such as Mixed-Integer Linear Programming (MILP), which require predefined models and struggle with real-time adaptability, the proposed approach leverages reinforcement learning to continuously learn and optimize actions based on evolving grid conditions. By integrating Q-learning with deep neural networks, the framework enables adaptive decision-making that efficiently handles uncertainties in renewable energy generation, load fluctuations, and grid disturbances. The model is evaluated across multiple case studies, demonstrating its superior flexibility and robustness compared to conventional optimization techniques. Results indicate that DQN-based optimization improves real-time adaptability, reduces computational overhead, and enhances resilience against unexpected disruptions, making it well-suited for modern power system operations. These findings highlight the potential of reinforcement learning in advancing intelligent, self-learning energy management strategies for resilient and cost-effective grid operations.</p>