An enhanced Deep Q-Network approach for load frequency control in hybrid renewable energy system

Load frequency control in hybrid renewable energy systems faces challenges from intermittent sources and reduced grid inertia. Traditional metaheuristic optimization techniques like Particle Swarm Optimization (PSO) and Grey Wolf Optimizer (GWO) suffer from computational complexity and premature convergence to local optima. This research presents an enhanced Deep Q-Network approach for optimizing PID controller parameters in load frequency control. The methodology analyzes a two-area system where Area 1 combines thermal and solar photovoltaic generation, while Area 2 integrates thermal and wind power. The novelty lies in integrating prioritized experience replay with importance sampling, domain-specific reward functions, and augmented state representation enabling model-free learning and real-time adaptation superior to conventional approaches. Experimental validation encompasses six comprehensive test scenarios, with baseline performance benchmarked against PSO-PID, GWO-PID and DDQN-PID controllers. Results demonstrate substantial improvements across key performance metrics. Settling times show reductions of 26.3% to 39.1%, while frequency deviations decrease by 29.0 to 51.3%. Total performance improvements range from 24.6 to 64% across different operating conditions. The controller demonstrates robust performance under challenging operational scenarios. Stability is maintained during parameter variations up to ± 50%, with effective management of load disturbances reaching 25%. The system also adapts effectively to renewable energy source variability across different time periods.