Adaptive Algorithmic Methods in Intelligent Control Systems: A Comparative Study of Reinforcement Learning and Model Predictive Control for Power System Optimization

Abstract

Adaptive algorithmic methods are essential for intelligent control systems in modern power grids and electrotechnical complexes, enabling real-time optimization, fault tolerance, and operational stability under varying conditions. This paper presents a comparative study of two adaptive control approaches-Reinforcement Learning (RL) based control and Model Predictive Control (MPC)-for power system optimization and voltage regulation. Using a simulated power distribution network with 118 buses, 3 renewable energy sources, and 5 energy storage systems, we evaluate both approaches on voltage regulation accuracy, transient response, power loss reduction, and computational efficiency. Experimental results demonstrate that RL-based adaptive control achieves 97.3% voltage regulation accuracy and 23.4% reduction in power losses, outperforming traditional MPC which achieves 94.1% accuracy and 17.8% loss reduction. RL-based control adapts faster to load variations (38.2 ms vs 142.7 ms settling time) but requires 12.5× more training time. Hybrid RL-MPC approach achieves 96.5% accuracy with 18.9% loss reduction, offering a balanced solution. Practical recommendations for selecting adaptive control methods based on application requirements and computational constraints are provided.