| Abstract: |
Power quality degradation in modern distribution networks has escalated with the proliferation of nonlinear loads, renewable energy sources, and distributed generation systems. Static Synchronous Compensators (STATCOMs), grounded in voltage-source converter (VSC) technology, represent a pivotal FACTS device for reactive power compensation and harmonic mitigation. However, conventional control strategies primarily proportional-integral (PI) regulators exhibit inherent limitations in adapting to rapidly varying load conditions and complex grid topologies. This paper proposes a novel Reinforcement Learning (RL)-based adaptive control framework for STATCOM, employing the Deep Deterministic Policy Gradient (DDPG) algorithm augmented with Prioritized Experience Replay (PER), to achieve superior power quality improvement in a 33 kV distribution network. The proposed RL agent perceives a twelve-dimensional state space encompassing bus voltages, reactive power flows, harmonic spectra, and load variations, and outputs continuous modulation indices and phase angle commands for the STATCOM's VSC. Simulation results obtained in MATLAB/Simulink on a modified IEEE 33-bus test system demonstrate that the RL-STATCOM reduces Total Harmonic Distortion (THD) to 2.18%, achieves a power factor of 0.99, and mitigates voltage sag by 96.7%, with a dynamic response time of 9.7 ms. These outcomes represent statistically significant improvements over PI, fuzzy logic, and model predictive control (MPC) baselines. |
