Performance enhancement of PI and PIDn controllers through deep reinforcement learning for frequency regulation in renewable-integrated power systems
Erdinç Şahin, Mustafa Şinasi Ayas
Abstract
Maintaining frequency stability has become a major difficulty for contemporary power systems when renewable energy sources – especially photovoltaic (PV) and wind power – are included into them. The intrinsic unpredictability of the supply of renewable energy causes this instability, which emphasises the great requirement of efficient load frequency management to guarantee stability of the power system. This work offers a new hybrid control architecture to improve frequency control in renewable-integrated power systems by means of deep reinforcement learning. Within a two-area power system encompassing three energy units: PV, wind, and thermal, the proposed approach consists in the installation of a deep deterministic policy gradient (DDPG) algorithm as a supplementary control mechanism in combination with optimised PI and PIDn controllers. Comprehensive simulations under many operating conditions – including step load changes, communication time delays, system parameter uncertainty, intermittent renewable energy supply, and random load disturbances – allow one to fully evaluate the effectiveness of the proposed control structure. Achieving up to 75% improvement in integral time-weighted absolute error values while preserving strong performance over several operating situations, the comparison analysis shows that the DDPG-assisted control system much beats conventional controllers. OPAL-RT based real-time simulations evaluate the practical feasibility of the proposed method by verifying its possibility for actual application in contemporary power systems.