Deep Reinforcement Learning-Based Cooperative Frequency Controller for Hydropower Dominated Systems

Hydropower dominated systems often experience negative damping and severe power flow oscillations due to improperly tuned governor parameters, even with power system stabilizers. Conventional methods optimize these parameters in a single-machine infinite bus model, neglecting interactions between generations during primary frequency control. This article proposes a deep reinforcement learning based cooperative primary frequency controller using a goal representation heuristic dynamic programming model to address this challenge. The controller minimizes power flow oscillations and frequency deviation through cooperative action among hydrogenerations. Case studies on an IEEE 9 bus system demonstrate that the proposed method achieves superior damping of power oscillations and achieves a much shorter regulation time in the primary frequency control compared to the traditional PID controllers.