Informer-embedded prioritized experience replay-based soft actor-critic for ultra-low frequency oscillation suppression of pumped hydropower storage systems

The occurrence of multiple ultra-low frequency oscillations in the hydro-dominated grid can lead to the instability of pumped hydropower storages and severely constrain energy delivery in the current trend of interconnected grids. The unreasonable parameter settings of governors exacerbate the negative damping phenomenon, which leads to continuous oscillation of pumped hydropower storages. Therefore, enhancing the damping torque of pumped hydropower storages by adapting governor parameters can suppress frequency oscillations. This study proposes an Informer-embedded prioritized experience replay-based soft actor-critic for hydropower unit governor parameter controllers. Prioritized experience replay improves training efficiency and convergence. Informer improves the ability to forecast systematic time series. After training, the proposed method provides optimal parameters for governors under different disturbance operating conditions. Furthermore, compared with conventional optimization and reinforcement learning algorithms, the proposed method provides superior control effects in single-machine and two-area four-machine systems.