The regulation speed and performance of automatic generation control (AGC) in hydropower stations to a certain extent determine the power quality of the grid. AGC is a key grid connected auxiliary service for large hydropower stations such as Monkey Rock, which can enhance the automatic control level of 40MW and above power stations, significantly reduce the burden of duty, and lower the risk of power misoperation. However, the current AGC of hydropower stations has problems such as low efficiency, high risk, and incomplete coverage in manual testing. Based on this, this article designs an adaptive control algorithm for the AGC testing system of hydropower stations. The algorithm first uses the Maximum Relevance Minimum Redundancy (mRMR) algorithm to extract features from the operating data of the unit, obtaining a set of relevant variables that affect the AGC control effect to improve modeling efficiency. On this basis, an AGC strategy deep learning (DL) model is constructed using long short-term memory networks (LSTM) as neurons to predict the regulatory ability of the AGC testing system. Finally, the algorithm is combined with reinforcement learning (RL) to achieve closed-loop adaptive optimization control of the testing process. Simulation experiments show that the algorithm can accurately evaluate the dynamic response capability and control performance of AGC testing systems.
Qiu et al. (Sun,) studied this question.