Unlocking the full-range regulation performance potential of hydropower systems: A condition-aware and parameter-adaptive cooperative optimization method

Conventional hydropower and pumped storage hydropower are important energy-support resources in renewable-dominated power systems. Under diverse operating states, particularly generating conditions, hydropower systems are more prone to deviate from their desirable operating region during frequent regulation, thereby leading to substantial financial penalties under grid assessment rules. To address this issue, this paper proposes an adaptive optimization method based on autoregressive conditional neural processes and meta-learning (AR-CNP + ML) to enhance the regulation performance of hydropower units and ensure compliance with grid assessment requirements. A condition-aware mapping among operating parameters, control parameters, and regulation performance enables rapid adaptive parameter generation after offline training. First, a regulation model is established using a comprehensive performance curve refined by radial basis function-Gaussian process regression (RBF-GPR). Then, a regulation performance evaluation index system is constructed in accordance with the grid assessment rules applied in Northwest China. Furthermore, the effects of key operating conditions, including initial water head, initial load, and disturbance type, are systematically analyzed, and a simulation dataset covering multiple operating scenarios is generated. Finally, AR-CNP + ML is employed to enable operating-condition-aware performance prediction and online self-tuning of PID parameters. The results show that the proposed method significantly improves regulation performance. Under frequency step disturbances, t 90% and t stable are reduced from 45.99 s and 76.41 s to 11.57 s and 38.63 s, respectively, while under power step disturbances, T p decreases from 49.69 s to 23.03 s. The proposed framework provides a scalable data-driven paradigm for adaptive control of hydropower units under high-frequency regulation demands.