An operating profit-oriented medium-term planning method for renewable-integrated cascaded hydropower

Abstract

For self-scheduling cascaded hydropower (S-CHP) facilities, medium-term planning decisions—such as end-of-day reservoir storage targets—set water usage boundaries for short-term operations, thus directly affecting operating profitability. However, existing medium-term planning methods generally disregard how their decisions will affect short-term operations, which can reduce ultimate profits, especially for S-CHPs integrated with variable renewable energy sources (VRESs). To this end, this paper customizes deep reinforcement learning to develop an operating profit-oriented medium-term planning method for VRES-integrated S-CHPs (VS-CHPs). This method leverages short-term contextual information and trains planning policies based on the operating profits they induce. Moreover, the proposed planning method offers two practical advantages: (i) its planning policies consider both seasonal reservoir storage requirements and the operating profit needs; (ii) it employs a multi-parametric programming strategy to accelerate the computationally intensive training process. Finally, the proposed method is validated on a real-world VS-CHP, demonstrating clear advantages over current practice.