融合人工智能与基于物理的建模用于极端高温事件下的长期级联水电调度

The operation of hydropower plants are significantly challenged by extreme heatwave events. This paper integrates artificial intelligence and a physics-based model to introduce an efficient long-term scheduling framework aimed at maximizing hydropower generation during extreme heat events. The water values derived from the proposed long-term scheduling model can be used in developing effective strategies for short-term hydropower scheduling. A physics-based evaporation model (PEM), which captures key land-atmosphere interactions, is developed to account for significant variations in reservoir evaporation rates during extreme heat events. A multivariate long short-term memory (M-LSTM) forecasting model is also utilized to predict the key input parameters required for both the PEM and the long-term scheduling problem. A regression-based machine learning algorithm is also utilized to estimate the hydropower production function, which enables linear integration of the nonlinear and nonconvex behavior of hydropower plant in the mixed-integer linear formulation of the scheduling problem. The proposed model is applied to a case study of eleven cascaded hydropower units located on the Columbia river. The numerical results demonstrate that the proposed long-term scheduling model effectively manages reservoir operations, mitigating the adverse impacts of extreme heat events on hydropower generation and operator profitability.