多尺度多模型的梯级水库优化调度决策支持系统

A Decision Support System for Optimal Scheduling of Cascade Reservoirs with Multiple Time Scales and Multiple Models

梯级水库群优化调度因来水预报不确定性强、水库群优化变量维数高、决策评估目标多等特点,一直是学术界和工程界的研究热点.本文分析了不同时间尺度的水文预报模型与优化调度模型的基本特征与数据需求,建立了多尺度多模型的梯级水库优化调度决策支持系统.该决策支持系统将数值天气预报的有效预报期(15d)作为短期调度模块与中长期调度模块的时间分界.短期调度模块采用数值天气预报降水量和数字流域模型预测未来15d的日径流过程,使用预报结果进行短期的梯级水库优化调度;中长期调度模块采用基于自回归和遥相关的组合预报模型,预报未来1年的月径流过程,用于中长期的优化调度.本文采用面向服务的软件架构(SOA),将年、月和日尺度的预报和调度模型集成,构建决策支持系统,可以实现多时间尺度的水情滚动预报和梯级水库优化调度.本系统在黄河上游干流梯级水库群和大通河流域梯级水库进行试验应用,结果表明该系统能够快速集成分析相关水情数据,开展梯级水库优化调度演算,及时为决策者提供清晰有效的决策辅助信息.

The optimal scheduling of cascade reservoirs still remains a challenging topic in academic and engineering fields,due to the strong uncertainty of inflow forecasting,high dimensionality of optimization variables,and multiple operation objectives. This paper analyzed the features and data requirements of inflow forecasting models and scheduling optimization of cascade reservoirs in different time scales,and established a decision support system comprised of multiple models for different time scales. The decision support system was built with the short-term( < 15 d,daily) and mid-to long-term( < 1 a,monthly) scheduling modules. The short-term module uses precipitation data from numerical weather predictions( NWPs) and Digital Yellow River Integrated Model to predict daily runoff in the next 15 days,which is limited by the effective forecast lead time of NWPs. Then the daily inflow is utilized for the short-term optimal scheduling model. The mid-to long-term module uses the combined forecasting model based on auto-regression and teleconnection method to predict the monthly runoff in the coming year,which is then utilized for the mid-and long-term optimal scheduling model. Finally,the decision support system was built by integrating the different models using the Service Oriented Architecture( SOA) technology,realizing the rolling forecasts at different time scales. The system was applied to two reservoir cascades in the upper Yellow River and the Datong River,respectively. The results showed that this system is effective for the integration and analysis of relevant hydrological data,efficient for inflow forecasting and scheduling optimization,and can finally facilitate the management of cascade reservoirs with clear and effective information in time.