基于决策因子选择的梯级水库多目标优化调度规则研究

运用极端随机森林挖掘水库调度信息,考虑各水库相关因子与决策目标的相互关系,提出了梯级水库输入数据选择(Cascade-reservoir Input-variables Selection,CIS)方法,引入Gaussian径向基函数建立水库调度规则,并采用PA-DDS多目标算法对水库调度规则参数进行优化,得到了同时考虑供水和发电的多目标优化调度规则集。汉江上游梯级水库的应用结果表明:CIS方法能优选得到较为稳定的决策因子,且能在充分利用已有优选信息的同时避免决策因子选取上的重复和冗余;经梯级多目标优化调度,优选的决策因子能得到比传统决策因子结果分布更优的非劣解集,在以发电为主要目标进行优化调度时,能有效调整汛期和非汛期的水量利用方式,梯级水库年均发电量和供水量分别提高了0.33%和7.84%。

Multi-objective cascade reservoir optimal operation rules based on decision factor selection

A cascade-reservoir input-variables selection (CIS) method considering the relations between in-put variables and decision-making in optimal reservoir operation was proposed using the Extra-Tree model. According to the selection results, the multi-objective reservoir operation rules for the maximization of wa-ter supply and power generation were extracted by Gaussian Radial Basis Functions and optimized by Pare-to-Archived Dynamically Dimensioned Search (PA-DDS) algorithm. With a case study of Hanjiang cascade reservoirs, it is shown that the CIS method can select stable input variables with less redundant informa-tion for the reservoir decision-making, and the input variables selected by CIS not only help to obtain bet-ter distributed non-dominated solutions in multi-objective optimization,but also can control the use of avail-able water in different seasons effectively. The annual mean power generation and water supply increase 0.33% and 7.84% in the operation mainly for power generation.