基于强化学习的非常规水资源优化配置模型  

Optimal allocation model of unconventional water resources based on reinforcement learning

作　　者：伍鑫 陈植欣 温庆博[1] 王忠静[1] 胡黎明[1] WU Xin;CHEN Zhixin;WEN Qingbo;WANG Zhongjing;HU Liming(State Key Laboratory of Hydroscience and Engineering,Tsinghua University,Beijing 100084;CITIC Construction Co.Ltd,Beijing 100027)

机构地区：[1]清华大学水沙科学与水利水电工程国家重点实验室,北京100084 [2]中信建设有限责任公司,北京100027

出　　处：《水力发电学报》2021年第7期23-31,共9页Journal of Hydroelectric Engineering

基　　金：国家重点研发计划课题(2017YFC0403500);水沙科学与水利水电工程国家重点实验室科研课题(2020-KY-01)。

摘　　要：目前我国非常规水资源开发利用程度较低,水资源配置的系统不确定性较大,水资源优化配置模型有待提升。本文将强化学习方法引入水资源配置模型,以水资源的经济效益为优化目标,建立了基于强化学习方法的水资源优化配置模型,采用Python语言编制计算分析软件;以北京市水资源优化配置作为案例分析,将非常规水资源纳入配置体系,得到了研究区水资源优化配置方案及收益区间。结果表明,本文建立的基于强化学习的非常规水资源优化配置模型能合理分析不同来水情况总体经济效益的收益范围,与区间两阶段随机规划模型相比,能在一定程度上提高预期经济收益;不同来水年的收益情况表明非常规水资源对缓解北京市水资源短缺有重要作用。Development and utilization degree of unconventional water resources in China are relatively low at present,and water resources allocation is complicated with a variety of uncertainties.Aimed at this problem,this paper applies the method of reinforcement learning to water resources allocation,and develops an optimal allocation model of Python coding with the objective function of maximizing the economic benefit.This model is applied in a case study of Beijing,taking the unconventional water resources into account,and focusing on comparison of different allocation schemes and their interval scales of economic profits.Results show it predicts the overall profits under different water inflow conditions satisfactorily,giving larger values than those predicted using the two-stage stochastic programming method.And utilizing the unconventional water resources would play a significant role in alleviating the city’s water shortage.

关 键 词：水资源配置 非常规水资源 强化学习 优化模型 经济效益 

分 类 号：TV213.9[水利工程—水文学及水资源]