基于深度强化学习与演化计算的风-水-火混合增强智能调度  

Wind-hydro-thermal hybrid-augmented intelligent scheduling based on deep reinforcement learning and evolutionary computation

作　　者：李远征[1] 郝国凯 杨东升[2] 赵勇[1] 周杰韩 曾志刚[1] LI YuanZheng;HAO GuoKai;YANG DongSheng;ZHAO Yong;ZHOU JieHan;ZENG ZhiGang(School of Artificial Inteigence and Automation,Huazhong University of Science and Technology Wuhan 430074,China;College of Information Science and Engineering,Northeast University,Shenyang I10006,China;Faculty of Information Technology and Electrical Engineering,University of Oulu,Oulu FI-90014,Finland)

机构地区：[1]华中科技大学人工智能与自动化学院,武汉430074 [2]东北大学信息科学与工程学院,沈阳110006 [3]奥卢大学信息技术与电子工程学院,芬兰奥卢FI-90014

出　　处：《中国科学：技术科学》2023年第7期1097-1113,共17页Scientia Sinica(Technologica)

基　　金：国家电网总部科技项目(编号:5108-202315041A-1-1-ZN)资助。

摘　　要：火电、水电和风电是我国电力工业系统的三大能源主体,根据风-水-火发电互补特性,建立联合优化调度模型对于降低电力系统运行成本以及促进新能源消纳具有重要意义.然而梯级水电站间的时空耦合性、风电的不确定性以及风-水-火多能源相互关联的复杂约束使得联合调度模型求解较为困难.因此,本文提出了一种基于深度强化学习(deep reinforcement learning,DRL)与演化计算的混合增强智能优化框架.该框架首先利用深度强化学习与风-水-火联合调度模型进行交互,并根据交互数据对联合调度模型复杂规律进行持续学习,优化自身控制策略,提高智能体泛化能力.此后,在解决实际调度问题时,为进一步提升算法的个性化能力,利用演化计算算法(particle swarm optimization,PSO)在经过训练的DRL上进一步优化调度方案,实现风-水-火联合调度的快速决策.算例分析表明,所提出的混合增强智能优化框架求解速度快、寻优能力强,提升了DRL优化性能的鲁棒性,提高了风-水-火系统运行的经济性及风电消纳能力.Thermal,hydro,and wind power are the primary energy sources of the power industry system in China.Thus,building a wind-hydrothermal complementary scheduling mode is important for reducing system operating costs and improving renewable energy accommodation.However,the temporal and spatial coupling of cascade hydropower stations,the uncertainty of wind speed,and complex constraints among multiple energy sources make the model difficult to solve.Thus,this paper proposed a hybrid-augmented intelligent optimization framework based on deep reinforcement learning(DRL)and evolutionary computation.The first step of the framework is using DRL to interact with the wind-hydro-thermal complementary scheduling model and learning the regularity of the model based on the interaction data,optimizing the control policy according to a trial-and-error strategy,and improving the generalization ability.Then,to enhance the algorithm's personalization ability when addressing industrial problems,particle swarm optimization is used to optimize the schedule of the wind-hydro-thermal system based on trained DRL.This step achieves rapid decision-making for wind-hydro-thermal complementary scheduling.Case studies show that the proposed optimization framework has outstanding responding speed and searching performance.Meanwhile,it improves the economical and renewable energy accommodation of the wind-hydro-thermal complementary scheduling model.

关 键 词：混合增强智能 深度强化学习 演化计算 风-水-火联合调度 滚动优化 

分 类 号：TM73[电气工程—电力系统及自动化] TP18[自动化与计算机技术—控制理论与控制工程]