Intelligent Demand Response Resource Trading Using Deep Reinforcement Learning  

作　　者：Yufan Zhang Qian Ai Zhaoyu Li 

机构地区：[1]Key Laboratory of Control of Power Transmission and Conversion,Ministry of Education,Department of Electrical Engineering,Shanghai Jiao Tong University,Shanghai 200240,China.

出　　处：《CSEE Journal of Power and Energy Systems》2024年第6期2621-2630,共10页中国电机工程学会电力与能源系统学报（英文）

基　　金：supported by the National Key R&D Program of China(2021YFB2401203).

摘　　要：With the liberalization of the retail market,customers can sell their demand response(DR)resources to the distribution company(Disco)through the DR aggregator(DRA).In this paper,an intelligent DR resource trading framework between Disco and DRA is proposed by exploiting the benefits of deep reinforcement learning(DRL).The hierarchical decision process of the two players is modeled as a Stackelberg game.In the game,Disco is the leader who determines the retail price while DRA is the follower who responds to it.To protect their privacy,a dueling deep Q-network(dueling DQN)is then constructed to model the bi-level Stackelberg game,such that the lower-level problem doesn’t need to reveal its detailed model to the upperlevel.In the learning process,the uncertainties from the DRA’s baseline load and wind power are considered.In order to boost the robustness against the estimation error,the baseline load is discretized into symbols before being used as the input states of the dueling DQN.And to mitigate the uncertainty of wind power,the scenario-based method is introduced when designing the reward.We demonstrate that the proposed dueling DQNbased method has good performance and is more robust against uncertainties.

关 键 词：Demand response economic interaction reinforcement learning stackelberg game UNCERTAINTY 

分 类 号：TK01[动力工程及工程热物理] TP39[自动化与计算机技术—计算机应用技术]