Research On Incentive-Based Demand Response Scheme Based On Multi-Armed Bandit

The report of the 20 th National Congress of the Communist Party of China emphasizes: "We should actively and steadily promote carbon peak and carbon neutrality,deeply promote the energy revolution,and accelerate the planning and construction of a new energy system." This points out the direction for the high-quality leapfrog development of energy and electricity in China in the new era.In recent years,with the continuous development of productivity in China,the electricity consumption of residential users has been increasing,and the potential for demand-side adjustable resources is enormous.Flexible aggregation and adjustment of demand-side resources through demand response has become an important means of optimizing the allocation of power resources.Incentive demand response is a means of flexibly dispatching demand side energy through subsidy signals,which has great potential for maintaining supply and demand balance in power systems and improving economic benefits.In the incentive demand response for residential users,when there is a shortage of electricity,power operators choose to reduce the potential of residential users and provide subsidies to encourage them to reduce their electricity.However,from the perspective of power operators,facing the unknown and uncertain electricity consumption behavior of residential users,how to dynamically set subsidy prices and how to identify and select residential users with high potential for power reduction are key challenges.This thesis discusses the application scenario of incentive demand response for residential users,and the specific research content is as follows:1)Analyzing and comparing existing research schemes for dynamic pricing and consumer selection in demand response.Currently,research schemes for dynamically setting subsidy prices in the field mainly include traditional optimization based methods,game theory based methods,and reinforcement learning based methods.Research schemes for participating in demand response for resident user identification and selection in the field mainly include clustering based methods and multi-armed bandit based methods.This thesis conducts in-depth research on the different schemes mentioned above,excavates their core ideas,analyzes their advantages and disadvantages,and summarizes and contemplates future research directions.2)Considering that the behavior of residential users in reducing electricity consumption in demand response is rational,that is,residential users respond to the incentive signals released by power operators based on their discomfort caused by electricity consumption reduction,psychological state,and environmental awareness.An incentive demand response algorithm based on online learning is proposed,which uses the multi-armed bandit machine framework to learn online the potential of residential users to reduce electricity consumption,Establish an operation cost optimization model for power operators,select the optimal resident users participating in demand response,and dynamically set subsidy prices.Simulation results show that the proposed algorithm IDR-OL can achieve a balance between supply and demand while reducing the operating costs of power operators to a greater extent.3)Considering that in real data sets,there are complex nonlinear relationships between residential user electricity consumption behavior patterns and environmental factors such as subsidy prices,historical behavior,and temperature,which cannot be simply assumed as linear relationships.Therefore,an incentive demand response algorithm based on context Gaussian process regression for multi-armed bandit is proposed to learn the nonlinear relationship between context feature information and residential users’ power reduction in demand response in a data-driven manner,thereby depicting the potential of residential users to reduce power in demand response,and helping power operators select appropriate residential user participation plans in demand response and dynamically establish the optimal subsidy price.Simulation results show that the proposed algorithm IDR-CGPB can accurately depict the behavior patterns of residential users’ electricity consumption,and can significantly save operating costs for power operators when compared with other comparison schemes in demand response.