Research On Optimal Operation Strategy And Method Of Energy Management And Demand Response In Smart Grid

Smart grid,as an emerging power technology,gradually replaces the traditional power grid and becomes the development direction of the future power grid.With the introduction of various forms of renewable energy,battery energy storage systems and other new distributed energy resources,the supply side of smart grid appears pluralism,intermittence,and randomness.At the same time,with the emergence of new load units such as the new energy vehicles and intelligent buildings,the demand side of smart grid shows diversity,randomness,and flexibility.Therefore,the optimal operation of energy management and demand response in the context of smart grid is facing new challenges.Addressing the energy management and demand response problems of smart grid is usually transformed into solving the optimization problems with a set of constraints.However,there exist some shortcomings in the current research works as they have not fully exploited the advantages of the introduction of distributed renewable energy and two-way power and information flow.To be specific:1)the existing optimization algorithms need to design extra algorithms to track the power imbalance between supply and demand sides,making it difficult to ensure real-time power balance between supply and demand sides;2)existing studies focus on modeling and designing optimization algorithms from the perspective of optimization to obtain the optimal operation strategy,but lack of effectiveky characterizing the interaction mechanism among all participating units,the decision-making order,and the coordination of individual interests and group interests;3)the existing works mainly investigate the operation strategies based on the deterministic models,resulting in that the constructed optimization models and the designed algorithms cannot adapt to the random uncertainty in the dynamic electricity market environment,and thus cannot fully stimulate the enthusiasm of users to participate in grid operation.Considering that game theory can depict the interactive relationship between multiple decision makers and reinforcement learning has the ability of interactive learning with the environment,this thesis investigates the optimal operation strategy and method of energy managment and demand response in smart grid from the perspectives of optimization,game,and learning.The main research contents and and contributions are summarized as follows:1.To address the challenges of power imbalance beween supply and demand sides caused by the intermittent renewable energy,this thesis adopts the multi-agentconsensus theory to propose distributed coordination algorithms based on Newton method and ADMM from the perspective of optimization in Chapters 3 and 4,respectively.The proposed algorithms are shown to effectively coordinate the power output of generation units and the energy consumption of load units,thereby minimizing the energy cost of the system while meeting the load demand of each area.Additionally,we utilize the complementary coordination of singlearea economic dispatch(ED)and multi-area ED to reduce the impact of intermittent renewable energy on power grid and ensure the real-time power balance between supply and demand sides,thus promoting the penetration of large-scale renewable energy in smart grid.2.To solve the problem of conflict between individual interests and group interests,this thesis studies the demand response problem from the point view of game theory in Chapters 5 and 6.Specifically,in Chapter 5,we first formulate a Stackelberg-Nash game framwork which can not only describe the hierarchical decision-making process between utilty company and end-users,but also the noncooperative competitive relationship between end-users.Then,the existence and uniqueness of Stackelberg-Nash equilibrium are rigorously proved.After that,a distributed equilibrium seeking algorithm with an adaptive step size is developed to achieve the coordination of individual interests and group interests.In Chapter 6,we propose a new mechanism to allow the interaction between EVs under a non-cooperative game framework and a distributed coordination algorithm based on Newton fixed point to solve the problems of demand surge and power grid collapse caused by the direct interaction between large-scale EVs and charging stations.3.To address the uncertainty of electricity price and energy consumption behavior of loads,this thesis studies the demand response problem from the learning perspective.Specifically,in Chapter 7,the price-based demand response problem is first modeled as a two-layer optimization problem.To address the randomness of retail prices,the upper problem is further reformulated as a Markov decision process with unknown state transition probabilities,and then the optimal dynamic retail prices are learned by a model-free reinforcement learning(RL)algorithm.After that,the optimal energy consumption strategies are determined based on the optimal retail prices.The learning-based algorithm can not only relax the dependence on the electricity price model,but also maximize the social welfare within an unknown electricity market environment.On this basis,in Chapter 8 we further study the EVs charging/discharging scheduling and develops a multi-agent deep RL method which enables the multiple decision makers to learn their optimal strategies by directly interacting with the environment and fully stimulates the enthusiasm of users to participate in power grid scheduling,thus improving the social benefits.