On Distributed Operation And Optimization In Smart Grid:An MAS Coordination Algorithm-Based Approach

With the integration of renewable energy resources(DERs),distributed generation(DG),and demand response(DR),the scale of the power grid grows drastically.In addition,the power grid is divided into different regions under the control of different entities and is not controlled by a single entity.Therefore,the conventional centralized organization and operation methods of power systems face various challenges.On the other hand,the developments of information communication technology(ICT)and the advances in optimization and control theory introduce promising approaches for the next generation power system,leading to the emergence of smart grid concept.In the smart grid,the economic dispatch(ED),unit commitment(UC),and opti-mal power flow(OPF)problems have regained a lot of research interests.The existing works generally focus on considering more new components such as DG and DR,im-proving the economic operation model,and developing distributed operation methods.More specifically,these works focus on:1)considering more components without los-ing the convexity of ED,and developing distributed operation algorithms,while little attention is devoted to the non-convex ED and the case when the cost function is not accurate;2)improving the modeling methods in the UC problem,while the distributed operation algorithm for UC is rarely investigated;and 3)decomposing the OPF and proposing distributed algorithms in cyber layer,leading to some inconsistency with the physical power grids.Thus,this dissertation studies the distributed operation of smart grid by a multi-agent system(MAS)based approach while considering the non-convex cost functions,inaccurate models,and features of power grids.The main works are summarized as follows.1.The ED problem with non-convex cost functions and inaccurate cost coefficients is investigated.First of all,the ED problem is reformulated as a derivative-free optimization(DFO)problem.Secondly,the multi-agent system based ED sys-tem is presented as well as the topology discovery algorithm.Then,a distributed pattern search algorithm(DPSA)is proposed for the DFO problem.Finally,theo-retical analysis and numerate simulations are presented to verify the convergence and efficiency of the proposed algorithm.On the other hand,the proposed algo-rithm can be applied to a class of DFO problems.The construction of positive spanning matrix by using the Laplacian matrix is potentially helpful to design other distributed derivative-free algorithms.2.The economic operation of smart grid considering ED and UC is studied with non-convex cost functions and inaccurate models.Firstly,the ED and UC is combined into a unified problem.Secondly,a centralized reinforcement learning(RL)based algorithm is proposed to solve the unified problem under discretiza-tion assumptions.Thirdly,a distributed RL algorithm is proposed by eliminating the requirement for global information,conducting cooperative exploration and exploitation,and establishing local action-value functions.Convergence analy-sis is also presented.Thanks to the RL,the proposed algorithms are capable of improving the operating policy by interacting with the power grid.The cooper-ative exploration and exploitation will potentially help to design other RL based distributed algorithms.3.The closed-loop optimal operation of microgrids is studied while considering DG,DR,and carbon emissions.Firstly,the economic operation problem is reformu-lated into an equivalent optimal switching problem of a hybrid system.The the-oretical analysis on the equivalence is presented.Then,a closed-loop optimal operation algorithm is proposed by employing the dynamic programming and neural network(NN)approximations,The convergence of the algorithm is also discussed.The algorithm is closed-loop in the sense that new optimal operation is given under different initial conditions and disturbance without retraining the NN.On the other hand,focusing on the optimal policy(function)instead of solution(point)may introduce a new approach for other similar problems.4.The cooperative OPF of multiple microgrids is discussed while considering the coupling between microgrids.Firstly,a detailed standalone microgrid model is introduced.Secondly,the cooperative OPF model of multiple interconnected mi-crogrids is presented.Then,a distributed cooperative OPF algorithm is proposed by employing Alternating Direction of Multipliers Method(ADMM).The con-vergence of the proposed algorithm is also discussed.Note that the proposed algorithm is fully distributed,requiring no central coordinators and global com-munications.One additional flexibility of the proposed algorithm is that:each microgrid is free to choose appropriate algorithms to solve the local OPF.This flexibility may be helpful to combine different optimization algorithms.