Distributed Optimization Of Social Welfare And Economic Dispatch In Smart Grid

As the awareness of ecological civilization receives more and more attention,the research and application of distributed renewable energy are gradually emerging.With energy storage equipment on the load side connected to the grid,the stringent requirements of power supply reliability and power quality make distributed computing be used widely in smart grid.This thesis studies the social welfare and economic dispatch problems in smart grid.The main works are as follows:In the first chapter,we introduce the background knowledge of smart grid,as well as the current research status in the field at home and abroad.Then the motivation and derivation of the main work is presented.In the second chapter,the main theoretical tools used in this paper are introduced,including algebraic graph theory and reinforcement learning.In the third chapter,we study the social welfare problem with transmission losses in smart grid.The addition of transmission losses make the feasible set of the formulated problem be non-convex,which induces that the methods of convex optimization can not effectively solve the problem.In order to overcome this,we find the same optimal solution to the problem.When solving the convex optimization problem,we design a distributed optimization algorithm based on the Lagrange multiplier method.As a result,the equilibrium point of the algorithm echoes the optimal solution of the convex optimization problem.Then,we prove the convergence of the algorithm.Finally,the effectiveness of the algorithm is verified by numerical simulation.In the fourth chapter,we study the unit commitment and economic dispatch problem in smart grid.In the mathematical model of the problem,we assumed that the mathematical expression of the generation cost functions are unknown.For this optimization problem with unknown functions,the optimization algorithm with gradients can no longer work.Since reinforcement learning can solve the problem with unknown environment,we first propose a centralized Q-learning algorithm.Considering that the centralized algorithm require a central agent and has lower robustness,we further propose a distributed Q-learning algorithm.Then we theoretically analyze the convergence and optimality of the distributed Q-learning algorithm.Finally,the simulation illustrates the correctness of the algorithm.In the last chapter,we make a summary and propose the relevant research work in the future.