Distributed Q-Learning In Power Grid Economic Dispatch And Firefighting Resource Allocation

In recent years,reinforcement learning has attracted much attention due to its wide application in unmanned driving,image processing,natural language processing,target recognition and other fields.Using Q-learning algorithm to solve problems in engineering has always been a research hotspot in the field of reinforcement learning.How to abstract practical problems into mathematical models and how to deal with different constraints are the core issues in this field.In this paper,a distributed Q-learning algorithm is designed for dynamic economic scheduling problem and forest fire resource allocation problem in smart grid.The main work is as follows:For the dynamic economic dispatch problem in smart grid,a mathematical model is established.The design goal of the algorithm is to configure the optimal generation and purchase power of each region at each time under the condition of unknown power cost function,and minimize the sum of power costs of multiple regions while satisfying multiple constraints.A local Q function is established for each region,and a distributed Q-learning algorithm based on information interaction between regions is designed.Under this algorithm,each region cooperates to find the optimal power combination satisfying the constraints of supply and demand balance.The designed algorithm tests its performance in the power distribution experiments of four regions,and further realizes the power distribution and optimization in the IEEE 39-Bus power system.On this basis,a new constraint and the size of the grid are added.The experimental results show that the algorithm still has good performance under the condition of considering transmission loss.A mathematical model is established for the fire scheduling problem of forest fire extinguishing.The goal of algorithm design is to allocate allocate the fire equipment of each fire unit to minimize the sum of the failure probabilities of fire fighting under various realistic constraints.In this algorithm,the agent observes the global information from the local information according to the consistency protocol,and uses the global information to find the optimal fire equipment allocation satisfying the constraint.Numerical simulation experiments verify the effectiveness of the algorithm.There are 17 figures,17 tables and 63 references in this paper.