Research On Smart Microgrid Optimization And Control Strategy Based On Deep Reinforcement Learning

With the rapid development of smart grid,energy internet,and ubiquitous power Internet of Things,a large number of distributed renewable energy power generation is connected to the grid,the electricity market is gradually deregulated,the amount of data obtained by measurement units is growing rapidly.The power system’s requirements for safe,reliable,economical,efficient,intelligent,and environmentally friendly operation control have gradually increased.Traditional methods face certain difficulties in this context.Being able to process higher-dimensional data,having adaptive learning capabilities,being able to handle more complex scenarios,being able to run online in real time,etc.are the requirements of power system development.Deep reinforcement learning is an emerging artificial intelligence method that combines the perception capabilities of deep learning and decision-making capabilities of reinforcement learning.It can achieve optimal control from input to output through end-to-end training and learning.It can sense high-dimensional data and perform real-time feedback and regulation.Meanwhile,it has strong universality,and has achieved great success in many fields.Deep reinforcement learning is also applicable to power systems,including optimization control problems in microgrids.This paper first studies the hybrid energy management problem in microgrids and the hierarchical electricity market trading problem considering demand response.And then this paper analyzes the main characteristics of the problems,establishes more detailed problems model,and illustrates the optimization control problems that need to be solved.Furthermore,the Deep Q-Network(DQN)and Deep Deterministic Policy Gradient(DDPG)algorithms in deep reinforcement learning are studied,and the mathematical principles,logical structure and applicable scenarios of the algorithms are analyzed.Using DQN and DDPG to conduct case studies on composite energy management issues and electricity market trading issues,respectively,and achieved ideal results.Through the practice in two different scenarios,the effectiveness,adaptability,and stability of the deep reinforcement learning method in the optimization control problem of the microgrid are explained,and the possibility of combining artificial intelligence and power systems is also explained.In addition,through comparison with classical game theory methods,it shows that deep reinforcement learning extracts high-dimensional data features and makes decisions based on real-time feedback,which has better dynamic performance and avoids some disadvantages of traditional model methods.