Collaborative Optimization Technology Of Active Distribution Network With SOP Based On Reinforcement Learning

With the proposed goal of ’carbon peak,carbon neutral’,an increasing proportion of renewable distributed power is connected to the distribution network in order to alleviate the energy crisis and meet people’s urgent demand for a better life.However,the inherent volatility and uncertainty of renewable energy such as wind power and photovoltaic bring new challenges to the safety and stablility of operation of distribution network.The insufficient regulation capacity of the traditional distribution network may lead to prominent problems such as large voltage fluctuation,low consumption level of renewable energy and poor economic indicators.Therefore,it is necessary to explore the flexible controllable resources within the system fully,improve the coordination and interaction ability among various resources,maximize the flexibility of the system,and effectively resist the adverse effects of renewable energy on the system.For the optimization of active distribution network,the specific work of this paper is as follows:(1)Explore the flexible and adjustable resources of the distribution network,explain the basic principles,methods and characteristics of different adjustment means,and construct the multi-period collaborative optimization operation model of the distribution network based on controllable resources both at the source and network side;Firstly,the power between the distribution network and the superior network,as well as controllable distributed power sources,such as micro gas turbines and other equipment,are adjusted at the source side.Secondly,the dynamic adjustment of the network structure is played at the network side to improve the power flow distribution.Finally,a new type of intelligent soft open point(SOP)switch is introduced to explore its flexible ability of power flow control.(2)This paper constructs a two-layer optimization model based on reinforcement learning,which is used to solve the cooperative optimization problem of multi-period dynamic reconstruction of distribution network with SOP.The upper layer of the model is used to optimize the network topology structure.The traditional contact switch is partially replaced by SOP.On this basis,the radial topology structure set is generated,which is used as the action space of reinforcement learning.The lower layer optimizes the operation of controllable active equipment with SOP based on the network structure of the upper layer,and constantly modifies the topological selection of the upper layer according to the optimization results.This multi-time scale mixed integer nonlinear programming problem can be efficiently solved,and the convergence rate of the algorithm is accelerated by deleting invalid networks and sampling historical information with priority.(3)Based on IEEE33 nodes system and IEEE123 nodes system examples,it is verified that the proposed double-layer collaborative optimization method based on reinforcement learning can effectively improve the operation of active distribution network,and improve the security and economy of system.In addition,deep reinforcement learning is a data-driven artificial intelligence method,which can better adapt to the uncertain scenarios of source and load within a certain fluctuation range through training.Moreover,a multi-task agent group based on scene clustering is further constructed,and corresponding agents are trained differentiated according to different typical scenarios,thus improving the consumption level of renewable energy.