Research On Scenario Construction And Planning Method For Multi-source Power System

With the publishment of the requirements of carbon neutrality and carbon peaking,the capacity of wind power,photovoltaic and other renewable energy units in the power system is bound to increase gradually,as well as the system complexity.The conventional planning method that is based on "typical days" cannot deal with the uncertainty contained in the renewable energy,and it could lead to waste and other problems.The power system planning method that is based on multi-scenario method can take the complicated operation conditions of the power system into consideration,so that the grid planning scheme is more adaptive.This paper focuses on the scenario construction and its corresponding planning methods that are suitable for multi-source power systems.First,a method for constructing time series scenarios suitable for power system planning is proposed.Resampling is used to generate multiple single-unit datasets,K-means clustering is used to construct typical scenarios from each single-unit dataset,and hierarchical clustering is used to perform secondary clustering on the results to construct the comprehensive single-unit typical scenario set.Then,the typical scenario sets of different units are connected with reduction based on the tabu search method.While satisfying the requirement of the size of the scenario set,the internal difference of the scenario set is enhanced,so that the scenario set can cover various operation conditions of power system.Considering the increase in the scale of typical scenario sets,an economic dispatch method based on reinforcement learning is established,where neural networks are used to construct the actor and the critic,the proximal policy optimization is used to train the agent,and then the scenario computing is performed by the trained agent.Finally,on the basis of scenario generation and scenario computing,a multi-scenario two-level planning model that accurately calculates operating costs is constructed.The first-level model is a long-term planning model that considers overall costs,and the second-level model is an economic dispatch model that considers operating costs.The objective function of the second model is used as the variable of the first model.The effectiveness and feasibility of the proposed methods are illustrated based on the IEEE 30-bus system and the SimBench platform.