电力系统混合整数线性规划问题的运筹决策关键技术综述与展望OA北大核心CSTPCD

The mixed-integer linear programming(MILP)problem is formulated to find the optimal decision in a discrete search decision space,which is wildly used in power systems,such as the unit commitment,maintenance scheduling,optimal transmission switching,power system planning,etc.MILP aims to achieve the best allocation of resources,whose accuracy and efficiency directly affect the security and economy of power systems.In a MILP model,both discrete and continuous decision variables are considered,and the physical constraints in power systems are linearly formulated because of the robustness and the convergence. Under the non-deterministic polynomial-time hardness,the number of feasible solutions to MILP is exponential to the scale of discrete decision variables,and cannot obtain the optimal solution by enumeration within the polynomial-time.Many methods and algorithms are proposed for MILP in power systems.Recently,the MILP solvers with the branch-and-bound(B&B)algorithm as the core is developing rapidly,and bring a considerable improvement to the solution efficiency,stability,and generality.Commercial MILP solvers,such as CPLEX,GUROBI,etc.,have been the dominated technology for MILP in power industries. However,MILP solvers still suffers from"combinatorial explosion"for large-scale MILP in power systems,and the independent intellectual property rights are restricted.Compared with commercial solvers,the open-source or the domestic MILP solvers has a gap in the solution efficiency.It is essential to develop a technological breakthrough for MILP in power systems.Therefore,this paper reviews the operation research in power systems and the latest developments in general MILP,and provides a future prospect on this topic,in order to offer a reference. First,this paper suggests several typical problems in power systems,including the unit commitment,maintenance scheduling,optimal transmission switching,power system planning,etc.,that can be formulated with a general MILP form.Second,this paper introduces a unified framework to solve the MILP model,while existing research is divided into different modules in the framework,i.e.,the solution process:(1)External model processing,including the processing of variables,constraints,and formulations;(2)presolve and cutting planes;(3)the branch-and-bound algorithm,including the variable selection strategy,the node selection strategy,and the heuristic;(4)parameter tuning,including the solution time prediction.Third,this paper summaries that the bottleneck of operation research in power systems is:(1)The wildly-studied external model processing methods cannot balance the optimality guarantee and the solution efficiency;(2)the internal algorithms are designed for general MILP instead of specific for MILP in power systems.Therefore,this paper presents the prospect for the future research of operations research in power systems that customized strategies should be developed based on both the physical feature of power systems and the mathematical information of MILP solvers.This idea is composed of three steps:(1)To build the physical feature of power systems;(2)to dig out the internal information during the solution process;(3)to embed the customized strategies into the solution process. In conclusion,this paper aims to draw the academic attention to the innovative research idea for MILP in power systems,and to provide a reference for related work in China.Although this paper does not consider the nonlinear nature of operation research in power systems,the main content should provide an inspiration for the general combinational problem.