基于图卷积神经网络的机组组合问题加速求解方法  

Acceleration Solving Method for Unit Commitment Problem Based on Graph Convolution Neural Network

作　　者：曾贵华 刘明波[1] ZENG Gui-hua;LIU Ming-bo(School of Electric Power Engineering,South China University of Technology,Guangzhou 510640,China)

机构地区：[1]华南理工大学电力学院,广东广州510640

出　　处：《电工电气》2024年第3期44-50,共7页Electrotechnics Electric

基　　金：国家自然科学基金项目(52077083)。

摘　　要：针对传统的精确优化算法求解规模较大的机组组合问题面临时间可行性的挑战,提出了一种基于图卷积神经网络的机组组合问题加速求解方法。将机组组合问题构建为一个混合整数线性规划模型,根据分支定界法的求解原理,将分支策略定义为从候选变量的特征到候选变量得分的映射关系;提出在离线阶段使用图卷积神经网络来模拟强分支策略的决策行为,并将学习到的映射关系应用到在线分支过程中,从而加速分支定界法求解机组组合问题。通过IEEE 39节点10机组和IEEE 118节点54机组系统的算例分析,验证了所提方法的有效性。To solve the challenge of time feasibility faced by traditional accurate optimization algorithms for solving large-scale Unit Commitment(UC)problems,this paper proposes an accelerated solution method for solving the UC problems based on graph convolution neural network.Firstly,the UC problem is constructed as a Mixed Integer Linear Programming(MILP)model.Next,according to the solution principle of the branch-and-bound method,we define the branching strategy as a mapping relationship from the features of candidate variables to the scores of candidate variables.Thus,we propose to mimic the decision-making behavior of strong branching strategy in the offline phase using Graph Convolutional Neural Network(GCNN)and apply the learned mapping relationship to the online branching process to accelerate the process of the branch and bound method to solve the UC problem.Finally,the effectiveness of the proposed method is verified by the analysis of IEEE 39-node 10-unit and IEEE 118-node 54-unit systems.

关 键 词：发电机 机组组合 分支定界法 分支策略 图卷积神经网络 

分 类 号：TM744[电气工程—电力系统及自动化]