基于机器学习-优化混合算法的离散交通网络双层规划模型  

A Bi-level Programming Model of the Discrete Transportation Network Design:A Hybrid Machine-learning and Optimization Method

作　　者：刘丹[1,2] 蒲自源 许晓晴[1] 伍佳妮[1] LIU Dan;PU Zi-yuan;XU Xiao-qing;WU Jia-ni(School of Economics and Management,Chang'an University,Xi'an 710064,China;Department of Civil and Environmental Engineering,University of Washington, Intelligent Transportation Research and Application Laboratory,Seattle 98105 ,USA)

机构地区：[1]长安大学经济与管理学院,陕西西安710064 [2]华盛顿大学土木与环境工程系,智能交通研究与应用实验室,美国西雅图98105

出　　处：《系统工程》2018年第8期114-122,共9页Systems Engineering

基　　金：国家自然科学基金青年项目(61806021;71701022);教育部人文社科项目(15YJCZH093;16YJC630141);陕西省社科基金资助项目(2014P08);陕西省重大理论与现实研究项目(2018Z158;2018Z013);中央高校基本科研业务专项(310823160645).

摘　　要：为求解离散交通网络设计的大规模双层规划问题(Bi-level Programming Program,BPP),提出一种机器学习-优化(ML-OP)混合算法。首先,基于分解方法的求解思路,将双层规划问题转换为:下层一个典型的非线性规划问题(NLP),上层一个整数规划问题来处理。研究设计的混合方法,上层问题用源于机器学习的替代问题代替,采用监督学习算法求解(SLA),下层问题为交通分配问题(TAP),是一个凸非线性规划问题,用Frank-Wolfe算法求解。其次,为了验证机器学习-优化算法的有效性,用一个真实的大规模数据集来测试,并与经典分支定界算法(Branch-and-Bound)进行比较。数据结果表明在处理大规模双层规划问题时,ML-OP混合算法在计算能力方面有较好的表现。A hybrid machine learning-optimization algorithm is proposed for solving the large-scale bi-level programming problem(BPP)in discrete transportation network.Based on the decomposition method,the Bi-level programming problem is transformed into a typical nonlinear programming problem(NLP)at the lower level,and an integer programming problem at the upper level.The upper level problem is treated as a substitute with a machine learning problems,using supervised learning algorithm(SLA).Frank-Wolfe algorithm is used to solve the traffic assignment problem(TAP)at the lower level,which is a convex nonlinear programming problem.A real large-scale dataset is used to test the effectiveness of the hybrid algorithm,also the machine learning-optimization algorithm is compared with the classic Branch-Bound algorithm.The results show that the hybrid algorithm of machine learning and optimization performs well in computing when dealing with large-scale bi-level programming problems.

关 键 词：离散交通网络设计 双层规划模型 交通分配 机器学习 混合算法 

分 类 号：U113[交通运输工程]