��ͨ˫��滮����:ͳһ��ѧģ�ͼ����㷨

主要讨论基于用户平衡原则的交通网络优化问题。这些问题大致上可以分为二大类:一类是涉及到确定性用户平衡原则；另一类是考虑随机性用户平衡原则。众所周知，运筹学中的双层规划模型能够完美地刻画这些问题，但是所建立的双层优化模型往拄属于不可微优化问题的范畴，这就给设计有效的算法带来了很大困难.此文首先从模型和算法的角度总结了有关这类问题已有的研究成果，接着介绍有关这方面的最新的研究进展，即如何把用户基于平衡原则下的交通网络优化问题的双层规划模型统一地转换为一个连续可微的单层最优化问题，并设计统一的算法。作为统一的算法方面的研究，我们可以看到增广的拉格朗日方法可以用来解上述的第一类问题，而基于灵敏度的分析的序列二次规划方法完全有能力解上述的第二类问题。

Transport Bilevel Programming Problems: Unified Models and Algorithms

This paper is concerned with a class of transport bilevel programming problems investigated by analytical analysis approaches. It begins with a state-of-art review on transport bilevel programming problems taking into account behavior of network users' routing choice. These problems in reality can be classifies into two major categories: transport bilevel programming problems with deterministic user equilibrium constraints and transport bilevel programming problems with stochastic user equilibrium constraints. It is well recognized that the bilevel programming model or mathematical program with equilibrium constraints as unified modeling approach can perfectly characterize these two categories of problems. Nevertheless, induced bilevel programming models for the former category usually belong to a subject of nondifferentiable optimization problems, whereas that for the latter category becomes the continuously differentiable optimization problems. It should be pointed out that designing an efficient solution method for a nondifferentiable optimization problem is not an easy task. This study thus introduces the recent unified modeling approach for the problems in the preceding category, which aims at transforming a bilevel programming model into a single level continuously differentiable optimization problem. As unified algorithms, it can be seen that the augmented Lagrangian method and sequential quadratic programming method based on the sensitivity analysis for the stochastic user equilibrium problem are capable of solving any problem in the first and second categories. Finally, three examples are provided to demonstrate the unified continuously differentiable optimization approach.