Improved Origin-Based Quasi-Newton Method for Traffic Assignment Problem

摘要: Recently Bar-Gera proposed a quasi-Newton method for the origin-based formulation of user-equilibrium traffic assignment problem. This paper shows that Bar-Gera's is likely to generate negative second-order derivatives. The drawback leads "wrong search direction" in Newton-type which seriously impacts overall convergence. will prove wrong direction can be avoided if Bertskeas's upper bound used approximate derivative. alternative not only appealing theory, but indeed demonstrates more satisfying computational performance as evidenced numerical results. Pros and cons various strategies expanding trimming subnetworks are also discussed.