In this paper we describe how the capacitated user equilibrium can be approximated by sequential uncapacitated models by the use of a penalty function. The efficiency of the method is governed by the algorithmic performance of the uncapacitated model. A skew gradient-based Newton method is used to solve the capacitated user equilibrium within the feasible region of path flows. In the path-flow region, the straight gradient is defined as the derivative of the objective function with respect to the flow of the corresponding path, while the skew gradient is defined for each particular origin destination pair and is characterized by the average cost of all the paths for that pair. Instead of movement of flow toward the shortest path, in the equilibration procedure path flows below the average decrease and path flows above the average increase. The characteristics of the Newton method with the column generation procedure are combined to achieve the efficient determination of the equilibrium point. Numerical experiments demonstrate the excellent performance of the proposed method and highlight its potential applications.
A set of constrained Newton methods were developed for static traffic assignment problems. The Newton formula uses the gradient of the objective function to determine an improved feasible direction scaled by the second-order derivatives of the objective function. The column generation produces the active paths necessary for each origin-destination pair. These methods then select an optimal step size or make an orthogonal projection to achieve fast, accurate convergence. These Newton methods based on the constrained Newton formula utilize path information to explicitly implement Wardrop＇s principle in the transport network modelling and complement the traffic assignment algorithms. Numerical examples are presented to compare the performance with all possible Newton methods. The computational results show that the optimal-step Newton methods have much better convergence than the fixed-step ones, while the Newton method with the unit step size is not always efficient for traffic assignment problems. Furthermore, the optimal-step Newton methods are relatively robust for all three of the tested benchmark networks of traffic assignment problems.
