Adaptive Algorithm For Signal Control And Traffic Assignment Collaboration Model

Signal control and traffic assignment are the core problems of traffic network control.The collaborative model of the two has great application value,It is normally formulated as a bi-level programming model,however,the bi-level structure make it difficult to solve.In this paper,the algorithm for solving the collaborative model is studied.At first,the relationship between signal control and traffic assignment is analyzed,and a bi-level programming model is established.Then,based on the iterative features of single level model and solution space features of global model,an adaptive iterative optimization and assignment algorithm and adaptive combination optimization algorithm are designed.The adaptive iterative optimization and assignment algorithm is the improvement of the traditional iterative optimization and assignment algorithm.An adaptive difference correction term and a local search strategy are designed to improve the optimality of the result while ensuring fast convergence.Simulation cases in Nguyen-Dupius network and Sioux-Falls network show that,the adaptive iterative optimization and assignment algorithm reduces 10% of time costs compared to the iterative optimization allocation algorithm with better solution results,which reduces the gap between the iterative optimization and assignment algorithm and the global optimal solution on average 51%.The adaptive combinatorial optimization algorithm is based on the features of the solution space of the model,by dividing the solution space into smooth "blocks" and non-smooth "boundaries" which connect the "blocks".Combining the characteristics of breadth search of genetic algorithms and depth search of gradient descent algorithms,an adaptive combination optimization algorithm is designed.Simulation cases in Rinaldi network and Nguyen-Dupius network show that the adaptive combination algorithm has stronger breadth and depth search capabilities in the global optimization model,which obtain the approximate global optimal solution with only 35% of the computational costs of the genetic algorithm.The transferability tests in the simulation cases show that the two algorithm proposed in this paper can be applied to a larger-scale road network,which reflects its ability to solve practical problems.