Genetic-algorithms-based approach for bilevel programming models

Many decision-making problems in transportation system planning and management can be formulated as bilevel programming models, which are intrinsically nonconvex and hence difficult to solve for the global optimum. Therefore, successful implementations of bilevel models rely largely on the development of an efficient algorithm in handling realistic complications. In spite of various intriguing attempts that were made in solving the bilevel models, these algorithms are unfortunately either incapable of finding the global optimum or very computationally intensive and impractical for problems of a realistic size. In this paper, a genetic-algorithms-based (GAB) approach is proposed to efficiently solve these models. The performance of the algorithm is illustrated and compared with the previous sensitivity-analysis-based algorithm using numerical examples. The computation results show that the GAB approach is efficient and much simpler than previous heuristic algorithms. Furthermore, it is believed that the GAB approach can more likely achieve the global optimum based on the globality and parallelism of genetic algorithms.