Bi-objective bi-level optimization for integrating lane-level closure and reversal in redesigning transportation networks

Traditionally, traffic congestion was alleviated through significantly upgrading the infrastructure of transportation networks. However, building new roads or adding more lanes to a main road needs huge expenses. A better cost-effective approach is to redesign and fine-tune transportation networks by closing and reversing existing lanes. This paper aims at developing an optimal scheme for lane-level closure and reversal to improve the performance of existing transportation networks with a fairly tight budget. We call this new problem Lane-level Closure and Reversal Problem (LCRP). By considering the capacities of all lanes of a road, two different bi-objective bi-level programs (called the arc-based and lane-based models) are developed to formulate the LCRP. Furthermore, our proposed formulations consider the elastic traffic demand and the elimination of conflict points resulting from reversing lanes. A hybrid machine learning and bi-objective optimization (MLBO) algorithm is developed to overcome the curse of dimensionality of the bi-level programs, especially for the land-based model that is more general but with higher computational complexity. The proposed methodology is illustrated by a small-size numerical example and verified by a real-world case study from Winnipeg (i.e., a benchmark transportation network). Computational results show that the integrated lane-level closure and reversal model can achieve a 0.52% reduction in the total travel time, which is significantly better than the 0.05% reduction individually obtained by arc-level closure or the 0.10% reduction obtained by arc-level reversal. The proposed methodology is beneficial for the traffic management bureau to make a more precise decision in redesigning transportation networks in practice.