Bi-level Optimization Algorithm for Dynamic Reversible Lane Control based on Short-term Traffic Flow Prediction

Abstract

Traffic congestion is a more and more serious problem all over the world. Reversible lanes have been used throughout the world to mitigate the effects of congestion and optimize roadway performance for more than 80 years. They have been applied on a variety of roadway types using different control methods to address an assortment of needs. However, the limited traditional control methods can not meet the increasing various demands. To address the needs of the freeway scenario, the study introduces a bi-level method based on short-term traffic flow prediction for the dynamic reversible lane control algorithm. The work improves the traditional traffic management method, reversible lane control, from static control to dynamic real-time traffic management. Taking advantage of the development of neural network technology, the input of the algorithm covers not only historical data and real-time data but also the predicted data. Advanced Bi-directional Long Short-term Memory (Abi-LSTM) model is employed for the short-term traffic flow prediction. For the control algorithm, the study introduces the bi-level optimization method to maximize the total traffic flow in both directions which determine the lane deployment. Also, the study considers the user costs in the lower level optimization formula. Finally, the study builds up a simulation to test the effect of the dynamic reversible lane control algorithm.