Intelligent Tidal Lane Setting Algorithm Based on LSTM-GASVR and Multi-objective Optimization

Abstract

To solve the problems of low utilization of tolled tidal lanes and the waste of resources caused by traffic congestion, this paper proposes a tidal lane setting model combining multi-scale LSTM-GASVR and multi-objective optimization. Based on survey data and historical traffic data from expressway toll stations in Shaanxi Province and the Caltrans Performance Measurement System (PeMS), a multi-scale LSTM-GASVR short-term traffic flow prediction model with workday indicators is developed to capture traffic time-series characteristics. Using the predicted traffic flow, a multi-objective optimization framework considering the trade-off between bidirectional lane service levels is constructed to determine tidal lane switching strategies. The multi-objective optimization-based model enables real-time tidal lane switching according to traffic queuing conditions. Experimental results show that the prediction model achieves R2 = 0.982, MAPE = 0.118, MAE = 20.156, and RMSE = 28.237. Further, the service levels of the balanced bidirectional lanes are adjusted to Level 3 or below. Moreover, four typical M/M/c queuing indicators, namely the average number of vehicles, average queue length, average residence time and average waiting time, are all significantly improved during peak hours. This model improves the traffic capacity of expressway toll stations during peak hours and optimizes the unbalanced service level of each time period.