Dynamic Pricing Strategy For High Occupancy Toll Lanes

As a form of congestion pricing,high occupancy toll(HOT)lanes have been widely introduced and applied in many advanced countries like America.In terms of mitigating the congestion,it is worthy of studying the effectiveness and feasibility of HOT lanes for the Chinese transportation scholars.HOT lanes combine the congestion pricing and high occupancy vehicle(HOV)lanes successfully and the low occupancy vehicles(LOV)on general purpose(GP)lanes can move to HOV lanes by paying some tolls so that the HOV lanes will be better utilized.In such HOT system,the toll rates play an important role in allocating LOVs over the HOT and GP lanes dynamically,which has a direct impact on the road system performance.This study will propose a novel pricing strategy for HOT lanes with single ingress and egress.The pricing strategy presents a method to model the lane choice behavior and determine the dynamic toll rates in response to the traffic conditions.Firstly,the paper introduces the architecture of HOT system and the key technique to realize the dynamic pricing strategy.Then,in the view of data-driven and machine learning methods,the support vector regression(SVR)and random forest regression(RFR)are used to model the lane choice behavior of LOV based on the historical traffic and toll data.In order to improve the prediction model,the grid search method(GSM)and particle swarm optimization(PSO)are compared and incorporated into two models to optimize the model parameters,while the k-fold cross validation(k-CV)is applied for evaluating the optimized models.After that,a novel nested pricing model is formulated,where the lower level is for calculating the optimal lane-changing ratio of LOV base on the link transmission model(LTM)to obtain the minimal total travel delay and the upper level is built to optimize the toll rates to minimize the absolute errors between the optimal lane-changing ratio and the predicted lane-changing ratio according the real-time traffic conditions.Finally,the raw data from I405 in U.S.state of Washington was collected for case study and experiments.The results reveal the good performance of PSO and indicate the proposed hybrid model PSO+SVR has the highest prediction accuracy of 90.60%,in comparison with other 5 models.Additionally,the dynamic toll rates,derived from the nested pricing model and real-time data,can decrease the total delay significantly.During the randomly chosen peak hours,the travel delay decreases the total delay by 59.7%,which turns out the effectiveness of the proposed dynamic pricing strategy.