Deep Learning-based Queue-aware Eco-Approach and Departure system for Plug-in Hybrid Electric Bus at signalized intersections: a simulation study 2020-01-0584
Eco-Approach and Departure (EAD) has been considered as a promising eco-driving strategy for vehicles traveling in an urban environment, where signal phase and timing (SPaT) and geometric intersection description (GID) information are well utilized to guide the vehicles passing through the intersection in a most energy efficient manner. Previous studies by the authors formulated the optimal trajectory planning problem as finding the shortest path on a graph model where the nodes define the reachable states of the host vehicle (e.g., speed, location) at each time step, the links govern the state reachability from previous time step, and the link costs represent the energy consumption rate due to state transition. This method is effective in energy saving, but its computation efficiency can be enhanced by machine learning techniques. In this paper, we propose an innovative Deep Learning-based Queue-aware Eco-Approach and Departure (DLQ-EAD) System for a Plug-in Hybrid Electric Bus (PHEB), to provide an online optimal vehicle trajectory considering both the downstream traffic conditions (i.e. traffic lights, queues) and vehicle powertrain efficiency. Based on the optimal solutions from the graph-based trajectory planning algorithm (GTPA), a deep neural network (DNN) is developed to learn the optimal speed for the next time step given the current vehicle state. The trained DNN can provide energy efficient trajectory with high computational efficiency and it can be easily adapted to dynamic changes in surrounding environment. To address the effect of downstream traffic, a queue prediction model is further developed using data from on-board radar, CVs, along with signal information from SPaT message. The proposed queue prediction model is of pragmatic significance, even under low penetration rate of CVs. A comprehensive simulation study in microscopic traffic modeling software PTV VISSIM showed that the proposed DLQ-EAD can achieve 18.7% - 24.0% energy efficiency improvements for a single PHEB over various traffic congestion levels. With 20% CAVs in the network and the proposed queue prediction model, additional energy savings can be further improved by 2.0%-8.2%.
Fei Ye, Peng Hao, Guoyuan Wu, Danial Esaid, Kanok Boriboonsomsin, Zhiming Gao, Tim LaClair, Matthew Barth
University of California, Oak Ridge National Laboratory