Deep Learning Based Proactive Multi-Objective Eco-Routing Strategies for Connected and Automated Vehicles

Lama Alfaseeh,Bilal Farooq

doi:10.3389/ffutr.2020.594608

Abstract

This study exploited the advancements in information and communication technology (ICT), connected and automated vehicles (CAVs), and sensing to develop proactive multi-objective eco-routing strategies for travel time and Greenhouse Gas (GHG) emissions reduction on urban road networks. For a robust application, several GHG costing approaches were examined. The predictive models for link level traffic and emission states were developed using the long short-term memory (LSTM) deep network with exogenous predictors. It was found that proactive routing strategies outperformed the reactive strategies regardless of the routing objective. Whether reactive or proactive, the multi-objective routing, with travel time and GHG minimization, outperformed the single objective routing strategies. Using a proactive multi-objective (travel time and GHG) routing strategy, we observed a reduction in average travel time (17%), average vehicle kilometer traveled (22%), total GHG (18%), and total nitrogen oxide (20%) when compared with the reactive single-objective (travel time).

Highlights

The transportation system in the U.S produced 28% of the total greenhouse gas (GHG) emissions in 2018, which was the largest share from a single source (EPA, 2020b)
With regards to the predictive models for speed and travel time, it was found in the literature that long short-term memory (LSTM) outperformed other predictive approaches, including the autoregressive integrated moving average (ARIMA) model (Ma et al, 2015)
Related to GHG predictive models, GHG emissions were predicted based on yearly data points of fuel (Zhao et al, 2011), gross domestic product, or other economic factors (Pao and Tsai, 2011; Ameyaw et al, 2019)

Summary

Introduction

The transportation system in the U.S produced 28% of the total greenhouse gas (GHG) emissions in 2018, which was the largest share from a single source (EPA, 2020b). According to the United States Environmental Protection Agency in 2017, 81% of the GHG was CO2 , which is a major contributor to climate change and global warming. CH4, N2O are converted to “CO2 equivalent” to estimate GHG emissions The common features between most of the aforementioned predictive models were the employed low temporal resolution and small case study. Even when a large network was employed as in Zhang et al (2019), the speed was predicted at a low level of temporal resolution. The predictive models varied from statistical (Tudor, 2016; Rahman and Hasan, 2017)

Objectives

Methods

Results

Conclusion