Spatiotemporal Fuel Consumption Forecasting

The Hong Kong University of Science and Technology
Department of Computer Science and Engineering

Final Year Thesis Oral Defense

Title: "Spatiotemporal Fuel Consumption Forecasting"

by

YAP Zhi Yun

Abstract:

Long term fuel consumption forecasting is the first step to recommending 
fuel efficient route reducing energy consumption of vehicle. We formulate 
fuel consumption prediction problem as a new spatiotemporal forecasting 
task in urban computing and developed a new benchmark dataset - Shenzhen 
Fuel Consumption (SZ-FC) dataset. In this paper, we propose the 
Spatiotemporal Multi-graph Atten- tion Network (ST-MGAN), a novel deep 
learning framework to predict regional fuel consumption level. We encode 
multiple inter-region correlations using multi- graph and attention-based 
aggregation mechanism, and leverage the computational efficient temporal 
convolution network (TCN) to capture the long-term dependency.?Experiments 
conducted on two real-world fuel consumption and traffic speed datasets 
(SZ-FC, PeMS-M)?show that the proposed ST-MGAN inhibit outstanding 
generalization and transferability property while outperforming the 
state-of-the-art model in 40% of the prediction intervals.


Date            : 3 May 2021 (Monday)

Time            : 17:00-17:40

Zoom Link:
https://hkust.zoom.us/j/99631956979?pwd=QkZLTzh2c1RmNEFnak00Tk4wWDZCQT09

Meeting ID      : 996 3195 6979
Passcode        : 756245

Advisor         : Dr. HUI Pan

2nd Reader      : Dr. CHATZOPOULOS Dimitris