Using Machine Learning to Customize Traffic Prediction for High Performance Traffic Analysis and Optimization

PI: Robert Heckendorn (UI),, ORCID: 0000-0002-1319-0670

Co PIs: none

AMOUNT & MATCH: $40,000 from PacTrans; $40,000 Match

PERFORMANCE PERIOD: 3/16/2021 – 3/15/2022

STATUS: Active

CATEGORIES: Machine Learning, Traffic Prediction, Optimization



FINAL PROJECT REPORT: will be available once completed

PROJECT DATA: will be available once completed

DESCRIPTION: This project explores using machine learning techniques to spatially and temporally customize predictive functions in queuing based macro simulations of traffic. Its objective is to replace much slower “one-size fits all” micro simulators so that reliable adaptive traffic control and optimization will be possible, which is a very practical end-goal.

Our first goal is to create machine learning algorithms for learning how to predict the travel time of a car on a specific segment which may include difficult segments which represent signaling such as intersections or toll booths. Data will ultimately come from car telemetry monitoring. The predictor functions will be further used in this project to make much more efficient, reliable, and location sensitive traffic simulator which is necessary for future optimization algorithms. Our initial experiments will attempt to reproduce VISSIM output but much more efficiently.

Our second goal is then to create traffic optimization algorithms using our simulator to estimate cost of a signaling strategy. We will be using several algorithms we have used in the past for evacuation planning as a starting point.

Research Project Progress Report #1 10/10/2021
No Cost Extension Request 1/15/2022
Draft Report 1/15/2022
Final Project Report 3/15/2022