EAGER: Optimization with Data Acquisition in Transportation Engineering

University Of Maryland, College Park, College Park MD

Investigators

Abstract

This EArly-concept Grant for Exploratory Research (EAGER) project will seek to acquire fundamental knowledge in determining the amount of data that should be purchased for statistically estimating the coefficients of optimization models. This will be done by examining the tradeoffs between the cost of data used for calibrating optimization models and the quality of solutions obtained with those models. Such tradeoffs will be explored mainly for problems relevant in transportation, such as vehicle assignment to tasks, vehicle routing, fleet sizing, and network flow models. The results should help transportation agencies and companies across the world in allocating resources and improving their professional practice. Beyond transportation problems, this work may yield methods that would be valuable in other applications, industries and disciplines, including engineering, operations research, economics and management. Traditional optimization approaches do not account for the cost of acquiring input data, which can be very important in many applications of engineering and operations research, including transportation. The need for considering the cost of data acquisition is even greater now and in the future when the ownership of the data for transportation is moving from publically owned to privately owned. The goals are to develop methods for optimizing the amounts of data acquired for calibrating some specific transportation optimization models and then explore whether and how such methods and guidelines may be generalized to other optimization problems in other fields. The quality and cost of data may determine how much should be obtained. In addition to extensive experiments involving real-world data, the project will examine the structures of datasets and their underlying distributions, to explore how a generic decision making framework may be developed for determining "how much data we should purchase in order to calibrate an optimization model"?