| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 27.89 MB | Adobe PDF |
Authors
Advisor(s)
Abstract(s)
This dissertation considers how a parcel delivery company can improve last mile delivery services’ performance using historical data. To tackle this challenge, we start by proposing a framework for data cleaning, in order to produce reliable data for vehicle routing problems. Data on the historical geographical location of clients is used to model a hierarchical districting problem, mid-level districts (named micro districts) are limited to an eight hour shift, representing a daily route. Using the districting solution as a procedure for package to driver/vehicle assignment, it is possible to achieve a 14% decrease in the number of vehicles needed, while keeping daily routes more balanced in terms of working times. Using a transition probabilities based TSP to sequence nano zones (the lower-level districts), the preferences of drivers are used as a cost function. The transition probabilities based TSP produces solutions with a total distance 12% higher, comparing with a distance based TSP. Moreover, sequencing the nano zones using the maximum likelihood routing enables the incorporation of the driver’s tacit knowledge.
Description
Keywords
Districting Data Driven TSP Maximum Likelihood Routing