Parameter tuning for a routing solver – PostNL
Overview
| Company Name / Department |
PostNL |
| Contact Person | Laurens Bliek |
| Location | Den Haag / online |
| Optional remote work | 100% remote possible |
| Travel expenses (own account or reimbursed by the company) | Own account |
| Housing arranged by company | No |
| Housing expenses (how much per month, own account or subsidized by the company) | Own account |
| Internship compensation | Yes, amount unknown |
| Study program | OML, AI-focused masters |
| ESCF community |
Full member |
| Start date |
September 2023 |
Company Description
We deliver parcels every day and mail five days a week. We are the indispensable link for our customers between senders and recipients, and the connector between the physical and the digital world.
Project Description
Project description:
At this large Dutch mail and parcel company, solving vehicle routing problems is a daily task. PostNL has developed their own local search algorithm to solve this task, but this algorithm has many hyperparameters that need to be tuned in order to efficiently solve the routing problems. Right now they are using a brute force approach, but with machine learning we can predict in advance how good certain hyperparameters are, and use a smarter approach to tune them. PostNL is interested in the best way to do this.
Goals of the project:
Develop and test an efficient parameter tuning framework for a vehicle routing solver.
Deliverables:
- A simplified version of PostNL’s vehicle routing solver that works on open data
- A parameter tuning algorithm for this solver
- Empirical or theoretical proof that it is more efficient than the current parameter tuning procedure
Essential student knowledge:
- Python
- Machine learning
- optimization
More information: escf@tue.nl
