Truck-Logistics Network Optimisation

01 / Context

The problem

The network needed secondary distribution hubs that could cover dealership demand without overbuilding infrastructure. Freight rates were not readily structured, and the team also had to compare road-only transport with a road-and-rail alternative.

02 / Approach

What I did

• Led a four-person cross-disciplinary team and built a site-selection and cost model from the coordinates of 764 dealerships.
• Applied simulated annealing with a 200 km coverage radius and a greedy algorithm to select seven secondary hubs.
• Scraped freight-rate data with EasySpider, extracted keyword features with ChatGLM and trained a Random Forest freight-rate model.
• Built a road-versus-rail cost model using RMB 1.8/km by road and RMB 1.0/km by rail plus delay cost.

03 / Outcome

What came out of it

• Selected seven secondary distribution hubs while reaching full dealership coverage under the modelled radius.
• Quantified the potential saving available through multi-modal road-and-rail transport.
• Combined operations research, NLP-assisted feature extraction and supervised learning in one decision model.

04 / Artifacts

Reports & code