Vehicle Logistics Network Optimisation

Spatial coverage

The coverage problem was treated as a candidate-site selection task. A simulated-annealing search explored combinations, while a greedy rule selected useful hubs from the candidate pool. The 200 km radius bounded the feasible service area and kept the model interpretable for operations stakeholders.

• The dataset contained the coordinates of 764 dealerships.
• Seven secondary hubs were selected, including Huzhou and Nanjing.
• The final set reached full coverage under the modelled radius.

Freight-rate prediction

Freight-rate data was scraped with EasySpider. ChatGLM extracted keyword features from the semi-structured listings, and a Random Forest model was tuned with random search. This handled nonlinear relationships between route characteristics and price better than a hand-written formula.

Modal economics

The road-versus-rail comparison combined distance, per-kilometre price and an explicit delay penalty. The point was not that rail is always cheaper, but that the break-even depends on shipment value, time sensitivity and the cost of delay.