Hybrid ML-ACO Route Optimisation for Last-Mile Delivery
A research project combining Ant Colony Optimisation with machine learning guidance models to improve vehicle routing decisions for last-mile delivery networks.
Hybrid ML-ACO Full Document
Document Preview
This document explains the research problem, routing approach, hybrid ML-ACO comparison, model findings and project outcome.
- Vehicle Routing Problem background
- ACO with KNN, ANN and LSTM guidance
- Route optimisation evaluation
- Key findings and outcome
LaDe delivery trajectory records referenced in the study.
Best performing hybrid model in the comparison.
Best route distance achieved under the simplified test setting.
Problem
The project addresses inefficiency in last-mile delivery, where route planning becomes a Vehicle Routing Problem. Traditional ACO is robust but can be slow and may converge to suboptimal routes in complex urban delivery networks.
Approach
- Built a hybrid optimisation framework where machine learning guidance influences ant route choices.
- Compared ANN-ACO, KNN-ACO and LSTM-ACO approaches.
- Analysed spatial delivery patterns, clustering behaviour and route convergence.
- Used GPU-oriented simulation concepts for scalable route evaluation.
Outcome
KNN-ACO produced the strongest results by using local spatial memory to guide route selection. The project showed that simpler memory-based learning can outperform more complex sequence models for static spatial routing problems.