Journal
COMPUTERS & INDUSTRIAL ENGINEERING
Volume 169, Issue -, Pages -Publisher
PERGAMON-ELSEVIER SCIENCE LTD
DOI: 10.1016/j.cie.2022.108282
Keywords
Pickup and delivery problem; Machine learning; Evolutionary neural network; Scalable information; Mixed-integer linear programming
Funding
- Basic Science Research Program through the National Research Foundation of Korea (NRF) [NRF-2021R1C1C1003433]
- Ministry of Education
Ask authors/readers for more resources
In this paper, a machine learning approach called ENSIGHT is proposed for solving the pickup and delivery problem with time windows and heterogenous vehicles. Experimental results show that ENSIGHT outperforms other machine learning techniques and current dispatching rules, and is effective in learning scalable heuristics.
In this paper, we address the pickup and delivery problem with time windows (PDP-TW) and heterogenous vehicles for minimisation of total tardiness by learning heuristics from a given set of solutions. In order to extract scalable heuristics from optimal or best feasible solutions, we propose a machine-learning (ML)-based approach called ENSIGHT (Evolutionary Neural network with Scalable Information for Generation of Heuristics for Transportation). ENSIGHT consists of three phases: solution generation, interpretation of solutions, and improvement of heuristics by an evolutionary neural network (ENN). First, a set of optimal or best feasible solutions for the training set of problem instances is acquired by using the proposed mathematical model. Second, as for the process interpreting those solutions, an approach for transforming them into training data by way of scalable input attributes as well as output discretisation is followed. Third, the ENN improves the learned heuristics by an evolutionary parameter optimisation process for minimization of total tardiness. To verify the performance of the proposed ENSIGHT, we conducted experiments and the results of which showed that it outperforms other ML techniques and the current dispatching rules (DRs). Moreover, the approach was demonstrated to be effective in learning scalable heuristics based on combined scalable inputs and discretisation as well as an evolutionary improvement process.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available