Two-stage heuristic algorithm for a new variant of the multi-compartment vehicle routing problem with stochastic demands

Juan Carlos Gonçalves-Dosantos,Laura Davila-Pena,Balbina Casas-Méndez

2024-10-23

Abstract:This paper presents a model for a vehicle routing problem in which customer demands are stochastic and vehicles are divided into compartments. The problem is motivated by the needs of certain agricultural cooperatives that produce various types of livestock food. The vehicles and their compartments have different capacities, and each compartment can only contain one type of feed. Additionally, certain farms can only be accessed by specific vehicles, and there may be urgency constraints. To solve the problem, a two-step heuristic algorithm is proposed. First, a constructive heuristic is applied, followed by an improvement phase based on iterated tabu search. The designed algorithm is tested on several instances, including an analysis of real-world datasets where the results are compared with those provided by the model. Furthermore, multiple benchmark instances are created for this problem and an extensive simulation study is conducted. Results are presented for different model parameters, and it is shown that, despite the problems' complexity, the algorithm performs efficiently. Finally, the proposed heuristic is compared to existing solution algorithms for similar problems using benchmark instances from the literature, achieving competitive results.

Optimization and Control

What problem does this paper attempt to address?

### What problems does this paper attempt to solve?

This paper aims to solve a new variant of the multi - compartment vehicle routing problem (MC - VRP), in which customer demands are stochastic (stochastic demands). Specifically, in response to the demands of some agricultural cooperatives, the paper proposes a two - stage heuristic algorithm to optimize feed distribution routes. These cooperatives produce multiple types of livestock feed, and each type of feed needs to be transported in different compartments to avoid mixing.

#### Main problem description:

1. **Multi - compartment vehicles**: Each vehicle is divided into multiple independent compartments, and each compartment can only load one type of feed, and each compartment has its capacity limit.
2. **Stochastic demands**: Customer demands are uncertain and stochastic. This means that the actual demand may be different from the planned demand, thus affecting the design of the distribution route.
3. **Urgent orders**: Some customers' orders have time - window constraints (time window constraints), that is, they must be delivered within a specific time.
4. **Access restrictions**: Some farms can only be visited by specific types of vehicles.
5. **Maximizing load and minimizing cost**: The cooperatives not only hope to minimize transportation costs, but also hope to maximize the load of each transportation.

#### Model characteristics:

- **NP - hard complexity**: Determining the optimal route is an NP - hard problem, so effective heuristic algorithms are required to solve it.
- **Heuristic algorithm**: The paper proposes a two - stage heuristic algorithm:
- Stage 1: Use a constructive heuristic algorithm to generate an initial solution.
- Stage 2: Improve based on iterated tabu search to avoid local optimal solutions and improve the quality of the solution.

#### Research background:

Traditional vehicle routing problems (VRP) usually assume that customer demands are deterministic, but in the real world, demands are often stochastic. In addition, the application of multi - compartment vehicles also increases the complexity of the problem. This paper combines these two aspects, proposes a stochastic - demand model of the multi - compartment vehicle routing problem (MC - VRPSD), and applies it in the actual scenario of agricultural cooperatives.

By solving these problems, this research aims to provide an efficient tool for agricultural cooperatives to plan future distribution routes, reduce operating costs, improve distribution efficiency, and prevent system overload.