A Matheuristic for the Electric Vehicle Routing Problem with Time Windows

Maurizio Bruglieri,Ferdinando Pezzella,Ornella Pisacane,Stefano Suraci

DOI: https://doi.org/10.48550/arXiv.1506.00211

2015-05-31

Abstract:The main goal of this paper is to time-effectively route and schedule a fleet of Electric Vehicles (EVs) on a road network in order to serve a set of customers. In particular, we aim to propose an optimized route planning by exploiting the advantages of these vehicles. Nowadays, in fact, electromobility plays a key role for reducing the harmful emissions due, instead, to the use of traditional vehicles. The starting point of this research is represented by the fact that the advanced recent technologies for the EVs allow also partially recharging their batteries. In this work, an Electric Vehicle Routing Problem with Time Windows (E-VRPTW) is addressed from a time effective point of view under the assumption that partial recharges are also allowed. For this purpose, the E-VRPTW is mathematically formulated as a Mixed Integer Linear Program in which both the total number of EVs used and the total time spent by them outside the depot are minimized. Due to the NP-hardness of the problem, a Variable Neighborhood Search Branching (VNSB) matheuristic is also designed for determining good quality solutions in reasonable computational times. Numerical results carried out on some benchmark instances taken from the literature provide useful insights regarding both the solution quality of the proposed formulation, compared to a previous one, and the performances of the VNSB.

Optimization and Control

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to effectively plan routes for a group of customers for electric vehicles (EVs) within a given time window, in order to reduce harmful emissions and optimize vehicle usage and time cost. Specifically, the paper focuses on the Electric Vehicle Routing Problem with Time Windows (E - VRPTW), and its objective is to minimize the total time cost while satisfying the following conditions:

1. **Service Time Windows**: Each customer has a specific service time window, that is, the time range for the start and end of service.
2. **Battery Capacity Limitations**: Electric vehicles need to consider battery power during driving and may need to be partially or fully charged at intermediate charging stations.
3. **Vehicle Capacity Limitations**: The load capacity of each vehicle is limited and cannot exceed its maximum capacity.

To achieve these goals, the paper proposes a Mixed Integer Linear Programming (MILP) model. This model not only considers traditional route - planning constraints (such as time windows, vehicle capacity, etc.), but also introduces the concepts of battery capacity and partial charging. In addition, due to the NP - hard nature of the problem, the paper also designs a heuristic algorithm based on Variable Neighborhood Search Branching (VNSB) to find high - quality solutions within a reasonable time.

### Main Contributions
1. **Partial Charging**: Different from the common assumption in previous studies that vehicles are fully charged at charging stations, this paper allows vehicles to be partially charged, which can reduce the total charging time and improve the flexibility of route planning.
2. **Optimization Objectives**: Besides minimizing the number of vehicles used, it also minimizes the time cost of vehicles outside, including driving time, waiting time and charging time.
3. **Model Improvement**: A new MILP model is proposed. By cloning charging stations and terminals, the model is simplified, the number of decision variables and constraints is reduced, and the solution efficiency of the model is improved.
4. **Heuristic Algorithm**: A VNSB algorithm is designed, which can find high - quality solutions within a reasonable time and is suitable for large - scale instances.

### Mathematical Model
The MILP model in the paper includes the following main parts:

- **Objective Function**:
\[
\min \sum_{i \in D_0, j \in N \cup F} x_{ij}+\frac{1}{l_0 |N| |D_0|} \sum_{i \in D_0, j \in D_0'}(\tau_j - \tau_i)
\]
where the first term represents the number of vehicles used, and the second term represents the time cost of vehicles outside, after normalization.

- **Constraint Conditions**:
- Each customer must be visited once:
\[
\sum_{j \in V''\setminus \{i\}} x_{ij} = 1\quad \forall i \in N
\]
- Each clone point of the charging station is used at most once:
\[
\sum_{j \in V''\setminus \{i\}} x_{ij}\leq 1\quad \forall i \in F
\]
- Each clone point of the starting point is used at most once:
\[
\sum_{j \in N \cup F} x_{ij}\leq 1\quad \forall i \in D_0
\]
- Each clone point of the ending point is used at most once:
\[
\sum_{i \in N \cup F} x_{ij}\leq 1\quad \forall j \in D_0'
\]
- Flow conservation constraint:
\[
\sum_{i \in V''\setmi