svrp-bench

# 🚚 SVRPBench

SVRPBench is an open and extensible benchmark for the Stochastic Vehicle Routing Problem (SVRP). It includes 500+ instances spanning small to large scales (10–1000 customers), designed to evaluate algorithms under realistic urban logistics conditions with uncertainty and operational constraints.

## 📌 Overview

Existing SVRP benchmarks often assume simplified, static environments, ignoring core elements of real-world routing such as time-dependent travel delays, uncertain customer availability, and dynamic disruptions. Our benchmark addresses these limitations by simulating urban logistics conditions with high fidelity:

- Travel times vary based on time-of-day traffic patterns, log-normally distributed delays, and probabilistic accident occurrences
- Customer time windows are sampled differently for residential and commercial clients using empirically grounded temporal distributions
- A systematic dataset generation pipeline that produces diverse, constraint-rich instances including multi-depot, multi-vehicle, and capacity-constrained scenarios

## 📦 Dataset Components

The dataset includes various problem instances:
- Problem sizes: 10, 20, 50, 100, 200, 500, 1000 customers
- Variants: CVRP (Capacitated VRP), TWCVRP (Time Window Constrained VRP)
- Configurations: Single/Multi-depot, Single/Multi-vehicle

Each instance includes:
- Customer locations
- Demand volumes
- Time window constraints
- Vehicle capacity limits
- Depot coordinates

## 🧪 Supported Algorithms

The benchmark includes implementations of several algorithms:
- OR-tools (Google's Operations Research tools)
- ACO (Ant Colony Optimization)
- Tabu Search
- Nearest Neighbor with 2-opt local search
- Reinforcement Learning models

## 📊 Benchmarking Results

Results compare algorithm performance across different problem sizes:

| Model | CVRP10 | CVRP20 | CVRP50 | CVRP100 | CVRP200 | CVRP500 | CVRP1000 |
|----------|--------|--------|--------|---------|---------|---------|----------|
| OR-tools | 1.4284 | 1.6624 | 1.3793 | 1.1513 | 1.0466 | 0.8642 | - |
| ACO | 1.5763 | 1.7843 | 1.5120 | 1.2998 | 1.1752 | 1.0371 | 0.9254 |
| Tabu | 1.4981 | 1.7102 | 1.4578 | 1.2214 | 1.1032 | 0.9723 | 0.8735 |
| NN+2opt | 1.6832 | 1.8976 | 1.6283 | 1.3844 | 1.2627 | 1.1247 | 1.0123 |

## 🛠️ Usage

```python
# Example of loading a dataset
from datasets import load_dataset
ds = load_dataset("MBZUAI/svrp-bench", split="test")
ds[0]
```

### Sample
```json
{'appear_times': [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
'demands': [0, 33, 52, 35, 85, 77, 68, 17, 61, 32, 23],
'file_name': 'cvrp_10_multi_depot_multi_vehicule_capacities.npz',
'instance_id': 0,
'locations': [[523, 497],
[394, 344],
[536, 599],
[341, 412],
[734, 652],
[492, 569],
[491, 238],
[419, 787],
[688, 422],
[708, 490],
[431, 454]],
'num_vehicles': 13,
'subset_name': 'cvrp_10_multi_depot_multi_vehicule_capacities',
'vehicle_capacities': [40.0]}
```

## 🔑 Features

- Comprehensive evaluation framework for VRP algorithms
- Realistic travel time modeling with time-dependent patterns
- Time window constraints based on empirical distributions
- Support for multi-depot and multi-vehicle scenarios
- Visualization tools for solution analysis
- Extensible architecture for adding new algorithms

## 📚 Citation

If you use this benchmark in your research, please cite:

```bibtex
@misc{svrbench2025,
author = {Heakl, Ahmed and Shaaban, Yahia Salaheldin and Takáč, Martin and Lahlou, Salem and Iklassov, Zangir},
title = {SVRPBench: A Benchmark for Stochastic Vehicle Routing Problems},
year = {2025},
publisher = {GitHub},
journal = {GitHub repository},
howpublished = {\url{https://github.com/yehias21/vrp-benchmarks}}
}
```

## 📄 License

This project is licensed under the MIT License.

- 📄 Paper: [Arxiv](https://arxiv.org/abs/2505.21887)

# 🚚 SVRPBench

SVRPBench是一个开源且可扩展的随机车辆路径问题（Stochastic Vehicle Routing Problem, SVRP）基准测试集。它包含500余个覆盖小至大规模（10~1000个客户节点）的测试实例，旨在评估具备不确定性与实际运营约束的城市物流场景下的算法性能。

## 📌 概述

现有SVRP基准测试集通常采用简化的静态环境假设，忽略了现实路径规划中的核心要素，如随时间变化的出行延误、不确定的客户可服务时段以及动态中断事件。本基准测试集通过高保真度的城市物流场景模拟解决了这些局限：
- 出行时长基于时段交通模式、对数正态分布的延误以及概率性事故发生情况动态变化
- 针对住宅与商业客户，采用基于经验的时间分布分别采样其服务时间窗约束
- 构建了系统化的数据集生成流水线，可生成包含多仓库、多车辆及容量约束等丰富约束的多样化实例

## 📦 数据集组成

本数据集包含各类问题实例：
- 问题规模：10、20、50、100、200、500、1000个客户节点
- 问题变体：带容量约束的车辆路径问题（Capacitated VRP, CVRP）、带时间窗约束的车辆路径问题（Time Window Constrained VRP, TWCVRP）
- 配置类型：单/多仓库、单/多车辆

每个实例包含以下信息：
- 客户节点坐标
- 货物需求量
- 时间窗约束
- 车辆容量限制
- 仓库坐标

## 🧪 支持的算法

本基准测试集集成了多款算法的实现：
- OR-tools（谷歌运筹学工具集）
- 蚁群优化算法（Ant Colony Optimization, ACO）
- 禁忌搜索算法
- 结合2-opt局部搜索的最近邻算法
- 强化学习模型

## 📊 基准测试结果

下表对比了不同算法在各规模问题上的性能表现：

| 算法模型 | CVRP10 | CVRP20 | CVRP50 | CVRP100 | CVRP200 | CVRP500 | CVRP1000 |
|----------|--------|--------|--------|---------|---------|---------|----------|
| OR-tools | 1.4284 | 1.6624 | 1.3793 | 1.1513 | 1.0466 | 0.8642 | - |
| ACO | 1.5763 | 1.7843 | 1.5120 | 1.2998 | 1.1752 | 1.0371 | 0.9254 |
| Tabu | 1.4981 | 1.7102 | 1.4578 | 1.2214 | 1.1032 | 0.9723 | 0.8735 |
| NN+2opt | 1.6832 | 1.8976 | 1.6283 | 1.3844 | 1.2627 | 1.1247 | 1.0123 |

## 🛠️ 使用方法

python
# 数据集加载示例
from datasets import load_dataset
ds = load_dataset("MBZUAI/svrp-bench", split="test")
ds[0]

### 样本示例
json
{'客户可服务时刻': [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
'货物需求量': [0, 33, 52, 35, 85, 77, 68, 17, 61, 32, 23],
'文件名': 'cvrp_10_multi_depot_multi_vehicule_capacities.npz',
'实例编号': 0,
'客户坐标': [[523, 497],
[394, 344],
[536, 599],
[341, 412],
[734, 652],
[492, 569],
[491, 238],
[419, 787],
[688, 422],
[708, 490],
[431, 454]],
'车辆总数': 13,
'子集名称': 'cvrp_10_multi_depot_multi_vehicule_capacities',
'车辆容量限制': [40.0]}

## 🔑 核心特性

- 面向车辆路径问题算法的全面评估框架
- 基于时段特征的真实化出行时长建模
- 基于经验分布的时间窗约束设置
- 支持多仓库与多车辆调度场景
- 集成解决方案可视化分析工具
- 可扩展的架构设计，支持新增算法集成

## 📚 引用声明

若您在研究中使用本基准测试集，请引用以下文献：

bibtex
@misc{svrbench2025,
author = {Heakl, Ahmed and Shaaban, Yahia Salaheldin and Takáč, Martin and Lahlou, Salem and Iklassov, Zangir},
title = {SVRPBench: A Benchmark for Stochastic Vehicle Routing Problems},
year = {2025},
publisher = {GitHub},
journal = {GitHub仓库},
howpublished = {url{https://github.com/yehias21/vrp-benchmarks}}
}

## 📄 许可协议

本项目采用MIT许可协议进行开源。

- 📄 相关论文: [Arxiv](https://arxiv.org/abs/2505.21887)

搜集汇总

背景与挑战

背景概述

svrp-bench是一个专注于随机车辆路径问题（SVRP）的基准测试数据集，包含500多个不同规模的实例，旨在模拟现实城市物流条件。数据集支持多种算法评估，并提供了详细的实例信息和算法性能比较。