cps-robustness-benchmark
收藏资源简介:
cps-robustness-benchmark是由汉堡赫尔穆特·施密特大学人工智能学院创建的一个基准测试数据集,旨在评估和比较深度学习模型在工业 cyber-physical systems (CPS) 时间序列预测任务中的鲁棒性。该数据集包含了多个真实世界CPS场景的时间序列数据,并针对不同的深度学习架构进行了鲁棒性测试。数据集通过模拟现实世界中的扰动,如传感器漂移、噪声和 irregular sampling 等,来评估模型在遭遇错误或不可预见输入时的性能。
The CPS-Robustness-Benchmark is a benchmark dataset developed by the Faculty of Artificial Intelligence at Helmut Schmidt University Hamburg. It is designed to evaluate and compare the robustness of deep learning models for industrial cyber-physical systems (CPS) time series forecasting tasks. This dataset contains time-series data from multiple real-world CPS scenarios, and is employed to perform robustness tests on various deep learning architectures. Specifically, it simulates real-world perturbations including sensor drift, noise, and irregular sampling to assess model performance when confronted with erroneous or unforeseen inputs.
AI Robustness Benchmark for Cyber-Physical Systems 数据集概述
数据集基本信息
- 名称: AI Robustness Benchmark for Cyber-Physical Systems
- 状态: 开发中(work in progress)
数据集结构
- 数据目录:
data/processed(存储最终的规范化数据集) - 其他相关目录:
logs: 存储训练和序列化的模型、模型预测或模型摘要models: 存储所有模型的Lightning模块notebooks: 存储Jupyter笔记本visualizations: 存储生成的图形和报告用图
数据生成与处理
- 数据生成: 通过运行
notebooks/1-generate-data.ipynb生成数据,并提供数据集的可视化分析
依赖与环境
- Python版本: 3.11
- 主要依赖:
- PyTorch(需指定CUDA版本,例如11.3)
- 通过
environment.yml配置环境
容器化支持
-
Docker镜像构建: sh docker build . --platform=linux/amd64 -t gitlab.kiss.space.unibw-hamburg.de:4567/kiss/robust-ai-validation:v31
-
镜像推送: sh docker push gitlab.kiss.space.unibw-hamburg.de:4567/kiss/robust-ai-validation:v31 --tls-varify=false




