PowerGraph
收藏资源简介:
PowerGraph数据集由苏黎世联邦理工学院的可靠性及风险工程实验室开发,专注于电力网络中级联故障的模拟与分析。该数据集包含144000条记录,涵盖了多个电力系统的状态,用于训练图神经网络模型进行多类分类、二分类和回归任务。数据集通过基于物理的级联故障模型生成,确保了操作和环境条件的普遍性,并通过模拟多种故障场景来增强数据集的多样性。此外,PowerGraph数据集还支持图神经网络解释方法的基准测试,通过提供边缘级解释的地面真值来促进模型的透明度和可解释性。该数据集的应用领域广泛,从化学到生物学,其中系统及过程可用图表描述,对于提升图级任务和解释性的图神经网络模型发展具有重要意义。
The PowerGraph dataset was developed by the Laboratory for Reliability and Risk Engineering at ETH Zurich, focusing on the simulation and analysis of cascading failures in power grids. Comprising 144,000 records, this dataset covers the states of multiple power systems and is used to train graph neural network (GNN) models for multi-class classification, binary classification, and regression tasks. Generated via a physics-based cascading failure model, the dataset guarantees the generality of operational and environmental conditions, and enhances its diversity by simulating a variety of fault scenarios. Furthermore, the PowerGraph dataset also supports benchmarking of GNN interpretation methods, facilitating model transparency and interpretability by providing ground truth for edge-level explanations. With a wide range of application domains spanning from chemistry to biology, where systems and processes can be represented as graphs, this dataset holds significant importance for advancing the development of graph-level tasks and interpretable graph neural network models.

- 1PowerGraph: A power grid benchmark dataset for graph neural networks苏黎世联邦理工学院 · 2024年



