PowerGraph
收藏资源简介：
我们创建了一个图数据集，用于模拟电力网络中的连锁故障事件，这是电力网络中停电的主要原因。为了为不同的电力网络生成一个全面的数据集，我们使用了一个基于物理的连锁故障模型，称为Cascades。该模型模拟了IEEE24、IEEE39、IEEE118和UK电力网络中故障的传播。数据集包括大量的电力网络状态，代表停电前的运行条件，并与初始触发停电（一个或多个故障元件）相关联。每个电力网络状态被表示为一个图，并根据基于物理的模型的结果分配一个图级标签。该数据集设计用于各种图级任务，如多类分类、二分类和回归。电力网络的元素是总线和分支，总线包括负载和发电机，代表图的节点，而分支包括输电线路和变压器，代表图的边。我们为每个节点提供三个特征：净有功功率、净视在功率和电压幅值。而每个边的特征是四个：有功功率流、无功功率流、线路电抗和线路额定值。
We have created a graph dataset to simulate cascading failure events in power networks, which are the primary cause of blackouts in power grids. To generate a comprehensive dataset for different power networks, we employed a physics-based cascading failure model known as Cascades. This model simulates the propagation of failures in IEEE24, IEEE39, IEEE118, and UK power networks. The dataset includes a vast number of power network states, representing operational conditions prior to blackouts, and is associated with the initial triggers of blackouts (one or more faulty components). Each power network state is represented as a graph and is assigned a graph-level label based on the results of the physics-based model. This dataset is designed for various graph-level tasks such as multi-class classification, binary classification, and regression. The elements of the power networks are buses and branches, where buses include loads and generators, representing the nodes of the graph, and branches include transmission lines and transformers, representing the edges of the graph. We provide three features for each node: net active power, net apparent power, and voltage magnitude. For each edge, the features are four: active power flow, reactive power flow, line reactance, and line rating.
数据集概述
数据集名称
- IEEE-24 (
ieee24) - IEEE-39 (
ieee39) - IEEE-118 (
ieee118) - UK (
uk)
数据集描述
- IEEE-24: IEEE-24 电力网络数据集。
- IEEE-39: IEEE-39 电力网络数据集。
- IEEE-118: IEEE-118 电力网络数据集。
- UK: UK 电力网络数据集。
数据集内容
- 文件结构:
blist.mat: 分支列表，也称为边序或边索引。of_bi.mat: 二分类标签。of_reg.mat: 回归标签。of_mc.mat: 多分类标签。Bf.mat: 节点特征矩阵。Ef.mat: 边特征矩阵。exp.mat: 地面真值解释。
数据集生成
- 使用物理基础的级联故障模型“Cascades”模拟IEEE24、IEEE39、IEEE118和UK电力网络中的故障传播。
- 输出为最终需求未服务（DNS）。
- 数据集包含大量电力网络状态，代表停电前的操作条件，并与初始触发停电（一个或多个故障元素）相关联。
节点和边特征
- 节点特征: 净有功功率、净视在功率和电压幅值。
- 边特征: 有功功率流、无功功率流、线路电抗和线路额定值。
数据集用途
- 用于图级任务，如多分类、二分类和回归。
解释性方法
- 非生成性解释器:
- Occlusion
- SA
- Grad-CAM
- Integrated Gradients
- GNNExplainer
- SubgraphX
- PGM-Explainer
- 生成性解释器:
- RCExplainer
- GSAT
- DiffExplainer
使用命令
bash python3 code/main.py --dataset_name [dataset-name] --model_name [gnn-model] --explainer_name [explainer-name]
- 参数:
--model_name: transformer / gin / gat--datatype: multiclass--dataset_name: uk_mc / ieee24_mc / ieee39_mc--explainer_name: random / sa / ig / gradcam / occlusion / basic_gnnexplainer / gnnexplainer / subgraphx / pgmexplainer / rcexplainer / gsat / diffexplainer
