witgaw/PEMS-BAY
收藏资源简介:
PEMS-BAY交通数据集包含用于时间序列预测任务的交通流量数据,常与图神经网络(GNN)结合使用,特别是用于扩散卷积循环神经网络(DCRNN)模型。数据集以Parquet文件格式存储,有效支持加载数据和分析。数据集被分为训练集(70%)、验证集(10%)和测试集(20%),并按时间顺序划分以保持数据的时序特性。每个时间点的数据包括所有传感器的信息,并保留了图结构。数据模式包括传感器节点标识、不同时间偏移量和维度的输入特征以及未来时间步的目标值。数据集包含大约34K个时间序列样本(METR-LA)和52K个(PEMS-BAY),记录数分别为7M和17M,每个样本包含207个(METR-LA)和325个(PEMS-BAY)传感器的数据。时间分辨率为5分钟,预测范围为1小时(12个时间步)。
The PEMS-BAY Traffic Dataset contains traffic flow data for time series forecasting tasks, commonly used with Graph Neural Networks (GNNs) especially with the Diffusion Convolutional Recurrent Neural Network (DCRNN) model. The dataset is stored in Parquet file format for efficient loading and analysis. It is split into training (70%), validation (10%), and test (20%) sets, with temporal splitting to maintain chronological order. Each time step includes data from all sensors, preserving the graph structure. The data schema includes a sensor/node identifier, input features at different time offsets and dimensions, and target values for future time steps. The dataset consists of approximately 34K time series samples (METR-LA) and 52K (PEMS-BAY), with 7M and 17M records respectively, and includes data from 207 (METR-LA) and 325 (PEMS-BAY) sensors per sample. The temporal resolution is 5-minute intervals, and the prediction horizon is 1 hour (12 time steps).




