PEMS04数据集

PEMS04数据集是加利福尼亚州交通部门通过其性能测量系统（PEMS）收集的2018年1月1日至2018年2月28日期间，加州某段高速公路的交通流量数据。具体而言，PEMS利用307个传感器以30秒为时间间隔进行采集，之后汇总为间隔为5分钟的样本。该数据集包含了307个节点和340条边的交通网络，时间步数量为16992，涵盖了丰富的时空交通信息。本项目采用PEMS04公开数据集作为交通流量预测的验证数据，并结合多尺度多智能体动态超图卷积网络（MSDHGCN）模型，对数据进行了多尺度分解和动态超图构建，以更好地捕捉交通数据中的高阶时空相关性。通过对原始数据的调整和处理，我们能够更有效地建模交通流量中的复杂动态系统，提升预测的准确性和鲁棒性。

The PEMS04 dataset comprises traffic flow data for a segment of a California highway, collected by the California Department of Transportation through its Performance Measurement System (PEMS) between January 1, 2018 and February 28, 2018. Specifically, PEMS employs 307 sensors to collect data at 30-second intervals, which is subsequently aggregated into 5-minute interval samples. This dataset contains a traffic network with 307 nodes and 340 edges, with a total of 16992 time steps, covering abundant spatiotemporal traffic information. This project adopts the publicly available PEMS04 dataset as validation data for traffic flow prediction, and combines it with the Multi-Scale Multi-Agent Dynamic Hypergraph Convolutional Network (MSDHGCN) model to conduct multi-scale decomposition and dynamic hypergraph construction on the data, thereby better capturing the high-order spatiotemporal correlations within traffic data. By adjusting and processing the raw data, we can more effectively model the complex dynamic system inherent in traffic flow, and enhance the accuracy and robustness of the prediction.

搜集汇总

背景与挑战

背景概述

PEMS04数据集是加利福尼亚州交通部门通过性能测量系统收集的2018年1月至2月期间高速公路交通流量数据，包含307个节点和340条边的网络结构，时间步为16992个，并以5分钟间隔汇总。该数据集被用于交通流量预测研究，通过多尺度分解和动态超图构建来捕捉高阶时空相关性。