A Large Scale PeMS Traffic Speed Dataset

This repository is a traffic speed data set collected by California Freeway Performance Measurement System (PeMS). In this data set, it has 11160 time series corresponding to 11160 sensors/locations. In three data files (i.e., pems-4w.csv, pems-8w.csv, and pems-12w.csv), each time series is indeed a time series with 288 (5-minute) time points per day during the first 4 weeks (PeMS-4W), first 8 weeks (PeMS-8W), and first 12 weeks (PeMS-12W) of 2018, respectively. Note that the original data was downloaded from PeMS system with 30-second traffic speed records. Our data is a subset of file speed.h5 created by article Transfer Learning with Graph Neural Networks for Short-Term Highway Traffic Forecasting and its original link is available at here.

Our research work on this data set is:

Xinyu Chen, Yixian Chen, Nicolas Saunier, Lijun Sun (2021). Scalable low-rank tensor learning for spatiotemporal traffic data imputation. Transportation Research Part C: Emerging Technologies, 129: 103226.  [DOI] [preprint] [data] [Python code]