Ride-Hailing Synthetic Dataset - 数据集

14,025,600

Time-stamped Records

4 Years

Time Span

200

Spatial Areas

30 min

Granularity

01Overview

Dataset Description

The Ride-Hailing dataset is synthetically generated based on Didi's ride-hailing scenarios and does not involve any real user personal information, behavioral traces, or other operational data. It simulates four consecutive years of data for 200 representative spatial areas at a 30-minute granularity. The dataset contains 14,025,600 time-stamped records indexed by area and time, covering nine endogenous market indicators together with exogenous covariates that capture weather, holiday, and large-scale event effects.

02Why It Matters

Research Challenges

Complex exogenous effects

Weather, holidays, and large-scale events reshape demand through complex, temporally structured effects.

Heterogeneous multi-area forecasting

Demand patterns and exogenous sensitivities vary substantially across 200 areas.

Long-horizon strategic planning

Long-horizon forecasts require balancing overall accuracy, trend fidelity, and short-term reliability, all of which are critical for downstream strategic planning.

03Structure

Data Schema

Field	Type	Description
`day_index`	Integer	Day index relative to the earliest observation date (0-1460).
`time`	Time	Time of day.
`area_id`	Integer	Spatial area identifier.
`endo_1`	Float	Anonymized endogenous target 1.
`endo_2`	Integer	Anonymized endogenous target 2.
`endo_3`	Float	Anonymized endogenous target 3.
`endo_4`	Float	Anonymized endogenous target 4.
`endo_5`	Integer	Anonymized endogenous target 5.
`endo_6`	Float	Anonymized endogenous target 6.
`endo_7`	Float	Anonymized endogenous target 7.
`endo_8`	Integer	Anonymized endogenous target 8.
`endo_9`	Float	Anonymized endogenous target 9.
`weather_factor`	Float	Weather disturbance impact factor.
`large_scale_event_1`	Binary	First large-scale event impact indicator.
`large_scale_event_2`	Binary	Second large-scale event impact indicator.
`half_hour_of_day`	Integer	Half hour slot index (0-47).
`day_of_week`	Integer	Day of week index (0-6).
`public_holiday`	Integer	Public holiday category code (0-9).
`traditional_festival`	Integer	Traditional festival category code (0-9).
`western_festival`	Integer	Western festival category code (0-14).

04Preview

Data Samples

Below is a sample of 10 rows from the Ride-Hailing dataset.(scroll horizontally →)

day_index	time	area_id	endo_1	endo_2	endo_3	endo_4	endo_5	endo_6	endo_7	endo_8	endo_9	weather_factor	large_scale_event_1	large_scale_event_2	half_hour_of_day	day_of_week	public_holiday	traditional_festival	western_festival
0	00:00:00	0	50.796	77	44.721	51.138	50	34.392	48.115	95	41.592	0.0	0	0	0	4	1	0	0
6	08:00:00	0	60.25	74	49.896	61.218	48	54.37	51.552	82	51.183	0.75	0	0	16	3	0	0	0
19	12:00:00	0	70.45	31	51.973	61.389	37	71.376	52.27	60	54.322	0.0	0	0	24	2	0	1	0
66	12:00:00	0	67.946	38	52.845	58.877	55	59.919	48.599	77	53.88	10.5	0	0	24	0	0	0	1
169	12:00:00	1	52.198	79	38.749	44.754	69	41.575	39.452	82	33.481	0.0	0	1	24	5	0	0	0
358	18:00:00	0	103.282	115	48.839	62.783	142	112.707	46.658	114	50.242	0.0	0	0	36	5	0	0	12
722	12:00:00	0	57.693	23	51.625	48.005	32	65.161	50.042	38	53.568	0.0	1	0	24	5	0	0	11
1003	19:30:00	0	80.226	97	50.011	52.158	65	56.617	47.521	97	49.032	2.25	0	0	39	6	8	0	0
1135	03:00:00	0	6.658	82	26.9	13.741	29	3.913	29.683	29	19.608	0.0	0	0	6	5	2	0	0
1460	23:30:00	199	43.888	62	33.311	46.119	57	42.517	36.059	46	34.318	0.0	0	0	47	1	0	0	0

05Reference

Citation

ridebench2027.bib

@misc{dibench2027,
    title = {DiBench: A Large-Scale Exogenous-Aware Benchmark for Ride-Hailing Time Series Forecasting},
    author = {Lin, Shengsheng and Hu, Jing and Hu, Zhengyang and Sun, Jiazheng and Cao, Zichun and Sun, Siwei and Zou, Zhichao and Li, Dongdong and Hu, Xinyi and Lin, Weiwei},
    year = {2027},
    note = {Manuscript submitted to the International Conference on Learning Representations (ICLR 2027)}
  }

Start Your Research

Ready to get started?

Apply now to access the dataset and accelerate your research.

Apply Now→