Large time-series data for learning

Question

I am looking for large time-series data for learning forecasting methods. Data should preferably be at hourly or minute level. Domain doesn't matter much but multiple types of data (financial, IOT, economic etc.) would be helpful.

---

A few previous posts have given relevant information.

1. See This post pointing to a resource at the UCI Data repository.

2. Another post is not asking for time series, but the first answer points to a resource with many time series.

3. You can get lots of weather data at ECMWF

Answer 1

Consider electricity consumption data, which is an interesting for forecasting due to having various types of seasonality (day of week, month, season), temperature, weather and daylight effects, holidays, etc.

Swissgrid, the Swiss Transmisssion System Operator (TSO), publishes a quarter-hourly timeseries for indidvidual regions as well as the entire country. The files are machine readable (Excel) and are divided into years, going back to 2011. The data is updated at the end of the month, for the previous month.

https://www.swissgrid.ch/swissgrid/en/home/experts/topics/energy_data_ch.html

See column B of the sheet named "Zeitreihen0h15"

Total energy consumed by end users in the Swiss controlblock"

Zeitstempel*    kWh**
01.01.2016 00:15    1525210.59
01.01.2016 00:30    1507491.133
01.01.2016 00:45    1539664.783
01.01.2016 01:00    1518853.125
01.01.2016 01:15    1539265.905

* Timestamp

** Kilowatt-hour

There are many other columns for regional and national consumption and production, as well as import/export, etc. Production is a little harder to forecast because it depends on market prices in Switzerland and neighboring countries (for example, if there is lots of solar and wind power in Germany, the hourly price can go quite low, even negative, and this would likely mean you'd see Switzerland importing from Germany and very little Swiss production.)

---

p.s. I used similar data from ENTSO-E, which is monthly-aggregated and for all countries in the EU market (including Switzerland).

my tweet

---

For yearly aggregated data for Switzerland, going back to 1910, check out worldenergy.ch

my tweet

Answer 2

I'm not sure what qualifies as "large" for you. If you're looking for a sizable collections of time series data, check out the M-series data which have been used in various forecast competitions. The latest one is M3, which have time series of different periodicity.

https://forecasters.org/resources/time-series-data/m-competition/

If you use R, the M datasets are built into the Mcomp package. To get it just type

install.packages("Mcomp")

Answer 3

The Monash Time Series Forecasting Repository contains multiple different time series datasets from various domains. These have been collected from forecasting competitions or other previous forecasting use (in particular, the M competition data recommended by horaceT is included), so you should also be able to learn something from previous work that used a particular dataset.

An introduction and description of the repository, along with more information about the datasets, is given by Godahewa et al. (2021).