THULab/m2oe2_timeseries

M2oE2 TimeSeries Data (TsFile)

Apache TsFile version of MuhaoGuo/M2oE2TimeSeriesData.

Overview

A collection of heterogeneous energy and weather time series across five sub-datasets: building energy metering, household power consumption, residential power usage, solar generation, and the Spanish electricity system. Each sub-dataset is converted to one or more TsFile tables.

• Tables: 8
• Rows: ~20.85 million total
• Sub-datasets: Building / Consumption / Residential / Solar / Spanish

Table	Source CSV	Description	Rows
building_train	Building/train.csv	Hourly meter readings (meter_reading) per building_id + meter	~20.2M
building_weather	Building/weather_train.csv	Hourly site weather per site_id	~139k
consumption_power	Consumption/powerconsumption.csv	Household power consumption zones + weather	~52k
residential_power_usage	Residential/power_usage_2016_to_2020.csv	Daily household kWh, tagged by notes	~36k
residential_weather	Residential/weather_2016_2020_daily.csv	Daily weather	~1.5k
solar_weather	Solar/solar_weather.csv	Minutely solar energy + weather	~197k
spanish_energy	Spanish/energy_dataset.csv	Hourly Spanish generation / price (28 series)	~35k
spanish_weather	Spanish/weather_features.csv	Hourly weather per city_name	~175k

Schema (TsFile structure)

• Time (INT64, milliseconds) — each table uses its own source time column (timestamp, Datetime, StartDate, Date, Time, time, or dt_iso), normalized to UTC.
• TAG columns are the low-cardinality device dimensions:
  • building_train: building_id, meter
  • building_weather: site_id
  • residential_power_usage: notes (weekday / weekend / COVID_lockdown / vacation)
  • spanish_weather: city_name
• FIELD columns are the numeric measurements (readings, weather, generation).

Column names with spaces or special characters (Value (kWh), Energy delta[Wh], SunlightTime/daylength, generation fossil gas, …) are renamed to underscore-separated forms.

Conversion notes

• Building/building_metadata.csv is a static building lookup table (no time column) and is therefore not converted to TsFile.
• Building/train.csv (~20M rows) is sharded into 11 TsFiles via row-group sizing.
• Spanish/weather_features.csv: weather_main, weather_description, and weather_icon are dropped as redundant text labels of the numeric weather_id. 3076 duplicate (city_name, Time) rows were removed.

Usage

Install the Apache TsFile Python SDK (pip install tsfile) and read a converted file:

from pathlib import Path
from tsfile import TsFileReader

path = Path("building_train_1.tsfile")
with TsFileReader(str(path)) as reader:
    schemas = reader.get_all_table_schemas()
    print("tables:", list(schemas))
    table_name = next(iter(schemas))
    table = schemas[table_name]
    columns = [column.get_column_name() for column in table.get_columns()]
    print("columns:", columns)
    field_names = [
        column.get_column_name()
        for column in table.get_columns()
        if column.get_column_name() not in {"Time", "time"}
    ]
    if field_names:
        with reader.query_table(table_name, field_names[:3], batch_size=1024) as result:
            batch = result.read_arrow_batch()
            if batch is not None:
                print(batch.to_pandas().head())

Source & license

• Original dataset: https://huggingface.co/datasets/MuhaoGuo/M2oE2TimeSeriesData
• Author / publisher: MuhaoGuo
• License: MIT