---
tags:
- tabular
- energy
- forecasting
- weather
- tsfile
- format:tsfile
- timeseries
pretty_name: REG-Forecasting_v2
size_categories:
- 100K<n<1M
---





# REG-Forecasting_v2 (TsFile)

Apache TsFile version of
[`wachawich/REG-Forecasting_v2`](https://huggingface.co/datasets/wachawich/REG-Forecasting_v2).

## Overview

Renewable Energy Generation (REG) forecasting data at **hourly** granularity, spanning
**2020-01-01 to 2025-11-19**. For each hour it records the generation (`value`) of two
generation types together with the shared meteorological and solar-geometry features.
Compared with the v1 dataset, this version adds extra columns such as `solar_count`,
`wind_turbine_count`, and cyclic time encodings (`sin_time`, `cos_time`,
`day_of_month_sin/cos`, `month_of_year_sin/cos`).

- **Two generation types:** `type_name` = `Solar` / `Wind`.
- **Forecast target:** `value` — generation in that hour.
- **Features:** radiation, cloud cover, temperature, humidity, pressure, wind
  speed/direction, precipitation, sunrise/sunset times, and cyclic calendar encodings.

## Schema (TsFile structure)

- **Time** (INT64, milliseconds) — the hourly timestamp.
- **type_name** (TAG) — the device dimension, `Solar` / `Wind`. Query one type with
  `WHERE type_name='Solar'`.
- All remaining columns are FIELDs: `value` plus the meteorological and calendar features.

Type mapping: floating-point → DOUBLE, integers → INT64, strings → STRING. A small number
of byte-for-byte duplicate rows (from daylight-saving transitions) were removed so that a
single device never carries two points at the same timestamp.

## Usage

Read `data.tsfile` with the Apache TsFile Java or Python SDK.

## Source & license

- Original dataset: https://huggingface.co/datasets/wachawich/REG-Forecasting_v2
- Author: wachawich
- License: not declared by the original dataset card; please defer to the original dataset.