OpenSynth/TUDelft-Electricity-Consumption-1.0

Timeseries Data Processing

This repository contains a script for loading and processing time series data using the datasets library and converting it to a pandas DataFrame for further analysis.

Dataset

The dataset used contains time series data with the following features:

• id: Identifier for the dataset, formatted as Country_Number of Household (e.g., GE_1 for Germany, household 1).
• datetime: Timestamp indicating the date and time of the observation.
• target: Energy consumption measured in kilowatt-hours (kWh).
• category: The resolution of the time series (e.g., 15 minutes, 30 minutes, 60 minutes).

Data Sources

The research utilizes high-resolution, open-source datasets from multiple countries, capturing diverse energy consumption patterns and system designs. The datasets are as follows:

---

1. Netherlands Smart Meter Data

Source: Liander Open Data
Description:

• Dataset Name: Zonnedael Dataset
• Time Period:
  • Electricity: May 1, 2012 – March 7, 2014
• Granularity:
  • Electricity: 60-minute intervals
• Scope:
  • ~80 addresses with user-consented smart meter data.
  • Data units: Wh/15min (electricity).
  • Data corrected for missing values, unrealistic consumption, or interpolation as necessary.
  • Includes metadata on household types, home age, and structure, linked via anonymized customer IDs.

---

2. UK Smart Meter Data

Source: London Datastore
Description:

• Dataset Name: Low Carbon London Project
• Time Period: November 2011 – February 2014
• Granularity: 30-minute intervals
• Scope:
  • Energy consumption data for 5,567 London households.
  • Approximately 167 million records in total (10GB uncompressed).
  • Two customer groups:
    1. 1,100 households participating in a Dynamic Time-of-Use (dToU) tariff trial in 2013, exposed to variable energy pricing (high, low, normal). Price schedules are provided alongside the dataset.
    2. 4,500 households on a flat-rate tariff.
  • Designed to test pricing effects on consumption behavior and distribution grid stress.

---

3. Germany Smart Meter Data

Source: Open Power System Data
Description:

• Dataset Name: CoSSMic Household Load and Solar Data
• Time Period: Varies by household; second release includes interpolated and processed data.
• Granularity:
  • 1-minute intervals (raw)
  • Aggregated to 15-minute and hourly intervals
• Scope:
  • 11 residential and small business households in southern Germany.
  • Includes household-level load and solar power generation data.
  • Covers single-device consumption, gaps interpolated or filled using prior-day data.
  • Cumulative energy generation/consumption retained.

---

4. Australian Smart Meter Data

Source: Smart-Grid Smart-City Customer Trial Data
Description:

• Dataset Name: SGSC Customer Trial Data
• Time Period: 2010 – 2014
• Granularity: 30-minute intervals
• Scope:
  • Part of a large-scale smart grid demonstration project in Australia.
  • Includes energy consumption data linked with demographic profiles, appliance use, peak event responses, and retail offers.
  • Data is anonymized and linked via customer IDs.
  • Accompanied by metadata such as climate conditions and product acceptance rates.

Requirements

• Python 3.6+
• datasets library
• pandas library

You can install the required libraries using pip:

python -m pip install "dask[complete]"    # Install everything

Usage

The following example demonstrates how to load the dataset and convert it to a pandas DataFrame.

import dask.dataframe as dd
# read parquet file
df = dd.read_parquet("hf://datasets/Weijie1996/load_timeseries/30m_resolution_ge/ge_30m.parquet")
# change to pandas dataframe
df = df.compute()

Output

        id            datetime    target category
0  NL_1  2013-01-01 00:00:00  0.117475      60m
1  NL_1  2013-01-01 01:00:00  0.104347      60m
2  NL_1  2013-01-01 02:00:00  0.103173      60m
3  NL_1  2013-01-01 03:00:00  0.101686      60m
4  NL_1  2013-01-01 04:00:00  0.099632      60m

Related Work

This dataset has been utilized in the following research studies:

1. Comparative Assessment of Generative Models for Transformer- and Consumer-Level Load Profiles Generation

   • GitHub Repository: Generative Models for Customer Profile Generation

2. A Flow-Based Model for Conditional and Probabilistic Electricity Consumption Profile Generation and Prediction

   • GitHub Repository: Full Convolutional Profile Flow

3. An Efficient and Explainable Transformer-Based Few-Shot Learning for Modeling Electricity Consumption Profiles Across Thousands of Domains

   • GitHub Repository: [GMMtransformer](https://github.com/xiaweijie1996/TransformerEM-GMM

Information

These datasets are made available via Weijie Xia and Pedro P. Vergara from the Delft University of Technology. Weijie Xia and Pedro P. Vergara are funded via the ALIGN4Energy Project (with project number NWA.1389.20.251) of the research programme NWA ORC 2020 which is (partly) financed by the Dutch Research Council (NWO), The Netherlands