Water Demand Dataset Collection

Introduction

A crucial part of optimizing the cost of water infrastructures is to look at their energy consumption. To reduce the cost for energy, it is possible to either reduce the energy consumption or to shift the energy consumption to times when energy is cheaper. With an increasing share of renewable energy, the latter becomes more and more important. To shift the energy consumption, one of the most important steps is to forecast the water demand. In this repository, we provide a collection of eleven water demand datasets.

Datasets

Key Information

• Number of Datasets: 11
• Granularity: 15 minutes
• Time Range: 2024-01-01 to 2026-01-28 (differs across datasets)
• Number of Records: 47448 – 70161 (differs across datasets)
• Features: 1 (water demand in m³/h)

Origin

The datasets are collected from three different water utilities in Germany. The datasets represent real water demand data that is collected from sensors placed at outflows of reservoirs. Because of this, the measurements are purely demand-driven, meaning that they are not influenced by any control actions such as pumps or valves. Each dataset represents the water demand of a different reservoir and each of these reservoirs provides water to a small, rural area without any large-scale industrial water consumers.

Preparation

The transformation from raw data to the provided datasets involved the following steps:

1. Consistent Formatting: All datasets were transformed to have a consistent format with a timestamp column and a water demand column.
2. Timezone Conversion: All timestamps were converted from the original timezone of Europe/Berlin to UTC to ensure consistency across datasets. The issues that arise from this conversion were resolved by inferring and shifting forward (see the Pandas function tz_localize for more details).
3. Resampling: In case the original data had a different granularity, it was resampled to 15-minute intervals using the mean value of the water demand within each interval.
4. Rounding: The water demand values were rounded to six decimal places.
5. Handling Missing Values: Missing values (timestamps or water demand values) were dropped before exporting.

Data Quality

The datasets contain a few data quality issues such as missing values, out