Using various data sources from the [Bosch Budapest Campus](https://www.homeofbosch.com/en/projects/campus-with-rooftop-gardens/) Building BP201, BP205, (BP106), we created this data repository containing the following types of data:

Time-Series Data from Sensors: This includes temperature, humidity, air quality, pressure, flow, energy consumption, valve and damper positions, pump and fan status, control system outputs, switches and relays status, enthalpy, operation counters, setpoints, control values, alarm, and fault indicators.

## About the included data:

The metadata file contains information about the sensors, including their names, types, descriptions, units and others. The metadata file is in the form of a .parquet file and a .xlsx file. The .parquet file contains the metadata in a structured format, while the .xlsx file contains the metadata in a human-readable format. The metadata file is used to interpret the time-series data from the sensors.

The compressed data files are in the form of .zip files, each containing a set of .parquet files that represent the time-series data from the sensors. The data is organized in a hierarchical structure, each .zip file containing data from a specific time period (e.g., a month). The .parquet files are named according to the sensor they represent, and each file contains the time-series data for that sensor. 

The data is stored in a columnar format, which allows for efficient querying and analysis. The time-series data is non-uniformly sampled, the sampling rate varies over time and change velocity. In case no data is available for a specific time period, the last known value is repeated. The dataset may also contain recording errors, missing data, and outliers. 

To have a simple example of how to use the sensor data, check the [example code](https://github.com/boschresearch/elias_aifbo).

## Files

The data repository contains the following files:

- **metadata.parquet**: The metadata file in the form of a .parquet file.
- **metadata.xlsx**: The metadata file in the form of a .xlsx file.
- **RBHU-2025-{01..07}.zip**: The compressed data files containing the time-series data from the sensors for each month of the year 2025. Regarding the separation into training and test set, and further details, see [Description section](www.kaggle.com/competitions/elias-uc-1-ai-for-building-optimisation/overview/description).


## Columns

The metadata file contains the following columns:

- object_id: The unique identifier of the sensor.
- class_id: The class identifier of the sensor.
- description: The description of the sensor.
- interpolation_mode: The interpolation mode of the time series (3: continuous, 2: discrete).
- pcl_type: The data type of the time series (REAL: float, INTEGER: enum, boolean)
- conversion_index: The conversion index contains information on how the raw value is interpreted by the data recording system.
- dimension_index: The dimension index of the time series.
- dimension_text: The dimension text of the time series.
- data_group_id: The data group corresponding to bde_channel_typ.
- bde_channel_typ: The channel type (AI: analog input, AO: analog output, DI: digital input, DO: digital output, PA: parameter, XX: internal data).
- file: The file name of the sensor.
- device_class: The device class contains group identifier of devices.
- device: The device id of the sensor.
- channel: The channel of the device.

## Further data available on Zenodo

To further improve your method, consider using the additional 2024 building data from Zenodo: [part 1](https://zenodo.org/records/12590466), [part 2](https://zenodo.org/records/14591934).

## Further notes

The original data contained leakage, such as some sensors measuring almost exactly the same object as the target sensor. Therefore, several sensors were removed from the dataset but may still occur in the metadata.