Hive Neural Network Forecasting: Code and Preprocessed Dataset

This repository contains a simulate dataset derived from the oiginal data and the R code illustrating the implementation described in the associated article:

Robustillo, M.C., Senger, D., Parra, M.I., Pérez, C.J. (2026). A Multivariate Autoregressive Multilayer Perceptron Model for Predicting Internal Beehive Conditions from Sensor Data. Computers and Electronics in Agriculture, Volume 246, pages 111593.
DOI: https://doi.org/10.1016/j.compag.2026.111593

The dataset (ExampleHive.txt) contains internal and external measurements simulated from a single hive (Hive 0) from the BeeObserver project (https://doi.org/10.5281/zenodo.10407693). It is based on the original hive data recorded at an hourly sampling frequency, complemented with meteorological information from the German Weather Service (DWD). A small amount of Gaussian noise was added to the measurements to prevent exact replication of the original data and to allow dataset sharing. All data were preprocessed following the procedures described in Robustillo et al. (2026) and are provided in a format ready for direct use with the R script included in this repository.

Variable Description Unit
_____________________________________________________________________________________
time2 Timestamp of the measurement -
t_i_1…t_i_5 Internal hive temperature sensors (DS18B20, 5 positions) °C
weight_kg Hive weight (Bosche H30/H40 load cell under hive) kg
h Relative humidity inside hive (BME280 sensor) %
t Internal temperature inside hive (BME280 sensor) °C
p Internal air pressure inside hive (BME280 sensor) hPa
to_imp External temperature (DS18B20 sensor) °C
ho_ms External relative humidity from DWD %
precip_ms Precipitation from DWD station mm
wind_ms Wind speed from DWD station m/s
pressure_ms Atmospheric pressure from DWD station hPa
hour Hour of the measurement
day Day of the measurement
month Month of the measurement
year Year of the measurement
ext External variable capturing large hive weight variations kg
______________________________________________________________________________________

This research was funded by Ministry of Science, Innovation and Universities - Spain, and State Research Agency - Spain (Projects PID2021-122209OB-C32 and PID2024-155179NB-C21), funded by MICIU/AEI/10.13039/501100011033 and European Union (European Regional Development Fund).