B1884
Title: Autoregressive neural networks for predicting temperature, relative humidity, and weight of beehives
Authors: Maria del Carmen Robustillo Carmona - Universidad de Extremadura (Spain) [presenting]
Carlos Javier Perez Sanchez - University of Extremadura (Spain)
M Isabel Parra Arevalo - Universidad de Extremadura (Spain)
Lizbeth Naranjo Albarran - Universidad Nacional Autonoma de Mexico UNAM (Mexico)
Abstract: Bee populations have been declining worldwide in recent decades, which is a serious problem for the environment since they help to maintain biodiversity through pollination. In this context, precision beekeeping emerges as a tool that allows greater control over the status of bees, helping the beekeeper to improve their care and maintenance. The objective is to predict internal temperature, relative humidity, and weight using sensor data of internal conditions and meteorological information. For this purpose, data obtained by the we4bee project in the Vohburg and Markt Indersdorf hives have been used. Two different scenarios have been considered: firstly, neural networks that use only climatological variables and, secondly, neural networks considering both climatological and internal variables. Predictions were made for one, three and seven days ahead. To validate this model, a rolling window 100 fold cross validation was performed, obtaining mean absolute errors lower than 0.750 Celsius degrees, 3.1\%, and 210 g in the one-day predictions of temperature, humidity, and weight, respectively. Improvements ranging from 2.28\% to 45.66\% have been observed when comparing both scenarios, concluding that the incorporation of internal hive information is important. These results show the power of autoregressive neural networks to make predictions about hive conditions.