Resumen de An LSTM Deep Learning Scheme for Prediction of Low Temperatures in Agriculture

M. Ángel Guillén Navarro, Raquel Martínez España , Andrés Bueno Crespo , Belén López Ayuso , José Luis Moreno, José M. Cecilia

• Precision agriculture adopts a set of techniques capable of increasingproductivity, yield and efficiency in work related to agriculture, producing a greaterbenefit for farmers. In this study we focus on the problem of predicting weatherconditions, specifically the prediction of low temperatures. Temperature predictionis a major problem in agriculture. Farmers can lose their crops if frost control tech-niques are not activated in time. The threshold for activating such techniques de-pends on the type of crop. A first preliminary study using deep learning is proposedto predict temperature, particularly a Long Short-Term Memory Network (LSTM)is used. The LSTM has been trained using real temperature data provided by an theInternet of Things (IoT) system, deployed in several plots and currently in opera-tion. The results obtained after testing the model created with this neural networkare quite satisfactory obtaining a determination coefficient (R2) of 99% and an av-erage quadratic error of less than 0.8 degrees Celsius. Given the goodness of themodel this can be implemented as an intelligent component of the IoT system, thuscomplementing its functionality.