Modelo de inferencia difusa evolutivo aplicado al pronóstico no estacionario de humedad interna en invernaderos

Abstract

Automated greenhouse agriculture serves as a tool for developing crops with high yield and quality indices, requiring proper management of internal climatic variables through prediction and control models. Due to the existence of measurable but uncontrollable disturbances that characterize the non-stationary behavior of these variables, models that can evolve and adapt to these changes are necessary. This document presents the development of a doctoral research project in which an evolutionary fuzzy inference model is constructed to forecast the non-stationary behavior of internal relative humidity in greenhouses for short-cycle crops, providing automatic adjustment of system parameters over time. The proposed methodology consists of three stages: the first is data analysis using the ETL (Extraction, Transformation, and Loading) process employed in data mining; the second is the definition of the model structure based on a systematic review of compatible structures using the PRISMA methodology; and the third is the development of the model based on the software prototyping paradigm. A software prototype of the model is established through various structures of evolutionary fuzzy inference systems, implementing optimized fuzzy systems with a hybrid Mamdani-type algorithm that handles floating-point values. This facilitates monitoring and control for individuals interested in agricultural activities, yielding results with high levels of interpretability and maximum precision in all proposals, with a mean squared error (MSE) of 1.20E-02.