Redes neurais artificiais aplicadas à predição da curva de retenção de água de substratos para plantas

• Ano de defesa: 2024
• Autor(a) principal: Pires, Fábio Soares
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• dARK ID: ark:/26339/001300001bsfp
• Idioma: por
• Instituição de defesa: Universidade Federal de Santa Maria; Brasil; Engenharia Agrícola; UFSM; Programa de Pós-Graduação em Engenharia Agrícola; Centro de Ciências Rurais
• Palavras-chave em Português: Agricultura; Modelagem; Sustentabilidade; Previsão; Otimização; Modeling; Sustainability; Prediction; Optimization; CNPQ::CIENCIAS AGRARIAS::ENGENHARIA AGRICOLA
• Link de acesso: http://repositorio.ufsm.br/handle/1/32955

The use of artificial intelligence has stood out as a powerful tool in predicting outcomes through machine learning, especially when dealing with large volumes of data. The integration of artificial intelligence techniques, such as neural networks, with traditional statistical methods, like principal component analysis (PCA) and clustering algorithms, such as k-means, in developing predictive models to understand hydrological processes in plant substrates, proves to be a promising approach to comprehend the relationship between volumetric moisture and different matrix potentials. By training these models with comprehensive and representative datasets, capturing complex patterns in the data and making more accurate predictions about water behavior is expected. Additionally, the combination of neural networks with clustering algorithms, such as k-means, allows for identifying patterns in the data that may not be easily perceptible to the naked eye, which is useful for grouping moisture data, enabling a detailed analysis of variations in water distribution. Principal component analysis (PCA) complements this process by aiding in reducing data dimensionality and identifying the main variables that influence water retention, facilitating result interpretation and identifying important relationships between variables. In the context of agriculture, these techniques can have broad applications, from efficient irrigation and drainage management to crop planning and yield prediction. Thus, the main objective of this work is to present methodological approaches with advanced artificial intelligence techniques to accurately predict the water retention curve of plant substrates, aiming to contribute to the advancement of precision agriculture and the development of more sustainable and efficient agricultural practices. The analysis of the prediction results conducted on the five clusters (K1 to K5) revealed valuable information about the relationship between matrix potentials and the moisture of formulations. Neural networks demonstrated an impressive ability to model and predict moisture under different conditions, as represented by the various matrix potentials. The coefficients of determination (R²) obtained for the training, testing, and validation data reflect the model's effectiveness in explaining variability in the data and providing accurate predictions. Identifying consistent patterns between observed and predicted values, even with small databases (LI et al., 2016), highlights the robustness and generalization capability of neural networks in generating the water retention curve in plant substrates. These results suggest that neural networks are a powerful and versatile tool for understanding and modeling moisture in different waste materials, providing important information for producers, aiding in water management, and environmental conservation. This study convincingly emphasizes the vital role of neural networks in advancing sciences related to irrigation, drainage, and agricultural practices by offering a deeper and more accurate understanding of the processes involved in substrate formulation, as well as the neural networks that promote a promising approach. Therefore, with this innovative tool, we have the ability to significantly improve our agricultural practices, driving efficiency, productivity, and sustainability.