Desenvolvimento de um modelo de previsão hidrológico usando máquina vetorial de suporte regressivo: uma aproximação computacional para modelagem da bacia do rio arkansas

• Autor(a) principal: Lima, Rômulo Galdino da Rocha
• Data de Publicação: 2018
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Repositório Digital da Universidade Federal Rural do Semi-Árido (RDU)
• Texto Completo: https://repositorio.ufersa.edu.br/handle/prefix/905
• topic: Recursos hídricos; Predição; Redes neurais; Árvore de regressão; SVR; Water resources; Prediction; Neural networks; Regression tree; SVR; CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO

Investments in natural water resources are used for different purposes among the governmental bodies that manage these resources. Among these purposes are water supply, power generation, irrigation and navigation. The prediction of the accuracy of such natural resources can have a great impact and relevance for society, since the precise estimation of the future behavior of these resources can guide the authorities in the decision-making process, in order to better direct investments in prevention and mitigation of the effects caused by the lack of these resources, it is necessary to develop reliable forecasting methods. This dissertation has as main objective the implementation, application and analysis of the algorithm SVR (Support Vector Regression) in order to have a more precise computational model of water predictability in relation to other two models of future prognosis (neural network and regression tree). Using global metrics MAE, RMSE and local, for the accuracy of the models, this proposed method applied predictive models of images, aiming to consider future changes of water resources using information obtained from the set of NDVI historical data. The images obtained showed the small modifications in the evaluated time interval of 16 days for the Arkansas river. Comparing the accuracy of the three methods and their results showed that the SVR algorithm was able to quantify more accurately statistically in comparison with two other models and to be more robust in computational performance to predict fluid movements in the Arkansas river basin