Dados do Trabalho

Título

ESTIMATION OF WATER INFILTRATION RATE IN CLAYEY TEXTURED SOILS BY MEANS OF AN ARTIFICIAL INTELLIGENCE

Resumo

Water infiltration into the soil is one of the major factors influencing surface runoff and provoking undesirable processes such as erosion and flooding. In this sense, the modeling of this process of infiltration brings great practical applications to soil management that contribute to define soil conservation techniques, plan and design irrigation mechanisms and drainage systems. Under this aspect, this study aimed to estimate the water infiltration rate in a clay textured soil. For this purpose, a modeling method based on Artificial Neural Networks (ANNs) Multi Layer Perceptron (MLP) with Backpropagation learning algorithm was developed using the Matlab® software on a volume of 214 data, 60% for training and 40% for validation and network testing. The network input data were: accumulated infiltration, infiltration variation, average infiltration rate, variation of time and accumulated time while the output data represents the water infiltration rate into the soil. This method obtained a mean square error (MSE) training performance of 1.03x10^-4 between the output obtained via ANN and the desired output via experiment, based on five iterations and one second of processing. As for the validation and testing phase of the network, the MSE value was of 1.7x10^-2 for data that were not part of the training. The results obtained were also compared with the empirical and theoretical models proposed in the literature and were shown to be promising, low MSE, which shows that this modeling method has great potential to assist in estimating the infiltration rate of water into the soil and may support significant contributions to the area, especially those regarding to the management and conservation of soil and water.  

Palavras-chave

Artificial Neural Networks, Prediction, Soil Management, Mathematical Modeling.

Instituição financiadora

São Paulo State University - UNESP

Agradecimentos

São Paulo State University - UNESP