Modelagem matemática na previsão de colheita de bananeira: regressão linear múltipla x redes neurais artificiais / Mathematical modeling to predict crop of banana: multiple regression x neural networks
AUTOR(ES)
Joyce Dória Rodrigues Soares
FONTE
IBICT - Instituto Brasileiro de Informação em Ciência e Tecnologia
DATA DE PUBLICAÇÃO
07/12/2010
RESUMO
One of the barriers relevant to the banana crop in Brazil is the lack of productive commercial varieties with adequate size, resistant to major pests and diseases and adapted to different ecosystems. The development of cultivars is the strategy for solving this problem through breeding programs, as well as its characterization and evaluation in areas of production when compared to traditional cultivars. The characters observed in the experimental areas have phenotypic nature and most often based solely on the experience of the producer, the measurement of bunch weight and the correlations of the variables involved are estimated in order to measure changes in a character when he altered another. This study aimed to: 1) Develop a model based in mathematic simple linear regression, 2) study the relationship between bunch weight and the variables that make up the model, 3) Create a prediction model based on neural networks harvest artificial; 4) Compare systems of predicting bunch weight: regression and neural networks. The experiment consisted of a uniformity trial, conducted in Guanambi, BA, with cultivar Tropical (YB42-21), AAAB tetraploid hybrid, planted at a spacing of 3 mx 2 m, consisting of 11 rows of 52 plants each and considered how useful the nine central rows with 40 plants per row, totaling 360 plants and an area of 2,160 m2. We assessed the vegetative characters, plant height, pseudostem circumference, number of children issued and number of green leaves at flowering and harvest, and the characters of yield, bunch weight, number of hands and fruits, weight of the second hand, length and diameter of the fruit in two growing seasons. In the evaluation, each plant was considered as a basic unit (bu), area of 6 m2 totaling well, 360 basic units (bu). It is inappropriate to use linear models for predicting the weight of the bunch. The RNA had implemented an efficient production forecast, and the network that obtained the best result has 10:10:1 architecture.
ASSUNTO(S)
banana inteligência computacional previsão de colheita fitotecnia banana computational intelligence forecasting crop
