Utilização de redes neurais artificiais para a aproximação de ganhos de peso de frango de corte

• Autor(a) principal: Sidney, Luiz Henrique Faleiro
• Data de Publicação: 2011
• Tipo de documento: Trabalho de conclusão de curso
• Idioma: por
• Título da fonte: Repositório Institucional da UFLA
• Texto Completo: http://repositorio.ufla.br/jspui/handle/1/30687
• topic: Redes neurais artificiais; Avicultura; Modelagem de sistemas biológicos

Among the challenges facing the poultry industry, highlights the importance of the authoring environment. When environmental conditions in domestic poultry shed are not adequate and the animals are exposed to conditions of heat stress, there is a decrease of well-being, which could compromise their growth and performance. An animal can demonstrate their comfort or discomfort in their productive responses. Given the above, this study aims to predict the overall weight gain of broilers during the first three weeks of life. To predict the productive performance of chickens, an artificial neural network was developed based on literature data and experimental, taking as input the age of the chicken (weeks), temperature (° C) and relative humidity (%) in the authoring environment, and as output variable weight gain of animals (g). To obtain the artificial neural network (ANN), it was necessary to make an initial treatment of the data. First, the data of temperature, relative humidity, temperature and weight gain were normalized and then normalized again, however, for the week of the bird's life. For the construction of RNA was used a topology composed of two networks, with the output of the first entries in the RNA of a second, that beyond this entrance also used the same inputs of the first RNA and treated all data related to temperature and relative humidity air, and its output weight gain. The proposed neural network was adequate to approximate the weight gains of chickens in the heating phase, with an R ! = 0.8442, an average standard deviation of 27.1 grams and average percentage error of 0.14.