Modelación del crecimiento de pollitas Lohmann LSL con redes neuronales y modelos de regresión no lineal

Abstract

RESUMEN Objetivo. Modelar la curva del crecimiento de aves de la línea Lohmann LSL utilizando modelos no lineales (MNL), no lineales mixtos (MNLM) y redes neuronales artificiales (RNA). Materiales y métodos.Periódicamente se pesaron 33 aves en promedio, desde el día 21 al 196 de vida para un total de 558 registros individuales de peso. En el ajuste de la curva de crecimiento se utilizaron los modelos: no lineal de Von Bertalanffy (MNL), no lineal Mixto de Von Bertalanffy (MNLM) y redes neuronales artificiales (RNA). Los modelos se compararon con coeficiente de correlación y medidas de presicion cuadrado medio del error (CME), desviación media absoluta (MAD) y porcentaje de la media absoluta del error (MAPE). Resultados. Los valores de correlación entre los datos reales y estimados, fueron 0.999, 0.990 y 0.986 para MNLM, RNA y MNL respectivamente. El modelo más preciso con base en los criterios MAPE, MAD y CME fue el MNLM, seguido por la RNA. La grafica de predicción generada la RNA es similar a la del MNLM. La RNA presentó un desempeño superior al MLN. Conclusiones. El mejor modelo para la predicción de curvas de crecimiento de aves comerciales de la línea Lohmman LSL hasta los 196 días de edad, con múltiples mediciones por animal en el tiempo, fue el MNLM. La RNA presento un desempeño superior al MNL.Palabras clave: Crecimiento, modelo no lineal, modelo no lineal mixto, redes neuronales artificiales (Fuente: MeSH). ABSTRACTObjective. Modeling the pullet growth curve of the Lohmann LSL line, by using nonlinear model (MNL), nonlinear mixed model (MNLM) and artificial neural networks (ANN). Materials and methods. An average of 33 birds, were weighed from day 21 to 196 of life for 558 individual weight records. To adjust the growth curve the following models were used: nonlinear Von Bertalanffy (MNL), nonlinear mixed Von Bertalanffy (MNLM) and artificial neural networks (RNA). The models were compared with a correlation coefficient and precision measurements: mean square error (MSE), Mean Absolute Deviation (MAD) and the mean absolute percentage error (MAPE). Results. Correlation values, between actual and estimated data, were 0.999, 0.990 and 0.986 for MNLM, RNA and MNL respectively. The most accurate model based on the MAPE, MAD and CME criteria was MNLM followed by RNA. The prediction graph for RNA was similar to MNLM. The RNA performance was higher than MLN. Conclusions. The best model for the prediction of growth curves of commercial Lohmman LSL birds to 196 days of age, was the MNLM, with multiple measurements per animal at the time. RNA performance was higher MLN.

Growth Description of Pure and Crossbred Turkeys Using Non-Linear Models in Hot and Humid Tropical Environment

“…The result of the study for the model with the lowest values (Galeano-Vasco et al, 2014), as predicted by AIC and BIC, showed that the Von Bertalanffy's growth model had the smallest predicted value in both criteria and the best fit for all the turkey males and local turkey females while Gompertz´s growth model had the best fit for the female exotic and female crossbred turkeys. However, the difference between Von Bertalanffy's and Gompertz's growth models is minimal, thus, making them fit for modeling the growth curve of domestic turkey.…”

A Simulation of the Broiler Growth Rate Using Artificial Neural Networks

“…An ANN model was developed by Galeano and Cerón [2] to estimate the weight of birds based only on the age of the bird, they found a very good performance with a correlation coefficient (R) of 0.99. The authors recommend the ANN as a viable option for modelling the animal production, because ANN has the ability of new variables inclusion and good adjustment between measured and predicted variables.…”