Journal
JOURNAL OF APPLIED POULTRY RESEARCH
Volume 18, Issue 3, Pages 440-446Publisher
POULTRY SCIENCE ASSOC INC
DOI: 10.3382/japr.2008-00064
Keywords
growth modeling; simulation; neural network
Categories
Funding
- NCRR NIH HHS [G12 RR013459] Funding Source: Medline
Ask authors/readers for more resources
Poultry growth is usually modeled with the Gompertz model or another nonlinear statistical model using average BW data over certain periods of time for a given strain of birds under specific farm management conditions. Constant selection in the genetic pool, nutritional factors, and environmental concerns, however, make such models limited in their utility because of the difficulty of fitting the growth curve across time, bird strains, and other determining variables. Moreover, generating data for every strain of birds under continually changing variables is difficult, expensive, and time consuming. The current model addresses 2 objectives: to simulate data using published literature for different growth periods, and to develop artificial intelligence models with various neural network architectures. By breaking down the actual broiler growth data into 5-d intervals, with known means and SD, normal distributions were generated for broiler growth using @Risk software. These simulated data were then used to recognize data patterns and model growth curves by using various neural networks. Three neural networks, namely, BackPropagation-3 (3 layers of back propagation, with each layer connected to the previous layer), BackPropagation-5 (5 layers of back propagation, with each layer connected to the previous layer), and Ward-5 (5 hidden slabs with various activation functions, using Neuro-Shell 2 Ward software) were used in this research. Once the networks were sufficiently trained, they were exposed to actual growth data to predict broiler growth over the next 50 d. The BackPropagation-3 neural network gave the best fitting line, with predictions fitting tightly to the actual data points. The R-2 was 0.998, and nearly perfect. The R-2 for the BackPropagation-5 and Ward-5 neural networks were 0.967 and 0.973, respectively. To test the approach further, the same methodology was applied in guinea fowl growth prediction, resulting in R-2 of 0.96 both for the general regression and Ward-5 neural networks.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available