An application of neural network and time series techniques in animal system modeling is presented, which sheds lights on strategies to deal with some typical peculiarities in biological system data. A neural network model was developed to describe the multi-input and multi-output relationship from live weight gain, age, breed, and rearing status to monthly wool growth rates of sheep. Constrained synaptic weights were imposed to avoid logically unrealistic functional relationships. ARMAX models were developed to describe the dynamics in wool growth rate and to analyze the time relationship between live weight gain and wool growth rate. The recursive least-squares parameter estimation algorithm was employed for effective use of short-term and multi-subject data in developing time series models.
展开▼
