Neural predictive model in the estimation process of somatic cell counts in milk

摘要

Neural models, generally occurring in the form of computer numerical simulators, are the effective tools used in research on complex empirical systems. Moreover, they are also useful in the acquisition of scientific knowledge related to implementation of the complex biological processes. Commonly recognized predictive abilities characterized for selected ANN (Artificial Neural Networks) topologies are widely used in practice. They often support the decision-making processes that occur in agri-alimentary processing, such as milk production. In this case very important is identification and then estimation of the qualitative attributes of the product, where the cows genotype and selected environmental factors are taken into account. The aim of the study was ANN application as a predictive tool in the estimation process of the influence of selected zootechnical characteristics of cows on the milk quality, which is determined by the standards defining the requirements compliance concerning the level of somatic cell counts in the obtained milk. The work resulted in creation of the optimum predictive model which is a neural topology of the MLP-4:17:1 (MultiLayer Perceptron). Generated multilayer perceptron was trained by the hybrid techniques with assistance of the standard optimization algorithms: BP (Back Propagation)) and CG (Conjugate Gradient). Generated topology has got 4 neurons in the input layer, 17 in the hidden layer and 1 neuron in the output layer, representing the predictive somatic cell counts in the obtained milk. The performed analysis of the generated neural model's sensitivity to the individual input variables has showed the impact of some of the zootechnical characteristics on somatic cell counts in the obtained milk.