Assessment of dairy cow feed intake based on BP neural network with polynomial decay learning rate

Abstract

To overcome the shortcomings of traditional dairy cow feed intake assessment model and BP neural network, this paper proposes a method of optimizing BP neural network using polynomial decay learning rate, taking the cow’s body weight, lying duration, lying times, walking steps, foraging duration and concentrate-roughage ratio as input variables and taking the actual feed intake is the output variable to establish a dairy cow feed intake assessment model, and the model is trained and verified by experimental data collected on site. For the sake of comparative study, feed intake is simultaneously assessed by SVR model, KNN logistic regression model, traditional BP neural network model, and multilayer BP neural network model. The results show that the established BP model using the polynomial decay learning rate has the highest assessment accuracy, the MSPE, RMSE, MAE, MAPE and R2 are 0.043 kg2/d and 0.208 kg/d, 0.173 kg/d, 1.37% and 0.94 respectively. Compared with SVR model and KNN mode, the RMSE value reduced by 43.9% and 26.5%, it is also found that the model designed in this paper has many advantages in comparison with the BP model and multilayer BP model in terms of precision and generalization. Therefore, this method is ready to be applied for accurately evaluating the dairy cow feed intake, and it can provide theoretical guidance and technical support for the precise-feeding and can also be of high significance in the improvement of dairy precise-breeding.

Keywords

• Cow
• Feed intake assessment
• BP neural network
• Polynomial decay learning rate