Optimization of Prediction Model for Milk Fat Content Based on Improved Whale Optimization Algorithm

Milk fat content of high and low will affect people's health. The experiment in milk fat content analysis indicators, application of image processing technology analysis of hyperspectral data, extracting the region of interest (ROI) from hyperspectral images using the ENVI software, different preprocessing methods were used to establish Partial Least Squares Regression (PLSR) model for spectral data and the best preprocessing method was obtained by comparison, Then, different numbers of principal components were used for feature extraction of the pre-processed data and Support Vector Regression (SVR) model was established. The optimal number of principal components was obtained through comparison. Finally, the SVR prediction model was established for the data after feature extraction to analyze the fat content in milk. Since the traditional SVR model has a poor prediction effect and cannot meet people's basic requirements, this paper proposes a hybrid strategy improved whale optimization algorithm to optimize the SVR prediction model. The evaluation parameters of the SVR model optimized by hybrid strategy whale optimization algorithm are compared with those optimized by genetic algorithm, traditional whale optimization algorithm and elite reverse learning whale optimization algorithm. The results show that the training set and prediction set coefficient of determination (R-2) of the SVR model optimized by hybrid strategy modified Whale optimization algorithm are 0. 998 and 0. 995, respectively. The reciprocal 1/RMSE values of Root Mean Square Error (RMSE) were 13. 766 and 6. 191, and the reciprocal 1/MAE values of Mean Absolute Error (MAE) were 13. 910 and 11. 422, respectively. The training set and prediction set parameters R-2 of the SVR model optimized by the traditional whale optimization algorithm are 0. 998 and 0. 989, 1/RMSE is 13. 526 and 5. 849, and 1/MAE is 13. 616 and 7. 037, respectively. The training set and prediction set parameters R-2 of the SVR model optimized by the whale optimization algorithm improved by reverse learning strategy are 0. 998 and 0. 988, 1/RMSE is 12. 474 and 6. 421, and 1/MAE is 15. 003 and 10. 554, respectively. The above results show that the hybrid strategy improved whale optimization algorithm is feasible to optimize the SVR prediction model, and the optimized SVR model has a better prediction effect.