Study On Deep Mixing Prediction Model Of Index Content Of Tangyuan Raw Materials

In this thesis,the prediction accuracy of raw material index content is as high as possible,and a deep mixed prediction model of raw material index content of Glutinous rice balls is proposed,which can provide help for food processing enterprises in the selection of raw materials and product quality testing.The research content of this thesis is divided into the following three parts:(1)Data preprocessing.Firstly,the indexes of tangyuan products and raw materials obtained were preprocessed,and then the characteristics of the indexes of tangyuan products and raw materials were extracted.Finally,factor analysis was conducted on the indexes of tangyuan raw materials,and the raw material index system under the three dimensions of taste composition,appearance composition and nutrition composition was obtained.(2)Secondary feature extraction and prediction model selection.Three feature extraction methods including multiple stepwise regression,principal component analysis and random forest were used to extract the secondary feature of tangyuan product index.Three kinds of classical regression models and four kinds of deep learning models were used to predict the content of tangyuan raw material.(3)Structure optimization of depth mixed prediction model.In this thesis,the optimization of the whole structure of the deep mixed prediction model is divided into three levels: Optimization of feature extraction model and prediction model selection;optimization of outer structure of depth mixed prediction model;optimization of internal hyperparameters of depth mixed prediction model.The innovation of this thesis is divided into the following two parts:(1)A multi-stage feature extraction + prediction model selection + model optimization framework was proposed to improve the prediction accuracy of tangyuan raw material index content.(2)When optimizing the structure of the whole depth mixed prediction model,the following improvements were made to the particle swarm optimization(PSO)algorithm: the inertia weight was nonlinear corrected with the loss value of the training set of tangyuan raw material index content,which enhanced the local searching ability of PSO;Combining the outer structure of the depth mixed prediction model with the velocity in the four dimensions of PSO,a new particle motion criterion is found.The idea of weighted optimization and difference optimization is introduced to the objective function,so that the loss function of the training set and test set of the index content of tangyuan raw material is as small as possible,and at the same time as close as possible.