Study On Prediction Of Raw Material Index Content Of Rice Wine Based On Deeply Mixing Of Feature Extraction Methods

This paper mainly aimed at improving the prediction accuracy of raw material index content of yellow rice wine,and proposed a prediction model based on deeply mixing of feature extraction methods,which could provide raw material preference help for wine making enterprises to produce high quality yellow wine.The research content of this thesis was divided into the following three parts:(1)Feature extraction method hybrid technology.In this paper,using the extraction characteristics of different feature extraction methods to introduce a combination of variables to construct a transversal combination of feature extraction methods.In this way,the adaptive feature extraction of yellow rice wine product index set was realized.And through the generalized combination optimization algorithm,the combination of feature extraction methods was preferred.(2)Deeply mixing of feature extraction methods.Based on the hybrid technology of feature extraction method,a hierarchical feature extraction structure was constructed by using combined variables.And the combination of feature extraction methods was transitioned from horizontal to vertical,by which could further improve the adaptability of feature extraction.(3)Prediction model structure optimization.In this paper,the optimization of the whole structure of the prediction model was divided into two levels.The optimization of the outer structure of the prediction model and the optimization of the internal weight of the prediction model.The innovation of this thesis was divided into the following two parts:(1)In terms of feature extraction,a technology for deeply mixing of feature extraction methods was proposed.According to the different extraction characteristics of feature extraction methods,combined variables were introduced to construct a multi-layer feature extraction structure.To provide combinations of different feature extraction methods from horizontal to vertical,which could enhance the adaptability of feature extraction and improve the adaptability of different prediction models to data sets,so as to improve the prediction accuracy.(2)In terms of model optimization,one or more different deep learning prediction models were used as an integral part of the architecture.Integrating the optimization idea,the structure and weight of the prediction model were optimized by the intelligent optimization algorithm to enhance the adaptability of the model and improve the prediction accuracy of the model.