A Research Of Liquor Yield Prediction And Production Process Parameters Optimization Algorithm

At present,the liquor industry mainly relies on traditional production technology.To achieve the goal of high-quality and high-yield of liquor,the liquor industry must promote the decoding of liquor production codes based on the traditional production experience.Therefore,according to the collected production data,it is particularly urgent to adjust the production process parameters to achieve the demand of high-quality and high-yield of liquor,but there is still a large gap in the application of optimizing the production process parameters.According the problem mentioned above,in this thesis,a new research approach is put forward by using association rule,machine learning,deep learning model to optimize the production process parameters.On the basis of collecting and processing production data,the research on optimizing the production process parameters of model is progressive and closely related,and lays the model foundation for the realization of high-quality and high-yield of liquor.The main work and innovations of this thesis are as follows:1)This thesis applies association rule to optimize the production process parameters.Firstly,the minimum acceptable support and confidence set manually are input into association rule,and output the support and confidence of each strong rule,then combines the support and confidence of each strong rule into a two-dimension data set.The clustering algorithm clusters the data set to obtain the special category with the highest average support and confidence.Finally,the minimum support and confidence of the special category are input into association rule again,and the strong rules of the first-class liquor label are screened,so that the influence of the combination of production process parameters on liquor quality can be analyzed.2)Combined with orthogonal experiment and other methods,this thesis uses machine learning to realize the quality classification and yield regression prediction of liquor,as well as the production process parameters optimization.In order to analyze the relationship between the production process parameters and liquor yield,this thesis builds quality classification and yield regression prediction models based on production features.Experiments show that liquor quality classification Accuracy is 0.916 and the yield regression prediction MAE is 3.450.For optimization of the production processparameters,the test samples are constructed in the following ways: constructing the full samples by exhausting the combination of production features,and taking part of samples from the full samples to represent the whole by orthogonal experiment.Considering the efficiency,this thesis puts the samples output from orthogonal experiment into the quality classification and yield regression prediction models,and the samples with high yield and labeled as first-class liquor are screened.Experiments show that the production process parameters optimization results are basically consistent with the results of the exhaustive way.3)Introducing the principles of deep learning algorithms such as residual network and TextCNN,this thesis proposes a production process parameters optimization algorithm based on the fully connected feature model,the shortcut connection feature model and the convolution feature model.Firstly,according to the liquor production process,the production features are separated and assigned to a special domain.Then the advanced features of each domain are generated from the full connected feature model,and the convolution feature model is used to simulate the interaction process of adjacent production processes and generates the convolution features required by the prediction model.Finally,all features are summarized in the feature merge layer by the shortcut connection feature model.Supplemented by methods such as full connection and dropout,the quality classification and yield regression prediction models are established.4)In order to further reduce the yield regression prediction error of deep learning model,this thesis uses model stacking to improve and optimize the deep learning model,and realizes the integration of machine learning and deep learning.Since the output of machine learning model is liquor yield,it can be considered as the parameter of liquor picking process,and uses them as an independent domain,thus,the corresponding advanced features and convolution features can be generated.Experiments show that the regression prediction effect of liquor yield has been greatly improved.In order to ensure the stability of the production process parameters optimization results obtained by the model,this thesis considers using association rule and machine learning to perform interval verification and intersection verification on the optimization results.