Research On Quality Evaluation Of Wine Based On GA-BP Neural Network

With the increasing popularity of wine consumption in China in recent years,the problems faced by the domestic wine market have followed,especially with respect to wine quality evaluation.As the current mainstream evaluation methods rely more on artificial tasting,and there is less concern about the physical and chemical data of wine,there is no unified and valid wine quality assessment model.Due to lack of assessment programs for physics and chemistry indicators of wines,researching a set of scientific and reasonable alcohol quality assessment models is inevitable.At present,mass spectrometers are commonly used in the wine industry to detect various chemical components in wine,but this method cannot analyze the effect of component content on overall quality.The currently used evaluation models are analytic hierarchy process and multiple logistic regression,but all have their own defects.For the above methods can not accurately reflect the problem of wine quality,this paper proposes an improved model based on genetic algorithm and BP neural network.The model is based on BP neural network.By determining the topology structure of the BP network,the wine sample set data is input into the network for training,and the quality of the wine sample is evaluated using the trained network model.In order to improve the network mapping ability and network convergence speed of the neural activation function in traditional BP neural network,this paper proposes KReLU and LogReLU activation function with independent hyperparameters,and then uses the wine sample set training network to obtain the optimal hyperparameter value.To determine a reasonable hidden layer neuron activation function.In order to improve the searching ability of global operator in GA,this paper chooses the real coding method,and optimizes the selection operation in GA,crossover operator and mutation operator respectively,in order to improve the overall population adaptability.At the same time,degree also maintains the diversity of population as much as possible,which is more conducive to the preservation of individuals with high fitness in the population,while low fitness individuals are more likely to be eliminated,thereby reducing the probability of falling into a local optimal solution.The experimental comparison proves that the improved OGA-BP model has a stronger global search network’s optimal weight and threshold than the improved GA-BP model.It has a more accurate assessment of wine quality and has a strong practicality value.