Artificial Neural Networks In The Application Of Beer Sensory Evaluation

Beer sensory evaluation is an important method for quality monitoring and testingdifferent flavors of beer. The assessment result of traditional method is influenced seriouslyby subjective opinions. The method of modern instruments is difficult to evaluation of beerflavor quality. If combination modern instruments method and beer sensory evaluation, willimprove the objectivity of beer sensory evaluation. Therefore, establishing a new type of beersensory evaluation method is very necessary.Using physicochemical indexes and beer flavor as input and sensory evaluation scores asoutput, artificial neural network back propagation（BP） algorithm was used to establish theprediction model. The grey associative analysis method, artificial neural network and geneticalgorithm were applied to beer sensory evaluation. A new method of sensory evaluation ofbeer is proposed. The main research contents and results are as follows:（1） The150kinds of beer were detected, including physicochemical indexes, beerflavor and sensory evaluation scores. The original experimental data was prepared for allresearch.（2） The experimental data as input, establish BP neural network prediction model.This model was used to predict the50beer’s sensory evaluation scores. The average absoluteerror is5.38; the predictive average relative error is8.69%.（3） Grey relationship analysis theory is used to filter the most important quantitativetechnical indices which can reflect beer sensory evaluation in order to optimize the inputparameters of the BP neural network. Establish Gray-BP neural network prediction model.This model was used to predict the50beer’s sensory evaluation scores. The average absoluteerror is3.23; the predictive average relative error is5.22%.（4） Genetic algorithm（GA）was used to optimize the neural network weights.Establish Gray-BP-GA neural network prediction model. This model was used to predict the50beer’s sensory evaluation scores. The average absolute error is2.53; the predictive averagerelative error is4.06%. Comparison of different models found that Gray-BP-GA neural network showed a good predictive ability.Therefore，the results indicated that the model was adoptable for beer sensory evaluationprediction.