The Research Of Optimize Numerical Blending Based On Artificial Neural Networks

As mean of artificial intelligence research, Computational Intelligence is widely used in some areas such as industry, agriculture and medicine. It provides a new design idea for the cigarette bending formulation, between the proportion of Tobacco and sensory characteristics have a complex and nonlinear relationship that is difficult to determine. Based on Computational Intelligence theory, meanwhile on the basis of summarizing predecessors' research results, this paper proposes to use Neural Network technology to design tobacco blending formulation that is called digital and intelligent Formulation design concept.First of all, the three aspects of the Total Blending formulation design are described in this paper, the three aspects are blending tobacco formula, Cutting and cigarettes parameters. Considering has not a good scientific classification for unknown samples of tobacco.combing two kinds of classical clustering method are proposed. Taking into account the experience of the experts, neural networks and fuzzy mathematics method, it is using 'have the teacher' support vector machines, these samples are divided into two categories of known and unknown sample Group. Two Groups are afterwards self-organizing clustering respectively. That existence of a Single sample group problem use C-mean method to choose the best alternative tobacco materials. New method to solving replacement of similar tobacco material is proposed by this paper. Now tobacco research designers Focus in the study of finding a more accurate comprehensive evaluation expression between formula's physical and chemical properties and sensory quality. Using this expression to predict whether the cigarettes have been met the requirements. Experimental data this thesis is based on experimental data. The input is tobacco's Physical and chemical properties. The output is tobacco's sensory characteristics. That a BP Neural Networks model which inflects the mapping relationship between the Physical and chemical properties and sensory characteristics is established and trained. Thereby, a mapping neural network model is established in this paper. The model Contrasts the remove second-degree polynomial regression equation that established. The results show that, the neural network method has a higher accuracy and much smaller error and more parallel processing capacity. It can solve the difficult to predict the sensory characteristics problem in cigarettes' production. At the same time this paper use established neural networks model and evolutionary computing methods to optimize tobacco formulation. formulation design idea that combines evaluating the smoking quality, computer-aided design and chemical analysis are proposed in this paper. According to the design requirements select tobacco material. And then it use uniform design to select the blending tobacco formulation. Smoking staffs assess the formulation's smoking quality. Combine BP Neural Network model and assessment results improve model predicted capacity. The neural network BP algorithm lakes predicted capacity and it difficult to find extreme points because sigmoid function values, the problem to find model's extreme points that based on expert experience are solved, narrowing the range of model aim to reduce the amount of low-grade tobacco. It use genetic algorithm optimize blending tobacco formulation eventually. Smoking improvement based on expert design amendments afterwards, it have designed a development of premium cost-effective blending formulation products with acceptable smoking qualities. The tobacco product expects a good market's affection.The design are used in computer to simulate Computational intelligence tobacco blending formulation, the design compared several formulation design methods. Through the analysis results, it verifies that using the digital and intelligent blending formula design is correct. The design idea has a certain guide of the experiments and actual production.