Location:Home > Engineering science > Motor and electrical > Application of Numerical Designing Cigarette Blending Formula on Computational Intelligence

Application of Numerical Designing Cigarette Blending Formula on Computational Intelligence

Downloads: []
Tutor: LiuMeiHong
School: Kunming University of Science and Technology
Course: Mechanical Design and Theory
Keywords: Computational Intelligence,Tobacco Classification,Forecast Sensory-Quality of Ci
Type: Master's thesis
Year:  2011
Facebook Google+ Email Gmail Evernote LinkedIn Twitter Addthis

not access Image Error Other errors

Firstly, cigarette formula design is the key in the entire cigarette design, basically containing two parts:cigarette blending formula design and charging, add fragrant formulate design. This paper using computational intelligence(CI) as theory, on the basis of tobacco data that provided by Kunming Shipbuilding Equipment Co, Ltd(KSEC) and the research conclusion with earlier researchers, putting forward CI to design numerical cigarette blending formula. To get the best sensory-quality, smoke yield of cigarette and the optimal cost via the most reasonable blending formula design.Secondly, the tobacco classification by using the traditional grade method was expounded in this paper, therefore, a brand-new tobacco classification method was put forward, which uses two clustering neural network combined with expert experience and fuzzy mathematics. First, on the basis of tobacco raw data, divided these tobacco into four categories under experience guidance. And then applying Learning Vector Quantization (LVQ) Artificial Neural Networks of Supervised Learning to make simulation experiment, and to verify the accuracy of the classification results of group expert;Second, if the classification of simulation result by LVQ Artificial Neural Networks and the experience category of experts are different, applying Adaptive Resonance Theory (ART) Artificial Neural Networks of Unsupervised Learning to make simulation experiment, getting new result to verify again. Finally, if above grouping results also are different, applying Fuzzy Center Means (FCM) to decide which kind of category do this attribute by the relationship of different samples degree membership. It is showed that this new classification method has 94.12% accurate results compare with the sort results of group experts, they combine with experiences for the 17 kind of tobacco. Thirdly, the evaluation of the quality of product of cigarette was introduced in this paper. Thus, on the basis of intrinsic chemical composition of tobacco, using Conjugate Gradient Back Propagation (BP) Artificial Neural Networks that based on numerical optimization method, to realize the prediction of the intrinsic chemical composition of tobacco, Therefore, it is a hot issue to find comprehensive evaluation that comparative accurate express the chemical composition of tobacco, sensory-quality and smoke yield. Meanwhile for the problem of local minimum points which is easily involved in of the BP Artificial Neural Networks, using Simulated Annealing (SA) to optimize the BP neural network weights and threshold to solve them. Results show that Conjugate Gradient BP neural network is more precise and smaller errors than standard BP neural network in fitting. The error between forecasting results and expected value is only 0.00000072. In addition, for forecasting sensory quality of 17th group tobacco with standard BP neural network, there existing the local minimum problem; using SA to optimize the BP neural network weights and threshold, making network final jump out of local minimum, obtains the optimal solution, the error between the obtained results expected results is only 1.66279e-026.Finally, design content of Chinese style cigarette blending formula design was introduce in this paper, thus, according to the existing tobacco raw data and the research conclusion, decide to use genetic algorithm (GA) to design three optimum cigarette blending formula designs. Then using the genetic algorithm, combine the hybrid optimization of BP neural network to forecast the sensory-quality of cigarette of the three groups of cigarette blending formula. Finally, comparing the prediction result with evaluation result of expert, eventually determines which group of leaf groups formula optimal. The final results shows, this paper starting from the design goal established with the cost for target function, origin as constraint conditions of mathematical model, through the introduction of the penalty function ideas, solving the binding optimization problems that genetic algorithms can¡¯t solve. Finally determine three groups of optimal formula scheme. By GA-BP method to forecast the sensory-quality of cigarette of the three groups of cigarette blending formula.
Related Dissertations
Last updated
Sponsored Links
Home |About Us| Contact Us| Feedback| Privacy | copyright | Back to top