| With the rapid development of textile industry technology, automation,intelligent instrument has been used in large scale. However, there are still some problems in the process of production, such as backward characterization methods,not highly adaptive intelligent degree during the control process, intelligent model uncertainty during the production process, and so on.In this article, according to the related issues of the cotton textile, we will research in complex adaptive characterization, product quality of intelligent prediction, dimensionality reduction processing of multidimensional problem,product quality prediction and evaluation, virtual manufacturing model based on the Agent. The mainly studies and explores in this paper are summarized as follows:(1) According to the strong openness of Petri net and the ability of handling multi factor problem, complex adaptive Petri net(CAPN) model is proposed and formal definition and reachability tree validation for the model on the basis of the characterize of its model constructed for multi dimension problem and its interactive model. At the same time, considering the influence factors, such as complexity of factors, complexity of relationship, complexity of environment and so on, we construct and operate the system complexity model. The results show that the ability of CAPN to solve the problem in a single basic unit is greatly improved, such as1.67 times in complexity of the elements, 1.5 times in complexity of the relationship,1.6times in complexity of the system, which means the whole system characterization ability enhanced to 1.6n times(n is the number of basic units)through CAPN. We compared the complexity of the two-processes by actual data,the result show that the absolute ratio of element complexity Ca, relationship complexity Cr and system complexity C is respectively reduced by 14%, 17% and14% after applying CAPN model.That is, the whole relative ratio of complexity is decreased by 40%. So it is proved that the CAPN model can be used to optimize the model and reduce the complexity of the model.(2) We define the trigger boundary of transition and network for CAPN,construct the energy flow model of continuous production and discrete production,and then propose the optimization strategy for the model. A single process of cotton spinning production is selected to model and test experimental data. The results showed that the system time of iteration through optimization model is 33.3%shorter than the original model. Combined with the prior period of data processing,the overall performance of the optimized system is proved to improve by 23%, and the computational efficiency of this system can be improved obviously by using the optimization strategy.(3) Firstly, the instance data of cotton quality index is pretreated by reducing dimension and classifying. Then neural network model is output for prediction and a hybrid intelligent index model is established. This model is a quality index analysis model combining with the grey correlation analysis theory, principal component analysis method and expert knowledge base. The result of experiments show that this hybrid model could reduce the raw cotton index from 15 to 6 during the process of cotton spinning production. The degree of dimensional reduction is up to 60 %.This method can improve the calculated efficiency to 60 % without lowering the accuracy. This process can be carried out to get major contribution index data in the stage of data acquisition. This focuses on the 40% data quality, which influence the result. The accurate results(+ 2 % error range) can be obtained by the prediction and the data acquisition accuracy can be improved by this method.A two-step information filtering algorithm is used in this article. The simulation result is improving about 10% by information accuracy than Bayes Classification algorithm. Compared with Decision-Theoretic Rough Sets(DTRST), the error ratio for finding new sample or unknown sample is reducing about 5%, both of that could prove the better accuracy of this algorithm.(4) To overcome problem of the result with large fluctuation caused by uncertainty of initial weights, a GA- BP Neural Network Algorithm is proposed in this article, in which BP neural network is optimized by genetic algorithm. The yarn quality index prediction and backward prediction were also carried out by using BP Neural Network and GA- BP algorithm. The actual tests show that GA- BP algorithm has the following advantages: 1) Convergence speed is significantly improved. Compared with the BP neural network, the efficiency is respectively improved by 55 % and 16 % under the condition of forward prediction model andreverse quality assessment model through the GA- BP algorithm. 2) Accuracy of prediction is also significantly improved. The accuracy of prediction is respectively improved by 26 % and 28 % under the condition of forward prediction model and reverse quality assessment model through the GA- BP algorithm. 3) The stability of prediction is very good. The error ratio is respectively lower 5% and 4.65 % under the condition of forward prediction model and reverse quality assessment model through the GA- BP algorithm, which is improved by 21% compared with BP network model(5.87%).(5) According to the characteristics of flexible manufacturing process, we put forward the framework of MAS-CAPN model. MAS-CAPN model can enhance the Agent capabilities of learning, interaction, collaboration through the process of positive data to generate knowledge and the reverse process of the knowledge to guide the Agent, which improves the learning ability the whole system coordination of Agent. X-KQML communication mechanism is proposed during the MAS-CAPN model based on message transmission. A MAS cross addressing method is also proposed, which method will serve Agent communication and operating Agent communication. According to the logical scope, this method effectively avoids length of address chain, low efficiency of addressing during the frequent activities in local area.(6) Based on the MAS-CAPN model, virtual machining system is operated.This system implements different types of process such as single, combined, custom and so on. According to the original data, this system was realized the incomplete information classification of index, dimensionality reduction of index, targets and other data confirm the pretreated. This system has realized the built- in algorithm to calculate the pluralistic composition of Kosovo, and support a custom algorithm or other programmed algorithm. The advantages including high efficiency, stable operation flexible prediction algorithm are confirmed by test results.In conclusion, in this article, firstly, we propose the characterizing method of production flow for CAPN, construct the system complexity model and test the capacity of this model. We define the trigger boundary of transition and network for CAPN, construct the energy flow model of continuous production and discrete production, propose the optimization strategy for the model, and then test the improving calculating efficiency by this model. We propose a hybrid intelligent index model and a two-step information filtering algorithm, and test theeffectiveness of dimension reduction and classification accuracy. We propose a GA-BP Neural Network Algorithm, and test the advantage in prediction accuracy. We propose a MAS-CAPN model, virtual machining system is operated on the basis of this model. |
PDF Full Text Request