Prediction Of Cotton Yarn Quality Based On Multiple Combinations Of Neural Network

Most of China’s cotton spinning enterprises now basically rely on the experience to design process, the mid-experiment processing to optimize processing parameter and artificial passive test to control quality. This will result in the blindness of the production, the volatility and inefficient of the production efficiency.By Yu Weidong, Yang Jianguo of Donghua University as the representatives of the team have made research on the virtual intelligent processing technology, with the emergence of the "Virtual Worsted Woven Processing System", "Cotton Textile Intelligent Process Design Platform", "Cotton Yarn Quality Prediction and Control System", "Textile Intelligent Process Design and Quality Prediction System" and "Textiles Intelligent Technology Aided Design System of WebPQC". These virtual intelligent processing systems using multiple intelligent algorithms to a certain extent to solve the above problems. Based on the above research, this paper further studies and analyzes the characteristics of the algorithm model and the high precision modeling, the main results in the following five aspects:(1) The establishment of CBR case base and its algorithm improvementBased Reasoning Case(CBR) in the design of textile technology need to solve the problems such as the representation of the process structure, the extraction of process features, process similarity search algorithm and so on. According to the collected data this paper extracts features of the yarn: yarn specification(Nm), moisture(%), twist(twist /m), twist factor and breaking strength(cN/tex), and build the case library if a yarn in the feature vector. Based on the purpose of case-based reasoning retrieval method is to find the most similar cases in the case base comparing with the new case, the algorithm is the most important, and based on the original algorithm, the improved algorithm is proposed. The algorithm does not contain variables that are independent of the case characteristics, so that the algorithm is applicable..(2) Verification of principal component analysis(PCA) to improve the accuracy of the BP neural networkPrincipal component analysis(PCA), a multivariate statistical analysis method, which is used to reduce the dimension and transform the multi index into a few non relevant comprehensive indexes. In this paper, by using the principal component analysis 13 raw cotton indexes such as micronaire value(x1), metric number(x2), breaking strength(x3), maturity index(x4), body length(x5), quartile length(x6), base(x7), evenness(x8), short fiber ratio(x9), moisture regain(x10), trash content percentage(x11), seed fragments(x12) and broken seeds(x13) have reduced to 5 comprehensive index, namely the length uniformity, fineness, mechanics and length, moisture and foreign impurities and ginning, and give the linear equations.Many studies pointed out that the principal component analysis can reduce the dimension and reduce the dimension of the input of BP neural network, and can improve the prediction accuracy of BP neural network. In this paper, the performance curve, error distribution and MEP analysis of a single BP neural network(model 1) and PCA and BP neural network(model 2), have proved the principal component analysis can only reduce the dimension, but it’s no help to improve the accuracy. By the MEP analysis of the 3 models, principal component analysis is that can reduce dimension, but the growth in the number of hidden nodes illustrate that the function of the variables involved in the principal component itself is still studied.(3) Using genetic algorithm(GA) to optimize the training parameters of neural networkBecause the parameters of the single neural network in the prediction are difficult to achieve the desired results, many scholars use genetic algorithm(GA) to optimize the weights and thresholds of the neural network. But author uses the advantages of Matlab2014 a in artificial intelligence and neural network toolbox function to analyze the factors that affect the accuracy of BP neural network, including data distribution, hidden layer node number and training parameters settings of trainlm, and these factors become the object of GA optimization.The prediction precision of breaking strength(Y2) of the combination of PCA and GA optimized BP neural network(model P-3) was improved significantly comparing with principal component analysis and neural networks(model P-2), with MEP of 1.21%, improvement rate of 61.09%, and the MEP of twist CV(Y3) was up to 2.2%, with improvement rate of 48.19%(model 1) and 50.65%(model P-2). The prediction precision of evenness CV value(Y1) of the combination of FBS and GA optimized BP neural network(model F-3) was improved significantly comparing with forward and backward stepwise and neural networks(model F-2), with MEP of 0.85%, improvement rate of 61.71%, and the MEP of twist CV(Y3) was up to 1.86%, with improvement rate of 57.92%(model 1) and 58.39%(model F-2), all of these prove the importance of introducing genetic algorithm(GA).When using the neural network for predicting, with genetic algorithm optimizing the factors that affect the prediction results to get their best combination, the prediction results are much more ideal than using fixed parameter settings of the neural network.(4) The superiority of FBS as the input of combination of neural networkWhen combinating neural network the input dimension of network need reduce, the principal component of principal component analysis(PCA) and variable combinations of multiple stepwise regression(FBS) are two types of input. Through the comparative MEP analysis of the two models, the variable combination obtained by multiple stepwise regression(FBS) can more improve the accuracy, the reason for this is that the variables are some factors that are independent and influence factor weight coefficient largely.(5) Interface development using Matlab GUIFirstly uses the Matlab GUI toolkit for the development of cotton yarn quality prediction systemt, with exploring the way of system development and giving full play to the advantages of Matlab in the aspect of numerical calculation. In this paper, based on the collected data and the realization of predict function, the system divided into user management, data management and prediction model, including user management and data management sector can well realize management functions. The symtem can enable the users to realize well interaction with the computer and facilitate the operation.