Research On Classificationalgorithm Of Batch Customization Garment Size

Batchcustomization is the new mode of production of garment enterprises that meet the needs of customers with low cost,fast and high quality and other characteristics.Size classification is an important prerequisite for batch customization production.In the traditional archiving methods,the number archives is according to the manual experience,the national standard and the enterprise’s own specifications,which is prone to data lost,wrong and long work time and it also increases business cost.In recent years,with the development of the apparel industry,the custom mode become the mainstream of production,it is the high coscharacteristics of a single production.There is little research on classification algorithm of batch customization,soit provides some reference value for improving the classification efficiency of customizationgarment and the fit degree of garment and establishireasonable number of plate based on classification algorithm of batch customization.the research object is classification algorithm of batch customization in this paper.Discuss the classification quality of K-means clustering with standard and non-standard initial clustering center and BackPropagation with six classified variables such as height,bust circumference,waist circumference,collar circumference,shoulder breadthand arm lengthbased on literature and enterprise survey.The main contents and conclusions of this research are as follows:(1)Research on size classification by improved K-means clustering algorithm.After data arepreprocessed,this experiment used the improved K-means clustering algorithmaccording tomaximum and minimum distance method selected the first initial clustering center of the classicalK-means clustering algorithm.Combined the Calin-ski-Harabasz index,the coefficient of variation and the relative deviation,the final clustering effect was evaluated and the clustemumber of non-GB methodwas determined according to the principle of the closest value of CH.The results show that there are no missing number and no abnormal number in the six control parts of 1000 data,and the data conformed to the criterion of normality,it can be followed by K-means clustering analysis and neural network data analysis.The numbers of clusters of the national size initial clustering centers was 25 by using SPSS software for data analysis,the 25 non-GB initial clustering centers was selected by suing the maximum and minimum distance method.According to the evaluation index,when the number of non-GB cluster is 13,the CH number of the national standard is the closest to the non-GB method,which plate number of the non-GB was less than the GB classification numbers,and the coefficient of variation and the relative deviation of the non-GB are less than the national standard.The result indicated that the deviation degree of sample and plate by classifying with non-GB is small,the clustering effect is better.(2)Research on size classification bythe BP neural network.Using Confusion Matrix estimated the classification performance through the traingd,traingdm,traingrpand Levenberg-Marquardt method of BP neural network,andanalyzing the reasons of the highest correct rate and the lowest accuracy rate and the low recognition rate.The results show that the recognition correct rateis more higher when hidden layer neure number is more and more for same method.The recognition rate of Levenberg-Marquardt method is highest for same neure number.The recognition rate of the purelin transfer function of the Levenberg-Marquardt method is similar to that of the tansig transfer function,which is up to 97%.As for traingrp,the recognition rate of the logsig transfer function of this method is highest,and the recognition rate of traingd and traingrp isvery low and they are equal nearly,soit does not have real value.Compared Levenberg-Marquardt with other three methods,time-consuming is longer,the reason is that the matrix of this methodtaken up more memory,so it has some requirements to system.Considering the main factors of recognition rateand the secondary factors of time-spending,Levenberg-Marquardt method has the best classification effect and it is suitable for the application of BP neural network,and it has a certain practical application significance.