【作者】 常丽霞；

【导师】 高卫东；

【作者基本信息】 江南大学 ， 纺织工程， 2013， 博士

【摘要】 服装流行色预测是指应用特定的方法判断并给出24个月后服装的色彩流行趋势定案。定案能为企业提供未来服装销售季节中色彩流行的相关信息，有益于服装企业合理制定生产方案和营销策略。服装流行色定案在传递中的滞后性和保守性制约了流行色商业价值的发挥，促动了服装流行色预测研究的发展。虽然一些理论和方法已被应用到此研究领域，并取得了一定的成果。但是，总体来说服装流行色预测研究还处于探索阶段，存在不足和争议，预测的有效性依然是纺织服装行业亟待解决的问题。色彩具有典型的视觉特征，文化学意义上的色彩流行具有模糊性，如何将色彩的感性转化为理性，构建比较严谨的数学模型是流行色趋势研究科学性的关键。本文首先探讨了将色彩转化为数据的量化方法，然后在此基础上展开了对服装流行色预测趋势的研究。本文以国际色彩委员会(INTERCOLOR)发布的2007-2013年国际春夏女装流行色定案为研究对象，通过对INTERCOLOR所使用的色彩体系PANTONE色彩体系的分析，提出了对色彩进行量化和分类的方法；通过对服装流行色定案统计数据的分析，给出了服装流行色色相、明度和纯度特征的变化。在上述基础上，分别采用灰色系统理论和BP神经网络构建了服装流行色色相预测模型，通过设计不同长度的时间序列，探讨了各色相预测模型的预测效果。最后，对服装流行色的明度和纯度进行了预测研究，最终实现了对服装流行色预测的系统性研究。本研究的创新点主要体现在：1）针对服装流行色预测研究中存在的色彩量化的无标准、不统一性，提出了基于PANTONE色彩体系的色彩量化和分类准则，为规范色彩的量化提供了参考，搭建了主观色彩与客观计算机色彩应用间的桥梁；2）探讨了不同长度的时间序列，即进行服装流行色预测的年份参数，对服装流行色预测模型预测效果的影响，为服装流行色流行预测研究提供了理论依据；3）提出了基于主导色相的服装流行色的明度和纯度的预测方法，丰富了服装流行色的研究理论体系。本文所做的内容主要包括：第一章为绪论部分。对服装流行色预测知识、研究现状进行介绍和评述。通过对国内外服装流行色预测研究文献的分析，概述了这一研究领域达到的研究水平、存在的不足，继而提出本文的研究目的、研究思路和研究方法。第二章研究了服装流行色定案的量化与分析。针对色彩的主观性和研究中存在数据来源不统一、色彩量化复杂等问题，提出以国际色彩委员会发布的2007-2013年国际春夏女装流行色定案为研究对象，以PANTONE色彩体系为色彩量化和分类的工具。确定了对PANTONE色彩体系的色相、明度、纯度区间分类的依据和标准，通过对量化数据分析获得定案的特征。第三章探讨了基于灰色系统理论的服装流行色色相的预测研究。根据色相在定案中的重要性，以及色相数据变化的非线性特征，提出采用灰色系统理论构建服装流行色色相预测模型的研究。通过设计不同长度的时间序列：4年、5年和6年，探讨了时间序列长度对预测模型预测效果的影响，研究以平均绝对误差(MAE)作为预测效果的评价指标。但灰色GM(1,1)模型对数据光滑性要求较高，进行色相预测时不能对所有数据完成预测，其普适性受到限制。第四章为基于BP神经网络理论的服装流行色色相的预测研究。针对灰色预测模型对服装流行色进行预测中存在的不足，构建了基于BP神经网络的服装流行色色相预测模型。通过分别设计输入层节点数为3、4和5，目标输出节点数为1的BP神经网络结构，系统地探讨了不同结构的BP神经网络模型对服装流行色色相的预测效果。第五章和第六章分别进行了服装流行色明度和纯度的预测研究。针对已有研究中缺少色相、明度和纯度内在联系不足，提出了建立在主导色相基础上的明度均值和纯度均值的预测研究。分别以定案中10类色相对应的明度和纯度数据为研究基础，完成了服装流行色明度和纯度的预测。第七章为结论与展望。给出了本文获得的主要结论，同时指出研究中尚存的问题以及未来研究中需要拓展的方向和思路。本文以服装流行色定案为研究对象，提出了对色彩进行量化和分类处理的方法。在对服装流行色定案数据的分析基础上，利用灰色系统理论和神经网络模型对服装流行色定案进行了预测研究，实现了对服装流行色较高精度的预测，研究结果可以为纺织服装行业掌握未来的流行色趋势变化提供参考。更多还原

【Abstract】 Fashion color prediction is to predict the future color trend of fashion in24months byspecial methods and to provide trend palettes. Obtained palettes could offer clothingenterprises valuable information on fashion color during the future selling seasons, which isbeneficial for drafting manufacturing and marketing plots. However, the lag and confidence ofpalettes during propagating restrict its commercial value and stimulate the development ofresearch on fashion color prediction. Though several theories and methods have been appliedinto this field and achievements are obtained, fashion color prediction is still in exploringstage. Deficiencies and controversies existed and the validity of prediction has already been apressing problem of the day in the textile and clothing industries.Color is with typically sensitive simulation and fashion color with ambiguous expression.How to translate the sensitive color into rational data and build serious mathematicalprediction models are the key points of fashion color prediction. In this research, the methodof color quantification was discussed, based on which the prediction researches on fashioncolor were investigated. Fashion color palettes were taken as the objects. The palettes werereleased by International Commission for Color Fashion and Textiles (INTERCOLOR) forwomen’s spring/summer, from2007to2013. Quantification and classification methods oncolors were put forward. The process was based on the analyses of PANTONE color system,which was applied by the palettes. Then features of palettes on hue, lightness and chromawere obtained respectively analyzing the statistical data. Consequently, hue prediction modelswere established using grey system theory and back-propagation neural networks respectively.The predicted effects of the models were investigated by setting different time series. Finally,lightness and chroma prediction models were established on the basis of hue. The systematicprediction on fashion color was developed.The innovations of this thesis are illustrated as follows:1) Quantification andclassification of color based on PANTONE color system were proposed in terms of thenon-standard and disunity of color quantification in fashion color prediction researches. It canprovide a valuable reference for the standardization of color and improve the communicationbetween colors on subjective perception and the objective computation application.2) Thepredicted effects of the prediction models were investigated by setting different time series,which can provide soundly basis for fashion color prediction researches.3) Prediction methodon lightness and chroma of color was put forward based on the dominant hues, which enrichesthe theory on fashion color research. The main studies of this paper were arranged as follows.Preface of the thesis was illustrated in Chapter1. Research background, methodologyand literature was introduced. Research levels and insufficiencies in this field are summarized.Consequently, the purpose, significance and methods of this research were proposed.Quantification and analyses of fashion color palettes were discussed in Chapter2.Inercolor palettes for women’s spring/summer, from2007to2013, were proposed as objectsand PANTONE color system was as the quantification tool, against the problems of disunityof data resources and complexity of quantification methods. Quantification and classification criteria of hue, lightness and chroma were defined. Consequently, features of palettes wereobtained by analyzing statistical data of hue, lightness and chroma.The predictions of hues using grey system theory were investigated in Chapter3.Prediction models on hue were established by grey model, in accordance with the significanceand the nonlinear features of hue. Validities of the prediction models were discussed bysetting different time series of historical fashion color data, as4,5and6years respectively.Mean absolute error (MAE) was considered as evaluation indicator in this thesis. Resultsdemonstrated that GM (1,1) cannot predict the hues continuously due to the high requirementon historical data and limit the generality of the prediction.The predictions on hue by using back-propagation neural networks (BPNN) wereprobed in Chapter4, in accordance with the deficiency of grey system theory. The validity ofthe prediction models were explored systematically by designing different structures of BPNNwith input neurons as3,4and5respectively.Consequently, prediction on lightness and chroma were investigated in Chapter5and6.Prediction models on mean values of lightness and chroma, based on dominant hues, wereproposed against the absence of internal relation among the three attributes of colors.Finally, the main achievements of the research were summed up in Chapter7of thethesis, meanwhile problems to be solved and explored in the future researches were pointedout.Quantification and classification methods were provided based on fashion color palettesin this thesis. Fashion color prediction was discussed by using grey system theory and BPNN,which was based on analyses of statistical data of fashion color palettes. It conducted theprediction with a higher accuracy and the prediction results could provide textile industriesreference on future fashion color trend.更多还原