Prediction of breathability performance of household apparel based on GRA-GA-BP neural network
摘要:
本文構(gòu)建了一種改進(jìn)BP神經(jīng)網(wǎng)絡(luò)模型來預(yù)測家居服面料的透氣性能,能為家居服設(shè)計提供重要的參考。首先,采用灰色關(guān)聯(lián)分析法(Grey Relation Analysis,GRA),選擇與透氣率關(guān)聯(lián)度較大的因素作為研究對象。其次,采用遺傳算法(GA)優(yōu)化BP神經(jīng)網(wǎng)絡(luò)的結(jié)構(gòu)參數(shù),構(gòu)建基于灰色關(guān)聯(lián)分析的遺傳算法優(yōu)化BP(GRA-GA-BP)神經(jīng)網(wǎng)絡(luò)預(yù)測模型。選取58種面料成分不同、織物組織各異的家居服面料,其中42種為模型訓(xùn)練樣本,16種為測試樣本對建立的模型進(jìn)行驗證。實驗結(jié)果表明,透氣率實測值與預(yù)測值平均相對誤差為8.39%;對透氣率實測值與預(yù)測值進(jìn)行相關(guān)性分析,擬合優(yōu)度R2為0.976。研究表明,該預(yù)測模型預(yù)測效果良好、預(yù)測精度高,在一定程度上可以精準(zhǔn)預(yù)測家居服面料的透氣率。
關(guān)鍵詞:
織物;家居服;灰色關(guān)聯(lián)分析;改進(jìn)BP神經(jīng)網(wǎng)絡(luò);透氣性預(yù)測
中圖分類號:
TS101.923.4
文獻(xiàn)標(biāo)志碼:
A
文章編號: 10017003(2024)10期數(shù)0046起始頁碼07篇頁數(shù)
DOI: 10.3969/j.issn.1001-7003.2024.10期數(shù).005(篇序)
收稿日期:
20240320;
修回日期:
20240912
基金項目:
國家自然科學(xué)基金項目(52203276)
作者簡介:
王彬霞(1997),女,碩士研究生,研究方向為服裝舒適性研究。通信作者:王春紅,教授,wangchunhong@tiangong.edu.cn。
隨著人們生活水平的提高,人們對于家居服的舒適性有了更廣泛的關(guān)注和更高的要求,而影響家居服面料舒適性的關(guān)鍵因素之一是織物的透氣性[1]。當(dāng)面料的兩側(cè)有壓力差異時,它們的通風(fēng)特性就被稱為透氣性[2-3]。影響面料透氣率的因素有纖維間的空隙、紗線直徑、織物密度、厚度、平方米質(zhì)量、織物組織結(jié)構(gòu)及面料組成成分等[4-5]。目前國內(nèi)外對于家居服研究的論文較少,家居服的舒適性研究還比較欠缺,對于透氣性的研究則更為重要。邵景峰等[6]構(gòu)建了一種基于支持向量機(jī)的精紡毛織物透氣性預(yù)測模型,與現(xiàn)有BP神經(jīng)網(wǎng)絡(luò)預(yù)測模型相比,其預(yù)測精度提高了3%;Zhu等[7]建立了幾種預(yù)測透氣性的分析模型,發(fā)現(xiàn)Hagen-Poiseuille方程比其他模型具有更好的預(yù)測性;徐瑤瑤等[8]擬合織物透氣率與紗線線密度、織物經(jīng)緯紗密度、孔徑dp之間的函數(shù)關(guān)系建立擬合函數(shù)來預(yù)測全棉織物的透氣性,所計算的實測值與預(yù)測值相關(guān)系數(shù)高。本文的研究對象屬于高度非線性問題,因此對于家居服面料透氣性能的預(yù)測更適合用神經(jīng)網(wǎng)絡(luò)來完成。然而,經(jīng)典的BP神經(jīng)網(wǎng)絡(luò)(Back-Propagation,BP)的訓(xùn)練與預(yù)測性能較差,并且容易出現(xiàn)陷入局部最優(yōu)解的情況,其他神經(jīng)網(wǎng)絡(luò)不適用于本文[9]。遺傳算法(Genetic Algorithm,GA)是一種模擬自然進(jìn)化過程以尋找最優(yōu)解的方法,適用于解決復(fù)雜的組合優(yōu)化問題,因為它能迅速提供較好的優(yōu)化結(jié)果[10-11]。利用遺傳算法改進(jìn)BP神經(jīng)網(wǎng)絡(luò)不僅能有效避免BP神經(jīng)網(wǎng)絡(luò)在選擇過程中的隨機(jī)性缺陷,還能顯著加快其收斂速度,進(jìn)而提升模型的預(yù)測精度和穩(wěn)定性[12-13]。
綜上所述,本文選取了涵蓋不同面料成分、不同織物組織的共58種家居服面料,首先采用灰色關(guān)聯(lián)分析,選取對家居服面料透氣率較大的影響因素作為研究對象,即作為預(yù)測模型的輸入層,然后構(gòu)建基于灰色關(guān)聯(lián)分析的遺傳算法優(yōu)化BP(GA-BP)神經(jīng)網(wǎng)絡(luò)預(yù)測模型,最后對構(gòu)建的模型進(jìn)行驗證。該模型可以完成對市場上不同類型家居服面料透氣性能的預(yù)測,在一定程度上節(jié)約了測試成本和時間,同時可為家居服的設(shè)計提供參考。
1 家居服面料透氣率影響因素的灰色關(guān)聯(lián)分析
織物透氣性與織物的經(jīng)緯紗密度、經(jīng)緯紗表觀直徑、織物厚度、單位體積質(zhì)量纖維性質(zhì)、紗線結(jié)構(gòu)和織物組織結(jié)構(gòu)等因素有關(guān)[4-5]。但各影響因素對家居服面料透氣率并未產(chǎn)生規(guī)律性的線性影響,因此,可以將其看作灰色系統(tǒng),而灰色關(guān)聯(lián)分析可對此類非確定性的動態(tài)過程發(fā)展態(tài)勢進(jìn)行量化分析,進(jìn)而得到各影響因素對家居服面料的透氣率的主次關(guān)系,找到主要影響因素[14]。
灰色關(guān)聯(lián)分析是一種衡量兩個因素關(guān)聯(lián)程度的方法,該方法基于因素之間發(fā)展趨勢的相似性或相異性。若兩個因素在系統(tǒng)的發(fā)展過程中相對變化大體相同,則它們之間的關(guān)聯(lián)度大;相反,關(guān)聯(lián)度就相對較?。?5-16]。利用灰色關(guān)聯(lián)分析法,對家居服透氣率各影響因素與透氣率的關(guān)聯(lián)度進(jìn)行計算和分析。
1.1 歸一化處理
首先確定系統(tǒng)因變量數(shù)列(透氣率Y)和自變量數(shù)列(影響透氣率的各因素),組成原始數(shù)據(jù)矩陣。因為各因素的數(shù)據(jù)之間存在較大差異,為了增加數(shù)據(jù)的可比性,對數(shù)據(jù)進(jìn)行標(biāo)準(zhǔn)化處理,從而得到標(biāo)準(zhǔn)化矩陣,i為指標(biāo)數(shù)量,k為樣本數(shù)量。
wki=xki-minixkimaxixki-minixki(1)
式中:maxixki與minixki分別為第i個指標(biāo)的最大值、最小值。
1.2 關(guān)聯(lián)系數(shù)計算
對因變量數(shù)列和自變量數(shù)列的關(guān)聯(lián)系數(shù)ξi(k)進(jìn)行計算,ξi(k)代表自變量數(shù)列與因變量數(shù)列在各個時間點(diǎn)的關(guān)聯(lián)程度值,其值越大,表示該因素在相應(yīng)時間點(diǎn)的影響力越顯著。對于一個因變量數(shù)列x0(k)(透氣率),有5個自變量x1(k)(經(jīng)/橫密),x2(k)(緯/縱密),x3(k)(紗線直徑),x4(k)(厚度),x5(k)(平方米質(zhì)量),其中k=1,2,…,5。各個自變量與因變量的關(guān)聯(lián)系數(shù)ξi(k)計算如下:
ξi(k)=mini(Δi(min))+ζmaxi(Δi(max))|x0(k)-xi(k)|+ζmaxi(Δi(max))(2)
式中:ζ為分辨系數(shù),ζ∈[0,1],一般取ζ=0.5;mini(Δi(min))與maxi(Δi(max))分別是兩級最小差、最大差。
mini(Δi(min))=minimink|x0(k)-xi(k)|(3)
maxi(Δi(max))=maximaxk|x0(k)-xi(k)|(4)
1.3 關(guān)聯(lián)度計算
用平均值關(guān)聯(lián)度ri來表示自變量數(shù)列與因變量數(shù)列之間的關(guān)聯(lián)程度,ri越大,因素的影響力就越大[17]。
ri=1N∑Nk=1ζi(k)(5)
按照上述灰色關(guān)聯(lián)分析方法,確定各影響因素對家居服面料透氣率的關(guān)聯(lián)度ri。
1.4 灰色關(guān)聯(lián)分析結(jié)果
透氣率各影響因素與家居服面料透氣率關(guān)聯(lián)度的計算結(jié)果表明,經(jīng)/橫密、緯/縱密、紗線直徑、厚度、平方米質(zhì)量共5個影響因素對透氣率的關(guān)聯(lián)度分別為0.76、0.70、0.68、0.73、0.62,與透氣率的關(guān)聯(lián)度都較高。因此,本文選擇這5個影響因素作為GA-BP神經(jīng)網(wǎng)絡(luò)的輸入?yún)?shù),對測試樣本的透氣率進(jìn)行預(yù)測。
2 實 驗
2.1 材 料
選取不同組織結(jié)構(gòu)、不同面料成分的家居服織物共58種(深圳全棉時代科技有限公司)。1#~18#試樣為100%棉機(jī)織平紋織物,19#~22#試樣為100%棉機(jī)織緞紋,23#~29#試樣為100%棉針織羅紋,30#~31#試樣為100%棉緯平針,32#~35#試樣為100%棉機(jī)織平紋,36#試樣為100%棉機(jī)織斜紋,37#試樣為100%棉緯平針,38#~42#試樣為95%棉5%氨綸襯緯組織,43#~46#試樣為95%棉5%氨綸緯平針,47#試樣為95%棉5%氨綸毛圈組織,48#試樣為44%棉49%莫代爾7%氨綸緯平針,49#試樣為4%棉49%莫代爾7%氨綸緯平針,50#試樣為95%聚酯纖維5%氨綸機(jī)織緞紋,51#試樣為92%黏膠8%氨綸羅紋,52#試樣為100%聚酯纖維機(jī)織緞紋,53#~54#試樣為96%聚酯纖維4%氨綸機(jī)織緞紋,55#試樣為91%聚酯纖維9%氨綸機(jī)織緞紋,56#試樣為46%棉54%莫代爾機(jī)織緞紋,57#試樣為31%棉37%莫代爾32%聚酯纖維針織芝麻點(diǎn)提花,58#試樣為36%棉38%莫代爾26%聚酯纖維針織芝麻點(diǎn)提花織物。其中1#~42#為訓(xùn)練集樣本(表1),43#~58#為測試集樣本。
2.2 方 法
2.2.1 織物密度測試
機(jī)織物:根據(jù)標(biāo)準(zhǔn)GB/T 4668—1995《機(jī)織物密度的測定》,將機(jī)織物平整地鋪在工作臺上,并運(yùn)用織物密度鏡來測量。具體操作為:在機(jī)織物上選取10 cm的長度,然后計算這一長度內(nèi)經(jīng)向的緯紗根數(shù)和緯向的經(jīng)紗根數(shù)[18]。
針織物:按照標(biāo)準(zhǔn)ASTM D 3887《針織物密度》,計算針織物橫縱向(橫向)10 cm長度內(nèi)的線圈數(shù)[19]。
2.2.2 紗線直徑測試
使用YG002型纖維細(xì)度綜合分析儀(溫州市大榮紡織儀器有限公司)來測量織物中紗線的表觀直徑,同一樣品分別取經(jīng)/縱向和緯/橫向的紗線進(jìn)行測試[20]。
2.2.3 織物厚度測試
使用YG(B)141D數(shù)字式織物厚度儀(溫州市大榮紡織儀器有限公司),根據(jù)GB/T 3820—1997《紡織品和紡織制品厚度的測定》標(biāo)準(zhǔn),將壓腳面積設(shè)定為2 000 mm2,施加200 cN的砝碼質(zhì)量,加壓壓力為1 kPa,加壓時間為10 s。對同一樣品進(jìn)行10次測試,并計算平均值[21]。
2.2.4 織物平方米質(zhì)量測試
根據(jù)GB/T 4669—2008《紡織品機(jī)織物單位長度質(zhì)量和單位面積質(zhì)量的測定》標(biāo)準(zhǔn),對織物進(jìn)行了平方米質(zhì)量測量。選取不同位置裁剪10 cm×10 cm的試樣重復(fù)測量5次,以平均值作為最終結(jié)果[22]。
2.2.5 織物透氣率測試
根據(jù)GB/T 5453—1997《紡織品織物透氣性的測定》的標(biāo)準(zhǔn)要求,使用YG(B)461D-Ⅱ型數(shù)字式織物透氣量儀(深圳市方源儀器有限公司),在預(yù)設(shè)的壓差條件下,測量規(guī)定時間內(nèi)試樣上垂直通過特定面積的氣流流量。每種樣品進(jìn)行10次測試,取平均值。實驗時,所選用的試樣面積為20 cm2,實驗壓差設(shè)定為100 Pa,并選用合適的噴嘴進(jìn)行測定[23]。
3 家居服面料透氣性預(yù)測模型建立
3.1 預(yù)測模型原理及過程
BP神經(jīng)網(wǎng)絡(luò)因其強(qiáng)大的學(xué)習(xí)能力和預(yù)測功能,以及較小的計算量和簡單的結(jié)構(gòu)而受到關(guān)注[24-27]。然而,BP神經(jīng)網(wǎng)絡(luò)也存在一些缺點(diǎn),如學(xué)習(xí)速度慢和容易陷入局部最小值,導(dǎo)致結(jié)果可能不是最優(yōu)解。與此同時,遺傳算法在解決復(fù)雜的組合優(yōu)化問題時表現(xiàn)出色,能夠迅速獲得優(yōu)質(zhì)的優(yōu)化結(jié)果[12]。因此,通
過結(jié)合遺傳算法,可以有效地解決BP神經(jīng)網(wǎng)絡(luò)在選擇上的隨機(jī)性缺陷,加快其收斂速度,并提升模型的預(yù)測精度和穩(wěn)定性?;诖?,本文采用GA對BP神經(jīng)網(wǎng)絡(luò)的初始權(quán)值和閾值分布進(jìn)行優(yōu)化,然后將優(yōu)化后的神經(jīng)網(wǎng)絡(luò)用于家居服面料透氣率預(yù)測[28]。
GA優(yōu)化BP神經(jīng)網(wǎng)絡(luò)的過程主要為:BP神經(jīng)網(wǎng)絡(luò)先初始化權(quán)值和閾值,然后對權(quán)值和閾值進(jìn)行編碼。個體的適應(yīng)度函數(shù)用BP神經(jīng)網(wǎng)絡(luò)預(yù)測誤差絕對值之和表示,如下式所示,適應(yīng)度函數(shù)值越小,則表示訓(xùn)練越準(zhǔn)確,模型的預(yù)測精度更好[28]。
Fi=(∑n1abs(yi-oi))(6)
式中:yi、oi分別為第i個節(jié)點(diǎn)的期望輸出和預(yù)測輸出。
3.2 模型參數(shù)確定
根據(jù)上述家居服面料透氣率影響因素的灰色關(guān)聯(lián)分析結(jié)果,選取篩選出的5個因素作為神經(jīng)網(wǎng)絡(luò)的輸入變量,家居服面料實測透氣率為輸出變量。即輸入層神經(jīng)元個數(shù)為6,輸出層神經(jīng)元個數(shù)為1,根據(jù)經(jīng)驗公式和Kolmogorov定理[29],隱藏層神經(jīng)元個數(shù)取值范圍為4~21,經(jīng)反復(fù)測試,確定隱含層節(jié)點(diǎn)數(shù)為6。遺傳算法參數(shù)設(shè)置如下:種群規(guī)模N=10,交叉概率Pc=0.4,變異概率Pm=0.05,最大迭代次數(shù)為20,選擇方式采用輪盤賭法[30]。
4 模型驗證
重新選取16種面料成分不同、織物組織各異的家居服面料,利用Matlab軟件的神經(jīng)網(wǎng)絡(luò)工具箱實現(xiàn)GA-BP的優(yōu)化和構(gòu)建,進(jìn)而完成了家居服面料透氣性能的預(yù)測,預(yù)測結(jié)果如表2所示。透氣率實測值與預(yù)測值相對誤差在0.80%~28.53%,平均相對誤差為8.39%,可知預(yù)測誤差很小,預(yù)測精度良好。
將透氣率實測值與預(yù)測值進(jìn)行對比,發(fā)現(xiàn)實測值與預(yù)測值曲線基本一致,如圖1所示,再一次驗證GA改進(jìn)BP神經(jīng)網(wǎng)絡(luò)模型的預(yù)測精度良好。
在OriginPro軟件中,將透氣率實測值與預(yù)測值進(jìn)行擬合,擬合優(yōu)度R2為0.976,擬合效果很好,再次驗證該預(yù)測模型預(yù)測的透氣率與實測值極強(qiáng)相關(guān),如圖2所示。
5 結(jié) 論
為了能實現(xiàn)基于織物結(jié)構(gòu)參數(shù)的織物透氣性能預(yù)測,本文采用灰色關(guān)聯(lián)分析篩選了5種關(guān)聯(lián)度較高的織物結(jié)構(gòu)參數(shù):經(jīng)/橫密、緯/縱密、紗線直徑、厚度、平方米質(zhì)量。以該5個影響因素作為BP神經(jīng)網(wǎng)絡(luò)的輸入?yún)?shù),用遺傳算法優(yōu)化BP神經(jīng)網(wǎng)絡(luò)的結(jié)構(gòu),從而實現(xiàn)了織物透氣性的預(yù)測。結(jié)果顯示:透氣率實測值與預(yù)測值的平均相對誤差為8.39%,透氣率實測值與預(yù)測值線性擬合度R2為0.976。該結(jié)果證明了本文提出的織物透氣性預(yù)測方法具有很高的可行性,可以為家居服的設(shè)計提供重要的參考依據(jù)。
《絲綢》官網(wǎng)下載
中國知網(wǎng)下載
參考文獻(xiàn):
[1]沈妙音. 抗菌熱濕舒適性服用機(jī)織物的研究與開發(fā)[D]. 杭州: 浙江理工大學(xué), 2021.
SHEN M Y. Research and Development of Antibacterial Thermo-wet Comfort Woven Fabric[D]. Hangzhou: Zhejiang Sci-Tech University, 2021.
[2]楊鑫, 陳慧敏, 柯達(dá), 等. 基于ALE方法研究面料透氣性對非接觸吸附效果的影響[J]. 東華大學(xué)學(xué)報(自然科學(xué)版), 2023, 49(6): 104-110.
YANG X, CHEN H M, KE D, et al. Effect of fabric permeability on non-contact adsorption based on ALE method[J]. Journal of Donghua University (Natural Science), 2023, 49 (6): 104-110.
[3]張鳳彩, 程小明, 呂治家, 等. 面料吸濕速干性能的影響因素研究[J/OL]. 棉紡織技術(shù), 2024: 61.1132.TS.20231013.1704.002.html.
ZHANG F C, CHENG X M, L Z J, et al. The influence factors of fabric moisture quick-drying properties research[J/OL]. Cotton Textile Technology, 2024: 61.1132.TS.20231013.1704.002.html.
[4]余佳文. 運(yùn)動型羊毛/滌綸針織物熱濕舒適性研究與綜合評價[D]. 上海: 東華大學(xué), 2023.
YU J W. Research and Comprehensive Evaluation on Thermal and Wet Comfort of Sports Wool/Polyester Knitted Fabric[D]. Shanghai: Donghua University, 2023.
[5]SANCHEZ-BOTERO L, AGRAWALA A, KRAMER-BOTTIGLIO R. Stretchable, breathable, and washable fabric sensor for human motion monitoring[J]. Advanced Materials Technologies, 2023, 8(17): 2300378.
[6]邵景峰, 王希堯. 基于支持向量機(jī)的精紡毛織物透氣性預(yù)測[J]. 毛紡科技, 2019, 47(8): 66-72.
SHAO J F, WANG X Y. Prediction of breathability of worsted fabrics based on support vector machines[J]. Woolen Textile Technology, 2019, 47(8): 66-72.
[7]ZHU G, FANG Y, et al. Prediction of structural parameters and air permeability of cotton woven fabric[J]. Textile Research Journal, 2018, 88(14): 1650-1659.
[8]徐瑤瑤, 朱俐莎, 杜磊, 等. 棉織物透氣性能的預(yù)測研究[J]. 現(xiàn)代紡織技術(shù), 2015, 23(3): 26-30.
XU Y Y, ZHU L S, DU L, et al. Prediction of breathability of cotton fabrics[J]. Advanced Textile Technology, 2015, 23(3): 26-30.
[9]馬海志. BP神經(jīng)網(wǎng)絡(luò)的改進(jìn)研究及應(yīng)用[D]. 哈爾濱: 東北農(nóng)業(yè)大學(xué), 2016.
MA H Z. Improvement Research and Application of BP Neural Network[D]. Harbin: Northeast Agricultural University, 2016.
[10]AIVALIOTIS-APOSTOLOPOULOS P, LOUKIDIS D. Swarming genetic algorithm: A nested fully coupled hybrid of genetic algorithm and particle swarm optimization[J]. PLoS One, 2022, 17(9): 0275094.
[11]NISHIDA K. Kernel density estimation by genetic algorithm[J]. Journal of Statistical Computation and Simulation, 2023, 93(8): 1263-1281.
[12]XU F, MO L, CHEN H, et al. Genetic algorithm to optimize the design of high temperature protective clothing based on BP neural network[J]. Frontiers in Physics, 2021, 9: 600564.
[13]石明, 汪舟, 甘進(jìn), 等. 基于GA-BP神經(jīng)網(wǎng)絡(luò)的噴丸樣品表層硬度預(yù)測模型[J]. 表面技術(shù), 2022, 51(1): 332-338.
SHI M, WANG Z, GAN J, et al. Microhardness prediction model of peened parts based on GA-BP neural network[J]. Surface Technology, 2022, 51(1): 332-338.
[14]ADAMU M A, SUMAILA M, DAUDA M, et al. Production and optimization of novel rice husk ash reinforced polycaprolactone/hydroxyapatite composite for bone regeneration using grey relational analysis[J]. Scientific African, 2023, 19: e01563.
[15]蔣榮超, 慈樹坤, 劉大維, 等. 基于灰色關(guān)聯(lián)分析的碳纖維增強(qiáng)樹脂復(fù)合材料控制臂鋪層優(yōu)化[J]. 復(fù)合材料學(xué)報, 2022, 39(1): 390-398.
JANG R C, CI S K, LIU D W, et al. Ply optimization of carbon fiber reinforced plastic control arm based on grey relational analysis[J]. Journal of Composite Materials, 2022, 39(1): 390-398.
[16]LIU S, ZHANG H, YANG Y. Explanation of terms of grey incidence analysis models[J]. Grey Systems: Theory and Application, 2017, 7(1): 136-142.
[17]水若倩. 基于灰色關(guān)聯(lián)分析法的美的集團(tuán)資產(chǎn)質(zhì)量研究[D]. 蘭州: 蘭州財經(jīng)大學(xué), 2022.
SHUI R Q. Research on Asset Quality of Midea Group Based on Grey Correlation Analysis[D]. Lanzhou: Lanzhou University of Finance and Economics, 2022.
[18]閆暢, 裴燕華. 一種高密度織物的織物密度測試方法研究[J]. 合成纖維, 2024, 53(1): 58-60.
YAN C, PEI Y H. Study on a fabric density testing method for high density fabrics[J]. Synthetic Fiber in China, 2019, 53(1): 58-60.
[19]丁雪婷, 王建萍, 潘婷, 等. 仿蜻蜓翅膀結(jié)構(gòu)的冬季針織面料研發(fā)及其性能[J]. 紡織學(xué)報, 2023, 44(9): 75-83.
DING X T, WANG J P, PAN T, et al. Development and performance of dragonfly wing structure like winter knitted fabrics[J]. Acta Textistrica Sinica, 2023, 44(9): 75-83.
[20]黃偉. 不同成紗結(jié)構(gòu)長絲紗的強(qiáng)力預(yù)測模型研究[D]. 上海: 東華大學(xué), 2017.
HUANG W. Research on Strength Prediction Model of Filament Yarn with Different Yarn Structure[D]. Shanghai: Donghua University, 2017.
[21]廖世豪, 張馨雨, 田飛, 等. 聚乳酸/聚酮共混織物的開發(fā)與設(shè)計[J]. 合成纖維, 2024, 53(2): 37-39.
LIAO S H, ZHANG X Y, TIAN F, et al. Development and design of polylactic acid/polyketone blend fabric[J]. Synthetic Fiber in China, 2024, 53(2): 37-39.
[22]聶梓言. 純氧化鋁長絲機(jī)織物開發(fā)工藝研究[D]. 上海: 東華大學(xué), 2024.
NIE Z Y. Research on the Development Process of Pure Alumina Filament Woven Fabric[D]. Shanghai: Donghua University, 2024.
[23]史洋濤. 基于石墨烯纖維交織物的性能研究與產(chǎn)品開發(fā)[D]. 杭州: 浙江理工大學(xué), 2023.
SHI Y T. Performance Research and Product Development of Graphene Fiber Interleaved Materials[D]. Hangzhou: Zhejiang Sci-Tech University, 2023.
[24]LIU X, LI Z. Influence mechanism of running sportswear fatigue based on BP neural network[J]. EURASIP Journal on Advances in Signal Processing, 2021(1): 1-15.
[25]王彥鑫, 李海巖, 崔世海, 等. 基于BP神經(jīng)網(wǎng)絡(luò)的兒童乘員頭部損傷預(yù)測模型及評估參數(shù)研究[J]. 汽車工程, 2024, 46(2): 329-336.
WANG Y X, LI H Y, CUI S H, et al. Research on prediction model and evaluation parameters of child occupant head injury based on BP neural network[J]. Automotive Engineering, 2024, 46(2): 329-336.
[26]史墨可, 路妍, 頡潭成, 等. 基于GA-BP神經(jīng)網(wǎng)絡(luò)的寬帶激光熔覆裂紋缺陷預(yù)測[J/OL]. 熱加工工藝, 2024: 10.14158/j.cnki.1001-3814.20223162.
SHI M K, LU Y, XIE T C, et al. Prediction of crack defects in broadband laser cladding based on GA-BP neural network[J/OL]. Hot Working Technology, 2024: 10.14158/j.cnki.1001-3814.20223162.
[27]呂哲, 賀利樂, 林育陽, 等. 基于鋁合金粉末成形本構(gòu)模型的BP神經(jīng)網(wǎng)絡(luò)參數(shù)預(yù)測[J]. 熱加工工藝, 2022, 51(4): 46-50.
L Z, HE L L, LIN Y Y, et al. BP neural network parameter prediction based on constitutive model of aluminum alloy powder forming[J]. Hot Working Technology, 2022, 51(4): 46-50.
[28]程曉娟, 韓慶蘭, 全春光. 基于灰關(guān)聯(lián)分析的GA-BP神經(jīng)網(wǎng)絡(luò)在汽車油耗估算中的應(yīng)用[J]. 數(shù)學(xué)的實踐與認(rèn)識, 2016, 46(8): 43-51.
CHENG X J, HAN Q L, QUAN C G. Vehicle fuel consumption forecast models of optimized BP neural network based on grey relation and genetic algorithm[J]. Mathematics in Practice and Theory, 2016, 46(8): 43-51.
[29]PAMUJI S. 服務(wù)質(zhì)量管理與收入和生活質(zhì)量之間的相關(guān)性: 以印度尼西亞語言村為例[D]. 武漢: 華中師范大學(xué), 2020.
PAMUJI S. Correlation Between Service Quality Management with Income and Quality of Life: The Case of Kampung Bahasa, Indonesia[D]. Wuhan: Central China Normal University, 2020.
[30]LU X, WANG Y, LI J, et al. Three-dimensional coordinate measurement algorithm by optimizing BP neural network based on GA[J]. Engineering Computations, 2019, 36(6): 2066-2083.
Prediction of breathability performance of household apparel based on GRA-GA-BP neural network
ZHANG Chi, WANG Xiangrong
WANG Binxia1a,b,c, WANG Chunhong1a,b,c, CHEN Yasong1d, ZHOU Jinxiang2, YIN Lanjun2, YANG Daopeng3
(1a.School of Textile Science and Engineering; 1b.Tianjin and Education Ministry Key Laboratory of Advanced Textile Composite Materials;1c.Key Laboratory of Hollow Fiber Membrane Materials and Membrane Processes; 1d.School of Mathematical Sciences,Tiangong University, Tianjin 300387, China; 2.Shenzhen Purcotton Co., Ltd., Shenzhen 518109, China;3.Shaoxing Zhongfanglian Inspection Technology Service Co., Ltd., Shaoxing 312000, China)
Abstract:
With the improvement of people’s living standards, people have higher requirements for the comfort of household apparel. Breathability is one of the key factors affecting the comfort of household apparel and is the most concerned by household apparel consumers. At present, research on the comfort of household apparel is still in a blank period both domestically and internationally. There is a lack of research on the breathability of various household apparel fabrics with different fabric compositions and textures, and there is relatively little research on predicting the comfort of household apparel. Based on this, this article selects 58 common household apparel fabrics with different fabric compositions and textures on the market, and constructs a genetic algorithm improved BP neural network model to predict the breathability performance of household apparel.
Firstly, to study the relationship between various influencing factors and air permeability of household apparel fabrics, the grey relational analysis (GRA) method was used to analyze the degree of influence of each influencing factor on the air permeability of household apparel fabrics. The factors with higher correlation were selected as input parameters for tc6713016caef53ec049a39674ec76f56e8164257f87ef9580de0076513520a67he model in this study, namely density, yarn diameter, thickness, and weight. Secondly, due to the shortcomings of BP neural network, such as proneness to local minima, slow learning rate, and long training time, this study used genetic algorithm (GA) to optimize the structural parameters of BP neural network, and constructed a genetic algorithm optimized BP (GRA-GA-BP) neural network prediction model based on grey correlation analysis. Genetic algorithm can optimize the structural parameters of the model, find the best parameter combination, and solve complex and high-dimensional problems, without being affected by local optimal solutions. 58 household apparel fabrics with different fabric compositions and textures were selected, of which 42 were model training samples and 16 were test samples to validate the established model. The parameters of each factor, including fabric density, yarn diameter, thickness, weight, and air permeability, were tested as input parameters for the GRA-GA-BP neural network.
The results show that the measured and predicted values of air permeability had a small error, with a relative error of between 0.80% and 28.53%, and an average relative error of 8.39%; a comparison chart between the measured and predicted values of air permeability was drawn, and it was found that the two curves are basically consistent, indicating high prediction accuracy of the model. Finally, OriginPro software was used to analyze the correlation between the measured and predicted values of air permeability, and the goodness of fit R2 was 0.976, very close to 1, indicating that the model’s prediction effect is good. The prediction model has a small prediction error, good prediction effect, high prediction accuracy, good fitting effect between the measured and predicted values of air permeability, and a strong correlation between the measured and predicted values.
This article enriches the research on predicting the comfort of household apparel. The model can accurately predict the breathability of household apparel fabrics to a certain extent, saving manpower and costs required for experiments. It has important reference significance for household apparel designers to design based on household apparel comfort performance. At the same time, it provides a reference route for predicting the comfort of household apparel. Researchers can start from the perspective of household apparel comfort, combine subjective and objective experiments, and construct corresponding household apparel comfort evaluation and prediction models.
Key words:
fabric; household apparel; grey correlation analysis; improved BP neural network; breathability prediction