【摘要】：針對(duì)潤(rùn)滑脂分類(lèi),提出了基于布谷鳥(niǎo)搜索的紅外光譜波段篩選方法,有效剔除了易受噪聲等環(huán)境影響的紅外光譜區(qū)域、實(shí)現(xiàn)了對(duì)龐大光譜數(shù)據(jù)進(jìn)行特征選擇和降維處理、通過(guò)篩選光譜最優(yōu)波段建立了更加準(zhǔn)確高效的潤(rùn)滑脂分類(lèi)模型。以三類(lèi)不同稠化劑潤(rùn)滑脂的紅外光譜數(shù)據(jù)為研究對(duì)象,采用主成分分析法(PCA),對(duì)不同波段的紅外光譜數(shù)據(jù)進(jìn)行壓縮,以提取的紅外光譜主要成分作為輸入,潤(rùn)滑脂稠化劑類(lèi)別作為輸出,通過(guò)布谷鳥(niǎo)搜索法(CS),對(duì)主要成分權(quán)重和分類(lèi)核參數(shù)進(jìn)行準(zhǔn)確度尋優(yōu)訓(xùn)練,建立分類(lèi)識(shí)別預(yù)測(cè)模型。對(duì)所建立的模型再進(jìn)行分類(lèi)準(zhǔn)確性測(cè)試,得到模型測(cè)試結(jié)果準(zhǔn)確度,建立紅外光譜波段和測(cè)試準(zhǔn)確度之間的聯(lián)系,得到潤(rùn)滑脂最優(yōu)類(lèi)別識(shí)別模型和最優(yōu)分類(lèi)波段。對(duì)所建立的模型再進(jìn)行分類(lèi)準(zhǔn)確性測(cè)試,結(jié)果顯示:經(jīng)過(guò)布谷鳥(niǎo)搜索法訓(xùn)練加權(quán)后的主要特征呈現(xiàn)明顯聚類(lèi)現(xiàn)象,可以得到分類(lèi)核,實(shí)現(xiàn)對(duì)潤(rùn)滑脂種類(lèi)的準(zhǔn)確識(shí)別;在搜索過(guò)程中提供了區(qū)分不同潤(rùn)滑脂的推薦波段和特征峰,使對(duì)潤(rùn)滑脂的正確鑒別概率由全波段建立分類(lèi)模型的94.44%提高到篩選后特征波段建立分類(lèi)模型的100%,并減少了運(yùn)算時(shí)間、提高了搜索運(yùn)行效率。
[Abstract]:Aiming at grease classification, an infrared spectral band screening method based on cuckoo search was proposed, which effectively eliminated the infrared spectral region which was susceptible to noise and other environmental effects, and realized the feature selection and dimensionality reduction of huge spectral data. A more accurate and efficient grease classification model was established by screening the spectral optimum band. Taking infrared spectrum data of three kinds of thickener greases as the research object, the infrared spectrum data of different bands were compressed by principal component analysis (PCA), and the main components of the extracted infrared spectra were used as input. The classification of grease thickener is used as output, and the weight of main components and the kernel parameters of classification are trained by cuckoo search method (CS), and the model of classification identification and prediction is established. The accuracy of the model is tested and the relationship between the infrared spectrum band and the test accuracy is established. The optimal classification model and the optimal classification band of lubricating grease are obtained. The classification accuracy of the model is tested. The results show that the main features after training and weighting by cuckoo search show obvious clustering phenomenon, and the classification kernel can be obtained to realize the accurate identification of grease types. In the process of searching, the recommended bands and characteristic peaks are provided to distinguish different greases, so that the correct identification probability of grease is increased from 94.44% of full-band classification model to 100% of the selected characteristic band classification model. And reduce the operation time, improve the efficiency of search.
【作者單位】： 華北電力大學(xué)能源動(dòng)力與機(jī)械工程學(xué)院;
【基金】：國(guó)家自然科學(xué)基金項(xiàng)目(5157181) 北京市自然科學(xué)基金項(xiàng)目(2172053)資助
【分類(lèi)號(hào)】：O657.33;TE626.4
，

本文編號(hào)：2470513