针对润滑脂分类, 提出了基于布谷鸟搜索的红外光谱波段筛选方法, 有效剔除了易受噪声等环境影响的红外光谱区域、 实现了对庞大光谱数据进行特征选择和降维处理、 通过筛选光谱最优波段建立了更加准确高效的润滑脂分类模型。 以三类不同稠化剂润滑脂的红外光谱数据为研究对象, 采用主成分分析法(PCA), 对不同波段的红外光谱数据进行压缩, 以提取的红外光谱主要成分作为输入, 润滑脂稠化剂类别作为输出, 通过布谷鸟搜索法(CS), 对主要成分权重和分类核参数进行准确度寻优训练, 建立分类识别预测模型。 对所建立的模型再进行分类准确性测试, 得到模型测试结果准确度, 建立红外光谱波段和测试准确度之间的联系, 得到润滑脂最优类别识别模型和最优分类波段。 对所建立的模型再进行分类准确性测试, 结果显示: 经过布谷鸟搜索法训练加权后的主要特征呈现明显聚类现象, 可以得到分类核, 实现对润滑脂种类的准确识别; 在搜索过程中提供了区分不同润滑脂的推荐波段和特征峰, 使对润滑脂的正确鉴别概率由全波段建立分类模型的94.44%提高到筛选后特征波段建立分类模型的100%, 并减少了运算时间、 提高了搜索运行效率。

A selection method of infrared spectral based on cuckoo search was proposed to meet the classification of grease. The method would help to remove the infrared spectral region affected by noise and environment effectively, and realized the feature selection and dimension reduction processing of large spectral data. A more accurate and efficient classification model of grease was established by selecting the optimal spectral bands. Regarding the infrared spectrum data of three different types of greases as research targets in this paper, the Principal Component Analysis (PCA) was applied to compress the Infrared spectrum data of different bands and extracted the main components. Using the extracted main components of IR spectra and the grease thickener category as input and output respectively, an accuracy optimization training for the weight of principal component and parameter of classification kernel was conducted by Cuckoo Search (CS) to establish the classification prediction model. The classification accuracy of the model was tested and obtained the accuracy of the test results of model. In addition, it established the link between the infrared spectral band and the accuracy to get the optimal class identification model and optimal classification bands. The classification accuracy of the model was tested, and the result showed that the main feature trained and weighted by Cuckoo Search presents obvious clustering phenomenon. The classification kernel could be found and the type of grease could be classified accurately. Furthermore, it provided recommended bands and characteristic peaks for distinguishing different greases in the process of searching. The correct identification probability of the grease was improved from 94.44% for the classification model by whole band to 100% for filtered feature band, reducing the operation time and improving the search efficiency.