GA-BPSO混合优化中红外光谱特征波段筛选的润滑油添加剂种类识别技术

针对各种设备润滑油中微量多品种添加剂种类识别问题，提出二进制粒子群算法结合遗传算法(GA-BPSO)混合优化中红外光谱特征波段筛选方法. 首先建立K近邻算法(KNN)和随机森林算法(RF)的润滑油添加剂种类识别基础分类模型；然后通过GA-BPSO混合优化算法在光谱全波段范围内筛选特征波段区域，消除干扰及无效信息，压缩庞大光谱数据集，降低搜索空间维度；再以模型识别准确率作为评价标准，用优选出的特征波段在基础分类模型上构建高性能增强分类模型. 选取硫化异丁烯(T321)、烷基二苯胺(T534)和硫化磷酸胺盐(T307) 三种润滑油添加剂作为测试对象，以不同配比混合在基础油中，采集配制样品的中红外光谱数据，并划分为训练集与测试集，分别导入基础分类模型与增强分类模型进行训练及测试. 结果显示，GA-BPSO优化筛选特征波段，使KNN的有效波段长度削减至原来的16.4%，识别准确率从70%提高到89.58%；RF的有效波段长度削减至原来的15.8%，识别准确率从85%提升至97.5%. 对比研究发现，GA-BPSO混合特征波段优选方法明显优于GA和BPSO单独使用时的筛选结果，在极大地减轻运行负担的同时，有效提高了模型多种类同步识别的准确率和稳定性.

Aiming at the problem of identifying trace multi-type additive species in various equipment lubricants, a hybrid optimization method combining binary particle swarm optimization and genetic algorithm (GA-BPSO) for mid-infrared spectral feature band screening is proposed. First, basic classification models for lubricant additive species identification, namely K-nearest neighbor (KNN) and random forest (RF), are established. Subsequently, the GA-BPSO hybrid optimization algorithm is employed to screen feature band regions across the full spectral range, eliminating interference and invalid information, compressing the large-scale spectral dataset, and reducing the dimensionality of the search space. Furthermore, taking the model recognition accuracy as the evaluation criterion, a high-performance enhanced classification model is constructed on the basis of the basic classification models using the optimized feature bands. Three lubricant additives, namely isobutylene sulfide (T321), alkyldiphenylamine (T534), and sulfurized amine phosphate salts (T307), were selected as test subjects, mixed with base oil at various ratios, and their mid-infrared spectral data of the prepared samples were collected. The dataset was then split into training and test sets, which were respectively fed into the basic classification models and enhanced classification models for training and evaluation. The results demonstrate that the GA-BPSO optimized feature band screening reduces the effective band length of KNN to 16.4% of the original, with the recognition accuracy increased from 70% to 89.58%; for RF, the effective band length is reduced to 15.8% of the original, and the recognition accuracy is improved from 85% to 97.5%. Comparative studies reveal that the GA-BPSO hybrid feature band optimization method significantly outperforms the screening results obtained when using GA or BPSO alone, effectively enhancing the accuracy and stability of the model's simultaneous identification of multiple lubricant additive types while greatly reducing the computational burden.

搜集汇总

背景与挑战

背景概述

该数据集聚焦于润滑油添加剂种类识别，通过GA-BPSO混合优化算法筛选中红外光谱特征波段，以提升分类模型性能。具体包括硫化异丁烯、烷基二苯胺和硫化磷酸胺盐三种添加剂的实验数据，结果显示该方法能大幅压缩光谱数据维度，同时提高KNN和RF模型的识别准确率，从70%提升至89.58%和85%提升至97.5%，有效增强了模型的稳定性和效率。