思维进化蝙蝠算法及其在混合气体红外光谱特征选择中的应用

Mind evolutionary bat algorithm and its application to feature selection of mixed gases infrared spectrum

为了解决混合气体多组分间特征吸收峰相互重叠引起的特征选择困难问题,提出了新型红外光谱特征选择方法,并对该方法的性能进行了分析与评价。首先,充分结合思维进化计算的并行机制、异化操作与蝙蝠算法的局部搜索能力,设计了思维进化蝙蝠算法。接着,通过实验采集两个混合气体数据库,利用思维进化蝙蝠算法对其目标组分的特征峰进行筛选。然后,从算法的收敛速度和筛选出的特征峰两个方面,将思维进化蝙蝠算法与基本蝙蝠算法、遗传算法、粒子群优化算法及并行萤火虫群优化算法等进行比较。最后,讨论了思维进化蝙蝠算法与无信息变量消除法相结合对结果的影响。实验结果表明：CO的特征峰范围包括2 090-2 110 cm^-1和2 115-2 125 cm^-1,共包含32个波长点;N2O的特征峰范围为2 225-2 250 cm^-1,共包含26个波长点。利用筛选出的特征波长点建立的浓度反演模型,测试集均方根误差为0.155,决定系数可达0.908。实验结果表明：思维进化蝙蝠算法收敛速度快、全局搜索能力强,适用于存在重叠特征峰的混合气体的特征选择,对应的浓度反演模型的泛化性能也有显著提升。

Due to the fact that the characteristic peaks of multi-component of mixed gases have overlapping problem, it was hard to implement feature selection for each target gas. To solve this problem, a novel feature selection method was introduced. First, by making full use of the parallel mechanism, dissimilation operator of mind evolutionary computation and local search ability of bat algorithm, the mind evolutionary bat algorithm was designed. Two different mixed gases databases were collected to validate the performance of proposed method. Then, from the aspects of convergence speed and characteristic peaks, the comparison with basic bat algorithm, genetic algorithm, particle swarm optimization and parallel glowworm swarm optimization algorithm was investigated. Finally, the influence of combination with uninformative variable elimination method was discussed. Experimental results show that the characteristic peaks of carbon monoxide include 2 090-2 110 cm-1and 2 115-2 125 cm-1, which total have 32 wavelength points while the characteristic peaks of nitrogen oxide were in range from 2 225 to 2 250 cm-1, which total have 26 wavelength points. Considering the concentration retrieve model established with the selected characteristic peaks, the root mean squared error of prediction set was 0.155, and the determined coefficient can reach as high as0.908. Experimental results show that the proposed method has the advantage of rapid convergence speed and well global search ability, which was adaptable to do the feature selection for those mixed gases with overlapping problem.