论文部分内容阅读
探讨了径向基核函数(RBF)支持向量机(RBF-SVR)建立定量分析模型时主要参数的优选方法及其在近红外光谱分析畜禽粪便堆肥产品含水率、挥发性固体含量和碳氮比中的应用,并与偏最小二乘回归法所建近红外定量分析模型的预测能力做了比较。供试样品为我国22省市不同种类的120个畜禽粪便堆肥产品样品,利用傅里叶变换型光谱仪获取样品在4000~10000 cm-1内的光谱数据信息。研究发现,逐步寻优循环优选支持向量机建模参数方法具有较好的可行性,其所建近红外定量分析模型均优于基于偏最小二乘法所建模型,所建立的含水率和挥发性固体近红外模型验证决定系数(r2)均大于0.90,相对分析误差(RPD)均大于4.0,具有实际应用价值;所建碳氮比近红外模型验证决定系数(r2)为0.85,RPD值大于2.5,也可用于定量分析,但精度有待于进一步提高。
This paper discusses the optimal method for establishing the main parameters of RBF-SVR (Quantitative Analysis Model) and its application in near-infrared spectroscopy to analyze the moisture content, volatile solid content and carbon nitrogen Compared with the application, and compared with partial least-squares regression model built near infrared quantitative analysis of the predictive power of comparison. The samples for testing were 120 samples of livestock manure composting from different provinces in 22 provinces and cities in China. The spectral data of samples in 4000 ~ 10000 cm-1 were obtained by Fourier transform spectrometer. It is found that the method of stepwise optimization of optimal SVM modeling parameters is more feasible and the model of near-infrared quantitative analysis is better than the model based on partial least square method. The established moisture content and volatility The determination coefficients (r2) of solid near-infrared model were all greater than 0.90 and the relative analytical errors (RPD) were both greater than 4.0, which had practical value. The established determination coefficient (r2) of the model was 0.85 and the RPD value was greater than 2.5 , Can also be used for quantitative analysis, but the accuracy needs to be further improved.