为了快速、准确检测出煤炭发热量指标,研究采集了150个烟煤样品的近红外光谱,并分别对800~1nm、1400~1nm、2000~2nm及2200~2nm下的光谱进行分析,采用学生式残差方法剔除了异常样品,建立了偏最小二乘回归(PLSR)定量检测模型。通过采用一阶微分、二阶微分、5点平滑、9点平滑、15点平滑、多元散射校正(MSC)和衰减全反射校正(ATR)进行光谱预处理,并结合不同建模波段进行建模结果分析。研究结果表明:采用5点平滑预处理结合1400~1nm建模波段的模型预测能力最佳,校正集和交叉验证集相关系数分别达到0.952和0.946,均方根误差分别为0.029和0.037,二者相差仅0.008,模型更具有一定代表性,模型预测精度较强。

In order to quickly and accurately detect the content of coal calorific value indicators, collected near-infrared spectra of bituminous coal samples, and analyze the spectra at 800~ 1 nm、1 400~ 1 nm、2 000~ 2 nm, and 200~ nm, respectively, a student style residual method was used to remove abnormal samples and a partial least squares regression (PLSR) quantitative detection model was established. By using first-order differentiation, second-order differentiation, 5-point smoothing, 9- point smoothing, 15 - point smoothing, multivariate scattering correction ( MSC), and attenuated total reflection correction(ATR) for spectral preprocessing, and combining different modeling bands for modeling result analysis. The research results show that the model with 5-point smoothing preprocessing combined with 400 ~ 1 nm modeling band has the best predictive ability. The correlation coefficients of the correction set and cross validation set reach 0. 952 and 0. 946, respectively, with root mean square errors of 0. 029 and 0. 037, and the difference between the two is only 0. 008. The model is more representative, with strong prediction accuracy.

               
                    1 试验部分
1.1 试验煤样
1.2 试验仪器
1.3 试验方案
1.4 评价指标
2 结果与分析
2.1 光谱分析
2.2 异常样品检测
2.3 建模结果分析
3 结 论