摘要
通过对沈海热电厂、阜新、抚顺等数处电厂锅炉用煤的近红外光谱建模实验,发现电厂入炉煤粉中的水分、灰分、挥发分和发热量等工业指标预测所采用的模型及其预处理方法直接关系到建模效果的优劣。预处理方法包括平滑处理、数据标准化、微分处理、信号校正等。建模过程中使用移动窗口PLSR、误差反向传播人工神经网络、径向基人工神经网络,发现除水分模型外在全光谱下采用FFT系数作为输入变量的径向基网络效果为最优。根据所得的预处理方法对部分煤粉工业分析进行预测,证实基于偏最小二乘回归分析煤粉工业分析的近红外光谱建模的重复性较好,其模型具有较高的应用价值。
Through the analysis on the near infrared spectroscopy （NIRS） modeling tests of some steam coals from Shenhai, Fuxin, Fushun power stations, it clearly described a direct relationship between the modeling effect and the pre-treatment method when calculating the coal quality indexes, such as moisture content, ash yield, volatile matter and calorific value. And the pre-treatment concerned included smooth treatment, data standardization, differential coefficient process and signal correction. Moreover, while the moving window PLSR, the error back propagation neural network and the radial basis function （RBF） network were applied in the modeling, there was the best modeling effect of RBF network using FFT coefficient as input variable under full-color spectrum except the moisture model. Finally, according to the prediction of the quality indexes of several coals based on the selected pre- treatment method, it indicated that there was a good repeatability of coal quality analysis by NIRS modeling method based on partial least-squares regression （PLSR）, which meant the NIRS model would be applied in a very wide range.
出处
《煤质技术》
2012年第5期1-4,共4页
Coal Quality Technology
关键词
煤质
近红外光谱
预处理方法
偏最小二乘回归分析
径向基网络
coal quality
near infrared spectroscopy （NIRS）
partial least-squares regression （PLSR）
pre-treatment
radial basis function （RBF） network
