iPLS结合BP-ANN的喷气燃料初馏点近红外分析模型研究

Study on Near Infrared Analysis Model of Initial Distillation Point of Jet Fuel Based on iPLS and BP-ANN

针对喷气燃料初馏点性质测试的近红外光谱干扰多、信息提取困难,定量分析模型预测准确度较低的问题,在1100~2500 nm的长波近红外光谱范围内采用间隔偏最小二乘法对建模区间进行优选,通过误差反向传播神经网络(BP-ANN)算法取代偏最小二乘法建立定量分析模型。结果表明:在区间1650~1705 nm建立初馏点的BP-ANN定量分析模型并预测验证集样本初馏点,预测决定系数(R2)为0.9361,验证标准偏差(SEV)为1.0726℃,预测准确度均优于全谱区及iPLS方法选择区间建立的PLS模型。

Aiming at solving the problems of more interference of near infrared spectroscopy(NIRS)on the properties of jet fuel’s initial distillation point,difficulty in extracting information and low prediction accuracy of quantitative analysis model,the interval partial least squares(iPLS)method was used to optimize the modeling interval in the range of 1100~2500 nm,and the partial least squares(PLS)method was replaced by the back propagation-artificial neural network(BP-ANN)algorithm to establish the quantitative analysis model.The BP-ANN quantitative analysis model of the initial distillation point was established in 1650~1705 nm,with the prediction correlation coefficient(R2v)of 0.9361 and the standard deviation of validation(SEV)of 1.0726℃.The prediction accuracy was superior to that of PLS model based on the full spectrum region and the interval selected by the iPLS method.