论文部分内容阅读
研究了基于遗传算法(GA)的波长选择方法结合反向传播神经网络(BP-ANN)建模用于在用航空润滑油-40℃运动粘度的近红外光谱分析。采集样品光谱经均值中心化和SavitzkyGolay平滑求导法预处理后,通过分段组合建模初选最优波段,利用遗传算法进一步筛选了对粘度预测敏感的波长点建模。该波长选择方法与相关系数法相比,所建模型预测准确度高。在建模采用的非线性BP-ANN法中,先通过主成分分析(PCA)分解光谱数据,将得分矩阵输入3层神经网络训练,通过参数优化建立最优模型。所建模型对8个在用油进行分析,各预测值与标准值的相对误差均低于2%,并且经t检验不存在显著性差异,模型预测能力较强,应用于在用润滑油质量的快速分析效果好,为油品在线监控提供了参考。
The method of wavelength selection based on genetic algorithm (GA) combined with backpropagation neural network (BP-ANN) modeling was applied to near-infrared spectroscopy analysis of kinematic viscosity at -40 ℃ using aviation lubricants. The spectra of the collected samples were averaged by the mean and SavitzkyGolay smoothing derivation method was preprocessed. The optimal bands of the primaries were modeled by piecewise combinations. Genetic algorithm was used to further screen the wavelength points which are sensitive to the viscosity prediction. Compared with the correlation coefficient method, the wavelength selection method has high predictive accuracy. In the nonlinear BP-ANN method used in modeling, the spectral data is first decomposed by principal component analysis (PCA), the score matrix is input into the 3-layer neural network training, and the optimal model is established through parameter optimization. The proposed model was used to analyze 8 oil samples. The relative errors of each predicted value and standard value were less than 2%. There was no significant difference by t-test. The rapid analysis of good results for oil online monitoring provides a reference.