| As one of the three major quality indicators of asphalt,the asphalt penetration index reflects the viscosity and shear resistance of asphalt,which is of great significance for evaluating asphalt quality.Currently,the penetration index is generally obtained by measuring with a penetrometer,a method that requires tedious sample preparation and consumes a significant amount of time for sample testing,with measurement accuracy being difficult to guarantee.Infrared spectroscopy(IR)analysis technology,as a highly efficient and low-cost online detection technology,can achieve rapid quantitative analysis of asphalt penetration index.However,due to the presence of nonlinearity,"dimensionality disaster," and noise interference in asphalt infrared spectral data,it is impossible to directly obtain quantitative information of substances from spectral signals.Nonlinear calibration is needed to correlate spectral feature information with relevant properties of substances,so the key to infrared spectroscopy quantitative analysis lies in establishing an efficient nonlinear calibration model.This paper focuses on the problems existing in infrared spectroscopy quantitative analysis,deeply analyzes the advantages and disadvantages of Kernel Partial Least Squares(KPLS)and the global optimization effect of Particle Swarm Optimization(PSO),and conducts research on the infrared spectroscopy nonlinear calibration method based on PSO-KPLS.The main research contents of the paper are as follows:(1)The characteristics curve of kernel functions in the "kernel method" is analyzed.Combining the local characteristics of Gaussian kernel functions and the global characteristics of polynomial kernel functions,a mixed kernel function method suitable for infrared spectroscopy nonlinear calibration is proposed.A diesel density nonlinear calibration model based on Kernel Partial Least Squares is constructed,proving that this method can improve the data partitioning ability of kernel functions.(2)In response to the problem that the optimal kernel function parameters cannot be selected in the modeling and analysis process of Kernel Partial Least Squares,Particle Swarm Optimization is used to globally optimize the kernel function parameters in Kernel Partial Least Squares.A nonlinear calibration model for asphalt penetration index based on PSO-KPLS is constructed,achieving rapid detection of asphalt penetration index.This method can effectively improve the prediction accuracy of the model. |
