Research And Application Of Synergy Interval Partial Least Squares Based On Monte Carlo Method

The partial least squares method is a commonly used algorithm in regression analysis,and has important applications in the field of spectral analysis.Since the synergy interval partial least squares method uses equal intervals to divide intervals,it is inevitable to discard some intervals containing characteristic variables when facing data with uneven distribution of characteristic variables,resulting in poor model results.Therefore,it is particularly important to choose an appropriate division method.The main work content of this paper is as follows:(1)This paper introduces the Monte Carlo method(MC)and used this method to replace the method of dividing intervals at equal intervals,and proposed a synergy interval partial least squares method based on the Monte Carlo method(MC-siPLS).Compared with other 5 similar wavelength interval selection algorithms on tomato,corn and beer data sets,the root mean square error and correlation coefficient of MCsiPLS on the three data sets were better than other algorithms.This shows that the performance of the siPLS algorithm has been significantly improved after introducing the MC method.(2)Since the final number of variables obtained by the MC-siPLS algorithm is still large.Therefore,based on the previous MC-siPLS algorithm,a competitive adaptive reweighted sampling algorithm based on anchor points(A-CARS)was proposed to reduce the number of variables.This method merges the intervals selected by each iteration by running MC-siPLS multiple times,and finally determines the interval boundaries,and then uses the CARS algorithm for variable reduction.The A-CARS algorithm was compared with 6 different algorithms on the near-infrared data set of corn protein content.The final results show that the A-CARS algorithm is far better than other algorithms,where the root mean square error and correlation coefficient on the test set are 0.9820 and 0.0688,respectively.In addition,A-CARS reduced the original 700 variables to 23,and gave a linear regression equation for corn protein content based on 23 characteristic variables.(3)For the lack of computing power of traditional embedded devices or portable devices,while expecting to obtain high-precision models.This paper proposes a fully automatic online model system,which transplants the algorithm studied in this paper to the server for remote computing to meet the needs of embedded devices.