The Increment Partial Least Squares Algorithm Based On Gradually Tuning The Error Of Regression Coefficient

Partial Least Squares algorithm(Partial Least Squares,PLS)as a multivariate calibration technology in infrared spectrum quantitative analysis,is widely applied to the pharmaceutical,food,petrochemical,and many other fields.However,when modeling the incrementally gained samples,all data must be ready,when adding new datas,the previous study results need to be abandoned,training and learning the original data and new data again,which can cause a lot of time and space resources consumption.According to the problem,this paper presents an increment partial least squares regression based on gradually tuning regression coefficient error,when detected new data,do not abandon the existing model any more,instead of updating the model with the new sample point on the basis of the existing model.In this paper,the main work and contributions are as follows:(1)It puts forward a sample composition method based on the increment partial least squares method.After analyzing the deficiencies of the traditional algorithm when measuring incremental samples,this algorithm uses approximate gradient descent method to update the IPLSR regression coefficient,selecting a threshold as coefficient updating standard for the incremental NIR data modeling process.By calculating the time complexity of PLSR and IPLSR,that time complexity of IPLSR is a level less than that of PSLR,so it can be concluded in theory that the incremental data processing time of IPLSR will be less than the processing time of PLSR.(2)It gives the method separately of the determination of the threshold in the increment partial least squares algorithm,the method adopts the K-fold cross validation method to select the best threshold parameter.(3)We do simulation experiment with IPLSR algorithm for grain,grass and soil data sets,and paeoniflorin concentration forecast instance based on the algorithm is also given.The experimental results show that although the regression coefficient of the two methods have different updateing processes,however,the regression coefficients obtained with incremental algorithm has the same direction with the regression coefficient of traditional algorithm,shows that the models are similar;Prediction accuracy of the incremental algorithm compared with the traditional algorithm is improved,and shows that the incremental algorithm has higher forecasting performance.The running time of the incremental algorithm is less than the traditional algorithm,shows that the algorithm has higher efficiency.Visible,the incremental algorithm in the analysis of simulation results and the actual process can effectively deal with incremental data,and the prediction precision,competence and efficiency compared with the traditional algorithm are improved.