Application Of GA-PSO-BP Neural Network In Edxrf Content Prediction Of Copper And Zinc

As a fast,non-destructive,safe and reliable detection method,energy dispersive X-ray fluorescence analysis(EDXRF)has been widely used in mineral exploration,environmental protection,archaeology,alloy analysis and other fields.Among them,the objective existence of matrix effect has always been the main factor affecting the accurate quantitative analysis of EDXRF.Traditional matrix correction methods such as experimental correction cannot avoid the problem of preparing a large number of standard samples,while the mathematical correction method is complex in calculation and depends on the similarity between the sample to be tested and the standard sample.In order to overcome the shortcomings of the traditional matrix correction method,the quantitative analysis error is too large due to the influence of the matrix effect when using the EDXRF technology to determine the element content of the sample.This paper focuses on the BP neural network prediction model optimized by the GA-PSO algorithm,and applies the established GA-PSO-BP neural network prediction model to the prediction of copper and zinc elements in EDXRF.The effectiveness of the method is verified,and a human-computer interaction system for energy dispersive X-fluorescence data processing based on MATLAB GUI platform is designed.The main work of this paper is as follows:(1)Aiming at the problem that the local minimization and slow convergence speed of the traditional BP neural network model have a great influence on the accurate quantitative analysis of the final sample.A hybrid algorithm with Genetic Algorithm(GA)as the main particle swarm algorithm(PSO)embedded as the genetic algorithm mutation operator is proposed.Combining GA-PSO algorithm with BP neural network model,to obtain the optimal weights and thresholds for subsequent training.(2)Aiming at the mathematical formulas and tools used in the de-spectrification process,a set of human-computer interaction interface for energy dispersive X-fluorescence data analysis is designed by using the GUI platform that comes with MATLAB,which consists of a data processing window and a result display window.The functions of the data processing window include data loading,spectral smoothing,background subtraction,peak position identification,overlapping peak decomposition and peak area calculation.The information displayed in the result display window is the peak position,left and right boundaries and net peak area.The obtained net peak area is used as the input variable of the subsequent neural network prediction model,and finally the operation of the entire interface is tested by loading the measured data.(3)The BP neural network model optimized by GA-PSO algorithm was applied to the prediction of copper and zinc content in EDXRF.Based on the four performance evaluation functions of mean absolute error(MAE),mean absolute percentage error(MAPE),mean square error(MSE)and root mean square error(RMSE),the BP neural network prediction model optimized by GA-PSO algorithm is compared with BP neural network prediction model,GA-BP and PSO-BP neural network prediction models were simulated and compared,and the results showed that the BP neural network model optimized by GA-PSO algorithm had the smallest prediction deviation for the content of copper and zinc.By comparing the predicted results with the instrumental spectral analysis values,the maximum relative error between the predicted results of Cu element content and the corresponding instrumental spectral analysis values is 0.6753%,the minimum relative error is 0.0091%,and the predicted results of Zn element content and the corresponding instrumental spectral analysis values.The maximum relative error is2.3964%,and the minimum relative error is 0.1131%.The GA-PSO-BP neural network prediction model can effectively improve the accuracy of quantitative analysis,and provides an effective method for quantitative analysis and prediction of copper alloying elements.