Research And Application Of Slurry Element Grade Prediction Based On Improved Partial Least Squares Regressio

With the development of big data mining and industrial data modeling technology,the traditional method of predicting the grade of mineral pulp elements based on experience can no longer meet the needs of today’s factories in terms of product quality,pharmaceutical savings,and production efficiency.Therefore,pulp element grade prediction based on data modeling has become a popular research direction in the field of mineral processing.Grade value is an important factor that affects the results of mineral flotation,and its accurate and rapid prediction is the prerequisite for improving the quality of mineral products.The traditional method of obtaining grade value is to bring a small part of the slurry flow sample back to the laboratory and use relevant equipment for chemical analysis.Although a higher accuracy grade value can be obtained,this process consumes a lot of time and The labor cost cannot adapt to the complexity and variability of working conditions.In recent years,the rapid development of X-ray fluorescence spectroscopy analysis technology,pulp X-ray fluorescence measurement data can be easily obtained,making it possible to achieve grade prediction through data modeling methods.This subject takes the time series of the X-ray fluorescence intensity of each metal element in the pulp measured in the actual production process of a mining factory as the research object,uses partial least squares regression as the core algorithm for modeling,and completes the modeling of the pulp fluorescence measurement sequence.Development of analysis system.The main content and innovative work of this thesis are as follows:(1)Aiming at the problems of high dimensionality and serious multi-correlation between variables in the time series of pulp fluorescence intensity,a correlation variable selection partial least squares regression(CVS-PLSR)modeling method is proposed.This method combines the advantages of the partial least squares regression modeling method in dealing with multiphase inertia,and at the same time uses the feature selection algorithm based on correlation to screen out the optimal feature subset,and achieves the dimensionality reduction of the data,and builds on this basis The prediction model of pulp element grade is simplified and more accurate.(2)Aiming at the situation that the offline model cannot be applied well due to the complex and variability of the mineral flotation process,On-line adaptive modeling method of correlation variable selection partial least squares regression based on Justin-Time Learning(JITL-CVS-PLSR)is proposed.The selection of similar samples is achieved through the combination of Euclidean distance and cosine distance,and the sample set most similar to the current working conditions is obtained.On this basis,the regression modeling method is combined to realize the online update of the pulp element grade prediction model,and the obtained online The model is more effective in the application of mineral processing.(3)On the basis of the above modeling method,a modeling and analysis system for slurry fluorescence measurement sequence is developed,which integrates complex algorithm modeling problems into interface tools that are convenient for operation and understanding,and realizes data import,data display,Data modeling,data analysis,and modeling report export and printing functions provide convenience for related workers.