Research On Intelligent Prediction Model Based On Depth Thickening Deductive Mechanism Of Unclassed Tailings

The deep thickening of unclassed tailings is the primary link of unclassed tailings paste filling process,the prerequisite for realizing paste filling and the favorable guarantee for continuous filling mining.At present,the research on the deep thickening behavior of total tailings focuses on particle flocculation,change of floc structure,shear drainage,etc.,but the factors affecting the thickening behavior are not fully considered,the influence mechanism of different factors on the deep thickening effect is not clear,and the universality of the mathematical model of deep thickening limited by specific mine data is low,It limits the popularization and application of unclassed tailings paste filling in mines.Therefore,based on a large number of model tests and industrial tests,this paper comprehensively uses the means of theoretical analysis,micro test,artificial intelligence algorithm,numerical simulation and field test to clarify the influencing factors and laws of unclassed tailings depth thickening behavior,reveals the evolution mechanism of unclassed tailings depth thickening behavior,and establishes an intelligent prediction model of unclassed tailings depth thickening.The main research results are as follows:（1）Through theoretical analysis,indoor dynamic settlement and semi industrial dynamic settlement tests,the depth and density test parameters of total tailings in 36 mines are mastered,and the influencing factors are put forward,which are the physical properties of total tailings（loose density,true density,porosity,particle size gradation）,feeding parameters（feeding speed,dilution concentration）and mud layer height,The qualitative relationship between each factor and underflow concentration and yield stress of thickener is analyzed.（2）Based on the theory of settlement concentration behavior of unclassed tailings,a semi quantitative evaluation system for the meso structure of unclassed tailings mortar is established,which takes the exposure state of coarse particles,floc size,floc compactness and structural porosity as the indexes;Combined with indoor static/dynamic settlement test,scanning electron microscope and CT scanning technology,the dynamic evolution laws of total tailings specific gravity,particle size gradation,mud layer height and evaluation indexes are revealed.（3）Based on the results of unclassed tailings thickening test under multi factor conditions,using advanced artificial intelligence technology,taking underflow concentration and yield stress as output factors,bulk density,true density,porosity,d₁₀,d₃₀,d₆₀,d₅₀,non-uniformity coefficient,curvature coefficient,dilution concentration and feeding speed as input factors,a PSO gradient elevator（GBM）based on particle swarm optimization is proposed The intelligent prediction model of unclassed tailings depth thickening based on the algorithm,and through the analysis of local dependence diagram and importance score,the weight distribution of various influencing factors is quantified,so as to realize the accurate prediction of unclassed tailings thickening behavior.（4）Based on COMSOL multiphysics software,the numerical model of unclassed tailings depth thickening of paste thickener is established,the depth thickening law of unclassed tailings in paste thickener under different specific gravity,particle size gradation and mud layer height is explored,and the mathematical relationship model between underflow average volume integral of thickener and mud layer height is constructed.（5）The intelligent prediction model is applied to predict the deep thickening behavior of total tailings in a gold mine in Hunan,and the engineering design of paste thickener is carried out.By monitoring the operation state of the filling system of the mine,it is found that the error between the predicted value of underflow concentration and the application result is 2.24%.The maximum working torque in the monitoring cycle is 315000 N·m,and the torque percentage is 28%,which meets the design requirements of general paste thickener,and verifies the accuracy of the intelligent prediction model of unclassed tailings depth concentration.There are 105 figures,27 tables and 205 references in this paper.