Construction And Optimization Of Multi-parameter Filling Slurry Yield Stress Prediction Model Based On Machine Learnin

Lucid waters and lush mountains is the main theme of green sustainable development of mines.Filling mining technology is an important way for green sustainable development of mines.Intelligent filling is an important trend in the development of contemporary filling mines.However,in the construction process of intelligent filling system,the whole process of filling design is mainly based on experience intervention and basic test.It is difficult to reflect the concept of intelligent filling by referring to the key parameters of paste filling in the same type of mine and combining with indoor experiments.In the whole process of filling process,due to the complex composition of filling slurry and many influencing factors,it is difficult to determine the key parameters.The rheological parameters of the slurry are important parameters for the selection,design and judgment of the rheological constitutive of the subsequent filling pipeline.The traditional rheological parameter analysis and measurement methods are insufficient.Based on the basic characteristics experiment,fluidity experiment and rheological experiment of eight kinds of tailings materials,this paper combines the Bayesian model optimization Stacking integrated learning algorithm to effectively evaluate the yield stress of slurry with different ratio design.The main contents include the following points:(1)A variety of tailings physical properties experiments were carried out to analyze the differences of tailings materials and the influence of material differences on rheological parameters.Through pycnometer method,bulk density test,porosity test,laser particle size analyzer test and specific surface area analyzer test,the physical basic characteristic parameters of eight kinds of tailings have been understood.Combined with flow and rheological tests,it is found that tailings with low density and high packing density usually have smaller particle gaps,resulting in greater interaction between particles.In macroscopic mechanics,the tailings with low density have larger Bingham yield stress and worse fluidity.(2)The influencing factors and differences of flow and rheological properties of filling slurry were analyzed.More than 200 groups of flow test and rheological test were completed by self-made micro slump cylinder and RST-SST slurry rheometer.The flow characteristic parameters and rheological parameters of Dahongshan copper mine tailings,Lala copper mine classified tailings,Lala copper mine tailings,Qinghai copper mine tailings,Huidong copper mine tailings,Beiya gold mine tailings,Fengzhengshan 5000 t mixed tailings and Fengzhengshan 600 t mixed tailings under 25 kinds of ratios were obtained.The influence of mass concentration and sand-cement ratio on the filling slurry was analyzed.The mass concentration had a great influence on the flow and rheology of the slurry.When the mass concentration increases to a certain value,the flow and rheological characteristic parameters will change abruptly.The sand-cement ratio has little effect on the flow parameters and rheological parameters.Then,combined with H test,the flow and rheological differences between different mine slurry under the same ratio design are qualitatively and quantitatively reviewed.There are great differences between different tailings filling slurry.(3)A slurry yield stress prediction model based on BOP-Stacking ensemble learning is constructed.Through a number of indoor experiments,it provides original data set support for machine learning model construction.In this paper,the density,porosity,non-uniformity coefficient and curvature coefficient are selected from the basic properties of tailings.The mass concentration and sand-cement ratio are selected from the ratio design,and the slump and expansion are selected from the fluidity test.The yield stress is selected from the rheological test,which together constitute the original data set of the prediction model.The distribution characteristics of the original data set and the correlation of each variable level were analyzed by Shapiro-Wilk normality test and Spearman rank correlation coefficient.The variables showed a non-normal overall distribution form,and each level variable had a certain correlation with yield stress.The principal component analysis method is used to reduce the dimensionality of the original data set.Four principal components are selected as the input indicators of the BOP-Stacking model,and the yield stress is the output indicator.After the original data set is disrupted,it is divided according to the ratio of training set:test set=9:1.Combined with five-fold cross-validation,the slurry yield stress prediction model of BOP-Stacking integrated learning method is constructed.Compared with the R~2 between the predicted value and the true value of each single model,the BOP-Stacking model has been significantly improved,with a maximum increase of 82.97%.