Research On Density Prediction Of Gas-solid Fluidized Bed Based On BP Neural Network Algorithm

Coal separation and processing is the most economical and effective way to make rational use of coal resources and protect the ecological environment in the era of green and low-carbon energy transformation.It is also a necessary means for deep coal processing and clean coal utilization.Dry coal separation technology,which does not require the use of water,has low cost and no pollution,thus providing an effective method for the efficient and clean separation and quality improvement of coal resources and the transformation and development of the coal industry.In recent years,intelligent separation has become a popular research direction in the coal industry,as it can increase the quality and added value of products,reduce production costs and energy consumption.In this study,experimental methods were used to obtain the initial fluidization velocity and spatial distribution of bed density data for multi-component mixed weighting materials,and the data of mixed weighting materials obtained by others were collected to construct a dataset of initial fluidization velocity and spatial distribution of bed density for multi-component mixed weighting materials.Based on this dataset and the BP neural network algorithm,a BP neural network model with 10 input variables(density,particle size,mass fraction,cross-sectional area of fluidized bed,static bed height)and initial fluidization velocity as the output variable was constructed,with a network structure of 10-13-1.The correlation coefficient of the network training is 0.9910,the mean absolute percentage error of the predicted data is12.59%,and the root mean square error is 1.9776.A multi-component mixed weighting material bed density prediction model with13 input variables(density,particle size,mass fraction,cross-sectional area of fluidized bed,static bed height,bed height,initial fluidization velocity,fluidization number)and bed density as output variable was established,with a network structure of13-14-1.The correlation coefficient of the network training is 0.9773,the mean absolute percentage error is 3.04%,and the root mean square error is 0.0844.Among them,the prediction accuracy of the model for single weighting material is higher,with a mean absolute percentage error of 4.17% and a root mean square error of 0.0298.Based on the multi-component mixed weighting material bed density prediction model,the genetic algorithm was used to optimize and adjust the model,resulting in the GA-BP neural network model.The correlation coefficient of the network training is 0.9824,the mean absolute percentage error is 2.46%,and the root mean square error is 0.0747.Among them,the optimization effect of the model for single weighting material prediction is the best,with a root mean square error reduced by 27.41% compared to the original model.Combining the minimum fluidized gas velocity and bed density prediction model,a multi element heavy bed density prediction mechanism is constructed,with the properties of heavy particles,key data of fluidized bed equipment,and operating gas velocity as key parameters to achieve the prediction of multi element heavy bed density.The parameters required for this prediction process are all basic parameters of gas-solid separation fluidized beds,which can be obtained without complex measurement methods.It can effectively solve the problem of difficult to accurately control the fluidization effect in the industrial application process of gassolid separation fluidized beds.This thesis contains 57 figures,11 tables,and 89 references.