Simulation And Modeling For The Air Dense Medium Fluidized Bed Coal Preparation Process

Coal resources are the main energy source of China and the main driving force of our economic and social development.However,the unreasonable processing of coal makes coal burning cause serious environmental pollution,so the coal preparation technology that can sort and improve the quality of coal is the key technology for the efficient and clean utilization of coal resources in China.As China’s coal resources are mostly distributed in arid and alpine areas,the application of wet coal preparation technology is restricted,so dry coal preparation technology is receiving more and more attention.Air Dense Medium Fluidized Bed(ADMFB)coal preparation technology is an efficient dry coal preparation technology that applies fluidized separation technology to the coal preparation process.Compared with the traditional wet coal separation method,the air dense medium fluidized bed coal separation method has the advantages of no water consumption,high sorting accuracy and less pollution,which is suitable for the sorting and quality improvement of coal resources in the arid and water-scarce areas of China and has a broad application prospect.Since the ADMFB coal preparation process is a complex long flow production process with complex process mechanism and variable production environment,it leads to the difficulty to realize the intelligent control and optimization of the operation status of the process.However,the prerequisite for control and optimization is a process model that reflects the main operating characteristics of the process.Therefore,this thesis focuses on the modeling problem of the ADMFB coal preparation process,and the main research elements are as follows:(1)In response to the problem that most of the existing modeling methods in the mechanistic modeling of ADMFB coal separation process only focus on the fluidized bed sorter itself,but ignore the dynamic characteristics of upstream and downstream processes of fluidized bed coal preparation.In this thesis,a research work is carried out to model and simulate the mechanism of ADMFB coal separation process.Firstly,the working mechanism of key equipment in each part of ADMFB coal preparation process is analyzed,and the dynamic mathematical model of each key equipment is established.Secondly,the mathematical models are linked together according to the process flow,and the construction and connection of the process models are completed in MATLAB/Simulink environment to form a complete dynamic model of the process.Then,the simulation platform of ADMFB coal beneficiation process is built based on Lab VIEW software,and the platform takes Simulink model as the data simulation background to realize dynamic simulation and simulation of the operating characteristics of air dense medium fluidized bed coal beneficiation process.Finally,the effectiveness of the proposed mechanism model is verified through the analysis of the overall simulation of the model and the simulation results of the key core equipment characteristics.(2)To address the problem of insufficient modeling data volume due to the short running time of the new process and the high cost of data collection in the data-driven modeling of the air dense medium fluidized bed coal beneficiation process.In thesis paper,Cycle GANs-based migration algorithms for old and new process data and multi-model migration strategies are applied to ADMFB coal preparation process modeling work.First,the mapping function between the old and new process data is learned by using Cycle GANs-based migration algorithm,and the new process data is correspondingly transformed into multiple sets of old process data containing new process information,and the old process data transformed from the new process data is added to the old process data,thus expanding the new process information in the old process data.Second,the expanded old process information is used to train multiple least squares support vector machines to obtain multiple old process models containing the new process information.Then,multiple old process models containing new process information are migrated using a multi-model migration strategy and trained with the new process data to obtain a prediction model of the new process.In addition,for the lack of model training conditions and supplementary training in the training process of this method,the stopping conditions and supplementary experiments of model training are added.Finally,the effectiveness of the proposed method is verified by conducting simulation experiments with the air dense medium fluidized bed coal preparation data set.(3)To address the problems of insufficient data volume,uneven data distribution and data privacy among fluidized bed coal preparation enterprises in modeling the air dense medium fluidized bed coal preparation process.Federated Learning(FL)and BP(Back Propagation,BP)neural networks are used to model the air dense medium fluidized bed coal preparation process.First,multiple fluidized bed coal processing plants are used as federation learning clients,BP neural networks as prediction models combined with a server side to establish horizontal federation learning.Second,the server side establishes a connection with the clients and transmits the initialized model parameters to each client,which uses local data for training and transmits the trained model parameters to the server side.Then,the server side uses Fed Avg algorithm to aggregate the model parameters of each client and sends the updated parameters to each client.Repeatedly,the client-server interaction training reaches the set conditions and completes the model building.This method can break the data exchange barrier among coal processing enterprises due to privacy protection and other issues,and combine the data information of multiple coal preparation plants or enterprises to collaboratively train the process prediction model.Finally,simulation experiments are conducted on the air dense medium fluidized bed coal preparation dataset to verify the effectiveness of the proposed modeling method.