Study Of Prediction Control On Heavy Medium Density In The Process Of Coal Preparation

Heavy medium coal preparation has many advantages of high separationprecision, easy operation and easily realizing automation, so it has a significantposition in coal production in china and it is the main direction of coal preparationdevelopment in the future. In recent years, control technology of coal preparation bydense medium has been made a rapid progress, but it was failed to solve somefundamental questions in the production process, such as feed-back control of coalash content apparatus, real-time prediction and intelligent control of separationparameters, which have seriously hampered the automation development of heavymedium separation. From the point of control, the study is carried out from thefollowing three aspects: information feedback, parameter given and density control ofheavy medium.This paper analyzed the measurement principle of coal ash content apparatus andconcluded that the ash error is related to the attenuation coefficient of radioactiveelements for coal, bulk density and thickness of coal, and the measurement sensitivityis mainly related to the quality of coal thickness according to the ash formula of coalash content apparatus derivation. Combing with actuality usage of coal ash contentapparatus, the paper concluded that the ash error is formed by three aspects includingcoal quality, coal flow conditions and daily maintenance and management. Accordingto raw coal and clean coal, the paper summarized the output errors about which issystematic error and controllable and which is random error and uncontrollable in thethree aspects. The final conclusion is that the output error of ash content apparatus isregular and can be corrected through the method of data mining.Selecting ash data of coal ash content apparatus and laboratory data fromdifferent coal mines as samples, calibration model of ash is established to make ashdata correct by training samples data using the LS-SVM and fuzzy LS-SVMalgorithm. Calibration results show that the correction effect of Fuzzy LS-SVM isbetter than LS-SVM owning to a fuzzy membership is introduced into FuzzyLS-SVM to improve the stability of the model and the sparsity of LS-SVM; Thecharacteristics of cleaned coal is relatively stable and the characteristics of raw coal ismore complex, so the actual correction effects of cleaned coal is better than raw coal.The methods of optimizing Kernel function parameter about LS-SVM have somedisadvantages, such as: parameter choice blindness, slow convergence speed, long running time, and so on. So the cuckoo (Cuckoo Search) search algorithm iscombined with fuzzy LS-SVM to optimize the parameter of kernel function, and itcan improve the response speed of ash monitor and achieve the calibration model ofoptimal performance.The paper concluded that cleaning coal ash would change in relation to rawcoal ash and density of heavy medium suspension. And the affection of raw coal ashto cleaning coal ash lags behind the affection of density of heavy medium suspensionto cleaning coal ash by analyzing the data change trend of raw coal ash, cleaning coalash and density of heavy medium suspension in the process of coal production.Selecting two hours of production data from Xingtai coal preparation plant, utilizingMean Completer algorithm and reconstructing by using phase space reconstructionmethod, density given prediction of heavy medium suspension is achieved byadopting time series LS-SVM algorithm after the production data is completed.Evaluating prediction effect by using root mean square error for different embeddingdimension and different sample size, the best training parameter of model is thatembedding dimension is7and sample size is80. Final calibration results show thatthe prediction effect based on LS-SVM without reconstructing data is poor, whileprediction effect based on time series LS-SVM is better, and the change trends ofprediction data about density given are consistent with the actual data, the predictionerror is small.This paper introduces the technological process of heavy media coal preparation.The transfer matrix of heavy medium density control system using the formula ofvolume balance and medium balance is deduced according to the relationshipsbetween input and output about heavy medium suspension in media bucket, finallythe parameter of transfer function matrix is determined through experimentalmodeling. The relative gain of the transfer function matrix is0.69, which shows thatthe density and liquid level control system of heavy medium suspension is a strongcoupling, long time-delay and multivariable system, so the decoupling controltechnology must be realized in this system.Decoupling controller is designed based on feed-forward decoupling algorithmin combination with GPC algorithm, the controller is carried by using decouplingcontrol on the density of heavy medium suspension and liquid level control system,the simulation result shows that the control effect based on the algorithm offeed-forward decoupling and GPC is better when the model fits on the actual system, the system output is still stable under the interference of liquid level output anddensity output, but the system stability deteriorates and even can’t maintain outputstability in the case of model mismatch. GPC algorithm of parameter self-reverse isused to solve the model mismatch issue for the decoupling control of density andliquid level system about heavy medium suspension, In order to eliminate couplingeffect, coupling relationship of the two subsystems can be eliminated by using othersubsystem control increment of previous time to replace current time controlincrement in the process of calculating one subsystem control increment. Thesimulation result shows that control effect based on GPC algorithm of parameterself-reverse is better, it has strong anti-interference capability, the system output isstill stable in the case of model mismatch. From the process of solving optimal controllaw, it is easy to see that U can be calculated by solving inverse matrix, and thecomputational complexity is increased. In order to reduce the calculation amountonline to meet the requirements of real-time optimization, control increment wasprocessed for constraint, and then solving inverse matrix was avoided to improve thesystem real-time, the system tracking speed was improved and the system overshootwas suppressed.Data exchange between VB and AB PLC can be realized using VB calleddynamic link library of OPC after configuring communications software of rslinx; theprediction algorithm is implemented through using VB called related functions, thedynamic link library of LS-SVM and GPC can be compiled by MATLAB, finally theprediction control software about the density and liquid level control system of heavymedium suspension is designed.