Intelligent Research On Contorl Of Density Of Heavy Medium Based On Data-driven

It clearly proposes that information technology represented by big data,cloud computing and artificial intelligence will become a new engine of economic growth in China-made 2025.Intelligent coal preparation plant is still in its infancy,and there are still many problems in equipment,technology,process and management.Building an intelligent coal preparation plant in an all-round way is a major strategic measure to enhance the core competitiveness,which is of great significance to accelerate the transformation and upgrading of the coal industry and cultivate new economic growth points.Therefore,this paper takes density as the main line throughout the whole paper.From the four aspects of ash analyzer upgrading,online density prediction,high-precision separation control,intelligent control system,based on online and offline data,the intelligent control of dense separation process is studied.This paper analyzes the measurement principle and detection structure of the active ash analyzer.For the traditional dual energy gamma ray ash analyzer,it is found that the special point of its structure is single point detection technology.Therefore,its hardware is optimized to multi-point full section detection,which realizes the basic full coverage of the cross-sectional area of the detected coal seam,eliminates the accidental error and improves the representativeness.At the same time,in order to integrate the ash content data of the multi-point detector,according to the multi-point installation mode and the distribution of coal flow,a variety of mathematical weight coefficient distribution calculation is applied,and it is found that the weight coefficient calculated by the least square method can integrate the error of multi-point detection and minimize the overall error,which is used as a static calibration.The results show that the calculated ash value is basically better than any single point ash data,which can greatly improve the accuracy of the ash analyzer from the hardware and software algorithm.The modified ash analyzer is connected to the site,and a group of density data of clean coal ash,raw coal ash and suspension collected at equal intervals during the production of the system are selected;through the analysis of heavy medium separation process,raw coal entering the system will go through many separation and classification equipment,so that the real-time raw coal information data,product information data and suspension density data collected at the same time are not relevant,Therefore,the ash content of clean coal product is the result of multivariate influence related to time.The time series LSTM long and short-term memory network is applied to the establishment of the density prediction model.By determining the optimal time steps(i.e.delay time),the number of hidden layers and nodes,and comparing with the traditional neural network without time series,the results show that The accuracy of LSTM model(MAPE 0.007g/cm~3)is 0.008 density points higher than that of traditional BP neural network model(MAPE 0.015g/cm~3).Therefore,the mathematical relationship among medium density,raw coal ash and clean coal ash can be established by using the collected historical data.Then,the suspension density can be predicted on-line according to the ash content of clean coal products.Set the output of online density prediction results to the production system.In order to ensure that the suspension can be stable within the allowable error of the set density,the traditional PID technology with good effect is used to control the pump root make-up valve with fast response speed.The modeling analysis is focused on the opening of the diverter valve,and the opening of the make-up valve is also introduced into the input variables to form the density of the suspension and the magnetic substance of the suspension GA-SVMR genetic support vector regression model with 4 inputs of content,qualified medium level,make-up valve opening and 1 output of diverter valve opening.The simulation results show that the predicted opening of the diverter valve is almost identical with the field opening,and the correlation coefficient is as high as0.9960,the average error is 0.15%,which is highly suitable for the field operation.The trained model is embedded in the system,and the real-time setting and control of the opening of the shunt valve are carried out according to the given predicted density and other parameters.On the basis of improving the precision of ash analyzer,on-line prediction of density and high precision control output of density,an intelligent control framework for dense medium separation process is proposed,which is composed of double closed-loop intelligent control system and data-driven system.Among them,data-driven includes three links:data preparation,data mining and application inheritance.According to the mass production data of coal preparation plant,data preparation is completed by setting storage rules and collection rules to prepare for the prediction of data mining.At the same time,all kinds of software and hardware in the system communicate with each other and interact with each other,and use iFix upper computer software and S7-300 lower computer programming to display the visual interface of the system production process,design the adjustment mode of the manual,automatic and intelligent control system,and realize the construction of the data driver system.Finally,the field application results are evaluated and analyzed,which shows that based on the data-driven heavy medium separation the effect of intelligent control system is pretty good.There are 41 figures,19 tables and 96references in this paper.