Study On Intelligent Control Of Coal Slurry Sedimentation Based On Data-driven

The influencing factors of coal slurry sedimentation process are numerous,the mechanism is complex and it is difficult to establish a mathematical model.Its nonlinear,multivariable and strong coupling characteristics bring great challenges to the process control and restricts the improvement of the efficiency of the whole coal slurry treatment process.Therefore,this paper collects the historical data of coal slurry sedimentation process,and studies the process control method based on data-driven.The purpose is to realize the intelligent control of the coal slurry sedimentation process.In this paper,the parameters of coal slurry sedimentation process in a coal preparation plant were detected and collected.Firstly,the influencing variables of the sedimentation process were analyzed.Combined with the existing detection instruments and methods,seven variables were selected as the input and output variable of the prediction model,including four sets of online data of raw coal quantity,flocculation dosage polycoagulant dosage and overflow turbidity,and three sets of offline data of raw coal ash content,underflow concentration and slime amount.Secondly,according to the existing control network of coal preparation plant,the data acquisition method and software were designed,and the process parameters were detected,collected and stored.In order to solve the missing and abnormal problem of the collected historical operation data of coal slurry sedimentation process,a systematic data quality control method was studied.First,the box plot method was used to detect and eliminate the abnormal data.Secondly,the cubic spline interpolation method was used to fill the missing values of four online data of raw coal quantity,flocculation dosage,polycoagulant dosage and overflow turbidity.For the offline data of raw coal ash content,underflow concentration and slime content,daily constant value was used to fill it.Finally,the extended Kalman filter was used to data noise reduction.After a series of data processing,the data used for modeling were obtained.Based on the historical data of the production process,a data-driven dosage prediction model for the coal slurry sedimentation process was established.Firstly,the control objective of the system was determined as the overflow turbidity,and the control quantity is the dosage of flocculant and polycoagulant.Secondly,three machine learning algorithms-BP neural network,RNN and LSTM were used to establish the prediction model,and the network parameters were adjusted to achieve the best prediction effect.The prediction results of the three models show that the performance of RNN and LSTM based on time series is better than BP neural network without time series,while the evaluation criteria of LSTM algorithm are all the best.Therefore,LSTM algorithm was finally selected to establish the reagent dosage prediction model for the coal slurry sedimentation process.On the basis of the reagent dosage prediction model,a set of intelligent control system of coal slurry sedimentation process was constructed.The prediction results of the reagent dosage prediction model were taken as feedforward input,and a fuzzy controller was designed to compensate the prediction results by using the feedback of real-time data of overflow turbidity.The intelligent control of the process was realized through the development of i Fix monitoring software of upper computer and PLC program of lower computer.There are 37 figures,15 tables and 92 references in this paper.