Cement Clinker F-CaO Soft Sensor Model Based On Time Series Analysis&Support Vector Machine

The content of free calcium oxide(f-CaO)in cement clinker is one of the most important indexes to estimate the quality of the cement clinker.At present,the content of f-CaO is mostly manually measured offline in the laboratory with a sampling period of 1 hour.Therefore time delay takes place between the control of the cement clinker quality and the measurement result,which makes the clinker quality real-time control impossible.Thus the automatic detection of f-CaO content is of great significance.The development of soft-sensor technology provides a new way to detect f-CaO content automatically.However,most of the present soft-senor models of detecting f-CaO(such as SVM model,ELM model,BPNN model)ignore the time lag and inertia in the cement production process.Aiming to automatically detect the content of f-CaO in cement clinker,this paper funded by NSFC project deeply analyzes the timing relationships between the process parameters based on a cement plant located in Jiang Xi province.Moreover an f-CaO soft-sensor model for cement clinker based on timing analysis and SVM is put forward in the present work,which equips high theoretical significance and important practical value.Main research accomplishments and contributions are given as follows:(1)Based on the introduction of NSP cement clinker calcination,the calcination mechanism of the material is analyzed in detail according to the fluxion of the material and the formation principle of f-CaO.Then six major influencing factors(mass feed rate,temperature of decomposition furnace,host current,rotation speed of the kiln,wind pressure of kiln head,wind pressure in second chamber of cooler)of f-CaO are selected as the inputs of the model.The field data from cement plant are preprocessed according to the time lag and inertia in the cement production process.Original data sets are denoised by mean filter and reduce the amount of data.Then,3σ law is used to detect the outliers in the data sets caused by equipment interruption and sensors failure.After preprocessing,690 reliable training and testing sets are obtained.(2)Considering the time lag and inertia in the cement production process and analyzing the sequential relationship among process parameters,the residence time of material in decomposition furnace,preheater,rotary kiln and the cooler is calculated respectively.And time series matching scheme among inputs are formulated.Aiming at the inertia in the cement production process,exponential decay function is applied to conduct local weighting process of every single input variable.Then,the f-CaO content soft-sensor model based on time-analysis and SVM is proposed,which is programmed in MATLAB.Then the former 670 data sets are used to train the model,and the latter 20 data sets are taken to test the model.Test results show that the evaluation index mean square error(MSE)is 0.10 and square correlation coefficient(R2)is 0.75,which verifies the proposed model.(3)The influence of time series parameters to the proposed model is discussed.Then the model proposed in this work is compared with the ELM model and BPNN model based on the same database.Final results present that both of the MSE and R2 of the proposed model are better than that in ELM model and BPNN model,which proves the availability of the model in this paper.