Soft Sensor Approaches Based On Deep Learning For Complex Industrial Process

The soft sensor technology reasonably selects other easily measurable variables according to the industrial process knowledge,and establishes a soft sensor model between the auxiliary variables and the dominant variables,and effectively estimates the dominant variables,which solves the problems that the important process variables cannot be directly measured due to the limitations of hardware sensors in the harsh environment such as high temperature and airtight.Soft sensor technology provides an effective means for the detection and control of existing sensors,and thus has become a research focus in the field of industrial control.In recent years,with the development of artificial intelligence,the deep learning for complex industrial sites has become a popular research direction.In this paper,using the strong learning ability of deep learning,combining soft sensor technology with deep learning,a novel soft sensor method of complex industrial process based on deep learning is proposed,and applies this novel model to predict rotor thermal deformation of rotary air-preheater,which provides more accurate deformation for rotor air leakage technology.The specific research of this paper are as follows:1.In this paper,the field data of the air preheater of a 600MW power plant boiler are collected first,and then the variables that have significant influence on the rotor thermal deformation are selected by the grey relational analysis(GRA)method to provide reliable input variables for the model training.Finally,the data are filtered and normalized,and the data after processing are selected as mutually exclusive training and testing sets by using the "reserve method" to provide the network model sample data.Through the selected sample data set,the comparative analysis of different model methods can be performed under the premise of the same sample data.2.Using the training set and test set obtained in(1),the network model of soft sensor method based on BP neural network and SVR algorithm are designed and simulated,and the deep learning soft sensor model based on DBN-DNN algorithm is constructed based on the analysis,and the advantages and disadvantages of the three models are analyzed by the mean square error of the training set and test set samples.The results demonstrate that the deep learning DBN-DNN soft sensor method has higher prediction accuracy,which provides a model basis for the construction of a novel type of soft sensor network.3.In order to further improve the prediction accuracy,the deep belief network(DBN)which can fully extract data features and the support vector regression(SVR)algorithm which has strong nonlinear regression ability are integrated to propose a novel soft sensor network structure.And the improved particle swarm optimization(IPSO)is employed to obtain the optimal parameter combination of support vector regression,which are applied to the new network,so as to propose the novel soft sensor model DBN-IPSO-SVR.The performance of the IPSO algorithm is compared with that of the PSO algorithm,the results demonstrate that the IPSO algorithm can improve the prediction performance.4.The new soft sensor model DBN-IPSO-SVR and the soft sensor model based on BP,SVR and DBN-DNN are illustrated and analyzed to highlight the advantages of the novel network structure.In addition,the new model is compared with the existing soft sensor method based on SAE for the prediction of rotor thermal deformation,and the soft sensor method based on MLP and ELM in different application fields,thereby highlighting the versatility and accuracy of the novel soft sensor model.The performance of the proposed new soft sensor model DBN-IPSO-SVR was evaluated.The experimental results demonstrate that the performance of the new soft sensor model is superior to other soft sensor methods.This novel soft sensor method significantly improves the performance of rotor thermal deformation prediction and is a valuable non-contact measuring tool for controlling air leakage of rotary air-preheater.