Study Of Multi-Model Fusion Modeling Method And Its Application In Chemical Process Soft Sensing

Soft sensor technology is an effective way to solve the problem that process industry's important quality indicators can't be detected online.One of the core research contents is to establish a soft sensor model with good predictive performance.Because the actual industrial process has the characteristics of complex working conditions,strong nonlinearity,strong coupling,and time-varying characteristics,the soft-sensing model established based on the single-model method generally can't guarantee the global-wide prediction accuracy and the model stability is poor.The multi-model fusion modeling method can significantly improve the prediction accuracy and stability of the soft sensor model by more fully describing the process characteristics of the object.This paper proposed two multi-model fusing modeling methods and applied it to melt index soft sensing of polypropylene production process.The specific work is as follows:(1)By reviewing a large number of related documents at home and abroad,the basic ideas,main contents and implementation steps of the soft sensor technology and multi-model fusion modeling method were summarized.The multi-model fusion modeling method was divided into three categories: cluster analysis,ensemble learning and hybrid modeling.The research status,advantages and disadvantages of various methods were reviewed in detail.(2)An FCM-ABC-MKRVM multi-model fusion modeling method was proposed.Firstly,the fuzzy C-means clustering algorithm was used to divide the training samples into multiple sub-classes.Then the multi-core correlation vector machine sub-models were established by training the samples of each sub-category.The artificial bee colony algorithm was used to optimize the parameters of the kernel function and the combined weighting factors.Finally,the test sample was calculated.The membership value of each cluster center was used as the weight coefficient of the output value of each sub-model,and the model prediction output was obtained through multi-model fusion.(3)A selective integration limit learning machine modeling method was proposed.First,the original sample set was divided into training sample set,evaluation sample set and test sample set.The Bagging integration algorithm was used to randomly sample the training sample set,establish multiple subsets of training samples,and train each sample subset to establish the limit learning machine model.Using the evaluation sample set to evaluate each extreme learning machine sub-model,using clustering selection method to screen the sub-models for integration;finally using the entropy weight method to calculate the weight coefficient of each sub-model,through multi-model fusion to obtain the model prediction output;(4)Applying the two proposed multi-model fusion soft modeling methods to melt index soft sensor of polypropylene production process.The results showed that the FCM-ABC-MKRVM based polypropylene melt index soft-sensing model had good prediction accuracy and could provide guidance for on-line prediction of product quality indicators in complex multi-process chemical processes.Compared with the single model,the fully integrated model and the average weighted model,the polypropylene melt index soft sensor model based on the selective integration limit learning machine has higher prediction accuracy,and overcomes the training and prediction time of the traditional ensemble learning method.The long problem had improved the model performance to some extent.This paper had carried out the study work of multi-model fusion modeling method and its application on chemical soft sensing.Two multi-model fusion modeling methods had been proposed and applied in the soft index measurement of polypropylene process melt index.Good results had been obtained.The research work had certain reference and guidance significance for the further research on the multi-model fusion soft-sensor modeling method;it provided important reference for the on-line prediction of the quality indicators of polypropylene production process and the application research of modeling,control and optimization of other chemical production processes.Helps to increase the level of production operations in the process industry,meets the requirements of high efficiency and low consumption in the production process,and further enhances the market competitiveness of the company.