Muti-Model Modeling Method And Its Applications In Soft Sensor

Soft-sensing technology is an important method to solve the real-time estimation of unmeasured variables in the field of process control and process detects. With the increasingly complex of the industrial process, the soft-sensor model established by single-model is often difficult to meet the accuracy of the complex system which is nonlinear, multivariable and time-varying. Multi-model soft sensor modeling methods can describe the process properties of the object better, while the prediction accuracy and the robustness of the soft sensor model can be improved significantly. A practical industrial process is taken as a background in this paper, and the methods of establishing multi-model soft sensor model are researched from model combination and the improvement of classification accuracy.Considering the limitations of the single modeling method, a multi-model modeling method based on nonlinear regression and support vector machine (SVM) is presented. The model is divided into two parts by this method, which one part of the model is the simple nonlinear model to estimate the overall trend of the object, and the other part is a combined model of support vector machines to describe the local variation characteristics of the object. Finally superimpose the combined model and the nonlinear regression model to form the multi-model soft sensor model. This modeling method is applied to a soft sensor modeling of the catalytic activity in a production process of Bisphenol A. The simulation results show the effectiveness of the method.The traditional multi-model modeling method does not consider the output error of the model during the clustering process, A multi-model modeling method based on supervised affinity propagation clustering algorithm is proposed. The principle is that the initial clusters are first obtained by the affinity propagation clustering algorithm, and then the clusters are adjusted cycled in accordance with the output errors until the minimum error. Finally the accurate clusters are got, and the sub-models are respectively built by least squares support vector machine so as to estimate the output.The value of K is difficult to be exactly determined in K nearest neighbor algorithm. A gaussian process multi-model modeling method is proposed based on the idea of AR model. The model output value of previous moments is introduced into the input set of the current moment. Calculate the mean minimum distance of the training samples to get a search radius. Determine the value of K according to the radius and calculate the weights of the output according to the K neighbor samples. Finally take the weighted output mode to get the output of combinational model. The method is used for the soft-sensor model to estimate the content of phenol at the outlet of a reaction vessel in a Bisphenol A production process. The simulation results show that the method has a higher accuracy and better model generalization ability.