Based On Ls-svm Soft Sensor Modeling Method

In modern complicated industrial process, some variables are very hard to be measured or even cannot be measured on-line by existing instruments and sensors. Soft sensor is an effective means of implementing the on-line evaluation of these variables. The rare-earth process by countercurrent extraction is complex, characterized of nonlinearity, time-variant properties, and severe lags. Component content is difficult to be measured online. To resolve this problem, an online prediction method of component content using soft sensors based on least squares support vector machine was proposed, through analysing the current state of component content online measurement and the disadvantages of traditional soft sensor modeling methods.The author mainly focuses on the research on soft sensor modeling method based on LS-SVM. The main works of the thesis are listed as follows:(1) The disadvantages of traditional soft sensor modeling methods were summarized, and an online prediction method of component content using soft sensors based on least squares support vector machine was proposed. Then, auxiliary variable selection and sample data processing were presented.(2) The LS-SVM for regression modeling was studied, since the drawback of sp(?)ness lost within the standard LS-SVM, an improved LS-SVM based on Suykens's sparseness algorithm was proposed.(3) A parameter selection method based on grid search was studied to select LS-SVM parameters, an improved soft sensor method based on LS-SVM is proposed to build component content soft sensor. Simulation results show that the model based on LS-SVM is characterized of stronger ability to generalization than that based on RBF neural network. However, this parameter selection method is time-consuming, and based on interpolation validation without global unified adjustment. So the parameters are not always optimum.(4) To resolve the disadvantages of parameter selection method based on grid search, quantum particle swarm optimization (QPSO) algorithm was presented to select the parameters. Simulation results indicate that the method based on QPSO is fast and precise. A method of soft sensor based on QPSO and LS-SVM was prososed, and soft sensor model of component content was established, simulation results show that the proposed model is precise.