The Study Of Modeling Algorithm Based On LS-SVM And Predictive Control Algorithm

The problems of nonlinear modeling and predictive control design based on Least Squares Support Vector Machines(LS-SVM) method are mainly studied in this dissertation. The main contents are outlined as follows,1. A modified least squares support vector machines (LS-SVM) approach, which treats the training data points differently according to their different degrees of importance, is proposed. On each data point, a support vector degree is defined and it is associated with the corresponding absolute value of Lagrange multiplier. The experiment of identification for a pH neutralization process indicates the validity of the presented method.2. A Recursive Least Squares Support Vector Machines (RLS-SVM) and a Recursive Fuzzy LS-SVM method are proposed. By recursively calculating the parameters of identified model, the huge calculation of matrices inversions can be avoided in case of the number of training data increases. The experiment of online modeling for a pH neutralizing process shows the validity and effectiveness of the presented algorithms.3. A sparse approximation algorithm based on projection is presented in order to overcome the limitation of non-sparsity of online LS- SVM. The new inputs are projected into the subspace spanned by previous basis vectors (BV) and those inputs whose squared distance from the subspace is higher than a threshold are added in the BV set while others are rejected. In addition, a recursive approach to deleting an exiting vector in the BV set is proposed. Then a sparse online LS-SVM algorithm, which can control the size of memory irrespective of the processed data size, is presented based on the online LS-SVM, sparse approximation and BV removal. The obtained algorithm is applied in the online modeling of a pH neutralizing process. The results show that the presented algorithm can greatly improve the sparsity on just little cost of precision.4. For non-uniformly distributed training data, a local weighted LS-SVM method for the online modeling of continuous process is proposed. At each period, only the samples similar to the current input are added into the training set, and the obtained model is just for predicting the current output. To distinguish the importance of the training data, weight is defined to each data according to the Euclidean distances between the training data and testing data. In this way, the training data participate the learning at different weights according to the similarity to the testing data. The presented LW-LSSVM algorithm is applied in a pH neutralization process and an isomerization process, respectively. The result shows the excellent performance of the presented algorithm in precision and predicting time.5. The modeling of multi-variable systems by LS-SVM is studied and the corresponding sparseness is discussed. The standard LS-SVM, nonsparse online LS-SVM and sparse online LS-SVM of different threshold are studied in the modeling for the isomerization of C₈ Aromatics. The results indicate the favorable performance of multi-variable sparse online LS-SVM.6. A generalized predictive control (GPC) algorithm based on online LS-SVM is proposed. The online LS-SVM algorithm is applied in GPC to deal with nonlinear modeling problems. At each sampling period, the algorithm recursively modifies the model to obtain a more accurate predictive output. The nonlinear LS-SVM model is lineared at each sampling period to the implementation of GPC. The experiments of LS-SVM based GPC on pH neutralizing process show the effectiveness and practicality of the proposed algorithm.