摘要
摘要
Subject:Research on Gas Ingredient Modeling Based On Support VectorMachine Specialty:The Technology of Computer Application Name:Zhao Hui Supervisor: Huang Jing-tao Associate Professor
ABSTRACT
Gas ingredient of boiler combustion directly reflects the combustion states, and it is an important indicator of the operation and economic combustion of the boiler, therefore, it’s necessary to analysis the gas ingredient.
Support vector machine (SVM) is a new machine learning method based on
statistical theory. SVM can effectively solve the practical problem with small sample size and high dimension. Now, it has been widely used to solve the problem of regression estimation, pattern recognize and so on. Based on these advantages, SVM has provided a theory foundation and ideas for the analysis of gas ingredient.
A data-based nonlinear system was proposed by using SVM theory in this paper,
and the model was built for the analysis of gas ingredient which was hardly online measured. Firstly, in order to solve the problem of the sparseness lacking in the least squares support vector machines (LS-SVM). A new least squares support vector machines based on the boundary samples was proposed. The new algorithm can properly obtain the sparse solutions to the LS-SVM, and the speed of computing was also improved. It also provided a new and effective method of selecting the training data for gas ingredient , Secondly, according to the relations of support vector, center distance ratio, incremental learning and margin vector, a new LS-SVM algorithm for the unbalanced samples was proposed. This method effectively improved the classification accuracy of LS-SVM for the unbalanced samples. The method was used
to build model for
NOx
in the gas ingredient, and provided a way to build model for
unbalanced data in the complex industrial systems, At last, a new LS-SVM modeling algorithm was proposed by combining sparsity algorithm and particle swarm opti