基于支持向量机的在线建模方法及应用

基金项目: 国家973计划资助项目（2002CB3122000）；国家自然科学基金资助项目（60074027）；国家863计划资助项目（2003AA412010）

详细信息

An SVM-based On-line Modeling Method and Its Application

Research Institute of Automation, East China University of Science and Technology, Shanghai 200237, China

摘要: 针对常规v支持向量回归用于在线建模时存在的问题，提出了一种支持向量回归在线建模方法．利用贝叶斯证据框架优化模型参数，通过判断新增观测值是否满足原来的KKT条件，并对历史数据给予不同程度的加权以充分利用最新的数据信息，使模型随着时间的推移在线更新．工业PTA氧化过程中4-CBA含量预测的实例表明，该方法能很好地跟踪4-CBA含量的变化趋势，是一种有效的在线建模方法．
- 支持向量机 /
- 回归 /
- 在线建模 /
- 4-CBA
Abstract: In order solve the problems in the application of v support vector regression(v-SVR) to on-line modeling,a support vector regression on-line modeling method is proposed.Bayesian evidence framework is used to optimize the model parameters.Through determining whether the new observation satisfies the original KKT conditions and assigning different weighting factors to the historical data,the latest data can be used sufficiently,and the model can be refreshed on-line as time passes by.The proposed approach is successfully applied to predict the concentration of 4-carboxybenzaldhyde(4-CBA) in industrial purified terephthalic acid(PTA) oxidation process.The results indicate that the proposed method can track the trend of 4-CBA and it is an effective method for on-line modeling.
- support vector machine(SVM) /
- regression /
- on-line modeling /
- 4-CBA

[1]	Vapnik V. The Nature of Statistical Learning Theory [M].New York:Springer, 1995.
[2]	Vapnik V. Statistical Learning Theory [M].New York:Springer,1998.
[3]	Scholkopf B, Smola A J. New support vector algorithms [J].Neural Computation, 2000, 12(5):1207～1245.
[4]	Martin M. On-line Support Vector Machines for Function Approximation [R].Barcelona:Software Department, Universitat Politecnica de Catalunya, 2002.
[5]	Ma J S, Theiler J, Perkins S. Accurate on-line support vector regression [J].Neural Computation, 2003, 15(11):2683～2704.
[6]	阎威武,常俊林,邵惠鹤.基于滚动时间窗的最小二乘支持向量机回归估计方法及仿真[J].上海交通大学学报,2004,38(4):524～532.
[7]	周伟达,张莉,焦李成.支撑矢量推广能力分析[J].电子学报,2001,29(5):1～5.
[8]	周伟达,张莉,焦李成.自适应支撑矢量机多用户检测[J].电子学报,2002,31(1):1～6.
[9]	Mackay D J C. Probable network and plausible predictions-a review of practical Bayesian methods for supervised neural networks [J].Network:Computation in Neural Systems, 1995,6(3):469～505.