摘要
针对全局建模方法很难精确描述实际生产过程,提出了一种模糊支持向量机回归建模算法,并推导出相应的增量与减量算法;在此基础上,提出了在线模糊支持向量机回归建模方法,该方法利用滚动时间窗内的数据优化建模,随着时间窗的滚动,在原有模糊支持向量机模型的基础上通过增量与减量算法实现参数的快速在线更新。通过将该方法用于丙烯腈收率的预测建模,结果表明,所提方法具有参数调整时间快、泛化能力强的优点,可以较好的跟踪丙烯腈收率的变化。
Since the global modeling approach is difficult to perfectly describe actual industrial process, a fuzzy support vector machines (FSVM) regression modeling method and its increment and decrement algorithms were proposed in this paper. Based on these, an on- line FSVM regression modeling method was also proposed, which used the samples in the time window to build the dynamic system model, and with the slide of the time window and based on the trained FSVM model, the proposed increment and decrement algorithms were used to update quickly on line. The proposed method was applied in predicting the yield of acrylonitrile. The results demonstrate that this method is effective, which can better trace the change of acrylonitrile yield.
出处
《石油化工高等学校学报》
EI
CAS
2005年第4期74-79,共6页
Journal of Petrochemical Universities
关键词
丙烯腈收率
模糊支持向量机
回归方法
Acrylonitrile yield
Fuzzy support vector machines(FSVM)
Regression modeling method