摘要
提出一种采用改进核Hebbian算法的加氢脱芳烃过程的递推产品质量建模方法,用于实时估计终端分馏产品的质量指标。通过利用核Hebbian算法的中间结果,计算中心化的核矩阵特征值,进而由核主元回归方法得到非线性动态质量模型。该递推滑动窗建模方法无需计算和保存整个核矩阵,并验证了所得到的闪点模型在正常和故障工况下均具有足够的精度。
A modified kernel Hebbian algorithm （MKHA） was proposed to integrate with the kernel principal component regression （KPCR） method for recursive product quality modeling of a two-stage hydro-dearomatization process. The approach to calculating the eigenvalues of centering kernel matrix was derived and the whole algorithm is formulated in a recursive mode. The proposed modeling strategy has an advantage of no need to calculate and store the kernel matrix. The obtained recursive nonlinear dynamic flash point model showed satisfying precision under both normal and faulty operations, and comparison studies with traditional offline KPCR modeling were presented.
出处
《化工学报》
EI
CAS
CSCD
北大核心
2007年第6期1518-1522,共5页
CIESC Journal
基金
国家自然科学基金项目(20576116)
德国洪堡基金会资助。~~
