摘要
谷氨酸发酵过程具有高度的非线性和时变性,其内在机理复杂,简单的数学模型难以很好地描述其反应过程。在线支持向量机回归(OSVR)是一种新型SVM学习算法,采用增量和减量训练算法在线校正模型参数,可提高预测模型的准确度。标准的OSVR算法中核函数运算且关于核函数的数据并非在每一步中都被更新,被更新的数据仅仅占据一小部分。文中提出利用缓存保存核函数运算结果的改进OSVR,它不需要重复计算核函数,只需对其中更新部分重新运算,并运用改进的OSVR建立谷氨酸发酵过程模型。仿真结果表明,改进的OSVR提高了建模精度和在线学习速度。
Due to the complexity, high nonlinearity and time-variation of Glutamate fermentation process, it is difficult for simple mathematical model to describe the reaction process. Online support vector regression （OSVR） is a new type of SVM learning algorithm, which updates model parameters online by the incremental and decremental training algorithm, the accuracy of prediction model is improved. However, the calculation of kernel function of the standard OSVR algorithm is complicated, and only a small part of the data about the kernel function needs to be updated at each step. Therefore, an improved OSVR algorithm by using cache to store the kernel function matrix results was proposed. It avoids calculating kernel function repeatedly, only the updating part is calculated at each step. To verify the improvements of the improved OSVR, it was applied to Glutamate fermentation process modeling.. The simulation results show that the improved OSVR has good accuracy and can improve the algorithm speed.
出处
《江南大学学报（自然科学版）》
CAS
2013年第1期13-16,共4页
Joural of Jiangnan University (Natural Science Edition)
基金
国家863计划项目(2009AA05Z203)
关键词
在线建模
在线支持向量机回归
核矩阵
谷氨酸发酵
on-line modeling, OSVR, kernel function matrix, glutamate fermentation process
