CAJ | 학술논문

针对间歇生产过程中，采集的数据存在非高斯、非线性的特征，本文将支持向量数据描述（Support Vector Data Description， SVDD）的方法应用到间歇过程故障监测中。首先，将数据按照批次展开并进行标准化，再按照变量展开；然后，建立 SVDD模型，应用核函数求出模型半径 R；对新的待检测样本，先计算其与模型中心的距离，再与半径比较，判断它是否正常。因为SVDD可以利用核函数替代向量内积的计算，所以能够解决非高斯、非线性数据的检测问题。最后，在青霉素发酵过程监测的成功应用，验证了该方法的有效性、准确性。
침대간헐생산과정중，채집적수거존재비고사、비선성적특정，본문장지지향량수거묘술（Support Vector Data Description， SVDD）적방법응용도간헐과정고장감측중。수선，장수거안조비차전개병진행표준화，재안조변량전개；연후，건립 SVDD모형，응용핵함수구출모형반경 R；대신적대검측양본，선계산기여모형중심적거리，재여반경비교，판단타시부정상。인위SVDD가이이용핵함수체대향량내적적계산，소이능구해결비고사、비선성수거적검측문제。최후，재청매소발효과정감측적성공응용，험증료해방법적유효성、준학성。
For batch production process, non-gaussian and nonlinear characteristics also exist dataset. Support vector data description (SVDD) method is used in this paper. Firstly, the dataset is first unfolded though the batch and standardization should also be performed nextly, then it is re-unfold through the variable direction. After that, a SVDD model can be built and kernel function is applied to solve the radius of the model, for the new samples to be detected, calculating the distance to the model center, comparing with the radius and then which can be determined whether it is normal. Because the SVDD could use kernel function instead of inner product of vector computation, it can solve the detection problem of nonlinear and non-gaussian data. Finally, in the monitoring of the successful application of penicillin fermentation process, SVDD is verified to be effective and accurate.