基于DMOLPP的间歇过程在线故障检测

为了保持过程数据集的局部结构,确保数据投影后的投影向量正交,降低数据误差重构方面的难度,提出了一种基于动态多向正交局部保持投影(DMOLPP)进行间歇过程故障检测的方法。该方法将滑动窗口技术和正交局部保持投影(OLPP)相结合用于间歇过程在线检测。首先,将批次数据展开成二维数据,利用滑动窗口技术分别在时间片内运用OLPP算法提取能表征过程正常数据内在局部近邻结构的特征;然后,对于新来批次数据标准化处理后分别在相应窗口内投影,提取特征向量;最后利用核密度估计(KDE)确定控制限进行过程检测。通过仿真结果表明,运用DMOLPP算法检测到故障发生的时刻早于动态多向局部保持投影(DMLPP)、动态多向邻域保持嵌入(DMNPE)方法。与动态多向主元分析(DMPCA)相比,具有较低或者无误报时刻,验证了该方法的有效性。

On-line fault detection of batch process based on DMOLPP

To keep the local structure of data set of a process, ensure that the projection vector of the data are orthogonal and reduce the difficulty of data error reconstruction, a dynamic multiway orthogonal locality preserving projections (DMOLPP) method for fault detection of batch process is proposed in this paper. Combining the moving window technology with orthogonal locality preserving projections (OLPP), this method is applied to on-line monitoring of batch process. Firstly, the batch dataset was unfolded into a two dimensional dataset, then the inherent characteristic of the local neighbor structure of normal process data was extracted within the time slice using OLPP algorithm with moving window technology. Secondly, after standardized, the new batch data was projected in the corresponding window and the characteristic vector was extracted. Lastly, the kernel density estimation (KDE) was used to determine the control limit and conduct process monitoring. The simulation results show that using DMOLPP algorithm, the time for detecting fault batch is earlier than those for the dynamic multiway locality preserving projections (DMLPP) and dynamic multiway neighborhood preserving embedding (DMNPE) methods. Compared with dynamic multiway principal component analysis (DMPCA), DMOLPP has lower or no fault detection numbers of normal time, which verifies the effectiveness of the proposed method.