基于动态MDONPE算法的间歇过程故障检测

针对间歇过程数据存在的非线性和动态特性导致故障检测效果不佳的问题,提出一种基于滑动窗(sliding window,SW)的多向差分正交邻域保持嵌入(multiway differential orthogonal neighborhood preserving embedded,MDONPE)算法.首先对间歇过程数据...

Fault detection of batch process based on dynamic MDONPE algorithm

To solve the problem of poor fault detection effect due to the nonlinear and dynamic characteristics of the data in the batch process, a multiway differential orthogonal neighborhood preserving embedding (MDONPE) algorithm based on the sliding window (SW) is proposed. Firstly, the data of the batch process is preprocessed to find the nearest neighbors of the samples, and the difference operations between the samples and the nearest neighbors is carried out. Then the orthogonal neighborhood preserving embedding algorithm with orthogonal constraints is obtained by orthogonalizing NPE algorithm, and the orthogonal neighborhood preserving embedding algorithm is used to reduce dimensions and extract features. The sliding window strategy is used to combine and achieve the error accumulation by selecting the sampling data within the window width, which can make the features of the fault samples more obvious. Finally, the fault is judged by detection, and the T² and SPE statistics are used to judge if faults have occurred. The data of the Penicillin fermentation simulation process are used and compared with MPCA and KNPE algorithms. The results show that the proposed algorithm has better detection effect than other algorithms in fault detection.