Abstract:
Aiming at the irregularity of data density in multi model processes, most of the existing methods have utilized Euclidean distance to detect the fault in the process by comparing the similarity between each other. However, because there may be different data density in complex multi modal processes, it is difficult to obtain complete information by using Euclidean distance. This paper presents a new weighted distance based strategy to select neighbors. Firstly, the distance is weighted reasonably, which is then utilized to re select the neighbors of each sample point. This can effectively avoid the incompleteness of data information. A comparison simulation with the method based on traditional Euclidean distance shows that the proposed method based weighted Euclidean distance can get better results of the monitoring. Finally, this paper combines the proposed with the local outlier factor (LOF) to monitor TE process, whose results illustrates the superiority of this new selecting neighbor method.