Detecting outliers for complex nonlinear systems with dynamic ensemble learning

Process data has been used in most industrial systems to facilitate process control and process monitoring. Even if outliers have been proved to have negative influence on those data-driven techniques, dedicated detection methods are still rare or at a junior phase. Furthermore, due to the fact that most industrial systems are complex and nonlinear, many outlier detection methods developed in the field of data mining are inefficient or cannot be applied directly. In this paper thereby, we propose an outlier detection method dedicated to complex and nonlinear industrial systems. This method is on the basis of dynamic ensemble learning. It is observed that ensemble learning has made great achievement recently, and dynamic ensemble learning usually outperforms other ensemble techniques. Experimental results prove that our dynamic ensemble outlier detection method has better performance for complex nonlinear industrial systems.