Outlier detection based on novel Robust Gaussian process: Applied to complex nonlinear system

As industries increasingly digitize, data-based technologies are becoming more widespread, placing higher demands on data quality. However, industrial processes are complex and subject to various interferences, resulting in detected data that inevitably contains outliers. To address this issue, outlier detection techniques are becoming increasingly indispensable. This paper presents a Gaussian process-based outlier detection method that is well-suited for complex and nonlinear industrial systems. This method combines a newly proposed robust Gaussian process with a specially designed outlier determination criterion. The approach allows for faster model training and more accurate outlier detection without the need for manual decision thresholds. To confirm the effectiveness of the proposed method, we compared it with various traditional outlier detection methods as well as mainstream robust Gaussian process-based outlier detection methods in the numerical simulation stage. The superior performance of the proposed method was confirmed in both artificial and real datasets.