Abstract: As a usual status monitoring method, four subspace method is only applicable under the condition that process data follows Gaussian process. However, most of industrial data are non-Gaussian, which makes the application of the four subspace method rather limited. This paper uses a kernel density estimation method to improve the traditional four subspace method, and designs a four subspace status monitoring method based on the kernel density estimation, which is suitable for general distributions. Finally, using the real data of high temperature superheater in some electric power plant, the empirical results show that the improved four subspace method is more universal, and it can significantly improve the status monitoring effect.

Key words: four subspace method, kernel density estimation, status monitoring, non-Gaussian distribution