Abstract
Complex slow feature analysis is a feature extraction technique that extracts slow oscillating patterns from the measured data. The measurement noise is usually assumed to follow a Gaussian distribution to obtain a closed-form solution. However, industrial process data is often characterized by measurement issues such as outliers, including asymmetric measurement noise. Such issues reduce the performance of the extracted features if not accounted for explicitly. Therefore, this article proposes a novel robust complex slow feature model to tackle the mentioned issues. In particular, this work considers a Skewed t-distribution for the measurement noise of the complex slow feature model. The parameters of the Skewed t-distribution, especially the degree of freedom and the shape parameters, account for the outliers and the asymmetric nature of the measurement noise. The parameters of the proposed model are jointly estimated using the expectation-maximization algorithm. The efficiency of the approach is demonstrated using simulated and industrial data.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have