Online monitoring for high-dimensional data streams based on spatial signs

Statistical Process Control (SPC) dynamically monitors production processes through control charts to ensure that the process remain in a statistically controllable state. As high-dimensional data streams increasingly permeate into various fields, their online monitoring technology has emerged as a core challenge for enabling efficient complex system operations. When facing the complexity, non-normal distribution data, and curse of dimensionality of high- dimensional data, traditional monitoring methods are severely limited in performance, leading to unreliable monitoring results. Aiming to tackle these specific difficulties, this paper proposes a nonparametric exponentially weighted moving average monitoring method for high-dimensional processes based on spatial signs. This method features robustness against outliers and noise, as well as good performance under the assumption of non-normal distribution, effectively solving issues in traditional methods such as the difficulty in estimating the covariance matrix and unknown data distribution. The effectiveness and applicability of the proposed method are demonstrated by numerical simulation results.