Autoencoder-K-Means Algorithm for Efficient Anomaly Detection to Improve Space Operations

As we venture further into space, the need for Artificial Intelligence (AI) based anomaly detection algorithms for space operations significantly escalates. These algorithms are essential for ground-based operators to issue accurate corrective commands that prevent potential disasters and spacecraft failure and reduce operation costs. To address this issue, this paper explores integrating two Deep Unsupervised Learning methods such, as Autoencoder (AE) and K-means. This approach integrates the reconstruction and clustering methods to detect anomalies. The AE method learns deep temporal dynamics by reducing the dimensions and denoising the data. K-means then clusters this compressed output from the AE. Performance of AE-K-means measures by calculating precision, recall, and Fl-score parameters. Precision is the percentage of detected anomalies that are true positives (TP); recall is the percentage of actual anomalies correctly detected; and the Fl-score is used to evaluate the algorithm's performance by applying precision and recall to a single value. The datasets used for this research comprise telemetry data expertly labeled and sourced from the Soil Moisture Active Passive (SMAP) satellite. Our results show that AE-K-means improves anomaly detection in complex multivariate time series datasets, like satellite telemetry, and it can provide an enhanced and smart solution to the challenges of detecting anomalies in space operations and systems.