Saliency Time-Series Anomaly Detection Service for AIOps

Artificial Intelligence for IT Operations (AIOps) platform needs to monitor the status of massive microservices to alert potential failures. In China Mobile Cloud, we develop a time-series anomaly detection service consisting of architecture and algorithm, that captures the anomaly metrics in time-series effectively. The architecture comprises four modules that work together: data collection, data storage, anomaly monitoring, and experimental platform. For enhancing the efficiency and accuracy of anomaly detection, a novel Saliency Time-Series Anomaly Detection Algorithm with minLSTM (STAD-minLSTM) is proposed to solve the challenges of high real-time, lack of labels, inefficiency, and poor accuracy in AIOps. A massive of experimental results on two real-world datasets demonstrate that our approach has substantially improved, which outperforms several state-of-the-art baselines.