Adaptive Anomaly Detection Strategy Based on Reinforcement Learning

Abstract

In complex and changeable cloud environment, the monitoring and anomaly detection of cloud platform become very important. Good anomaly detection can help cloud platform managers to make quick adjustments to ensure a good user experience. Although many anomaly detection models have been put forward by researchers in recent years, the application of these anomaly detection models to a given service still faces the challenge of parameter adjustment, which is time-consuming and exhausting, and still fails. In order to solve the problem of parameter adjusting, in this paper, an adaptive anomaly detection framework is proposed, the process of parameter adjustment is transformed into a general Markov decision process by means of reinforcement learning, which realized the automation of parameter adjustment, reducing the workload of operator and the effective detection rate of the anomaly detection model is improved, we compared it on three typical KPI (Key Performance Indicator) curves with artificial adjustment mode and other optimization strategies, in the end, we verified the effectiveness of the strategy used in this paper.