Deep Reinforcement Learning Based Command Control System for Automating Fault Diagnosis

Due to the recent growth of network services of telecommunication carriers, their communication networks have become increasingly large and complex, and their network operations have also become complicated. For this problem, fault diagnosis and anomaly detection using log data of network devices (e.g., router syslog) have been extensively studied. However, there has been little research on automating the fault diagnosis of failures that are reported by users and whose causes do not appear in logs. In this paper, we propose a deep reinforcement learning (DRL)-based command control system for automating fault diagnosis of communication services. The proposed system first runs a sequence of device commands (e.g., show interfaces) for fault diagnosis on network devices, and estimates the fault-type based on the output of the commands using a supervised classifier. The sequence of device commands executed on network devices is optimized by DRL to identify the fault type with the least number of device commands. We evaluate the proposed system using real data obtained from a commercial service network and demonstrate its accuracy and effectiveness.