<![CDATA[The limited adaptivity of Quality of Service (QoS) systems on MikroTik RouterOS, which still rely on static configurations and manual intervention, demands a network management mechanism capable of automatically adapting to traffic fluctuations. This study designs a self-healing system using Proximal Policy Optimization (PPO) integrated directly into MikroTik RouterOS via API within a closed-loop control architecture to realize adaptive QoS management. The agent is trained using a 12 feature state space that combines QoS parameters (latency, jitter, packet loss) and queue tree statistics (queued bytes, queued packets, dropped) for each traffic type (bulk, game, streaming), as well as a multidiscrete action space to adapt the queue type, limit-at, and max-limit parameters. The reward function is composed of anomaly penalties and a health bonus with clipping values in the range of [-5.0, 2.0] for training stability. Evaluation is conducted under three load scenarios (normal, static QoS, adaptive QoS) using a combination of Python generator scripts for streaming and game traffic, as well as real downloads via IDM for bulk traffic, with primary performance metrics including response time, recovery time, and recovery quality. Test results show that the self-healing system is capable of adjusting QoS configurations with an average response time of 22.11 ms, recovering network conditions with an average recovery time of 5.65 seconds, and achieving a recovery quality of 100%. These findings indicate that integrating PPO via the RouterOS API can realize an adaptive QoS self-healing system capable of responding to disruptions and restoring network conditions automatically on MikroTik infrastructure.]]>
