A Novel Deep Reinforcement Learning Framework for Self-Healing Telecommunications Networks with Application to Nigerian Operator Outage Management

Self-healing network capability enables telecommunications systems to automatically detect, diagnose, and compensate for component failures without manual intervention, reducing mean time to repair and maintaining service continuity for subscribers. Deep reinforcement learning provides a framework for learning optimal healing action policies from operational experience, potentially surpassing rule-based approaches in complex multi-failure scenarios. This doctoral study develops a novel deep reinforcement learning framework for self-healing telecommunications networks, with experimental validation and customization targeting the outage management challenges of Nigerian mobile network operations. The framework formulates the self-healing problem as a partially observable Markov decision process, with a state space incorporating real-time performance measurements from neighboring cells, action space defining compensatory parameter adjustments available to the network, and a reward function balancing subscriber quality of service restoration against unnecessary parameter disruption. A novel actor-critic architecture incorporating a graph attention network for exploiting network topology structure in the state representation is proposed, representing an original methodological contribution to self-healing network literature. Training uses operational outage scenarios derived from twelve months of alarm and performance data from a major Nigerian operator's network, enabling the policy to address failure types and network topology characteristics representative of Nigerian deployments. Simulation evaluation demonstrates that the proposed framework reduces coverage hole duration following base station failures by 58% compared to rule-based compensating SON algorithms, with 23% lower neighbor cell interference generation. Keywords: deep reinforcement learning, self-healing network, outage management, SON, Nigeria.