AI-Powered Resilience: A Dual-Approach for Outage Management in Dense Cellular Networks

Abstract

As 5G evolves to 6G, network management faces growing challenges with increasing base station density, leading to more frequent outages. To address this, we introduce a robust, automated two-tier framework for outage management. The first tier involves an artificial intelligence-based outage detection scheme using an enhanced XGBoost model (Impv-XGBoost), which incorporates autoencoder outputs for hyperparameter tuning. The analysis shows Impv-XGBoost’s superior performance in high shadowing conditions and with sparse data, outperforming existing methods. The second tier adopts an actor–critic reinforcement learning strategy for outage compensation by adjusting the tilt of the neighboring base station and power. To prevent service declines to connected user equipment, our compensation scheme accounts for both outage-affected users and those connected to compensating base stations. We design a reward scheme that combines Jain’s fairness index and the geometric mean of the reference signal received power to ensure fairness and enhance convergence. Performance evaluations for single and multiple base station failures show coverage improvements for outage-affected users without compromising the coverage of the users in compensating base stations.