Network Maintenance Planning Via Multi-Agent Reinforcement Learning

Within this work, the challenge of developing maintenance planning solutions for networked assets is considered. This is challenging due to the very nature of these systems which are often heterogeneous, distributed and have complex co-dependencies between the constituent components for effective operation. We develop a Multi-Agent Reinforcement Learning (MARL) solution for this domain and apply it to a simulated Radio Access Network (RAN) comprising of nine Base Stations (BS). Through empirical evaluation we show that our model outperforms fixed corrective and preventive maintenance policies in terms of network availability whilst generally utilizing less than or equal amounts of maintenance resource.