An Optimization Approach to DWDM Network Reconfiguration through Reinforcement Learning

Optical networks must support dynamic reconfiguration to fulfill the requirements of the upcoming generationof data-hungry applications. Reconfiguration can lead to increased network utilization, higher bit rates, increased capacity, and lower QoS violations. With the objective of maximization of utilization under multiple constraints of latency, jitter and transmission quality, this paper proposes a reinforcement-learning-based reconfiguration for dense wavelength division multiplexing (DWDM) Networks. The reconfiguration is addressed both at the local switch and global network level for higher performance gains. A mixed strategy with strategy selection based on reinforcement learning is proposed in this study, as opposed to earlier works that used a single reconfiguration strategy such as wavelength selection, transceiver parameter tuning, route path planning, bandwidth adjustment, etc. A simulation of the proposed solution is performed by integrating MatLab with the OptiSystem Simulator. As compared to existing work, the proposed solution increased network utilization by 4.10% and reduced QoS violations by 2.2 times.