Transformer-Assisted Service-Attribute-Aware Lightpath Provisioning in Multi-Granularity Optical Networks

Multi-granularity optical networks (MGONs) provide flexible switching capabilities at the waveband, wavelength, and sub-wavelength levels, which are promising for accommodating heterogeneous service requests with different bandwidth, delay, and quality requirements. However, existing lightpath provisioning strategies usually rely on fixed routing or cost-oriented rules, making it difficult to match dynamic network resources with service-specific requirements. To address this problem, this paper proposes a service-aware framework for joint routing, modulation format, granularity, and wavelength assignment (RMGWA) in MGONs, which integrates a Transformer-assisted service-attribute-aware (TSAA) strategy with a waveplane-based RMGWA algorithm. In the proposed strategy, the Transformer module takes service attributes and present network states as inputs and generates service-specific link weights to guide candidate route selection. Then, the proposed waveplane-based RMGWA algorithm is applied to determine feasible resource assignments. Also, a mixed-integer linear programming (MILP) model is formulated as a benchmark for small-scale scenarios. Simulation results on n6s9 and USNET show that the proposed framework improves service satisfaction and spectrum utilization while maintaining low establishment cost. Moreover, it is observed that the proposed TSAA does not increase average completion time of services; instead, it supports faster service-aware routing decisions. These results demonstrate the effectiveness of integrating TSAA with the waveplane-based RMGWA algorithm for adaptive lightpath provisioning in dynamic MGON environments.