The application discloses a
physical layer perception dynamic resource allocation method and
system for an elastic optical network, and solves the technical problems of existing schemes, such as conservative modulation selection, lack of
physical layer perception, poor stability of
reinforcement learning, low spectrum
utilization rate and high service blocking rate. The method receives a dynamic service request, obtains a network state, generates a candidate path, extracts service, spectrum,
physical layer and transmission feasibility features to construct a
state vector, outputs a path-modulation joint decision through a
reinforcement learning model, calculates a spectrum slot according to a modulation format, searches for a spectrum block meeting a constraint, performs QoT
transmission quality verification, generates a reward according to a service establishment result, and iteratively updates
model parameters through a reward centralization mechanism. The application improves the learning stability and decision accuracy of the model, realizes dynamic and flexible selection of modulation formats, fully excavates the physical layer transmission margin, improves the spectrum
utilization rate and reduces the service blocking rate, and can be adapted to multiple scenes such as backbone networks,
metropolitan area networks and
data center interconnections.