The spectrum distribution method, device, storage medium and equipment of the elastic light network

Abstract

The present application discloses a spectrum allocation method, device, storage medium and device for an elastic optical network. The method includes: when a target traffic request is received, acquiring the target traffic request and the current environmental state of the elastic optical network of the partition to be adjusted and the trained target spectrum allocation action network; convert each current candidate path of the current environment state into the current link topology information; input the current link topology information into the target spectrum allocation action network, through the first graph convolutional neural network. Feature extraction, and then perform feature aggregation through the first recurrent neural network to obtain the path-level features of each current candidate path, and then perform probability conversion through the first fully connected layer and the flexible maximum transfer function layer to obtain the corresponding current candidate path spectrum allocation probability, and output the target spectrum allocation action according to the spectrum allocation probability. The blocking probability of the elastic optical network during and after spectrum allocation can be reduced.

Description

technical field [0001] The present application relates to the technical field of spectrum partitioning of elastic optical networks, and in particular, to a spectrum allocation method, apparatus, storage medium and device for elastic optical networks. Background technique [0002] Elastic Optical Network (EON) has attracted extensive attention due to its advantages of high throughput and transmission flexibility, and has become the most promising next-generation backbone network architecture. However, routing and spectrum allocation (RSA) is a key issue for elastic optical networks. Finding the optimal route and spectrum for a traffic request is NP-hard (one of the most important complexity classes in computational complexity theory), which means that it is extremely difficult to obtain optimal solutions suitable for large network environments. [0003] To overcome the above limitations, deep reinforcement learning (DRL) has been extensively studied in recent years. DRL is ...

AI Technical Summary

Problems solved by technology

In the DRL-based RSA framework, input state representation and feature extraction are essential. However, in the prior art network feature extraction, feature extraction is only performed on nodes, resulting in unsatisfactory feature extraction and utilization effects. , so that the spectrum allocation result of the elastic optical network is less effective in reducing the blocking probability

Images