CN-122027918-A - Optical network routing and spectrum allocation algorithm based on GCN-LSTM

CN122027918ACN 122027918 ACN122027918 ACN 122027918ACN-122027918-A

Abstract

The invention discloses an optical network routing and spectrum allocation algorithm based on GCN-LSTM, which comprises the steps of S1, defining an optical network environment and initializing the algorithm, constructing a basic framework of a flow prediction GCN-LSTM model, S2, predicting network flow data according to historical flow data in a network by combining a machine learning method, S3, determining an optimal modulation format in advance based on a predicted flow result, and S4, performing efficient RSA decision according to the determined modulation format. The optical network routing and spectrum allocation algorithm based on the GCN-LSTM disclosed by the invention optimizes and improves the aspects of flow data prediction, flow dynamic threshold setting, modulation format determination and the like, and can solve the problems of lower spectrum resource utilization rate and the like in the traditional heuristic algorithm.

Inventors

ZHANG QUN
CHEN XIUQIN
MA JINGXIN
GUO XUENING
WANG ZHENYU

Assignees

浙江大学绍兴研究院

Dates

Publication Date: 20260512
Application Date: 20251230

Claims (5)

1. An optical network routing and spectrum allocation algorithm based on GCN-LSTM, characterized by comprising the steps of: Step S1, defining an optical network environment and initializing an algorithm, and constructing a basic framework of a flow prediction GCN-LSTM model; step S2, according to historical flow data in the network, predicting the flow data of the network by combining a machine learning method; step S3, an optimal modulation format is predetermined based on a predicted flow result; And S4, performing efficient RSA decision according to the determined modulation format.
2. The GCN-LSTM based optical network routing and spectrum allocation algorithm according to claim 1, wherein step S1 is implemented as the following steps: Step S1.1, capturing spatial characteristics in a network topological structure by adopting a GCN network, and aggregating node information by utilizing side information to generate new node representation, wherein the propagation modes among layers in the network are as follows: ; Wherein the method comprises the steps of A is an input adjacent matrix, I is an identity matrix, Is that Is a diagonal matrix, each diagonal element is the number of the node connected with other nodes, and the calculation formula is that I.e., summing each row of adjacency matrix a, H is the feature matrix of all nodes in the GCN network, one node for each row, H is updated continuously as the number of network layers increases, A non-linear activation function is used, Trainable weights for the network; S1.2, inputting spatial characteristics output by a GCN (global control network) into an LSTM (local area network) and utilizing a gating mechanism thereof to learn time characteristics of flow data, wherein the gating mechanism of the LSTM comprises a forgetting gate, an input gate, an updating gate and an output gate, the forgetting gate has the functions of receiving state information h (t-1) transmitted by a last neuron and information x (t) newly transmitted, all the information is processed through the sigmoid function to obtain f (t) transmitted cell states c (t), f (t) is between 0 and 1, more forgetting is represented by more approaching 0, more remembering is represented by more approaching 1, the input gate has the functions of respectively using the sigmoid and the tanh functions to determine data needing to be updated, the updating gate has the functions of multiplying the data needing to be forgotten and the data needing to be updated into the cell states c (t) through a matrix, and updating the cell states by using the tanh function to update the cell states c (t) and outputting the new cell states to the LST; And S1.3, defining a mapping rule of a flow threshold and a modulation format, presetting a plurality of flow threshold intervals, defining candidate modulation formats corresponding to each interval, and defining constraint conditions to be considered by the mapping rule, including transmission distance, signal quality requirements and spectrum efficiency, so as to provide rule basis for the selection of the subsequent modulation formats, thereby completing the basic framework of the GCN-LSTM model.
3. The GCN-LSTM based optical network routing and spectrum allocation algorithm according to claim 2, wherein for step S2: And (3) inputting connection relation and historical flow data of nodes in the network by adopting a GCN-LSTM model, extracting time and space characteristics of the flow data in the collaborative capture network through double characteristics, and predicting to obtain network flow data.
4. A GCN-LSTM based optical network routing and spectrum allocation algorithm according to claim 3, characterized by for step S3: setting a plurality of flow threshold intervals according to the predicted network flow data, and matching the flow threshold interval and the modulation format mapping rule defined in the step S1.3, wherein when the predicted flow is in a lower interval, a high-order modulation format is selected to improve the spectrum efficiency; Combining constraint conditions of transmission distance, signal quality requirement and spectrum efficiency, determining an optimal scheme from the matched candidate modulation formats; calculating the number of spectral slots required for establishing an optical connection according to the selected modulation format; the routing uses a K-shortest path algorithm to generate a set of candidate paths, and spectrum allocation is performed on the candidate paths.
5. The GCN-LSTM based optical network routing and spectrum allocation algorithm of claim 4, wherein for step S4: training a GCN-LSTM model using the historical traffic data and the network topology data; Aiming at each new light path request, calling a trained GCN-LSTM model to obtain flow prediction data, performing spectrum allocation on a candidate path set according to the modulation format and the spectrum slot number requirement determined in the step S3, adopting a first fitting spectrum allocation method, sequentially searching from the lowest index of spectrum resources, finding a first continuous idle spectrum block meeting the constraints of continuity, consistency and non-overlapping, and allocating the first continuous idle spectrum block to the current service request; And in the training and decision-making process, recording the prediction result of each node, the occupancy rate of the frequency spectrum channel and the selection duty ratio of different modulation formats, and reducing the occupancy rate of the frequency spectrum channel in the optical network and continuously improving the resource utilization rate by dynamically adjusting the modulation format selection.

Description

Optical network routing and spectrum allocation algorithm based on GCN-LSTM Technical Field The invention belongs to the technical field of optical network resource allocation, and particularly relates to an optical network routing and spectrum allocation (Routing and Spectrum Allocation, RSA) algorithm based on GCN-LSTM (Graph Convolutional Network-Long Short-Term Memory). Background With the rapid development of technologies such as mobile internet, cloud computing, internet of things and the like, the optical network of the conventional technology has been difficult to meet the demands. At this time, the elastic optical network is a new optical network technology, which improves the bandwidth utilization rate and flexibility of the optical network by flexible spectrum allocation, variable modulation format and other technical means, while RSA is one of the key problems in the elastic optical network and is also one of the key technologies for optical network resource scheduling. RSA is largely divided into two parts, routing and spectrum allocation. The paper "Online routing and spectrum allocation in elastic optical networks based on dueling Deep Q-network" indicates that the routing refers to determining a feasible transmission path from a source node to a destination node on the premise of meeting service requirements, and the spectrum allocation refers to allocating spectrum resources meeting requirements for the service according to specific requirements of the service on the basis of the selected path. I.e. an optical connection is established between the source node and the destination node, and an appropriate number of contiguous, consecutive, non-overlapping spectral Slots (FSs) are allocated. Wherein a continuity (continuity) indicates a continuity constraint that the allocation of a traffic demand frequency band must consist of a continuity FSs, a continuity (continuity) indicates a continuity constraint that the allocation of a continuity FSs with the same position index on each link of the traffic demand path must be made, and a non-overlapping indicates a non-overlapping constraint that an occupied FS cannot be allocated to another traffic demand. Currently, a widely applied traditional RSA method, such as a Shortest path and first fitting spectrum allocation combined method (short PATH WITH FIRST-Fit), a Mixed Integer Linear Programming (MILP) method, and the like have corresponding disadvantages. The paper "Efficient statistical QoT-aware resource allocation in EONs over the C+L-band：a multi-period and low-margin perspective" indicates that although the method based on the shortest path and the first fitting spectrum allocation has the advantages of simple implementation, high stability and the like, the spectrum allocation strategy of the method often causes low bandwidth resource utilization rate and has the problems of resource waste and the like, and the mixed integer linear programming method proposed in the paper "Nonlinear Impairment-Aware RMSA Under the Sliding Scheduled Traffic Model for EONs Based on Deep Reinforcement Learning" can optimize resource allocation and improve the utilization rate in theory. However, the method has higher computational complexity and poorer real-time performance, and is difficult to effectively adapt to dynamically-changed network environments and real-time service requests. The main disadvantage of the prior art is that it is difficult to dynamically and cooperatively process network traffic changes, and they are often based on fixed rules or only optimize a single link, so that the trend of traffic changes in time and space cannot be accurately predicted, and the prediction result cannot be combined with modulation format selection, which becomes a key bottleneck for restricting the improvement of spectrum efficiency. The invention provides an optical network routing and spectrum allocation algorithm based on GCN-LSTM aiming at the core problem. Accordingly, the above problems are further improved. Disclosure of Invention The invention mainly aims to provide an optical network routing and spectrum allocation algorithm based on GCN-LSTM, which optimizes and improves the aspects of flow data prediction, flow dynamic threshold setting, modulation format determination and the like, and can solve the problems of low spectrum resource utilization rate and the like in the traditional heuristic algorithm. To achieve the above object, the present invention provides an optical network routing and spectrum allocation algorithm based on GCN-LSTM, comprising the steps of: Step S1, defining an optical network environment and initializing an algorithm, and constructing a basic framework of a flow prediction GCN-LSTM model; step S2, according to historical flow data in the network, predicting the flow data of the network by combining a machine learning method; step S3, an optimal modulation format is predetermined based on a predicted flow result; And S4, perform