Wavelength adaptive configuration method of CWDM / DWDM hybrid wavelength division system

Abstract

The invention relates to the technical field of optical communication, in particular to a wavelength adaptive configuration method of a CWDM / DWDM hybrid wavelength division system, which comprises the following steps of: acquiring physical layer parameters of an optical fiber link in real time; outputting the OSNR margin and the nonlinear noise prediction value of each wavelength position; dynamically adjusting the reservation proportion of the channel; collaboratively optimizing bandwidth, time delay, damage and cost constraints in path calculation through a reinforcement learning algorithm; splitting the CWDM channel into virtual sub-channels, or combining the virtual sub-channels to generate a super-channel; compared with the prior art which depends on an empirical formula to estimate transmission damage and has the defects of extensive modeling and cross-band crosstalk missing detection, the method provided by the invention has the advantages that a power spectrum, a temperature gradient and nonlinear interaction are analyzed in real time through an LSTM-GNN combined architecture, and a mixed channel damage map covering a full frequency band is constructed; the method has the advantages that cross damage of a CWDM wide channel and a DWDM dense area is accurately quantified, and high-crosstalk wavelength combination is dynamically avoided.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of optical communication technology, in particular to a wavelength adaptive configuration method of a CWDM / DWDM hybrid wavelength division system. BACKGROUND

[0002] As a new generation of optical network core technology, the CWDM / DWDM hybrid wavelength division multiplexing system aims to balance the economy of the metropolitan access network and the large capacity demand of the backbone network by co-transmitting low-cost wide channels of coarse wavelength division multiplexing (CWDM) and high-spectral efficiency subchannels of dense wavelength division multiplexing (DWDM) in a single optical fiber. The system needs to dynamically adapt to business flow fluctuations and physical layer damage changes, realize intelligent scheduling of wavelength resources and optimization of transmission performance, and become the key carrier for improving spectral efficiency and reducing latency in the field of optical communication.

[0003] The traditional implementation scheme adopts a hierarchical management architecture, and considers CWDM and DWDM as independent resource pools for static allocation, which leads to serious fragmentation of the spectrum and lack of cross-system cooperation. The path calculation relies on a fixed rule library or heuristic algorithm, which is difficult to accurately quantify the nonlinear crosstalk (such as stimulated Raman scattering tilt and four-wave mixing effect) between CWDM wide channels and DWDM dense subchannels in the hybrid system, resulting in high bit error rate and bandwidth waste. In the operation process, manual identification of low-load periods is required to perform fragment consolidation, which has inherent defects such as high risk of business interruption and slow response speed. These shortcomings seriously restrict the practicality and potential for large-scale deployment of the hybrid wavelength division system in the super-100G business scenario.

[0004] The present application proposes an adaptive configuration mechanism based on deep reinforcement learning and spectral virtualization, which constructs a physical layer damage knowledge graph through an LSTM-GNN joint model, and generates a cross-band crosstalk risk heat map in real time. Combined with multi-layer topology modeling technology, the CWDM fixed grid is split into programmable virtual subchannels, which are dynamically stitched with idle DWDM spectrum to form super channels. A threshold-triggered fully automatic fragment migration algorithm is introduced to complete resource reorganization within a preset time window. This breaks through the closed loop from physical perception to decision execution in the hybrid wavelength division system, achieving a generational breakthrough in transmission performance while maintaining a low-cost architecture. SUMMARY

[0005] In order to overcome the problems proposed in the above background art, the present application proposes a wavelength adaptive configuration method of a CWDM / DWDM hybrid wavelength division system.

[0006] The technical solution of the present application is: a wavelength adaptive configuration method of a CWDM / DWDM hybrid wavelength division system, comprising the following steps: S11: Real-time acquisition of physical layer parameters of the optical fiber link, including wavelength power spectrum, optical fiber type identification, temperature and modulation format; S12: Generate hybrid channel damage maps based on machine learning models, and output the OSNR margin and nonlinear noise prediction values ​​at each wavelength position; S13: Execute time-series prediction based on historical business data and dynamically adjust the reserved ratio of CWDM wide channels and DWDM sub-channels; S14: Construct a multi-layer graph model containing CWDM, DWDM and hybrid layers, and use reinforcement learning algorithms to collaboratively optimize bandwidth, latency, impairment and cost constraints in path computation. S15: In the spectrum virtualization module, split the CWDM channel into virtual sub-channels, or merge the DWDM sub-channels and CWDM spectrum gaps to generate a super channel; S16: Initiate fragmentation during a preset low-load period to migrate discrete DWDM services to the CWDM channel and reassemble continuous spectrum blocks.

[0007] Preferably, when generating hybrid channel impairment maps based on a machine learning model and outputting the OSNR margin and nonlinear noise prediction values ​​at each wavelength position, the machine learning model adopts a joint architecture of long short-term memory network and graph neural network, wherein: A11: The input time-varying parameter sequence of the long short-term memory network module includes cross-segment power fluctuations and temperature gradients sampled every 10 seconds; A12: The graph neural network module models the nonlinear interactions between adjacent wavelengths in the fiber optic topology, and the output damage map includes: (1) Stimulated Raman scattering tilt level of CWDM channel; (2) Risk coefficient of four-wave mixing in dense DWDM areas; (3) Frequency domain coordinate identification of cross-band crosstalk sensitive areas.

[0008] As a preferred option, the spectrum virtualization module specifically includes the following functions: A21: Divide a single CWDM channel into N virtual sub-channels at 12.5 GHz intervals, where N is greater than or equal to 2, and configure the modulation format independently for each sub-channel; A22: Identify the idle spectrum in the gaps between CWDM grids and stitch it with the adjacent DWDM sub-channels in the frequency domain to form a super channel with a bandwidth greater than 100GHz; A23: In a multi-layer graph model, connection edges are established for virtualized resources, and their weights are calculated jointly by the cumulative dispersion of the substrate fiber and the nonlinear Gran coefficient.

[0009] As a preferred embodiment, when constructing a multi-layer graph model comprising CWDM, DWDM, and hybrid layers, and using a reinforcement learning algorithm to collaboratively optimize bandwidth, latency, impairment, and cost constraints in path computation, the specific configuration of the reinforcement learning algorithm is as follows: A31: The state space is defined as a triple consisting of a hybrid channel impairment map, a real-time spectrum occupancy matrix, and service QoS requirements. A32: The action space includes: resource allocation strategies for selecting pure CWDM path, pure DWDM path, or hybrid superchannel path; A33: The expression for calculating the reward function R is: ; in, , , and These are the trainable weight parameters.

[0010] Preferably, when defragmentation is initiated during a preset low-load period, the activation conditions for defragmentation are as follows: S21: Calculate the DWDM spectrum fragmentation rate of the entire network : ; S22: When Send alarm signals to the network management system in a timely manner; S23: After manual confirmation, perform the migration operation between 02:00 and 04:00 local time. This operation includes: (1) Aggregate DWDM services with a utilization rate of less than 50% into the same CWDM virtual sub-channel group; (2) The released DWDM spectrum is sorted according to wavelength continuity to construct Flexi-Grid blocks with a minimum width of 100 GHz.

[0011] As a preferred approach, when performing time-series forecasting based on historical business data and dynamically adjusting the reservation ratio of CWDM wide channels and DWDM sub-channels, the time-series forecasting adopts the Holt-Winters three-parameter exponential smoothing model, including: S31: The input parameter is the historical 72-hour business traffic matrix aggregated by hour; S32: Outputs the predicted load values ​​for each link over the next 4 hours and maps them to resource pre-adjustment commands: (1) If the predicted peak load is greater than 80% of the link capacity, reserve more than 2 adjacent DWDM sub-channels; (2) If the predicted valley load is less than 30% of the link capacity, switch the DWDM service to the CWDM channel and turn off the idle amplifier.

[0012] Preferably, when constructing a multi-layer graph model that includes CWDM layers, DWDM layers, and hybrid layers, the multi-layer graph model includes: A41: CWDM layer, nodes represent coarse wavelength division channels with 20nm spacing, and edge weights are determined by the configuration cost of the dispersion compensation module; A42: DWDM layer, where nodes represent dense wavelength division channels at the 12.5 GHz granularity, and edge weights are calculated by nonlinear Shannon limit capacity constraints; A43: Hybrid layer, containing virtual nodes and hyperedges. Virtual nodes correspond to sub-channels generated by spectrum virtualization. Hyperedges connect frequency domain stitching regions spanning CWDM and DWDM gratings, and their weights are the product of stitching loss and crosstalk suppression factor.

[0013] Preferably, when collaboratively optimizing bandwidth, latency, impairment, and cost constraints in path computation using reinforcement learning algorithms, cross-layer impairments are dynamically avoided in path computation, including: S41: When the damage spectrum indicates the CWDM wavelength With DWDM wavelength When the crosstalk risk value is greater than the threshold X: (1) Remove from multi-layer graph model and The connecting edges between them; (2) Regarding the use The CWDM channel automatically reduces the transmit power by 3dB; S42: Insert an optical isolator for high-risk links. The insertion position is determined by the node importance ranking of the GNN module.

[0014] Preferably, the hardware is controlled to perform the following during spectrum virtualization: S51: When splitting the CWDM channel, drive the tunable laser to perform sweep frequency modulation within the target sub-channel bandwidth; S52: When generating a super channel, configure the filter shape of the programmable wavelength selection switch to be a flat-top response curve with an edge roll-off coefficient of less than 0.5dB / GHz; S53: During the fragmentation and migration phase, synchronous switching technology is adopted to ensure that the service interruption time is less than 50ms.

[0015] Preferably, when fragmentation defragmentation is initiated during a preset low-load period to migrate discrete DWDM services to CWDM channels and reassemble continuous spectrum blocks, the specific migration operation of fragmentation defragmentation includes: S61: Identify source fragment service set Each of them Bandwidth usage ≤ 12.5GHz; S62: Select target CWDM channel C, whose idle spectrum capacity satisfies: ; S63: Establish a temporary optical path from S to C at the ROADM node and execute synchronously: (1) Activate the C-band tunable laser at the source end; (2) Configure a broadband coherent receiver at the destination; S64: After the migration is complete, release the discrete DWDM spectrum occupied by S and mark it as a continuous Flexi-Grid block.

[0016] The beneficial effects of this invention are: 1. Compared with existing technologies that rely on empirical formulas to estimate transmission damage, which have the disadvantages of coarse modeling and missed detection of cross-band crosstalk, this solution uses an LSTM-GNN joint architecture to analyze the power spectrum, temperature gradient and nonlinear interaction in real time to construct a hybrid channel damage map covering the entire frequency band; it has the advantages of accurately quantifying the cross-damage between CWDM wide channels and DWDM dense areas and dynamically avoiding high crosstalk wavelength combinations. 2. Compared with existing technologies that manage CWDM and DWDM as independent resource pools, which suffer from grid fragmentation and severe instability, this solution uses spectrum virtualization technology to decompose CWDM channels into programmable sub-channels and supports stitching them with idle DWDM spectrum to form super channels; it achieves cross-system resource scheduling without discrimination, and greatly improves the flexibility of large-granularity service carrying and spectrum utilization. 3. Compared with existing path calculation algorithms that rely on static weights, which have the disadvantages of a single optimization objective and slow convergence speed, this solution embeds a reinforcement learning strategy into a multi-layer graph model. It integrates damage constraints and business intentions through state space and designs a multi-factor reward function to dynamically balance efficiency and quality. It has the adaptive decision-making ability to generate the global optimal solution in milliseconds under complex constraints. 4. Compared with the disadvantages of manual defragmentation operations, which have the disadvantages of misjudging the timing and high risk of business interruption, this solution automatically triggers migration based on threshold alarms and time window mechanisms. It releases discrete spectrum and reassembles continuous blocks through business aggregation, forming an intelligent maintenance closed loop that is unattended during low-load periods, which significantly reduces the complexity of operation and maintenance and the rate of human error. 5. Compared with the traditional solution that uses a fixed power template, which leads to uncontrolled nonlinear effects and mutual interference between high-power CWDM channels and dense DWDM services, this solution dynamically cuts off high-risk links based on the damage map, adaptively adjusts the transmit power, and inserts isolators; achieving a generational improvement in signal transmission stability, especially ensuring Q-factor stability in long-distance hybrid networking scenarios. 6. Compared with the existing system's passive expansion strategy for sudden business, which has the disadvantages of high resource idle rate and huge energy consumption, this solution integrates time series prediction model and hardware linkage mechanism to pre-release redundant DWDM channels and intelligently shut down idle amplifiers; it constructs an "on-demand supply" resource supply mode, systematically reducing the unit bit transmission cost and carbon emissions. Attached Figure Description

[0017] Fig. 1The diagram shows a flowchart of the wavelength adaptive configuration method for the CWDM / DWDM hybrid wavelength division system of the present invention. Fig. 2 The diagram shows the architecture of the machine learning model used in the wavelength adaptive configuration method of the CWDM / DWDM hybrid wavelength division system of the present invention. Detailed Implementation

[0018] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0019] Please see Figs. 1-2 The present invention provides an embodiment of a wavelength adaptive configuration method for a CWDM / DWDM hybrid wavelength division system, comprising the following steps: S11: Real-time acquisition of physical layer parameters of fiber optic links, including wavelength power spectrum, fiber type identification, temperature and modulation format; S12: Generate hybrid channel damage maps based on machine learning models, and output the OSNR margin and nonlinear noise prediction values ​​at each wavelength position; S13: Execute time-series prediction based on historical business data and dynamically adjust the reserved ratio of CWDM wide channels and DWDM sub-channels; S14: Construct a multi-layer graph model containing CWDM, DWDM and hybrid layers, and use reinforcement learning algorithms to collaboratively optimize bandwidth, latency, impairment and cost constraints in path computation. S15: In the spectrum virtualization module, split the CWDM channel into virtual sub-channels, or merge the DWDM sub-channels and CWDM spectrum gaps to generate a super channel; S16: Initiate fragmentation during a preset low-load period to migrate discrete DWDM services to the CWDM channel and reassemble continuous spectrum blocks.

[0020] As described above, this invention generates a damage map by real-time acquisition of optical fiber physical layer parameters and inputting them into an ML model. It dynamically adjusts the resource reservation strategy by combining historical traffic prediction, optimizes path calculation by using a multi-layer graph model and reinforcement learning, implements spectrum virtualization to achieve CWDM / DWDM resource interoperability, and triggers a fragmentation defragmentation mechanism during low-load periods. This significantly improves the spectrum utilization efficiency (>25%), service deployment success rate (>95%), and damage avoidance accuracy (OSNR prediction error <0.5dB) of the hybrid wavelength division system.

[0021] Preferably, when generating hybrid channel impairment maps based on a machine learning model and outputting the OSNR margin and nonlinear noise prediction values ​​at each wavelength position, the machine learning model adopts a joint architecture of long short-term memory network and graph neural network, wherein: A11: The input time-varying parameter sequence of the long short-term memory network module includes cross-segment power fluctuations and temperature gradients sampled every 10 seconds; A12: The graph neural network module models the nonlinear interactions between adjacent wavelengths in the fiber optic topology, and the output damage map includes: (1) Stimulated Raman scattering tilt level of CWDM channel; (2) Risk coefficient of four-wave mixing in dense DWDM areas; (3) Frequency domain coordinate identification of cross-band crosstalk sensitive areas.

[0022] As described above, this invention models cross-layer impairment mechanisms using a joint LSTM-GNN architecture, captures the impact of power / temperature transient changes on CWDM wide channels through temporal features, and analyzes the nonlinear interference propagation path in DWDM dense areas based on graph structure. This enables the system to have the ability to quantitatively predict impairments unique to hybrid systems, such as SRS tilt (accuracy ±0.1dB / km) and FWM products (identification sensitivity -35dB), thereby reducing the bit error rate of services over 100G by 1-2 orders of magnitude.

[0023] As a preferred option, the spectrum virtualization module specifically includes the following functions: A21: Divide a single CWDM channel into N virtual sub-channels at 12.5 GHz intervals, where N is greater than or equal to 2, and configure the modulation format independently for each sub-channel; A22: Identify the idle spectrum in the gaps between CWDM grids and stitch it with the adjacent DWDM sub-channels in the frequency domain to form a super channel with a bandwidth greater than 100GHz; A23: In a multi-layer graph model, connection edges are established for virtualized resources, and their weights are calculated jointly by the cumulative dispersion of the substrate fiber and the nonlinear Gran coefficient.

[0024] As described above, this invention breaks the fixed grid limitation of CWDM through spectrum virtualization operations, divides a single CWDM channel into N independently modulated sub-channels (bandwidth utilization ↑ 40%), generates a cross-system super channel (supporting 400G+ services) through frequency domain stitching, constructs hybrid layer connection relationships in multi-layer graphs and introduces nonlinear weight constraints, eliminating the spectrum fragmentation problem caused by resource pool division in traditional solutions.

[0025] As a preferred embodiment, when constructing a multi-layer graph model comprising CWDM, DWDM, and hybrid layers, and using a reinforcement learning algorithm to collaboratively optimize bandwidth, latency, impairment, and cost constraints in path computation, the specific configuration of the reinforcement learning algorithm is as follows: A31: The state space is defined as a triple consisting of a hybrid channel impairment map, a real-time spectrum occupancy matrix, and service QoS requirements. A32: The action space includes: resource allocation strategies for selecting pure CWDM path, pure DWDM path, or hybrid superchannel path; A33: The expression for calculating the reward function R is: ; in, , , and These are the trainable weight parameters.

[0026] As described above, the reinforcement learning algorithm in this invention integrates physical layer constraints and business requirements through state space triples, covers all resource type selection strategies in the action space, and dynamically balances spectrum efficiency, transmission quality and cost based on trainable weights of the reward function (optimization speed increased by 5 times), achieving millisecond-level decision-making for the global optimal solution under complex constraints.

[0027] Preferably, when defragmentation is initiated during a preset low-load period, the activation conditions for defragmentation are as follows: S21: Calculate the DWDM spectrum fragmentation rate of the entire network : ; S22: When Send alarm signals to the network management system in a timely manner; S23: After manual confirmation, perform the migration operation between 02:00 and 04:00 local time. This operation includes: (1) Aggregate DWDM services with a utilization rate of less than 50% into the same CWDM virtual sub-channel group; (2) The released DWDM spectrum is sorted according to wavelength continuity to construct Flexi-Grid blocks with a minimum width of 100 GHz.

[0028] As described above, this invention uses an alarm mechanism based on a fragmentation rate threshold (η>30%) and a timed window operation to release discrete DWDM spectrum through service aggregation and construct continuous Flexi-Grid blocks, thereby increasing the available spectrum resources for large-granular services by more than 30% and reducing the frequency of ROADM node port switching (operation and maintenance costs ↓15%).

[0029] As a preferred approach, when performing time-series forecasting based on historical business data and dynamically adjusting the reservation ratio of CWDM wide channels and DWDM sub-channels, the time-series forecasting adopts the Holt-Winters three-parameter exponential smoothing model, including: S31: The input parameter is the historical 72-hour business traffic matrix aggregated by hour; S32: Outputs the predicted load values ​​for each link over the next 4 hours and maps them to resource pre-adjustment commands: (1) If the predicted peak load is greater than 80% of the link capacity, reserve more than 2 adjacent DWDM sub-channels; (2) If the predicted valley load is less than 30% of the link capacity, switch the DWDM service to the CWDM channel and turn off the idle amplifier.

[0030] As described above, this invention uses the Holt-Winters model to predict peak / valley values ​​of link load, drives pre-adjustment commands to dynamically control DWDM sub-channel reservation and amplifier start / stop, reduces reconfiguration latency for bursty services by 70%, and reduces system idle energy consumption by 40%.

[0031] Preferably, when constructing a multi-layer graph model that includes CWDM layers, DWDM layers, and hybrid layers, the multi-layer graph model includes: A41: CWDM layer, nodes represent coarse wavelength division channels with 20nm spacing, and edge weights are determined by the configuration cost of the dispersion compensation module; A42: DWDM layer, where nodes represent dense wavelength division channels at the 12.5 GHz granularity, and edge weights are calculated by nonlinear Shannon limit capacity constraints; A43: Hybrid layer, containing virtual nodes and hyperedges. Virtual nodes correspond to sub-channels generated by spectrum virtualization. Hyperedges connect frequency domain stitching regions spanning CWDM and DWDM gratings, and their weights are the product of stitching loss and crosstalk suppression factor.

[0032] As described above, in the multilayer graph model of this invention, the CWDM layer adopts DCM cost weight (compensation accuracy ↑ 50%), the DWDM layer is based on Shannon capacity constraint, and the hybrid layer provides a physically realizable topological basis for cross-system path computation through hyperedge quantization stitching loss (< -0.2dB) and crosstalk suppression factor (> 20dB).

[0033] Preferably, when collaboratively optimizing bandwidth, latency, impairment, and cost constraints in path computation using reinforcement learning algorithms, cross-layer impairments are dynamically avoided in path computation, including: S41: When the damage spectrum indicates the CWDM wavelength With DWDM wavelength When the crosstalk risk value is greater than the threshold X: (1) Remove from multi-layer graph model and The connecting edges between them; (2) Regarding the use The CWDM channel automatically reduces the transmit power by 3dB; S42: Insert an optical isolator for high-risk links. The insertion position is determined by the node importance ranking of the GNN module.

[0034] As described above, this invention improves the stability of long-distance transmission by dynamically cutting off links with high crosstalk risk (threshold X=0.15) and automatically adjusting power, combined with the GNN node importance deployment strategy of optical isolators, thereby controlling the Q factor degradation caused by cross-band interference within ±0.3dB.

[0035] Preferably, the hardware is controlled to perform the following during spectrum virtualization: S51: When splitting the CWDM channel, drive the tunable laser to perform sweep frequency modulation within the target sub-channel bandwidth; S52: When generating a super channel, configure the filter shape of the programmable wavelength selection switch to be a flat-top response curve with an edge roll-off coefficient of less than 0.5dB / GHz; S53: During the fragmentation and migration phase, synchronous switching technology is adopted to ensure that the service interruption time is less than 50ms.

[0036] As described above, the flat-top response curve (roll-off factor <0.5dB / GHz) in the tunable hardware operation of this invention ensures the integrity of the super channel signal, and the same-frequency resonance switching technology reduces the service migration interruption time to within 50ms, meeting the SLA requirements of financial-grade services.

[0037] Preferably, when fragmentation defragmentation is initiated during a preset low-load period to migrate discrete DWDM services to CWDM channels and reassemble continuous spectrum blocks, the specific migration operation of fragmentation defragmentation includes: S61: Identify source fragment service set Each of them Bandwidth usage ≤ 12.5GHz; S62: Select target CWDM channel C, whose idle spectrum capacity satisfies: ; S63: Establish a temporary optical path from S to C at the ROADM node and execute synchronously: (1) Activate the C-band tunable laser at the source end; (2) Configure a broadband coherent receiver at the destination; S64: After the migration is complete, release the discrete DWDM spectrum occupied by S and mark it as a continuous Flexi-Grid block.

[0038] As described above, the fragment migration process of the present invention prevents crosstalk by reserving a protection interval (>10GHz), resolves the mixed rate compatibility problem by configuring a broadband coherent receiver, and compresses the configuration time of subsequent 100G services to the second level by marking discrete spectrum as continuous blocks.

[0039] Example 1: Burst 400G Service Bearing in Metropolitan Area Core Network A smart city's core ring network deploys a hybrid CWDM (8 wavelengths @ 20nm spacing) and DWDM (C-band 96 wavelengths @ 50GHz) system. At 2:00 AM, the network monitoring platform detected a fragmentation rate of 32% (exceeding the 30% threshold) and automatically triggered an email alert. After a one-click confirmation by maintenance personnel, the system performed defragmentation. Spectrum sensing: Scanning revealed 10 10G DWDM services scattered across 12.5GHz sub-channels; Virtualization migration: Aggregate and migrate it to an idle 1550nm CWDM channel (split into 8 12.5GHz virtual sub-channels); Resource reorganization: The released DWDM spectrum forms a continuous 200 GHz Flexi-Grid block.

[0040] At 9:00 AM that day, the government cloud suddenly requested 400G of data transfer: Damage prediction: LSTM-GNN model analysis of real-time power spectrum identifies high SRS risk in the vicinity of 1570nm CWDM; Path calculation: In the multilayer diagram, the hybrid layer enables the 1570nm DWDM side free block and the 1530nm CWDM gap, and stitches them together to generate a super channel; Reinforcement learning decision-making: Avoid high-risk wavelengths and choose low-dispersion paths (reward function weighted delay factor γ=0.8). Coordinated control: The WSS is configured with a flat-top filter (roll-off factor 0.4dB / GHz), and the tunable laser performs sweep frequency modulation.

[0041] Services are established within 50ms, with a measured OSNR of 25dB (predicted value 24.8dB), and the utilization rate of spectrum resources released by fragmentation reaches 95%.

[0042] Example 2: Dynamic Damage-Resistant Reconstruction of Long-Distance Backbone Networks The 1500km inter-provincial backbone network adopts a hybrid architecture of CWDM (1610nm / 1590nm) + DWDM (L-band 80-wavelength). Heavy rainfall caused micro-bending of optical fibers, leading to regional optical power fluctuations. Real-time monitoring: Distributed sensors captured a sudden temperature rise of 8°C and a power fluctuation of ±2dB at 1610nm in segment 3; Damage profile update: The GNN module resolved the enhancement of the four-wave mixing products of 1590nm CWDM and L-band DWDM; Predictive scheduling: The Holt-Winters model predicts that the link load will drop by 40% and automatically shuts down redundant amplifiers; Cross-layer avoidance: Cut off the high-risk wavelength combination edge (crosstalk coefficient X=0.17), insert an optical isolator to attenuate by -35dB; Path Restructuring: Agent Selection in Reinforcement Learning The affected 100G services will be migrated from DWDM to 1610nm CWDM virtual subchannel group; The new video service uses 1590nm CWDM to split sub-channels (ensuring QoS weight β=0.9).

[0043] The system completes a lossless switch within 80ms, and the fragmentation module simultaneously integrates the released L-band spectrum into 3×150GHz continuous blocks, ensuring zero service interruption across the entire network during the rainstorm.

[0044] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.

Claims

1. A wavelength adaptive configuration method for a CWDM / DWDM hybrid wavelength division multiplexing system, characterized in that: Includes the following steps: S11: Real-time acquisition of physical layer parameters of fiber optic links, including wavelength power spectrum, fiber type identification, temperature and modulation format; S12: Generate hybrid channel damage maps based on machine learning models, and output the OSNR margin and nonlinear noise prediction values ​​at each wavelength position; S13: Execute time-series prediction based on historical business data and dynamically adjust the reserved ratio of CWDM wide channels and DWDM sub-channels; S14: Construct a multi-layer graph model containing CWDM, DWDM and hybrid layers, and use reinforcement learning algorithms to collaboratively optimize bandwidth, latency, impairment and cost constraints in path computation. S15: In the spectrum virtualization module, split the CWDM channel into virtual sub-channels, or merge the DWDM sub-channels and CWDM spectrum gaps to generate a super channel; S16: Initiate fragmentation during a preset low-load period to migrate discrete DWDM services to the CWDM channel and reassemble continuous spectrum blocks.

2. The wavelength adaptive configuration method for a CWDM / DWDM hybrid wavelength division system according to claim 1, characterized in that: When generating hybrid channel impairment maps based on a machine learning model and outputting the OSNR margin and nonlinear noise prediction values ​​at each wavelength, the machine learning model employs a joint architecture of a long short-term memory network and a graph neural network, wherein: A11: The input time-varying parameter sequence of the long short-term memory network module includes cross-segment power fluctuations and temperature gradients sampled every 10 seconds; A12: The graph neural network module models the nonlinear interactions between adjacent wavelengths in the fiber optic topology, and the output damage map includes: (1) Stimulated Raman scattering tilt level of CWDM channel; (2) Risk coefficient of four-wave mixing in dense DWDM areas; (3) Frequency domain coordinate identification of cross-band crosstalk sensitive areas.

3. The wavelength adaptive configuration method for a CWDM / DWDM hybrid wavelength division system according to claim 2, characterized in that: The specific functions of the spectrum virtualization module include: A21: Divide a single CWDM channel into N virtual sub-channels at 12.5 GHz intervals, where N is greater than or equal to 2, and configure the modulation format independently for each sub-channel; A22: Identify the idle spectrum in the gaps between CWDM grids and stitch it with the adjacent DWDM sub-channels in the frequency domain to form a super channel with a bandwidth greater than 100GHz; A23: In a multi-layer graph model, connection edges are established for virtualized resources, and their weights are calculated jointly by the cumulative dispersion of the substrate fiber and the nonlinear Gran coefficient.

4. The wavelength adaptive configuration method for a CWDM / DWDM hybrid wavelength division system according to claim 3, characterized in that: When constructing a multi-layer graph model containing CWDM, DWDM, and hybrid layers, and using reinforcement learning algorithms to collaboratively optimize bandwidth, latency, impairment, and cost constraints in path computation, the specific configuration of the reinforcement learning algorithm is as follows: A31: The state space is defined as a triple consisting of a hybrid channel impairment map, a real-time spectrum occupancy matrix, and service QoS requirements. A32: The action space includes: resource allocation strategies for selecting pure CWDM path, pure DWDM path, or hybrid superchannel path; A33: The expression for calculating the reward function R is: ； in, , , and These are the trainable weight parameters.

5. The wavelength adaptive configuration method for a CWDM / DWDM hybrid wavelength division system according to claim 4, characterized in that: When defragmentation is initiated during a preset low-load period, the activation conditions are as follows: S21: Calculate the DWDM spectrum fragmentation rate of the entire network : ; S22: When Send alarm signals to the network management system in a timely manner; S23: After manual confirmation, perform the migration operation between 02:00 and 04:00 local time. This operation includes: (1) Aggregate DWDM services with a utilization rate of less than 50% into the same CWDM virtual sub-channel group; (2) The released DWDM spectrum is sorted according to wavelength continuity to construct Flexi-Grid blocks with a minimum width of 100 GHz.

6. The wavelength adaptive configuration method for a CWDM / DWDM hybrid wavelength division system according to claim 5, characterized in that: When performing time-series forecasting based on historical business data and dynamically adjusting the reservation ratio of CWDM wide channels and DWDM sub-channels, the time-series forecasting adopts the Holt-Winters three-parameter exponential smoothing model, including: S31: The input parameter is the historical 72-hour business traffic matrix aggregated by hour; S32: Outputs the predicted load values ​​for each link over the next 4 hours and maps them to resource pre-adjustment commands: (1) If the predicted peak load is greater than 80% of the link capacity, reserve more than 2 adjacent DWDM sub-channels; (2) If the predicted valley load is less than 30% of the link capacity, switch the DWDM service to the CWDM channel and turn off the idle amplifier.

7. The wavelength adaptive configuration method for a CWDM / DWDM hybrid wavelength division system according to claim 6, characterized in that: When constructing a multi-layer graph model that includes CWDM layers, DWDM layers, and hybrid layers, the multi-layer graph model includes: A41: CWDM layer, nodes represent coarse wavelength division channels with 20nm spacing, and edge weights are determined by the configuration cost of the dispersion compensation module; A42: DWDM layer, where nodes represent dense wavelength division channels at the 12.5 GHz granularity, and edge weights are calculated by nonlinear Shannon limit capacity constraints; A43: Hybrid layer, containing virtual nodes and hyperedges. Virtual nodes correspond to sub-channels generated by spectrum virtualization. Hyperedges connect frequency domain stitching regions spanning CWDM and DWDM gratings, and their weights are the product of stitching loss and crosstalk suppression factor.

8. The wavelength adaptive configuration method for a CWDM / DWDM hybrid wavelength division system according to claim 7, characterized in that: When using reinforcement learning algorithms to collaboratively optimize bandwidth, latency, impairment, and cost constraints in path computation, dynamic avoidance of cross-layer impairments in path computation includes: S41: When the damage spectrum indicates the CWDM wavelength With DWDM wavelength When the crosstalk risk value is greater than the threshold X: (1) Remove from multi-layer graph model and The connecting edges between them; (2) Regarding the use The CWDM channel automatically reduces the transmit power by 3dB; S42: Insert an optical isolator for high-risk links. The insertion position is determined by the node importance ranking of the GNN module.

9. The wavelength adaptive configuration method for a CWDM / DWDM hybrid wavelength division system according to claim 8, characterized in that: Controlling hardware execution during spectrum virtualization: S51: When splitting the CWDM channel, drive the tunable laser to perform sweep frequency modulation within the target sub-channel bandwidth; S52: When generating a super channel, configure the filter shape of the programmable wavelength selection switch to be a flat-top response curve with an edge roll-off coefficient of less than 0.5dB / GHz; S53: During the fragmentation and migration phase, synchronous switching technology is adopted to ensure that the service interruption time is less than 50ms.

10. The wavelength adaptive configuration method for a CWDM / DWDM hybrid wavelength division system according to claim 9, characterized in that: When fragmentation defragmentation is initiated during a pre-set low-load period to migrate discrete DWDM services to CWDM channels and reassemble continuous spectrum blocks, the specific migration operation of fragmentation defragmentation includes: S61: Identify source fragment service set Each of them Bandwidth usage ≤ 12.5GHz; S62: Select target CWDM channel C, whose idle spectrum capacity satisfies: ; S63: Establish a temporary optical path from S to C at the ROADM node and execute synchronously: (1) Activate the C-band tunable laser at the source end; (2) Configure a broadband coherent receiver at the destination; S64: After the migration is complete, release the discrete DWDM spectrum occupied by S and mark it as a continuous Flexi-Grid block.

Images