Endogenous holographic perception reinforcement learning collaborative optical circuit autonomous switching method and system

Abstract

The application relates to the technical field of optical communication, and provides an optical circuit autonomous switching method and system based on endogenous holographic perception and reinforcement learning collaboration, which comprises the following steps: a sending end collects a main service optical signal spectrum in real time, calculates a dynamic security masking threshold, determines probe emission power, and superimposes and sends an endogenous holographic perception probe; a receiving end separates the probe signal and calculates a link health score; a dynamic optical network graph containing a health score, energy consumption and a time penalty factor is constructed; an optical path reconfiguration action is output through a graph neural network and a deep reinforcement learning network; an optical path is established and enters a pre-connection state, a high-energy probe is deeply scanned to verify link physical quality, if the verification is passed, the service is cut in, otherwise, automatic rollback is performed. The application solves the problems that a traditional optical circuit switching (OCS) is invisible in a physical state, lacks physical perception in scheduling, and is slow in fault recovery, and realizes intelligent autonomy of an optical network.

Description

Technical Field

[0001] This application relates to the field of optical communication technology, specifically to an endogenous holographic perception reinforcement learning collaborative optical circuit autonomous switching method and system. Background Technology

[0002] Business requirements: For AI large model training and high-performance computing cluster scenarios, cloud service providers require OCS to have capabilities such as network topology reconfigurability, second-level / millisecond-level holographic perception of link physical status, autonomous decision-making for handling burst traffic, and atomic SLA guarantees.

[0003] Existing problems: Problem 1: Lack of dynamic scheduling capabilities. Existing OCSs are mostly pre-configured and cannot respond to traffic changes, leading to congestion or idle resources. Electrical layer monitoring cannot detect micro-damage in the optical layer (such as micro-bends or dirty connectors), causing scheduling algorithms to be based on an incorrect global view, which may easily schedule services to links that are 'logically connected but physically degraded'. Problem 2: Faults are not visible. Optical signals lack headers, traditional SNMP / NetFlow fail, and link degradation cannot be detected in a timely manner. Traditional OTDRs require service interruption or occupy a dedicated fiber core, which cannot meet the 'zero-interruption' monitoring requirements of production networks; SNMP polling cycles are long (seconds), making it impossible to capture millisecond-level optical transients. Issue 3: Operation and maintenance rely on manual labor. After the optical path is established, online inspection is not possible. Fault location requires OTDR or physical plugging and unplugging. The average repair time (MTTR) for manual troubleshooting is over an hour, which is orders of magnitude lower than the second / millisecond (e.g., <50ms) fault recovery capability required by AI computing power clusters. This has become a bottleneck restricting the large-scale commercial use of OCS.

[0004] These issues hinder the large-scale adoption of OCS in production networks. Summary of the Invention

[0005] To help solve the above-mentioned technical problems, this application provides an endogenous holographic perception reinforcement learning collaborative optical circuit autonomous switching method and system.

[0006] Firstly, this application provides an autonomous switching method for optical circuits based on the synergy of endogenous holographic perception and reinforcement learning, comprising: S1: The transmitting end collects the spectral density distribution of the main service optical signal in real time, calculates the dynamic security masking threshold based on the spectral density distribution, searches for the optimal frequency point in the guard band or spectral depression of the main service optical signal spectrum, determines the transmission power of the endogenous holographic sensing probe signal according to the dynamic security masking threshold, and superimposes the endogenous holographic sensing probe signal on the main service optical signal for transmission. S2: The receiving end filters the received main service optical signal to separate the endogenous holographic sensing probe signal, converts the endogenous holographic sensing probe signal into a digital signal, extracts the feature parameters of the digital signal and calculates the link health score. S3: Treat the ports of the optical circuit switching chip as nodes and the physical links as edges. Combine link health scores, optical switch action energy consumption and time penalty factors to define the multidimensional composite weights of the edges and construct a dynamic optical network graph. S4: Feature encoding of the dynamic optical network map is performed through a graph neural network, and combined with a deep reinforcement learning network, the optical path reconfiguration action is output based on a preset reward function; S5: Based on the optical path reconfiguration action, establish an optical path and enter the pre-connection state. In the pre-connection state, perform an endogenous holographic sensing probe depth scan with power increased to the upper limit of the safety threshold to verify the physical quality of the link. If the verification is successful, switch to service traffic. If the verification fails, perform automatic rollback and re-trigger path calculation.

[0007] S1 includes: The dynamic security masking threshold is calculated using the following method: ; This indicates the wavelength of the main service optical signal acquired in real time. Spectral density distribution at time t; K safe This represents the dynamic safety margin coefficient, used to offset the nonlinear effects caused by four-wave mixing and cross-phase modulation. L comp This represents the insertion loss compensation value of the receiver filter, used to calibrate power measurement deviations caused by filter characteristics.

[0008] S1 includes: The transmission power of the endogenous holographic sensing probe signal is calculated using the following method: : P probe This indicates the transmission power of the probe signal; P max This indicates the preset maximum safe output power; Q is a safety margin factor to ensure that the probe power is always below a preset safety threshold; This indicates the optimal frequency point obtained by searching in the guard band or spectral recess of the main service optical signal spectrum; This represents the dynamic security masking threshold at the optimal frequency point.

[0009] S2 includes: The characteristic parameters include at least the instantaneous power P.rx Phase The link health score is calculated using the rate of change dp / dt as follows: ; H e This represents the link health score; a higher value indicates a healthier link. P rx This represents the instantaneous power measured at the receiving end; This indicates the phase measured at the receiving end; dp / dt represents the rate of change of received power.

[0010] S3 includes: The multidimensional composite weight W is calculated using the following method. e : ; W e This represents the multidimensional composite weight of the edge; the smaller the value, the better the link. H e Represents the link health score, (1−H) e ) represents a health penalty item; P switch This indicates the estimated energy consumption of the optical switching operations involved in this link; T history Indicates the time penalty factor; , , These are the weighting coefficients.

[0011] S4 includes: Design the preset reward function in the following way: ; R represents the reward function value; R sla This indicates a positive reward for business operations that meet Service Level Agreement (SLA) requirements; R drop This represents the penalty value for packet loss or link degradation. N switch This represents a motion smoothing penalty term, used to provide negative feedback for frequent optical switching mechanical actions; W1, W2, and W3 are the weight coefficients of the corresponding items.

[0012] S4 includes: Graph neural networks are spatiotemporal graph convolutional networks. They encode the node and edge features of the dynamic optical network graph. Node features include at least the port ID and the port's real-time physical fingerprint. The physical fingerprint includes at least temperature and bias current. Edge features include at least the multidimensional composite weight W. e and historical traffic load sequences.

[0013] S5 includes: The process of performing an endogenous holographic sensing probe depth scan with power increased to the upper limit of the safety threshold in the pre-connected state to verify the physical quality of the link specifically includes: The pre-connected state refers to the state where the optical path is physically connected but the service traffic has not yet been switched in. In the pre-connected state, the power of the endogenous holographic sensing probe signal is increased to the upper limit of the safety threshold to obtain a higher signal-to-noise ratio. Detect received optical power, phase noise, polarization mode dispersion (PMD), and spectral waveform parameters; The test results are compared with the dynamic service level agreement (SLA) threshold library. If all indicators are within the threshold range, the verification is deemed successful; otherwise, the verification is deemed unsuccessful.

[0014] S5 includes: If verification fails, an automatic rollback will be executed, specifically including: Block the inbound business traffic, keep the business traffic on the original path, or trigger emergency protection switching; Add the characteristics of links that fail verification to a blacklist cache, and prevent the scheduler from selecting that path for a preset time. This triggers a new round of path calculations until a valid and usable path is found.

[0015] Secondly, this application provides an autonomous switching system for optical circuits based on the synergy of endogenous holographic perception and reinforcement learning, comprising: Optical circuit switching chip, edge intelligent controller, self-healing management unit and SDN northbound interface; The optical circuit switching chip integrates an IOTT transmit / receive unit for: Using an IOTT transmitting unit as the transmitting end, the spectral density distribution of the main service optical signal is collected in real time. Based on the spectral density distribution, a dynamic security masking threshold is calculated, and the optimal frequency point is searched in the guard band or spectral depression of the main service optical signal spectrum. The transmission power of the endogenous holographic sensing probe signal is determined according to the dynamic security masking threshold, and the endogenous holographic sensing probe signal is superimposed on the main service optical signal for transmission. Using an IOTT receiving unit as the receiving end, the received main service optical signal is filtered to separate the endogenous holographic sensing probe signal, the endogenous holographic sensing probe signal is converted into a digital signal, the feature parameters of the digital signal are extracted and the link health score is calculated. The edge intelligent controller is used for: The ports of the optical circuit switching chip are regarded as nodes, and the physical links are regarded as edges. The multidimensional composite weight of the edges is defined by combining the link health score, the energy consumption of optical switch action and the time penalty factor, and a dynamic optical network graph is constructed. The dynamic optical network map is feature-encoded using a graph neural network, and combined with a deep reinforcement learning network, an optical path reconfiguration action is output based on a preset reward function. The self-healing management unit is used to trigger optical path reconstruction transactions according to the optical path reconfiguration action, control the optical circuit switching chip to establish an optical path and enter a pre-connection state, and perform an endogenous holographic sensing probe depth scan with power increased to the upper limit of the safety threshold in the pre-connection state. If the verification is successful, the service traffic is switched in. If the verification fails, automatic rollback is performed and path calculation is re-triggered. The SDN northbound interface is used to connect to an external controller and receive service traffic and Service Level Agreement (SLA) requirements.

[0016] In summary, compared with the prior art, this application has the following advantages: 1) The first endogenous optical telemetry (endogenous holographic sensing probe IOTT) mechanism breaks through the limitation of fixed wavelength probes, realizes sub-millisecond link anomaly detection with zero additional insertion loss, uninterrupted, end-to-end optical layer telemetry, and fills the OCS sensing gap; 2) For the first time, GNN+DQN is used for dynamic scheduling of OCS, constructing a spatiotemporal graph neural network driven by both physics and logic. For the first time, the optical layer physical damage model is embedded into the AI ​​decision input, supporting multi-objective SLA optimization, and the blocking rate is significantly reduced compared with traditional heuristic algorithms. 3) Construct an autonomous closed loop to achieve "atomic transaction" processing for optical path establishment. Through a two-phase commit and pre-verification process, completely eliminate the hidden danger of "path construction is failure". The mean time to repair (MTTR) can be reduced from hours to seconds, ensuring that the business runs 100% on a healthy link. 4) Compatible with OpenConfig / YANG standards (where YANG is a data modeling language: RFC 6020 / RFC7950, used to define the structure of network device configuration data, status data, event notifications, etc.; OpenConfig is an open source project that provides a vendor-neutral collection of YANG models covering routing, interfaces, optical transmission, telemetry, etc.), and can also extend to customize the Telemetry field, achieving plug-and-play with mainstream SDN controllers, solving the interoperability problem of heterogeneous vendors, and having a foundation for industrial application; 5) Extremely low power consumption design, single-port sensing power consumption <1mW, and further reduction of nonlinear impact on the main signal through dynamic power control; 6) Define the next-generation architecture standard for OCS, evolving from a "static pipeline" to an "intelligent autonomous network," and promote the evolution of OCS from experimental networks to core production networks. Attached Figure Description

[0017] Figure 1 This is a diagram showing the overall architecture of an optical circuit autonomous switching system based on the synergy of endogenous holographic perception and reinforcement learning, as described in this application. Figure 2 This is a schematic diagram of the endogenous holographic sensing mechanism of this application; Figure 3 This is a schematic diagram of the GNN intelligent scheduling process in this application; Figure 4 This is a flowchart illustrating the self-healing verification closed loop of this application.

[0018] Figure reference numerals: 101-Silicon-based OCS switching chip, 102-Edge intelligent controller, 103-SDN northbound interface, 104-Self-healing management unit, 201-L-band filter, 202-Transimpedance amplifier (TIA), 203-Analog-to-digital converter (ADC). Detailed Implementation

[0019] The present application will be further described below with reference to the accompanying drawings. The principles of the present application are very clear to those skilled in the art. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application.

[0020] like Figure 1As shown, this application provides an optical circuit autonomous switching system based on the synergy of endogenous holographic perception and reinforcement learning. This system is an OCS architecture with endogenous perception, intelligent decision-making, and autonomous operation and maintenance capabilities, solving the structural operation and maintenance problems of traditional OCS, such as the black-box nature of physical state, static topology reconstruction, and passive fault operation and maintenance. The system includes: an optical circuit switching chip (preferably a silicon-based OCS switching chip 101 in this embodiment), an edge intelligent controller 102, a self-healing management unit 104, and an SDN northbound interface 103.

[0021] 1. Optical circuit switching chip (silicon-based OCS switching chip 101).

[0022] This chip is the core of the system's physical layer, integrating an IOTT transmit / receive unit for performing optical path switching and transmitting and receiving signals from the endogenous holographic sensing probe.

[0023] The IOTT transmitting unit (transmitter) is used to collect the spectral density distribution of the main service optical signal in real time, calculate the dynamic security masking threshold based on the spectral density distribution, search for the optimal frequency point in the guard band or spectral depression of the main service optical signal spectrum, determine the transmission power of the endogenous holographic sensing probe signal according to the dynamic security masking threshold, and superimpose the endogenous holographic sensing probe signal on the main service optical signal for transmission.

[0024] The IOTT receiving unit (receiving end) is used to filter the received main service optical signal to separate the endogenous holographic sensing probe signal, convert the endogenous holographic sensing probe signal into a digital signal, extract the feature parameters of the digital signal, and calculate the link health score.

[0025] In practice, the IOTT transmitter unit uses the Dynamic Power Spectrum Masking (DPSM) algorithm to superimpose IOTT probes onto the service light, achieving zero-interference, high-sensitivity end-to-end link health real-time telemetry.

[0026] 2. Edge intelligent controller 102.

[0027] The edge intelligent controller 102 is the core of the system's intelligent decision-making, and it is equipped with graph neural network (GNN) and deep reinforcement learning network (DQN) models. It not only processes the current state, but also predicts the evolution trend of link quality through time series analysis.

[0028] The edge intelligent controller 102 is used to treat the ports of the optical circuit switching chip as nodes and the physical links as edges. It combines the link health score, the energy consumption of optical switching action and the time penalty factor to define the multidimensional composite weight of the edges and construct a dynamic optical network graph. The dynamic optical network graph is feature-encoded by a graph neural network and combined with a deep reinforcement learning network to output the optical path reconfiguration action based on a preset reward function.

[0029] It also integrates an experience replay pool module to store state-action pair data, supporting online continuous iterative optimization of the model.

[0030] 3. Self-healing management unit 104.

[0031] The self-healing management unit 104 is the core of the system's execution and security, used to achieve a self-governing closed loop of "build-test-heal".

[0032] The self-healing management unit 104 is used to trigger optical path reconstruction transactions based on optical path reconfiguration actions, control the optical circuit switching chip to establish an optical path and enter a pre-connection state, and perform an endogenous holographic sensing probe depth scan with power increased to the upper limit of the safety threshold in the pre-connection state. If the verification is successful, the service traffic is switched in; if the verification fails, automatic rollback is performed and path calculation is re-triggered.

[0033] 4. SDN Northbound Interface 103.

[0034] SDN Northbound Interface 103 is used to interface with external controllers, receiving service traffic and Service Level Agreement (SLA) requirements. Specifically, it can interface via the gNMI protocol, is compatible with the OpenConfig / YANG standard, and achieves plug-and-play functionality with mainstream SDN controllers.

[0035] Business traffic and SLA requirements enter the edge controller through the SDN interface. The controller generates scheduling instructions and sends them to the OCS chip. At the same time, the telemetry data (IOTT) of the OCS chip is uploaded to the self-healing unit for health assessment. The assessment results are fed back to the controller to update the map weights, thus forming a closed-loop feedback.

[0036] The method described in this application will now be explained in detail.

[0037] 1. Endogenous holographic perception mechanism (see appendix) Figure 2 ).

[0038] The transmitter superimposes an IOTT probe (1610 nm, <10 μW) onto the service light (C-band, 1528–1567 nm). The receiver separates the IOTT using an on-chip L-band filter 201, calculating not only the power but also analyzing its phase noise and polarization state changes. The controller executes a dynamic power spectral masking (DPSM) algorithm to adaptively adjust the probe's transmit power in real time. The specific steps are as follows: First, spectral sensing: The spectral density distribution Smain(λ,t) of the main service signal is acquired in real time using the Optical Performance Monitoring Module (OPM); Secondly, threshold calculation: Based on the current spectral distribution, the dynamic security masking threshold is calculated using the following formula: K safeThis is a dynamic safety margin factor used to compensate for nonlinear effects such as four-wave mixing (FWM) and cross-phase modulation (XPM). Its preferred value range is 15 dB to 20 dB (higher values ​​for higher-order modulation formats such as 64-QAM, and lower values ​​for lower-order formats such as QPSK); L comp This is the insertion loss compensation value for the receiver filter (unit: dB), used to calibrate power measurement deviations caused by filter characteristics; Next, frequency locking and power injection: searching for the optimal frequency in the guard band or spectral recess of the main service spectrum. And lock the probe's emission power to , where P max The maximum safe output power of the laser is given by a coefficient Q (preferably 0.9 in this embodiment) as an additional safety margin factor to ensure that the probe power is always more than 10% below the theoretical safety threshold. Finally, closed-loop feedback protection: Establish a closed-loop feedback mechanism. If a degradation trend in the main service bit error rate (BER) is detected, immediately trigger the protection strategy: prioritize step-wise reduction of probe power (in 1dB steps). If this is ineffective, quickly switch to the backup frequency.

[0039] Signal digitization and health score calculation: The weak probe optical signal separated at the receiving end is first converted into an analog electrical signal by a transimpedance amplifier TIA202, and then sampled and quantized at high speed by a high-precision analog-to-digital converter ADC203. This ADC203 is equipped with a sampling rate and effective bit depth sufficient to capture minute phase changes, converting the analog signal into a digital sequence in real time, thereby accurately extracting the instantaneous power P. rx and phase And its rate of change dp / dt. Based on these digitized parameters, the controller executes an algorithm to calculate the link health score. The entire process is completed in the digital domain without interrupting main business traffic, achieving end-to-end real-time monitoring.

[0040] 2. GNN Intelligent Scheduling Process (see appendix) Figure 3 ).

[0041] As attached Figure 3 As shown, the edge smart controller 102 performs the following steps: Inputs: Business traffic matrix F, Service Level Agreement (SLA) requirements (from SDN northbound interface 103), Link health score H e (From self-healing management unit 104); Constructing the topology graph: Treating OCS ports as nodes V and physical links as edges E, constructing a physically aware dynamic optical network graph G. t = (V, E) tThe graph's topology evolves dynamically over time t to reflect real-time changes in link status.

[0042] Define weight: Define the multidimensional composite weight W of the edge. e It not only includes health factors but also incorporates energy consumption and stability constraints, employing a multi-objective composite weighting: .

[0043] in: H e Health status (0 to 1) calculated in real time by the IOTT module; P switch The estimated energy consumption of the optical switching operation involved in this link (encouraging the reuse of existing paths); T history Time penalty factor: If the link has experienced jitter recently (e.g., within the past minute), even if the current H... e This is normal, and it is also given a high weight to avoid "oscillating routing".

[0044] GNN encoding: Encodes node and edge features using a spatiotemporal graph convolutional network (ST-GCN).

[0045] Node characteristics: In addition to the port ID, the real-time physical fingerprint of the port (temperature, bias current) is also embedded. Edge features: Input the composite weight W calculated above e and historical traffic load sequences.

[0046] GNNs not only capture current spatial topological dependencies, but also capture the temporal evolution trend of link quality through temporal convolutional layers, predicting future trends. The probability of congestion at any given moment.

[0047] DQN Decision Making: The DQN agent is based on the state features S extracted by GNN. t Output optical path reconfiguration action A t Its reward function is designed as follows: ,in: R sla Positive rewards for services that meet SLAs (latency and bandwidth); R drop Severe penalties for packet loss or link degradation; N switch The action smoothing penalty term provides negative feedback to frequent mechanical actions of light switching, guiding the agent to learn a "less movement, more stability" strategy and extending the lifespan of the hardware.

[0048] The configuration is sent to the silicon-based OCS switching chip 101 for execution: The controller converts the optimal action output by the DQN into a specific OCS cross-connect matrix and sends it to the driver layer for execution. Simultaneously, the state and action of this decision are... The data is stored in the experience replay pool (integrated inside the edge intelligent controller 102) for continuous online iterative optimization of the model.

[0049] 3. Self-healing verification closed loop (see appendix) Figure 4 ).

[0050] Establishing an optical path: The system triggers an optical path reconstruction transaction and enters the 'Pre-ConnectionState'. At this time, the micromirrors inside the OCS have completed physical deflection, and the new optical path is physically connected, but service traffic has not yet been interrupted (or only low-priority probe packets are allowed to pass through).

[0051] Probe Scan: After path establishment is completed, the self-healing management unit 104 immediately increases the IOTT power (e.g., to 50μW) to trigger the high-energy probe depth scan mode. Power Boost: Temporarily increase the IOTT probe power to the upper limit of the safety threshold (e.g., from nW to μW) to obtain a higher signal-to-noise ratio; Multidimensional detection: It not only detects the received optical power, but also simultaneously analyzes the phase noise, polarization mode dispersion (PMD) and spectral waveform of the probe signal to comprehensively evaluate the physical quality of the link.

[0052] Judgment logic: The deep scan results are compared with a dynamic SLA threshold library. If all indicators (power, OSNR, PMD, etc.) are within the threshold range, it is judged as "verification passed"; otherwise, it is judged as "verification failed", and a fault fingerprint containing specific degradation dimensions is generated. Branch processing: "If verification fails, the controller executes the 'Rollback' instruction:" Block business traffic ingress: Ensure that business traffic remains on the original path (if the original path is not completely interrupted) or trigger emergency protection failover; Add the characteristics (fault fingerprint) of this failed link to the 'blacklist cache' and prevent the GNN scheduler from selecting this path again within the next T time period; This triggers a new round of path calculation, skipping blacklisted links until a valid and usable path is found.

[0053] It should be noted that the silicon-based OCS chip described in this application is only a preferred embodiment. This solution is also applicable to optical switch matrices based on MEMS, liquid crystal (LCoS), or other material systems; the probe wavelength is not limited to 1610nm, but can also be any wavelength within the L-band, such as 1625nm, 1650nm, or any other wavelength, as long as it meets the conditions of being isolated from the main service signal spectrum and can be separated by the receiver, all of which fall within the protection scope of this application. Furthermore, all technical solutions based on the core ideas of "dynamic power spectrum masking" and "endogenous holographic sensing," which adjust probe parameters to adapt to the main service spectrum characteristics to achieve uninterrupted real-time monitoring, should be considered equivalent embodiments of this application.

[0054] The technical scope of this application is not limited to the contents of the above specification. Those skilled in the art can make various modifications and variations to the above embodiments without departing from the technical concept of this application, and all such modifications and variations should fall within the scope of this application.

Claims

1. A method for autonomous switching of optical circuits based on the synergy of endogenous holographic perception and reinforcement learning, characterized in that, include: S1: The transmitting end collects the spectral density distribution of the main service optical signal in real time, calculates the dynamic security masking threshold based on the spectral density distribution, searches for the optimal frequency point in the guard band or spectral depression of the main service optical signal spectrum, determines the transmission power of the endogenous holographic sensing probe signal according to the dynamic security masking threshold, and superimposes the endogenous holographic sensing probe signal on the main service optical signal for transmission. S2: The receiving end filters the received main service optical signal to separate the endogenous holographic sensing probe signal, converts the endogenous holographic sensing probe signal into a digital signal, extracts the feature parameters of the digital signal and calculates the link health score. S3: Treat the ports of the optical circuit switching chip as nodes and the physical links as edges. Combine link health scores, optical switch action energy consumption and time penalty factors to define the multidimensional composite weights of the edges and construct a dynamic optical network graph. S4: Feature encoding of the dynamic optical network map is performed through a graph neural network, and combined with a deep reinforcement learning network, the optical path reconfiguration action is output based on a preset reward function; S5: Based on the optical path reconfiguration action, establish an optical path and enter the pre-connection state. In the pre-connection state, perform an endogenous holographic sensing probe depth scan with power increased to the upper limit of the safety threshold to verify the physical quality of the link. If the verification is successful, switch to service traffic. If the verification fails, perform automatic rollback and re-trigger path calculation.

2. The method according to claim 1, characterized in that, S1 includes: The dynamic security masking threshold is calculated using the following method: ； This indicates the wavelength of the main service optical signal acquired in real time. Spectral density distribution at time t; K safe This represents the dynamic safety margin coefficient, used to offset the nonlinear effects caused by four-wave mixing and cross-phase modulation. L comp This represents the insertion loss compensation value of the receiver filter, used to calibrate power measurement deviations caused by filter characteristics.

3. The method according to claim 2, characterized in that, S1 includes: The transmission power of the endogenous holographic sensing probe signal is calculated using the following method: ： P probe This indicates the transmission power of the probe signal; P max This indicates the preset maximum safe output power; Q is a safety margin factor to ensure that the probe power is always below a preset safety threshold; This indicates the optimal frequency point obtained by searching in the guard band or spectral recess of the main service optical signal spectrum; This represents the dynamic security masking threshold at the optimal frequency point.

4. The method according to claim 1, characterized in that, S2 includes: The characteristic parameters include at least the instantaneous power P. rx Phase The link health score is calculated using the rate of change dp / dt as follows: ； H e This represents the link health score; a higher value indicates a healthier link. P rx This represents the instantaneous power measured at the receiving end; This indicates the phase measured at the receiving end; dp / dt represents the rate of change of received power.

5. The method according to claim 1, characterized in that, S3 includes: The multidimensional composite weight W is calculated using the following method. e : ； W e This represents the multidimensional composite weight of the edge; the smaller the value, the better the link. H e Represents the link health score, (1−H) e ) represents a health penalty item; P switch This indicates the estimated energy consumption of the optical switching operations involved in this link; T history Indicates the time penalty factor; , , These are the weighting coefficients.

6. The method according to claim 1, characterized in that, S4 includes: Design the preset reward function in the following way: ； R represents the reward function value; R sla This indicates a positive reward for business operations that meet Service Level Agreement (SLA) requirements; R drop This represents the penalty value for packet loss or link degradation. N switch This represents a motion smoothing penalty term, used to provide negative feedback for frequent optical switching mechanical actions; W1, W2, and W3 are the weight coefficients of the corresponding items.

7. The method according to claim 1, characterized in that, S4 includes: Graph neural networks are spatiotemporal graph convolutional networks. They encode the node and edge features of the dynamic optical network graph. Node features include at least the port ID and the port's real-time physical fingerprint. The physical fingerprint includes at least temperature and bias current. Edge features include at least the multidimensional composite weight W. e and historical traffic load sequences.

8. The method according to claim 1, characterized in that, S5 includes: The process of performing an endogenous holographic sensing probe depth scan with power increased to the upper limit of the safety threshold in the pre-connected state to verify the physical quality of the link specifically includes: The pre-connected state refers to the state where the optical path is physically connected but the service traffic has not yet been switched in. In the pre-connected state, the power of the endogenous holographic sensing probe signal is increased to the upper limit of the safety threshold to obtain a higher signal-to-noise ratio. Detect received optical power, phase noise, polarization mode dispersion (PMD), and spectral waveform parameters; The test results are compared with the dynamic service level agreement (SLA) threshold library. If all indicators are within the threshold range, the verification is deemed successful; otherwise, the verification is deemed unsuccessful.

9. The method according to claim 1, characterized in that, S5 includes: If verification fails, an automatic rollback will be executed, specifically including: Block the inbound business traffic, keep the business traffic on the original path, or trigger emergency protection switching; Add the characteristics of links that fail verification to a blacklist cache, and prevent the scheduler from selecting that path for a preset time. This triggers a new round of path calculations until a valid and usable path is found.

10. An autonomous switching system for optical circuits based on the synergy of endogenous holographic perception and reinforcement learning, characterized in that, include: Optical circuit switching chip, edge intelligent controller, self-healing management unit and SDN northbound interface; The optical circuit switching chip integrates an IOTT transmit / receive unit for: Using an IOTT transmitting unit as the transmitting end, the spectral density distribution of the main service optical signal is collected in real time. Based on the spectral density distribution, a dynamic security masking threshold is calculated, and the optimal frequency point is searched in the guard band or spectral depression of the main service optical signal spectrum. The transmission power of the endogenous holographic sensing probe signal is determined according to the dynamic security masking threshold, and the endogenous holographic sensing probe signal is superimposed on the main service optical signal for transmission. Using an IOTT receiving unit as the receiving end, the received main service optical signal is filtered to separate the endogenous holographic sensing probe signal, the endogenous holographic sensing probe signal is converted into a digital signal, the feature parameters of the digital signal are extracted and the link health score is calculated. The edge intelligent controller is used for: The ports of the optical circuit switching chip are regarded as nodes, and the physical links are regarded as edges. The multidimensional composite weight of the edges is defined by combining the link health score, the energy consumption of optical switch action and the time penalty factor, and a dynamic optical network graph is constructed. The dynamic optical network map is feature-encoded using a graph neural network, and combined with a deep reinforcement learning network, an optical path reconfiguration action is output based on a preset reward function. The self-healing management unit is used to trigger optical path reconstruction transactions according to the optical path reconfiguration action, control the optical circuit switching chip to establish an optical path and enter a pre-connection state, and perform an endogenous holographic sensing probe depth scan with power increased to the upper limit of the safety threshold in the pre-connection state. If the verification is successful, the service traffic is switched in. If the verification fails, automatic rollback is performed and path calculation is re-triggered. The SDN northbound interface is used to connect to an external controller and receive service traffic and Service Level Agreement (SLA) requirements.

Images