Optical transmission network hidden danger detection method and device and computer program product

Abstract

The invention discloses an optical transmission network hidden danger detection method and device and a computer program product, which can realize dynamic adjustment of differentiated bandwidth requirements by acquiring optical network state information in real time, evaluating service priority and establishing an optical channel resource allocation model. And when a topology anomaly is detected, constructing a service quality evaluation model by using a long-short-term memory neural network, and quantitatively analyzing the influence of the anomaly on the service. When a fault occurs, distributed link fault detection and a fault diagnosis engine based on a decision tree are adopted, and fault hidden dangers are quickly positioned and classified. And then, through an elastic recovery mechanism oriented to service sensitivity, the service quality level of a standby optical path is adjusted, and an optical layer topology connection relationship is optimized. According to the method, the resource utilization efficiency, the fault recovery capability and the service quality of the optical network are remarkably improved, and an innovative solution is provided for intelligent optical network management.

Description

Technical Field

[0001] The present invention relates to the technical field of power systems, and in particular to a method, device and computer program product for detecting hidden dangers in an optical transmission network. Background Art

[0002] Large-scale data center interconnection has become an indispensable core pillar of the modern information technology system. It carries the operational requirements of key applications such as cloud computing, big data analysis and artificial intelligence, and directly affects the stability and efficiency of the global digital economy. As the main technical means of data center interconnection, dynamic optical networks are highly relied on due to their high bandwidth and low latency characteristics. However, with the growing demand for collaboration between data centers, potential failures and hidden dangers in network operation have become increasingly prominent threats to system continuity and real-time performance, becoming a major issue that needs to be addressed urgently.

[0003] At present, traditional methods for optical network fault prediction mostly rely on static models or rule-driven monitoring methods. These solutions often seem powerless when faced with traffic bursts and dynamic topology changes. They are usually unable to quickly adapt to real-time fluctuations in network status, resulting in delayed recognition of abnormal optical channel resource reallocation or optical layer performance degradation, which in turn affects the accuracy of prediction and the timeliness of early warning. This limitation is particularly evident in high-load, variable data center interconnection scenarios, and it is difficult to meet the stringent requirements of modern applications for high reliability and low latency. In this area, the core challenge focuses on how to effectively deal with abnormal optical channel resource reallocation caused by traffic bursts and topology changes, as well as the resulting optical layer performance degradation. These two technical factors are coupled with each other, increasing the complexity of network operation risks. The unpredictability of traffic bursts may lead to imbalanced resource allocation, while topology changes further exacerbate the uncertainty of optical channel status. Under the combined effect of the two, the collaborative work between data centers is prone to interruption or delay. Traditional monitoring and analysis methods are difficult to achieve accurate risk capture in a dynamic environment. Therefore, how to build a mechanism that can perceive optical channel resource reallocation anomalies in real time and predict the performance degradation trend of the optical layer in response to traffic bursts and topology changes in dynamic optical networks has become a key issue in ensuring the continuity and real-time performance of data center interconnection and collaborative work. Summary of the invention

[0004] The technical problem to be solved by the embodiments of the present invention is to provide a method, device and computer program product for detecting hidden dangers in an optical transmission network, so as to realize real-time perception of optical channel resource reallocation anomalies and predict the optical layer performance degradation trend in response to traffic bursts and topology changes in a dynamic optical network.

[0005] In order to solve the above technical problems, the present invention provides a method for detecting hidden dangers in an optical transmission network, comprising:

[0006] Step S1: Obtain the link bandwidth utilization rate, node resource occupancy rate, and topological status of the optical network in real time, evaluate the service priority by combining the user service level agreement, service type, and bandwidth requirement, and establish a differential bandwidth allocation model;

[0007] Step S2: Statistically analyze the idle wavelength resources of each optical fiber link. When the resources cannot meet the differential bandwidth requirements, calculate the service importance score based on the service type and service level, and trigger the resource preemption mechanism for wavelength reallocation;

[0008] Step S3: Monitor the wavelength occupancy rate and optical power difference of the optical fiber link, analyze the time-series data of the buffer usage of the all-optical switching node, identify the topological anomaly feature vector through the support vector machine, and generate hierarchical alarms;

[0009] Step S4: Collect the service interruption delay, optical channel packet loss rate, and link blockage data at the time of topological anomaly alarms, construct a service quality evaluation model using the long short-term memory neural network, and quantify the impact level of the anomaly on service continuity;

[0010] Step S5: When the service continuity is lower than the threshold, locate the abnormal section of the optical transmission channel through distributed link fault detection, generate a protection switching instruction, and execute the standby optical path switching;

[0011] Step S6: Analyze the service load intensity, signal attenuation rate, and optical amplifier parameters of the faulty link, use a fault diagnosis engine combining decision tree and random forest to identify cross-layer signaling anomalies and optical fiber aging hazard types, and output the demarcation area and repair suggestions;

[0012] Step S7: Dynamically adjust the service quality level of the standby optical path according to the fault severity and service sensitivity score, reconstruct the optical layer topological connection relationship, and reallocate the optical channel resources in combination with the real-time network status.

[0013] Preferably, the step S1 specifically includes:

[0014] Step S11: Periodically collect the link bandwidth utilization rate data and perform time-series smoothing processing, predict the bandwidth requirement based on the deep neural network, and establish a dynamic data set;

[0015] Step S12: Real-time monitor the node computing resource occupancy rate, combine the service level agreement and service type weights, and generate the resource priority allocation coefficient through the competitive neural network;

[0016] Step S13: Based on the network topology connectivity modeling and shortest path constraint, construct a linear programming model to achieve differential bandwidth allocation, and dynamically adjust the resource guarantee ratio according to the service priority.

[0017] Preferably, the step S2 specifically includes:

[0018] Step S21, scan the wavelength status of the optical fiber link, mark the idle wavelengths based on the optical power threshold, and construct a multi-dimensional state feature vector by combining the link utilization rate, delay jitter, and bit error rate;

[0019] Step S22, calculate the wavelength resource occupancy rate, predict the wavelength demand through a deep neural network, and trigger a priority evaluation if the predicted value exceeds the existing idle resources;

[0020] Step S23, construct a feature matrix based on the service type, service level, and bandwidth demand, generate a service importance score using the random forest algorithm, and form a wavelength resource preemption sequence;

[0021] Step S24, calculate the wavelength switching cost according to the preemption sequence, dynamically adjust the resource allocation according to the principle of the minimum switching cost, and periodically release the wavelengths occupied for over time.

[0022] Preferably, the step S3 specifically includes:

[0023] Step S31, monitor the difference between the input and output optical powers of the optical fiber link, count the wavelength occupancy rate and the non-uniformity of resource distribution between nodes, and generate a node load feature vector;

[0024] Step S32, collect the timing data of the buffer usage of the all-optical switching node, and predict the buffer overflow probability curve through a deep learning model;

[0025] Step S33, construct a support vector machine classification model based on the load feature vector and the overflow probability curve, identify the topological anomaly pattern and compare it with the hierarchical alarm threshold, and generate a three-level anomaly alarm message.

[0026] Preferably, the step S4 specifically includes:

[0027] Step S41, collect the service interruption delay, optical path failure duration, and link congestion data through periodic end-to-end detection and buffer queue monitoring, and construct a service quality monitoring data set;

[0028] Step S42, construct a multi-dimensional time series feature matrix based on the service interruption duration, recovery duration, switching delay, and link congestion degree, and train a service continuity prediction model through a long short-term memory neural network;

[0029] Step S43, classify the link congestion characteristics using the decision tree algorithm, calculate the influence quantification index in combination with the service reliability benchmark deviation, and output a hierarchical service quality evaluation result.

[0030] Preferably, the step S5 specifically includes:

[0031] Step S51: Monitor the optical power attenuation time series data through distributed link segmentation, generate a fault feature map using a long short-term memory neural network, locate the abnormal optical link, and mark the quality degradation status;

[0032] Step S52: Construct a multi-dimensional evaluation rule base based on optical power attenuation, dispersion, and polarization mode dispersion indicators, calculate the link status score through a deep belief network, trigger a connectivity alarm, and perform standby optical path switching;

[0033] Step S53: Periodically collect link optical power, bit error rate, and delay indicators, use the sliding window statistical features and support vector machine clustering analysis to verify the abnormal link, and generate a fault priority score matrix in combination with the service level.

[0034] Preferably, the step S6 specifically includes:

[0035] Step S61: Collect the service loading intensity, signal attenuation rate, and cross-layer signaling interaction data of the fault node, and construct a multi-dimensional fault symptom feature set in combination with optical amplifier parameters and fiber performance indicators;

[0036] Step S62: Generate a fault feature vector based on the optical amplifier gain deviation threshold and signaling interaction complexity, and use a decision tree model to evaluate the occurrence probabilities of cross-layer signaling anomalies, optical amplifier failures, and fiber aging respectively;

[0037] Step S63: Use the random forest ensemble method to perform confidence weighting on multiple fault probabilities, locate the fault link section, and output the damage level and repair suggestions according to the quantization standard.

[0038] Preferably, the step S7 specifically includes:

[0039] Step S71: Extract the fault severity score and service priority mark according to the fault diagnosis conclusion, analyze the user impact range through a deep neural network, and generate a fault feature vector including service sensitivity indicators;

[0040] Step S72: Construct a multi-dimensional service sensitivity evaluation matrix, screen the standby optical path based on the service quality level threshold, use a support vector machine to optimize the optical layer topology connection relationship, and generate an optical path switching sequence;

[0041] Step S73: Real-time monitor the optical path delay, jitter, and packet loss rate data, evaluate the performance degradation characteristics through a time series neural network, and dynamically adjust the node port mapping to generate an optimized topology structure.

[0042] The present invention also provides an optical transmission network hidden danger detection device, including:

[0043] One or more processors;

[0044] A memory;

[0045] One or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the one or more processors, and the one or more applications are configured to execute the optical transmission network hidden danger detection method described above.

[0046] The present invention also provides a computer program product, including computer instructions, and the computer instructions direct a computer device to execute the operations corresponding to the method.

[0047] Implementing the present invention has the following beneficial effects: By integrating multi-algorithm collaborative mechanisms such as deep neural networks, support vector machines, and decision trees, an intelligent optical network hidden danger detection and recovery system is constructed, achieving the following beneficial effects: Based on real-time bandwidth utilization prediction and dynamic scoring of service priorities, a multi-constraint optimization model is used to achieve differential resource allocation, significantly improving network bandwidth utilization and ensuring the continuity of high-priority services; Through the joint analysis of time series modeling and anomaly classification, the accuracy of topology anomaly detection is enhanced and a hierarchical alarm mechanism is established to achieve accurate quantitative assessment of service impacts; Combining random forest integrated diagnosis and elastic recovery mechanisms, various types of hidden dangers such as cross-layer signaling anomalies and optical device damages are synchronously identified, and the standby optical path switching efficiency is improved through topology dynamic reconstruction technology to ensure the reliability of critical services; This solution effectively solves problems such as rigid resource allocation, low fault location efficiency, and single recovery strategy in traditional optical networks, and while reducing operation and maintenance costs and reducing the risk of service interruption, provides self-healing and adaptive efficient management capabilities for intelligent optical networks. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or in the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention, and for those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0049] Figure 1 It is a schematic flowchart of a method for detecting hidden dangers in an optical transmission network according to Embodiment 1 of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0050] The following descriptions of the embodiments are with reference to the drawings to illustrate specific embodiments in which the present invention can be implemented.

[0051] Please refer to Figure 1 As shown, Embodiment 1 of the present invention provides a method for detecting hidden dangers in an optical transmission network, including:

[0052] Step S1, obtain the link bandwidth utilization rate, node resource occupancy rate, and topology status of the optical network in real time, evaluate the service priority in combination with the user service level agreement, service type, and bandwidth requirements, and establish a differentiated bandwidth allocation model;

[0053] Step S2, count the idle wavelength resources of each optical fiber link. When the resources cannot meet the differentiated bandwidth requirements, calculate the service importance score based on the service type and service level, and trigger the resource preemption mechanism for wavelength reallocation;

[0054] Step S3, monitor the wavelength occupancy rate and optical power difference of the optical fiber link, analyze the time-series data of the buffer usage of the all-optical switching node, and identify the topology anomaly eigenvectors through the support vector machine and generate hierarchical alarms;

[0055] Step S4, collect the service interruption delay, optical channel packet loss rate, and link blockage data when the topology anomaly alarm occurs, and construct a service quality evaluation model using the long short-term memory neural network to quantify the impact level of the anomaly on service continuity;

[0056] Step S5, when the service continuity is lower than the threshold, locate the abnormal section of the optical transmission channel through distributed link fault detection, generate a protection switching instruction and execute the standby optical path switching;

[0057] Step S6, analyze the service load intensity, signal attenuation rate, and optical amplifier parameters of the faulty link, and use a fault diagnosis engine combining decision tree and random forest to identify cross-layer signaling anomalies and types of optical fiber aging hazards, and output the demarcation area and repair suggestions;

[0058] Step S7, according to the fault severity and service sensitivity scores, dynamically adjust the service quality level of the standby optical path, reconstruct the optical layer topology connection relationship, and reallocate the optical channel resources in combination with the real-time network status.

[0059] Specifically, in step S1, during the operation of the optical network, obtain the real-time status information of the optical network, including the link bandwidth utilization rate, node computing resource occupancy rate, topology connectivity, and user service level agreement. At the same time, combine the user level, service type, and bandwidth requirements to evaluate the service priority and construct an optical channel resource allocation model to determine the differentiated bandwidth demand. It specifically includes:

[0060] Step S11, periodically collect the link bandwidth utilization rate data and perform time-series smoothing processing, predict the bandwidth demand based on the deep neural network and establish a dynamic data set.

[0061] In an embodiment of the present invention, data on link bandwidth utilization is regularly collected from the optical network resource monitoring layer. The collection period is set to 300 seconds. The data content includes link identifiers, bandwidth utilization values, and timestamps. The exponential weighted moving average method is used to smooth the collected data, and the smoothing coefficient is set to 0.8 to effectively eliminate short-term fluctuations in traffic bursts. Historical bandwidth data is analyzed through a deep neural network. The number of hidden layer nodes is set to 128, and the mean squared error is used as the loss function to train the predicted value of the bandwidth utilization within the next 300 seconds, and a bandwidth demand dataset is established based on the predicted value.

[0062] Step S12: Monitor the node computing resource occupancy rate in real time. Combine the service level agreement and the business type weight, and generate a resource priority allocation coefficient through a competitive neural network.

[0063] Regarding the node computing resource occupancy rate, the CPU usage rate, memory usage rate, and storage usage rate are monitored in real time. The warning thresholds are set to 75%, 80%, and 85% respectively. If any index exceeds the threshold, the resource constraint mechanism is triggered. The neuron competition mechanism implemented by the self-organizing feature mapping network is used to calculate the optical channel resource priority allocation coefficient. The input is the node resource occupancy data, and the output is the allocation coefficient with a value range of 0 to 1. At the same time, according to the user service level agreement, three levels of bronze, silver, and gold are defined, and the weight coefficients 1, 2, and 3 are assigned respectively. The business types are divided into ordinary data services, real-time video services, and critical services, and the weight coefficients are 1, 2, and 3 respectively. The service level and the business type weight are multiplied to obtain the business priority score value.

[0064] Step S13: Based on the network topology connectivity modeling and the shortest path constraint, construct a linear programming model to achieve differential bandwidth allocation, and dynamically adjust the resource guarantee ratio according to the business priority.

[0065] After determining the business priority, an optical channel resource allocation model is constructed based on the bandwidth prediction dataset and the priority score value. The adjacency matrix method is used to calculate the optical network topology connectivity. The matrix element values are 0 or 1 to represent the link connectivity state. The shortest path between nodes is calculated through the Floyd algorithm and the path link capacity is used as a constraint condition. The linear programming method is used to solve the model. The goal is to maximize the total network bandwidth utilization. The constraints include link capacity, node resources, and business priority. According to the preset bandwidth guarantee upper limit, resources are allocated to different priority services. For example, the critical services of gold users guarantee 100% of the bandwidth demand when resources are sufficient and 80% when resources are tight. The ordinary services of bronze users are allocated 80% when resources are sufficient and 50% when resources are tight, so as to obtain the differential bandwidth allocation result.

[0066] In the embodiment of the present invention, the real-time perception of the optical network state and the dynamic adjustment of resource allocation are achieved through the above steps. The bandwidth allocation scheme is updated every 300 seconds. If the deviation between the actual bandwidth utilization rate and the predicted value exceeds 20%, the allocation scheme is recalculated based on the optical channel resource allocation model. This mechanism can effectively cope with traffic bursts and topological changes, ensure the rationality of resource allocation and the satisfaction of service requirements, lay a foundation for subsequent anomaly detection and fault recovery, and ultimately improve the operation efficiency and reliability of the optical network.

[0067] It can be understood that the embodiment of the present invention does not overly limit the specific data collection period or algorithm parameters, which can be adjusted by technicians according to the actual scenario to adapt to the requirements of different network environments.

[0068] In step S2, the idle wavelength resources of each optical fiber link are counted according to the real-time network state information, and it is judged whether the differential bandwidth requirements are met. If not, the resource preemption and reallocation mechanism based on the service importance level is triggered, and the optical channel wavelength resources are dynamically adjusted to ensure the efficient operation of the optical network. It specifically includes:

[0069] Step S21, scan the wavelength state of the optical fiber link, mark the idle wavelengths based on the optical power threshold, and construct a multi-dimensional state feature vector in combination with the link utilization rate, delay jitter, and bit error rate.

[0070] During the operation of the optical network, the wavelength channel state data of each optical fiber link is scanned by using an optical power detector. The optical power threshold value is set to -30 dBm. When the optical power of a certain wavelength channel is lower than this value, it is marked as the idle state. Each optical fiber link contains 40 wavelength channels, and the detection interval is 100 milliseconds. An idle wavelength resource list is generated through the scanning result. The list records the number and availability identifier of each wavelength. At the same time, the real-time link state information is collected from the network manager, including the link bandwidth utilization rate expressed as a percentage, the delay jitter in microseconds, and the bit error rate expressed as a negative power of 10. A 16-dimensional optical fiber link state feature vector is constructed in combination with the number of idle wavelength resources.

[0071] Step S22, calculate the wavelength resource occupancy rate, predict the wavelength demand through a deep neural network, and trigger a priority evaluation if the predicted value exceeds the existing idle resources.

[0072] After determining the optical fiber link status eigenvector, calculate the wavelength resource occupancy rate, which is the ratio of the number of allocated wavelengths to the total number of wavelengths, and predict the wavelength resource demand within the next 500 milliseconds through a deep neural network. The network adopts a three-layer structure, with 16 nodes in the input layer, 32 nodes in the hidden layer, and 8 nodes in the output layer. The predicted demand is output through the trained model and compared with the current number of idle wavelengths. If the predicted demand exceeds the idle resources, it indicates that the current resources cannot meet the differential bandwidth allocation requirements, and the service priority needs to be further evaluated to optimize the resource allocation.

[0073] Step S23: Based on the service type, service level, and bandwidth demand, construct a feature matrix, and use the random forest algorithm to generate the service importance score to form a wavelength resource preemption sequence.

[0074] For an optical fiber link with a wavelength resource occupancy rate exceeding the preset threshold, construct a feature matrix based on the service type, service level, and bandwidth demand. The service type is divided into real-time video, data transmission, and voice call, and is assigned 3, 2, and 1 respectively. The service level is divided into high, medium, and low levels, and is assigned 3, 2, and 1 respectively. The bandwidth demand is in megabits per second. Process the feature matrix through the random forest algorithm, which contains 100 decision trees. Each tree is split according to the dimension of the feature matrix. Finally, the service importance score is output, and the score range is from 0 to 100. Subsequently, generate a priority weight coefficient table according to the score, and the weight coefficient range is from 0 to 1. For every 10-point increase in the score, the weight increases by 0.1. Arrange in descending order according to the weight coefficient to form a wavelength resource preemption sequence, and the sequence is used to guide resource reallocation.

[0075] Step S24: Calculate the wavelength switching cost according to the preemption sequence, dynamically adjust the resource allocation according to the principle of the minimum switching cost, and periodically release the wavelengths occupied for overtime.

[0076] During the resource preemption and reallocation process, calculate the wavelength switching cost of each service based on the preemption sequence. The switching cost comprehensively considers the switching delay and the service interruption duration. The switching delay is calculated in milliseconds, and the service interruption duration is controlled within 50 milliseconds. Dynamically adjust the wavelength resource allocation according to the principle of the minimum switching cost. When the occupancy duration of a certain wavelength reaches the preset 300-second cycle, automatically release the wavelength for other services to use. Through this mechanism, ensure that the bandwidth requirements of high-priority services are preferentially met while maintaining the overall stability of the network.

[0077] In the embodiments of the present invention, by the collaborative work of the optical power detector and the network manager, the wavelength resource status of the optical fiber link can be grasped in real time. For example, for a certain optical fiber link, the current bandwidth utilization rate is 85%, the delay jitter is 50 microseconds, the bit error rate is 10 to the power of negative 9, and the idle wavelength resource list shows that there are 3 remaining wavelengths. When a new high-priority service request predicts a need for 8 wavelengths, the system triggers the resource preemption mechanism, identifies 5 low-priority voice call services through the service importance degree scoring, and releases the wavelengths they occupy, so as to ensure the operation requirements of the newly added real-time video service. This method significantly improves the flexibility and efficiency of resource allocation.

[0078] It can be understood that the prediction of wavelength resource requirements depends on the training process of the deep neural network. The construction of its input feature vector fully considers multi-dimensional indicators such as bandwidth utilization rate, delay jitter, and bit error rate to ensure the accuracy of the prediction results. When calculating the service importance degree by the random forest algorithm, through the integrated voting mechanism of multiple decision trees, the overfitting problem that may occur in a single model is avoided, making the scoring results more robust. The embodiments of the present invention do not overly limit the detection interval or the threshold setting method, and can be flexibly adjusted according to the actual network scale.

[0079] In an actual scenario, if 15 real-time video services, 10 data transmission services, and 5 voice call services have been allocated on a 40-wavelength optical fiber link, when 5 new high-priority real-time video services are added, the system quickly identifies the resource gap and adjusts the allocation plan through the above steps to ensure the continuity of key services. This dynamic adjustment mechanism not only optimizes the utilization efficiency of wavelength resources.

[0080] In the embodiments of the present invention, the design of the wavelength resource reallocation rule fully considers the actual cost of service switching, and realizes the smooth transition of resources by minimizing the switching cost. For example, low-priority services with a long switching delay may be preferentially retained, while high-priority services sensitive to the interruption duration are preferentially allocated new wavelengths. This strategy effectively reduces the service interruption risk and improves the overall service quality of the optical network.

[0081] In step S3, execute the adjusted optical channel wavelength resources and analyze the change in the number of wavelengths occupied by the optical fiber link of the node and the buffer usage of the all-optical switching node buffer. Determine whether it exceeds the preset threshold by identifying and vectorizing the topological anomaly features. If it exceeds, generate a topological anomaly alarm to ensure the stability of the optical network.

[0082] In the embodiments of the present invention, the wavelength resource adjustment counter samples and records the change in the number of wavelengths of the optical fiber link every 300 seconds. The wavelength occupancy monitoring threshold is set to 75%. When the wavelength occupancy rate of a certain optical fiber link exceeds this threshold, the optical power detector is used to collect the input and output optical power values of the node, and the power difference is calculated in dBm. Under normal circumstances, the difference should be controlled within 3 dB. If it exceeds, a wavelength resource utilization statistical table is generated and the abnormal state is marked. At the same time, the buffer usage data of the optical buffer is obtained from the all-optical switching node, and the potential risks of the network operation state are analyzed based on this.

[0083] Therefore, step S3 specifically includes:

[0084] Step S31, monitor the input-output optical power difference of the optical fiber link, count the wavelength occupancy rate and the non-uniformity of resource distribution between nodes, and generate a node load feature vector.

[0085] For the generation of the wavelength resource utilization statistical table, the optical power change amount collected by the optical power detector is used to count the wavelength occupancy of each optical fiber link. For example, if the input optical power of an optical fiber link is -10 dBm and the output is -14 dBm, it is marked as abnormal when the difference exceeds 3 dB. The statistical table records the number of used wavelengths and the total number of wavelengths of each node, and calculates the non-uniformity of wavelength resource distribution between nodes. This index is obtained by comparing the wavelength occupancy rates of adjacent nodes. If the occupancy rate of a certain node reaches 90% while that of the adjacent node is only 40%, the non-uniformity value increases, reflecting the imbalance state of resource allocation. Subsequently, the non-uniformity is compared with the preset threshold to form a node load feature vector.

[0086] Step S32, collect the cache usage time series data of the all-optical switching node, and predict the cache overflow probability curve through a deep learning model.

[0087] When analyzing the buffer of the all-optical switching node, the cache usage time series data of the optical buffer is obtained. The optical buffer is implemented by an optical fiber delay line, with a fixed cache capacity. The upper limit of the queue length is set to 1000 optical packets, and the time delay of each optical packet is 1 microsecond. A deep learning method is used to train the prediction model. The model is a three-layer feedforward neural network. The input layer takes the cache usage of the past 10 sampling periods, the hidden layer has 20 nodes, and the output layer predicts the cache overflow probability curve of the next 5 sampling periods. For example, when the cache usage continuously approaches 80%, it is predicted that the overflow probability may rise to 85% within the next 5 minutes. This curve is used to evaluate the buffer pressure and assist in abnormal judgment.

[0088] Step S33, construct a support vector machine classification model based on the load feature vector and the overflow probability curve, identify the topological abnormal pattern and compare it with the hierarchical alarm threshold to generate a three-level abnormal alarm message.

[0089] Construct a support vector machine model based on the node load feature vector and the cache overflow probability curve to identify abnormal wavelength occupancy patterns. This model classifies by training normal and abnormal wavelength occupancy samples. The feature dimensions include the unevenness of wavelength resource distribution, the cache overflow probability, and the optical power fluctuation value. The radial basis kernel function is used to map the features to a high-dimensional space. The output topological anomaly feature vector includes the anomaly type and the degree score. Subsequently, the changes in fiber optic link delay, the cache overflow rate, and the node connectivity measurement are extracted and compared with the preset alarm thresholds. The alarms are divided into three levels. Slight anomalies correspond to a delay change within 50 microseconds, an overflow rate below 5%, and a connectivity decrease of no more than 10%. Moderate anomalies correspond to 100 microseconds, 10%, and 20%. Severe anomalies exceed the above ranges. According to the classification results, topological anomaly alarm information is generated and the anomaly flag bit in the wavelength resource utilization statistics table is updated.

[0090] In the embodiment of the present invention, the anomaly flag bit records the anomaly occurrence time, the duration, and the affected range. For example, if the optical power fluctuation of a certain node exceeds 5 dB within 10 minutes and the cache usage reaches 80%, and the predicted overflow probability exceeds 90%, it is identified as a moderate anomaly and an alarm is triggered. This mechanism ensures the timely discovery of topological anomalies.

[0091] It can be understood that the training process of the support vector machine model optimizes the classification boundary through historical data, and the selection of its kernel function enhances the discrimination ability of non-linear features. The deep learning prediction model captures the temporal pattern of cache usage through a multi-layer neural network. The combination of the two improves the accuracy of anomaly recognition. The present invention does not make excessive limitations on the threshold or the sampling period, which can be adjusted according to the network scale. In practical applications, if a certain all-optical switching node detects uneven wavelength resource distribution and an increase in the cache overflow probability, the system quickly generates an alarm and marks the abnormal link through the above steps, providing support for the wavelength resource adjustment feedback mechanism. This method effectively improves the adaptability of the optical network to dynamic changes and ensures service continuity.

[0092] In step S4, key service quality index data at the time of obtaining the topological anomaly alarm are acquired, including the service interruption perception delay, the optical channel packet loss rate, the optical path failure duration, and the link blocking moment. A service quality evaluation model is constructed through a long short-term memory neural network to quantify the impact of topological anomalies on service continuity and reliability and output the analysis results. Specifically, it includes:

[0093] Step S41, through periodic end-to-end detection and cache queue monitoring, collect the service interruption delay, the optical path failure duration, and the link blocking data, and construct a service quality monitoring data set.

[0094] During the operation of the optical network, the value of the service interruption perception delay when the topology anomaly alarm is triggered is extracted from the network monitor. The round-trip delay is measured by sending end-to-end probe packets every 100 milliseconds. If the delay exceeds 500 milliseconds, it is determined as a service interruption. The optical path failure duration is statistically calculated by combining the sliding window method. The window length is set to 60 seconds, and the cumulative duration of the optical path interruption within the window is recorded. At the same time, the optical channel packet loss rate is calculated with a sampling period of 1 second. The packet loss ratio is obtained by statistically counting the number of optical packets sent and received, and the length of the link buffer queue is monitored. When the queue occupancy rate exceeds 90%, it is marked as the link congestion moment. These data together constitute the basis for service quality monitoring and are used as the input for the long short-term memory neural network model.

[0095] Step S42: Construct a multi-dimensional time series feature matrix based on the service interruption duration, recovery duration, handover delay, and link congestion degree, and train the service continuity prediction model through the long short-term memory neural network.

[0096] Construct an input feature matrix of the long short-term memory neural network based on the service quality monitoring data. The matrix includes four dimensions: service interruption duration, recovery duration, optical channel handover delay, and link congestion degree. Among them, the service interruption duration records the duration of the interruption, with a range of 0 to 3600 seconds; the recovery duration is the time interval from interruption to recovery, with a range of 0 to 7200 seconds; the optical channel handover delay is the time consumption of service rerouting, usually about 50 milliseconds; the link congestion degree is expressed as the percentage of the cache queue occupancy rate. These data are normalized to map the values to the interval from 0 to 1 to ensure the stability of model training. Subsequently, the long short-term memory neural network is used to train this feature matrix. The network includes an input layer, a hidden layer, and forget gates, input gates, and output gates. The number of nodes in the hidden layer is set to 128, and the sequence length is 10 to capture the dynamic changes in the previous 10 time points, and the service continuity prediction result is output.

[0097] In the embodiment of the present invention, the long short-term memory neural network adjusts the retention ratio of historical information through the forget gate, the input gate screens the importance of the current input, and the output gate determines the output weight of the state information. The training data includes normal operation and known interruption cases, and the time series features are captured through iterative optimization. For example, when an anomaly occurs in a certain optical fiber link, the service interruption duration is 300 seconds, the recovery duration reaches 600 seconds, and the packet loss rate rises to 15%. The model predicts a significant decline in service continuity.

[0098] Step S43: Use the decision tree algorithm to classify the link congestion characteristics, calculate the impact quantification index in combination with the service reliability benchmark deviation, and output the hierarchical service quality evaluation result.

[0099] Extract link congestion features based on the business continuity prediction results, and use the decision tree method for classification. The decision tree adopts a binary tree structure and divides the congestion level based on three indicators: cache queue length, packet loss rate, and delay jitter, into three categories: mild, moderate, and severe. Assign values to different categories according to the preset business continuity scoring standard. Scores from 90 to 100 indicate good service quality, scores from 70 to 89 are average, and scores below 69 are poor. At the same time, calculate the business reliability index value. Based on the deviation between the prediction result and the service reliability benchmark value of 99.999%, it is divided into five intervals after logarithmic function conversion to generate the service quality assessment result. Further construct a topological anomaly impact quantification index, comprehensively consider the service interruption frequency, average recovery duration, and cumulative packet loss rate, and calculate the impact degree score through weighted calculation to output the hierarchical quantification analysis result.

[0100] In the embodiment of the present invention, the splitting process of the decision tree selects the optimal feature through information gain to ensure the accuracy of congestion classification. For example, when the occupancy rate of the link cache queue exceeds 90%, the packet loss rate reaches 15%, and the delay jitter is significant, the score is 65 points, indicating poor service quality and serious topological anomaly impact.

[0101] It can be understood that the sequence length and the number of hidden layer nodes of the long short-term memory neural network can be adjusted according to the network scale. Its design aims to capture the long-term dependencies of service quality indicators, while the classification of the decision tree improves the robustness of reliability assessment through comprehensive judgment of multiple indicators. The present invention does not strictly limit the sampling period or threshold, and can be optimized according to actual needs.

[0102] In an actual scenario, if a fiber optic link causes service interruption due to topological anomaly, the system quickly analyzes the degree of service quality degradation through the above steps and quantifies the anomaly impact, providing precise support for subsequent protection measures, thereby effectively reducing the service interruption risk and improving the overall reliability of the network.

[0103] In step S5, if the topological anomaly significantly affects business continuity, trigger a one-key service protection switch based on fault management. Analyze the link status of each section of the optical transmission channel through distributed link fault detection, screen the optical layer connectivity alarm links and generate the final alarm set, and at the same time execute the optical path protection switch to ensure service reliability.

[0104] In an embodiment of the present invention, by detecting in real time the service continuity monitoring value of the optical transmission channel and comparing it with the protection switching trigger threshold of 90%, when the monitoring value is lower than the threshold, a switching control instruction is generated. For example, a channel with an identification number of OC-192-001 spans 4 nodes, and the monitoring value drops to 85%. The system then initiates protection switching. The optical power attenuation index of each link segment is obtained by sampling once per second using an optical power detector. The normal attenuation should be less than 0.25 dB / km. If it exceeds this value, it indicates a potential fault risk. Subsequently, a long short-term memory neural network is used to model the time series data to generate an optical layer connectivity fault feature map for anomaly localization.

[0105] Accordingly, step S5 specifically includes:

[0106] Step S51, monitoring the optical power attenuation time series data of each link segment through distributed link segmentation, using a long short-term memory neural network to generate a fault feature map, locating the abnormal optical link and marking the quality degradation state.

[0107] In the fault feature analysis, the number of optical link segments is obtained from the distributed detection node database, and the channel is divided into detection regions. For example, the above-mentioned channel consists of 3 link segments with lengths of 80 km, 120 km, and 100 km respectively. The optical power detector detects that the attenuation value of the second link segment rises to 0.4 dB / km and lasts for more than 60 seconds. The long short-term memory neural network takes a data sequence of 10 time points as input, with 64 nodes in the hidden layer, and outputs a fault feature map. The map shows that the connectivity of this segment may have decreased due to microbending or stress damage. The segment quality assessment data is generated through link quality detection once every 30 seconds. After three consecutive anomalies, this segment is marked as a quality degradation state.

[0108] Step S52, constructing a multi-dimensional evaluation rule base based on optical power attenuation, dispersion, and polarization mode dispersion indicators, calculating the link state score through a deep belief network, triggering a connectivity alarm, and performing standby optical path switching.

[0109] In the link state scoring, a state evaluation rule base is established, including multi-dimensional indicators such as optical power attenuation, dispersion, and polarization mode dispersion. Evaluation thresholds are set, and a deep belief network is used to calculate the score. This network has a 5-layer structure, with 16 nodes in the input layer and 32, 64, 32, and 16 nodes in the middle layers in sequence. If the score of a certain link segment is lower than the preset 85 points, for example, the second link segment drops to 75 points, a level 2 connectivity alarm mark is generated. The switching control instruction carries the channel identification number, fault location information, and service priority parameters, triggering standby optical path switching. The standby path bypasses the faulty segment, and the routing length increases by 50 km. The switching duration is controlled within 50 milliseconds. After completion, the channel status information is updated.

[0110] Step S53: Periodically collect link optical power, bit error rate, and delay metrics. Use a sliding window to statistically analyze features and perform support vector machine clustering analysis to verify abnormal links, and generate a fault priority scoring matrix in combination with the service level.

[0111] In the verification of abnormal links, collect link status monitoring metric data from the optical transmission channel, including optical power, bit error rate, and delay. Sample once every 60 seconds. The preset connectivity thresholds are -25 dBm, 10^(-9), and 50 ms respectively. If the optical power of a certain link drops to -28 dBm and the bit error rate rises to 10^(-7), it is included in the initial fault marking set. Calculate the mean and variance of optical power loss, delay jitter, and bit error rate within 10 cycles through the sliding window method. The support vector machine performs clustering analysis with a radial basis kernel function to generate a quantization table of the abnormal degree, and constructs a fault priority scoring matrix based on the service level and abnormal degree, and calculates the processing priority score.

[0112] In the embodiment of the present invention, for a channel spanning 5 nodes, if the abnormal degree score of the 3rd link is 85 and it carries a special-level service, the score reaches 85, exceeding the 80-point threshold, then it enters the sequence of fault links to be verified. The deep convolutional network takes 60 cycles of time-series data as input, extracts features through the convolutional layer, reduces the dimension through the pooling layer, and outputs the abnormal probability through the fully connected layer. If the probability reaches 0.92, higher than the 0.8 threshold, then generate an optical layer connectivity alarm mark, record the time as 10:30 on March 17, 2025, locate the link between node 3 and node 4, the identification number is OC-192-003-4, and mark it as severely abnormal.

[0113] It can be understood that the long short-term memory neural network captures the fault trend through time-series modeling, the multi-layer structure of the deep belief network improves the scoring accuracy, and the support vector machine and the deep convolutional network ensure the alarm accuracy through clustering and verification. The present invention does not strictly limit the detection frequency or threshold, and can be adjusted according to actual needs.

[0114] In practical applications, protection switching and fault detection work together. If the optical power attenuation returns to normal and stabilizes for more than 300 seconds after the 2nd link is switched, the score gradually increases, and the system continuously monitors to optimize resource allocation. This mechanism significantly improves the fault response ability of the optical network and the service continuity guarantee level.

[0115] In step S6, obtain the service loading intensity and signal attenuation rate of the fault node of the alarm link, analyze the fault symptom characteristics of the optical network through a decision tree-based fault diagnosis engine, diagnose the types of fault hidden dangers including cross-layer abnormal signaling interaction, abnormal optical amplifier gain, and aging deterioration of the optical fiber link, and output a fault diagnosis conclusion including the hidden danger delimitation area, fault damage level, and repair operation suggestions. Specifically, it includes:

[0116] Step S61: Collect the service loading intensity, signal attenuation rate, and cross-layer signaling interaction data of the faulty node, and construct a multi-dimensional fault symptom feature set by combining the optical amplifier parameters and fiber performance indicators.

[0117] In the fault data collection stage, obtain the service loading intensity data from the faulty node of the alarm link. The load level is reflected by calculating the ratio of the current service traffic to the maximum processing capacity of the node. The normal value should be lower than 70%. At the same time, collect the signal attenuation detection values according to the optical power monitoring period. The normal attenuation rate should be less than 0.25 dB / km. Use a multi-source data collector to record the node fault flag bit, and obtain the cross-layer signaling data through a signaling interaction collector. Extract features such as signaling type, occurrence frequency, and timing relationship. Combine the input and output power, gain value, noise figure of the optical amplifier, and the dispersion value and polarization mode dispersion of the fiber to construct a multi-dimensional fault symptom feature set.

[0118] In an embodiment of the present invention, taking a certain faulty optical link as an example, the node service loading intensity reaches 85%, the signal attenuation rate rises to 0.4 dB / km, the abnormal signaling interaction frequency exceeds 100 times per minute, the feature set shows that the signaling interaction between the control layer and the data layer is disordered, the gain deviation of the optical amplifier exceeds 3 dB, and the fiber performance is abnormal, indicating that there may be multiple fault hazards.

[0119] Step S62: Generate a fault feature vector based on the gain deviation threshold of the optical amplifier and the signaling interaction complexity, and respectively evaluate the occurrence probabilities of cross-layer signaling anomalies, optical amplifier faults, and fiber aging through a decision tree model.

[0120] In the fault feature analysis, construct a decision tree diagnosis model based on the fault symptom feature set. By setting the abnormal detection threshold of the optical amplifier gain parameter, for example, a gain deviation greater than 2 dB is regarded as abnormal, calculate the signaling interaction complexity between nodes, and generate a fault feature vector including signaling anomaly degree, gain deviation value, and attenuation rate. Among them, the signaling anomaly degree measures the coordination between control signaling and data forwarding, the gain deviation value reflects the working state of the amplifier, and the attenuation rate reveals the degree of fiber aging. According to the vector values, respectively judge the occurrence probabilities of cross-layer abnormal signaling interaction, optical amplifier gain anomaly, and fiber link aging deterioration.

[0121] Step S63: Use the random forest integration method to perform confidence weighting on the multi-fault probabilities, locate the faulty link section, and output the damage level and repair suggestions according to the quantization standard.

[0122] In the fault type diagnosis, the random forest method is used to comprehensively evaluate the above three fault probabilities. This method constructs multiple decision trees, and each tree independently judges based on a feature subset. For example, when the signaling anomaly degree reaches 0.82, the gain deviation exceeds the standard, and the attenuation rate is abnormal, the cross-layer anomaly probability reaches 0.85, and the optical fiber aging probability is 0.78. A confidence threshold of 0.8 is set to screen high-confidence fault types, and the fault location is determined to be the optical fiber link section between node 4 and node 5, with the affected range covering 3 adjacent network elements. Subsequently, by measuring the insertion loss, dispersion value, and polarization mode dispersion, it is evaluated as a level 3 moderate damage according to the damage level quantification standard.

[0123] In the embodiment of the present invention, the random forest improves the diagnostic robustness through the voting mechanism of integrating multiple decision trees, avoids the deviation of a single model, and ensures the accurate identification of fault types. The classification of the damage level provides a basis for the repair priority.

[0124] It can be understood that the decision tree model optimizes the judgment rule through the feature importance ranking. For example, it preferentially analyzes the signaling anomaly degree to quickly locate cross-layer problems. The present invention does not make fixed limitations on the monitoring period or threshold, and can be adjusted according to the actual network requirements.

[0125] In the generation of repair suggestions, according to the hidden danger delimitation result and the damage level, the process is extracted from the preset repair operation procedure library. For the above case, it is recommended to first optimize the control layer signaling protocol to restore the interaction stability, and secondly arrange the optical fiber replacement plan, and preferentially process the moderately damaged link, generating a complete diagnostic conclusion including fault location, cause analysis, and processing suggestions, providing accurate guidance for the optical network maintenance and effectively reducing the continuous impact of the fault.

[0126] In step S7, the fault diagnosis conclusion is parsed and the fault information is extracted, including the severity, relevant service types, and the user affected range. If it is necessary to optimize the optical channel bandwidth allocation, the elastic recovery mechanism oriented to service sensitivity is started. By adjusting the service quality level of the standby optical path and changing the optical layer topology connection relationship, the distributed link fault detection is combined to locate the fault link and perform the repair according to the preset repair priority. At the same time, the optical path performance is re-evaluated based on the network state data to determine the new topology structure. It specifically includes:

[0127] Step S71, extract the fault severity score and service priority mark according to the fault diagnosis conclusion, and analyze the user affected range through the deep neural network to generate a fault feature vector including the service sensitivity index.

[0128] In the fault impact assessment, the severity score of the fault is extracted according to the fault diagnosis conclusion. For example, the score of a core business optical path is 85 points, which belongs to the high-risk level. The business priority mark is obtained from the business database. This optical path carries financial transaction services and is marked as the highest priority level 9. The number and distribution of affected users are analyzed through a deep neural network. The input features include fault location, service type, and user distribution. The hidden layer has 128 nodes, and the output shows that 2,000 users are affected and cover 3 cities, generating a fault feature vector containing business priority and impact assessment indicators.

[0129] Step S72: Construct a multi-dimensional service sensitivity assessment matrix, screen alternative optical paths based on the service quality level threshold, and use a support vector machine to optimize the optical layer topology connection relationship and generate an optical path switching sequence.

[0130] In the elastic recovery mechanism, a service sensitivity assessment matrix is constructed for the fault feature vector. The matrix comprehensively includes dimensions such as a delay sensitivity of 5 milliseconds, a packet loss sensitivity of 0.01%, and a bandwidth sensitivity of 95%. Based on the service quality level threshold, the alternative optical paths are divided into four levels. It is required that the delay of the first-level optical path is less than 10 milliseconds and the packet loss rate is less than 0.1%. Three qualified paths are screened out from the list of alternative optical path resources. The original optical path has a bandwidth of 10 gigabits per second and a service load of 8 gigabits per second. The bandwidths of the alternative optical paths are 12, 8, and 6 gigabits per second respectively. The path with a bandwidth of 12 gigabits per second is selected for switching, and 4 gigabits per second of redundancy is reserved. The optical layer topology is optimized through a support vector machine. Input features such as the distance between nodes, link bandwidth, and optical power loss are input, and the connection relationship between 2 optical cross-connect nodes is adjusted through kernel function mapping to generate an optical path switching sequence.

[0131] In the embodiment of the present invention, 200 detection points are deployed for the distributed link fault detector to collect data on optical power, bit error rate, and delay. The fault is located at 78 kilometers of the optical fiber segment and is confirmed as a micro-bending damage. The repair priority score is 95 points, ranking first. The switching duration is controlled within 50 milliseconds. After completion, fiber replacement is arranged to ensure the continuity of high-priority services.

[0132] Step S73: Real-time monitor the data of optical path delay, jitter, and packet loss rate, evaluate the performance degradation characteristics through a time series neural network, and dynamically adjust the node port mapping to generate an optimized topology structure.

[0133] In the optimization of optical path performance, the network detector collects data on network delay monitoring values, jitter change rates, and packet loss rates every 100 milliseconds. For example, the delay of a certain optical path rises to 25 milliseconds, the jitter is 5 milliseconds, and the packet loss rate is 0.1%. The long short-term memory neural network is used to extract temporal features, and the data of 10 time points are input. The three-layer structure outputs a performance degradation feature table, constructs a handover evaluation index set, combines the service priority level 9, the degradation degree 85 points, and the spare resource adequacy 95%, and calculates the handover priority score of 92 points according to the weighting rule, ranking first. After performing a non-interrupt handover, the optical power detection value is -15 dBm, the signal-to-noise ratio is 25 dB, and the bit error rate is 10 to the power of -12. The support vector machine analyzes with a Gaussian kernel function and outputs a transmission performance score of 95 points. The optical layer topology is adjusted, 2 node port mappings are changed, and a new connection scheme is generated.

[0134] It can be understood that the deep neural network improves the impact evaluation accuracy through multi-layer feature extraction, the long short-term memory neural network captures the performance change trend, the support vector machine optimizes the topological connection, and the elastic recovery mechanism realizes efficient resource allocation through sensitivity evaluation. The present invention does not strictly limit the detection frequency or threshold, and can be adjusted according to the network scale.

[0135] In practical applications, the repair timing first completes the optical path handover, and then replaces the faulty optical fiber. The optimized topological structure improves the resource utilization efficiency while ensuring the service quality.

[0136] Corresponding to the optical transmission network hidden danger detection method described in the foregoing Embodiment 1 of the present invention, Embodiment 2 of the present invention further provides an optical transmission network hidden danger detection device, including:

[0137] One or more processors;

[0138] A memory;

[0139] One or more applications, wherein the one or more applications are stored in the memory and are configured to be executed by the one or more processors, and the one or more applications are configured to execute the optical transmission network hidden danger detection method described in the foregoing Embodiment 1 of the present invention.

[0140] Corresponding to the optical transmission network hidden danger detection method described in the foregoing Embodiment 1 of the present invention, Embodiment 3 of the present invention further provides a computer program product, including computer instructions, and the computer instructions instruct the computer device to execute the operations corresponding to the optical transmission network hidden danger detection method described in the foregoing Embodiment 1 of the present invention.

[0141] Preferably, the processor may be a Central Processing Unit (CPU), or may also be other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor, or the processor may also be any conventional processor. The processor is the control center of the device, and connects various parts of the device through various interfaces and circuits.

[0142] The memory mainly includes a program storage area and a data storage area. Among them, the program storage area can store an operating system, application programs required for at least one function, etc., and the data storage area can store relevant data, etc. In addition, the memory may be a high-speed random access memory, or may also be a non-volatile memory, such as a plug-in hard disk, a SmartMedia Card (SMC), a Secure Digital (SD) card, a Flash Card, etc., or the memory may also be other volatile solid-state storage devices.

[0143] It should be noted that the above device may include, but is not limited to, a processor and a memory, which can be understood by those skilled in the art.

[0144] From the above description, it can be seen that compared with the prior art, the beneficial effects of the present invention are as follows: By integrating multi-algorithm cooperation mechanisms such as deep neural networks, support vector machines, and decision trees, an intelligent optical network hidden danger detection and recovery system is constructed, achieving the following beneficial effects: Based on real-time bandwidth utilization prediction and dynamic scoring of service priorities, a multi-constraint optimization model is used to achieve differential resource allocation, significantly improving network bandwidth utilization and ensuring the continuity of high-priority services; Through joint analysis of time series modeling and anomaly classification, the accuracy of topology anomaly detection is enhanced and a hierarchical alarm mechanism is established to achieve accurate quantitative assessment of service impacts; Combining random forest integrated diagnosis and elastic recovery mechanisms, various types of hidden dangers such as cross-layer signaling anomalies and optical device damages are synchronously identified, and the standby optical path switching efficiency is improved through topology dynamic reconstruction technology to ensure the reliability of critical services; This solution effectively solves problems such as rigid resource allocation, low fault location efficiency, and single recovery strategy in traditional optical networks, and while reducing operation and maintenance costs and reducing the risk of service interruption, provides self-healing and adaptive high-efficiency management capabilities for intelligent optical networks.

[0145] The above-disclosed is only the preferred embodiment of the present invention. Of course, it cannot be used to limit the scope of the rights of the present invention. Therefore, equivalent changes made according to the claims of the present invention still fall within the scope covered by the present invention.

Claims

1. A method for detecting potential hazards in an optical transmission network, characterized in that, Including: Step S1: Obtain the link bandwidth utilization rate, node resource occupancy rate, and topology status of the optical network in real time, evaluate the service priority by combining the user service level agreement, service type, and bandwidth demand, and establish a differentiated bandwidth allocation model. Step S2: Count the idle wavelength resources of each optical fiber link. When the resources cannot meet the differentiated bandwidth demand, calculate the service importance score based on the service type and service level, and trigger the resource preemption mechanism for wavelength reallocation. Step S3: Monitor the wavelength occupancy rate and optical power difference of the optical fiber link, analyze the time-series data of the buffer usage of the all-optical switching node, identify the topology anomaly feature vector through the support vector machine, and generate hierarchical alarms. Step S4: Collect the service interruption delay, optical channel packet loss rate, and link blockage data at the time of topology anomaly alarm, construct a service quality assessment model using the long short-term memory neural network, and quantify the impact level of the anomaly on service continuity. Step S5: When the service continuity is lower than the threshold, locate the abnormal section of the optical transmission channel through distributed link fault detection, generate a protection switching instruction, and execute the standby optical path switching. Step S6: Analyze the service load intensity, signal attenuation rate, and optical amplifier parameters of the faulty link, use a fault diagnosis engine combining decision tree and random forest to identify cross-layer signaling anomalies and optical fiber aging hidden danger types, and output the demarcation area and repair suggestions. Step S7: Dynamically adjust the service quality level of the standby optical path according to the fault severity and service sensitivity score, reconstruct the optical layer topology connection relationship, and reallocate optical channel resources in combination with the real-time network status.

2. The method according to claim 1, wherein The specific steps of Step S1 include: Step S11: Periodically collect link bandwidth utilization rate data and perform time-series smoothing processing, predict the bandwidth demand based on the deep neural network, and establish a dynamic data set. Step S12: Monitor the node computing resource occupancy rate in real time, combine the service level agreement and service type weights, and generate a resource priority allocation coefficient through the competitive neural network. Step S13: Based on the network topology connectivity modeling and shortest path constraint, construct a linear programming model to achieve differentiated bandwidth allocation, and dynamically adjust the resource guarantee ratio according to the service priority.

3. The method according to claim 1, wherein The specific steps of Step S2 include: Step S21: Scan the wavelength status of the optical fiber link, mark the idle wavelengths based on the optical power threshold, and construct a multi-dimensional state feature vector by combining the link utilization rate, delay jitter, and bit error rate. Step S22: Calculate the wavelength resource occupancy rate, predict the wavelength demand through the deep neural network, and trigger the priority evaluation if the predicted value exceeds the existing idle resources. Step S23: Construct a feature matrix based on the service type, service level, and bandwidth demand, use the random forest algorithm to generate the service importance score, and form a wavelength resource preemption sequence. Step S24: Calculate the wavelength switching cost according to the preemption sequence, dynamically adjust the resource allocation according to the principle of the minimum switching cost, and periodically release the wavelength occupied for overtime.

4. The method according to claim 1, wherein The specific steps of Step S3 include: Step S31: Monitor the input-output optical power difference of the optical fiber link, count the wavelength occupancy rate and the unevenness of resource distribution between nodes, and generate a node load feature vector. Step S32: Collect the timing data of the buffer usage of the all-optical switching node, and predict the buffer overflow probability curve through a deep learning model; Step S33: Based on the load feature vector and the overflow probability curve, construct a support vector machine classification model, identify the topological anomaly pattern, compare it with the hierarchical alarm threshold, and generate a three-level anomaly alarm message.

5. The method according to claim 1, wherein The specific steps of step S4 include: Step S41: Through periodic end-to-end detection and buffer queue monitoring, collect the service interruption delay, optical path failure duration, and link congestion data, and construct a quality of service monitoring data set; Step S42: Based on the service interruption duration, recovery duration, handover delay, and link congestion degree, construct a multi-dimensional time series feature matrix, and train a service continuity prediction model through a long short-term memory neural network; Step S43: Use the decision tree algorithm to classify the link congestion characteristics, calculate the influence quantification index in combination with the service reliability benchmark deviation, and output the hierarchical quality of service evaluation result.

6. The method according to claim 1, characterized in that The specific steps of step S5 include: Step S51: Monitor the optical power attenuation time series data through distributed link segmentation, generate a fault feature map using a long short-term memory neural network, locate the abnormal optical link, and mark the quality degradation state; Step S52: Based on the optical power attenuation, dispersion, and polarization mode dispersion indicators, construct a multi-dimensional evaluation rule base, calculate the link state score through a deep belief network, trigger a connectivity alarm, and execute the standby optical path switching; Step S53: Periodically collect the link optical power, bit error rate, and delay indicators, use the sliding window statistical features and support vector machine clustering analysis to verify the abnormal link, and generate a fault priority score matrix in combination with the service level.

7. The method according to claim 1, characterized in that, The specific steps of step S6 include: Step S61: Collect the service loading intensity, signal attenuation rate, and cross-layer signaling interaction data of the faulty node, and construct a multi-dimensional fault symptom feature set in combination with the optical amplifier parameters and fiber performance indicators; Step S62: Based on the optical amplifier gain deviation threshold and the signaling interaction complexity, generate a fault feature vector, and respectively evaluate the occurrence probabilities of cross-layer signaling anomalies, optical amplifier faults, and fiber aging through a decision tree model; Step S63: Use the random forest integration method to perform confidence weighting on multiple fault probabilities, locate the faulty link section, and output the damage level and repair suggestions according to the quantization standard.

8. The method according to claim 1, characterized in that, The specific steps of step S7 include: Step S71: Extract the fault severity score and service priority mark according to the fault diagnosis conclusion, analyze the user influence range through a deep neural network, and generate a fault feature vector containing the service sensitivity index; Step S72: Construct a multi-dimensional service sensitivity evaluation matrix, screen the standby optical path based on the quality of service level threshold, use a support vector machine to optimize the optical layer topology connection relationship, and generate an optical path switching sequence; Step S73: Real-time monitor the optical path delay, jitter, and packet loss rate data, evaluate the performance degradation characteristics through a time series neural network, and dynamically adjust the node port mapping to generate an optimized topology structure.

9. An optical transmission network hidden trouble detection device, characterized in that, Including: One or more processors; A memory; One or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the one or more processors, and the one or more applications are configured to perform the optical transmission network hidden danger detection method according to any one of claims 1 to 8.

10. A computer program product, characterized in that, Comprising computer instructions, the computer instructions instructing the computer device to perform the operations corresponding to the method according to any one of claims 1 to 8.

Images