Fault self-checking method and system of PON (Passive Optical Network) gateway

Abstract

The invention relates to a PON gateway fault self-checking method and system. A three-dimensional monitoring system of an optical module physical layer, an MAC protocol layer and an application service layer is constructed to respectively collect optical transmission performance indexes, optical module state parameters and link physical characteristics; ethernet frame statistical information and link layer protocol state information; service flow characteristics, user experience indexes and service states; extracting abnormal features of each layer, and training an abnormal recognition model based on the labeled PON gateway historical data; inputting the collected data into an exception recognition model to extract exception features, and generating an exception feature vector; recognizing fault categories through a fault recognition model on the basis of the abnormal feature vectors, and realizing cross-setting fault diagnosis knowledge sharing on the basis of a federated learning framework; and matching corresponding fault solving measures based on the fault type identification result. According to the PON gateway fault self-checking method, cross-layer data association analysis is achieved by constructing a multi-dimensional monitoring system, and the PON gateway fault self-checking method has the dynamic optimization capacity.

Description

Technical Field

[0001] The present invention belongs to the technical field of gateway fault detection, and particularly relates to a fault self-checking method and system for a PON gateway. Background Technique

[0002] With the rapid popularization of fiber-optic broadband networks, passive optical network (PON), as the core access technology for the "last mile", occupies a key position in communication networks. As the core device of the PON network, the stable operation of the PON gateway directly affects network service quality and user experience. Traditional PON gateway fault detection methods mainly rely on single-dimensional monitoring, such as simply judging the optical power of the optical module's physical layer by a threshold, or identifying faults based on a small amount of error statistics at the MAC protocol layer. This single-dimensional detection method has obvious defects and is difficult to comprehensively capture fault characteristics in a complex network environment, resulting in inaccurate fault location and a high false alarm rate.

[0003] Some existing technologies have tried to adopt multi-dimensional monitoring, but each monitoring dimension is independent of each other, lacking effective data association and collaborative analysis. For example, in the data processing of the optical module's physical layer and the MAC protocol layer, the fluctuation of optical power and the CRC error rate at the protocol layer are not deeply associated, making it difficult to detect potential compound faults in a timely manner. In addition, existing fault diagnosis models are mostly based on fixed rules or static models and cannot adapt to the dynamic changes of the network environment. When the network traffic pattern or device performance changes, the diagnosis accuracy drops significantly.

[0004] At the same time, existing fault resolution measures are mostly preset static strategies, lacking dynamic adaptability to actual fault scenarios and unable to achieve efficient fault repair in a complex fault environment.

[0005] Therefore, there is an urgent need for a PON gateway fault self-checking method that can construct a multi-dimensional monitoring system, achieve cross-layer data association analysis, ensure data privacy and security, and have the ability of dynamic optimization, so as to improve the intelligence and efficiency of network operation and maintenance. Summary of the Invention

[0006] The purpose of the present invention is to provide a fault self-checking method and system for a PON gateway, which can construct a multi-dimensional monitoring system, achieve cross-layer data association analysis, ensure data privacy and security, and have the ability of dynamic optimization, so as to improve the intelligence and efficiency of network operation and maintenance.

[0007] To solve the above technical problems, the technical solution adopted by the present invention is as follows:

[0008] In the first aspect, a fault self-checking method for a PON gateway is provided, including the following steps:

[0009] S1: Construct a three-dimensional monitoring system for the physical layer, MAC protocol layer, and application service layer of the optical module;

[0010] S2: Collect optical transmission performance indicators, optical module status parameters, and link physical characteristics through the physical layer of the optical module, collect Ethernet frame statistical information and link layer protocol status information through the MAC protocol layer, and the application service layer collects service traffic characteristics, user experience indicators, and service status;

[0011] S3: Create an anomaly recognition model to extract anomaly characteristics from each layer, and train the anomaly recognition model based on the labeled historical data of the PON gateway;

[0012] S4: Input the relevant data collected by the physical layer, MAC protocol layer, and application service layer of the optical module into the anomaly recognition model to extract anomaly characteristics and generate anomaly feature vectors;

[0013] S5: Identify the fault category based on the anomaly feature vector through the fault recognition model, and realize cross-setting fault diagnosis knowledge sharing based on the federated learning framework;

[0014] S6: Match corresponding fault resolution measures based on the fault type recognition result, and create a dynamic fault resolution model based on reinforcement learning to resolve faults.

[0015] Preferably, an optical time domain reflectometry module is provided in the physical layer of the optical module in step S1. The optical time domain reflectometry module emits short pulse light into the optical fiber. The wavelength of the short pulse light is 1310 nm or 1550 nm or 1625 nm. The optical fiber loss and fault point are detected by analyzing the time and intensity information of the reflected light;

[0016] A deep packet inspection module is set in the MAC protocol layer. The deep packet inspection module captures data through the data plane development kit, preprocesses the captured data, and performs protocol parsing and preliminary anomaly detection on the preprocessed data;

[0017] The application service layer uses a multi-mode matching algorithm to identify application layer protocols.

[0018] Preferably, the detailed execution process of data capture, preprocessing, and preliminary anomaly detection by the MAC protocol layer is as follows:

[0019] The MAC protocol layer performs zero-copy data capture, performs preliminary filtering based on the MAC address, VLAN ID, and IP address, reassembles IP fragmented packets, and adds a timestamp to each packet;

[0020] Parse the Ethernet frame header of the data packet, parse the IP header based on the Ethernet frame header type, and parse the TCP header and extract TCP features according to the IP protocol type. When there is TCP payload in the data, parse the application layer protocol.

[0021] Preferably, the specific process of creating an anomaly recognition model to extract and label anomaly features at each layer in step S3 is as follows:

[0022] S31: The anomaly recognition model extracts historical statistical features, including flow-level features, time features, and content features. The flow-level features include connection duration, number of bytes / number of packets, byte rate / packet rate, and direction ratio. The time features include inter-packet interval, burst frequency, and periodicity. The content features include URI length, request method distribution, and response status code distribution.

[0023] S32: Build a normal behavior model based on the historical statistical features and initialize the normal behavior model. The initialization content includes protocol distribution features, flow statistical features, and time series features.

[0024] S33: The anomaly recognition model performs protocol distribution anomaly detection, statistical feature anomaly detection, and time series anomaly detection based on the protocol distribution features, flow statistical features, and time series features of the normal behavior model to obtain protocol distribution anomaly features, statistical feature anomaly features, and time series anomaly features, and create an abnormal behavior pattern through the protocol distribution anomaly features, statistical feature anomaly features, and time series anomaly features.

[0025] Preferably, the detailed execution process of training the anomaly recognition model based on the labeled historical data of the PON gateway in step S3 is as follows:

[0026] S34: Divide the protocol distribution anomaly features, statistical feature anomaly features, and time series anomaly features into a training set and a test set according to a specified ratio, and input the training set into the anomaly recognition model to train the model.

[0027] S35: Input the test set into the trained anomaly recognition model for model testing, and judge whether the trained anomaly recognition model meets the specified requirements according to the test results. If so, execute step S4. If not, recreate the training set for model training.

[0028] Preferably, the specific process of realizing cross-setting fault diagnosis knowledge sharing based on the federated learning framework in step S5 is as follows:

[0029] S51: Call the corresponding federated learning framework based on the recognition result of the fault category.

[0030] S52: Create a central server and initialize the creation of a global model, and each edge PON gateway registers with the central server.

[0031] S53: The central server broadcasts the abnormal feature vector to each edge PON gateway, and each edge PON gateway performs normalization processing based on the abnormal feature vector;

[0032] S54: Each edge PON gateway performs local local model training based on its local data and the abnormal feature vector, calculates the gradient and updates the local local model parameters;

[0033] S55: Set the Gaussian noise-differential mechanism in the gradient, encrypt and transmit after over-updating the gradient, calculate the parameter difference between the local local model and the global model, and encrypt and transmit it to the central server;

[0034] S56: Perform weighted averaging on the local local models of each edge PON gateway and update the global model.

[0035] Preferably, step S56 is iteratively executed until the loss of the global model converges on the validation set or reaches the preset maximum number of iterations.

[0036] In a second aspect, a fault self-checking system for a PON gateway is provided, which is used to implement any one of the described PON gateway fault self-checking methods, including a monitoring system construction module, an abnormal recognition model, a fault recognition model, and a fault solution module. The monitoring system construction module is connected to the abnormal recognition model, the abnormal recognition model is connected to the fault recognition model, and the fault recognition model is connected to the fault solution module;

[0037] The monitoring system construction module is used to construct a three-dimensional monitoring system for the optical module physical layer, the MAC protocol layer, and the application service layer. The optical transmission performance indicators, optical module status parameters, and link physical characteristics are collected through the optical module physical layer, and the Ethernet frame statistical information and link layer protocol status information are collected through the MAC protocol layer. The application service layer collects service traffic characteristics, user experience indicators, and service status;

[0038] The abnormal recognition model is used to extract abnormal features of each layer, and train the abnormal recognition model based on the labeled historical data of the PON gateway. After the relevant data collected by the optical module physical layer, the MAC protocol layer, and the application service layer are input into the abnormal recognition model, abnormal features are extracted and an abnormal feature vector is generated;

[0039] The fault recognition model is used to identify the fault category through the fault recognition model based on the abnormal feature vector, and realize cross-setting fault diagnosis knowledge sharing based on the federated learning framework;

[0040] The model solving module is used to match corresponding fault solving measures based on the fault type recognition result, and create a dynamic fault solving model based on reinforcement learning to solve the fault.

[0041] The beneficial effects of the present invention include:

[0042] The fault self-checking method and system for a PON gateway provided by the present invention construct a three-dimensional monitoring system for the optical module physical layer, MAC protocol layer, and application service layer to respectively collect optical transmission performance indicators, optical module status parameters, and link physical characteristics; Ethernet frame statistical information and link layer protocol status information; service traffic characteristics, user experience indicators, and service status; extract abnormal characteristics of each layer, and train the abnormal recognition model based on the labeled historical data of the PON gateway; input the collected data into the abnormal recognition model to extract abnormal characteristics and generate an abnormal characteristic vector; identify the fault category through the fault recognition model based on the abnormal characteristic vector, and realize cross-setting fault diagnosis knowledge sharing based on the federated learning framework; match corresponding fault solving measures based on the fault type recognition result. It is a PON gateway fault self-checking method that realizes cross-layer data correlation analysis by constructing a multi-dimensional monitoring system and has the ability of dynamic optimization.

[0043] First, through the hierarchical monitoring of the optical module physical layer, MAC protocol layer, and application service layer, the full-link data collection from the physical link to the protocol layer and then to the service layer is realized, avoiding the limitations of single-dimensional monitoring, and different levels of faults can be accurately located, including fiber breakage in the physical layer, packet loss in the protocol layer, and application anomalies in the service layer.

[0044] Second, in the physical layer, short pulse light with wavelengths of 1310nm / 1550nm / 1625nm is used. By analyzing the time and intensity of the reflected light, quantitative detection of fiber loss and accurate positioning of the fault point are realized, improving the timeliness of physical layer fault diagnosis. The MAC protocol layer realizes zero-copy data capture and high-speed parsing based on the data plane development kit, supports IP fragmentation recombination, TCP feature extraction, and application layer protocol parsing, can quickly identify protocol layer anomalies, and reduce the missed detection rate of protocol layer faults. The multi-mode matching algorithm in the application layer identifies the application layer protocol through pattern matching, and combines with service traffic characteristics to monitor service layer anomalies in real time and ensure the user experience.

[0045] Third, by constructing a normal behavior baseline through flow-level features, time features, and content features, protocol distribution anomalies, statistical feature deviations, and time series anomalies can be effectively identified, covering a variety of potential fault modes. By dividing the training set and the test set and iteratively optimizing the model based on the labeled historical data, the accuracy and robustness of abnormal feature extraction are ensured, avoiding false alarms / missed alarms caused by model overfitting or underfitting.

[0046] Finally, through the cooperation between the central server and the edge PON gateway, cross-device fault diagnosis knowledge sharing is achieved. The edge device trains a local model based on local data, and the central server aggregates and updates the global model, which not only protects data privacy, avoids the computing power bottleneck of centralized training, but also improves the generalization ability of the model for cross-scenario faults. By iteratively weighted averaging the local model parameters, the global model continuously evolves with cross-device fault data to adapt to the diagnostic requirements of new or rare faults. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is a schematic flow chart of the fault self-checking method of the PON gateway of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0048] The following will further elaborate on the present invention in conjunction with the attached Figure 1 drawings:

[0049] Embodiment 1

[0050] Referring to the attached Figure 1 drawings, a fault self-checking method for a PON gateway includes the following steps:

[0051] S1: Construct a three-dimensional monitoring system for the optical module physical layer, MAC protocol layer, and application service layer. Deploy an optical time domain reflectometry module and an optical performance monitoring unit at the optical module physical layer;

[0052] S2: Collect optical transmission performance indicators, optical module status parameters, and link physical characteristics through the optical module physical layer, collect Ethernet frame statistical information and link layer protocol status information through the MAC protocol layer, and the application service layer collects service traffic characteristics, user experience indicators, and service status;

[0053] S3: Create an anomaly recognition model to extract anomaly features from each layer, and train the anomaly recognition model based on the labeled historical data of the PON gateway;

[0054] S4: Input the relevant data collected by the optical module physical layer, MAC protocol layer, and application service layer into the anomaly recognition model to extract anomaly features and generate an anomaly feature vector;

[0055] S5: Identify the fault category based on the anomaly feature vector through a fault recognition model, and achieve cross-setting fault diagnosis knowledge sharing based on the federated learning framework;

[0056] S6: Match corresponding fault resolution measures based on the fault type recognition result, and create a dynamic fault resolution model based on reinforcement learning to resolve the fault.

[0057] In this embodiment, an optical time domain reflectance module is provided in the physical layer of the optical module in step S1. The optical time domain reflectance module emits short pulse light into the optical fiber. The wavelength of the short pulse light is 1310 nm, 1550 nm or 1625 nm. The optical fiber loss and the fault point are detected by analyzing the time and intensity information of the reflected light. When the optical time domain reflectance module uses short pulse light with a wavelength of 1550 nm and the pulse width is set to 100 ns, it is connected to the optical fiber to be measured through an optical switch. After the reflected light is received by an APD avalanche diode, it is subjected to analog-to-digital conversion by an ADC circuit with a sampling rate of 1 GHz and a resolution of 12 bits to obtain the reflected light intensity sequence. For example, when a sudden drop in the reflected light intensity is detected and there is no subsequent reflection peak, the optical fiber break point can be located.

[0058] The MAC protocol layer is provided with a deep packet inspection module. The deep packet inspection module captures data through a data plane development kit, preprocesses the captured data, and performs protocol parsing and preliminary anomaly detection on the preprocessed data.

[0059] The MAC protocol layer realizes high-speed data processing based on the data plane development kit. The zero-copy technology is used to capture data packets through the network card driver. First, preliminary filtering is performed based on the MAC address, VLAN ID, and IP address, and irrelevant traffic is discarded. For IP fragmented data packets, they are reorganized by maintaining a fragmentation reassembly table, and the reassembly timeout is set to 200 ms.

[0060] The data packet parsing process is as follows:

[0061] Parse the Ethernet frame header and extract the EtherType field to determine the upper-layer protocol;

[0062] If it is the IP protocol, further parse the IP header to obtain information such as the source IP, destination IP, and protocol type;

[0063] If it is the TCP protocol, parse the TCP header, extract the source port, destination port, and flag bits, and calculate TCP features including window size and sequence number;

[0064] If there is TCP payload, a multi-mode matching algorithm is used to identify the application-layer protocol, and the protocol feature library contains feature strings of HTTP, HTTPS, and DNS protocols.

[0065] The application service layer uses a multi - mode matching algorithm to identify application - layer protocols. The application service layer collects service data through traffic mirroring and analyzes service traffic characteristics using the NetFlow protocol. The following metrics are calculated in real - time: Flow - level characteristics: connection duration, number of bytes / number of packets, byte rate / packet rate, direction ratio; Time characteristics: inter - packet interval, burst frequency, periodicity; Content characteristics: URI length, request method distribution, response status code distribution. At the same time, user experience metrics are obtained through active probing and passive monitoring. When the average delay exceeds 100 ms or the packet loss rate exceeds 1%, it is marked as abnormal service quality.

[0066] The detailed execution process of data capture, pre - processing, and preliminary anomaly detection in the MAC protocol layer is as follows:

[0067] The MAC protocol layer performs zero - copy data capture, conducts preliminary filtering based on MAC address, VLAN ID, and IP address, reassembles IP - fragmented data packets, and adds a timestamp to each data packet. The Ethernet frame header of the data packet is parsed, the IP header is parsed based on the Ethernet frame header type, and the TCP header is parsed and TCP features are extracted according to the IP protocol type. When there is TCP payload in the data, the application - layer protocol is parsed.

[0068] Embodiment 2

[0069] Based on Embodiment 1, the specific process of creating an anomaly recognition model to extract anomaly features of each layer and label them in step S3 is as follows:

[0070] S31: The anomaly recognition model extracts historical statistical features, including flow - level features, time features, and content features. The flow - level features include connection duration, number of bytes / number of packets, byte rate / packet rate, direction ratio. The time features include inter - packet interval, burst frequency, periodicity. The content features include URI length, request method distribution, response status code distribution;

[0071] S32: A normal behavior model is constructed based on the historical statistical features, and the normal behavior model is initialized. The initialization content includes protocol distribution features, flow statistical features, and time - series features;

[0072] S33: The anomaly recognition model performs protocol distribution anomaly detection, statistical feature anomaly detection, and time - series anomaly detection based on the protocol distribution features, flow statistical features, and time - series features of the normal behavior model to obtain protocol distribution anomaly features, statistical feature anomaly features, and time - series anomaly features, and creates an abnormal behavior pattern through the protocol distribution anomaly features, statistical feature anomaly features, and time - series anomaly features.

[0073] In this embodiment, the detailed execution process of training the anomaly recognition model based on the labeled historical data of the PON gateway in step S3 is as follows:

[0074] S34: Divide the protocol distribution anomaly features, statistical feature anomaly features, and time series anomaly features into a training set and a test set at a ratio of 8:2, and input the training set into the anomaly recognition model to train the model;

[0075] S35: Input the test set into the trained anomaly recognition model for model testing, and determine whether the trained anomaly recognition model meets the specified requirements according to the test results. If so, execute step S4. In this embodiment, if the accuracy rate of the test set is lower than 90%, adjust the hyperparameters or increase the training data to retrain and recreate the training set for model training.

[0076] Embodiment 3

[0077] Based on Embodiment 1 or Embodiment 2, the specific process of realizing cross-setting fault diagnosis knowledge sharing in step S5 is as follows:

[0078] S51: Invoke the corresponding federated learning framework based on the recognition result of the fault category;

[0079] S52: Create a central server and initialize and create a global model, and each edge PON gateway registers with the central server;

[0080] S53: The central server broadcasts the anomaly feature vector to each edge PON gateway, and each edge PON gateway performs normalization processing based on the anomaly feature vector;

[0081] S54: Each edge PON gateway performs local local model training based on its local data and the anomaly feature vector, calculates the gradient, and updates the local local model parameters;

[0082] S55: Set Gaussian noise-differential mechanism in the gradient, encrypt and transmit after over-updating the gradient, calculate the parameter difference between the local local model and the global model, and encrypt and transmit it to the central server;

[0083] S56: Perform weighted averaging on the local local models of each edge PON gateway to update the global model.

[0084] Iteratively execute step S56 until the loss of the global model converges on the validation set or reaches the preset maximum number of iterations.

[0085] In this embodiment, the federated learning framework adopts an architecture of "central server + edge PON gateway". The central server is built based on the Flask framework and provides API interfaces such as model parameter distribution and aggregation; the edge gateway is based on the ARM Cortex-A72 processor and runs TensorFlow Lite to implement local model training.

[0086] The training process includes initialization: the central server initializes the global random forest model, and the edge gateway registers with the central server and uploads information such as device ID and data scale; model distribution: the central server broadcasts the global model parameters to the edge gateways; local training: the edge PON gateway trains the global model based on local labeled data, uses the Mini-Batch SGD optimizer, sets the learning rate to 0.01, and the batch size to 32. Parameter upload: Calculate the parameter difference between the local model and the global model, such as the updated value of the tree node splitting threshold, use the PyNaCl library to encrypt the gradient with AES-256, and add Gaussian noise to achieve differential privacy, with the noise intensity ∈ set to 1.0, and upload it to the central server. Model aggregation: The central server uses the FedAvg algorithm to perform weighted averaging on the parameter updates of each edge gateway, with the weight being the proportion of the data volume of each gateway, and updates the global model. Iterative training: Repeat steps 2-5 until the loss of the global model on the validation set drops by less than 0.1% for 5 consecutive rounds or reaches the maximum number of iterations of 100 rounds.

[0087] In step S6, corresponding fault solution measures are matched according to the fault type recognition result, and a dynamic fault solution model based on reinforcement learning is created to solve the fault. The dynamic fault solution model based on reinforcement learning is used as the dynamic fault solution model. Its network structure contains 2 hidden layers, with 256 neurons in each layer, and the activation function is ReLU. The state space contains a total of 100-dimensional features such as fault type, the current execution stage of the solution measure, and the real-time state of the network; the action space contains 10 fault solution measures; the reward function is designed as follows:

[0088] Successfully solve the fault: +100;

[0089] Partially relieve the fault: +30;

[0090] The measure execution fails: -50;

[0091] Timeout and not resolved: -100.

[0092] The decision-making process of the dynamic fault solution model is as follows:

[0093] After the fault recognition model outputs the fault type, obtain the current network state to construct the state vector st;

[0094] Input the state vector st into the DQN model. Select the action at through the ε-greedy policy, where the initial value of ε is 0.1 and it decays by 0.001 every 1000 steps.

[0095] Execute the action at, obtain the next state st+1 and the reward rt, and store them in the experience replay pool.

[0096] Randomly sample 32 samples from the experience replay pool and calculate the target Q value:

[0097]

[0098] Among them, γ is the discount factor with a value of 0.99, and θ- is the target network parameter; use the mean square error (MSE) loss function to update the DQN model parameter θ:

[0099]

[0100] Synchronize the main network parameters to the target network every 100 steps.

[0101] A fault self-checking system for a PON gateway, which is used to implement the fault self-checking method of any one of the above-mentioned PON gateways, including a monitoring system construction module, an anomaly recognition model, a fault recognition model, and a fault resolution module. The monitoring system construction module is connected to the anomaly recognition model, the anomaly recognition model is connected to the fault recognition model, and the fault recognition model is connected to the fault resolution module. The monitoring system construction module is used to construct a three-dimensional monitoring system for the optical module physical layer, MAC protocol layer, and application service layer. Collect optical transmission performance indicators, optical module status parameters, and link physical characteristics through the optical module physical layer, collect Ethernet frame statistical information and link layer protocol status information through the MAC protocol layer, and collect service traffic characteristics, user experience indicators, and service status through the application service layer.

[0102] The anomaly recognition model is used to extract anomaly features at each layer and train the anomaly recognition model based on the labeled historical data of the PON gateway. After the relevant data collected at the optical module physical layer, MAC protocol layer, and application service layer are input into the anomaly recognition model, anomaly features are extracted and an anomaly feature vector is generated. The fault recognition model is used to identify the fault category based on the anomaly feature vector through the fault recognition model and realize cross-setting fault diagnosis knowledge sharing based on the federated learning framework. The model solution module is used to match corresponding fault resolution measures based on the fault type recognition result and create a dynamic fault resolution model based on reinforcement learning to solve the fault.

[0103] In summary, the fault self - detection method and system for the PON gateway provided by the present invention construct a three - dimensional monitoring system for the optical module physical layer, MAC protocol layer, and application service layer to respectively collect optical transmission performance indicators, optical module status parameters, and link physical characteristics; Ethernet frame statistical information, link - layer protocol status information; service traffic characteristics, user experience indicators, and service status; extract abnormal characteristics of each layer, and train the abnormal recognition model based on the labeled historical data of the PON gateway; input the collected data into the abnormal recognition model to extract abnormal characteristics and generate abnormal feature vectors; identify the fault category based on the abnormal feature vectors through the fault recognition model, and achieve cross - setting fault diagnosis knowledge sharing based on the federated learning framework; match corresponding fault - solving measures based on the fault type recognition result. It is a PON gateway fault self - detection method that realizes cross - layer data correlation analysis by constructing a multi - dimensional monitoring system and has the ability of dynamic optimization.

[0104] Through hierarchical monitoring of the optical module physical layer, MAC protocol layer, and application service layer, full - link data collection from the physical link to the protocol layer and then to the service layer is realized, avoiding the limitations of single - dimension monitoring, and different - level faults can be accurately located, including fiber breakage in the physical layer, packet loss in the protocol layer, and application anomalies in the service layer. In the physical layer, short - pulse light with wavelengths of 1310nm / 1550nm / 1625nm is used. By analyzing the time and intensity of the reflected light, quantitative detection of fiber loss and accurate positioning of the fault point are achieved, improving the timeliness of physical - layer fault diagnosis. The MAC protocol layer realizes zero - copy data capture and high - speed parsing based on the data plane development kit, supports IP fragmentation recombination, TCP feature extraction, and application - layer protocol parsing, can quickly identify protocol - layer anomalies, and reduce the missed - detection rate of protocol - layer faults. The multi - mode matching algorithm in the application layer identifies application - layer protocols through pattern matching, combined with service traffic characteristics, can monitor service - layer anomalies in real time, and ensure the user experience.

[0105] By constructing a normal behavior baseline through flow - level characteristics, time characteristics, and content characteristics, abnormal protocol distribution, statistical feature deviation, and time - series anomalies can be effectively identified, covering a variety of potential fault modes. By dividing the training set and the test set and iteratively optimizing the model based on the labeled historical data, the accuracy and robustness of abnormal feature extraction are ensured, avoiding false alarms / missed alarms caused by model over - fitting or under - fitting. Through the cooperation between the central server and the edge PON gateway, cross - device fault diagnosis knowledge sharing is achieved. The edge device trains a local model based on local data, and the central server aggregates and updates the global model, which not only protects data privacy, avoids the computing power bottleneck of centralized training, improves the generalization ability of the model for cross - scenario faults, but also makes the global model evolve continuously with cross - device fault data through iterative weighted averaging of local model parameters to adapt to the diagnostic requirements of new or rare faults.

Claims

1. A fault self-checking method for a PON gateway, characterized in that, It includes the following steps: S1: Construct a three-dimensional monitoring system for the physical layer, MAC protocol layer, and application service layer of the optical module; S2: Collect optical transmission performance indicators, optical module status parameters, and link physical characteristics through the physical layer of the optical module, collect Ethernet frame statistical information and link layer protocol status information through the MAC protocol layer, and the application service layer collects service traffic characteristics, user experience indicators, and service status; S3: Create an anomaly recognition model to extract anomaly characteristics of each layer, and train the anomaly recognition model based on the labeled historical data of the PON gateway; S4: Input the relevant data collected by the physical layer, MAC protocol layer, and application service layer of the optical module into the anomaly recognition model to extract anomaly characteristics and generate anomaly feature vectors; S5: Identify the fault category based on the anomaly feature vector through the fault recognition model, and achieve cross-setting fault diagnosis knowledge sharing based on the federated learning framework; S6: Match corresponding fault resolution measures based on the fault type recognition result, and create a dynamic fault resolution model based on reinforcement learning to resolve faults.

2. The fault self-checking method of a PON gateway according to claim 1, wherein, In step S1, the physical layer of the optical module is provided with an optical time domain reflectometry module, which emits short pulse light into the optical fiber. The wavelength of the short pulse light is 1310nm or 1550nm or 1625nm, and the optical fiber loss and fault points are detected by analyzing the time and intensity information of the reflected light; The MAC protocol layer is provided with a deep packet inspection module. The deep packet inspection module captures data through the data plane development kit, preprocesses the captured data, and performs protocol parsing and preliminary anomaly detection on the preprocessed data; The application service layer uses a multi-pattern matching algorithm to identify application layer protocols.

3. The fault self-checking method of a PON gateway according to claim 2, characterized in that, The detailed execution process of data capture, preprocessing, and preliminary anomaly detection by the MAC protocol layer is as follows: The MAC protocol layer performs zero-copy data capture, performs preliminary filtering based on MAC address, VLAN ID, and IP address, reorganizes IP fragmented packets, and adds a timestamp to each packet; Parse the Ethernet frame header of the packet, parse the IP header based on the Ethernet frame header type, and parse the TCP header and extract TCP features according to the IP protocol type. When there is TCP payload in the data, parse the application layer protocol.

4. A fault self-checking method for a PON gateway according to claim 1, characterized in that, The specific process of creating an anomaly recognition model to extract and label anomaly characteristics of each layer in step S3 is as follows: S31: The anomaly recognition model extracts historical statistical features, including flow-level features, time features, and content features. The flow-level features include connection duration, number of bytes / number of packets, byte rate / packet rate, and direction ratio. The time features include inter-packet interval, burst frequency, and periodicity. The content features include URI length, request method distribution, and response status code distribution; S32: Construct a normal behavior model based on historical statistical features and initialize the normal behavior model. The initialization content includes protocol distribution features, flow statistical features, and time series features; S33: The abnormal recognition model performs protocol distribution anomaly detection, statistical feature anomaly detection, and time series anomaly detection based on the protocol distribution features, flow statistical features, and time series features of the normal behavior model to obtain protocol distribution anomaly features, statistical feature anomaly features, and time series anomaly features, and creates an abnormal behavior pattern through the protocol distribution anomaly features, statistical feature anomaly features, and time series anomaly features.

5. A fault self-checking method for a PON gateway according to claim 4, characterized in that The detailed execution process of training the abnormal recognition model based on the labeled historical data of the PON gateway in step S3 is as follows: S34: Divide the protocol distribution anomaly features, statistical feature anomaly features, and time series anomaly features into a training set and a test set according to a specified ratio, and input the training set into the abnormal recognition model to train the model; S35: Input the test set into the trained abnormal recognition model for model testing, and judge whether the trained abnormal recognition model meets the specified requirements according to the test results. If so, execute step S4; if not, recreate the training set for model training.

6. The fault self-checking method of a PON gateway according to claim 1, characterized in that, The specific process of realizing cross-setting fault diagnosis knowledge sharing based on the federated learning framework in step S5 is as follows: S51: Call the corresponding federated learning framework based on the recognition result of the fault category; S52: Create a central server and initialize the creation of a global model, and each edge PON gateway registers with the central server; S53: The central server broadcasts the abnormal feature vector to each edge PON gateway, and each edge PON gateway performs normalization processing based on the abnormal feature vector; S54: Each edge PON gateway performs local local model training based on its local data and the abnormal feature vector, calculates the gradient, and updates the local local model parameters; S55: Set the Gaussian noise-differential mechanism in the gradient, encrypt and transmit after over-updating the gradient, calculate the parameter difference between the local local model and the global model, and encrypt and transmit it to the central server; S56: Perform weighted averaging on the local local models of each edge PON gateway and update the global model.

7. A fault self-checking method for a PON gateway according to claim 6, characterized in that, Iteratively execute step S56 until the loss of the global model converges on the validation set or reaches the preset maximum number of iterations.

8. A fault self-checking system for a PON gateway, which is used to implement the fault self-checking method for a PON gateway according to any one of claims 1-7, and is characterized in that, It includes a monitoring system construction module, an abnormal recognition model, a fault recognition model, and a fault resolution module. The monitoring system construction module is connected to the abnormal recognition model, the abnormal recognition model is connected to the fault recognition model, and the fault recognition model is connected to the fault resolution module; The monitoring system construction module is used to construct a three-dimensional monitoring system for the optical module physical layer, MAC protocol layer, and application service layer. It collects optical transmission performance indicators, optical module status parameters, and link physical characteristics through the optical module physical layer, collects Ethernet frame statistical information and link layer protocol status information through the MAC protocol layer, and collects service traffic characteristics, user experience indicators, and service status through the application service layer; The abnormal recognition model is used to extract abnormal features at each layer, and train the abnormal recognition model based on the labeled historical data of the PON gateway. After the relevant data collected at the physical layer, MAC protocol layer, and application service layer of the optical module are input into the abnormal recognition model, abnormal features are extracted and an abnormal feature vector is generated; The fault recognition model is used to identify the fault category through the fault recognition model based on the abnormal feature vector, and realize cross-setting fault diagnosis knowledge sharing based on the federated learning framework; The model solution module is used to match corresponding fault solution measures based on the fault type recognition result, and create a dynamic fault solution model based on reinforcement learning to solve the fault.

Images