A communication fault location system and corresponding method based on machine learning technology

Abstract

This invention relates to the field of optical communication network fault diagnosis technology, and discloses a communication fault location system and corresponding method based on machine learning technology. The method includes the following steps: collecting raw data from the optical communication network according to a fixed sampling period; processing the raw data to obtain node features of each node in the optical communication network; generating a feature set, which includes node features of all nodes in the optical communication network, and these node features originate from raw data within the same sampling period, generating multiple training samples. This invention utilizes interpretable algorithms to analyze the contribution of node features to the output of the network fault identification model, providing more interpretable fault location results, helping maintenance personnel understand the causes and impact range of faults, accelerating problem location and repair, improving work efficiency, better handling large-scale data, and possessing real-time capabilities.

Description

Technical Field

[0001] This invention relates to the field of optical communication network fault diagnosis, and more specifically, to a communication fault location system and corresponding method based on machine learning technology. Background Technology

[0002] For example, a mobile optical communication network fault analysis and location assistance system with publication number CN104469832B utilizes the scenario of fault handling and summarization by operation and maintenance personnel to assist them in quickly locating faults on-site. Traditional fault location methods mainly rely on manual analysis and experience accumulation, which are inefficient and prone to errors.

[0003] Current methods for fault location in communication networks suffer from a lack of interpretability and targeted analysis of fault location results, leading to high costs of human intervention and trial and error during the troubleshooting process. When dealing with large-scale data, it is impossible to guarantee the efficiency of model training, real-time monitoring, and fault location. Summary of the Invention

[0004] This invention provides a communication fault location system and corresponding method based on machine learning technology, which solves the technical problem in related technologies that the lack of interpretability and targeted analysis of fault location results leads to high human intervention and trial-and-error costs in the fault diagnosis process.

[0005] This invention provides a communication fault location method based on machine learning technology, comprising the following steps:

[0006] S100 collects raw data from the optical communication network according to a fixed sampling period;

[0007] S200 processes raw data to obtain the node characteristics of each node in the optical communication network;

[0008] S300, Generate a feature set, which includes the node features of all nodes in the optical communication network. These node features are derived from the raw data within the same sampling period. Generate multiple training samples and label the training samples with optical communication network fault types. Each training sample includes a feature set.

[0009] S400 trains a network fault identification model using training samples. The network fault identification model takes a feature set as input and outputs the optical communication network fault type. The optical communication network fault type represents the type of optical communication network fault within the sampling period corresponding to the feature set. The optical communication network fault types include: no fault, single link congestion fault, single router fault, and comprehensive fault.

[0010] S500 uses an interpretable algorithm to analyze the contribution of node features to the output of the network fault identification model and obtain contribution data.

[0011] S600: Input the feature set corresponding to the sampling period to be identified into the network fault identification model trained in step S400 to obtain the optical communication network fault type of the sampling period to be identified. If the identification result is no fault, the step is terminated; otherwise, proceed to the next step.

[0012] S700 inputs the contribution data and node features into the fault location model, and the fault location model outputs the fault type of the node corresponding to the node feature.

[0013] Further, in step S100, the collected raw data includes: the interface traffic rate, interface packet error rate, forwarding throughput and internal processing latency of each node, as well as the end-to-end packet loss rate and end-to-end latency between nodes.

[0014] Further, in step S200, the node features are represented in the form of a vector, and the first to fourth components of the i-th node feature represent the interface traffic rate, interface packet error rate, forwarding throughput and internal processing latency of the i-th node, respectively.

[0015] The network contains N nodes. The 5th to N+4th components of the feature of the i-th node represent the end-to-end packet loss rate of the feature of the i-th node and the other N-1 nodes, respectively. The N+5th to 2N+4th components of the feature of the i-th node represent the end-to-end latency of the feature of the i-th node and the other N-1 nodes, respectively.

[0016] Further, step S300 includes the following steps:

[0017] S310: Generate training samples: For the data in each sampling period, combine the node features of all nodes into a feature set as a training sample. The features of all nodes in the generated training samples all come from the same sampling period.

[0018] S320: Labeling Fault Types: For each generated training sample, label the corresponding fault type according to the actual situation and known optical communication network fault types. This labeling is done using known fault datasets or based on the experience of network maintenance personnel.

[0019] S330: Dataset preparation: Combine the generated training samples and their corresponding fault types into a training dataset.

[0020] Further, in step S400, the calculation formula for the network fault identification model is as follows:

[0021] ;

[0022] in This represents the encoded feature of the i-th node. This represents the first weight parameter. and Let the node features of the i-th and j-th nodes be represented respectively. This represents the set of nodes directly connected to the i-th node. express The total number of nodes in the system tanh represents the hyperbolic tangent function. Indicates an adjustable parameter; The default value is 1 / v, where v is the dimension of the node feature. Let T denote the activation function;

[0023] ;

[0024] This represents the second weighting parameter. This represents the concatenated vector of the encoded features of all nodes. Indicates the predicted output. , , , , These represent the probability values ​​for four types of optical communication network faults.

[0025] Further, in step S400, the steps for training the network fault identification model are as follows:

[0026] S410: Data processing: For the input feature set, perform preprocessing operations such as feature scaling and normalization, and divide the training dataset into training set and validation set;

[0027] S420: Model Training: The network fault identification model is trained using the training set, and the model parameters are updated through the backpropagation algorithm;

[0028] S430: Model Evaluation: Validate the performance of the network fault identification model using a validation set;

[0029] S440: Model tuning: Tune the network fault identification model based on the verification results.

[0030] Furthermore, the contribution of node features to the output of the network fault identification model is analyzed by using the SHAP algorithm as an interpretable algorithm.

[0031] The expression for the SHAP algorithm is:

[0032]

[0033] in, This represents the SHAP value of the feature at node i in the model output. This represents the total number of features in the feature set. express It is a subset of the feature set, and It does not include the feature of the i-th node. Represents a set The base number, The model output representing the feature subset. This indicates adding the feature of the i-th node to the feature subset. The output results of the model after the middle and later stages, This represents the number of combinations, used to calculate the impact of feature subset combinations on the contribution value. Let H be the cardinality of set H. This is a factorial operation.

[0034] Further, in step S700, the contribution data is represented in the form of a vector, where the i-th component represents the SHAP value of the i-th node feature. The contribution data and the node feature are concatenated and then input into the fault location model.

[0035] A communication fault location system based on machine learning technology includes:

[0036] A data acquisition module, used to acquire raw data from an optical communication network;

[0037] Feature extraction module: It is used to obtain the node features of each node in the optical communication network from the raw data;

[0038] Training sample generation module: It is used to generate feature sets and label the training samples with optical communication network fault types;

[0039] Network fault identification model training module: It is used to train a network fault identification model through training samples to identify fault types in optical communication networks;

[0040] Feature contribution analysis module: It is used to analyze the contribution of node features to the output of the network fault identification model through interpretable algorithms to obtain contribution data;

[0041] Fault identification module: It is used to input the set of features to be identified into the trained network fault identification model to obtain the type of optical communication network fault to be identified;

[0042] Fault location module: It is used to input contribution data and node features into the fault location model and output the fault type of the node corresponding to the node feature.

[0043] A storage medium storing non-transitory computer-readable instructions for performing one or more steps in the aforementioned communication fault location method based on machine learning techniques.

[0044] The beneficial effects of this invention are as follows: This invention utilizes interpretable algorithms to analyze the contribution of node features to the output of the network fault identification model, which can provide more interpretable fault location results, help operation and maintenance personnel understand the cause and scope of the fault, speed up the problem location and repair, improve work efficiency, better handle large-scale data and have real-time performance, identify fault signals in advance by training the model, enhance the network's adaptability and rapid response capability, minimize the impact of faults on services, and improve the stability and reliability of the network. Attached Figure Description

[0045] Figure 1 This is a flowchart of a communication fault location method based on machine learning technology proposed in this invention;

[0046] Figure 2 This is a structural block diagram of a communication fault location system based on machine learning technology proposed in this invention;

[0047] Figure 3 This is the invention Figure 1 Flowchart of step S300;

[0048] Figure 4 This is the invention Figure 1 Flowchart of step S400;

[0049] Figure 5 This is a structural block diagram of a communication fault location method based on machine learning technology proposed in this invention.

[0050] In the diagram: 101, Data Acquisition Module; 102, Feature Extraction Module; 103, Training Sample Generation Module; 104, Network Fault Identification Model Training Module; 105, Feature Contribution Analysis Module; 106, Fault Identification Module; 107, Fault Location Module. Detailed Implementation

[0051] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, features described in some examples may be combined in other examples.

[0052] See Figures 1-5 As shown, a communication fault location method based on machine learning technology includes the following steps:

[0053] S100 collects raw data from the optical communication network according to a fixed sampling period;

[0054] In one embodiment of the present invention, the sampling period is 1 to 24 hours, with a default value of 1 hour.

[0055] In step S100, the specific process is as follows:

[0056] Determine the sampling period: Determine a fixed sampling period, which is determined based on actual needs and network conditions. You can choose to use a few minutes or a few hours as a sampling period.

[0057] Setting up a data acquisition point: Select a fixed data acquisition point in the optical communication network, which can be a node in the network or a central control node;

[0058] Data acquisition and storage: During each sampling period, raw data is collected from selected data acquisition points. Data acquisition can be performed through optical network monitoring equipment or sensors, and the raw data is stored in the database.

[0059] Data processing and feature extraction: Process the stored raw data and extract the node features of each node as needed. Data mining, machine learning and other methods can be used to analyze the raw data and extract the feature information of the nodes.

[0060] Feature analysis and application: Finally, the extracted node features are analyzed and applied, which can be used for network performance monitoring, fault diagnosis, load balancing and other applications. Models can be built to monitor and predict node features, and optimize the performance and stability of optical communication networks.

[0061] In step S100, the collected raw data includes:

[0062] The interface traffic rate, interface packet error rate, forwarding throughput, and internal processing latency of each node;

[0063] End-to-end packet loss rate and end-to-end latency between nodes;

[0064] The above parameters can be instantaneous values ​​or average values ​​within the sampling period.

[0065] Topology data: Records the topology between nodes in the network, including the location of nodes, neighbor relationships, and routing information.

[0066] S200 processes raw data to obtain the node characteristics of each node in the optical communication network;

[0067] In one embodiment of the present invention, the raw data collected in step S100 is the mean value within the sampling period;

[0068] Node features are represented in vector form. The first to fourth components of the i-th node feature represent the interface traffic rate, interface packet error rate, forwarding throughput, and internal processing latency of the i-th node, respectively.

[0069] The network contains N nodes. The 5th to N+4th components of the feature of the i-th node represent the end-to-end packet loss rate of the feature of the i-th node and the other N-1 nodes, respectively. The N+5th to 2N+4th components of the feature of the i-th node represent the end-to-end latency of the feature of the i-th node and the other N-1 nodes, respectively.

[0070] S300, Generate a feature set, which includes the node features of all nodes in the optical communication network. These node features are derived from the raw data within the same sampling period. Generate multiple training samples and label the training samples with optical communication network fault types. Each training sample includes a feature set.

[0071] Step S300 includes the following steps:

[0072] S310: Generate training samples:

[0073] For the data in each sampling period, the node features of all nodes are combined into a feature set as a training sample;

[0074] This method generates multiple training samples, ensuring that the features of all nodes in each training sample come from the same sampling period;

[0075] S320: Mark fault type:

[0076] For each generated training sample, based on the actual situation and known types of optical communication network faults, the corresponding fault type is labeled for each training sample. This labeling is done using known fault datasets or based on the experience of network operators.

[0077] S330: Dataset Preparation

[0078] The generated training samples and their corresponding fault types are combined to form a training dataset;

[0079] In one embodiment of the present invention, the training samples are labeled with feature sets corresponding to optical communication network fault types within the sampling period.

[0080] S400 trains a network fault identification model using training samples. The network fault identification model takes a feature set as input and outputs the optical communication network fault type. The optical communication network fault type represents the type of optical communication network fault within the sampling period corresponding to the feature set. The optical communication network fault types include: no fault, single link congestion fault, single router fault, and comprehensive fault.

[0081] In one embodiment of the present invention, the calculation formula for the network fault identification model is as follows:

[0082]

[0083] in This represents the encoded feature of the i-th node. This represents the first weight parameter. and Let the node features of the i-th and j-th nodes be represented respectively. This represents the set of nodes directly connected to the i-th node. express The total number of nodes in the system tanh represents the hyperbolic tangent function. Indicates an adjustable parameter; The default value is 1 / v, where v is the dimension of the node feature. Let T denote the activation function;

[0084]

[0085] This represents the second weighting parameter. This represents the concatenated vector of the encoded features of all nodes. Indicates the predicted output. , , , , These represent the probability values ​​for four types of optical communication network faults.

[0086] In one embodiment of the present invention, a multilayer perceptron (MLP) is used as the network fault identification model. Given a feature set, the MLP includes an input layer, a first layer, a second layer, and an output layer, and is computed as follows:

[0087]

[0088] The input vector of the input layer is: The input vector of the input layer is formed by concatenating the node features of all nodes;

[0089] First layer weight parameters: , dimension ,in Indicates the size of the hidden layer;

[0090] First layer bias parameters: , dimension ;

[0091] Activation function: Its expression is ;

[0092] Second-layer weight parameters: , dimension ,in Indicates the number of fault types output;

[0093] Second layer bias parameters: , dimension ;

[0094] Predicted output of the output layer: , , , , , These represent the probability values ​​for four types of optical communication network faults.

[0095] In one embodiment of the present invention, step S400, the step of training the network fault identification model includes the following:

[0096] S410: Data Processing: For the input feature set, perform preprocessing operations such as feature scaling and normalization to enable the model to better learn the relationships between features; and divide the training dataset into training set and validation set;

[0097] S420: Model Training: The network fault identification model is trained using the training set. The model parameters are updated through the backpropagation algorithm to continuously reduce the loss function, so that the network fault identification model can accurately predict the fault type of optical communication network.

[0098] S430: Model Evaluation: Use the validation set to validate the performance of the network fault identification model and evaluate its generalization ability on unseen data. Methods such as cross-validation can be considered.

[0099] S440: Model Tuning: Based on the validation results, the network fault identification model will be tuned. This may require adjusting the structure, learning rate, regularization parameters, etc., of the network fault identification model to improve its performance.

[0100] S500 uses an interpretable algorithm to analyze the contribution of node features to the output of the network fault identification model and obtain contribution data.

[0101] In one embodiment of the present invention, the contribution data is represented in the form of a vector, and the value of the i-th component of the vector represents the contribution of the i-th node to the output of the network fault identification model.

[0102] In one embodiment of the present invention, the SHAP (SHapley Additive exPlanations) algorithm is used as an interpretability algorithm to analyze the contribution of node features to the output of the network fault identification model.

[0103] Calculate the SHAP value for each node feature to measure the contribution of each node feature to the prediction result;

[0104] The SHAP value represents the impact, or contribution, of each node feature on the model's prediction results given a model and dataset.

[0105] The contribution data of node features are obtained through interpretability algorithms. These data include the SHAP value or other interpretable indicators for each node feature, which represent the importance and contribution of the node features in the model prediction.

[0106] A detailed table or visualization of the weights or contributions of each node's features can be obtained, showing the importance and contribution of each node's features in network fault identification;

[0107] The expression for the SHAP algorithm is as follows:

[0108]

[0109] in, This represents the SHAP value of the feature at node i in the model output. This represents the total number of features in the feature set. express It is a subset of the feature set, and It does not include the feature of the i-th node. Represents a set The base number, The model output representing the feature subset. This indicates adding the feature of the i-th node to the feature subset. The output results of the model after the middle and later stages, This represents the number of combinations, used to calculate the impact of feature subset combinations on the contribution value. Let H be the cardinality of set H. This is a factorial operation.

[0110] S600: Input the feature set corresponding to the sampling period to be identified into the network fault identification model trained in step S400 to obtain the optical communication network fault type of the sampling period to be identified. If the identification result is no fault, the step is terminated; otherwise, proceed to the next step.

[0111] In step S600, the specific implementation steps are as follows:

[0112] Input feature set: The feature set of the sampling period to be identified is used as input and input into the network fault identification model that has been trained in step 400;

[0113] Model prediction: The network fault identification model trained in step 400 is used to predict the feature set to be identified. The model will output a probability distribution representing the fault type of the optical communication network, including various fault types such as no fault, single link congestion fault, single router fault, and comprehensive fault.

[0114] Determine the identification result: Based on the model's prediction results, determine the type of optical communication network fault in the sampling period to be identified. If the identification result is no fault, the step is terminated.

[0115] If the identification result is not a fault-free fault type, it indicates that there is a fault in the network. Once it is determined that there is a fault in the network, the process can proceed to the next step S700 depending on the specific circumstances.

[0116] S700 inputs the contribution data and node features into the fault location model, and the fault location model outputs the fault type of the node corresponding to the node feature.

[0117] In one embodiment of the present invention, the contribution data is represented in the form of a vector, wherein the i-th component represents the SHAP value of the feature of the i-th node;

[0118] The contribution data and node features are concatenated and then input into the fault location model;

[0119] In one embodiment of the present invention, the fault location model employs a multilayer perceptron (MLP) to output the fault type of the node corresponding to the node features.

[0120] In one embodiment of the present invention, the fault types of a node include the presence of a fault and the absence of a fault.

[0121] In one embodiment of the present invention, node faults include absence faults, hardware faults, and software faults. More specific fault type classifications facilitate maintenance personnel in performing repairs, but correspondingly place higher performance demands on the fault location model.

[0122] In one embodiment of the present invention, three fault location models are configured, namely fault location model A, fault location model B, and fault location model C. If the result identified in step 500 is a single link congestion fault, then fault location model A is enabled to identify the fault type of the node. If the result identified in step 500 is a single router fault, then fault location model B is enabled to identify the fault type of the node. If the result identified in step 500 is a combined fault, then fault location model C is enabled to identify the fault type of the node.

[0123] The fault tracing logic differs under different fault types. Therefore, in this embodiment, a fault location model is trained independently for each fault type, and fault location models with different parameters are used to locate faults under different fault tracing logics.

[0124] In one embodiment of the present invention, an iteration period is set, and the following operation is performed at the beginning of each iteration period:

[0125] The iterative training samples are generated by using the feature set from the previous iteration cycle. The iterative training samples are then added to the set of training samples used in the previous training of the network fault identification model to obtain a new set of training samples. The parameters of the network fault identification model are then trained and updated using the new set of training samples.

[0126] The contribution of node features to the updated network fault identification model output is analyzed using an interpretability algorithm, and the contribution data is updated (the update is performed by replacing the original contribution data).

[0127] The topology of optical communication networks is not fixed but dynamic, and the network environment is also changing dynamically. Therefore, the network fault identification model needs to be dynamically updated to maintain the timeliness of the contributed data and avoid the decline in fault location performance.

[0128] Based on the above-mentioned communication fault location method based on machine learning technology, the following communication fault location system based on machine learning technology can be obtained, including:

[0129] Data acquisition module 101 is used to acquire raw data from the optical communication network;

[0130] Feature extraction module 102: It is used to obtain the node features of each node in the optical communication network from the raw data;

[0131] Training sample generation module 103: It is used to generate a feature set and label the training samples with optical communication network fault types;

[0132] Network fault identification model training module 104: It is used to train a network fault identification model through training samples to identify fault types in optical communication networks;

[0133] Feature contribution analysis module 105: It is used to analyze the contribution of node features to the output of the network fault identification model through an interpretable algorithm to obtain contribution data;

[0134] Fault identification module 106: It is used to input the set of features to be identified into the trained network fault identification model to obtain the type of optical communication network fault to be identified;

[0135] Fault location module 107: It is used to input contribution data and node features into the fault location model and output the fault type of the node corresponding to the node feature.

[0136] In one embodiment of the present invention, the communication fault location system based on machine learning technology further includes a fault repair module, which is used to switch the node identified as faulty to a backup node of that node.

[0137] In one embodiment of the present invention, the communication fault location system based on machine learning technology further includes a network coordination module, which is used to assign the communication tasks of nodes identified as faulty to nodes without faults.

[0138] At least one embodiment of this disclosure provides a storage medium storing non-transitory computer-readable instructions for performing one or more steps in the aforementioned communication fault location method based on machine learning technology.

[0139] Computer programs can be stored / distributed on suitable media, such as optical storage media or solid-state media supplied together with or as part of other hardware, but they can also be distributed in other forms, such as via the Internet or other wired or wireless telecommunications systems.

[0140] The embodiments of this example have been described above. However, this example is not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms based on the guidance of this example, and all of them are within the protection scope of this example.

Claims

1. A communication fault location method based on machine learning technology, characterized in that, Includes the following steps: S100 collects raw data from the optical communication network according to a fixed sampling period; S200 processes raw data to obtain the node characteristics of each node in the optical communication network; In step S200, the node features are represented in the form of a vector, and the first to fourth components of the i-th node feature represent the interface traffic rate, interface packet error rate, forwarding throughput and internal processing latency of the i-th node, respectively. The network contains N nodes. The 5th to N+4th components of the feature of the i-th node represent the end-to-end packet loss rate of the feature of the i-th node and the other N-1 nodes, respectively. The N+5th to 2N+4th components of the feature of the i-th node represent the end-to-end latency of the feature of the i-th node and the other N-1 nodes, respectively. S300, Generate a feature set, which includes the node features of all nodes in the optical communication network. These node features are derived from the raw data within the same sampling period. Generate multiple training samples and label the training samples with optical communication network fault types. Each training sample includes a feature set. S400 trains a network fault identification model using training samples. The network fault identification model takes a feature set as input and outputs the optical communication network fault type. The optical communication network fault type represents the type of optical communication network fault within the sampling period corresponding to the feature set. The optical communication network fault types include: no fault, single link congestion fault, single router fault, and comprehensive fault. In step S400, the calculation formula for the network fault identification model is as follows: ； in This represents the encoded feature of the i-th node. This represents the first weight parameter. and Let the node features of the i-th and j-th nodes be represented respectively. This represents the set of nodes directly connected to the i-th node. express The total number of nodes in the system tanh represents the hyperbolic tangent function. Indicates an adjustable parameter; The default value is 1 / v, where v is the dimension of the node feature. Let T denote the activation function; ； in, This represents the second weighting parameter. This represents the concatenated vector of the encoded features of all nodes. Indicates the predicted output. , , , , These represent the probability values ​​for four types of optical communication network faults; S500 uses an interpretable algorithm to analyze the contribution of node features to the output of the network fault identification model and obtain contribution data. S600: Input the feature set corresponding to the sampling period to be identified into the network fault identification model trained in step S400 to obtain the optical communication network fault type of the sampling period to be identified. If the identification result is no fault, the step is terminated; otherwise, proceed to the next step. S700 inputs the contribution data and node features into the fault location model, and the fault location model outputs the fault type of the node corresponding to the node feature.

2. The communication fault location method based on machine learning technology according to claim 1, characterized in that, In step S100, the collected raw data includes: the interface traffic rate, interface packet error rate, forwarding throughput and internal processing latency of each node, as well as the end-to-end packet loss rate and end-to-end latency between nodes.

3. The communication fault location method based on machine learning technology according to claim 1, characterized in that, Step S300 includes the following steps: S310: Generate training samples: For the data in each sampling period, combine the node features of all nodes into a feature set as a training sample. The features of all nodes in the generated training samples all come from the same sampling period. S320: Labeling Fault Types: For each generated training sample, label the corresponding fault type according to the actual situation and known optical communication network fault types. This labeling is done using known fault datasets or based on the experience of network maintenance personnel. S330: Dataset preparation: Combine the generated training samples and their corresponding fault types into a training dataset.

4. The communication fault location method based on machine learning technology according to claim 1, characterized in that, In step S400, the steps for training the network fault identification model are as follows: S410: Data processing: For the input feature set, perform preprocessing operations such as feature scaling and normalization, and divide the training dataset into training set and validation set; S420: Model Training: The network fault identification model is trained using the training set, and the model parameters are updated through the backpropagation algorithm; S430: Model Evaluation: Validate the performance of the network fault identification model using a validation set; S440: Model tuning: Tune the network fault identification model based on the verification results.

5. The communication fault location method based on machine learning technology according to claim 1, characterized in that, In step S500, the SHAP algorithm is used as the interpretability algorithm, where the expression for the SHAP algorithm is: in, This represents the SHAP value of the feature at node i in the model output. This represents the total number of features in the feature set. express It is a subset of the feature set, and It does not include the feature of the i-th node. Represents a set The base number, The model output representing the feature subset. This indicates adding the feature of the i-th node to the feature subset. The output results of the model after the middle and later stages, This represents the number of combinations, used to calculate the impact of feature subset combinations on the contribution value. Let H be the cardinality of set H. This is a factorial operation.

6. The communication fault location method based on machine learning technology according to claim 1, characterized in that, In step S700, the contribution data is represented in the form of a vector, where the i-th component represents the SHAP value of the i-th node feature. The contribution data and the node feature are concatenated and then input into the fault location model.

7. A communication fault location system based on machine learning technology, which realizes communication fault location according to any one of claims 1-6, characterized in that, include: Data acquisition module (101): It is used to acquire raw data from the optical communication network; Feature extraction module (102): It is used to obtain the node features of each node in the optical communication network from the raw data; Training sample generation module (103): It is used to generate a feature set and label the optical communication network fault type for the training samples; Network fault identification model training module (104): It is used to train the network fault identification model through training samples to identify the fault types of optical communication networks; Feature contribution analysis module (105): It is used to analyze the contribution of node features to the output of the network fault identification model through an interpretable algorithm to obtain contribution data; Fault identification module (106): It is used to input the set of features to be identified into the trained network fault identification model to obtain the type of optical communication network fault to be identified; Fault location module (107): It is used to input contribution data and node features into the fault location model and output the fault type of the node corresponding to the node feature.

8. A storage medium, characterized in that, It stores non-transitory computer-readable instructions for performing one or more steps in the communication fault location method based on machine learning technology as described in any one of claims 1-6.

Images