A method and system for fault location in communication networks based on fault propagation relationships

Abstract

This invention discloses a method and system for fault location in communication networks based on fault propagation relationships, belonging to the field of communication operator operation and maintenance technology. It represents the alarm at the moment of a fault using a propagation relationship graph, identifies the root cause device and root cause alarm from the graph, and completes the location of the fault root cause. The method uses fault handling work order data for training model correction, continuously iterating and updating for optimization. The implementation includes: constructing an alarm type dependency graph; constructing a topology-based alarm propagation relationship graph based on the alarm type dependency graph and network element device resource topology graph; constructing a fault training dataset; training a neural network model; and iteratively optimizing the trained model. This invention can improve fault location accuracy, reduce fault location time, and reduce the dependence of operation and maintenance work on expert experience.

Description

Technical Field

[0001] This invention relates to the field of telecommunications operator operation and maintenance technology, specifically a method and system for fault location in communication networks based on fault propagation relationships. Background Technology

[0002] Operator networks face complex situations involving collaborative services from multiple equipment models from different manufacturers and the coexistence and integration of multiple standards in network architecture. Furthermore, driven by new technologies and services such as 5G, NFV, SDN, cloud-network convergence, and computing power networks, the network architecture is evolving from planar networking to a three-dimensional networking model, leading to an exponential increase in the difficulty of network operation and maintenance management.

[0003] When a network element malfunctions, alarm data is reported through underlying network management platforms such as the OMC (Operation and Maintenance Center) and EMS (Electronic Element Management System). Simultaneously, neighboring network elements are also affected and issue related alarm data. Currently, upper-level network management receives alarm data from various underlying network management platforms for centralized monitoring. Alarm reports from different platforms influence each other, resulting in complex and intertwined relationships between alarms. Manual troubleshooting is time-consuming, heavily reliant on expert experience, and has low accuracy. Furthermore, continuous network and service updates increase the learning cost of manual experience. Accurately locating the actual faulty network element based on alarm data and determining the appropriate handling measures has become a major pain point for customers. Summary of the Invention

[0004] The technical objective of this invention is to address the above-mentioned shortcomings by providing a communication network fault location method and system based on fault propagation relationships, which can improve fault location accuracy, reduce fault location time, and reduce the dependence of operation and maintenance work on expert experience.

[0005] The technical solution adopted by this invention to solve its technical problem is:

[0006] A communication network fault location method based on fault propagation relationships represents alarms at the time of a fault using a propagation relationship diagram. The root cause device and root cause alarm are identified from the propagation relationship diagram, thus locating the root cause of the fault. The method uses fault handling work order data for training model correction, continuously iterating, updating, and optimizing. The implementation of this method includes:

[0007] Construct an alarm type dependency graph;

[0008] Construct a topology-based alarm propagation relationship graph based on alarm type dependency graph and network element device resource topology graph;

[0009] By connecting the fault work order data and the alarm propagation relationship diagram, the fault alarm data and work order processing conclusion data are extracted to construct a fault training dataset;

[0010] Three-layer neural network model training: The fault training dataset is used as input to train the model using a three-layer neural network, and the model is evaluated using precision, recall, and F-score.

[0011] The training model is optimized and iterated. By continuously acquiring fault data and work order data, the dependency graph of alarm types is constantly enriched, and transfer learning is used to continuously optimize the model accuracy.

[0012] This method utilizes alarm data, resource data, work order data, and existing rules based on expert experience from the operator's network management system for daily fault handling. First, it organizes an alarm dependency graph using expert experience rules. Then, it instantiates existing historical faults to establish a fault propagation graph. Finally, it inputs the fault propagation graph instances into a neural network for training to generate a fault location model. Work order data is used as annotation data for continuous iterative optimization. When a fault occurs again, the fault propagation graph instances are input into the location model to complete the delimitation and location of the fault root cause.

[0013] Embedding this fault location method into production and maintenance processes enables rapid identification of the root cause of faults, solving the problems of long manual fault location and troubleshooting cycles and high reliance on human experience, and improving the level of automated operation and maintenance of networks.

[0014] Preferably, the alarm type dependency graph is constructed, with nodes {Node} as the basis for the graph. i}, i = 1, ..., N represent alarm types, where N is the number of nodes, i.e., the number of alarm types; with directed edges<A,B> This indicates the triggering relationship from node A to node B (two alarm types or system components).

[0015] Preferably, the construction of the alarm propagation relationship graph based on topology,

[0016] The network element device resource topology diagram describes the physical connection relationships between devices. Based on the alarm type dependency relationship diagram, an alarm propagation (sub) diagram of the fault device alarm type set is constructed on the network element device resource topology diagram. The alarm propagation relationship diagram is a set of network element device nodes with a center radius ≤ R starting from the fault device. Each alarm is traversed at least once within R steps. If the number of steps exceeds R, it is considered that the fault type of the network element device is unrelated to the current fault type.

[0017] Preferably, the default value for R is 2, but this may change depending on the dependencies.

[0018] Preferably, the fault work order data and the alarm information of each node in the alarm propagation relationship diagram are concatenated by depth-first or breadth-first traversal to extract the fault alarm data and work order processing conclusion data.

[0019] Preferably, the three-layer neural network includes an input layer, a hidden layer, and an output layer. It is a multi-layer feedforward network trained using the error backpropagation algorithm. The input data of the input layer is fault alarm data, and the output layer outputs possible conclusions (device nodes) and the probability of each possible conclusion. The neural network has three layers.

[0020] The number of neurons in the input layer is related to the dimension of the fault alarm data, and the number of neurons in the output layer is related to the type of conclusion. The number of neurons is adjusted according to the business situation. The data is randomly disordered, and 60% of the data is extracted as training data, 20% as validation data, and 20% as test data. The neural network is trained using the above data. The model is evaluated by using precision, recall, and F-score to complete the model training.

[0021] Preferably, the training model is optimized and iterated to accurately locate faults in the communication network through the fault location model, reducing the reliance on expert experience in the fault location process and making it a powerful support for the daily operation and maintenance of network faults.

[0022] Preferably, the implementation of the neural network model training includes data acquisition, data storage, model training, and model management.

[0023] The data acquisition process involves acquiring labeled training data and real data.

[0024] The data storage corresponds to labeled training data and real data, respectively forming a training dataset and a prediction dataset;

[0025] The model training involves training the neural network model and iteratively optimizing the training model based on the training dataset; and combining the prediction dataset to form a multi-classification model.

[0026] The model management includes model loading, model deployment, and model prediction.

[0027] Furthermore, the neural network model can be directly invoked through the model call API interface to locate the root cause of the fault.

[0028] This invention also claims a communication network fault location system based on fault propagation relationships, comprising a module for constructing an alarm type dependency graph, a module for constructing a topology-based alarm propagation graph, a fault training dataset module, a neural network model training module, and an optimization module for the trained model.

[0029] The module for constructing an alarm type dependency graph constructs an alarm type dependency graph.

[0030] The module for constructing a topology-based alarm propagation relationship diagram constructs a topology-based alarm propagation relationship diagram based on the alarm type dependency relationship diagram and the network element device resource topology diagram.

[0031] The fault training dataset module connects fault work order data and alarm propagation relationship diagram, extracts fault alarm data and work order processing conclusion data, and constructs a fault training dataset.

[0032] The neural network model training module takes the fault training dataset as input, uses a three-layer neural network for model training, and uses precision, recall, and F-score for model evaluation.

[0033] The training model optimization and iteration module continuously acquires fault data and work order data to enrich the alarm type dependency graph and uses transfer learning to continuously optimize the model accuracy.

[0034] The system achieves fault root cause localization through the aforementioned fault propagation-based fault root cause localization method.

[0035] Compared with the prior art, the communication network fault location method and system based on fault propagation relationship of the present invention have the following advantages:

[0036] This method uses machine learning algorithms to analyze the handling process of historical faults to determine the true cause of the fault, which is then used to guide the root cause localization of subsequent faults, shorten the time required for fault localization in operator networks, improve the accuracy of fault localization, and enhance the efficiency of fault handling.

[0037] This method starts with historical fault data from operators and combines it with the current status of operator network management. Using the expert experience rules accumulated in the existing network as a prototype, it expresses the alarm at the time of the fault using a propagation relationship diagram. From the propagation relationship diagram, it finds the root cause device and root cause alarm of the fault, thus completing the location of the fault root cause. It uses fault handling work order data to train and correct the model, and continuously iterates, updates and optimizes it. It supplements and improves the fault location rules, reducing the dependence of operation and maintenance work on expert experience. Attached Figure Description

[0038] Figure 1 This is a flowchart illustrating the algorithm implementation of the communication network fault location method based on fault propagation relationship provided in this embodiment of the invention;

[0039] Figure 2 This is a diagram illustrating a three-layer neural network in a communication network fault location method based on fault propagation relationships provided in an embodiment of the present invention. Detailed Implementation

[0040] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0041] This invention provides a communication network fault location method based on fault propagation relationships. The alarm at the time of the fault is expressed using a propagation relationship diagram. The root cause device and root cause alarm are found from the propagation relationship diagram to complete the location of the fault root cause. The model is trained and corrected using fault handling work order data, and continuously iterated, updated and optimized.

[0042] Starting from historical fault data of operators and combining it with the current status of operator network management; taking the expert experience rules accumulated in the existing network as the prototype, the alarm at the time of the fault is expressed by a propagation relationship diagram. The root cause device and root cause alarm are found from the propagation relationship diagram, and the root cause of the fault is located. The model is trained and corrected by fault handling work order data, and iteratively updated, optimized and adjusted; the fault location rules are supplemented and improved to reduce the dependence of operation and maintenance work on expert experience.

[0043] The implementation of this method includes five steps: constructing an alarm type dependency graph, constructing a topology-based alarm propagation graph, fault training dataset, neural network model training, and training model optimization iteration.

[0044] 1. Construct an alarm type dependency graph.

[0045] A global alert type dependency graph is maintained using expert experience rules. This graph is based on nodes {Node i}, i = 1, ..., N represent alarm types, where N is the number of nodes, i.e., the number of alarm types; with directed edges<A,B> This indicates the triggering relationship from node A to node B (two alarm types or system components).

[0046] 2. Construct a topology-based alarm propagation relationship diagram based on the alarm type dependency relationship diagram and the network element device resource topology diagram.

[0047] The network element device resource topology diagram describes the physical connections between devices. Based on the alarm type dependency graph, an alarm propagation (sub-graph) of the fault device alarm type set is constructed on the network element device resource topology diagram. The alarm propagation graph is a set of network element device nodes with a radius ≤ R centered on the fault device. Each alarm is traversed at least once within R steps; if it exceeds R, it is considered that the fault type of the network element device is unrelated to the current fault type. The default value of R is 2, and it may change later based on the dependency relationship.

[0048] 3. Fault training dataset: Connect fault work order data and alarm propagation relationship diagram, extract fault alarm data and work order processing conclusion data, and construct fault training dataset.

[0049] The fault work order data and alarm information of each node in the alarm propagation relationship diagram are concatenated using a depth-first / breadth-first traversal to extract fault alarm data and work order processing conclusion data.

[0050] 4. Neural network model training: Using the fault training dataset as input, a three-layer neural network is used for model training, and the model is evaluated using precision, recall, and F-score.

[0051] The three-layer neural network includes an input layer, a hidden layer, and an output layer. It is a multi-layer feedforward network trained using an error backpropagation algorithm, and its structure is as follows: Figure 2 The diagram shown is a BP neural network; the input layer takes fault alarm data as input, and the output layer outputs the probability of each conclusion (device node). The neural network has three layers.

[0052] The number of neurons in the input layer is related to the dimension of the fault alarm data, and the number of neurons in the output layer is related to the type of conclusion. The number of neurons is adjusted according to the business situation. The data is randomly disordered, and 60% of the data is extracted as training data, 20% as validation data, and 20% as test data. The neural network is trained using the above data. The model is evaluated by using precision, recall, and F-score to complete the model training.

[0053] 5. Train the model and optimize iteratively. Continuously acquire fault data and work order data to enrich the alarm type dependency graph. Use transfer learning to continuously optimize the model accuracy.

[0054] By using fault location models to accurately locate faults in communication networks, the reliance on expert experience in the fault location process is reduced, making it a powerful support for daily network fault maintenance.

[0055] Embedding this fault location method into production and maintenance processes enables rapid identification of the root cause of faults, solving the problems of long manual fault location and troubleshooting cycles and high reliance on human experience, and improving the level of automated operation and maintenance of networks.

[0056] refer to Figure 1 As shown, the implementation of the neural network model training includes data acquisition, data storage, model training, and model management.

[0057] The data acquisition process involves acquiring labeled training data and real data.

[0058] The data storage corresponds to labeled training data and real data, respectively forming a training dataset and a prediction dataset;

[0059] The model training involves training the neural network model and iteratively optimizing the training model based on the training dataset; and combining the prediction dataset to form a multi-classification model.

[0060] The model management includes model loading, model deployment, and model prediction.

[0061] The neural network model can be directly invoked through the model call API interface to locate the root cause of the fault.

[0062] This method utilizes alarm data, resource data, work order data, and existing rules based on expert experience from the operator's network management system for daily fault handling. First, it organizes an alarm dependency graph using expert experience rules. Then, it instantiates existing historical faults to establish a fault propagation graph. Finally, it inputs the fault propagation graph instances into a neural network for training to generate a fault location model. Work order data is used as annotation data for continuous iterative optimization. When a fault occurs again, the fault propagation graph instances are input into the location model to complete the delimitation and location of the fault root cause.

[0063] This invention also provides a communication network fault location system based on fault propagation relationships, including a module for constructing an alarm type dependency graph, a module for constructing a topology-based alarm propagation graph, a fault training dataset module, a neural network model training module, and an optimization module for the trained model. This system achieves fault root cause location through the fault propagation relationship-based fault root cause location method described in the above embodiments.

[0064] The module for constructing the alarm type dependency graph builds an alarm type dependency graph; it maintains a global alarm type dependency graph using expert experience rules, with nodes {Node...} as the basis for this graph. i}, i = 1, ..., N represent alarm types, where N is the number of nodes, i.e., the number of alarm types; with directed edges<A,B> This indicates the triggering relationship from node A to node B (two alarm types or system components).

[0065] The module for constructing a topology-based alarm propagation relationship diagram constructs a topology-based alarm propagation relationship diagram based on the alarm type dependency relationship diagram and the network element device resource topology diagram.

[0066] The network element device resource topology diagram describes the physical connections between devices. Based on the alarm type dependency graph, an alarm propagation (sub-graph) of the fault device alarm type set is constructed on the network element device resource topology diagram. The alarm propagation graph is a set of network element device nodes with a radius ≤ R centered on the fault device. Each alarm is traversed at least once within R steps; if it exceeds R, it is considered that the fault type of the network element device is unrelated to the current fault type. The default value of R is 2, and it may change later based on the dependency relationship.

[0067] The fault training dataset module connects fault work order data and alarm propagation relationship graph, extracts fault alarm data and work order processing conclusion data, and constructs a fault training dataset; wherein the alarm information of each node in the fault work order data and alarm propagation relationship graph is connected by depth-first / breadth-first traversal to extract fault alarm data and work order processing conclusion data.

[0068] The neural network model training module takes the fault training dataset as input, uses a three-layer neural network for model training, and uses precision, recall, and F-score for model evaluation.

[0069] The three-layer neural network includes an input layer, a hidden layer, and an output layer. It is a multi-layer feedforward network trained using an error backpropagation algorithm, and its structure is as follows: Figure 2 The diagram shown is of a BP neural network; the input layer takes fault alarm data as input, and the output layer outputs possible conclusions (device nodes) and the probability of each possible conclusion. The neural network has three layers.

[0070] The number of neurons in the input layer is related to the dimension of the fault alarm data, and the number of neurons in the output layer is related to the type of conclusion. The number of neurons is adjusted according to the business situation. The data is randomly disordered, and 60% of the data is extracted as training data, 20% as validation data, and 20% as test data. The neural network is trained using the above data. The model is evaluated by using precision, recall, and F-score to complete the model training.

[0071] The training model optimization and iteration module continuously acquires fault data and work order data to enrich the alarm type dependency graph and uses transfer learning to continuously optimize the model accuracy.

[0072] By using fault location models to accurately locate faults in communication networks, the reliance on expert experience in the fault location process is reduced, making it a powerful support for daily network fault maintenance.

[0073] The implementation of the neural network model training includes data acquisition, data storage, model training, and model management. The neural network model is directly invoked through the model call API interface for fault root cause localization.

[0074] Through the specific embodiments described above, those skilled in the art can easily implement the present invention. However, it should be understood that the present invention is not limited to the specific embodiments described above. Based on the disclosed embodiments, those skilled in the art can arbitrarily combine different technical features to achieve different technical solutions.

[0075] Except for the technical features described in the specification, all other technologies are known to those skilled in the art.

Claims

1. A method for fault location in communication networks based on fault propagation relationships, characterized in that, The alarms at the moment of failure are represented by a propagation diagram. The root cause device and root cause alarm are found from the propagation diagram to complete the location of the root cause of the failure. The model is trained and corrected using fault handling work order data, and continuously iterated, updated, and optimized. The implementation of this method includes: Construct an alarm type dependency graph; Construct a topology-based alarm propagation relationship graph based on alarm type dependency graph and network element device resource topology graph; By connecting the fault work order data and the alarm propagation relationship diagram, the fault alarm data and work order processing conclusion data are extracted to construct a fault training dataset; Three-layer neural network model training: The fault training dataset is used as input to train the model using a three-layer neural network, and the model is evaluated using precision, recall, and F-score. The training model is optimized and iterated. By continuously acquiring fault data and work order data, the dependency graph of alarm types is constantly enriched. Transfer learning is used to continuously optimize the model accuracy. The alarm type dependency graph is constructed, with nodes in the graph. Indicated as alarm type, where It refers to the number of nodes, i.e., the number of alarm types; with directed edges This indicates the triggering relationship from node A to node B; The aforementioned construction of a topology-based alarm propagation relationship graph, The network element device resource topology diagram describes the physical connection relationships between devices. Based on the alarm type dependency relationship diagram, an alarm propagation diagram of the fault device alarm type set is constructed on the network element device resource topology diagram. The alarm propagation relationship diagram is a set of network element device nodes with a radius ≤ R centered on the fault device. Each alarm is traversed at least once within R steps. If the number of steps exceeds R, it is considered that the fault type of the network element device is unrelated to the current fault type. The fault work order data and alarm information of each node in the alarm propagation relationship diagram are traversed and concatenated in a depth-first / breadth-first manner to extract the fault alarm data and work order processing conclusion data. The three-layer neural network includes an input layer, a hidden layer, and an output layer. It is a multi-layer feedforward network trained using the error backpropagation algorithm. The input data of the input layer is fault alarm data, and the output layer outputs possible conclusions and the probability of each possible conclusion. The neural network has three layers. The number of neurons in the input layer is related to the dimension of the fault alarm data, and the number of neurons in the output layer is related to the type of conclusion. The number of neurons is adjusted according to the business situation. The data is randomly disordered, and 60% of the data is extracted as training data, 20% as validation data, and 20% as test data. The neural network is trained using the above data. The model is evaluated by using precision, recall, and F-score to complete the model training.

2. A communication network fault location method based on fault propagation relationship according to any one of claims 1, characterized in that, The default value for R is 2.

3. The communication network fault location method based on fault propagation relationship according to claim 1, characterized in that, The training model is optimized and iterated to accurately locate faults in the communication network through the fault location model.

4. The communication network fault location method based on fault propagation relationship according to claim 3, characterized in that, The implementation of the neural network model training includes data acquisition, data storage, model training, and model management. The data acquisition process involves acquiring labeled training data and real data. The data storage corresponds to labeled training data and real data, respectively forming a training dataset and a prediction dataset; The model training involves training the neural network model and iteratively optimizing the training model based on the training dataset; and combining the prediction dataset to form a multi-classification model. The model management includes model loading, model deployment, and model prediction.

5. The communication network fault location method based on fault propagation relationship according to claim 1, characterized in that, The neural network model can be directly invoked through the model call API interface to locate the root cause of the fault.

6. A communication network fault location system based on fault propagation relationships, characterized in that, This includes modules for constructing alarm type dependency graphs, topology-based alarm propagation graphs, fault training datasets, neural network model training, and optimization of the trained models. The module for constructing an alarm type dependency graph constructs an alarm type dependency graph. The module for constructing a topology-based alarm propagation graph constructs a topology-based alarm propagation graph based on the alarm type dependency graph and the network element device resource topology graph. The fault training dataset module connects fault work order data and alarm propagation relationship diagram, extracts fault alarm data and work order processing conclusion data, and constructs a fault training dataset. The neural network model training module takes the fault training dataset as input, uses a three-layer neural network for model training, and uses precision, recall, and F-score for model evaluation. The training model optimization and iteration module continuously acquires fault data and work order data to enrich the alarm type dependency graph and uses transfer learning to continuously optimize the model accuracy. The system achieves fault root cause localization using the fault propagation relationship-based fault root cause localization method as described in any one of claims 1-5.

Images