A fault analysis method and system for 5G base station transmission links

Abstract

The present invention discloses a fault analysis method and system applied to 5G base station transmission links, wherein the method includes the following steps: obtaining alarm information of 5G base station transmission links by integrating real-time data stream processing mechanism and incremental learning model; preprocessing the alarm information; converting the preprocessed alarm information into a multi-dimensional vector representation based on time dimension, topology dimension and text dimension; performing group aggregation in combination with DBSCAN algorithm; performing deep suspected root cause analysis on the alarm information after group aggregation by Incident root cause analysis algorithm to determine the suspected root cause alarm; performing fault diagnosis by knowledge graph and diagnostic rules, generating a repair process and recording the whole process. It can significantly improve the accuracy and speed of fault location, enhance the intelligence and automation level of fault analysis, and at the same time ensure the real-time and flexibility of fault handling, thereby optimizing the operation and maintenance management of 5G networks and providing solid guarantee for the stability and reliability of communication services.

Description

Technical Field

[0001] The present invention relates to the field of 5G communication networks, and more specifically to a fault analysis method and system applied to 5G base station transmission links. Background Art

[0002] In 5G communication networks, base stations serve as key nodes connecting user devices and the core network, and the stability and reliability of their transmission links are crucial. With the large-scale deployment of 5G networks and the diversification of application scenarios, data traffic on base station transmission links is growing exponentially, placing higher demands on fault detection, diagnosis, and repair.

[0003] From the perspective of fault location accuracy and efficiency, traditional fault analysis methods mostly rely on manual experience and a step-by-step troubleshooting process. This process is not only time-consuming but also susceptible to subjectivity and experience limitations, leading to inaccurate fault location or missing key information. It is often difficult to quickly identify and accurately locate the fault point, which prolongs troubleshooting and recovery time, affecting the continuity and stability of network services.

[0004] Secondly, existing technologies also have significant shortcomings in the intelligence and automation of fault analysis. Despite the rapid development of artificial intelligence and machine learning technologies in recent years, achieving significant results in various fields, their application in 5G base station fault analysis and processing still faces numerous challenges. On the one hand, the complexity of algorithmic models and the acquisition and annotation of training data limit the intelligence level of existing technologies. On the other hand, the consumption of computing resources and the optimization of algorithmic models also affect the degree of automation of fault analysis. These factors combined result in existing technologies still requiring a high degree of manual intervention and judgment during the fault analysis process, reducing the efficiency and accuracy of fault handling.

[0005] Finally, from the perspective of real-time and flexibility in troubleshooting, traditional troubleshooting methods typically follow a fixed series of steps, including fault information collection, analysis, location, repair, and verification. The connection and transition between these steps are often not smooth and efficient, resulting in a lengthy and inflexible troubleshooting process.

[0006] Therefore, how to design a fault analysis method for 5G base station transmission links, improve the accuracy and efficiency of fault location, and enhance the intelligence and automation of fault analysis is an urgent problem that technicians in this field need to solve. Summary of the Invention

[0007] In view of this, the present invention provides a fault analysis method applied to 5G base station transmission links. By integrating real-time data stream processing, incremental learning, multi-dimensional vector construction and clustering, root cause analysis algorithm and knowledge graph, it realizes the rapid location, diagnosis and repair of faults. Compared with the existing technology, it significantly improves the real-time, intelligence, accuracy and operability of fault handling.

[0008] In order to achieve the above object, the present invention adopts the following technical solutions:

[0009] In a first aspect, the present invention provides a fault analysis method applied to a 5G base station transmission link, comprising the following steps:

[0010] S1. Obtain alarm information of 5G base station transmission links by integrating real-time data stream processing mechanism and incremental learning model;

[0011] S2. Preprocessing the alarm information to obtain preprocessed alarm information;

[0012] S3, based on the time dimension, topology dimension and text dimension, convert the pre-processed warning signal into a multi-dimensional vector representation;

[0013] S4, based on multi-dimensional vector representation, combined with DBSCAN algorithm for grouping aggregation;

[0014] S5. Use the Incident root cause analysis algorithm to perform in-depth suspected root cause analysis on the grouped and aggregated alarm information to determine the suspected root cause alarm;

[0015] S6. Use knowledge graphs and diagnostic rules to diagnose suspected root cause alarms, generate repair processes, and record the entire process.

[0016] Preferably, the S1 includes:

[0017] Receive alarm information from each node on the 5G base station transmission link through the Kafka message queue model; wherein, multiple consumer groups are set in the Kafka message queue model, and each consumer group processes corresponding types of alarm information;

[0018] Combined with online learning algorithms, an incremental learning model is constructed to dynamically adjust internal parameters of alarm information received based on the Kafka message queue model;

[0019] Apply timestamps to dynamically adjusted alarm information, record the exact time each alarm message occurs and when it arrives at the system, and classify and store the alarm information with timestamps by type.

[0020] Preferably, the dynamic adjustment of internal parameters of the alarm information received based on the Kafka message queue model includes:

[0021] Monitor the alarm information flow of the Kafka message queue in real time and dynamically adjust the number of consumer groups based on traffic changes; also, analyze the type distribution of alarm information and dynamically adjust the number of consumer instances in each consumer group based on the number of different types of alarm information.

[0022] Preferably, said S2 includes:

[0023] Read alarm information with timestamps and perform standardization and normalization;

[0024] Based on the standardized and normalized alarm information, the pre-trained NLP model is used to parse the text alarm content, extract key information, and convert it into structured data;

[0025] Based on structured data, construct alarm time series and network topology diagrams.

[0026] Preferably, the S3 includes:

[0027] Converting the timestamp into a relative time value relative to a reference time, and normalizing the relative time value to a preset interval to form a time dimension vector;

[0028] Determine the network node location corresponding to the alarm information using a network topology diagram, and encode the network node location into a topology dimension vector;

[0029] Parse the text alert content, extract key information, and convert it into word vectors or sentence vectors to form a text dimension vector;

[0030] The time dimension vector, topology dimension vector and text dimension vector are combined, and linear dimensionality reduction is performed in conjunction with the PCA algorithm to generate a multi-dimensional vector representation of the alarm information.

[0031] Preferably, the generating of the multi-dimensional vector representation of the alarm information further comprises: performing vector similarity calculation; the vector similarity calculation comprises:

[0032] Calculate the cosine similarity between the multidimensional vector representation of each alarm information and the multidimensional vector representation of all other alarm information;

[0033] Based on a preset similarity threshold, when the cosine similarity calculation result of any two alarm messages exceeds the threshold, the two alarm messages are determined to be duplicate or highly similar;

[0034] After duplicate or highly similar alarm information is identified, one of the duplicate or highly similar alarm information is selected as the representative alarm information, and the remaining duplicate or highly similar alarm information is deduplicated.

[0035] Preferably, the S4 includes:

[0036] loading the multidimensional vector representation;

[0037] Based on the distribution characteristics of alarm information and clustering requirements, the neighborhood size in the DBSCAN algorithm and the minimum number of points required to form a dense area are set;

[0038] Conduct cluster analysis and obtain cluster analysis results;

[0039] Based on the cluster analysis results, the alarm information belonging to the same group is merged into one alarm group, and a unique identifier is assigned to each alarm group.

[0040] Preferably, the S5 includes:

[0041] Load the alarm group and its identifier into the Incident root cause analysis algorithm;

[0042] Extract key features related to root cause analysis from each alarm and convert them into numerical vectors; the key features include: alarm type, severity, occurrence time, impact range, and associated network elements;

[0043] Combined with the pre-trained gradient boosting tree model, root cause analysis is performed to predict the suspected root cause alarms for each alarm group and generate a ranked list of suspected root cause alarms.

[0044] Based on the ranked list of suspected root cause alarms, a final suspected root cause alarm and a corresponding alarm group identifier are determined.

[0045] Preferably, the S6 includes:

[0046] Load the knowledge graph and diagnostic rule base; the knowledge graph includes base station equipment, transmission links, alarm types, and their relationships in the 5G base station transmission link; the diagnostic rule base contains preset rules and logical reasoning rules for different alarm types, equipment status, and historical fault handling experience;

[0047] Match suspected root cause alarm information with the nodes and relationships in the loaded knowledge graph, and analyze associated nodes and relationships to determine possible fault propagation paths and impact ranges;

[0048] Combined with the diagnostic rule base, logical reasoning is performed to analyze the relevance, time sequence, and equipment status of alarm information, predict the cause of the fault, and generate a preliminary fault diagnosis report;

[0049] Based on the fault diagnosis report and pre-defined repair strategies, a repair process for the fault is generated;

[0050] Perform the corresponding repair operations, record the operation results, timestamps and possible abnormal conditions of each step in real time, and obtain a complete fault handling log.

[0051] In a second aspect, the present invention provides a fault analysis system for a 5G base station transmission link, comprising:

[0052] Data acquisition module: used to obtain alarm information of 5G base station transmission links by integrating real-time data stream processing mechanism and incremental learning model;

[0053] Preprocessing module: used to preprocess the alarm information to obtain preprocessed alarm information;

[0054] Multidimensional vector representation generation module: used to convert pre-processed alarm signals into multidimensional vector representations based on time, topology, and text dimensions;

[0055] Grouping aggregation module: used for grouping aggregation based on multidimensional vector representation combined with DBSCAN algorithm;

[0056] Root cause analysis module: Uses the incident root cause analysis algorithm to conduct in-depth suspected root cause analysis on grouped and aggregated alarm information to identify suspected root cause alarms.

[0057] Fault diagnosis and repair module: Used to diagnose suspected root cause alarms through knowledge graphs and diagnostic rules, generate repair processes, and record the entire process.

[0058] It can be seen from the above technical solution that compared with the prior art, the technical solution of the present invention has the following advantages:

[0059] Beneficial effects:

[0060] 1. Utilizing the Kafka message queue model and online learning algorithms, internal parameters can be dynamically adjusted to accommodate changes in different types of alarm information traffic, ensuring the system maintains efficient processing capabilities and responsiveness even when dealing with large amounts of data. Furthermore, through timestamp tagging and categorized storage, the temporal attributes of alarm information are preserved, helping to more accurately locate the time of fault occurrence and thus improving the accuracy of fault analysis.

[0061] 2. By converting alarm information into multi-dimensional vectors and combining PCA dimensionality reduction technology with vector similarity calculation, we enhance correlation analysis between alarms, effectively reduce data redundancy, and improve the efficiency and quality of cluster analysis. In particular, we remove duplicates from repeated or highly similar alarms, avoiding unnecessary resource waste and improving the speed and accuracy of fault diagnosis.

[0062] 3. Leveraging the knowledge graph and diagnostic rule base, we can deeply explore the complex relationships behind alerts, predict possible fault propagation paths and their impact, and generate a more scientific and rational repair process. The entire process emphasizes automation and intelligence, optimizing every step from alert reception to fault repair, ensuring timely and effective troubleshooting and improving the efficiency and reliability of operations and maintenance management. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0064] Figure 1 A flowchart of a fault analysis method for a 5G base station transmission link provided by an embodiment of the present invention;

[0065] Figure 2 A schematic diagram of a specific implementation process of the fault analysis method provided in an embodiment of the present invention;

[0066] Figure 3 A structural framework diagram of a fault analysis system for 5G base station transmission links provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0067] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0068] Example 1;

[0069] like Figure 1 As shown, this embodiment provides a fault analysis method applied to a 5G base station transmission link, including the following steps:

[0070] S1. Obtain alarm information of 5G base station transmission links by integrating real-time data stream processing mechanism and incremental learning model;

[0071] S2. Preprocessing the alarm information to obtain preprocessed alarm information;

[0072] S3, based on the time dimension, topology dimension and text dimension, convert the pre-processed warning signal into a multi-dimensional vector representation;

[0073] S4, based on multi-dimensional vector representation, combined with DBSCAN algorithm for grouping aggregation;

[0074] S5. Use the Incident root cause analysis algorithm to perform in-depth suspected root cause analysis on the grouped and aggregated alarm information to determine the suspected root cause alarm;

[0075] S6. Use knowledge graphs and diagnostic rules to diagnose suspected root cause alarms, generate repair processes, and record the entire process.

[0076] This method acquires alarm information by integrating a real-time data stream processing mechanism with an incremental learning model. After preprocessing, multidimensional vector representation conversion, group aggregation based on the DBSCAN algorithm, deep suspected root cause analysis, and knowledge graph-assisted fault diagnosis, it ultimately generates a repair process and documents the entire process. This method significantly improves the accuracy and speed of fault location, reduces reliance on manual intervention, enhances the automation and intelligence of fault handling, ensures the continuity and stability of network services, and provides an efficient, intelligent, and flexible operation and maintenance management solution for 5G base station transmission links.

[0077] The following is a further detailed description of each of the above steps and related technical features:

[0078] In this embodiment S1, alarm information of a 5G base station transmission link is obtained by integrating a real-time data stream processing mechanism and an incremental learning model. Specifically, the following steps are performed:

[0079] Alarm information from each node on the 5G base station transmission link is received through the Kafka message queue model. Multiple consumer groups are set up in the Kafka message queue model, and each consumer group processes corresponding types of alarm information. This design not only improves processing efficiency, but also ensures isolation between different types of alarm information, avoiding mutual interference.

[0080] An incremental learning model is constructed in conjunction with an online learning algorithm to dynamically adjust internal parameters of alarm information received based on the Kafka message queue model. This includes: real-time monitoring of the alarm information flow of the Kafka message queue and dynamic adjustment of the number of consumer groups based on traffic changes; for example, increasing the number of consumer groups during peak traffic periods to distribute the load, and reducing the number of consumer groups during low traffic periods to save resources.

[0081] Furthermore, the system analyzes the distribution of alarm types and dynamically adjusts the number of consumer instances within each consumer group based on the number of different alarm types. If a certain type of alarm suddenly increases, the system automatically increases the number of corresponding consumer instances to speed up processing; otherwise, it reduces the number of instances to avoid wasting resources.

[0082] Apply timestamps to dynamically adjusted alarm information, record the exact time each alarm message occurs and when it arrives at the system, and classify and store the alarm information with timestamps by type.

[0083] This step enables efficient collection and preliminary processing of 5G base station transmission link alarm information, laying a solid data foundation for subsequent fault analysis and diagnosis. This process emphasizes real-time, automated, and intelligent features, aiming to improve the speed and accuracy of fault handling while reducing operational costs and complexity. In particular, the combination of the Kafka message queue model and incremental learning ensures the system's ability to flexibly respond to changes in alarm information traffic of varying types. Dynamically adjusting the number of consumer groups and consumer instances further optimizes system performance and resource utilization.

[0084] In this embodiment, S2 pre-processes the alarm information to obtain pre-processed alarm information, which specifically includes:

[0085] Read alarm information with timestamps and perform standardization and normalization. Standardization converts data to zero mean and unit variance, while normalization scales the data to a specific interval (such as [0, 1]). This eliminates dimensionality effects, enables comparison of different types of alarm information on the same scale, and improves the accuracy of data analysis.

[0086] Based on standardized and normalized alarm information, the pre-trained NLP model parses the text alarm content, extracts key information and converts it into structured data. After being trained on a large amount of text data, the NLP model can effectively understand and extract key elements in the alarm information, such as the fault type, affected equipment or components, and possible causes.

[0087] Based on structured data, we construct alarm time series and network topology diagrams. Time series analysis helps identify patterns and trends in alarm occurrence, which is crucial for predicting future failures. The network topology diagram not only shows how network nodes are connected but also provides information related to the alarm, such as its specific location and impact range.

[0088] The above preprocessing steps improve the quality and consistency of alarm information and lay a solid foundation for subsequent multi-dimensional analysis and fault diagnosis. In particular, the use of NLP models to parse text alarm content enables the conversion of unstructured to structured data, greatly facilitating further data processing and mining.

[0089] In this embodiment S3, based on the time dimension, topology dimension and text dimension, the pre-processed warning signal is converted into a multi-dimensional vector representation; specifically, the following steps are performed:

[0090] The timestamp is converted into a relative time value relative to a reference time, and the relative time value is normalized to a preset interval to form a time dimension vector; the time dimension vector is generated by converting the timestamp in the alarm information into a relative time value relative to a reference time (such as the system startup time or a fixed reference point), and then normalizing this relative time value to a preset interval to ensure that the time attributes of different alarms are compared on the same scale.

[0091] The network topology diagram is used to determine the network node locations corresponding to the alarm information, and these network node locations are encoded as topology dimension vectors. The topology dimension vectors are used to determine the network node locations associated with each alarm information, and these locations are encoded into vector form. The encoding method can be based on the specific location, connectivity, or other topological characteristics of the network nodes in the network structure, using one-hot encoding or embedded encoding.

[0092] Parse the text alarm content, extract key information and convert it into word vectors or sentence vectors to form a text dimension vector; the generated text dimension vector can quantitatively describe the specific content and nature of the alarm.

[0093] The time dimension vector, topology dimension vector, and text dimension vector are combined and linearly reduced using the PCA algorithm to generate a multidimensional vector representation of the alarm information. To reduce data redundancy and improve computational efficiency, the principal component analysis (PCA) algorithm is used to perform linear dimensionality reduction on the multidimensional vector. PCA identifies the main directions of change in the data, retaining the most important feature information while reducing dimensionality and simplifying the complexity of subsequent analysis.

[0094] Furthermore, generating a multi-dimensional vector representation of the alarm information further includes: performing vector similarity calculation; the vector similarity calculation includes:

[0095] Calculate the cosine similarity between the multidimensional vector representation of each alarm information and the multidimensional vector representation of all other alarm information;

[0096] Based on a preset similarity threshold, when the cosine similarity calculation result of any two alarm messages exceeds the threshold, the two alarm messages are determined to be duplicate or highly similar;

[0097] After duplicate or highly similar alarm information is identified, one of the duplicate or highly similar alarm information is selected as the representative alarm information, and the remaining duplicate or highly similar alarm information is deduplicated.

[0098] This step combines information from three dimensions: time, topology, and text, making alarm analysis more comprehensive and in-depth. It also enhances the intelligence level of fault diagnosis and lays a solid foundation for subsequent advanced functions such as cluster analysis and root cause analysis.

[0099] In this embodiment S4, grouping and aggregation are performed based on multidimensional vector representation and combined with the DBSCAN algorithm; specifically, the following steps are performed:

[0100] loading the multidimensional vector representation;

[0101] Based on the distribution characteristics of alarm information and clustering requirements, the neighborhood size and the minimum number of points required to form a dense area in the DBSCAN algorithm are set. The neighborhood size determines the local density range considered by the algorithm during clustering, while the minimum number of points determines whether an area is dense enough to form a cluster. The setting of these two parameters requires a comprehensive consideration of the actual alarm information and the clustering objectives to ensure the accuracy and effectiveness of the clustering results.

[0102] Perform cluster analysis to obtain cluster analysis results. The DBSCAN algorithm clusters closely connected points in space, while points far from any dense areas are labeled as noise. After the cluster analysis is complete, the system generates cluster results, including the cluster label for each alarm. If an alarm does not belong to any cluster, it is labeled as a noise point.

[0103] Based on cluster analysis results, alarms belonging to the same group are grouped together and a unique identifier is assigned to each alarm group. Alarms that are temporally and spatially adjacent and semantically similar are aggregated, simplifying fault analysis. Each alarm group is assigned a unique identifier for subsequent tracking and management, facilitating the rapid location and handling of specific alarm sets.

[0104] It not only effectively groups and aggregates alarm information, but also improves the accuracy and efficiency of fault analysis. In particular, the combination of multidimensional vector representation and the DBSCAN algorithm can automatically identify alarm clusters with similar characteristics, helping operations personnel identify potential problems more quickly and improving the speed and accuracy of fault handling.

[0105] In this embodiment, S5 uses the Incident root cause analysis algorithm to perform in-depth suspected root cause analysis on the grouped and aggregated alarm information to determine the suspected root cause alarm, specifically including:

[0106] Load the alarm group and its identifier into the Incident root cause analysis algorithm;

[0107] Extract key features related to root cause analysis from each alarm and convert them into numerical vectors; the key features include: alarm type, severity, occurrence time, impact scope, and associated network elements; specifically, alarm type: describes the specific category or nature of the fault. Severity: reflects the importance and urgency of the alarm and is usually divided into different levels. Occurrence time: records the exact time of occurrence of the alarm for time series analysis. Impact scope: assesses the scope of the impact of the alarm on the network and services, such as how many users or services are affected. Associated network elements: identify the network devices or components involved and understand the possible propagation paths of the fault.

[0108] A pre-trained gradient boosting tree model is used to perform root cause analysis, predicting suspected root cause alarms for each alarm group and generating a ranked list of suspected root cause alarms. The gradient boosting tree model effectively handles nonlinear relationships and complex feature interactions, making it suitable for classification and regression tasks. For each alarm group, the gradient boosting tree model is used to predict suspected root cause alarms and generate a ranked list containing each alarm's likelihood score. A higher score indicates a greater likelihood that the alarm is the root cause.

[0109] Based on the ranked list of suspected root cause alarms, a final suspected root cause alarm and a corresponding alarm group identifier are determined.

[0110] This step enables in-depth suspected root cause analysis of grouped and aggregated alarm information, significantly improving the accuracy and efficiency of fault diagnosis. In particular, the combination of the Incident Root Cause Analysis algorithm and the pre-trained Gradient Boosting Tree model enables the system to automatically identify the most likely root cause alarm, reducing the need for manual intervention and improving the automation level of operations and maintenance management.

[0111] In this embodiment S6, the suspected root cause alarm is diagnosed through the knowledge graph and diagnostic rules, a repair process is generated, and the entire process is recorded, which specifically includes:

[0112] Load the knowledge graph and diagnostic rule base; the knowledge graph includes base station equipment, transmission links, alarm types, and their relationships in the 5G base station transmission link; the diagnostic rule base contains preset rules and logical reasoning rules for different alarm types, equipment status, and historical fault handling experience;

[0113] Match suspected root cause alarm information with the nodes and relationships in the loaded knowledge graph, and analyze associated nodes and relationships to determine possible fault propagation paths and impact ranges;

[0114] Combined with the diagnostic rule base, logical reasoning is performed to analyze the relevance, time sequence, and equipment status of alarm information, predict the cause of the fault, and generate a preliminary fault diagnosis report;

[0115] Based on the fault diagnosis report and pre-defined repair strategies, a repair process for the fault is generated;

[0116] Perform the corresponding repair operations, record the operation results, timestamps and possible abnormal conditions of each step in real time, and obtain a complete fault handling log.

[0117] This step, combined with the application of knowledge graphs and diagnostic rule bases, can intelligently analyze the root cause of the fault and provide scientific and reasonable repair solutions. This enhances the automation and intelligence level of fault handling, improves the efficiency and reliability of operation and maintenance management, and ensures the stable operation of 5G base station transmission links.

[0118] This embodiment provides a fault analysis method for 5G base station transmission links. It obtains alarm information by integrating a real-time data stream processing mechanism and an incremental learning model. It then undergoes preprocessing, multidimensional vector representation conversion, group aggregation based on the DBSCAN algorithm, deep suspected root cause analysis, and knowledge graph-assisted fault diagnosis to ultimately generate a repair process and record the entire process. This method significantly improves the accuracy and speed of fault location, reduces reliance on manual intervention, enhances the automation and intelligence of fault handling, ensures the continuity and stability of network services, and provides an efficient, intelligent, and flexible operation and maintenance management solution for 5G base station transmission links.

[0119] Example 2;

[0120] A carrier operates a 5G network covering a wide area, comprising multiple base stations. Due to the network's large scale and dense traffic, various types of alarms frequently appear on base station transmission links. To ensure network stability and service quality, an efficient fault analysis method is required to quickly locate and fix problems.

[0121] like Figure 2 As shown, the specific implementation steps include:

[0122] 1) Obtain alarm information;

[0123] Receive alarm information from each base station node in real time through the Kafka message queue model. Each alarm message contains the following data types:

[0124] Timestamp: a time string in ISO 8601 format; Alarm ID: a unique string identifier; Alarm type: an enumeration type, specifically "transmission link interruption"; Severity: an integer ranging from 1 to 5, where 5 is the most severe; Device ID: a string identifying a specific base station device; Text description: a string describing the alarm content in detail;

[0125] In Kafka, a consumer group is set up to handle transmission link related alarms, specifically the transmission link alarm group.

[0126] The traffic of the Kafka message queue is monitored in real time. Consumer groups are added during peak hours to distribute the load and resource consumption is reduced during off-peak hours. Specific implementations include: using Kafka's built-in monitoring tools to check the depth of the message queue every minute. If the message queue depth exceeds the set threshold (such as 1,000 messages), a consumer group is added; if it is below the threshold for 10 consecutive minutes, a consumer group is removed. Each alarm message is timestamped and stored by type. Alarm information is stored in different tables in the distributed database Cassandra, each corresponding to a different alarm type.

[0127] 2) Pre-processing alarm information;

[0128] Convert the timestamp from ISO 8601 format to Unix timestamp (seconds), normalize the severity, and scale it to the range [0, 1]. Use the BERT model to parse the text description, extract key information such as the fault type and affected devices, and convert it into structured data. Specifically, extract {"type":"transmission link","status":"interruption"} from "transmission link interruption."

[0129] Based on structured data, we build alarm time series and network topology diagrams to help understand the alarm occurrence patterns and potential impact scope.

[0130] 3) Multidimensional vector representation;

[0131] Time dimension vector: Converts Unix timestamps to relative time values ​​relative to a base time and normalizes them to the [0, 1] range. Topology dimension vector: Uses a network topology diagram to determine the location of the network node corresponding to the alarm and encodes it as a one-hot vector or embedded vector. Text dimension vector: Parses the text alarm content, extracts key information, and converts it into a word vector or sentence vector.

[0132] The principal component analysis algorithm is combined to perform linear dimensionality reduction on the above multidimensional vectors to simplify the data complexity and retain the first 100 principal components.

[0133] When calculating the cosine similarity between alarm messages, duplicate or highly similar alarms are identified and deduplication is performed. Specifically, a similarity threshold is set to 0.9. When the cosine similarity of any two alarm messages exceeds this threshold, they are considered duplicate alarms.

[0134] 4) Group aggregation;

[0135] Load the multidimensional vector representation obtained in step 3) of multidimensional vector representation. Based on the alarm distribution characteristics and clustering requirements, set the neighborhood size in the DBSCAN algorithm to 0.5 and the minimum number of points to 5. Perform DBSCAN cluster analysis to obtain clustering results. Alarms belonging to the same group are merged into one alarm group and assigned the unique identifier GROUP_001.

[0136] 5) Root cause analysis;

[0137] The alarm groups and their identifiers generated in the grouping and aggregation step (4) are loaded into the Incident root cause analysis algorithm. Key features (such as alarm type, severity, and occurrence time) are extracted from each alarm and converted into numerical vectors.

[0138] Combined with the pre-trained gradient boosting tree model, root cause analysis is performed to generate a ranked list of suspected root cause alarms, and ultimately determine the most likely root cause alarm and its corresponding alarm group identifier.

[0139] 6) Fault diagnosis and repair;

[0140] Load a knowledge graph containing base station equipment, transmission links, alarm types, and their relationships in 5G base station transmission links, as well as a diagnostic rule base for different alarm types. For example, the knowledge graph records the connection relationships between base station equipment, and the diagnostic rule base contains specific repair strategies for different alarm types.

[0141] Match suspected root cause alarm information with nodes and relationships in the knowledge graph to analyze possible fault propagation paths and impact ranges; perform logical reasoning based on the diagnostic rule base to generate a preliminary fault diagnosis report.

[0142] Furthermore, based on the diagnostic report and pre-defined repair strategies, a specific repair process is generated. For the fault type of "transmission link interruption", the repair process may include the following steps:

[0143] Check the physical connection: Dispatch a technician to the affected base station to check for loose or damaged fiber connections. Test signal strength: Use an optical power meter to measure the signal strength at both ends of the fiber to confirm any attenuation. Replace damaged fiber: If any fiber is significantly damaged, immediately replace the damaged section. Restart related devices: Restart devices involved in the transmission link to ensure correct configuration. Verify repair results: Retest the transmission link performance to ensure normal operation.

[0144] Finally, when performing repair operations, the operation results, timestamps and abnormal conditions of each step are recorded in real time to form a complete fault handling log.

[0145] By applying these technical solutions, especially for specific alarm types like "transmission link interruption," operators can achieve efficient and accurate fault analysis and rapid repairs. This process not only reduces the need for manual intervention and improves O&M efficiency, but also ensures the continuity and stability of network services.

[0146] Example 3;

[0147] like Figure 3 As shown, this embodiment provides a fault analysis system applied to a 5G base station transmission link, including:

[0148] Data acquisition module: used to obtain alarm information of 5G base station transmission links by integrating real-time data stream processing mechanism and incremental learning model;

[0149] Preprocessing module: used to preprocess the alarm information to obtain preprocessed alarm information;

[0150] Multidimensional vector representation generation module: used to convert pre-processed alarm signals into multidimensional vector representations based on time, topology, and text dimensions;

[0151] Grouping aggregation module: used for grouping aggregation based on multidimensional vector representation combined with DBSCAN algorithm;

[0152] Root cause analysis module: Uses the incident root cause analysis algorithm to conduct in-depth suspected root cause analysis on grouped and aggregated alarm information to identify suspected root cause alarms.

[0153] Fault diagnosis and repair module: Used to diagnose suspected root cause alarms through knowledge graphs and diagnostic rules, generate repair processes, and record the entire process.

[0154] This embodiment provides a fault analysis system for 5G base station transmission links, which includes a data acquisition module, a preprocessing module, a multi-dimensional vector representation generation module, a group aggregation module, a root cause analysis module, and a fault diagnosis and repair module. It obtains alarm information by integrating a real-time data stream processing mechanism and an incremental learning model, and after preprocessing, multi-dimensional vector representation conversion, group aggregation based on the DBSCAN algorithm, deep suspected root cause analysis, and knowledge graph-assisted fault diagnosis, it finally generates a repair process and records the entire process. The system significantly improves the accuracy and speed of fault location, reduces the need for manual intervention, enhances the automation and intelligence level of fault handling, and ensures the stable operation and service quality of the 5G base station transmission link.

[0155] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. References to the same or similar parts between the various embodiments are sufficient. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple. For relevant parts, refer to the method description.

[0156] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A fault analysis method applied to a 5G base station transmission link, characterized in that: The following steps are involved: S1. Acquire alarm information for 5G base station transmission links by integrating a real-time data stream processing mechanism and an incremental learning model. This includes: receiving alarm information from each node on the 5G base station transmission link through a Kafka message queue model; setting up multiple consumer groups in the Kafka message queue model, with each consumer group processing corresponding types of alarm information; constructing an incremental learning model using an online learning algorithm to dynamically adjust internal parameters of alarm information received based on the Kafka message queue model; applying a timestamp to the dynamically adjusted alarm information, recording the exact time each alarm information occurred and when it arrived at the system, and categorizing and storing the alarm information with timestamps by type. S2. Preprocessing the alarm information to obtain preprocessed alarm information; S3. Based on the time dimension, topology dimension and text dimension, the pre-processed alarm information is converted into a multi-dimensional vector representation; including: converting the timestamp into a relative time value relative to the reference time, and normalizing the relative time value to a preset interval to form a time dimension vector; using the network topology relationship diagram to determine the network node position corresponding to the alarm information, and encoding the network node position into a topology dimension vector, including based on the specific position, connectivity and related topological features of the network node in the network structure, using One-Hot encoding or embedded encoding to achieve; parsing the text alarm content, extracting key information and converting it into a word vector or sentence vector to form a text dimension vector; combining the time dimension vector, topology dimension vector and text dimension vector, and combining the PCA algorithm to perform linear dimensionality reduction to generate a multi-dimensional vector representation of the alarm information; S4, based on multi-dimensional vector representation, combined with DBSCAN algorithm for grouping aggregation; S5. Use the Incident root cause analysis algorithm to perform in-depth suspected root cause analysis on the grouped and aggregated alarm information to determine the suspected root cause alarm; S6. Use knowledge graphs and diagnostic rules to diagnose suspected root cause alarms, generate repair processes, and record the entire process.

2. A fault analysis method applied to a 5G base station transmission link according to claim 1, characterized in that: The dynamic adjustment of internal parameters of the alarm information received based on the Kafka message queue model includes: Monitor the alarm information flow of the Kafka message queue in real time and dynamically adjust the number of consumer groups based on traffic changes; also, analyze the type distribution of alarm information and dynamically adjust the number of consumer instances in each consumer group based on the number of different types of alarm information.

3. A fault analysis method applied to a 5G base station transmission link according to claim 1, characterized in that: Said S2 comprises: Read alarm information with timestamps and perform standardization and normalization; Based on the standardized and normalized alarm information, the pre-trained NLP model is used to parse the text alarm content, extract key information, and convert it into structured data; Based on structured data, construct alarm time series and network topology diagrams.

4. A fault analysis method applied to a 5G base station transmission link according to claim 1, characterized in that: The generating of the multi-dimensional vector representation of the alarm information further includes: performing vector similarity calculation; the vector similarity calculation includes: Calculate the cosine similarity between the multidimensional vector representation of each alarm information and the multidimensional vector representation of all other alarm information; Based on a preset similarity threshold, when the cosine similarity calculation result of any two alarm messages exceeds the threshold, the two alarm messages are determined to be duplicate or highly similar; After duplicate or highly similar alarm information is identified, one of the duplicate or highly similar alarm information is selected as the representative alarm information, and the remaining duplicate or highly similar alarm information is deduplicated.

5. A fault analysis method applied to a 5G base station transmission link according to claim 1, characterized in that: Said S4 comprises: loading the multidimensional vector representation; Based on the distribution characteristics of alarm information and clustering requirements, the neighborhood size in the DBSCAN algorithm and the minimum number of points required to form a dense area are set; Conduct cluster analysis and obtain cluster analysis results; Based on the cluster analysis results, the alarm information belonging to the same group is merged into one alarm group, and a unique identifier is assigned to each alarm group.

6. A fault analysis method for a 5G base station transmission link according to claim 1, characterized in that: Said S5 comprises: Load the alarm group and its identifier into the Incident root cause analysis algorithm; Extract key features related to root cause analysis from each alarm and convert them into numerical vectors; the key features include: alarm type, severity, occurrence time, impact range, and associated network elements; Combined with the pre-trained gradient boosting tree model, root cause analysis is performed to predict the suspected root cause alarms for each alarm group and generate a ranked list of suspected root cause alarms. Based on the ranked list of suspected root cause alarms, a final suspected root cause alarm and a corresponding alarm group identifier are determined.

7. A fault analysis method for a 5G base station transmission link according to claim 1, characterized in that: Said S6 comprises: Load the knowledge graph and diagnostic rule base; the knowledge graph includes base station equipment, transmission links, alarm types, and their relationships in the 5G base station transmission link; the diagnostic rule base contains preset rules and logical reasoning rules for different alarm types, equipment status, and historical fault handling experience; Match suspected root cause alarm information with the nodes and relationships in the loaded knowledge graph, and analyze associated nodes and relationships to determine possible fault propagation paths and impact ranges; Combined with the diagnostic rule base, logical reasoning is performed to analyze the relevance, time sequence, and equipment status of alarm information, predict the cause of the fault, and generate a preliminary fault diagnosis report; Based on the fault diagnosis report and pre-defined repair strategies, a repair process for the fault is generated; Perform the corresponding repair operations, record the operation results, timestamps and possible abnormal conditions of each step in real time, and obtain a complete fault handling log.

8. A fault analysis system applied to 5G base station transmission links, characterized in that: include: Data acquisition module: used to obtain alarm information of 5G base station transmission links by integrating real-time data stream processing mechanism and incremental learning model. This module includes: receiving alarm information of each node on the 5G base station transmission link through Kafka message queue model; setting up multiple consumer groups in the Kafka message queue model, each consumer group processes corresponding types of alarm information; constructing incremental learning model based on online learning algorithm, dynamically adjusting internal parameters of alarm information received based on Kafka message queue model; applying timestamp to the dynamically adjusted alarm information, recording the exact time of occurrence and arrival of each alarm information in the system, and classifying and storing the alarm information with timestamp by type; Preprocessing module: used to preprocess the alarm information to obtain preprocessed alarm information; Multidimensional vector representation generation module: used to convert the pre-processed alarm information into a multidimensional vector representation based on the time dimension, topology dimension and text dimension; including: converting the timestamp into a relative time value relative to the reference time, and normalizing the relative time value to a preset interval to form a time dimension vector; using the network topology relationship diagram to determine the network node position corresponding to the alarm information, and encoding the network node position into a topology dimension vector, including based on the specific position, connectivity and related topological features of the network node in the network structure, using One-Hot encoding or embedded encoding to achieve; parsing the text alarm content, extracting key information and converting it into a word vector or sentence vector to form a text dimension vector; combining the time dimension vector, topology dimension vector and text dimension vector, and combining the PCA algorithm to perform linear dimensionality reduction to generate a multidimensional vector representation of the alarm information; Grouping aggregation module: used for grouping aggregation based on multidimensional vector representation combined with DBSCAN algorithm; Root cause analysis module: Uses the incident root cause analysis algorithm to conduct in-depth suspected root cause analysis on grouped and aggregated alarm information to identify suspected root cause alarms. Fault diagnosis and repair module: Used to diagnose suspected root cause alarms through knowledge graphs and diagnostic rules, generate repair processes, and record the entire process.

Images