Railway 5G-R network-oriented root alarm intelligent positioning method and system
By collecting and enhancing data for railway 5G-R networks and performing multi-dimensional correlation analysis using the FP-Growth algorithm, a root cause alarm knowledge base is generated, solving the problem of difficult root cause fault location in railway 5G-R networks and achieving rapid and accurate fault location and panoramic decision support.
Patent Information
- Application Number
- CN202511720955.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-21
- Publication Date
- 2026-02-03
AI Technical Summary
The existing alarm management system of railway 5G-R network lacks intelligent correlation analysis capabilities, making it difficult to identify the root cause of the fault from the alarm storm. It also lacks the utilization of historical data and the analysis model does not match the actual operating conditions of the railway, resulting in low efficiency in fault location.
It adopts data collection and enhanced cleaning for railway operations, uses a distributed computing framework to deeply mine massive historical alarm data, combines the FP-Growth algorithm to perform multi-dimensional correlation analysis, generates a root cause alarm knowledge base, and visualizes the fault propagation path in real time, providing panoramic decision support.
It enables precise location of root cause alarms, improves the accuracy of fault location and operation and maintenance efficiency, can proactively issue early warnings and quickly converge the number of alarms, and ensures the safety and smooth operation of railway transportation.
Smart Images

Figure CN121462999A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of mobile communication, in particular to a root cause alarm intelligent positioning method and system for railway 5G-R network. BACKGROUND
[0002] Railway 5G-R network is the core infrastructure to ensure the safety and efficiency of future railway operation, and its business has strict requirements of high reliability, low latency and wide connection. The network structure is complex, covering wireless access network (such as BBU-RHU, pRRU, etc. Access unit covering stations, trackside and tunnel) and core network (5GC network element), and carrying key business such as train control (CTCS), dispatching communication, passenger information service (PIS) and so on.
[0003] In the existing railway network management system, the alarm management faces the following severe challenges: 1. Hidden and complex alarm correlation: A root cause failure (such as core network user plane function UPF abnormality) may quickly derive a series of cross-system and cross-network element alarms such as wireless access abnormality, train-ground communication interruption, and even train control business degradation along the business link. The existing system lacks intelligent correlation analysis capability, and it is difficult to distinguish the root cause from the alarm storm.
[0004] 2. Insufficient value mining of historical data: Traditional railway network management is limited by the performance of relational database, and usually only stores short-term alarm data, resulting in a large amount of valuable data reflecting network history being discarded. Lack of long-term and massive historical data for training and learning makes it difficult to improve the accuracy of experience-based root cause analysis.
[0005] 3. Disconnected from railway business scenarios: The general analysis method does not fully consider the communication characteristics of special scenarios such as linear deployment, high-speed mobility, tunnels and mountainous areas of railway network, resulting in mismatch between analysis model and actual operation conditions of railway, and the positioning result often deviates from the true root cause.
[0006] Therefore, the existing technology cannot meet the stringent requirements of railway 5G-R network for fault rapid positioning and high reliability of business, and an intelligent operation and maintenance solution that can deeply integrate railway business features and has big data analysis capability is urgently needed. SUMMARY
[0007] The purpose of the present application is to provide a root cause alarm intelligent positioning method and system for railway 5G-R network to solve the problems raised in the background technology.
[0008] In order to solve the above technical problems, the present application provides the following technical solutions: a root cause alarm intelligent positioning method for railway 5G-R network, comprising S1, data acquisition and enhanced cleaning for railway business: collecting original alarm data from railway 5G-R wireless network management, core network management and related business systems such as train control and dispatching; S2, mass historical alarm data persistence: the enhanced alarm data is stored in the data warehouse of the railway special big data platform, and a high-throughput distributed column database such as HBase, Kudu and ClickHouse is used to establish a complete alarm medical record for the railway network, and the historical data for years is supported for backtracking analysis; S3, FP-Growth correlation rule mining based on railway scene and optimization: using a distributed computing framework such as Apache Spark to deeply mine the alarm data with time and space and business tags in the data warehouse, and adaptively modifying and optimizing the classic FP-Growth algorithm to cope with the particularity of railway alarm data; S4, multi-dimensional correlation analysis and rule extraction: taking the FP-Growth algorithm as the core engine, combining with the railway operation experience, performing multi-batch and multi-dimensional correlation calculation, the key dimensions include: time and space dimensions including line, kilometer marker, time window, business dimension, affected business system, business importance level, network topology dimension including logical networking and physical link; S5, rule application and root cause alarm panoramic visualization: integrating the generated root cause alarm knowledge base into the railway network management system, when an alarm occurs, the system matches the knowledge base in real time, automatically locates and highlights the root cause alarm, and on the visualization interface, not only the alarm itself is displayed, but also the fault propagation path is intuitively displayed in the form of a topology diagram, and the influence on railway transportation business is evaluated, providing panoramic and business-oriented decision support for dispatchers and operation and maintenance personnel.
[0009] According to the above technical solutions, in the S1, on the basis of basic cleaning, deduplication and standardization, the alarm data is enhanced for business, and is labeled with business impact tags, such as affecting train control, affecting dispatching, location tags such as a line and a kilometer marker, a tunnel and mobility tags such as being associated with a train.
[0010] According to the above technical solutions, in the S3, the adaptive modification and optimization of the classic FP-Growth algorithm are as follows: A, transaction set construction: transforming the traditional one-time shopping basket transaction concept into all co-occurring alarm sets in a fault event window, dividing the alarm stream through a sliding time window, for example, 5 minutes and 15 minutes, according to the railway business response time configuration, regarding all alarm items occurring in the same window as a transaction, so as to capture the correlation of alarms in time; B, FP-Tree construction and compression: the algorithm first scans all transaction sets to generate frequent 1-item sets, i.e., single high-frequency alarms, and ranks them in descending order of support. Then, the second scan of the data is performed to construct a frequent pattern tree, FP-Tree. In this process, the business hierarchical relationship of railway alarms, such as the network element-device-port level, is used to preprocess and merge alarm items, effectively compressing the size of the tree and improving storage and computing efficiency. C, conditional pattern base mining and rule generation: condition pattern bases are recursively mined from the FP-Tree, and a conditional FP-Tree is constructed. Finally, all frequent item sets are obtained. On this basis, the confidence and lift of the association rules between frequent item sets, such as A and B-C, are calculated. A minimum confidence threshold is set to ensure rule reliability, and the lift is used to filter out false associations that are frequent but independent events, ensuring that the generated rules can truly reflect the root-derived causal relationship.
[0011] According to the above technical solution, in S4, the association calculation specifically performs FP-Growth mining according to the scene, and separately mines all alarm transaction sets occurring in the tunnel area to discover exclusive root rules in this specific scene. Through multi-dimensional fusion analysis, a high-confidence root alarm judgment rule set that conforms to the actual railway operation is refined.
[0012] According to the above technical solution, the S2 further includes a data access and real-time cleaning enhancement step. A big data access tool of Flume and Kafka is used to access the multi-source original alarm data collected in step S1 in a streaming manner to a railway special big data processing platform. A streaming processing framework of SparkStreaming and Flink is used to perform real-time deduplication, merge alarms with highly overlapped time and content, and perform format standardization processing on the original alarm data. The processed alarms are uniformly packaged as structured fields containing alarm identification, network element identification, alarm type, occurrence time, severity, location label, and business impact label. In this process, business enhancement and label marking are completed, and high-quality, standardized alarm data input is provided for the persistent storage of step S2.
[0013] According to the above technical solution, in S2, the alarm data stream processed by the data access and real-time cleaning enhancement step is first written to a distributed file system HDFS as an original backup. Then, through a preset ETL process or real-time writing mode, the alarm data is loaded into the data warehouse of the railway special big data platform. The data warehouse is preferably stored in a Kudu distributed columnar database, and is configured to build a historical alarm medical record library with a storage period of several months or even several years. The transaction set used for performing the FP-Growth association rule mining in the steps S3 and S4 and the multi-dimensional data based on the space-time dimension, the business dimension and the network topology dimension are all read from the historical alarm medical record library according to a preset time window, and a batch processing calculation job is triggered periodically in a railway business low peak period to perform incremental update and retraining on the root cause alarm knowledge base.
[0014] According to the above technical solution, in the S5, the root cause alarm knowledge base formed in the step S4 is exported and loaded into the memory of a real-time stream processing engine and a railway network management system application server, when receiving a new real-time alarm stream, the real-time stream processing engine performs pattern matching on alarm combination based on the root cause alarm knowledge base, when a certain root cause alarm association rule is matched, the alarms corresponding to the antecedent of the rule are marked as root cause alarms in an alarm list interface and highlighted with a prominent color, and the derived alarms corresponding to the consequent of the rule are automatically merged under the root cause alarm entry or their display priority is reduced to realize alarm convergence. The fault propagation path from the root cause alarm to each derived alarm is drawn on a topology view, and a fault impact analysis result is generated in combination with the business importance level of the business affected by the root cause alarm; and when a partial alarm combination similar to the historical pattern in the root cause alarm knowledge base is detected and all derived alarms have not yet appeared, early warning information is output in advance and pushed to an operation and maintenance personnel or a dispatcher to realize early warning of a potential root cause fault.
[0015] A root cause alarm intelligent positioning system for a railway 5G-R network, comprising: A collection and enhanced cleaning module collects original alarm data streams from wireless network OMC, core network management, train control, dispatching and other business systems through a standardized interface, and performs real-time cleaning using SparkStreaming and Flink: removes completely repeated alarms, merges alarms with highly overlapped time and content, and unifies them into a structured standard format, alarm ID, network element ID, alarm type, time, severity, location label, and business impact label. An alarm medical record library management module loads data into a data warehouse of a big data platform through a timing ETL or real-time writing mode, and the embodiment preferably uses a distributed columnar database such as Kudu for persistent storage; A root cause alarm knowledge base construction module starts a Spark batch processing job periodically during a business low peak period to read historical alarm data of a recent period from the alarm medical record library; A real-time root cause alarm positioning module exports and loads the root cause alarm knowledge base into a real-time processing engine and a network management application server memory; The alarm convergence module highlights the located root cause alarm in the alarm list interface, and automatically merges the derived alarm under the root cause alarm entry or reduces the display priority of the derived alarm, so as to realize alarm convergence.
[0016] Compared with the prior art, the beneficial effects achieved by the present application are: the present application provides a root cause alarm intelligent positioning method and system for a railway 5G-R network, to solve the problems of difficult root cause fault positioning and low efficiency caused by complex network, large amount of alarms, hidden correlation relationship and insufficient use of historical data; Accurate positioning, in line with the actual railway: by introducing railway business, space-time and mobility labels, the correlation analysis model is closer to the real railway operation scene, and the accuracy of root cause alarm positioning is significantly improved.
[0017] Data-driven, scientific decision-making: using long-term accumulated "alarm medical history" for mining, the fault positioning is changed from relying on personal experience to relying on data-driven scientific decision-making.
[0018] Active early warning, improve safety: through learning from historical patterns, the system can predict the derived alarm chain that may be triggered by a specific root cause fault, and realize early warning.
[0019] Improve operation efficiency and ensure smooth transportation: it can quickly converge the number of alarms, directly hit the root cause of the fault, and greatly shorten the fault troubleshooting time.
[0020] Strong system scalability: based on an open big data technology stack, the system can be smoothly expanded as the railway 5G-R network scale expands. BRIEF DESCRIPTION OF DRAWINGS
[0021] The accompanying drawings are used to provide a further understanding of the present application, and constitute a part of the specification, together with the embodiments of the present application, to explain the present application, and do not constitute a limitation on the present application. In the drawings: Figure 1 is a root cause alarm intelligent positioning flowchart of the present application; Figure 2 is a root cause alarm intelligent positioning system architecture diagram of the present application. DETAILED DESCRIPTION
[0022] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.
[0023] Please refer to Figure 1 and Figure 2The application provides a technical scheme: a root cause alarm intelligent positioning method for a railway 5G-R network, comprising the following steps of S1, data acquisition and enhanced cleaning for railway services: collecting original alarm data from a wireless network management of the railway 5G-R, a core network management and related service systems such as train control and dispatching; S2, mass historical alarm data persistence: storing the enhanced alarm data in a data warehouse of a railway special big data platform, using a high-throughput distributed columnar database such as HBase, Kudu and ClickHouse, establishing a complete alarm medical record for the railway network, and supporting backtracking analysis on historical data for years; S3, FP-Growth correlation rule mining based on the railway scene and optimization: using a distributed computing framework such as Apache Spark to deeply mine the alarm data with time and space and business tags in the data warehouse, and adaptively modifying and optimizing the classical FP-Growth algorithm to cope with the particularity of the railway alarm data; S4, multi-dimensional correlation analysis and rule extraction: taking the FP-Growth algorithm as a core engine, combining railway operation and maintenance experience, and performing multi-batch and multi-dimensional correlation calculation, the key dimensions include: time and space dimensions including a line, a kilometer marker, a time window, a business dimension, an affected business system, a business importance level, and a network topology dimension including logical networking and a physical link; S5, rule application and root cause alarm panoramic visualization: integrating the generated root cause alarm knowledge base into a railway network management system, when an alarm occurs, the system matches the knowledge base in real time, automatically locates and highlights the root cause alarm, and on a visual interface, not only the alarm itself is displayed, but also the fault propagation path is intuitively displayed in the form of a topology diagram, and the influence on railway transportation business is evaluated, thereby providing panoramic and business-oriented decision support for dispatchers and operation and maintenance personnel; In S1, on the basis of basic cleaning, deduplication and standardization, the alarm data is enhanced in business, and is labeled with a business impact label, such as an impact on train control, an impact on dispatching, a position label such as a line and a kilometer marker, a tunnel and a mobility label such as a train number; In S3, the adaptive modification and optimization of the classical FP-Growth algorithm are as follows: A, transaction set construction: converting the traditional one-time shopping basket transaction concept into all co-occurring alarm sets in a fault event window, dividing the alarm stream through a sliding time window, for example, 5 minutes and 15 minutes, according to the railway business response time length configuration, regarding all alarm items occurring in the same window as a transaction, so as to capture the correlation of the alarm in time; B. FP-Tree construction and compression: The algorithm first scans all the transaction sets to generate frequent 1-item sets, i.e., single high-frequency alarms, and ranks them in descending order of their support. Then, the second scan of the data is performed to construct the frequent pattern tree, FP-Tree. In this process, the hierarchical relationship of railway alarms, such as the network element-device-port level, is used to preprocess and merge the alarm items, effectively compressing the size of the tree and improving the storage and computing efficiency. C. Conditional pattern base mining and rule generation: The conditional pattern base is recursively mined from the FP-Tree, and the conditional FP-Tree is constructed. Finally, all frequent item sets are obtained. On this basis, the confidence and lift of the association rules between frequent item sets, such as A and B-C, are calculated. A minimum confidence threshold is set to ensure the reliability of the rules, and the lift is used to filter out false associations that are frequent but independent events, ensuring that the generated rules can truly reflect the root-derived causal relationship. In S4, the association calculation specifically performs FP-Growth mining by scene, and separately mines all alarm transaction sets occurring in the tunnel area to discover exclusive root rules in this specific scenario. Through multi-dimensional fusion analysis, a set of high-confidence root alarm judgment rules that conform to the actual railway operation is refined. S2 also includes data access and real-time cleaning enhancement steps. The multi-source raw alarm data collected in step S1 is accessed in a streaming manner to the railway dedicated big data processing platform using Flume and Kafka big data access tools. The raw alarm data is processed in real time using the streaming processing frameworks of SparkStreaming and Flink to remove duplicates, merge alarms with highly overlapping time and content, and perform format standardization. The processed alarms are uniformly packaged as structured fields containing alarm identification, network element identification, alarm type, occurrence time, severity, location label, and business impact label. Business enhancement and label tagging are completed in this process, providing high-quality, standardized alarm data input for the persistent storage of step S2. In S2, the alarm data stream processed by the data access and real-time cleaning enhancement steps is first written to the distributed file system HDFS as a raw backup. Then, through a pre-set ETL process or real-time writing method, the alarm data is loaded into the data warehouse of the railway dedicated big data platform. The data warehouse is preferably stored in a Kudu distributed columnar database, configured to build a historical alarm medical record library with a storage period of several months or even years. The transaction sets used to perform FP-Growth association rule mining in steps S3 and S4, as well as the multi-dimensional data based on the time and space dimensions, business dimensions, and network topology dimensions, are all read from the historical alarm medical record library in a pre-set time window, and batch processing jobs are periodically triggered during the railway business off-peak period to incrementally update and retrain the root alarm knowledge base. In S5, the root cause alarm knowledge base formed in step S4 is exported and loaded into the memory of the real-time stream processing engine and the railway network management system application server. When the real-time stream processing engine receives a new real-time alarm stream, it performs pattern matching on alarm combination based on the root cause alarm knowledge base. When a root cause alarm correlation rule is matched, the alarms corresponding to the antecedent of the rule are marked as root cause alarms in the alarm list interface and highlighted in a prominent color, and the derived alarms corresponding to the consequent of the rule are automatically merged under the root cause alarm entry or their display priority is reduced to achieve alarm convergence. The fault propagation path from the root cause alarm to each derived alarm is drawn on the topology view, and a fault impact analysis result is generated in combination with the service importance level of the service affected by the root cause alarm. When a partial alarm combination similar to the historical pattern in the root cause alarm knowledge base is detected but not all derived alarms have occurred, early warning information is output in advance and pushed to the operation and maintenance personnel or dispatchers to achieve early warning of potential root cause faults. A root cause alarm intelligent positioning system for a railway 5G-R network, comprising: A collection and enhancement cleaning module collects original alarm data streams from wireless network OMC, core network management, train control, dispatching and other business systems through a standardized interface, and performs real-time cleaning using SparkStreaming and Flink: removes completely repeated alarms, merges alarms with highly overlapping time and content, and unifies them into a structured standard format, alarm ID, network element ID, alarm type, time, severity, location label, and service impact label. An alarm medical record library management module loads data into the data warehouse of the big data platform through timed ETL or real-time writing. In this embodiment, a distributed columnar database such as Kudu is preferably used for persistent storage. A root cause alarm knowledge base construction module periodically starts a Spark batch processing job during a business low peak period to read historical alarm data from the alarm medical record library for a period of time. A real-time root cause alarm positioning module exports and loads the root cause alarm knowledge base into the memory of the real-time processing engine and the network management application server. An alarm convergence module highlights the positioned root cause alarm in the alarm list interface, and automatically merges the derived alarms under the root cause alarm entry or reduces their display priority to achieve alarm convergence.
[0024] The specific implementation of the present application in the railway 5G-R network will be described in detail below with reference to the accompanying drawings and embodiments.
[0025] Step one: multi-source data collection Through the standardized data interface, the original and real-time alarm data streams are collected from the wireless network operation and maintenance center (OMC) of the 5G-R live network, the core network management system, and related train control (CTC), dispatching communication and other business systems.
[0026] Step two: data access and real-time cleaning enhancement ① Using big data access tools such as Flume or Kafka, the multi-source data collected in step one is real-time accessed to the railway special big data processing platform.
[0027] ② Using SparkStreaming or Flink, the original alarm data is preliminarily cleaned, including: removing completely repeated alarms and merging alarms with highly overlapping time and content.
[0028] ③ The cleaned alarm data is standardized and business enhanced, and is unified into a structured standard format (alarm ID, network element ID, alarm type, occurrence time, severity, location label, business impact label,...).
[0029] Step three: mass data persistent storage ① The high-quality alarm data stream processed in step two is first written into a distributed file system (such as HDFS) as an original backup.
[0030] ② Through a timed ETL (extraction, transformation, loading) process or real-time writing method, the data is loaded into the data warehouse of the big data platform, and the distributed columnar database Kudu is preferably used for persistent storage.
[0031] ③ The data storage period can be flexibly configured according to the operation and maintenance requirements, and is usually set to several months or even several years to build a historical "alarm medical record library" for long-term analysis.
[0032] Step four: association rule mining and knowledge base construction ① A preset Spark batch processing job is started regularly (for example, during the low business peak period in the early morning).
[0033] ② The job reads the historical alarm data of a certain period (such as 30 days) from the data warehouse (Kudu).
[0034] ③ Transaction set construction: according to the preset sliding time window (such as 10 minutes), the continuous alarm stream is cut into multiple "alarm transactions", and each transaction contains all co-occurring alarm items within the time window.
[0035] (4) Execute FP-Growth algorithm: call the FP-Growth algorithm in the SparkMLlib library to perform frequent pattern mining on the above transaction set. The algorithm automatically discovers the frequent co-occurrence relationship between the alarm items and generates association rules.
[0036] Output example: The algorithm may output the rule {AMF node communication interruption}->{UPF link alarm, SMF service timeout} with a confidence of 95%.
[0037] (5) Multi-dimensional analysis and rule screening: start another Spark job in parallel to group the data from multiple dimensions such as time and space, business, topology, etc., and perform FP-Growth mining respectively to generate multiple rule subsets with scene characteristics. Finally, through rule merging and screening algorithms (such as weighted sorting based on confidence and lift), the final comprehensive root alarm knowledge base is formed.
[0038] Step five: rule application and real-time root location (1) The root alarm knowledge base generated in step four is exported and loaded into the memory of the real-time processing engine (such as FlinkCEP) and the application server of the network management system for high-speed query.
[0039] (2) When the network management system receives a new real-time alarm stream, the real-time matching module will immediately match the new alarm with the rules in the knowledge base.
[0040] Step six: root alarm panoramic visualization (1) Once a certain association rule is matched (for example, {UPF link alarm, SMF service timeout} appears again), the system will automatically locate and highlight the root alarm {AMF node communication interruption} according to the rule antecedent.
[0041] (2) In the alarm list interface of the network management system, the root alarm is marked as "root" and highlighted in a prominent color (such as red).
[0042] (3) At the same time, the system will automatically merge its derivative alarms (such as UPF and SMF alarms) under the root alarm entry or reduce their display priority, thereby achieving alarm convergence.
[0043] (4) On the topology view, the system can visually display the fault propagation path from the root alarm to the derivative alarms and generate a preliminary fault impact analysis report to provide panoramic decision support for the operation and maintenance personnel.
[0044] Through the implementation of the above six steps, the invention achieves the goal of quickly, automatically and accurately locating the root fault in the massive alarms of the railway 5G-R network, significantly improving the operation and maintenance efficiency and network reliability.
[0045] The scheme combines the frequent pattern growth (FP-Growth) algorithm with the innovation of railway business feature deep integration and multi-dimensional analysis mechanism, and constructs a complete and efficient railway 5G-R network root alarm intelligent positioning system. The combination innovatively adapts the general data mining algorithm to the special operation and maintenance scene of railway communication network with high reliability, linear deployment and strong business association. Through business-enhanced data preprocessing, algorithm optimization for railway scene and spatio-temporal-business-topology multi-dimensional correlation analysis, the root fault is automatically, accurately and quickly located from massive, heterogeneous and cross-domain alarm data, effectively solving the inherent problems of traditional methods in correlation discovery, historical data utilization and business scene fitting degree, and significantly improving the intelligent level and business guarantee capability of railway special mobile communication system.
[0046] It should be noted that in this paper, relational terms such as first and second are used merely to distinguish one entity or action from another, without necessarily requiring or implying any such actual relationship or order between such entities or actions. Moreover, the terms "comprising", "including" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or apparatus including a series of elements includes not only those elements, but also other elements not explicitly listed, or inherent to such a process, method, article or apparatus.
[0047] Finally, it should be noted that: the above only describes the preferred embodiments of the present application, and does not limit the present application. Although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions recorded in the foregoing embodiments, or replace some technical features with equivalent ones. Any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application shall be included in the protection scope of the present application.
Claims
1. A root cause alarm intelligent location method for railway 5G-R networks, characterized in that: include: S1. Data collection and enhanced cleaning for railway business: Collect raw alarm data from railway 5G-R wireless network management, core network management, and related business systems such as train control and dispatching; S2. Persistence of massive historical alarm data: The enhanced alarm data is persistently stored in the data warehouse of the railway-specific big data platform. High-throughput distributed columnar databases such as HBase, Kudu, and ClickHouse are used to establish a complete alarm medical record for the railway network and support retrospective analysis of historical data for several years. S3. FP-Growth Association Rule Mining Based on Railway Scenarios and Optimization: Utilize distributed computing frameworks such as Apache Spark to perform in-depth mining of alarm data with spatiotemporal and business tags in the data warehouse, and adapt and optimize the classic FP-Growth algorithm to cope with the special characteristics of railway alarm data. S4. Multi-dimensional correlation analysis and rule extraction: Using the FP-Growth algorithm as the core engine, combined with railway operation and maintenance experience, multi-batch, multi-dimensional correlation calculations are performed. Key dimensions include: spatiotemporal dimensions including lines, kilometer markers, time windows; business dimensions including affected business systems, business importance levels; and network topology dimensions including logical networking and physical links. S5. Rule Application and Root Cause Alarm Panoramic Visualization: The generated root cause alarm knowledge base is integrated into the railway network management system. When an alarm occurs, the system matches the knowledge base in real time, automatically locates and highlights the root cause alarm. On the visualization interface, not only is the alarm itself displayed, but the fault propagation path is also intuitively shown in the form of a topology diagram, and its impact on railway transportation operations is assessed, providing dispatchers and maintenance personnel with panoramic and business-oriented decision support.
2. The root cause alarm intelligent location method for railway 5G-R networks according to claim 1, characterized in that: In S1, on top of basic cleaning, deduplication, and standardization, alarm data is enhanced by adding business impact tags, such as affecting train control, affecting scheduling, location tags such as a certain line and a certain kilometer mark, a certain tunnel, and mobility tags such as associating with a certain train number.
3. The root cause alarm intelligent location method for railway 5G-R networks according to claim 2, characterized in that: In S3, the adaptive modification and optimization of the classic FP-Growth algorithm are specifically as follows: A. Transaction set construction: The traditional concept of a shopping basket transaction is transformed into a set of all co-occurring alarms within a fault event window. By using sliding time windows, such as 5 minutes and 15 minutes, and configuring them according to the railway business response time, the alarm flow is segmented, and all alarm items occurring within the same window are regarded as a transaction, thereby capturing the temporal correlation of alarms. B. FP-Tree Construction and Compression: The algorithm first scans all transaction sets to generate frequent 1-itemsets, i.e., single high-frequency alarms, and sorts them in descending order of their support. Then, it scans the data a second time to construct a frequent pattern tree, FP-Tree. In this process, the business hierarchy relationship of railway alarms, such as the network element-device-port hierarchy, is used to preprocess and merge alarm items, effectively compressing the size of the tree and improving storage and computing efficiency. C. Conditional Pattern Base Mining and Rule Generation: Conditional pattern bases are recursively mined from the FP-Tree, and conditional FP-Trees are constructed to obtain all frequent itemsets. Based on this, the association rules between frequent itemsets are calculated, such as the confidence and lift of A and BC. A minimum confidence threshold is set to ensure the reliability of the rules, and the lift is used to filter out false associations that are frequent but are actually independent events, ensuring that the generated rules can truly reflect the root-derived causal relationship.
4. The root cause alarm intelligent location method for railway 5G-R networks according to claim 3, characterized in that: In S4, the correlation calculation specifically involves performing FP-Growth mining in different scenarios, mining all alarm transaction sets occurring in the tunnel area separately to discover the unique root cause rules under this specific scenario, and extracting a set of root cause alarm judgment rules with high confidence and in line with the actual operation of railways through multi-dimensional fusion analysis.
5. The root cause alarm intelligent location method for railway 5G-R networks according to claim 4, characterized in that: S2 also includes a data access and real-time cleaning and enhancement step. Using Flume and Kafka, big data access tools, the multi-source raw alarm data collected in step S1 is streamed into the railway-specific big data processing platform. Using Spark Streaming and Flink, the raw alarm data is deduplicated in real time, alarms with highly overlapping time and content are merged, and format standardization is performed. The processed alarms are uniformly encapsulated into structured fields including alarm identifier, network element identifier, alarm type, occurrence time, severity, location tag, and business impact tag. In this process, business enhancement and tagging are completed to provide high-quality, standardized alarm data input for persistent storage in step S2.
6. The root cause alarm intelligent location method for railway 5G-R networks according to claim 5, characterized in that: In step S2, the alarm data stream processed by the data access and real-time cleaning and enhancement steps is first written to the distributed file system HDFS as an original backup; then, through a preset ETL process or real-time writing method, the alarm data is loaded into the data warehouse of the railway-specific big data platform. The data warehouse preferably uses Kudu distributed columnar database for persistent storage and is configured to build a historical alarm record database with storage cycles of several months or even several years. In steps S3 and S4, the transaction set used to perform FP-Growth association rule mining, as well as the multi-dimensional data based on spatiotemporal, business, and network topology dimensions, are all read from the historical alarm medical record database according to a preset time window. During the off-peak period of railway business, batch processing calculations are periodically triggered to incrementally update and retrain the root cause alarm knowledge base.
7. The root cause alarm intelligent location method for railway 5G-R networks according to claim 6, characterized in that: In step S5, the root cause alarm knowledge base formed in step S4 is exported and loaded into the memory of the real-time stream processing engine and the railway network management system application server. When the real-time stream processing engine receives a new real-time alarm stream, it performs pattern matching on the alarm combination based on the root cause alarm knowledge base. When a root cause alarm association rule is matched, the alarm corresponding to the antecedent of the rule is marked as a root cause alarm in the alarm list interface and highlighted with a prominent color. The derived alarms corresponding to the consequent of the rule are automatically merged under the root cause alarm entry or their display priority is reduced to achieve alarm convergence. The fault propagation path from the root cause alarm to each derivative alarm is plotted on the topology view, and the fault impact analysis results are generated by combining the business importance level of the services affected by the root cause alarm; when a partial alarm combination similar to the historical pattern in the root cause alarm knowledge base is detected but all derivative alarms have not yet appeared, early warning information is output in advance and pushed to operation and maintenance personnel or dispatchers to achieve early warning of potential root cause faults.
8. A root cause alarm intelligent positioning system for railway 5G-R networks according to any one of claims 1-7, characterized in that: include: The data acquisition and enhanced cleaning module collects raw alarm data streams from business systems such as wireless network OMC, core network management, train control, and scheduling through standardized interfaces. It uses Spark Streaming and Flink to perform real-time cleaning: removing completely duplicate alarms, merging alarms with highly overlapping time and content, and unifying them into a structured standard format, including alarm ID, network element ID, alarm type, time, severity, location label, and business impact label. The alarm medical record database management module loads data into the data warehouse of the big data platform through timed ETL or real-time writing. In this embodiment, it is preferred to use a distributed columnar database such as Kudu for persistent storage. The root cause alarm knowledge base building module periodically starts Spark batch processing jobs during off-peak business periods to read historical alarm data from the alarm history database over a recent period. The real-time root cause alarm location module exports the root cause alarm knowledge base and loads it into the memory of the real-time processing engine and the network management application server; The alarm convergence module highlights the root cause alarms identified in the alarm list interface, and automatically merges derivative alarms into the root cause alarm entry or reduces their display priority, thereby achieving alarm convergence.
