Multi-agent cooperative identification ENodeB batch out-of-service root cause positioning method and system

Abstract

The invention relates to the technical field of network management, in particular to an ENodeB batch out-of-service root cause positioning method and system based on multi-agent cooperative identification, and the method comprises the following steps: a problem rewriting agent collects sentence pattern information of an operator query performance intention; carrying out keyword extraction and intention recognition on the fault description input by a user by utilizing a natural language processing technology, and generating a structured query statement containing a fault time range, base station group characteristics and a performance index threshold value; the method has the beneficial effects that the conversion from a natural language to a structured query is realized, and key parameters such as a time range and base station cluster characteristics are extracted by adopting a semantic recombination technology. Then performing intelligent delimitation, executing'scene classification 'and'path planning', and dynamically generating a detection path tree; and finally, in a data fusion stage, an API tool set is called through a network management system, cross-system data alignment is realized, and a root cause of a fault is inferred. The situation that tasks of complex scenes are completed through a large model is achieved.

Description

Technical Field

[0001] This invention relates to the field of network management technology, specifically to a method and system for multi-agent collaborative identification of the root cause of batch ENodeB outages. Background Art

[0002] Traditional ENodeB fault location mainly relies on the experience of operation and maintenance experts and single-system log analysis, which presents three major challenges: 1. Excessive reliance on manual labor: It requires collaboration among experts from multiple fields such as transmission, wireless, and core network, with an average location time exceeding 4 hours; 2. Data silo effect: The data formats of systems such as network management, environmental monitoring, and performance are heterogeneous, and the alignment error rate of key indicators reaches 23%; 3. Lack of dynamic adaptation: Differences in the version / configuration of existing base stations result in more than 35% of faults not being able to match the preset detection template.

[0003] Existing technical shortcomings: 1. Linear and rigid process: The serial troubleshooting mechanism cannot dynamically switch detection paths, resulting in a false judgment rate as high as 27.6%; 2. Weak intelligent decision-making: The traditional rule engine only covers 68% of known scenarios and lacks a confidence verification mechanism; 3. Poor readability of results: The location report contains more than 80% technical terms, requiring front-line operation and maintenance personnel to translate them. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for multi-agent collaborative identification of the root cause of ENodeB batch outages, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for multi-agent collaborative identification of the root cause of batch ENodeB outages, comprising the following steps:

[0006] S1: The "problem rewriting agent" collects sentence structure information of operators' performance query intentions, uses natural language processing technology to extract keywords and identify intents from the fault descriptions input by users, and generates structured query statements containing fault time range, base station group characteristics, and performance indicator thresholds.

[0007] S2: The "Scene Classification Intelligent Agent" implements scene classification decisions, identifies fault types through a pre-trained classification model, determines whether it is a wireless professional ENodeB base station outage scenario, and calls the historical case library to match typical scenarios with a similarity greater than 85%.

[0008] S3: The "Path Planning Agent" dynamically generates a diagnostic path tree based on the classification results. The diagnostic paths are as follows: Business Engineering Status → Circuit Type → Transmission Type → Uplink BRAS Fault Type → Business Complaint Status → Online User Count Status → Self-Equipment Alarm Fault Status → Environmental Alarm Fault Status → Event Resource Error Correction Status.

[0009] S4: The "Tool Recall and Invocation Agent" constructs a tool invocation matrix to recall associated API toolkits from the network management system, performance platform, and configuration library. These tools include circuit-related transmission equipment fault verification (Tool-101), optical path-related transmission equipment fault verification (Tool-102), uplink BRAS fault troubleshooting (Tool-103), core service complaint status (Tool-104), obtaining online user count (Tool-105), self-equipment alarm fault troubleshooting capability (Tool-106), environmental alarm fault troubleshooting capability (Tool-107), verifying service engineering status (Tool-108), and verifying event resource errata (Tool-109).

[0010] S5: The "Tool Recall and Invocation Agent" executes multiple rounds of tool invocation in a "request-response-decision" pattern.

[0011] S6: The "Root Cause Localization Agent" constructs a multi-dimensional feature space, maps the alarm data, performance indicators, and configuration parameters returned by the tool to a unified feature space, and uses a sliding window algorithm to detect abnormal mutation points;

[0012] S7: The "Root Cause Localization Agent" performs association rule mining, discovers strong association rules of fault features through the improved Apriori algorithm, with support greater than 0.7 and confidence greater than 0.85, and outputs the TOP3 suspected root causes;

[0013] S8: The "Report Generation Agent" uses a template engine to automatically generate interpretable reports;

[0014] S9: The "Report Generation Agent" optimizes the readability of the output results through the BERT semantic matching model, ensuring a 100% conversion rate of technical terms.

[0015] Preferably, the sentence structure information collected by the "problem rewriting agent" regarding the operator's query performance intent covers a variety of common fault description patterns. Natural language processing technology adopts a combination of lexical analysis, syntactic analysis, and semantic understanding to accurately extract keywords and identify user intent.

[0016] Preferably, the pre-trained classification model used by the "scene classification agent" is trained based on a large amount of historical fault data. It classifies fault types through deep learning algorithms. When calling the historical case library, it uses a similarity calculation algorithm to comprehensively consider multi-dimensional information such as fault description, time, and location to match typical scenarios with a similarity greater than 85%.

[0017] Preferably, when constructing a multi-dimensional feature space, the "root cause localization agent" standardizes alarm data, performance indicators, and configuration parameters to ensure that different types of data are comparable in a unified feature space; when using the sliding window algorithm to detect abnormal mutation points, it dynamically adjusts the window size and sliding step size according to the data characteristics.

[0018] Preferably, when the "report generation agent" automatically generates an interpretable report using a template engine, it selects the appropriate report template according to different suspected root cause types. The report content includes fault description, diagnosis process, suspected root cause, and recommended measures information. When optimizing the readability of the output results through the BERT semantic matching model, it transforms technical terms into easy-to-understand expressions and ensures that the transformed content is consistent with the original technical meaning.

[0019] A system for a multi-agent collaborative method to identify the root causes of ENodeB batch outages includes:

[0020] The problem rewriting module has a "problem rewriting intelligent agent" which collects the sentence structure information of the operator's query intent for performance. It uses natural language processing technology to extract keywords and identify intent from the fault description input by the user, and generates a structured query statement that includes the fault time range, base station group characteristics, and performance index thresholds.

[0021] The scenario classification module is equipped with a "scenario classification intelligent agent" to implement scenario classification decisions. It identifies the fault type through a pre-trained classification model, determines whether it is a wireless professional ENodeB base station outage scenario, and calls the historical case library to match typical scenarios with a similarity greater than 85%.

[0022] The path planning module has a "path planning intelligent agent" which is used to dynamically generate a location path tree based on the classification results. The diagnostic paths are as follows: business engineering status → circuit type → transmission type → uplink BRAS fault type → business complaint status → number of online users status → self-equipment alarm fault status → environmental alarm fault status → event resource correction status.

[0023] The tool recall and invocation module has a "tool recall and invocation intelligent agent" used to build a tool invocation matrix. It recalls related API toolkits from the network management system, performance platform, and configuration library, including circuit-related transmission equipment fault verification (Tool-101), optical path-related transmission equipment fault verification (Tool-102), uplink BRAS fault troubleshooting (Tool-103), core service complaint status (Tool-104), online user count acquisition (Tool-105), self-equipment alarm fault troubleshooting capability (Tool-106), environmental alarm fault troubleshooting capability (Tool-107), business engineering status verification (Tool-108), and event resource correction verification (Tool-109). It executes multiple rounds of tool invocation in a loop according to the "request-response-decision" pattern.

[0024] The root cause localization module is equipped with a "root cause localization agent" to construct a multi-dimensional feature space. It maps the alarm data, performance indicators, and configuration parameters returned by the tool to a unified feature space, uses a sliding window algorithm to detect abnormal mutation points, performs association rule mining, and discovers strong association rules of fault features through an improved Apriori algorithm, with support greater than 0.7 and confidence greater than 0.85, and outputs the top 3 suspected root causes.

[0025] The report generation module includes a "report generation agent" that uses a template engine to automatically generate interpretable reports and optimizes the readability of the output results using a BERT semantic matching model to ensure a 100% conversion rate of technical terms.

[0026] Preferably, in the problem rewriting module, the sentence structure information collected by the "problem rewriting agent" regarding the operator's query performance intent covers a variety of common fault description patterns. Natural language processing technology adopts a combination of lexical analysis, syntactic analysis, and semantic understanding to accurately extract keywords and identify user intent, and the generated structured query statements can adapt to the data processing needs of subsequent modules.

[0027] Preferably, in the scene classification module, the pre-trained classification model used by the "scene classification agent" is trained based on a large amount of historical fault data and classifies fault types through deep learning algorithms. When calling the historical case library, a similarity calculation algorithm is used to comprehensively consider multi-dimensional information such as fault description, time, and location to match typical scenarios with a similarity greater than 85%, and the matching results are fed back to the path planning module to assist in generating a location path tree.

[0028] Preferably, in the root cause localization module, the "root cause localization agent" standardizes alarm data, performance indicators, and configuration parameters when constructing a multi-dimensional feature space to ensure that different types of data are comparable in a unified feature space; when using the sliding window algorithm to detect abnormal mutation points, it dynamically adjusts the window size and sliding step size according to the data characteristics; and when the improved Apriori algorithm mines association rules, it filters and optimizes fault features to improve the accuracy and effectiveness of association rules.

[0029] Preferably, in the report generation module, when the "report generation agent" automatically generates an interpretable report using a template engine, it selects the appropriate report template according to different suspected root cause types. The report content includes fault description, diagnostic process, suspected root cause, and recommended measures information. When optimizing the readability of the output results through the BERT semantic matching model, it transforms technical terms into easy-to-understand expressions and ensures that the transformed content is consistent with the original technical meaning. Furthermore, the generated report can be output in multiple formats, such as PDF and Word, to meet the needs of different users.

[0030] Compared with the prior art, the present invention has the following beneficial effects:

[0031] This invention proposes a multi-agent collaborative method and system for identifying the root causes of batch ENodeB outages. It converts natural language into structured queries and employs semantic recombination technology to extract key parameters such as time range and base station cluster characteristics. Then, intelligent delimitation is performed, executing "scene classification" and "path planning," and dynamically generating a detection path tree. Finally, in the data fusion stage, cross-system data alignment is achieved by calling API tools through the network management system to infer the root cause of the fault. This enables the use of a large model to complete tasks in complex scenarios. Attached Figure Description

[0032] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0033] To make the objectives, technical solutions, and advantages of the present invention clear and complete, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only some, not all, embodiments of the present invention, and are merely illustrative of the embodiments of the present invention. They are not intended to limit the embodiments of the present invention. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0034] Example 1, please refer to Figure 1 This invention provides a technical solution: a method for multi-agent collaborative identification of the root cause of batch ENodeB outages, comprising the following steps:

[0035] I. Input Processing Stage

[0036] Step S1: The "problem rewriting agent" receives the original problem input and performs semantic reconstruction.

[0037] Natural language processing (NLP) techniques are used to extract keywords and identify intent from user-inputted fault descriptions, generating structured query statements that include fault time range, base station group characteristics, and performance indicator thresholds.

[0038] II. Scene Delineation Stage

[0039] Step S2: Implement scenario classification decision

[0040] The "Scene Classification Intelligent Agent" identifies the fault type through a pre-trained classification model, determining whether it is a wireless professional ENodeB base station outage scenario, and calls the historical case library to match typical scenarios with a similarity of >85%.

[0041] Step S3: Generate the location path tree

[0042] The "path planning agent" dynamically generates diagnostic paths based on classification results.

[0043] Business Engineering Status → Circuit-related → Transmission-related → Uplink BRAS Fault-related → Business Complaint Status → Online User Count Status → Self-equipment Alarm Fault Status → Environmental Alarm Fault Status → Event Resource Errata Status.

[0044] Explanation of each category:

[0045] Business engineering status: Cutover time window matching → Expansion operation log tracing → License resource verification → Neighbor cell parameter modification records

[0046] Circuitry: Power Module → Baseband Board → RF Unit Detection Chain

[0047] Transmission Category: Optical Path Status → Transmission Error Rate → Routing Configuration Detection Chain

[0048] Uplink BRAS fault types: abnormal optical power → BRAS port CRC packet error rate → routing protocol oscillation detection → QoS policy conflict verification

[0049] Business Complaint Analysis: Complaint Type Clustering → User Geographic Distribution Heatmap → Business Type Correlation Analysis → VIP User Impact Assessment

[0050] Online user statistics: Analysis of sudden user surges → IP address pool exhaustion detection → APN congestion coefficient calculation → AAA authentication log tracing

[0051] Internal device alarm faults: Main control board heartbeat loss → Single board CPU overload alarm → Version consistency verification → Critical process liveness detection

[0052] Environmental alarm fault conditions: DC power supply fluctuation → excessive temperature and humidity in the computer room → unauthorized access alarm → battery pack discharge curve

[0053] Event resource correction process: Alarm work order conflict detection → Network element resource identifier verification → Topology connection relationship verification → GIS latitude and longitude correction

[0054] III. Tool Collaboration Phase

[0055] Step S4: Construct the tool call matrix

[0056] The "Tool Recall and Invocation Agent" recalls associated API toolkits from network management systems, performance platforms, and configuration repositories, establishing a system that includes:

[0057] Tool-101: Circuit-related Transmission Equipment Fault Troubleshooting

[0058] Tool-102: Fault Diagnosis of Optical Path Associated Transmission Equipment

[0059] Tool-103: Troubleshooting Uplink BRAS Faults

[0060] Tool-104: Verify business complaint information

[0061] Tool-105: Get Online User Count

[0062] Tool-106: Troubleshooting capabilities for device alarms and faults

[0063] Tool-107: Troubleshooting capabilities for environmental alarms (single computer room)

[0064] Tool-108: Verify Business Engineering Status

[0065] Tool-109: Verify Event Resource Errata

[0066] Step S5: Perform multiple rounds of tool calls

[0067] The "tool recall and invocation agent" is executed cyclically according to the "request-response-decision" pattern:

[0068] ① Select the best detection tool currently available (based on information entropy evaluation)

[0069] ②Analyze the output parameter format constraints

[0070] ③ Call the API to obtain cross-system data

[0071] ④ When the data confidence level is <90%, a retry mechanism is triggered.

[0072] IV. Root Cause Locating Stage

[0073] Step S6: Construct a multidimensional feature space

[0074] The "root cause localization agent" maps the alarm data, performance indicators, and configuration parameters returned by the tool to a unified feature space, and uses a sliding window algorithm to detect abnormal mutation points.

[0075] Step S7: Perform association rule mining

[0076] The "root cause localization agent" uses an improved Apriori algorithm to discover strong correlation rules of fault features (support > 0.7, confidence > 0.85) and outputs the top 3 suspected root causes.

[0077] V. Results Generation Stage

[0078] Step S8: Generate an interpretable report

[0079] The "Report Generation Agent" uses a template engine to automatically generate reports that include: the impact of the fault (number of base stations / traffic loss); the root cause probability distribution (Bayesian network calculation); and a knowledge graph of repair suggestions.

[0080] Step S9: Implement answer optimization

[0081] The "report generation agent" optimizes the readability of the output results through the BERT semantic matching model, ensuring a 100% conversion rate of technical terms.

[0082] Step S1, the "problem rewriting agent" collects the sentence structure information of the operator's query intent for performance, and uses natural language processing technology to extract keywords and identify intent from the fault description input by the user, generating a structured query statement that includes the fault time range, base station group characteristics, and performance indicator thresholds.

[0083] Step S2, the "Scene Classification Intelligent Agent" implements scene classification decision-making, identifies the fault type through a pre-trained classification model, determines whether it is a wireless professional ENodeB base station outage scenario, and calls the historical case library to match typical scenarios with a similarity of >85%.

[0084] Step S3, the "path planning agent," generates a location path tree. Based on the classification results, a diagnostic path is dynamically generated: Business engineering status → Circuit type → Transmission type → Uplink BRAS fault type → Business complaint status → Online user count status → Self-equipment alarm fault status → Environmental alarm fault status → Event resource correction status.

[0085] Step S4, "Tool Recall and Intelligent Agent Invocation," constructs a tool invocation matrix, recalling associated API toolkits from the network management system, performance platform, and configuration library. These tools include: circuit-related transmission equipment fault verification (Tool-101), optical path-related transmission equipment fault verification (Tool-102), uplink BRAS fault troubleshooting (Tool-103), core service complaint status (Tool-104), obtaining online user count (Tool-105), self-equipment alarm fault troubleshooting capability (Tool-106), environmental alarm fault troubleshooting capability (Tool-107), verifying service engineering status (Tool-108), and verifying event resource errata (Tool-109).

[0086] Step S5, "Tool Recall and Invocation Agent", executes multiple rounds of tool invocation, following a "request-response-decision" pattern.

[0087] Step S6, the "Root Cause Localization Agent" constructs a multi-dimensional feature space, maps the alarm data, performance indicators, and configuration parameters returned by the tool to a unified feature space, and uses a sliding window algorithm to detect abnormal mutation points.

[0088] Step S7, the "Root Cause Localization Agent" performs association rule mining, discovers strong association rules of fault features (support > 0.7, confidence > 0.85) through the improved Apriori algorithm, and outputs the TOP3 suspected root causes.

[0089] Step S8, “Report Generation Agent,” uses a template engine to automatically generate an interpretable report.

[0090] Step S9, "Report Generation Agent," optimizes the answer by using the BERT semantic matching model to improve the readability of the output, ensuring a 100% conversion rate for technical terms.

[0091] Example 2, based on Example 1, proposes a system for a multi-agent collaborative identification method for root cause localization of ENodeB batch outages, comprising:

[0092] The problem rewriting module includes a "problem rewriting agent" that collects sentence structure information from operators' performance query intentions. It uses natural language processing (NLP) technology to extract keywords and identify intent in the user-input fault description, generating a structured query statement that includes the fault time range, base station group characteristics, and performance indicator thresholds. The sentence structure information collected by the "problem rewriting agent" covers various common fault description patterns. NLP technology combines lexical analysis, syntactic analysis, and semantic understanding to accurately extract keywords and identify user intent, and the generated structured query statement can adapt to the data processing needs of subsequent modules.

[0093] The scenario classification module includes a "scenario classification agent" that makes scenario classification decisions. It identifies fault types using a pre-trained classification model, determines whether a scenario involves the outage of a wireless professional ENodeB base station, and calls upon a historical case library to match typical scenarios with a similarity greater than 85%. The pre-trained classification model used by the "scenario classification agent" is trained on a large amount of historical fault data and classifies fault types using deep learning algorithms. When calling the historical case library, a similarity calculation algorithm is used, comprehensively considering fault description, time, location, and other multi-dimensional information to match typical scenarios with a similarity greater than 85%. The matching results are then fed back to the path planning module to assist in generating a location path tree.

[0094] The path planning module has a "path planning intelligent agent" which is used to dynamically generate a location path tree based on the classification results. The diagnostic paths are as follows: business engineering status → circuit type → transmission type → uplink BRAS fault type → business complaint status → number of online users status → self-equipment alarm fault status → environmental alarm fault status → event resource correction status.

[0095] The tool recall and invocation module has a "tool recall and invocation intelligent agent" used to build a tool invocation matrix. It recalls related API toolkits from the network management system, performance platform, and configuration library, including circuit-related transmission equipment fault verification (Tool-101), optical path-related transmission equipment fault verification (Tool-102), uplink BRAS fault troubleshooting (Tool-103), core service complaint status (Tool-104), online user count acquisition (Tool-105), self-equipment alarm fault troubleshooting capability (Tool-106), environmental alarm fault troubleshooting capability (Tool-107), business engineering status verification (Tool-108), and event resource correction verification (Tool-109). It executes multiple rounds of tool invocation in a loop according to the "request-response-decision" pattern.

[0096] The root cause localization module includes a "root cause localization agent" used to construct a multi-dimensional feature space. This agent maps alarm data, performance metrics, and configuration parameters returned by the tool to a unified feature space. It employs a sliding window algorithm to detect anomalous mutation points and performs association rule mining. An improved Apriori algorithm is used to discover strong association rules for fault features, with support greater than 0.7 and confidence greater than 0.85, and outputs the top 3 suspected root causes. When constructing the multi-dimensional feature space, the "root cause localization agent" standardizes alarm data, performance metrics, and configuration parameters to ensure comparability of different types of data within the unified feature space. When using the sliding window algorithm to detect anomalous mutation points, it dynamically adjusts the window size and sliding step based on data characteristics. The improved Apriori algorithm filters and optimizes fault features during association rule mining to improve the accuracy and effectiveness of the association rules.

[0097] The report generation module includes a "report generation agent" that automatically generates interpretable reports using a template engine and optimizes the readability of the output using a BERT semantic matching model, ensuring a 100% conversion rate of technical terminology. When automatically generating interpretable reports using the template engine, the "report generation agent" selects the appropriate report template based on different suspected root cause types. The report content includes a fault description, diagnostic process, suspected root causes, and recommended measures. When optimizing the readability of the output using the BERT semantic matching model, it transforms technical terms into easily understandable expressions, ensuring that the converted content remains consistent with the original technical meaning. The generated reports can be output in multiple formats, such as PDF and Word, to meet the needs of different users.

[0098] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for multi-agent collaborative identification and root cause localization of ENodeB batch outages, characterized in that: Includes the following steps: S1: The "problem rewriting agent" collects sentence structure information of operators' performance query intentions, uses natural language processing technology to extract keywords and identify intents from the fault descriptions input by users, and generates structured query statements containing fault time range, base station group characteristics, and performance indicator thresholds. S2: The "Scene Classification Intelligent Agent" implements scene classification decisions, identifies fault types through a pre-trained classification model, determines whether it is a wireless professional ENodeB base station outage scenario, and calls the historical case library to match typical scenarios with a similarity greater than 85%. S3: The "Path Planning Agent" dynamically generates a diagnostic path tree based on the classification results. The diagnostic paths are as follows: Business Engineering Status → Circuit Type → Transmission Type → Uplink BRAS Fault Type → Business Complaint Status → Online User Count Status → Self-Equipment Alarm Fault Status → Environmental Alarm Fault Status → Event Resource Error Correction Status. S4: "Tool Recall and Invocation Agent" constructs a tool invocation matrix to recall associated API toolkits from the network management system, performance platform, and configuration library, including circuit-related transmission equipment fault verification (Tool-101), optical path-related transmission equipment fault verification (Tool-102), uplink BRAS fault troubleshooting (Tool-103), core service complaint status (Tool-104), obtaining online user count (Tool-105), self-equipment alarm fault troubleshooting capability (Tool-106), environmental alarm fault troubleshooting capability (Tool-107), verifying service engineering status (Tool-108), and verifying event resource errata (Tool-109); S5: The "Tool Recall and Invocation Agent" executes multiple rounds of tool invocation in a loop according to the "request-response-decision" pattern; S6: The "Root Cause Localization Agent" constructs a multi-dimensional feature space, maps the alarm data, performance indicators, and configuration parameters returned by the tool to a unified feature space, and uses a sliding window algorithm to detect abnormal mutation points; S7: The "Root Cause Localization Agent" performs association rule mining, discovers strong association rules of fault features through the improved Apriori algorithm, with support greater than 0.7 and confidence greater than 0.85, and outputs the TOP3 suspected root causes; S8: The "Report Generation Agent" uses a template engine to automatically generate interpretable reports; S9: The "Report Generation Agent" optimizes the readability of the output results through the BERT semantic matching model, ensuring a 100% conversion rate of technical terms.

2. The method for multi-agent collaborative identification and root cause localization of ENodeB batch outages according to claim 1, characterized in that: The "problem rewriting agent" collects sentence structure information on operators' query performance intent, covering a variety of common fault description patterns. Natural language processing technology combines lexical analysis, syntactic analysis, and semantic understanding to accurately extract keywords and identify user intent.

3. The method for multi-agent collaborative identification and root cause localization of ENodeB batch outages according to claim 2, characterized in that: The "Scene Classification Intelligent Agent" uses a pre-trained classification model trained on a large amount of historical fault data. It classifies fault types through deep learning algorithms. When calling the historical case library, it uses a similarity calculation algorithm to comprehensively consider multi-dimensional information such as fault description, time, and location to match typical scenarios with a similarity greater than 85%.

4. The method for multi-agent collaborative identification and root cause localization of ENodeB batch outages according to claim 3, characterized in that: When constructing a multi-dimensional feature space, the "Root Cause Localization Agent" standardizes alarm data, performance indicators, and configuration parameters to ensure that different types of data are comparable in a unified feature space. When using the sliding window algorithm to detect abnormal mutation points, it dynamically adjusts the window size and sliding step size according to the characteristics of the data.

5. The method for multi-agent collaborative identification and root cause localization of ENodeB batch outages according to claim 4, characterized in that: When the "report generation agent" automatically generates interpretable reports using a template engine, it selects the appropriate report template based on different suspected root cause types. The report content includes fault description, diagnostic process, suspected root cause, and recommended action information. When optimizing the readability of the output results using the BERT semantic matching model, technical terms are transformed into easy-to-understand expressions, and the transformed content is kept consistent with the original technical meaning.

6. A system for the method of multi-agent collaborative identification of ENodeB batch outage root cause localization according to claim 5, characterized in that: include: The problem rewriting module has a "problem rewriting intelligent agent" which is used to collect sentence structure information of operators' query intent for performance. It uses natural language processing technology to extract keywords and identify intent from the fault description input by the user, and generates a structured query statement that includes fault time range, base station group characteristics and performance index thresholds. The scenario classification module is equipped with a "scenario classification intelligent agent" to implement scenario classification decisions. It identifies the fault type through a pre-trained classification model, determines whether it is a wireless professional ENodeB base station outage scenario, and calls the historical case library to match typical scenarios with a similarity greater than 85%. The path planning module includes a "path planning intelligent agent" used to dynamically generate a location path tree based on the classification results. The diagnostic paths are as follows: business engineering status → circuit type → transmission type → uplink BRAS fault type → business complaint status → number of online users status → self-equipment alarm fault status → environmental alarm fault status → event resource correction status. The tool recall and invocation module has a "tool recall and invocation intelligent agent" used to build a tool invocation matrix. It recalls related API toolkits from the network management system, performance platform, and configuration library, including circuit-related transmission equipment fault verification (Tool-101), optical path-related transmission equipment fault verification (Tool-102), uplink BRAS fault troubleshooting (Tool-103), core service complaint status (Tool-104), online user count acquisition (Tool-105), self-equipment alarm fault troubleshooting capability (Tool-106), environmental alarm fault troubleshooting capability (Tool-107), business engineering status verification (Tool-108), and event resource correction verification (Tool-109). It executes multiple rounds of tool invocation in a loop according to the "request-response-decision" pattern. The root cause localization module is equipped with a "root cause localization agent" to construct a multi-dimensional feature space. It maps the alarm data, performance indicators, and configuration parameters returned by the tool to a unified feature space, uses a sliding window algorithm to detect abnormal mutation points, performs association rule mining, and discovers strong association rules of fault features through an improved Apriori algorithm, with support greater than 0.7 and confidence greater than 0.85, and outputs the top 3 suspected root causes. The report generation module includes a "report generation agent" that uses a template engine to automatically generate interpretable reports and optimizes the readability of the output results using a BERT semantic matching model to ensure a 100% conversion rate of technical terms.

7. The system according to claim 6, characterized in that: In the problem rewriting module, the "problem rewriting agent" collects sentence information on the operator's query performance intent, covering a variety of common fault description patterns. Natural language processing technology uses a combination of lexical analysis, syntactic analysis, and semantic understanding to accurately extract keywords and identify user intent. The generated structured query statements can adapt to the data processing needs of subsequent modules.

8. The system according to claim 7, characterized in that: In the scene classification module, the "scene classification agent" uses a pre-trained classification model trained on a large amount of historical fault data. It classifies fault types through deep learning algorithms. When calling the historical case library, it uses a similarity calculation algorithm to comprehensively consider multi-dimensional information such as fault description, time, and location to match typical scenarios with a similarity greater than 85%. The matching results are then fed back to the path planning module to assist in generating a location path tree.

9. A system according to claim 8, characterized in that: In the root cause localization module, the "root cause localization agent" standardizes alarm data, performance indicators, and configuration parameters when constructing a multi-dimensional feature space to ensure that different types of data are comparable in a unified feature space. When using the sliding window algorithm to detect abnormal mutation points, it dynamically adjusts the window size and sliding step size according to the data characteristics. The improved Apriori algorithm filters and optimizes fault features when mining association rules to improve the accuracy and effectiveness of association rules.

10. A system according to claim 9, characterized in that: In the report generation module, when the "report generation agent" automatically generates an interpretable report using a template engine, it selects the appropriate report template according to different suspected root cause types. The report content includes fault description, diagnosis process, suspected root cause, and recommended measures information. When optimizing the readability of the output results using the BERT semantic matching model, technical terms are transformed into easy-to-understand expressions, and the transformed content is kept consistent with the original technical meaning. The generated report can be output in multiple formats, such as PDF and Word, to meet the needs of different users.

Images