Machine learning based wireless network device fault root cause localization method and system

Abstract

The application discloses a wireless network equipment fault root cause positioning method and system based on machine learning, belongs to the field of communication technology, and aims to solve the technical problem of how to quickly perform root cause analysis and positioning on wireless equipment. The method comprises the following steps: acquiring equipment and topological relation data required for wireless network equipment fault root cause positioning; performing data preprocessing on the acquired equipment and topological relation data; storing the acquired equipment log data and the preprocessed data into a distributed message queue system; forming an alarm data hierarchical relationship, constructing a knowledge base based on alarm levels, alarm descriptions and root cause descriptions; constructing a root cause prediction model based on deep learning, predicting a root cause of a new alarm description through the trained root cause prediction model to obtain a root cause level; and calculating the root cause level based on the obtained root cause level according to the similarity between the root cause and the alarm.

Description

Technical Field

[0001] This invention relates to the field of communication technology, specifically to a method and system for locating the root cause of faults in wireless network equipment based on machine learning. Background Technology

[0002] Troubleshooting wireless equipment is a crucial aspect of modern communication maintenance. In modern communication networks, the failure rate of wireless equipment is significantly higher than that of wired equipment; therefore, wireless equipment fault location techniques are receiving increasing attention and research.

[0003] Currently, the main technologies used for locating the causes of wireless device failures include the following:

[0004] 1. Wireless positioning technology is a geolocation-based technology that locates the cause of a malfunction by pinpointing the location of the device. This technology uses information such as GPS or base station signal strength, combined with mathematical algorithms and data processing techniques, to calculate the precise location of the device and thus determine the cause of the malfunction.

[0005] 2. Wireless signal analysis technology

[0006] Wireless signal analysis technology is a technique that analyzes wireless signals to determine the cause of a fault. This technology collects and analyzes various parameters of the wireless signal, such as signal strength, signal amplitude, and signal frequency, to identify anomalies and thus determine the cause of the fault.

[0007] 3. Network analysis techniques

[0008] Network analysis technology is a technique based on network analysis that uses methods such as network topology, network load, and traffic analysis to determine the cause of a fault. This technology mainly includes several aspects such as network topology analysis, traffic analysis, and load analysis, using analysis of network topology, data traffic, and load conditions to identify the cause of a fault.

[0009] Currently, the following problems exist in the location of wireless device faults: First, the accuracy and precision of wireless device fault location technology need to be improved; second, the cost of wireless device fault location technology is high, requiring a large investment of manpower and resources; in addition, the real-time performance and reliability of wireless device fault location technology also need to be further improved.

[0010] The main reason for the above problems and shortcomings is the complexity and diversity of wireless device fault location technologies. Furthermore, the application scenarios for wireless device fault location technologies are also quite complex, requiring consideration of various factors such as environmental factors, equipment factors, and human factors.

[0011] How to quickly perform root cause analysis and localization of wireless devices is a technical problem that needs to be solved. Summary of the Invention

[0012] The technical objective of this invention is to address the above-mentioned shortcomings by providing a machine learning-based wireless network device fault location system to solve the technical problem of how to quickly perform root cause analysis and location of wireless devices.

[0013] In a first aspect, the present invention provides a method for locating the root cause of faults in wireless network devices based on machine learning, comprising the following steps:

[0014] Data acquisition: Acquire device and topology relationship data required for root cause localization of wireless network device faults. The device and topology relationship data includes device information, topology relationship information, device operation data, and environmental data.

[0015] Data preprocessing: The acquired device and topology relationship data are preprocessed to remove null values, outliers, and noise, resulting in preprocessed data.

[0016] Data storage: The acquired device log data and preprocessed data are stored in a distributed message queue system;

[0017] Alarm Analysis: Based on time windows and clustering algorithms, the data stored in the distributed message queue system is clustered according to the alarms within a predetermined time period. Then, the association rule mining algorithm is used to summarize alarms with the same root cause into generalized warnings that can cover the alarm content, forming a hierarchical relationship of alarm data.

[0018] Knowledge base construction: Feature extraction and cluster analysis are performed on alarm data to obtain alarm descriptions. Based on the alarm data, the root cause of the fault is located through visualization tools. A knowledge base is constructed based on alarm level, alarm description and root cause description.

[0019] Fault Root Cause Prediction: Sample data is constructed based on alarm description and root cause description. A root cause prediction model is constructed based on deep learning. The root cause prediction model is used to predict the output root cause level based on alarm description. The root cause prediction model is trained based on sample data. The root cause prediction model is used to predict the root cause of new alarm descriptions to obtain the root cause level.

[0020] Fault root cause localization: Based on the obtained root cause level, calculate the root cause level according to the similarity between the root cause and the alarm.

[0021] Preferably, the equipment information is obtained from the OMC, including equipment model, equipment status and equipment configuration, which is used to determine the type and location of equipment failure;

[0022] The equipment operation data includes the equipment's operating status, signal strength, and data transmission rate, which are used to predict potential equipment failures and provide a basis for fault diagnosis and location.

[0023] The device topology information is obtained from the resource management system, including the connection relationships between devices and signal transmission paths, which is used to determine the specific location of device failures.

[0024] The environmental information refers to the data on the computer room environment obtained from the environmental monitoring system, including the computer room temperature and humidity.

[0025] Preferably, the acquired device and topology relationship data is preprocessed, including the following steps:

[0026] Remove outliers and missing values ​​from device and topology relationship data;

[0027] Normalize the device and topology relationship data;

[0028] Noisy data is removed from device and topology relationship data based on clustering methods;

[0029] Noise data in device and topology relationship data is predicted based on regression methods, and the predicted noise data is then removed.

[0030] As a preferred approach, alarm analysis includes the following steps:

[0031] The data stored in the distributed message queue system is divided into multiple time periods according to the time window. The data in each time period includes alarm data.

[0032] For alarm data within each time period, a clustering algorithm is used to cluster the alarm data;

[0033] The Apriori algorithm is used to discover frequent itemsets and association rules among alarm data. Alarm data with the same root cause are grouped into one category, and the association rules are used as the content of generalized warnings. The generalized warnings are used as a summary of the alarm data.

[0034] Hierarchical clustering algorithm is used to cluster alarm data according to different levels.

[0035] As a preferred approach, knowledge base construction includes the following steps:

[0036] Key features of alarm data are extracted using feature engineering methods. Based on these key features, clustering methods are used to merge and cluster similar alarm data to obtain an alarm summary. The key features include the data center name, network element name, and alarm description. The alarm summary includes the data center name, network element name, alarm description, and number of alarms. The alarm description is the specific content of the alarm.

[0037] Alarms are classified into multiple levels according to their importance;

[0038] Based on alarm data, the root cause of the fault is located through visualization tools to obtain a root cause description;

[0039] A knowledge base is built based on alarm level, alarm description, and root cause description.

[0040] As a preferred method, root cause failure prediction includes the following steps:

[0041] Root causes are divided into multiple levels, and a hierarchical relationship is established between the root cause levels;

[0042] A root cause prediction model is constructed based on the seq2seq model. The root cause prediction model is used to predict the root cause level with alarm summary as input.

[0043] The alarm summary and root cause level are encoded, and the encoded alarm summary and root cause level are used as training samples to train the root cause prediction model, thus obtaining the trained root cause prediction model.

[0044] The new alarm summary is encoded, and the encoded alarm summary is used as the input sequence to input the trained root cause prediction model. The trained root cause prediction model outputs the root cause level corresponding to the new alarm summary.

[0045] The new alarm summary and the root cause level of the predicted output are added to the training samples to train the root cause prediction model.

[0046] In a second aspect, the present invention provides a machine learning-based wireless network device fault location system, used to locate the root cause of a wireless network device fault using the machine learning-based wireless network device fault root cause location method as described in any one of the first aspects, the system comprising:

[0047] The data acquisition module is used to acquire device and topology relationship data required for root cause localization of wireless network device faults. The device and topology relationship data includes device information, topology relationship information, device operation data, and environmental data.

[0048] The data preprocessing module is used to preprocess the acquired device and topology relationship data, and remove null values, outliers and noise through data preprocessing to obtain preprocessed data;

[0049] The data storage module is used to store the acquired device log data and preprocessed data into a distributed message queue system;

[0050] The alarm analysis module is used to cluster alarms stored in the distributed message queue system according to a predetermined time period based on time windows and clustering algorithms, and to use association rule mining algorithms to summarize alarms with the same root cause into generalized warnings that can cover the alarm content, thus forming a hierarchical relationship of alarm data.

[0051] The knowledge base construction module is used to perform feature extraction and cluster analysis on alarm data to obtain alarm descriptions, and to locate the root cause of the fault based on the alarm data and visualization tools, and to construct a knowledge base based on alarm level, alarm description and root cause description.

[0052] The fault root cause prediction module is used to construct sample data based on alarm description and root cause description, construct a root cause prediction model based on deep learning, and use the root cause prediction model to predict the output root cause level based on alarm description. The root cause prediction model is trained based on sample data, and the root cause prediction model is used to predict the root cause of new alarm descriptions to obtain the root cause level.

[0053] A fault root cause localization module is used to calculate the root cause level based on the obtained root cause level and the similarity between the root cause and the alarm.

[0054] Preferably, the equipment information is obtained from the OMC, including equipment model, equipment status and equipment configuration, which is used to determine the type and location of equipment failure;

[0055] The equipment operation data includes the equipment's operating status, signal strength, and data transmission rate, which are used to predict potential equipment failures and provide a basis for fault diagnosis and location.

[0056] The device topology information is obtained from the resource management system, including the connection relationships between devices and signal transmission paths, which is used to determine the specific location of device failures.

[0057] The environmental information refers to the data on the computer room environment obtained from the environmental monitoring system, including the computer room temperature and humidity.

[0058] The data preprocessing module is used to perform the following data preprocessing on the acquired device and topology relationship data:

[0059] Remove outliers and missing values ​​from device and topology relationship data;

[0060] Normalize the device and topology relationship data;

[0061] Noisy data is removed from device and topology relationship data based on clustering methods;

[0062] Noise data in device and topology relationship data is predicted based on regression methods, and the predicted noise data is then removed.

[0063] Preferably, the alarm analysis module is used to perform the following:

[0064] The data stored in the distributed message queue system is divided into multiple time periods according to the time window. The data in each time period includes alarm data.

[0065] For alarm data within each time period, a clustering algorithm is used to cluster the alarm data;

[0066] The Apriori algorithm is used to discover frequent itemsets and association rules among alarm data. Alarm data with the same root cause are grouped into one category, and the association rules are used as the content of generalized warnings. The generalized warnings are used as a summary of the alarm data.

[0067] Hierarchical clustering algorithm is used to cluster alarm data according to different levels.

[0068] Preferably, the knowledge base construction module is used to perform the following:

[0069] Key features of alarm data are extracted using feature engineering methods. Based on these key features, clustering methods are used to merge and cluster similar alarm data to obtain an alarm summary. The key features include the data center name, network element name, and alarm description. The alarm summary includes the data center name, network element name, alarm description, and number of alarms. The alarm description is the specific content of the alarm.

[0070] Alarms are classified into multiple levels according to their importance;

[0071] Based on alarm data, the root cause of the fault is located through visualization tools to obtain a root cause description;

[0072] A knowledge base is built based on alarm level, alarm description, and root cause description;

[0073] The root cause failure prediction module is used to perform the following:

[0074] Root causes are divided into multiple levels, and a hierarchical relationship is established between the root cause levels;

[0075] A root cause prediction model is constructed based on the seq2seq model. The root cause prediction model is used to predict the root cause level with alarm summary as input.

[0076] The alarm summary and root cause level are encoded, and the encoded alarm summary and root cause level are used as training samples to train the root cause prediction model, thus obtaining the trained root cause prediction model.

[0077] The new alarm summary is encoded, and the encoded alarm summary is used as the input sequence to input the trained root cause prediction model. The trained root cause prediction model outputs the root cause level corresponding to the new alarm summary.

[0078] The new alarm summary and the root cause level of the predicted output are added to the training samples to train the root cause prediction model.

[0079] The machine learning-based method and system for locating the root cause of wireless network device faults of the present invention have the following advantages:

[0080] 1. A knowledge base is built based on alarm level, alarm description and root cause description. A root cause prediction model is trained based on alarm description and root cause description. The root cause prediction model is used to predict the root cause of new alarm descriptions. The root cause level is calculated based on the predicted root cause level and the similarity between the root cause and the alarm. This achieves fast and accurate root cause prediction.

[0081] 2. Based on time windows and clustering algorithms, the device log data stored in the distributed message queue system and the preprocessed device and topology relationship data are analyzed to obtain the hierarchical relationship of alarm data. Through multi-level alarms and multi-level root causes, the accuracy of root cause prediction is improved.

[0082] 3. For alarm data, key features are extracted using feature engineering methods. Based on the key features, similar alarm data are merged and clustered using clustering methods to obtain alarm summaries. When performing root cause prediction using the root cause prediction model, the alarm summaries are used as the input sequence and the root cause level is used as the output sequence. The samples constructed based on the alarm summaries and root cause levels enable the root cause prediction model to be adapted to multiple wireless network devices.

[0083] 4. After predicting the new alarm summary by the root cause prediction model, the new alarm summary and the corresponding root cause level are added to the training samples, which realizes the retraining of the root cause prediction model after training and improves the accuracy of the model. Attached Figure Description

[0084] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0085] The invention will be further described below with reference to the accompanying drawings.

[0086] Figure 1This is a flowchart of a machine learning-based method for locating the root cause of wireless network device faults, as described in Example 1. Detailed Implementation

[0087] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments are not intended to limit the present invention. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.

[0088] This invention provides a method and system for root cause localization of wireless network device faults based on machine learning, which is used to solve the technical problem of how to quickly perform root cause analysis and localization of wireless devices.

[0089] Example 1:

[0090] This invention provides a machine learning-based method for locating the root cause of faults in wireless network devices, comprising the following steps:

[0091] S100, Data Acquisition: Acquire device and topology relationship data required for root cause localization of wireless network device faults. The device and topology relationship data includes device information, topology relationship information, device operation data, and environmental data.

[0092] S200, Data Preprocessing: Preprocess the acquired device and topology relationship data to remove null values, outliers and noise, and obtain the preprocessed data.

[0093] S300, Data Storage: The acquired device log data and preprocessed data are stored in a distributed message queue system;

[0094] S400 Alarm Analysis: Based on time windows and clustering algorithms, the data stored in the distributed message queue system is clustered according to alarms within a predetermined time period. Then, the association rule mining algorithm is used to summarize alarms with the same root cause into generalized warnings that can cover the alarm content, forming a hierarchical relationship of alarm data.

[0095] S500 Knowledge Base Construction: Perform feature extraction and cluster analysis on alarm data to obtain alarm descriptions, and locate the root cause of the fault based on the alarm data and visualization tools. Construct a knowledge base based on alarm level, alarm description and root cause description.

[0096] S600, Fault Root Cause Prediction: Based on alarm description and root cause description, sample data is constructed, and a root cause prediction model is constructed based on deep learning. The root cause prediction model is used to predict the output root cause level based on alarm description. The root cause prediction model is trained based on sample data. The root cause prediction model is used to predict the root cause of new alarm descriptions through the trained root cause prediction model to obtain the root cause level.

[0097] S700, Fault Root Cause Location: Based on the obtained root cause level, calculate the root cause level according to the similarity between the root cause and the alarm.

[0098] Locating the cause of a wireless device malfunction requires inputting various types of data. In this embodiment, step S100 collects data including device information, topology information, device operating data, and environmental data. Specifically:

[0099] Equipment information: Equipment information needs to be obtained from OMC, including equipment model, equipment status, equipment configuration, etc. This information can help determine the type and location of equipment failure;

[0100] Equipment operation data: This includes the equipment's operating status, signal strength, data transmission rate, etc. This data can help predict potential equipment failures and provide a basis for fault diagnosis and location.

[0101] Obtaining equipment topology information, equipment configuration information, optical path information, circuit information, and topology relationship information from the resource management system, including the connection relationship between devices and signal transmission paths, can help determine the specific location of equipment failure.

[0102] Data on the computer room environment, such as temperature and humidity, needs to be obtained from the environmental monitoring system. This data can be obtained through environmental monitoring equipment or sensors.

[0103] When collecting data required for fault location of wireless devices, noise may be present in the data due to various reasons. To improve the accuracy and reliability of the data, step S200 uses clustering and regression methods to remove noise from the data.

[0104] This embodiment uses clustering to separate noise from the data. Clustering is an unsupervised learning method that divides data into multiple categories, each with similar characteristics. When noise is present in the data, clustering can separate the noisy data, thereby improving the accuracy and reliability of the data. The K-means clustering algorithm can be used to divide the data into multiple categories, and then the noisy data can be removed.

[0105] Regression methods are used to predict noise in data. Regression is a supervised learning method that predicts noise in data by modeling and analyzing it. A linear regression model can be used to predict noise in the data, and then the predicted noise can be removed from the data.

[0106] Before using clustering and regression methods to remove noise from data, it is necessary to remove outliers and null values, and to normalize the data to better suit the requirements of clustering algorithms and regression models. Furthermore, the clustering results and regression models need to be evaluated and validated to ensure their accuracy and reliability.

[0107] Meanwhile, in this embodiment, step S100 also obtains device log data from the device interface of the wireless network device, and step S300 stores the obtained device log data and preprocessed data in the distributed message queue system.

[0108] This embodiment uses Kafka as the distributed message queue system, which can achieve high throughput and low latency message transmission. When using the Kafka Producer API, a Producer object needs to be created and its configuration information configured. Then, the Producer object can be used to send the received log content to the Kafka queue. When sending a message, information such as the message's Topic and Partition needs to be specified.

[0109] If the parsing format is consistent, data can be stored in the same Topic. In a Kafka team, a Topic is a logical category of messages; messages of the same type can be stored in the same Topic. Similarly, if the parsing format is consistent, noise-removed data can be stored in the same Topic. When storing data, it's necessary to specify information such as the Topic and Partition for better data management and maintenance.

[0110] The alarm analysis in step S400 includes the following operations:

[0111] (1) Divide the data stored in the distributed message queue system into multiple time periods according to the time window. The data in each time period includes alarm data.

[0112] (2) For alarm data within each time period, clustering algorithm is used to cluster the alarm data;

[0113] (3) Discover frequent itemsets and association rules among alarm data through the Apriori algorithm, classify alarm data with the same root cause into one category, use association rules as the content of generalized warnings, and use generalized warnings as a summary of alarm data.

[0114] (4) Use hierarchical clustering algorithm to cluster alarm data according to different levels.

[0115] In this embodiment, the alarm analysis uses time windows and clustering algorithms to cluster stored data based on alarms within a certain time period. A time window is a method of segmenting data by time, dividing data into multiple time periods at specific intervals. Within each time period, a clustering algorithm is used to cluster the alarm data.

[0116] Algorithms using association rule mining are used to group alarms with the same root cause into generalized alerts that encompass the alarm content. The Apriori algorithm is used to discover frequent itemsets and association rules among alarm data. Algorithms with the same root cause are then grouped together, and these association rules are used as the content of the generalized alerts. Finally, the generalized alerts are used as a summary of the alarm data for better understanding and analysis.

[0117] Hierarchical clustering algorithms are used to cluster alarm data according to different levels, thus preserving details while maintaining strong generalization. Association rule mining algorithms can discover association rules between alarm data, thereby providing a better understanding of the relationships between alarm data.

[0118] To obtain the root cause of abnormal warnings for wireless devices, the process can be divided into the following four parts: collecting warning information, extracting key features of warning information, clustering, and displaying warning summaries.

[0119] Part 1: Collecting Early Warning Information. Collecting early warning information is the first step in identifying the root cause of wireless device anomalies. This information can be collected through multiple systems, such as device management systems, network management systems, and environmental monitoring systems. When collecting early warning information, it is necessary to select appropriate data sources and preprocess and clean the data to ensure its accuracy and reliability. This embodiment uses alarm data stored in a distributed message queue system as early warning information.

[0120] Part Two: Extracting Key Features from Warning Information. Extracting key features from warning information is a crucial step in identifying the root cause of wireless device anomaly warnings. Feature engineering methods can be used to extract these key features. Feature engineering is a method that transforms raw data into features that can be used in machine learning algorithms.

[0121] Part Three: Clustering Processing. Clustering algorithms are used to divide the warning information into multiple categories, with each category containing warning information exhibiting similar characteristics.

[0122] Part Four: Presenting Alarm Summary. Presenting alarm summary is a method of visualizing clustered alarm data. Visualization tools can be used to present alarm data graphically, thereby better understanding the relationships between alarm data.

[0123] In this embodiment, feature engineering methods are used to extract key features of alarm information, such as data center name, network element name (number), and alarm description. Clustering algorithms are then used to merge and cluster similar alarm information.

[0124] Clustering was performed using three dimensions: data center name, network element name (number), and alarm description. Feature-based clustering is a method that groups data according to features, allowing data with similar characteristics to be grouped together. The resulting alarm summaries are shown in Table 1 below.

[0125] Table 1:

[0126]

[0127]

[0128] In this embodiment, the knowledge base is a system for storing and managing knowledge, which can be used to record and share knowledge. When defining the knowledge base, its structure and content need to be considered. The knowledge base can be divided into three parts: levels, alarm descriptions, and root cause descriptions.

[0129] Level: The level indicates the importance of an alert, and can be divided into multiple levels, such as urgent, important, and general. When defining the knowledge base, alerts of different levels need to be categorized and their level information recorded.

[0130] Alarm Description: The alarm description is the specific content of the alarm, which may include information such as the alarm name, description, and type.

[0131] Root cause description: The root cause description is the underlying reason for the alarm, and may include information such as the source of the alarm, its impact, and solutions. The knowledge base presentation structure is shown in Table 2.

[0132] Table 2:

[0133]

[0134] Based on the above analysis, step S500 of this embodiment constructs a knowledge base, including the following steps:

[0135] (1) Extract key features of alarm data through feature engineering method, and merge and cluster similar alarm data based on key features using clustering method to obtain alarm summary. Key features include computer room name, network element name and alarm description. The alarm summary includes computer room name, network element name, alarm description and number of alarms. The alarm description is the specific content of the alarm.

[0136] (2) According to the importance of the alarm, the alarms are divided into multiple levels;

[0137] (3) Based on alarm data, use visualization tools to locate the root cause of the fault and obtain the root cause description;

[0138] (4) Construct a knowledge base based on alarm level, alarm description and root cause description.

[0139] In this embodiment, after receiving an alarm, the optimization personnel use visualization tools to quickly locate the general direction of the fault. Visualization tools can present alarm data graphically, thereby better understanding the relationships between alarm data. Scatter plots, heatmaps, and other methods are used to present alarm data to facilitate a better understanding of the relationships between alarm data. Simultaneously, search tools can also be used to quickly locate the general direction of the fault. Search tools can search alarm data based on keywords, time, location, and other conditions, thereby better pinpointing the general direction of the fault.

[0140] Step S600 uses a seq2seq model to predict the root cause of alarms. This includes the following steps:

[0141] (1) Divide the root causes into multiple levels and establish the hierarchical relationship between the root cause levels;

[0142] (2) Construct a root cause prediction model based on the seq2seq model. The root cause prediction model is used to predict the root cause level with alarm summary as input.

[0143] (3) Encode the alarm summary and root cause level, use the encoded alarm summary and root cause level as training samples, train the root cause prediction model through the training samples, and obtain the trained root cause prediction model.

[0144] (4) Encode the new alarm summary, use the encoded alarm summary as the input sequence to input the trained root cause prediction model, and output the root cause level corresponding to the new alarm summary through the trained root cause prediction model.

[0145] (5) Add the new alarm summary and the root cause level of the prediction output to the training samples to train the root cause prediction model.

[0146] As a specific implementation, root cause prediction involves the following steps:

[0147] Data preprocessing: Alarm descriptions and root causes need to be encoded and decoded in order to be input into the seq2seq model for training and prediction.

[0148] Model training: Training data is needed to train the seq2seq model so that the model can learn the relationship between alarm descriptions and root causes;

[0149] Model prediction: A trained seq2seq model is needed to predict the new alarm description in order to predict its primary and secondary root causes.

[0150] Results Display: The prediction results need to be displayed so that users can view and understand them. Additionally, writing the prediction results into a knowledge base as subsequent training data can improve the model's intelligent performance.

[0151] This embodiment uses a clustering algorithm and a seq2seq text generation algorithm to output the root cause level based on the device's alarm summary. The specific steps are as follows:

[0152] (1) Data preprocessing: The alarm summaries of the equipment need to be preprocessed and cleaned;

[0153] (2) Clustering: Clustering algorithms can be used to group device alarms according to their similarity, thereby better understanding the relationship between alarms;

[0154] (3) seq2seq text generation algorithm: The seq2seq text generation algorithm is used to convert the alarm summary of the device into an alarm description. When using the seq2seq text generation algorithm for fault root cause prediction, the alarm summary can be used as the input sequence and the root cause level can be used as the output sequence. Then, the seq2seq model is used for training and prediction.

[0155] (4) Results display: The prediction results need to be displayed so that users can view and understand them.

[0156] If the transmission data and environmental data have poor matching, hierarchical clustering can be used for data matching analysis. Hierarchical clustering is a method of grouping data according to similarity, which can group data with similar characteristics into one category. Then, clustering algorithms can be used to group data with similar characteristics into one category.

[0157] For cases with poor matching, text generation technology can be used to generate alarm causes. When using text generation technology to generate alarm causes, transmission data and environmental data can be used as input sequences, and alarm causes can be used as output sequences. Then, a text generation model can be used for training and prediction.

[0158] In step S700, during fault root cause localization, the primary and secondary root cause faults are determined based on the similarity calculation of root cause alarms. The specific steps are as follows:

[0159] (1) Root cause alarm similarity calculation. The root cause alarm similarity calculation method can be used to calculate the similarity between faults. The root cause alarm similarity calculation method is a method to calculate the similarity between the root causes of faults, which can be used for fault root cause localization;

[0160] (2) Primary and Secondary Root Cause Faults: Based on the clustering and root cause alarm similarity calculations, primary and secondary root cause faults can be identified. Primary root cause faults refer to the root cause of the fault, while secondary root cause faults refer to the secondary causes. The results are presented in Table 3.

[0161] Machine room name Network element Alarm summary Number of items Primary root cause Secondary root cause A X1 xxxx XXX xxx xxx A X2 xxxx XXX xxx xxx B … … XXX … …

[0162] This embodiment performs root cause analysis and location of base station problems discovered through centralized maintenance, which can bring multiple benefits, including improving the service quality and user satisfaction of the communication network, reducing base station maintenance and labor costs, and improving the accuracy and efficiency of base station fault diagnosis.

[0163] Example 2:

[0164] The present invention discloses a fault location system for wireless network devices based on machine learning, comprising a data acquisition module, a data preprocessing module, a data storage module, an alarm analysis module, a knowledge base construction module, a fault root cause prediction module, and a fault root cause location module. The system can locate the fault root cause of wireless network devices using the method disclosed in Embodiment 1.

[0165] The data acquisition module is used to acquire device and topology relationship data needed for root cause localization of wireless network device faults. The device and topology relationship data includes device information, topology relationship information, device operation data, and environmental data.

[0166] Locating the cause of wireless device malfunctions requires inputting various types of data. In this embodiment, the data acquisition module collects data including device information, topology information, device operating data, and environmental data. Specifically:

[0167] Equipment information: Equipment information needs to be obtained from OMC, including equipment model, equipment status, equipment configuration, etc. This information can help determine the type and location of equipment failure;

[0168] Equipment operation data: This includes the equipment's operating status, signal strength, data transmission rate, etc. This data can help predict potential equipment failures and provide a basis for fault diagnosis and location.

[0169] Obtaining equipment topology information, equipment configuration information, optical path information, circuit information, and topology relationship information from the resource management system, including the connection relationship between devices and signal transmission paths, can help determine the specific location of equipment failure.

[0170] Data on the computer room environment, such as temperature and humidity, needs to be obtained from the environmental monitoring system. This data can be obtained through environmental monitoring equipment or sensors.

[0171] The data preprocessing module is used to preprocess the acquired device and topology relationship data. By removing null values, outliers, and noise, the preprocessed data is obtained.

[0172] When collecting data for wireless device fault location, noise may be present in the data due to various reasons. To improve the accuracy and reliability of the data, a data preprocessing module is used to remove noise from the data using clustering and regression methods.

[0173] This embodiment uses clustering to separate noise from the data. Clustering is an unsupervised learning method that divides data into multiple categories, each with similar characteristics. When noise is present in the data, clustering can separate the noisy data, thereby improving the accuracy and reliability of the data. The K-means clustering algorithm can be used to divide the data into multiple categories, and then the noisy data can be removed.

[0174] Regression methods are used to predict noise in data. Regression is a supervised learning method that predicts noise in data by modeling and analyzing it. A linear regression model can be used to predict noise in the data, and then the predicted noise can be removed from the data.

[0175] Before using clustering and regression methods to remove noise from data, it is necessary to remove outliers and null values, and to normalize the data to better suit the requirements of clustering algorithms and regression models. Furthermore, the clustering results and regression models need to be evaluated and validated to ensure their accuracy and reliability.

[0176] The data storage module is used to store the acquired device log data and preprocessed data into a distributed message queue system.

[0177] In this embodiment, the data acquisition module is also used to obtain device log data from the device interface of the wireless network device, and the data storage module is used to store the obtained device log data and preprocessed data into the distributed message queue system.

[0178] This embodiment uses Kafka as the distributed message queue, which can achieve high throughput and low latency message transmission. When using the Kafka Producer API, a Producer object needs to be created and its configuration information configured. Then, the Producer object can be used to send the received log content to the Kafka queue. When sending a message, information such as the message's Topic and Partition needs to be specified.

[0179] If the parsing format is consistent, data can be stored in the same Topic. In a Kafka team, a Topic is a logical category of messages; messages of the same type can be stored in the same Topic. Similarly, if the parsing format is consistent, noise-removed data can be stored in the same Topic. When storing data, it's necessary to specify information such as the Topic and Partition for better data management and maintenance.

[0180] The alarm analysis module is used to cluster alarms stored in the distributed message queue system according to a predetermined time period based on time windows and clustering algorithms. It also uses association rule mining algorithms to summarize alarms with the same root cause into generalized warnings that can cover the alarm content, forming a hierarchical relationship of alarm data.

[0181] As a specific implementation of the alarm analysis module, this embodiment uses the module to perform alarm analysis as follows:

[0182] (1) Divide the data stored in the distributed message queue system into multiple time periods according to the time window. The data in each time period includes alarm data.

[0183] (2) For alarm data within each time period, clustering algorithm is used to cluster the alarm data;

[0184] (3) Discover frequent itemsets and association rules among alarm data through the Apriori algorithm, classify alarm data with the same root cause into one category, use association rules as the content of generalized warnings, and use generalized warnings as a summary of alarm data.

[0185] (4) Use hierarchical clustering algorithm to cluster alarm data according to different levels.

[0186] In this embodiment, the alarm analysis uses time windows and clustering algorithms to cluster stored data based on alarms within a certain time period. A time window is a method of segmenting data by time, dividing data into multiple time periods at specific intervals. Within each time period, a clustering algorithm is used to cluster the alarm data.

[0187] Algorithms using association rule mining are used to group alarms with the same root cause into generalized alerts that encompass the alarm content. The Apriori algorithm is used to discover frequent itemsets and association rules among alarm data. Algorithms with the same root cause are then grouped together, and these association rules are used as the content of the generalized alerts. Finally, the generalized alerts are used as a summary of the alarm data for better understanding and analysis.

[0188] Hierarchical clustering algorithms are used to cluster alarm data according to different levels, thus preserving details while maintaining strong generalization. Association rule mining algorithms can discover association rules between alarm data, thereby providing a better understanding of the relationships between alarm data.

[0189] The knowledge base construction module is used to extract features and perform cluster analysis on alarm data to obtain alarm descriptions. Based on the alarm data, the root cause of the fault is located through visualization tools, and a knowledge base is built based on the alarm level, alarm description, and root cause description.

[0190] To obtain the root cause of abnormal warnings for wireless devices, the process can be divided into the following four parts: collecting warning information, extracting key features of warning information, clustering, and displaying warning summaries.

[0191] Part 1: Collecting Early Warning Information. Collecting early warning information is the first step in identifying the root cause of wireless device anomalies. This information can be collected through multiple systems, such as device management systems, network management systems, and environmental monitoring systems. When collecting early warning information, it is necessary to select appropriate data sources and preprocess and clean the data to ensure its accuracy and reliability. This embodiment uses alarm data stored in a distributed message queue system as early warning information.

[0192] Part Two: Extracting Key Features from Warning Information. Extracting key features from warning information is a crucial step in identifying the root cause of wireless device anomaly warnings. Feature engineering methods can be used to extract these key features. Feature engineering is a method that transforms raw data into features that can be used in machine learning algorithms.

[0193] Part Three: Clustering Processing. Clustering algorithms are used to divide the warning information into multiple categories, with each category containing warning information exhibiting similar characteristics.

[0194] Part Four: Presenting Alarm Summary. Presenting alarm summary is a method of visualizing clustered alarm data. Visualization tools can be used to present alarm data graphically, thereby better understanding the relationships between alarm data.

[0195] In this embodiment, feature engineering methods are used to extract key features of alarm information, such as data center name, network element name (number), and alarm description. Clustering algorithms are then used to merge and cluster similar alarm information.

[0196] Clustering was performed using three dimensions: data center name, network element name (number), and alarm description. Feature-based clustering is a method that groups data according to features, which can group data with similar features into one category.

[0197] In this embodiment, the knowledge base is a system for storing and managing knowledge, which can be used to record and share knowledge. When defining the knowledge base, its structure and content need to be considered. The knowledge base can be divided into three parts: levels, alarm descriptions, and root cause descriptions.

[0198] Level: The level indicates the importance of an alert, and can be divided into multiple levels, such as urgent, important, and general. When defining the knowledge base, alerts of different levels need to be categorized and their level information recorded.

[0199] Alarm Description: The alarm description is the specific content of the alarm, which may include information such as the alarm name, description, and type.

[0200] Root cause description: The root cause description is the underlying reason for the alarm, which may include information such as the source of the alarm, its impact, and solutions.

[0201] Based on the above analysis, the knowledge base construction module in this embodiment is used to construct the knowledge base as follows:

[0202] (1) Extract key features of alarm data through feature engineering method, and merge and cluster similar alarm data based on key features using clustering method to obtain alarm summary. Key features include computer room name, network element name and alarm description. The alarm summary includes computer room name, network element name, alarm description and number of alarms. The alarm description is the specific content of the alarm.

[0203] (2) According to the importance of the alarm, the alarms are divided into multiple levels;

[0204] (3) Based on alarm data, use visualization tools to locate the root cause of the fault and obtain the root cause description;

[0205] (4) Construct a knowledge base based on alarm level, alarm description and root cause description.

[0206] In this embodiment, upon receiving an alarm, a visualization tool is used to quickly locate the approximate direction of the fault. The visualization tool can present alarm data graphically, thereby improving the understanding of the relationships between alarm data. Scatter plots, heatmaps, and other methods are used to present alarm data to better illustrate these relationships. Simultaneously, a search tool can also be used to quickly locate the approximate direction of the fault. The search tool can search for alarm data based on keywords, time, location, and other criteria, thereby further improving the ability to pinpoint the approximate direction of the fault.

[0207] The fault root cause prediction module is used to construct sample data based on alarm description and root cause description, and to construct a root cause prediction model based on deep learning. The root cause prediction model is used to predict the output root cause level based on alarm description. The root cause prediction model is trained based on sample data, and the root cause prediction is performed on new alarm descriptions through the trained root cause prediction model to obtain the root cause level.

[0208] In this embodiment, the fault root cause prediction module is used to predict alarm root causes using a seq2seq model. Specifically, the following operations are performed:

[0209] (1) Divide the root causes into multiple levels and establish the hierarchical relationship between the root cause levels;

[0210] (2) Construct a root cause prediction model based on the seq2seq model. The root cause prediction model is used to predict the root cause level with alarm summary as input.

[0211] (3) Encode the alarm summary and root cause level, use the encoded alarm summary and root cause level as training samples, train the root cause prediction model through the training samples, and obtain the trained root cause prediction model.

[0212] (4) Encode the new alarm summary, use the encoded alarm summary as the input sequence to input the trained root cause prediction model, and output the root cause level corresponding to the new alarm summary through the trained root cause prediction model.

[0213] (5) Add the new alarm summary and the root cause level of the prediction output to the training samples to train the root cause prediction model.

[0214] As a specific implementation, root cause prediction involves the following steps:

[0215] Data preprocessing: Alarm descriptions and root causes need to be encoded and decoded in order to be input into the seq2seq model for training and prediction.

[0216] Model training: Training data is needed to train the seq2seq model so that the model can learn the relationship between alarm descriptions and root causes;

[0217] Model prediction: A trained seq2seq model is needed to predict the new alarm description in order to predict its primary and secondary root causes.

[0218] Results Display: The prediction results need to be displayed so that users can view and understand them. Additionally, writing the prediction results into a knowledge base as subsequent training data can improve the model's intelligent performance.

[0219] This embodiment uses a clustering algorithm and a seq2seq text generation algorithm to output the root cause level based on the device's alarm summary. The specific steps are as follows:

[0220] (1) Data preprocessing: The alarm summaries of the equipment need to be preprocessed and cleaned;

[0221] (2) Clustering: Clustering algorithms can be used to group device alarms according to their similarity, thereby better understanding the relationship between alarms;

[0222] (3) seq2seq text generation algorithm: The seq2seq text generation algorithm is used to convert the alarm summary of the device into an alarm description. When using the seq2seq text generation algorithm for fault root cause prediction, the alarm summary can be used as the input sequence and the root cause level can be used as the output sequence. Then, the seq2seq model is used for training and prediction.

[0223] (4) Results display: The prediction results need to be displayed so that users can view and understand them.

[0224] If the transmission data and environmental data have poor matching, hierarchical clustering can be used for data matching analysis. Hierarchical clustering is a method of grouping data according to similarity, which can group data with similar characteristics into one category. Then, clustering algorithms can be used to group data with similar characteristics into one category.

[0225] For cases with poor matching, text generation technology can be used to generate alarm causes. When using text generation technology to generate alarm causes, transmission data and environmental data can be used as input sequences, and alarm causes can be used as output sequences. Then, a text generation model can be used for training and prediction.

[0226] The root cause localization module is used to calculate the root cause level based on the obtained root cause level and the similarity between the root cause and the alarm.

[0227] During root cause localization, the root cause localization module is used to calculate the primary and secondary root cause faults based on the similarity of root cause alarms. Specifically, it performs the following operations:

[0228] (1) Root cause alarm similarity calculation. The root cause alarm similarity calculation method can be used to calculate the similarity between faults. The root cause alarm similarity calculation method is a method to calculate the similarity between the root causes of faults, which can be used for fault root cause localization;

[0229] (2) Primary root cause faults and secondary root cause faults: Based on the results of clustering and root cause alarm similarity calculation, primary root cause faults and secondary root cause faults can be identified. Primary root cause faults refer to the root cause of the fault, while secondary root cause faults refer to the secondary cause of the fault.

[0230] The present invention has been shown and described in detail above with reference to the accompanying drawings and preferred embodiments. However, the present invention is not limited to these disclosed embodiments. Based on the above multiple embodiments, those skilled in the art will know that more embodiments of the present invention can be obtained by combining the means in the different embodiments described above, and these embodiments are also within the protection scope of the present invention.

Claims

1. A method for locating the root cause of faults in wireless network devices based on machine learning, characterized in that, Includes the following steps: Data acquisition: Acquire device and topology relationship data required for root cause localization of wireless network device faults. The device and topology relationship data includes device information, topology relationship information, device operation data, and environmental data. Data preprocessing: The acquired device and topology relationship data are preprocessed to remove null values, outliers, and noise, resulting in preprocessed data. Data storage: The acquired device log data and preprocessed data are stored in a distributed message queue system; Alarm Analysis: Based on time windows and clustering algorithms, the data stored in the distributed message queue system is clustered according to the alarms within a predetermined time period. Then, the association rule mining algorithm is used to summarize alarms with the same root cause into generalized warnings that can cover the alarm content, forming a hierarchical relationship of alarm data. Knowledge base construction: Feature extraction and cluster analysis are performed on alarm data to obtain alarm descriptions. Based on the alarm data, the root cause of the fault is located through visualization tools. A knowledge base is constructed based on alarm level, alarm description and root cause description. Fault Root Cause Prediction: Sample data is constructed based on alarm description and root cause description. A root cause prediction model is constructed based on deep learning. The root cause prediction model is used to predict the output root cause level based on alarm description. The root cause prediction model is trained based on sample data. The root cause prediction model is used to predict the root cause of new alarm descriptions to obtain the root cause level. Fault root cause localization: Based on the obtained root cause level, calculate the root cause level according to the similarity between the root cause and the alarm.

2. The method for locating the root cause of wireless network device faults based on machine learning according to claim 1, characterized in that, The equipment information is obtained from the OMC, including equipment model, equipment status, and equipment configuration, which is used to determine the type and location of equipment failure; The equipment operation data includes the equipment's operating status, signal strength, and data transmission rate, which are used to predict potential equipment failures and provide a basis for fault diagnosis and location. The topology information is obtained from the resource management system, including the connection relationships between devices and signal transmission paths, which is used to determine the specific location of device failures. The environmental data refers to the data of the computer room environment obtained from the environmental monitoring system, including the computer room temperature and humidity.

3. The method for locating the root cause of wireless network device faults based on machine learning according to claim 1, characterized in that, The acquired device and topology relationship data undergoes data preprocessing, including the following steps: Remove outliers and missing values ​​from device and topology relationship data; Normalize the device and topology relationship data; Noisy data is removed from device and topology relationship data based on clustering methods; Noise data in device and topology relationship data is predicted based on regression methods, and the predicted noise data is then removed.

4. The method for locating the root cause of wireless network device faults based on machine learning according to claim 1, characterized in that, Alarm analysis includes the following steps: The data stored in the distributed message queue system is divided into multiple time periods according to the time window. The data in each time period includes alarm data. For alarm data within each time period, a clustering algorithm is used to cluster the alarm data; The Apriori algorithm is used to discover frequent itemsets and association rules among alarm data. Alarm data with the same root cause are grouped into one category, and the association rules are used as the content of generalized warnings. The generalized warnings are used as a summary of the alarm data. Hierarchical clustering algorithm is used to cluster alarm data according to different levels.

5. The machine learning-based root cause localization method for wireless network equipment faults according to any one of claims 1-4, characterized in that, Knowledge base construction includes the following steps: Key features of alarm data are extracted using feature engineering methods. Based on these key features, clustering methods are used to merge and cluster similar alarm data to obtain an alarm summary. The key features include the data center name, network element name, and alarm description. The alarm summary includes the data center name, network element name, alarm description, and number of alarms. The alarm description is the specific content of the alarm. Alarms are classified into multiple levels according to their importance; Based on alarm data, the root cause of the fault is located through visualization tools to obtain a root cause description; A knowledge base is built based on alarm level, alarm description, and root cause description.

6. The method for locating the root cause of wireless network device faults based on machine learning according to claim 5, characterized in that, Root cause failure prediction includes the following steps: Root causes are divided into multiple levels, and a hierarchical relationship is established between the root cause levels; A root cause prediction model is constructed based on the seq2seq model. The root cause prediction model is used to predict the root cause level with alarm summary as input. The alarm summary and root cause level are encoded, and the encoded alarm summary and root cause level are used as training samples to train the root cause prediction model, thus obtaining the trained root cause prediction model. The new alarm summary is encoded, and the encoded alarm summary is used as the input sequence to input the trained root cause prediction model. The trained root cause prediction model outputs the root cause level corresponding to the new alarm summary. The new alarm summary and the root cause level of the predicted output are added to the training samples to train the root cause prediction model.

7. A machine learning-based fault location system for wireless network devices, characterized in that, The system is used to locate the root cause of a wireless network device failure using the machine learning-based wireless network device failure root cause localization method as described in any one of claims 1-6, the system comprising: The data acquisition module is used to acquire device and topology relationship data required for root cause localization of wireless network device faults. The device and topology relationship data includes device information, topology relationship information, device operation data, and environmental data. The data preprocessing module is used to preprocess the acquired device and topology relationship data, and remove null values, outliers and noise through data preprocessing to obtain preprocessed data; The data storage module is used to store the acquired device log data and preprocessed data into a distributed message queue system; The alarm analysis module is used to cluster alarms stored in the distributed message queue system according to a predetermined time period based on time windows and clustering algorithms, and to use association rule mining algorithms to summarize alarms with the same root cause into generalized warnings that can cover the alarm content, thus forming a hierarchical relationship of alarm data. The knowledge base construction module is used to perform feature extraction and cluster analysis on alarm data to obtain alarm descriptions, and to locate the root cause of the fault based on the alarm data and visualization tools, and to construct a knowledge base based on alarm level, alarm description and root cause description. The fault root cause prediction module is used to construct sample data based on alarm description and root cause description, construct a root cause prediction model based on deep learning, and use the root cause prediction model to predict the output root cause level based on alarm description. The root cause prediction model is trained based on sample data, and the root cause prediction model is used to predict the root cause of new alarm descriptions to obtain the root cause level. A fault root cause localization module is used to calculate the root cause level based on the obtained root cause level and the similarity between the root cause and the alarm.

8. The machine learning-based wireless network device fault location system according to claim 7, characterized in that, The equipment information is obtained from the OMC, including equipment model, equipment status, and equipment configuration, which is used to determine the type and location of equipment failure; The equipment operation data includes the equipment's operating status, signal strength, and data transmission rate, which are used to predict potential equipment failures and provide a basis for fault diagnosis and location. The topology information is obtained from the resource management system, including the connection relationships between devices and signal transmission paths, which is used to determine the specific location of device failures. The environmental data refers to the data of the computer room environment obtained from the environmental monitoring system, including the computer room temperature and humidity; The data preprocessing module is used to perform the following data preprocessing on the acquired device and topology relationship data: Remove outliers and missing values ​​from device and topology relationship data; Normalize the device and topology relationship data; Noisy data is removed from device and topology relationship data based on clustering methods; Noise data in device and topology relationship data is predicted based on regression methods, and the predicted noise data is then removed.

9. The machine learning-based wireless network device fault location system according to claim 7, characterized in that, The alarm analysis module is used to perform the following: The data stored in the distributed message queue system is divided into multiple time periods according to the time window. The data in each time period includes alarm data. For alarm data within each time period, a clustering algorithm is used to cluster the alarm data; The Apriori algorithm is used to discover frequent itemsets and association rules among alarm data. Alarm data with the same root cause are grouped into one category, and the association rules are used as the content of generalized warnings. The generalized warnings are used as a summary of the alarm data. Hierarchical clustering algorithm is used to cluster alarm data according to different levels.

10. The machine learning-based wireless network device fault location system according to any one of claims 7-9, characterized in that, The knowledge base construction module is used to perform the following: Key features of alarm data are extracted using feature engineering methods. Based on these key features, clustering methods are used to merge and cluster similar alarm data to obtain an alarm summary. The key features include the data center name, network element name, and alarm description. The alarm summary includes the data center name, network element name, alarm description, and number of alarms. The alarm description is the specific content of the alarm. Alarms are classified into multiple levels according to their importance; Based on alarm data, the root cause of the fault is located through visualization tools to obtain a root cause description; A knowledge base is built based on alarm level, alarm description, and root cause description; The fault root cause prediction module is used to perform the following: Root causes are divided into multiple levels, and a hierarchical relationship is established between the root cause levels; A root cause prediction model is constructed based on the seq2seq model. The root cause prediction model is used to predict the root cause level with alarm summary as input. The alarm summary and root cause level are encoded, and the encoded alarm summary and root cause level are used as training samples to train the root cause prediction model, thus obtaining the trained root cause prediction model. The new alarm summary is encoded, and the encoded alarm summary is used as the input sequence to input the trained root cause prediction model. The trained root cause prediction model outputs the root cause level corresponding to the new alarm summary. The new alarm summary and the root cause level of the predicted output are added to the training samples to train the root cause prediction model.

Images