Fault detection method for wireless communication network

Abstract

The invention relates to the technical field of fault detection for a wireless communication network, and discloses a fault detection method for the wireless communication network. By collecting data of base station signal strength, network delay, data packet loss rate and user equipment signal quality, the problem of insufficient fault detection caused by a single index is solved. Through multi-dimensional data acquisition, the system can comprehensively evaluate network performance, the fault detection accuracy is improved, and different fault types can be effectively identified especially in a complex fault mode. By storing the data according to the time sequence, the problem of time sequence chaos is avoided, and accurate detection of time-varying faults is ensured. The comprehensive data vector generated by each time of sampling comprises a plurality of network performance indexes, so that the fault detection and positioning capability is improved. Abnormal values are eliminated through data preprocessing, the data quality is ensured, an accurate basis is provided for subsequent analysis, and the reliability of a detection result is ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of fault detection for wireless communication networks, and in particular to a fault detection method for wireless communication networks. Background Art

[0002] With the continuous development of wireless communication technology, especially the increasingly complex application scenarios of 5G and future 6G networks, the reliability, stability and efficiency of communication networks have become particularly important. In wireless communication networks, network failures occur frequently due to various factors, such as hardware failures, changes in network topology, changes in the external environment, etc. These failures not only affect the user experience, but may even paralyze the entire system. Therefore, timely detection and location of network failures are of great significance to improving network operation and maintenance efficiency, reducing maintenance costs, and ensuring network service quality.

[0003] Existing fault detection methods usually rely on rule-based systems or experience-based diagnostic methods. For example, some methods rely on threshold settings, and when the network signal or quality indicator exceeds the preset threshold, it is considered that a fault has occurred. Although these traditional methods can provide certain help in simple network environments, as network environments and fault modes become more diverse, these methods are often unable to accurately and timely detect the occurrence of complex faults. For example, in large-scale wireless communication networks, traditional fault detection methods may cause false alarms and missed alarms due to their reliance on fixed rules, especially when faced with a variety of complex network fault types, their accuracy and adaptability are relatively insufficient.

[0004] Therefore, this case aims to propose a fault detection method for wireless communication networks, which is mainly based on a data-driven fault detection and location method, aiming to solve the problems of insufficient accuracy and high processing complexity existing in the prior art. Summary of the invention

[0005] The present invention provides a fault detection method for a wireless communication network, which facilitates solving the problems mentioned in the above background technology.

[0006] The present invention provides the following technical solution: a fault detection method for a wireless communication network, comprising: Collecting data from wireless communication networks; The collected data includes: Base station signal strength, denoted as , represents the signal strength from the base station to the user equipment at time t, in dBm; Network delay, denoted as , represents the communication delay between the user equipment and the base station at time t, in ms; Data packet loss rate, denoted as , represents the packet loss ratio of data transmission at time t, expressed as a percentage; The signal quality of the user equipment is denoted as , represents the quality index of the signal received by the user terminal equipment at time t, in dBm; Set the sampling period to The total observation time is , calculate the number of sampling points: ; in, is the total number of sampling points; Each sampling generates a data vector, which is recorded as: ; in, is the comprehensive data vector at time t, including signal strength, network delay, data packet loss rate and signal quality; Store the sampled data in the database in chronological order; Preprocess the collected data.

[0007] Optionally, the preprocessing of the collected data specifically includes: The data in the database are classified and extracted according to base station signal strength, network delay, data packet loss rate and user equipment signal quality, the data of the same type are grouped into a data set, and the classified and extracted data are sorted in the order of storage time; For any kind of dataset, perform the following steps: S1. Use the triple standard deviation method to detect outliers, specifically: Calculate the mean of all values ​​in a dataset: ; in, is the mean of the data set; M is the total number of all elements in the data set; is the value of the i-th data point in the data set; Calculate the standard deviation: ; in, is the standard deviation of the data set; When the data point Requirements: , then the data point Mark as outlier; in, For data points With the mean The absolute value of the difference; S2. Use linear interpolation to fill in missing values: When the data point When data is missing, linear interpolation is used to fill in the missing values, specifically: ; in, is the data value after filling; is the data point before the missing value; is the data point after the missing value; Perform feature extraction on the preprocessed dataset.

[0008] Optionally, the feature extraction in the preprocessed data set specifically includes: Set the sliding window size to k and extract the following features at each time point: S3. Extract sliding mean: ; in, is the sliding mean at time t; k is the sliding window size; is the rth data point in the window; S4. Extract sliding standard deviation: ; in, is the sliding standard deviation at time t; S5. Extract change rate: ; in, is the rate of change at time t; is the data point at time t; is the data point at the starting time of the window; Combine the original data and features to form an extended feature vector: ; in, is the extended feature vector at time t; Based on the extended feature vector, a fault detection strategy is executed.

[0009] Optionally, executing a fault detection strategy based on the extended feature vector specifically includes: Set the normal operating range for each data type metric: S6, ; S7, ; S8, ; S9, ; Set the fault label: ; in, is the fault status label at time t; when , it indicates a fault; when , indicating normal.

[0010] Optionally, executing a fault detection strategy based on the extended feature vector specifically includes: Compute the similarity of the extended feature vectors: ; in, is the extended feature vector and similarity; , are the i-th and j-th extended feature vectors respectively; is the Gaussian kernel width parameter; Initialize cluster centers For randomly selected sample points, iterative update: ; in, is the k-th class center after the t+1th iteration; is the sample set contained in the kth class; is the number of samples in the kth category.

[0011] Optionally, executing a fault detection strategy based on the extended feature vector specifically includes: All extended feature vectors Randomly assign to K initial clusters, with the cluster center being a randomly selected sample point: ; in, is the center of the initial cluster k; p is a random index; Calculate the similarity between each sample and all cluster centers and assign the sample to the nearest cluster: ; in, is the sample set contained in the kth cluster; For sample To cluster center similarity; Update the cluster centers based on the sample set of each cluster: ; in, is the k-th class center after the t+1th iteration; is the k-th class sample set; is the number of samples in the kth category; Set the clustering convergence threshold to ; Check if the cluster center change is below a threshold : ; If the changes in all cluster centers are below the threshold, the clustering process terminates; Classify the faults.

[0012] Optionally, the fault classification specifically includes: After clustering is completed, each cluster is mapped to a failure mode or normal state, specifically: Predefined Failure Mode ; Use majority voting to select each cluster Assign fault labels: ; in, is the label of cluster k; is the indicator function, if If yes, the value is 1, otherwise the value is 0; When there is a cluster that does not match the existing fault mode, it is marked as an unknown fault. , and record its characteristics; Locate the detected fault.

[0013] Optionally, locating the detected fault specifically includes: Set the set of identified fault modes, denoted as set Y; Get all identified failure modes and add them to set Y; For each failure mode , calculated and collected data The similarity between: ; in, and Respectively represent the failure mode and data The value of the jth feature in ; and Failure modes and data The mean of is the weight of the jth feature; By calculating the similarity of each failure mode, the priority of each failure mode is calculated: ; in, Indicates failure mode The priority of the fault mode; d is the total number of fault modes; Is the failure mode With current data similarity; Prioritize all failure modes Sort from high to low; Output the highest priority failure mode and the data with the highest similarity to the highest priority failure mode.

[0014] The present invention has the following beneficial effects: 1. By collecting data on base station signal strength, network delay, data packet loss rate, and user equipment signal quality from the wireless communication network, the problem of insufficient fault detection that relies solely on a single indicator is solved. Traditional methods usually rely solely on a single network performance indicator, which may lead to low accuracy of fault detection, especially when facing complex fault modes, and cannot fully reflect the network status. Collecting multi-dimensional network data enables the system to comprehensively evaluate network performance and capture more fine-grained information, which helps to improve the accuracy of fault detection, especially in complex environments. It can effectively identify and distinguish different types of fault modes. By storing data in chronological order, the problem of chaotic data time sequence is solved. The state of a wireless network is a dynamically changing process, and the time series of data is very important. Many faults may occur instantaneously and have a certain time dependency. If the data storage does not maintain the order of the time series, it may cause incorrect time series judgments during the analysis process, affecting the accuracy of fault detection. By ensuring the time sequence storage of data, the system can accurately capture the dynamic process of network status changing over time, thereby enhancing the perception of time-varying faults and effectively improving the accuracy of data analysis and fault prediction. By generating a comprehensive data vector for each sampling, including base station signal strength, network delay, data packet loss rate and user equipment signal quality, the problem of missing information that cannot fully express the network status is solved. Data vectorization enables the system to use more comprehensive information for fault detection and location in the subsequent processing and analysis stages, improving the system's analysis capabilities and judgment accuracy. Data preprocessing can effectively eliminate invalid data and outliers, improve data quality, and provide a clearer and more accurate basis for subsequent fault analysis and pattern recognition. By removing interference, the accuracy of subsequent steps is guaranteed and the fault detection results are more credible.

[0015] 2. By classifying and extracting the data in the database according to base station signal strength, network delay, data packet loss rate and user equipment signal quality, and sorting them according to the order of storage time, the problem of data clutter and difficulty in unified analysis is solved. Different types of performance indicators in wireless communication networks have different data characteristics and time dependencies. If they are not distinguished and sorted, it may lead to information loss or misinterpretation in the data analysis process. By classifying, extracting and sorting, it can ensure that each type of data is processed in the correct time series. By using the triple standard deviation method to detect outliers, the problem of interference caused by noise data or outliers on subsequent analysis is solved. Data in wireless communication networks are affected by many factors, such as equipment failure, environmental interference, etc., and extreme outliers may appear. If these outliers are not processed, they may mislead the fault detection model and lead to incorrect fault diagnosis results. By detecting and eliminating outliers, the quality and consistency of the data are guaranteed, providing a reliable data basis for subsequent feature extraction and fault detection. The elimination of abnormal data reduces the detection error caused by noise data interference, thereby improving the accuracy and reliability of fault detection. By using linear interpolation to fill in missing values, the problem of incomplete analysis caused by data loss is solved. In wireless communication networks, missing values ​​may occur during data collection due to reasons such as unstable signals and equipment failures. If these missing data are not processed, data may be missing in subsequent analysis, affecting the overall analysis results. Linear interpolation estimates missing values ​​by using the values ​​of adjacent data points, thereby smoothing the time series of data and ensuring the continuity and integrity of the data. After filling in the missing values, the data set is more complete, ensuring the continuity of feature extraction and subsequent analysis. This not only avoids the calculation difficulties caused by missing data, but also improves the accuracy of the overall analysis results, ensures the full use of data, and improves the reliability of fault detection. By performing feature extraction in the preprocessed data set, the problem of interference between redundancy of original data and irrelevant information is solved. The data in the wireless communication network contains a lot of information, but not all data is useful for fault detection. Through feature extraction, the most valuable features for fault diagnosis can be screened out from a large amount of raw data, thereby improving the efficiency of subsequent fault location and analysis. Feature extraction makes data analysis more focused on key parameters, reduces the interference of irrelevant data, can improve the efficiency and accuracy of subsequent analysis algorithms, and further improve the overall performance of the fault detection system.

[0016] 3. By extracting the sliding mean, the influence of local fluctuation interference in the data is resolved. Data in wireless communication networks are usually disturbed by instantaneous fluctuations and noise. These short-term fluctuations may not represent the actual status of the system and can easily mislead fault detection. By calculating the sliding mean, short-term fluctuations in the data can be smoothed and a more stable trend can be extracted. The sliding mean helps reduce noise and random fluctuations in the data, ensuring that the fault detection algorithm focuses on real trend changes rather than short-term fluctuations. This process makes subsequent fault detection more stable and improves sensitivity to long-term trend changes. By extracting the sliding standard deviation, the problem of measuring the amplitude and degree of fluctuation of data changes is solved. In wireless communication networks, the changes in indicators such as signal strength and delay are not always uniform, and some fault modes may manifest as abnormal increases in indicator fluctuations. The sliding standard deviation can effectively quantify these fluctuations and help detect abnormal changes. The sliding standard deviation provides a dynamic fluctuation metric that can reflect the degree of change in network performance within a certain time window. This helps detect abnormal fluctuations in the system, identify possible faults in advance, and enhance the sensitivity and accuracy of the fault detection algorithm. By extracting the rate of change, the problem of not being able to capture rapid changes and emergencies in network indicators is solved. In wireless communication networks, some faults may appear as rapid signal changes, which may not be accurately captured by traditional static indicators. Rate of change extraction can measure the speed of change between a certain time point and the previous time point, helping to identify sudden faults. By calculating the rate of data change, the system can detect the speed of data change and sudden events, which is crucial for timely identification of sudden faults in the network. The extraction of the rate of change improves the timeliness of fault detection, allowing the system to respond to rapid changes in the network and quickly locate possible fault sources. By combining the original data with the extracted features, such as the sliding mean, sliding standard deviation, and rate of change to form an extended feature vector, the problem of insufficient information in a single data point is solved. A single raw data point often cannot provide enough information to accurately judge the fault, while combining multiple features to form an extended feature vector can more comprehensively represent the network status. The construction of the extended feature vector provides richer data input for subsequent fault detection. By integrating information from multiple dimensions, the detection system can make more accurate fault judgments based on a more comprehensive network status. The fusion of features enhances the overall robustness of the system and improves the diversity and accuracy of fault detection.

[0017] 4. By setting the normal operating range of each data type, such as base station signal strength, network delay, data packet loss rate, user equipment signal quality, etc., the problem of how to distinguish between normal and abnormal states is solved. In wireless communication networks, different performance indicators have their own normal fluctuation ranges. Too wide or inappropriate normal ranges may lead to the inability to accurately identify faults. By presetting the normal operating range of each indicator, it is possible to more accurately determine whether the data is within the normal range, thereby effectively distinguishing between normal and abnormal states. This operation effectively improves the accuracy of fault detection. By clarifying the normal range, the system can more clearly distinguish between fault states and normal states, avoid misjudgment and missed judgment, and improve the stability and reliability of the fault detection system. By setting a fault label to identify the fault state of each data point, the problem of being unable to clearly identify the fault location in the data is solved. In actual applications, the fault state in the network may be instantaneous, sudden, or continuous. If there is no clear fault label to identify the occurrence of the fault, subsequent positioning and repair work will become difficult. By setting the fault label, the system can clearly identify whether the data point at each moment has a fault, thereby providing a clear mark for fault location and recovery. The setting of fault labels makes fault identification more accurate and clear, providing very important support for subsequent fault location. This labeled fault representation method simplifies the operation process of the fault detection system and improves the speed and accuracy of fault response. By judging the fault label, the system can quickly identify the time when the fault occurs in the real-time data stream, avoiding the defects of traditional methods that require manual judgment or delayed judgment. Such a real-time fault marking and judgment mechanism improves the efficiency of network fault detection and speeds up the speed of fault response, providing network operation and maintenance personnel with instant alarm information, and further improving network stability and user experience.

[0018] 5. By calculating the similarity between extended feature vectors, the problem of how to identify similar fault modes in multidimensional data space is solved. Each extended feature vector contains a variety of indicators, such as signal strength, delay, packet loss rate, signal quality, etc. Through similarity calculation, samples with similar features can be effectively identified, thereby helping the system understand which data points belong to the same type of fault mode. The introduction of Gaussian kernel function can make the similarity measurement smoother and can handle the complex relationship between different feature dimensions. By calculating similarity through Gaussian kernel, it is possible to effectively focus on fault modes with similar performance, thereby improving the accuracy of fault detection. Compared with traditional Euclidean distance or other simple similarity calculation methods, Gaussian kernel function can better capture the nonlinear relationship and potential complex patterns of data, effectively improving the accuracy of fault mode recognition. By initializing the cluster center as a randomly selected sample point, the problem of how to efficiently start the clustering process is solved. In cluster analysis, the selection of the initial cluster center has an important influence on the final clustering result. By randomly selecting some sample points from the data set as the initial center, it is possible to avoid over-reliance on certain prior assumptions and increase the flexibility and adaptability of the clustering process. This step enables the clustering process to start from a variety of initial conditions, enhances the diversity and exploratory nature of the clustering results, and avoids the problem of local optimal solutions caused by improper selection of initial points. Randomly initializing the cluster center helps to improve the effectiveness of the clustering algorithm and prevent clustering from falling into the local optimal solution. By iteratively updating the cluster center, the problem of how to continuously optimize the results in the clustering process is solved. By continuously adjusting the cluster center, the system can better adapt to data changes and improve the accuracy of clustering. In each iteration, the cluster center is updated according to the sample set of the current class, so as to gradually converge to the real fault mode. Iterative updates enable the clustering results to be optimized as the data is continuously adjusted, avoiding errors caused by inaccurate initial cluster centers. Iterative updates of cluster centers greatly enhance the flexibility and accuracy of clustering algorithms. Each update enables the class center to more accurately reflect the distribution of data, thereby improving the accuracy of fault mode recognition. In wireless communication network fault detection, real-time data and network conditions may change at any time. Iterative updates enable clustering algorithms to dynamically adapt to these changes and maintain efficient fault identification capabilities. Through multiple iterative updates of cluster centers, the problem of how to ensure that the clustering algorithm eventually converges to the correct fault mode is solved. In multiple iterations, the cluster center will gradually approach the true distribution of the samples to which it belongs, making the clustering results more and more accurate. The convergence of clustering ensures that the final fault detection results can reflect the true structure of the data and avoid errors caused by insufficient iterations or early stopping. The process of cluster center convergence ensures the stability and accuracy of the clustering results, so that fault mode identification can accurately reflect the actual state of the network and reduce misjudgments caused by clustering failure to converge or unstable results.

[0019] 6. By randomly assigning all extended feature vectors to K initial clusters and initializing the cluster centers to randomly selected sample points, the problem of how to reasonably initialize the cluster centers is solved. In traditional clustering algorithms, the selection of the initial cluster center is crucial to the clustering effect. Randomly selecting the cluster center can increase the diversity of clustering and avoid over-reliance on certain prior knowledge or assumptions. Randomly initializing the cluster center makes the clustering process more flexible, can explore more potential fault modes, and avoids the deviation caused by over-reliance on preliminary data analysis. Through this initialization method, the algorithm can better adapt to the complex and changeable fault modes in the network, and improve the exploration and accuracy of clustering. By calculating the similarity between each sample and all cluster centers and assigning the sample to the closest cluster, the problem of how to reasonably classify the sample according to its characteristics is solved. By calculating the similarity between the sample and the cluster center, the sample can be classified into its most likely fault mode to ensure the accurate attribution of the fault mode. This process can efficiently cluster similar fault samples together, avoiding misclassification and misjudgment. In wireless communication networks, different fault modes may have similar manifestations. Clustering similarity calculation can help the algorithm quickly identify and accurately group them, thereby improving the accuracy of fault detection. By updating the cluster center according to the sample set of each cluster, the problem of how to optimize the clustering results through iteration is solved. By continuously updating the cluster center, the clustering can gradually converge to a more accurate state, thereby better reflecting the actual distribution of the samples. This step improves the adaptability and accuracy of the clustering algorithm, allowing the clustering to be optimized as the data changes. By updating the cluster center, the system can continuously optimize the clustering results, making the fault mode represented by each cluster more accurate. The updated cluster center can better capture the changes in data, increase the stability of clustering, and ensure the accurate identification of real-time fault modes in wireless communication networks. By setting the cluster convergence threshold and checking whether the change of the cluster center is lower than the threshold, the problem of how to judge whether the clustering algorithm has reached convergence is solved. By setting the threshold, it is possible to automatically judge whether the clustering has stabilized at the end of the clustering process, avoiding unnecessary calculations and iterations, and improving the efficiency of the algorithm. Setting the cluster convergence threshold enables the clustering process to terminate in time after reaching a stable state, avoiding the waste of calculations caused by excessive iterations, while ensuring the accuracy and efficiency of the clustering results. In wireless communication network fault detection, quickly terminating the clustering process helps to improve the real-time performance and response speed of the system. By classifying the faults after the clustering process is completed, the problem of how to extract specific fault modes from the clustered samples is solved. By classifying the samples within each cluster, each fault mode can be accurately located and marked, providing clear guidance for subsequent troubleshooting and processing.

[0020] 7. The problem of how to convert clustering results into specific fault mode labels is solved. After the clustering step is completed, the clustering results can be converted into specific fault types by corresponding each cluster to a predefined fault mode or normal state, thereby providing clear guidance for subsequent processing. This step ensures that each cluster can be accurately labeled as a specific fault mode, avoiding ambiguity or misunderstanding of the clustering results. This step enables the clustering results to be docked with the actual fault mode, improving the practical application value of fault detection. By corresponding to the actual fault mode, each network problem can be accurately labeled, which enhances the interpretability and operability of the detection and provides network maintenance personnel with clearer fault information. The problem of how to deal with the fault label assignment for each cluster is solved. In practical applications, some clusters may contain multiple different fault modes. The use of majority voting can effectively avoid the influence of human intervention on the label, making the label assignment more objective and fair. The majority voting method counts the labels of each sample in the cluster and selects the label with the highest frequency as the final cluster label, thereby ensuring the reliability of the cluster label. This step effectively reduces the randomness and errors in label assignment and improves the accuracy of classification. In particular, when faced with diverse fault samples, the majority voting method can ensure that the main fault modes within the cluster are accurately reflected, enhancing the robustness of the system. It solves the problem of how to deal with unknown faults. In wireless communication networks, due to the diversity and rapid changes of fault types, there may be some unknown fault modes. By identifying clusters that do not match the predefined fault modes during the classification process and marking them as "unknown faults", new fault types that are not foreseen by the database can be discovered in a timely manner. This step provides the necessary flexibility for comprehensive monitoring of network faults. This step effectively improves the adaptability of fault detection, ensuring that the system can handle and record unknown faults and avoid missing potential problems. With the passage of time and the discovery of new faults, the characteristics of these unknown fault clusters can be further accumulated, ultimately providing data support for the establishment of a more comprehensive fault mode library, thereby continuously improving the accuracy and comprehensiveness of the system.

[0021] 8. By setting the identified fault mode set Y and obtaining all fault modes, the problem of how to systematically manage and organize the identified fault modes is solved. In wireless communication networks, the number of fault modes is huge and may continue to increase over time. Therefore, centrally storing all identified fault modes in set Y can help the system effectively track historical fault modes and provide a complete fault mode reference for subsequent fault location. The maintenance and update of this set ensures that the system can continuously optimize and enrich the fault mode library as the network status changes. By building a complete fault mode set Y, the system can flexibly respond to new fault types and modes, and use existing fault modes for rapid comparison and analysis, thereby improving the accuracy and efficiency of fault location. The management of historical fault modes helps to reduce the detection and resolution time of similar problems in the future. By calculating the similarity between each fault mode and the collected data. The problem of how to compare the fault mode with the real-time collected data to accurately determine the fault type is solved. In actual networks, fault modes are diverse, and the characteristics of each fault mode may change at different time points or under different environmental conditions. By calculating the similarity between each fault mode and the collected data, the fault mode most relevant to the current data can be efficiently identified, avoiding simple rule matching or overly rough judgment methods. This step makes the matching between the fault mode and the real-time data more accurate by quantifying the similarity index. The similarity score between each fault mode and the data provides a clear basis for the subsequent priority calculation, further improving the accuracy of fault detection. This method can reduce the false alarm rate and the missed alarm rate, and improve the intelligent level of fault location. By calculating the priority of each fault mode. The problem of how to evaluate the severity and urgency of the fault mode based on similarity is solved. In wireless communication networks, different fault modes may have different impacts on the network. Some faults may cause major system crashes, while others may be small-scale problems. By calculating the priority of each fault mode, it can be ensured that the fault location system can give priority to faults that have a greater impact on the network, thereby improving the response speed and processing efficiency of network operation and maintenance. This step quantifies the priority of each fault mode, so that the system can automatically judge and focus on the most critical faults, avoiding waste of resources and redundancy in processing. Through priority sorting, network operation and maintenance personnel can quickly handle high-priority faults, reduce network interruption time, and improve network availability and stability. By sorting all fault modes from high to low priority, the problem of how to sort multiple fault modes and select the most urgent fault to handle is solved. Since multiple fault modes may be detected at the same time during the fault detection process, how to sort these fault modes and determine the processing order directly affects the efficiency and effectiveness of fault handling. By sorting fault modes by priority, the system can ensure that the most critical problems are solved first, thereby reducing the impact on network services.The sorting mechanism enables network operation and maintenance to quickly identify the most important faults and respond in the shortest possible time. In this way, the system can allocate resources more efficiently when facing multiple faults, improve the overall fault repair efficiency, reduce the recovery time of network faults, and optimize the quality of network services. By outputting the highest priority fault mode and the data with the highest similarity, the problem of how to associate the detected fault with specific data features and further analyze and locate it is solved. After completing the sorting of the fault modes, outputting the highest priority fault mode and the most similar collected data helps to accurately lock the time and specific problem of the fault, providing a more intuitive basis for troubleshooting. This step helps operation and maintenance personnel quickly locate the cause and background of the fault by outputting the highest priority fault mode and the most similar data. Through precise docking with data, operation and maintenance personnel can quickly make fault repair decisions based on specific historical data and similar network conditions, improving the efficiency and accuracy of fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0023] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0024] Example, see Figure 1 , a fault detection method for a wireless communication network, comprising: Collecting data from wireless communication networks; The collected data includes: Base station signal strength, denoted as , represents the signal strength from the base station to the user equipment at time t, in dBm; Network delay, denoted as , represents the communication delay between the user equipment and the base station at time t, in ms; Data packet loss rate, denoted as , represents the packet loss ratio of data transmission at time t, expressed as a percentage; The signal quality of the user equipment is denoted as , represents the quality index of the signal received by the user terminal equipment at time t, in dBm; Set the sampling period to The total observation time is , calculate the number of sampling points: ; in, is the total number of sampling points; Each sampling generates a data vector, which is recorded as: ; in, is the comprehensive data vector at time t, including signal strength, network delay, data packet loss rate and signal quality; Store the sampled data in the database in chronological order; Preprocess the collected data.

[0025] By collecting data on base station signal strength, network delay, data packet loss rate, and user equipment signal quality from the wireless communication network, the problem of insufficient fault detection that relies solely on a single indicator is solved. Traditional methods usually rely solely on a single network performance indicator, which may lead to low fault detection accuracy, especially when facing complex fault modes, and cannot fully reflect the network status. Collecting multi-dimensional network data enables the system to comprehensively evaluate network performance and capture more fine-grained information, which helps to improve the accuracy of fault detection, especially in complex environments. It can effectively identify and distinguish different types of fault modes. By storing data in chronological order, the problem of chaotic data time sequence is solved. The state of a wireless network is a dynamically changing process, and the time series of data is very important. Many faults may occur instantaneously and have a certain time dependency. If the data storage does not maintain the order of the time series, it may cause incorrect time series judgments during the analysis process, affecting the accuracy of fault detection. By ensuring the time order of data storage, the system can accurately capture the dynamic process of network status changing over time, thereby enhancing the perception of time-varying faults and effectively improving the accuracy of data analysis and fault prediction. By generating a comprehensive data vector for each sampling, including base station signal strength, network delay, data packet loss rate and user equipment signal quality, the problem of missing information that cannot fully express the network status is solved. Data vectorization enables the system to use more comprehensive information for fault detection and location in the subsequent processing and analysis stages, improving the system's analysis capabilities and judgment accuracy. Data preprocessing can effectively eliminate invalid data and outliers, improve data quality, and provide a clearer and more accurate basis for subsequent fault analysis and pattern recognition. By removing interference, the accuracy of subsequent steps is guaranteed and the fault detection results are more credible.

[0026] The preprocessing of the collected data specifically includes: The data in the database are classified and extracted according to base station signal strength, network delay, data packet loss rate and user equipment signal quality, the data of the same type are grouped into a data set, and the classified and extracted data are sorted in the order of storage time; For any kind of dataset, perform the following steps: S1. Use the triple standard deviation method to detect outliers, specifically: Calculate the mean of all values ​​in a dataset: ; in, is the mean of the data set, reflecting the overall average level of the data; M is the total number of all elements in the data set; is the value of the i-th data point in the data set; Calculate the standard deviation: ; in, is the standard deviation of the data set, indicating the degree of fluctuation or dispersion of the data; When the data point Requirements: , then the data point Mark as outlier; in, For data points With the mean The absolute value of the difference; S2. Use linear interpolation to fill in missing values: When the data point When data is missing, linear interpolation is used to fill in the missing values, specifically: ; in, is the data value after filling; is the data point before the missing value; is the data point after the missing value; Perform feature extraction on the preprocessed dataset.

[0027] By classifying and extracting the data in the database according to base station signal strength, network delay, data packet loss rate, and user equipment signal quality, and sorting them according to the order of storage time, the problem of data clutter and difficulty in unified analysis is solved. Different types of performance indicators in wireless communication networks have different data characteristics and time dependencies. If they are not distinguished and sorted, information loss or misinterpretation may occur during data analysis. Classification, extraction, and sorting can ensure that each type of data is processed in the correct time series. By using the triple standard deviation method to detect outliers, the problem of interference caused by noise data or outliers on subsequent analysis is solved. Data in wireless communication networks are affected by many factors, such as equipment failure, environmental interference, etc., and extreme outliers may appear. If these outliers are not processed, they may mislead the fault detection model and lead to incorrect fault diagnosis results. By detecting and eliminating outliers, the quality and consistency of the data are guaranteed, providing a reliable data basis for subsequent feature extraction and fault detection. The elimination of abnormal data reduces the detection error caused by noise data interference, thereby improving the accuracy and reliability of fault detection. By using linear interpolation to fill in missing values, the problem of incomplete analysis caused by data loss is solved. In wireless communication networks, missing values ​​may occur during data collection due to reasons such as unstable signals and equipment failures. If these missing data are not processed, data may be missing in subsequent analysis, affecting the overall analysis results. Linear interpolation estimates missing values ​​by using the values ​​of adjacent data points, thereby smoothing the time series of data and ensuring the continuity and integrity of the data. After filling in the missing values, the data set is more complete, ensuring the continuity of feature extraction and subsequent analysis. This not only avoids the calculation difficulties caused by missing data, but also improves the accuracy of the overall analysis results, ensures the full use of data, and improves the reliability of fault detection. By performing feature extraction in the preprocessed data set, the problem of interference between redundancy of original data and irrelevant information is solved. The data in the wireless communication network contains a lot of information, but not all data is useful for fault detection. Through feature extraction, the most valuable features for fault diagnosis can be screened out from a large amount of raw data, thereby improving the efficiency of subsequent fault location and analysis. Feature extraction makes data analysis more focused on key parameters, reduces the interference of irrelevant data, can improve the efficiency and accuracy of subsequent analysis algorithms, and further improve the overall performance of the fault detection system.

[0028] The feature extraction in the preprocessed data set specifically includes: Set the sliding window size to k and extract the following features at each time point: S3. Extract sliding mean: ; in, is the sliding mean at time t; k is the sliding window size; is the rth data point in the window; S4. Extract sliding standard deviation: ; in, is the sliding standard deviation at time t; S5. Extract change rate: ; in, is the rate of change at time t; is the data point at time t; is the data point at the starting time of the window; Combine the original data and features to form an extended feature vector: ; in, is the extended feature vector at time t; Based on the extended feature vector, a fault detection strategy is executed.

[0029] By extracting the sliding mean, the influence of local fluctuation interference in the data is resolved. Data in wireless communication networks are often disturbed by instantaneous fluctuations and noise. These short-term fluctuations may not represent the actual status of the system and can easily mislead fault detection. By calculating the sliding mean, short-term fluctuations in the data can be smoothed and a more stable trend can be extracted. The sliding mean helps reduce noise and random fluctuations in the data, ensuring that the fault detection algorithm focuses on real trend changes rather than short-term fluctuations. This process makes subsequent fault detection more stable and improves the sensitivity to long-term trend changes. By extracting the sliding standard deviation, the problem of measuring the amplitude and degree of fluctuation of data changes is solved. In wireless communication networks, the changes in indicators such as signal strength and delay are not always uniform, and some fault modes may manifest as abnormal increases in indicator fluctuations. The sliding standard deviation can effectively quantify these fluctuations and help detect abnormal changes. The sliding standard deviation provides a dynamic fluctuation metric that can reflect the degree of change in network performance within a certain time window. This helps detect abnormal fluctuations in the system, identify possible faults in advance, and enhance the sensitivity and accuracy of the fault detection algorithm. By extracting the rate of change, the problem of not being able to capture rapid changes and emergencies in network indicators is solved. In wireless communication networks, some faults may appear as rapid signal changes, which may not be accurately captured by traditional static indicators. Rate of change extraction can measure the speed of change between a certain time point and the previous time point, helping to identify sudden faults. By calculating the rate of data change, the system can detect the speed of data change and sudden events, which is crucial for timely identification of sudden faults in the network. The extraction of the rate of change improves the timeliness of fault detection, allowing the system to respond to rapid changes in the network and quickly locate possible fault sources. By combining the original data with the extracted features, such as the sliding mean, sliding standard deviation, and rate of change to form an extended feature vector, the problem of insufficient information in a single data point is solved. A single raw data point often cannot provide enough information to accurately judge the fault, while combining multiple features to form an extended feature vector can more comprehensively represent the network status. The construction of the extended feature vector provides richer data input for subsequent fault detection. By integrating information from multiple dimensions, the detection system can make more accurate fault judgments based on a more comprehensive network status. The fusion of features enhances the overall robustness of the system and improves the diversity and accuracy of fault detection.

[0030] The fault detection strategy is executed based on the extended feature vector, specifically including: Set the normal operating range for each data type metric: S6, ; S7, ; S8, ; S9, ; Set the fault label: ; in, is the fault status label at time t; when , it indicates a fault; when , indicating normal.

[0031] By setting the normal operating range of each data type, such as base station signal strength, network delay, data packet loss rate, user equipment signal quality, etc., the problem of how to distinguish between normal and abnormal states is solved. In wireless communication networks, different performance indicators have their own normal fluctuation ranges. Too broad or inappropriate normal ranges may lead to inaccurate fault identification. By presetting the normal operating range of each indicator, it is possible to more accurately determine whether the data is within the normal range, thereby effectively distinguishing between normal and abnormal states. This operation effectively improves the accuracy of fault detection. By clarifying the normal range, the system can more clearly distinguish between fault states and normal states, avoid misjudgment and missed judgment, and improve the stability and reliability of the fault detection system. By setting a fault label to identify the fault state of each data point, the problem of being unable to clearly identify the fault location in the data is solved. In actual applications, the fault state in the network may be instantaneous, sudden, or continuous. If there is no clear fault label to identify the occurrence of the fault, subsequent positioning and repair work will become difficult. By setting the fault label, the system can clearly identify whether the data point at each moment has a fault, thereby providing a clear mark for fault location and recovery. The setting of fault labels makes fault identification more accurate and clear, providing very important support for subsequent fault location. This labeled fault representation method simplifies the operation process of the fault detection system and improves the speed and accuracy of fault response. By judging the fault label, the system can quickly identify the time when the fault occurs in the real-time data stream, avoiding the defects of traditional methods that require manual judgment or delayed judgment. Such a real-time fault marking and judgment mechanism improves the efficiency of network fault detection and speeds up the speed of fault response, providing network operation and maintenance personnel with instant alarm information, and further improving network stability and user experience.

[0032] The fault detection strategy is executed based on the extended feature vector, specifically including: Compute the similarity of the extended feature vectors: ; in, is the extended feature vector and similarity; , are the i-th and j-th extended feature vectors respectively; is the Gaussian kernel width parameter; Initialize cluster centers For randomly selected sample points, iterative update: ; in, is the k-th class center after the t+1th iteration; is the sample set contained in the kth class; is the number of samples in the kth category.

[0033] By calculating the similarity between extended feature vectors, the problem of how to identify similar fault modes in multidimensional data space is solved. Each extended feature vector contains multiple indicators, such as signal strength, delay, packet loss rate, signal quality, etc. Through similarity calculation, samples with similar features can be effectively identified, thereby helping the system understand which data points belong to the same type of fault mode. The introduction of Gaussian kernel function can make the similarity measurement smoother and can handle the complex relationship between different feature dimensions. By calculating similarity by Gaussian kernel, it is possible to effectively focus on fault modes with similar performance, thereby improving the accuracy of fault detection. Compared with traditional Euclidean distance or other simple similarity calculation methods, Gaussian kernel function can better capture the nonlinear relationship and potential complex patterns of data, effectively improving the accuracy of fault mode recognition. By initializing the cluster center as a randomly selected sample point, the problem of how to efficiently start the clustering process is solved. In cluster analysis, the selection of the initial cluster center has an important influence on the final clustering result. By randomly selecting some sample points from the data set as the initial center, it is possible to avoid over-reliance on certain prior assumptions and increase the flexibility and adaptability of the clustering process. This step enables the clustering process to start from a variety of initial conditions, enhances the diversity and exploratory nature of the clustering results, and avoids the problem of local optimal solutions caused by improper selection of initial points. Randomly initializing the cluster center helps to improve the effectiveness of the clustering algorithm and prevent clustering from falling into the local optimal solution. By iteratively updating the cluster center, the problem of how to continuously optimize the results in the clustering process is solved. By continuously adjusting the cluster center, the system can better adapt to data changes and improve the accuracy of clustering. In each iteration, the cluster center is updated according to the sample set of the current class, so as to gradually converge to the real fault mode. Iterative updates enable the clustering results to be optimized as the data is continuously adjusted, avoiding errors caused by inaccurate initial cluster centers. Iterative updates of cluster centers greatly enhance the flexibility and accuracy of clustering algorithms. Each update enables the class center to more accurately reflect the distribution of data, thereby improving the accuracy of fault mode recognition. In wireless communication network fault detection, real-time data and network conditions may change at any time. Iterative updates enable clustering algorithms to dynamically adapt to these changes and maintain efficient fault identification capabilities. Through multiple iterative updates of cluster centers, the problem of how to ensure that the clustering algorithm eventually converges to the correct fault mode is solved. In multiple iterations, the cluster center will gradually approach the true distribution of the samples to which it belongs, making the clustering results more and more accurate. The convergence of clustering ensures that the final fault detection results can reflect the true structure of the data and avoid errors caused by insufficient iterations or early stopping. The process of cluster center convergence ensures the stability and accuracy of the clustering results, so that fault mode identification can accurately reflect the actual state of the network and reduce misjudgments caused by clustering failure to converge or unstable results.

[0034] The fault detection strategy is executed based on the extended feature vector, specifically including: All extended feature vectors Randomly assign to K initial clusters, with the cluster center being a randomly selected sample point: ; in, is the center of the initial cluster k; p is a random index; Calculate the similarity between each sample and all cluster centers and assign the sample to the nearest cluster: ; in, is the sample set contained in the kth cluster; For sample To cluster center similarity; Update the cluster centers based on the sample set of each cluster: ; in, is the k-th class center after the t+1th iteration; is the k-th class sample set; is the number of samples in the kth category; Set the clustering convergence threshold to ; Check if the cluster center change is below a threshold : ; If the changes in all cluster centers are below the threshold, the clustering process terminates; Classify the faults.

[0035] By randomly assigning all extended feature vectors to K initial clusters and initializing the cluster centers to randomly selected sample points, the problem of how to reasonably initialize the cluster centers is solved. In traditional clustering algorithms, the selection of initial cluster centers is crucial to the clustering effect. Randomly selecting cluster centers can increase the diversity of clustering and avoid over-reliance on certain prior knowledge or assumptions. Randomly initializing cluster centers makes the clustering process more flexible, can explore more potential fault modes, and avoids the deviation caused by over-reliance on preliminary data analysis. With this initialization method, the algorithm can better adapt to the complex and changeable fault modes in the network, and improve the exploration and accuracy of clustering. By calculating the similarity between each sample and all cluster centers and assigning the sample to the closest cluster, the problem of how to reasonably classify the sample according to its characteristics is solved. By calculating the similarity between the sample and the cluster center, the sample can be classified into its most likely fault mode to ensure the accurate attribution of the fault mode. This process can efficiently cluster similar fault samples together and avoid misclassification and misjudgment. In wireless communication networks, different fault modes may have similar manifestations. Clustering similarity calculation can help the algorithm quickly identify and accurately group them, thereby improving the accuracy of fault detection. By updating the cluster center according to the sample set of each cluster, the problem of how to optimize the clustering results through iteration is solved. By continuously updating the cluster center, the clustering can gradually converge to a more accurate state, thereby better reflecting the actual distribution of the samples. This step improves the adaptability and accuracy of the clustering algorithm, allowing the clustering to be optimized as the data changes. By updating the cluster center, the system can continuously optimize the clustering results, making the fault mode represented by each cluster more accurate. The updated cluster center can better capture the changes in data, increase the stability of clustering, and ensure the accurate identification of real-time fault modes in wireless communication networks. By setting the cluster convergence threshold and checking whether the change of the cluster center is lower than the threshold, the problem of how to judge whether the clustering algorithm has reached convergence is solved. By setting the threshold, it is possible to automatically judge whether the clustering has stabilized at the end of the clustering process, avoiding unnecessary calculations and iterations, and improving the efficiency of the algorithm. Setting the cluster convergence threshold enables the clustering process to terminate in time after reaching a stable state, avoiding the waste of calculations caused by excessive iterations, while ensuring the accuracy and efficiency of the clustering results. In wireless communication network fault detection, quickly terminating the clustering process helps to improve the real-time performance and response speed of the system. By classifying the faults after the clustering process is completed, the problem of how to extract specific fault modes from the clustered samples is solved. By classifying the samples within each cluster, each fault mode can be accurately located and marked, providing clear guidance for subsequent troubleshooting and processing.

[0036] The fault classification specifically includes: After clustering is completed, each cluster is mapped to a failure mode or normal state, specifically: Predefined Failure Mode ; Use majority voting to select each cluster Assign fault labels: ; in, is the label of cluster k; is the indicator function, if If yes, the value is 1, otherwise the value is 0; When there is a cluster that does not match the existing fault mode, it is marked as an unknown fault. , and record its characteristics; Locate the detected fault.

[0037] The problem of how to convert clustering results into specific fault mode labels is solved. After the clustering step is completed, the clustering results can be converted into specific fault types by corresponding each cluster to a predefined fault mode or normal state, thereby providing clear guidance for subsequent processing. This step ensures that each cluster can be accurately labeled as a specific fault mode, avoiding ambiguity or misunderstanding of the clustering results. This step enables the clustering results to be connected with the actual fault mode, thereby improving the practical application value of fault detection. By corresponding to the actual fault mode, each network problem can be accurately labeled, which enhances the interpretability and operability of the detection and provides network maintenance personnel with clearer fault information. The problem of how to deal with the fault label assignment of each cluster is solved. In practical applications, some clusters may contain multiple different fault modes. The use of the majority voting method can effectively avoid the influence of human intervention on the label, making the label assignment more objective and fair. The majority voting method counts the labels of each sample in the cluster and selects the label with the highest frequency as the final cluster label, thereby ensuring the reliability of the cluster label. This step effectively reduces the randomness and errors in label assignment and improves the accuracy of classification. In particular, when faced with diverse fault samples, the majority voting method can ensure that the main fault modes within the cluster are accurately reflected, enhancing the robustness of the system. It solves the problem of how to deal with unknown faults. In wireless communication networks, due to the diversity and rapid changes of fault types, there may be some unknown fault modes. By identifying clusters that do not match the predefined fault modes during the classification process and marking them as "unknown faults", new fault types that are not foreseen by the database can be discovered in a timely manner. This step provides the necessary flexibility for comprehensive monitoring of network faults. This step effectively improves the adaptability of fault detection, ensuring that the system can handle and record unknown faults and avoid missing potential problems. With the passage of time and the discovery of new faults, the characteristics of these unknown fault clusters can be further accumulated, ultimately providing data support for the establishment of a more comprehensive fault mode library, thereby continuously improving the accuracy and comprehensiveness of the system.

[0038] The fault locating of the detected fault specifically includes: Set the set of identified fault modes, denoted as set Y; Get all identified failure modes and add them to set Y; For each failure mode , calculated and collected data The similarity between: ; in, and Respectively represent the failure mode and data The value of the jth feature in ; and Failure modes and data The mean of is the weight of the jth feature; By calculating the similarity of each failure mode, the priority of each failure mode is calculated: ; in, Indicates failure mode The priority of the fault mode; d is the total number of fault modes; Is the failure mode With current data similarity; Prioritize all failure modes Sort from high to low; Output the highest priority failure mode and the data with the highest similarity to the highest priority failure mode.

[0039] By setting the identified fault mode set Y and obtaining all fault modes, the problem of how to systematically manage and organize the identified fault modes is solved. In wireless communication networks, the number of fault modes is huge and may continue to increase over time. Therefore, centrally storing all identified fault modes in set Y can help the system effectively track historical fault modes and provide a complete fault mode reference for subsequent fault location. The maintenance and update of this set ensures that the system can continuously optimize and enrich the fault mode library as the network status changes. By building a complete fault mode set Y, the system can flexibly respond to new fault types and modes, and use existing fault modes for rapid comparison and analysis, thereby improving the accuracy and efficiency of fault location. The management of historical fault modes helps to reduce the detection and resolution time of similar problems in the future. By calculating the similarity between each fault mode and the collected data. The problem of how to compare the fault mode with the real-time collected data to accurately determine the fault type is solved. In actual networks, fault modes are diverse, and the characteristics of each fault mode may change at different time points or under different environmental conditions. By calculating the similarity between each fault mode and the collected data, the fault mode most relevant to the current data can be efficiently identified, avoiding simple rule matching or overly rough judgment methods. This step makes the matching between the fault mode and the real-time data more accurate by quantifying the similarity index. The similarity score between each fault mode and the data provides a clear basis for the subsequent priority calculation, further improving the accuracy of fault detection. This method can reduce the false alarm rate and the missed alarm rate, and improve the intelligent level of fault location. By calculating the priority of each fault mode. The problem of how to evaluate the severity and urgency of the fault mode based on similarity is solved. In wireless communication networks, different fault modes may have different impacts on the network. Some faults may cause major system crashes, while others may be small-scale problems. By calculating the priority of each fault mode, it can be ensured that the fault location system can give priority to faults that have a greater impact on the network, thereby improving the response speed and processing efficiency of network operation and maintenance. This step quantifies the priority of each fault mode, so that the system can automatically judge and focus on the most critical faults, avoiding waste of resources and redundancy in processing. Through priority sorting, network operation and maintenance personnel can quickly handle high-priority faults, reduce network interruption time, and improve network availability and stability. By sorting all fault modes from high to low priority, the problem of how to sort multiple fault modes and select the most urgent fault to handle is solved. Since multiple fault modes may be detected at the same time during the fault detection process, how to sort these fault modes and determine the processing order directly affects the efficiency and effectiveness of fault handling. By sorting fault modes by priority, the system can ensure that the most critical problems are solved first, thereby reducing the impact on network services.The sorting mechanism enables network operation and maintenance to quickly identify the most important faults and respond in the shortest possible time. In this way, the system can allocate resources more efficiently when facing multiple faults, improve the overall fault repair efficiency, reduce the recovery time of network faults, and optimize the quality of network services. By outputting the highest priority fault mode and the data with the highest similarity, the problem of how to associate the detected fault with specific data features and further analyze and locate it is solved. After completing the sorting of the fault modes, outputting the highest priority fault mode and the most similar collected data helps to accurately lock the time and specific problem of the fault, providing a more intuitive basis for troubleshooting. This step helps operation and maintenance personnel quickly locate the cause and background of the fault by outputting the highest priority fault mode and the most similar data. Through precise docking with data, operation and maintenance personnel can quickly make fault repair decisions based on specific historical data and similar network conditions, improving the efficiency and accuracy of fault diagnosis.

[0040] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0041] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A fault detection method for a wireless communication network, characterized in that: include: Collecting data from wireless communication networks; The collected data includes: Base station signal strength, denoted as , represents the signal strength from the base station to the user equipment at time t, in dBm; Network delay, denoted as , represents the communication delay between the user equipment and the base station at time t, in ms; Data packet loss rate, denoted as , represents the packet loss ratio of data transmission at time t, expressed as a percentage; The signal quality of the user equipment is denoted as , represents the quality index of the signal received by the user terminal equipment at time t, in dBm; Set the sampling period to The total observation time is , calculate the number of sampling points: ； in, is the total number of sampling points; Each sampling generates a data vector, which is recorded as: ； in, is the comprehensive data vector at time t, including signal strength, network delay, data packet loss rate and signal quality; Store the sampled data in the database in chronological order; Preprocess the collected data.

2. A fault detection method for a wireless communication network according to claim 1, characterized in that: The preprocessing of the collected data specifically includes: The data in the database are classified and extracted according to base station signal strength, network delay, data packet loss rate and user equipment signal quality, the data of the same type are grouped into a data set, and the classified and extracted data are sorted in the order of storage time; For any kind of dataset, perform the following steps: S1. Use the triple standard deviation method to detect outliers, specifically: Calculate the mean of all values ​​in a dataset: ； in, is the mean of the data set; M is the total number of all elements in the data set; is the value of the i-th data point in the data set; Calculate the standard deviation: ； in, is the standard deviation of the data set; When the data point Requirements: , then the data point Mark as outlier; in, For data points With the mean The absolute value of the difference; S2. Use linear interpolation to fill in missing values: When the data point When data is missing, linear interpolation is used to fill in the missing values, specifically: ； in, is the data value after filling; is the data point before the missing value; is the data point after the missing value; Perform feature extraction on the preprocessed dataset.

3. A method for fault detection in a wireless communication network according to claim 2, characterized in that: The feature extraction in the preprocessed data set specifically includes: Set the sliding window size to k and extract the following features at each time point: S3. Extract sliding mean: ； in, is the sliding mean at time t; k is the sliding window size; is the rth data point in the window; S4. Extract sliding standard deviation: ； in, is the sliding standard deviation at time t; S5. Extract change rate: ； in, is the rate of change at time t; is the data point at time t; is the data point at the starting time of the window; Combine the original data and features to form an extended feature vector: ； in, is the extended feature vector at time t; Based on the extended feature vector, a fault detection strategy is executed.

4. A method for fault detection in a wireless communication network according to claim 3, characterized in that: The fault detection strategy is executed based on the extended feature vector, specifically including: Set the normal operating range for each data type metric: S6、 ； S7、 ； S8、 ； S9、 ； Set the fault label: ； in, is the fault status label at time t; when , it indicates a fault; when , indicating normal.

5. A method for fault detection in a wireless communication network according to claim 4, characterized in that: The fault detection strategy is executed based on the extended feature vector, specifically including: Compute the similarity of the extended feature vectors: ； in, is the extended feature vector and similarity; , are the i-th and j-th extended feature vectors respectively; is the Gaussian kernel width parameter; Initialize cluster centers For randomly selected sample points, iterative update: ； in, is the k-th class center after the t+1th iteration; is the sample set contained in the kth class; is the number of samples in the kth category.

6. A method for fault detection in a wireless communication network according to claim 5, characterized in that: The fault detection strategy is executed based on the extended feature vector, specifically including: All extended feature vectors Randomly assign to K initial clusters, with the cluster center being a randomly selected sample point: ； in, is the center of the initial cluster k; p is a random index; Calculate the similarity between each sample and all cluster centers and assign the sample to the nearest cluster: ； in, is the sample set contained in the kth cluster; For sample To cluster center similarity; Update the cluster centers based on the sample set of each cluster: ； in, is the k-th class center after the t+1th iteration; is the k-th class sample set; is the number of samples in the kth category; Set the clustering convergence threshold to ; Check if the cluster center change is below a threshold : ； If the changes in all cluster centers are below the threshold, the clustering process terminates; Classify the faults.

7. A method for fault detection in a wireless communication network according to claim 6, characterized in that: The fault classification specifically includes: After clustering is completed, each cluster is mapped to a failure mode or normal state, specifically: Predefined Failure Mode ; Use majority voting to select each cluster Assign fault labels: ； in, is the label of cluster k; is the indicator function, if If yes, the value is 1, otherwise the value is 0; When there is a cluster that does not match the existing fault mode, it is marked as an unknown fault. , and record its characteristics; Locate the detected fault.

8. A method for fault detection in a wireless communication network according to claim 7, characterized in that: The fault locating of the detected fault specifically includes: Set the set of identified fault modes, denoted as set Y; Get all identified failure modes and add them to set Y; For each failure mode , calculated and collected data The similarity between: ； in, and Respectively represent the failure mode and data The value of the jth feature in ; and Failure modes and data The mean of is the weight of the jth feature; By calculating the similarity of each failure mode, the priority of each failure mode is calculated: ； in, Indicates failure mode The priority of the fault mode; d is the total number of fault modes; Is the failure mode With current data similarity; Prioritize all failure modes Sort from high to low; Output the highest priority failure mode and the data with the highest similarity to the highest priority failure mode.

Images