Base station vulnerability prediction methods, devices, equipment, media, and software products

Abstract

This invention provides a method, apparatus, device, medium, and program product for predicting base station vulnerabilities, relating to the field of computer technology. The method includes: preprocessing base station alarm data and performance data using base station resource data; inputting the preprocessed alarm data into a first time-series prediction model to output a first probability that the base station has vulnerabilities; inputting the preprocessed performance data into a second time-series prediction model to output a second probability that the base station has vulnerabilities; and determining the base station vulnerability information based on the first and second probabilities. This invention introduces intelligent AI model capabilities, multi-dimensionally selecting alarm data strongly correlated with base station vulnerabilities and base station performance data aggregated from a service usage perspective that directly characterizes user experience to form a vulnerability mining data system. The combination of these two systems generates a more accurate and automated base station vulnerability mining capability, thereby improving the accuracy and efficiency of base station vulnerability prediction.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to a method, apparatus, equipment, medium, and program product for predicting potential risks of base stations. Background Technology

[0002] With the rapid development of 5G networks, network architectures are becoming increasingly complex, making the stable operation of base stations particularly important. As a crucial component of wireless communication networks, the normal operation of base stations directly impacts user experience and network service quality. However, current wireless maintenance work faces numerous challenges and pain points.

[0003] First, the complexity of network management increases the difficulty of operation and maintenance (O&M) work, especially in the management of large base station clusters. Manual inspection is not only time-consuming and labor-intensive, but also difficult to guarantee efficiency and accuracy. Second, the lack of effective proactive O&M methods means that O&M personnel can often only respond passively after problems occur, unable to proactively identify and resolve potential risks. Furthermore, the current level of intelligence in O&M processes is still insufficient, failing to effectively utilize big data and artificial intelligence technologies for real-time monitoring and analysis, resulting in poor timeliness in risk identification and handling. Therefore, how to accurately and efficiently predict base station risks has become an urgent problem to be solved. Summary of the Invention

[0004] This invention provides a method, apparatus, device, medium, and program product for predicting base station potential hazards, in order to solve the defects of low accuracy and low efficiency in the prediction of base station potential hazards in the prior art, and to improve the accuracy and efficiency of base station potential hazard prediction.

[0005] This invention provides a method for predicting potential risks in base stations, comprising the following steps:

[0006] Obtain alarm data, performance data, and resource data from the base station;

[0007] The resource data is used to preprocess the alarm data and the performance data;

[0008] The preprocessed alarm data is input into a first time-series prediction model to obtain a first probability that the base station has a potential vulnerability, output by the first time-series prediction model; and the preprocessed performance data is input into a second time-series prediction model to obtain a second probability that the base station has a potential vulnerability, output by the second time-series prediction model; the first time-series prediction model is trained based on the preprocessed alarm sample data; the second time-series prediction model is trained based on the preprocessed performance sample data.

[0009] Based on the first probability and the second probability, the potential risks of the base station are determined.

[0010] According to a method for predicting potential risks of a base station provided by the present invention, determining the potential risk information of the base station based on the first probability and the second probability includes:

[0011] Based on the first probability, a first base station vulnerability score is determined for the base station in the current time period, and based on the second probability, a second base station vulnerability score is determined for the base station in the current time period.

[0012] The pre-set professional experience-based hazard score and the second base station hazard score are weighted and summed to obtain the third base station hazard score for the current time period;

[0013] The base station vulnerability scores of the first base station and the third base station are weighted and summed to obtain the base station vulnerability score of the base station in the current time period.

[0014] The base station vulnerability score in the current time period and the base station vulnerability score in the previous time period are weighted and summed to obtain the final vulnerability score of the base station.

[0015] Based on the final hazard score of the base station, the hazard information of the base station is determined.

[0016] According to a base station vulnerability prediction method provided by the present invention, the resource data is used to preprocess the alarm data, including:

[0017] Based on the resource data, obtain the engineering status of the base station;

[0018] Invalid alarm data is filtered out based on the engineering status of the base station to obtain valid alarm data;

[0019] Based on the valid alarm data, construct a training data matrix;

[0020] The training data matrix is ​​shrunk according to the alarm level of the valid alarm data to obtain the preprocessed alarm data.

[0021] According to the base station vulnerability prediction method provided by the present invention, the alarm level of the effective alarm data is determined based on the following method:

[0022] Based on the alarm labels of the valid alarm data, determine the alarm level of the valid alarm data; or...

[0023] Calculate the alarm percentage of each type of valid alarm data;

[0024] The alarm level of the valid alarm data is determined based on the alarm percentage.

[0025] According to a base station vulnerability prediction method provided by the present invention, the resource data is used to preprocess the performance data, including:

[0026] The performance data is cleaned to remove invalid, duplicate, or abnormal data.

[0027] The resource data is used to integrate the cleaned performance data to obtain the preprocessed performance data.

[0028] According to the base station vulnerability prediction method provided by the present invention, the first time-series prediction model is trained in the following manner:

[0029] Using the preprocessed alarm sample data, a preset time-series feature model is trained to obtain the first time-series prediction model.

[0030] The second time-series prediction model was trained in the following way:

[0031] Using the preprocessed performance sample data, a preset time-series feature model is trained to obtain the second time-series prediction model.

[0032] The present invention also provides a base station hazard prediction device, comprising the following modules:

[0033] The acquisition module is used to acquire alarm data, performance data, and resource data from the base station;

[0034] The preprocessing module is used to preprocess the alarm data and the performance data using the resource data;

[0035] The first potential hazard prediction module is used to input preprocessed alarm data into a first time-series prediction model to obtain a first probability that the base station has a potential hazard, output by the first time-series prediction model; and to input preprocessed performance data into a second time-series prediction model to obtain a second probability that the base station has a potential hazard, output by the second time-series prediction model; the first time-series prediction model is trained based on preprocessed alarm sample data; the second time-series prediction model is trained based on preprocessed performance sample data.

[0036] The second hazard prediction module is used to determine the hazard information of the base station based on the first probability and the second probability.

[0037] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement any of the base station vulnerability prediction methods described above.

[0038] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the base station vulnerability prediction method as described above.

[0039] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements any of the base station vulnerability prediction methods described above.

[0040] The base station vulnerability prediction method, apparatus, equipment, medium, and program products provided by this invention acquire alarm data, performance data, and resource data of the base station; preprocess the alarm data and performance data using the resource data; input the preprocessed alarm data into a first time-series prediction model to obtain a first probability of a base station vulnerability output by the first time-series prediction model; and input the preprocessed performance data into a second time-series prediction model to obtain a second probability of a base station vulnerability output by the second time-series prediction model. The first time-series prediction model is trained based on the preprocessed alarm sample data; the second time-series prediction model is trained based on the preprocessed performance sample data; and the vulnerability information of the base station is determined based on the first and second probabilities. This invention introduces intelligent AI model capabilities, selects alarm data strongly correlated with base station vulnerabilities from multiple dimensions, and forms a vulnerability mining data system by combining base station performance data aggregated from the perspective of service use that directly represents user perception. The combination of these two data sources generates a more accurate and automated base station vulnerability mining capability, thereby improving the accuracy and efficiency of base station vulnerability prediction. Attached Figure Description

[0041] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0042] Figure 1 This is one of the flowcharts of the base station hidden danger prediction method provided by the present invention.

[0043] Figure 2 This is the second flowchart of the base station hidden danger prediction method provided by the present invention.

[0044] Figure 3 This is a schematic diagram of the base station hidden danger prediction device provided by the present invention.

[0045] Figure 4 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0047] The following is combined Figures 1-4 This invention describes the base station vulnerability prediction method, apparatus, equipment, medium, and program products.

[0048] Figure 1 This is one of the flowcharts illustrating the base station vulnerability prediction method provided by the present invention, such as... Figure 1 As shown, the method includes the following:

[0049] Step 101: Obtain alarm data, performance data, and resource data from the base station.

[0050] Alarm data is provided by the fault management system and contains information about anomalies or faults occurring during base station operation. When collecting alarm data, it is necessary to filter and select alarms based on domain knowledge and alarm titles, retaining only those relevant to identifying potential base station vulnerabilities. Simultaneously, key information such as alarm time, alarm title, alarm object name, and alarm level must be recorded for subsequent analysis and processing.

[0051] The performance data primarily comes from two sources: 1) Data sharing platform: This platform mainly provides DPI (Deep Packet Inspection) performance data. This data originates from XDR (eXtensible Data Record) documents generated by users during service usage. XDR documents record detailed information about users' mobile internet, voice, and SMS services, such as traffic, TCP (Transmission Control Protocol) connections, and HTTP (Hypertext Transfer Protocol) requests. Analyzing this data allows for understanding the base station's performance under different service scenarios. 2) Network optimization platform: This platform mainly provides raw OMC (Operation and Maintenance Center) performance data and processed base station anomaly tag data. Raw OMC data is performance indicators directly extracted from the equipment side, reflecting the real-time operating status of the base station. Processed base station anomaly tag data is obtained by analyzing and processing the raw data, used to identify potential anomalies in the base station, such as high load, weak coverage, and high interference. This tag data helps to more accurately determine the potential problems of the base station. Performance data needs to be obtained from two channels: data sharing platform and network optimization platform. By integrating these two types of data, a more comprehensive and accurate performance indicator system can be provided for the base station hidden danger discovery module, thereby improving the accuracy of hidden danger prediction.

[0052] Resource data primarily originates from the resource management system and includes basic information and engineering status of base stations. The basic information provides fundamental details about the base station, such as its location, type, and equipment model. This information helps in understanding the base station's basic characteristics and operating environment, providing crucial reference value for subsequent hazard identification and analysis. The engineering status describes the current engineering condition of the base station, indicating whether it is undergoing maintenance or upgrades. When identifying potential hazards, base stations under construction must be excluded, as their performance and alarm data may be affected by these activities, leading to inaccurate data analysis results. Therefore, the engineering status of base stations allows for filtering, retaining only those not under construction for hazard identification. Resource data plays a crucial role in screening and filtering during base station hazard identification. Understanding the basic information and assessing the engineering status of base stations ensures the accuracy and effectiveness of hazard identification.

[0053] Step 102: Use the resource data to preprocess the alarm data and the performance data.

[0054] In the preprocessing of alarm and performance data, the first step is to perform group filtering on base stations. This ensures that subsequent data processing only targets base stations not in an engineering state, as base stations in an engineering state may have data anomalies due to maintenance, upgrades, or other reasons, making them unsuitable for hazard identification. Performance data provides basic information and engineering status data for base stations. This data allows us to identify base stations currently in an engineering state and exclude them from the dataset. This leaves only base station data in a non-engineering state, providing an accurate data foundation for subsequent hazard identification.

[0055] During the preprocessing of alarm data, it is necessary to filter base station alarms based on their specialty and alarm title. At this point, basic base station information from resource data, such as base station name and type, can be used as one of the filtering criteria to ensure that the filtered alarm data is relevant to the target base station. Furthermore, the filtering of alarm data can be further optimized based on the base station's engineering status data. For example, if a base station frequently issues alarms within a short period and is under engineering maintenance, these alarms may be due to temporary problems caused by engineering operations, rather than genuine hidden dangers. Therefore, excluding base stations under engineering maintenance can improve the accuracy and reliability of the alarm data.

[0056] During the preprocessing of performance data, the geographical location information in the basic information of the base station can help understand the performance differences in different areas; while the base station engineering status data can be used to explain whether abnormal fluctuations in certain performance indicators are related to engineering operations.

[0057] In addition, resource data also provides other auxiliary information related to base station performance, such as base station load and coverage. This information can serve as a reference factor in the performance data processing process, helping to more accurately assess the base station performance and identify potential problems.

[0058] Step 103: Input the preprocessed alarm data into the first time series prediction model to obtain the first probability that the base station has a hidden danger, output by the first time series prediction model; and input the preprocessed performance data into the second time series prediction model to obtain the second probability that the base station has a hidden danger, output by the second time series prediction model.

[0059] It should be noted that both the first and second time-series prediction models were pre-trained. The first time-series prediction model was trained based on pre-processed alarm sample data, while the second time-series prediction model was trained based on pre-processed performance sample data.

[0060] In one embodiment, preprocessed alarm sample data is used to train a preset time-series feature model to obtain a first time-series prediction model; preprocessed performance sample data is used to train the preset time-series feature model to obtain a second time-series prediction model. For example, firstly, data for model training needs to be prepared. This data includes alarm data strongly correlated with base station vulnerabilities and base station performance data directly representing user perception. These two types of data are modeled separately, therefore, separate datasets for each type need to be prepared. Based on data characteristics and application scenarios, a time-series feature model is selected as the model to be used. Among these models, LSTM, TCN, nheats, and Nhits are selected for comparative verification. By comparing the performance of these models, the Nhits algorithm, which performs better, is ultimately selected. The advantages of the Nhits algorithm include: adding a downsampling layer before the fully connected structure, reducing memory overhead and computational complexity while ensuring long sequence dependencies; Nhits can hierarchically synchronize input and output scales, allowing blocks to focus on predicting their own time series; and compared to nheats, Nhits does not require setting stack_types to directly construct the stack. Then, using weekly data as a baseline, a training cycle of four weeks is employed. This means the model will be trained within a four-week time window and then predict the probability of base station vulnerabilities in the next cycle. During model training, the model's performance needs to be evaluated. This can be achieved by calculating the error between the model's predicted base station vulnerability probability and the actual base station vulnerabilities that occur. If the error is large, it indicates that the model's prediction is inaccurate, and the model needs to be adjusted and optimized. Optimization methods may include adjusting model parameters, increasing the amount of training data, and adopting a more complex model structure. Through continuous trial and optimization, the model configuration best suited to the current data and task can be found.

[0061] After training the first time-series prediction model and the second time-series prediction model, the preprocessed alarm data is input into the first time-series prediction model to obtain the first probability that the base station has hidden dangers, and the preprocessed performance data is input into the second time-series prediction model to obtain the second probability that the base station has hidden dangers, as output by the second time-series prediction model.

[0062] Step 104: Determine the potential risks of the base station based on the first probability and the second probability.

[0063] The first and second probabilities are combined to comprehensively determine the potential risks of the base station, such as the likelihood of the base station experiencing a malfunction or anomaly in the future. In practical applications, this probability value can be used to determine whether the base station needs further inspection and maintenance. Generally, a high probability value indicates a significant potential risk that requires priority attention; a low probability value may mean that minimal intervention is not necessary at the moment.

[0064] In one embodiment, a first base station vulnerability score is determined for the base station in the current time period based on a first probability, and a second base station vulnerability score is determined for the base station in the current time period based on a second probability. A weighted sum of the preset professional experience vulnerability score and the second base station vulnerability score is then obtained to obtain a third base station vulnerability score for the base station in the current time period. A weighted sum of the first and third base station vulnerability scores is then obtained to obtain the base station vulnerability score for the base station in the current time period. A weighted sum of the base station vulnerability score for the base station in the current time period and the base station vulnerability score for the base station in the previous time period is then obtained to obtain the final vulnerability score for the base station. Based on the final vulnerability score, the vulnerability information of the base station is determined. Here, the current time period can be understood as this week or the current period, and the previous time period can be understood as the previous week.

[0065] Understandably, both the first and second time-series prediction models output a value representing the probability of a base station having a potential hazard. This value can be a number between 0 and 1, with a higher probability of hazard being closer to 1. The hazard score, on the other hand, is a specific numerical value used to quantify this probability. Generally, the higher the probability of a hazard, the higher the corresponding hazard score, and vice versa.

[0066] refer to Figure 2 For example, assuming the probability of a hazard is P (ranging from 0 to 1) and the hazard score is S (a non-negative integer), a linear function can be defined to map the hazard probability to the hazard score:

[0067] S = f(P) = a × P + b;

[0068] Here, a and b are constants used to adjust the specific shape and position of the mapping relationship.

[0069] Assuming the first probability is P1, the corresponding first base station vulnerability score is S1; assuming the second probability is P2, the corresponding second base station vulnerability score is S2; and the preset professional experience vulnerability score is S3, then the third base station vulnerability score S4 predicted based on performance data for the current week is:

[0070] S4 = S2 × a2 + S3 × a3;

[0071] Where a2 is the weight of the second base station hidden danger score S2, and a3 is the weight of the professional experience hidden danger score S3.

[0072] The base station's vulnerability score (S5) for the current week is:

[0073] S5 = S1 × a1 + S4 × a4;

[0074] Where a1 is the weight of the first base station's hidden danger score S1, and a4 is the weight of the third base station's hidden danger score S4.

[0075] The final vulnerability score S6 for the base station is:

[0076] S6 = S5 × a5 + S7 × a7;

[0077] Where a5 is the weight of the base station vulnerability score S5 in the current week, S7 is the base station vulnerability score in the previous week, and a7 is the weight of the base station vulnerability score S7 in the previous week.

[0078] Finally, based on the final hazard score of the base station, the hazard information of the base station is determined.

[0079] The base station vulnerability prediction method provided in this invention acquires alarm data, performance data, and resource data of the base station; preprocesses the alarm data and performance data using the resource data; inputs the preprocessed alarm data into a first time-series prediction model to obtain a first probability of a base station vulnerability output by the first time-series prediction model, and inputs the preprocessed performance data into a second time-series prediction model to obtain a second probability of a base station vulnerability output by the second time-series prediction model; the first time-series prediction model is trained based on the preprocessed alarm sample data; the second time-series prediction model is trained based on the preprocessed performance sample data; and the vulnerability information of the base station is determined based on the first and second probabilities. This invention introduces intelligent AI model capabilities, multi-dimensionally selects alarm data strongly correlated with base station vulnerabilities, and base station performance data aggregated from a service usage perspective that directly characterizes user perception to form a vulnerability mining data system. The combination of these two systems generates a more accurate and automated base station vulnerability mining capability, thereby improving the accuracy and efficiency of base station vulnerability prediction.

[0080] Based on the above embodiments, the alarm data is preprocessed using the resource data, including:

[0081] Step 210: Obtain the engineering status of the base station based on the resource data;

[0082] Step 211: Filter invalid alarm data according to the engineering status of the base station to obtain valid alarm data;

[0083] Step 212: Construct a training data matrix based on the valid alarm data;

[0084] Step 213: According to the alarm level of the valid alarm data, shrink the training data matrix to obtain the preprocessed alarm data.

[0085] By analyzing the base station's engineering status in the resource data, alarm data is initially filtered, retaining only alarm data generated by base stations currently in a non-engineering state (i.e., not undergoing maintenance or upgrades) (i.e., valid alarm data). This is because base stations in an engineering state may generate a large amount of temporary or atypical alarm data (i.e., invalid alarm data) due to construction or adjustments, which may interfere with normal hazard identification and analysis. Therefore, filtering invalid alarm data can improve the accuracy and effectiveness of subsequent analysis.

[0086] Optionally, basic base station information from resource data, such as base station type, location, and equipment model, can be used to further classify alarm data. Different types or locations of base stations may face different operating environments and challenges, thus their alarms may have different characteristics and meanings. Classifying alarm data allows for a better understanding of the characteristics and distribution patterns of various alarms, providing more targeted input for subsequent model training. On the other hand, in the alarm screening process, in addition to considering the nature of the alarm itself (such as severity and duration), other relevant information from the resource data can be combined for comprehensive judgment. For example, certain types of base stations or base stations located in specific geographical locations may be more susceptible to certain factors that trigger alarms. By comprehensively considering these factors, truly noteworthy alarm information can be identified more accurately. Based on this, a filtered and optimized alarm dataset can be obtained. This dataset contains alarm information highly relevant to base station hazard discovery and has had most noise and redundant data removed. Such a dataset is more suitable for subsequent model training and hazard discovery work.

[0087] Furthermore, from the perspective of the base station, the effective alarm data is used to generate an n×m matrix, where n represents the alarm data used for training over n days, and m represents the alarm types of the base station over n days, as shown in Table 1 below.

[0088] Table 1

[0089]

[0090] In this training data matrix, the data in the i-th row and j-th column indicates whether a base station issued alarm number j on day i (1 for yes, 0 for no). This type of training data matrix contains a large number of zeros, making it very sparse and unfavorable for identifying patterns in potential problems. Therefore, the training data needs to be processed further.

[0091] The problem of matrix sparsity is solved by shrinking the training data matrix by converting the alarm titles in Table 1 into high alarm, medium alarm, and low alarm (as shown in Table 2 below).

[0092] Table 2

[0093]

[0094] In one embodiment, the alarm level of valid alarm data is determined based on the following methods: determining the alarm level of valid alarm data according to the alarm label of the valid alarm data; or, calculating the alarm proportion of each type of valid alarm data; and determining the alarm level of valid alarm data based on the alarm proportion. For example, narrowing down the alarms based on their own labeled alarm levels: taking the i-th row as an example, alarms at level one are defined as high alarms, alarms at levels two and three are defined as medium alarms, and alarms below level three are uniformly defined as low alarms. This method can classify alarms according to their own severity, which helps to more accurately reflect the actual situation of alarms. Narrowing down the alarms based on the alarm quantity proportion: taking base station A as an example, the alarm proportion of base station A is calculated by dividing the number of alarms generated by base station A by the total number of alarms generated by all base stations, and the alarm proportion is divided into three intervals, thus converting them into high alarms, medium alarms, and low alarms. This method can classify alarms according to their distribution across all base stations, which helps to more comprehensively consider the scope and severity of alarm impact. After shrinking the data into two training matrices using the two methods described above, they can be trained separately and complement each other. This fully utilizes the advantages of different shrinkage methods, improving the model's accuracy and robustness. Furthermore, since both training data matrices are generated from the same original data, they possess a degree of complementarity and consistency, which helps in better identifying and preventing potential base station vulnerabilities.

[0095] The embodiments of the present invention can effectively remove noise and redundant information from the original alarm data by filtering base stations in non-engineering states, classifying alarms based on the basic information of base stations, and optimizing alarm filtering by combining resource data. This makes the preprocessed alarm data more accurate and targeted, which helps to improve the accuracy and effectiveness of subsequent model training.

[0096] Based on the above embodiments, the performance data is preprocessed using the resource data, including:

[0097] Step 220: Perform data cleaning on the performance data to remove invalid, duplicate, or abnormal data;

[0098] Step 221: Using the resource data, integrate the cleaned performance data to obtain the preprocessed performance data.

[0099] Performance data includes DPI performance data and OMC performance data, and the preprocessing methods for the two types of data are different:

[0100] 1) Base Station-Level DPI Performance Data Processing: DPI performance data is obtained by aggregating and processing XDR (Extended Data Recording) documents generated by users using mobile internet, voice, and SMS services. These documents record user behavior data in various service scenarios, including metrics such as traffic, TCP connections, and HTTP requests. The collected DPI performance data is cleaned to remove invalid, duplicate, or abnormal data. For example, this involves checking for duplicate XDR documents, analyzing missing data for each metric, and using statistical methods to detect and handle outliers. Through this data processing, base station-level performance metrics can be obtained, such as registration counts, request counts, handover counts, location update counts, and carrying capacity. Control plane metrics mainly include registration, request, handover, location update, and carrying capacity, reflecting the base station's performance at the network control layer. User plane metrics include traffic, TCP connections, and HTTP requests, reflecting the user's actual experience and behavioral patterns when using base station services. During processing, key metrics affecting base station performance need to be selected as the metric system for model training to facilitate subsequent risk identification and prediction.

[0101] 2) OMC data processing at the base station level: OMC data is a performance metric directly extracted from the equipment side. These metrics complement DPI data, together forming a more complete metric system. OMC data provides more comprehensive and accurate equipment performance information, helping to compensate for the shortcomings of DPI data in certain aspects.

[0102] Furthermore, resource data is used to integrate performance data from different sources (such as DPI and OMC) into a unified dataset to facilitate subsequent analysis. For example, base station IDs are obtained based on resource data, and DPI performance data and OMC data are merged according to the base station ID and timestamp to form a comprehensive dataset containing all relevant performance indicators. At the same time, indicators in different formats or units are standardized, such as converting traffic from MB to GB to ensure unit consistency.

[0103] Optionally, base station hazard labels professionally developed by the network optimization system, such as high load, weak coverage, and high interference, can be directly used to improve the model training results. These labels are based on professional knowledge and experience in assessing base station status and have high reference value. Meanwhile, to ensure the accuracy of hazard prediction, the hazard probability from model training and the existing hazard labels can each be assigned a certain weight (e.g., 5%), and then the final hazard score of the base station can be obtained by weighting them. This comprehensively considers the influence of multiple factors on base station hazards, improving the accuracy and reliability of the prediction.

[0104] By processing and analyzing DPI performance data and OMC performance data, this invention can obtain a comprehensive base station performance index system. Furthermore, by combining the hidden danger labels of the network optimization system and the model training results, the accuracy of hidden danger prediction can be further improved.

[0105] To further explain the base station vulnerability prediction method proposed in this invention, please refer to the following embodiments.

[0106] The base station vulnerability prediction method proposed in this invention innovatively introduces intelligent AI model capabilities through an encapsulated base station data acquisition module, model training preprocessing module, and base station vulnerability mining module, thereby better improving the precision management and control capabilities and the level of intelligence in operation and maintenance. The main functions of each module are as follows:

[0107] Base station data acquisition module: Primarily provides automatic data acquisition capabilities, enabling the scheduled collection of alarm, performance, and resource data. Alarm data is provided by the fault management system, with base station alarms filtered based on specialty and alarm title. Performance data comes from two sources: one is DPI performance data (XDR documents generated during user service usage); the other is raw OMC performance data and processed base station anomaly tag data. Resource data mainly consists of basic base station information and base station engineering status.

[0108] Model training preprocessing module: mainly performs data preprocessing on the training data of the base station hidden danger mining module, performs matrix convergence on the sparsity characteristics of alarm data through high, medium and low alarms to ensure the model effect, and integrates and processes the performance data of directly collected equipment and the performance data of business usage to form a complete training wide table.

[0109] Base Station Hazard Discovery Module: This module provides AI-based hazard discovery capabilities, encompassing algorithm selection and model training. To ensure accurate hazard discovery, the training data is multi-dimensionally selected, including alarm data strongly correlated with hazard discoveries and base station performance data aggregated from a business usage perspective that directly reflects user experience. These two types of data are modeled separately, and the final hazard probability outputs from each model are weighted to form a final hazard score.

[0110] This invention innovatively deploys AI algorithms for base station vulnerability detection in frontline base station maintenance, enabling proactive and rapid operation and maintenance (O&M). It serves closed-loop intelligent O&M for base stations, improving the efficiency of O&M resource utilization and enhancing customer experience. Key benefits include: precise inspection by outsourced maintenance providers (shifting from full-site inspections to on-demand, precise inspections, effectively reducing inspection frequency and costs); proactive O&M reducing base station failure rates (predicting and addressing potential base station vulnerabilities in advance, preventing service outages, ensuring base station operational stability, and improving user experience); a complete and closed-loop intelligent O&M system (filling the gap in proactive O&M methods, using AI to locate base station vulnerabilities, saving significant manpower in indicator analysis, and effectively supplementing existing fault monitoring systems); and improved O&M efficiency and reduced costs (precisely allocating limited O&M resources to sub-optimal sites, improving work efficiency through intelligent means, and reducing O&M costs such as fault handling and inspection).

[0111] The base station hazard prediction device provided by the present invention is described below. The base station hazard prediction device described below can be referred to in correspondence with the base station hazard prediction method described above.

[0112] refer to Figure 3 The base station hazard prediction device provided by the present invention includes an acquisition module 301, a preprocessing module 302, a first hazard prediction module 303, and a second hazard prediction module 304.

[0113] The acquisition module 301 is used to acquire alarm data, performance data and resource data of the base station;

[0114] Preprocessing module 302 is used to preprocess the alarm data and the performance data using the resource data;

[0115] The first potential hazard prediction module 303 is used to input preprocessed alarm data into a first time-series prediction model to obtain a first probability that the base station has a potential hazard, output by the first time-series prediction model; and to input preprocessed performance data into a second time-series prediction model to obtain a second probability that the base station has a potential hazard, output by the second time-series prediction model; the first time-series prediction model is trained based on preprocessed alarm sample data; the second time-series prediction model is trained based on preprocessed performance sample data.

[0116] The second hazard prediction module 304 is used to determine the hazard information of the base station based on the first probability and the second probability.

[0117] The base station vulnerability prediction device provided in this invention acquires alarm data, performance data, and resource data of the base station; preprocesses the alarm data and performance data using the resource data; inputs the preprocessed alarm data into a first time-series prediction model to obtain a first probability of a base station vulnerability output by the first time-series prediction model; and inputs the preprocessed performance data into a second time-series prediction model to obtain a second probability of a base station vulnerability output by the second time-series prediction model. The first time-series prediction model is trained based on the preprocessed alarm sample data; the second time-series prediction model is trained based on the preprocessed performance sample data; and the vulnerability information of the base station is determined based on the first and second probabilities. This invention introduces intelligent AI model capabilities, selects alarm data strongly correlated with base station vulnerabilities from multiple dimensions, and forms a vulnerability mining data system by combining base station performance data aggregated from the perspective of service use that directly characterizes user perception. The combination of these two data sources generates a more accurate and automated base station vulnerability mining capability, thereby improving the accuracy and efficiency of base station vulnerability prediction.

[0118] In one embodiment, the second hazard prediction module 304 is specifically used for:

[0119] Based on the first probability, a first base station vulnerability score is determined for the base station in the current time period, and a second base station vulnerability score is determined for the base station in the current time period based on the second probability. A weighted sum of a preset professional experience vulnerability score and the second base station vulnerability score is obtained to get a third base station vulnerability score for the base station in the current time period. A weighted sum of the first base station vulnerability score and the third base station vulnerability score is obtained to get a base station vulnerability score for the base station in the current time period. A weighted sum of the base station vulnerability score for the base station in the current time period and the base station vulnerability score for the base station in the previous time period is obtained to get a final vulnerability score for the base station. Based on the final vulnerability score of the base station, the vulnerability information of the base station is determined.

[0120] In one embodiment, the preprocessing module 302 is specifically used for:

[0121] Based on the resource data, the engineering status of the base station is obtained; invalid alarm data is filtered based on the engineering status of the base station to obtain valid alarm data; a training data matrix is ​​constructed based on the valid alarm data; the training data matrix is ​​shrunk according to the alarm level of the valid alarm data to obtain the preprocessed alarm data.

[0122] In one embodiment, the preprocessing module 302 is specifically used for:

[0123] Based on the alarm labels of the valid alarm data, determine the alarm level of the valid alarm data; or, calculate the alarm percentage of each type of valid alarm data; and determine the alarm level of the valid alarm data based on the alarm percentage.

[0124] In one embodiment, the preprocessing module 302 is specifically used for:

[0125] The performance data is cleaned to remove invalid, duplicate, or abnormal data; the resource data is then used to integrate the cleaned performance data to obtain the preprocessed performance data.

[0126] In one embodiment, the first hazard prediction module 303 is further configured to:

[0127] Using the preprocessed alarm sample data, a preset time-series feature model is trained to obtain the first time-series prediction model; the second time-series prediction model is trained based on the following method: using the preprocessed performance sample data, a preset time-series feature model is trained to obtain the second time-series prediction model.

[0128] Figure 4 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 4 As shown, the electronic device may include: a processor 410, a communications interface 420, a memory 430, and a communication bus 440, wherein the processor 410, the communications interface 420, and the memory 430 communicate with each other through the communication bus 440. The processor 410 can call logical instructions in the memory 430 to execute a base station vulnerability prediction method. The method includes: acquiring alarm data, performance data, and resource data of the base station; using the resource data to preprocess the alarm data and the performance data; inputting the preprocessed alarm data into a first time-series prediction model to obtain a first probability that the base station has a vulnerability, output by the first time-series prediction model; and inputting the preprocessed performance data into a second time-series prediction model to obtain a second probability that the base station has a vulnerability, output by the second time-series prediction model; the first time-series prediction model is trained based on the preprocessed alarm sample data; the second time-series prediction model is trained based on the preprocessed performance sample data; and determining the vulnerability information of the base station based on the first probability and the second probability.

[0129] Furthermore, the logical instructions in the aforementioned memory 430 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0130] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the base station vulnerability prediction method provided by the above methods. The method includes: acquiring alarm data, performance data, and resource data of a base station; using the resource data to preprocess the alarm data and the performance data; inputting the preprocessed alarm data into a first time-series prediction model to obtain a first probability that the base station has a vulnerability, output by the first time-series prediction model; and inputting the preprocessed performance data into a second time-series prediction model to obtain a second probability that the base station has a vulnerability, output by the second time-series prediction model; the first time-series prediction model is trained based on the preprocessed alarm sample data; the second time-series prediction model is trained based on the preprocessed performance sample data; and determining the vulnerability information of the base station based on the first probability and the second probability.

[0131] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements the base station vulnerability prediction method provided by the above methods. The method includes: acquiring alarm data, performance data, and resource data of a base station; preprocessing the alarm data and performance data using the resource data; inputting the preprocessed alarm data into a first time-series prediction model to obtain a first probability that the base station has a vulnerability, output by the first time-series prediction model; and inputting the preprocessed performance data into a second time-series prediction model to obtain a second probability that the base station has a vulnerability, output by the second time-series prediction model; the first time-series prediction model is trained based on the preprocessed alarm sample data; the second time-series prediction model is trained based on the preprocessed performance sample data; and determining the vulnerability information of the base station based on the first probability and the second probability.

[0132] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0133] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0134] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A base station hazard prediction method, characterized by, include: Acquire alarm data, performance data, and resource data from the base station; the performance data includes deep packet inspection (DPI) performance data and / or operation and maintenance center (OMC) performance data. The resource data is used to preprocess the alarm data and the performance data; The preprocessed alarm data is input into a first time-series prediction model to obtain a first probability that the base station has a potential vulnerability, output by the first time-series prediction model; and the preprocessed performance data is input into a second time-series prediction model to obtain a second probability that the base station has a potential vulnerability, output by the second time-series prediction model; the first time-series prediction model is trained based on the preprocessed alarm sample data; the second time-series prediction model is trained based on the preprocessed performance sample data. Based on the first probability and the second probability, the potential risks of the base station are determined; The step of determining the potential risks of the base station based on the first probability and the second probability includes: Based on the first probability, a first base station vulnerability score is determined for the base station in the current time period, and based on the second probability, a second base station vulnerability score is determined for the base station in the current time period. The pre-set professional experience-based hazard score and the second base station hazard score are weighted and summed to obtain the third base station hazard score for the current time period; The base station vulnerability scores of the first base station and the third base station are weighted and summed to obtain the base station vulnerability score of the base station in the current time period. The base station vulnerability score in the current time period and the base station vulnerability score in the previous time period are weighted and summed to obtain the final vulnerability score of the base station. Based on the final hazard score of the base station, the hazard information of the base station is determined.

2. The base station hazard prediction method of claim 1, wherein Using the resource data, the alarm data is preprocessed, including: Based on the resource data, obtain the engineering status of the base station; Invalid alarm data is filtered out based on the engineering status of the base station to obtain valid alarm data; Based on the valid alarm data, construct a training data matrix; The training data matrix is ​​shrunk according to the alarm level of the valid alarm data to obtain the preprocessed alarm data.

3. The base station hidden danger prediction method according to claim 2, characterized in that, The alarm level of the valid alarm data is determined based on the following method: Based on the alarm labels of the valid alarm data, the alarm level of the valid alarm data is determined; or, Calculate the alarm percentage of each type of valid alarm data; The alarm level of the valid alarm data is determined based on the alarm percentage.

4. The base station hidden danger prediction method according to claim 1, characterized in that, Using the resource data, the performance data is preprocessed, including: The performance data is cleaned to remove invalid, duplicate, or abnormal data. The resource data is used to integrate the cleaned performance data to obtain the preprocessed performance data.

5. The base station hidden danger prediction method according to any one of claims 1 to 4, characterized in that, The first time-series prediction model was trained in the following way: Using the preprocessed alarm sample data, a preset time-series feature model is trained to obtain the first time-series prediction model. The second time-series prediction model was trained in the following way: Using the preprocessed performance sample data, a preset time-series feature model is trained to obtain the second time-series prediction model.

6. A base station hazard prediction device, characterized in that, include: The acquisition module is used to acquire alarm data, performance data, and resource data of the base station; the performance data includes deep packet inspection (DPI) performance data and / or operation and maintenance center (OMC) performance data. The preprocessing module is used to preprocess the alarm data and the performance data using the resource data; The first potential hazard prediction module is used to input preprocessed alarm data into a first time-series prediction model to obtain a first probability that the base station has a potential hazard, output by the first time-series prediction model; and to input preprocessed performance data into a second time-series prediction model to obtain a second probability that the base station has a potential hazard, output by the second time-series prediction model; the first time-series prediction model is trained based on preprocessed alarm sample data; the second time-series prediction model is trained based on preprocessed performance sample data. The second hazard prediction module is used to determine the hazard information of the base station based on the first probability and the second probability. The second hidden danger prediction module is further configured to determine the first base station hidden danger score of the base station in the current time period based on the first probability, and to determine the second base station hidden danger score of the base station in the current time period based on the second probability. The pre-set professional experience-based hazard score and the second base station hazard score are weighted and summed to obtain the third base station hazard score for the current time period; The base station vulnerability scores of the first base station and the third base station are weighted and summed to obtain the base station vulnerability score of the base station in the current time period. The base station vulnerability score in the current time period and the base station vulnerability score in the previous time period are weighted and summed to obtain the final vulnerability score of the base station. Based on the final hazard score of the base station, the hazard information of the base station is determined.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the base station vulnerability prediction method as described in any one of claims 1 to 5.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the base station vulnerability prediction method as described in any one of claims 1 to 5.

9. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the base station vulnerability prediction method as described in any one of claims 1 to 5.

Images