Base station coverage anomaly detection method based on big data analysis

Abstract

The invention discloses a base station coverage anomaly detection method based on big data analysis, and belongs to the technical field of base station coverage detection in the technical field of communication. Aiming at the problems that the existing base station coverage anomaly detection is insufficient in accuracy and cannot adapt to dynamic environment change, the method comprises the following steps of: acquiring operation data such as received signal strength indication values of all base stations in a target area for continuous 1-30 days and 10-60 minutes, performing dimension reduction processing, and calculating a coverage characteristic value obtained by multi-dimensional index linear weighting; the method comprises the following steps: acquiring a coverage anomaly probability value, inputting the coverage anomaly probability value into an isolated forest model which is trained by historical data and can be dynamically updated, outputting the coverage anomaly probability value, and when the coverage anomaly probability value is in an interval greater than 0.7 and less than or equal to 0.9, marking a corresponding base station position on an electronic map as a coverage anomaly area. The method can accurately position the abnormal coverage area of the base station, is suitable for real-time monitoring and abnormity identification of the coverage state of the base station in communication network maintenance, and improves the network optimization efficiency and the service quality.

Description

Technical Field

[0001] The present invention belongs to the field of communication technologies, and particularly relates to a method for detecting abnormal base station coverage based on big data analysis. Background Art

[0002] In the operation and maintenance scenario of a wireless communication network, the accurate detection of the base station coverage status is the core link to ensure the network service quality. With the evolution of communication technologies, the types of base stations are becoming increasingly rich. For example, 4G / 5G dual-mode base stations, 5G NR base stations supporting millimeter wave bands, etc. The operation data generated by them shows multi-source heterogeneous characteristics. These data not only have multiple dimensions, but also are continuously collected at minute-level intervals, forming a large amount of time-series data in continuous time periods. As a result, traditional detection methods based on single-index thresholds or simple statistical analysis are difficult to establish effective associations at the level of heterogeneous data fusion, and cannot capture in real time the abnormal coverage patterns co-represented by multi-dimensional features, resulting in insufficient accuracy and timeliness of anomaly recognition.

[0003] At the level of big data processing, the real-time analysis of a large amount of time-series data faces the dual challenges of computing resource consumption and processing efficiency. On the one hand, there is a large amount of redundant information in the original data. For example, the repeated signal features continuously collected for a long time will significantly increase the computational complexity if directly processed, resulting in an increase in detection latency. On the other hand, traditional dimensionality reduction techniques are difficult to balance the retention of feature information and the optimization of computational efficiency when dealing with high-dimensional time-series data. The above problems make the existing technologies unable to quickly complete data processing and anomaly discrimination in the face of scenarios such as sudden traffic peaks and meteorological environment changes, and it is difficult to meet the requirements of network operation and maintenance for minute-level responses.

[0004] In a dynamic network environment, the base station coverage status shows non-linear changes under the influence of multiple factors. Geographical environment differences, such as dense urban building clusters and open suburban terrains, time-periodic service loads, such as traffic fluctuations during morning and evening rush hours, and extreme meteorological conditions, such as strong winds and heavy rains during typhoon weather, will all cause changes in the distribution laws of base station operation characteristics. Traditional detection models usually adopt fixed weighting coefficients, such as the weighting coefficient range of each dimension feature being 0.1 - 0.5; static thresholds, such as the abnormal coverage probability threshold being 0.7, and lack the ability to perceive and adaptively adjust to environmental variables. For example, during regional major events, the base station traffic load may exceed 70% of the designed capacity. At this time, if the conventional weighting coefficient allocation continues to be used, the contribution degree of the traffic load feature to anomaly detection may be underestimated; while during typhoon weather, environmental features such as wind speed and rainfall intensity are strongly correlated with signal penetration loss and link stability, and traditional models may not be able to accurately identify the coverage anomalies caused by environmental changes. Therefore, the existing technologies have significant limitations in aspects such as the real-time processing of a large amount of heterogeneous data and the adaptive update of models in a dynamic environment. Summary of the Invention

[0005] The present invention provides a method for detecting abnormal base station coverage based on big data analysis, which can accurately locate the abnormal base station coverage area, is applicable to the real-time monitoring and abnormal identification of the base station coverage status in communication network maintenance, improves the network optimization efficiency and service quality, and can solve the problems that traditional methods for detecting abnormal base station coverage cannot effectively integrate multi-dimensional operation data, have insufficient real-time performance, and low detection accuracy.

[0006] To achieve these and other advantages of the present invention, the present invention provides a method for detecting abnormal base station coverage based on big data analysis, including the following steps: S1. Collect the operation data of all base stations in the target area within a continuous time period. The operation data includes: received signal strength indication value, reference signal received quality value, handover success rate value, and traffic load value. The time period is 1 - 30 days, and the collection time interval is 10 - 60 minutes; S2. Calculate the coverage characteristic value of each base station according to the operation data. The coverage characteristic value is linearly weighted by the 10th percentile of the received signal strength indication value, the coefficient of variation of the reference signal received quality value, the 24-hour moving average of the handover success rate value, and the peak-to-valley difference of the traffic load value; S3. Input the coverage characteristic value into an isolated forest model trained with historical data, and output the coverage abnormal probability value of each base station; S4. When the coverage abnormal probability value is within the range interval of the probability threshold, mark the position of the corresponding base station on the electronic map as the abnormal coverage area, where the range interval of the probability threshold is greater than 0.7 and less than or equal to 0.9.

[0007] Preferably, in step S2, before calculating the coverage characteristic value, the operation data needs to be dimensionally reduced. The dimensional reduction processing specifically includes: S210. Use the principal component analysis method to compress the received signal strength indication value, the reference signal received quality value, the handover success rate value, and the traffic load value into a 2 - 4 dimensional feature space; S220. Adopt the piecewise aggregate approximation algorithm to compress the time series data to 10% - 20% of the original length; In step S2, the weighting coefficients of the linear weighting are determined by the following method: Establish a dynamic adjustment model for weighting coefficients, and the input parameters include base station type, geographical environment, and time period; Use the adaptive weighting coefficient algorithm, with the historical coverage abnormal detection accuracy as the objective function, and periodically optimize the weighting coefficients of each feature; Each feature weighting coefficient satisfies the following range and the sum is 1.0: the weighting coefficient of the 10th percentile of the received signal strength indication value is 0.3 - 0.5, the weighting coefficient of the coefficient of variation of the reference signal reception quality value is 0.2 - 0.3, the weighting coefficient of the 24-hour moving average value of the handover success rate is 0.1 - 0.2, and the weighting coefficient of the peak-to-valley difference of the traffic load value is 0.1 - 0.2.

[0008] Preferably, in step S3, it further includes updating the isolation forest model, and the updating method of the isolation forest model includes: S310. Periodically receive incremental operation data; S320. When the cumulative amount of incremental operation data reaches 5% - 15% of the original training data volume, trigger the subtree update operation; S330. The subtree update operation includes: selecting the subtrees to be updated with a depth of 3 - 5 in the current isolation forest model, generating replacement subtrees based on the incremental operation data, and atomically replacing the subtrees to be updated with the replacement subtrees. Specifically: calculating the coverage difference degree of the subtrees for the incremental operation data; selecting the top K subtrees with the highest coverage difference degree, where K is 10% - 30% of the total number of current subtrees. Among them, the calculation method of the coverage difference degree is: , is an indicator function, which outputs 1 when the condition is satisfied and 0 otherwise. T is the anomaly judgment threshold, and D i is the coverage difference degree of the i-th subtree, n is the number of incremental operation data samples, is the anomaly score prediction value of the i-th subtree for the incremental operation data sample , is the incremental operation data sample 's actual anomaly label, 1 for anomaly and 0 for normal; The atomic replacement of the replacement subtrees includes: creating a replacement subtree version identifier; pointing the pointer of the subtree to be updated to the replacement subtree through an atomic transaction.

[0009] Preferably, in step S4, after marking the coverage anomaly area, perform a positioning compensation step: Retrieve the user measurement report data within a range of 300 - 500 meters from the abnormal base station radius; When the proportion of sampling points where the received signal strength indication value in the user measurement report is continuously lower than -110 dBm exceeds 20% and does not exceed 40%, add the corresponding geographical coordinates to the boundary of the anomaly area; After step S4, it further includes: S5. Perform spatial clustering analysis and spatial clustering verification on the marked coverage abnormal base stations, specifically: S510. Extract the geographical coordinates of the abnormal base stations and construct a spatial feature matrix, including longitude and latitude, base station type, and coverage feature values; S520. Use the DBSCAN algorithm to perform spatial clustering on abnormal base stations, and set the parameters as follows: the neighborhood radius ε, with a value range of 500 - 1500 meters; the minimum number of samples MinPts, taking 3 - 5 base stations; S530. When there is a cluster containing ≥ MinPts base stations in the clustering result, it is determined as continuous regional coverage attenuation. The specific verification rules are as follows: a. Calculate the average value of the coverage anomaly probability values of all base stations within the cluster , if ≥ 0.7, and the spatial distribution of the base stations within the cluster is continuous and sheet-like, then trigger a regional-level anomaly alarm; output the spatial range of the clustering cluster, and mark it as a continuous coverage attenuation area on the electronic map; S540. For isolated abnormal base stations that do not form an effective cluster, retain the original single-point anomaly mark.

[0010] Preferably, in step S4, it further includes setting a dynamic threshold adjustment method for the coverage anomaly probability value, specifically including: Statistically calculate the standard deviation of the coverage anomaly probability values of all base stations in the entire region within the previous 24 hours; When the standard deviation exceeds 0.15 and does not exceed 0.25, increase the probability threshold by 0.05 - 0.1; When the standard deviation is lower than 0.05, decrease the probability threshold by 0.03 - 0.08.

[0011] Preferably, when collecting operation data, for 5G NR base stations, increase the collection of millimeter wave band characteristics, and obtain the beamforming failure rate value and the millimeter wave signal penetration loss compensation value; When calculating the coverage feature value, increase the millimeter wave feature weighting coefficient factor, and the weighting coefficient factor is 0.2 - 0.4.

[0012] Preferably, the processing of the user measurement report data includes: When the abnormal base station is a 4G / 5G dual-mode base station, perform multi-mode data verification: Compare the difference between the 4G reference signal received power value and the 5G synchronization signal reference signal received power value; If the difference exceeds 8 dB and does not exceed 12 dB and the duration ratio exceeds 30% and does not exceed 50%, then trigger neighbor base station collaborative detection. The neighbor base station collaborative detection includes: a. Send a collaborative request instruction to 1 - 3 co-frequency base stations adjacent to the target base station; b. Obtain the beam scanning measurement data of the neighbor base stations. The beam scanning measurement data includes: the user equipment reference signal received power difference and the signal delay spread value. Among them, the user equipment reference signal received power difference is the difference between the reference signal received powers measured by the same user equipment at the target base station and the neighbor base station. The calculation formula is: , RSRP target is the reference signal received power of the target base station, RSRP neighbor is the reference signal received power measured by the neighboring base station; c. When the following conditions are met, it is confirmed that the coverage anomaly is effective: the difference in the reference signal received power of the user equipment between the target base station and the neighboring base station is greater than 6 dB and does not exceed 10 dB; and the proportion of sampling points where the signal delay spread value is greater than 100 nanoseconds and does not exceed 150 nanoseconds exceeds 25% and does not exceed 40%.

[0013] Preferably, the training of the isolation forest model includes: The historical data includes a typhoon weather scenario dataset, and meteorological feature injection is performed during training: Taking the wind speed value of 12 - 20 m / s and the rainfall intensity value of 30 - 50 mm / h as environmental feature dimensions; Adding a meteorological feature judgment branch at the model splitting node.

[0014] Preferably, the dynamic threshold adjustment method further includes: Enabling a load compensation algorithm during regional major events: When the base station traffic load value exceeds 70% and does not exceed 90% of the design capacity, generating a temporary coverage attenuation coefficient of 0.6 - 0.8; Multiplying the coverage anomaly probability value by the attenuation coefficient and then comparing it with the probability threshold.

[0015] Preferably, marking the coverage anomaly area includes: Performing spatial topology verification on the marking result: Detecting the wireless backhaul link quality value between the abnormal base station and the adjacent base station; If the bit error rate of the backhaul link exceeds 10 -4 and does not exceed 10 -3 and the transmission delay increases by 30 - 50 ms, then freeze the abnormal marking of the base station.

[0016] The present invention has at least the following beneficial effects: First, through multi - dimensional operation data fusion and isolation forest model detection, breaking through the limitations of traditional single indicators, realizing multi - feature collaborative analysis of base station coverage anomalies. Continuous time - period data collection combined with probability threshold determination can capture long - term trends and short - term fluctuations, improving the comprehensiveness and accuracy of anomaly recognition, providing a quantitative basis for base station fault location, avoiding the blindness of manual inspections, and reducing operation and maintenance costs.

[0017] Second, the combination of principal component analysis and piecewise aggregate approximation algorithm effectively reduces the computational complexity of high-dimensional time-series data, compresses the data scale while retaining key features, improves processing efficiency, and meets the requirements of real-time detection. The dynamic weighted coefficient model adaptively adjusts the weighted coefficients according to the base station type, environment, and time period, so that the feature importance in different scenarios matches the actual abnormal contribution degree, avoids detection bias caused by fixed weighted coefficients, and improves the generalization ability of the model.

[0018] Third, the sub-tree update mechanism based on incremental operation data can optimize local parameters of the model without full retraining, shorten the update cycle, and ensure that the model can respond to network state changes in a timely manner. By screening the sub-trees to be updated through the coverage difference degree and atomically replacing them, the consumption of computing resources is reduced. At the same time, the interference of old data that may be introduced by full update is avoided, and the stability of the detection accuracy of the model is maintained, which is especially suitable for scenarios with dynamic traffic or environmental changes. The positioning compensation step corrects the abnormal boundary by combining the actual measured data of users to enhance the accuracy of single-point detection; spatial clustering analysis identifies continuous regional attenuation through the DBSCAN algorithm, distinguishes single-point faults from regional problems, provides a hierarchical disposal strategy for operation and maintenance, precisely repairs isolated anomalies, and initiates joint optimization for regional attenuation to improve the efficiency of resource allocation and avoid the increase in the overall maintenance cost caused by local misjudgment.

[0019] Fourth, the dynamic threshold adjustment mechanism adaptively corrects the threshold according to the fluctuation of the abnormal probability distribution in the entire region, tightens the detection standard to reduce false alarms when the network state is stable, and relaxes the threshold to avoid missed alarms when the fluctuation intensifies. The global abnormal dispersion degree is quantified through the standard deviation index, so that the threshold matches the real-time network characteristics, improves the flexibility and scenario adaptability of the detection strategy, and reduces the misjudgment risk in different time periods. The acquisition and weighting of the millimeter wave band characteristics of 5G NR base stations fill the dimension gap in high-frequency communication detection. The introduction of the beamforming failure rate and the millimeter wave signal penetration loss compensation value can accurately identify specific problems in millimeter wave signal propagation, such as occlusion and penetration loss, and improve the proportion of high-frequency anomalies in the eigenvalue by combining the exclusive weighted coefficient factor, ensuring the comprehensiveness of the coverage quality assessment of 5G new base stations and promoting the unified operation and maintenance of heterogeneous networks.

[0020] Fifthly, a multi-system data verification and neighbor cell collaborative detection mechanism excludes interference factors outside the base station through cross-system signal difference analysis and adjacent base station linkage measurement, achieving precise positioning of the abnormal source. It avoids false alarms caused by misjudgment of single-system data or neighbor cell interference, and improves the reliability of anomaly confirmation. The meteorological feature injection mechanism incorporates environmental variables such as wind speed and rainfall intensity into model training, enabling the Isolation Forest model to have the ability to identify meteorologically sensitive anomalies. In extreme weather such as typhoons, the model can predict in advance the coverage risks caused by environmental factors through the meteorological feature branches of split nodes, assisting maintenance personnel in formulating preventive maintenance strategies and reducing the impact of natural disasters on network services. The load compensation algorithm corrects probability values for high-traffic load scenarios during major events, avoiding misjudging fluctuations in base station performance caused by business peaks as coverage anomalies. Spatial topology verification excludes false anomaly markers caused by transmission layer failures, such as increased bit error rate and increased latency, by detecting the quality of the backhaul link, and avoids misjudging non-coverage problems as base station coverage failures. Freezing the markers of abnormal base stations with frozen links can guide maintenance personnel to prioritize troubleshooting of the transmission layer, improve the accuracy of fault classification, and ensure the efficiency of network maintenance.

[0021] Other advantages, objectives, and features of the present invention will be partially reflected by the following description and partially understood by those skilled in the art through the research and practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is a schematic flowchart of the base station coverage anomaly detection method based on big data analysis of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0023] The following further details the present invention so that those skilled in the art can implement it with reference to the description in the specification.

[0024] It should be understood that terms such as "having", "including", and "comprising" used herein do not exclude the presence or addition of one or more other elements or their combinations.

[0025] As Figure 1 shown, the present invention provides a base station coverage anomaly detection method based on big data analysis, including the following steps: S1. Collect the operation data of all base stations in the target area within a continuous time period. The operation data includes: received signal strength indication value, reference signal received quality value, handover success rate value, and traffic load value. The time period is 1 - 30 days, and the collection time interval is 10 - 60 minutes.

[0026] S2. Calculate the coverage eigenvalue of each base station according to the operating data, where the coverage eigenvalue is linearly weighted by the 10th percentile of the received signal strength indication value, the coefficient of variation of the reference signal received quality value, the 24-hour moving average of the handover success rate value, and the peak-to-valley difference of the traffic load value.

[0027] S3. Input the coverage eigenvalue into the Isolation Forest model trained with historical data, and output the coverage anomaly probability value of each base station.

[0028] S4. When the coverage anomaly probability value is within the range of the probability threshold, mark the location of the corresponding base station as a coverage anomaly area on the electronic map, where the range of the probability threshold is greater than 0.7 and less than or equal to 0.9.

[0029] Among them, marking the coverage anomaly area includes: Perform spatial topology verification on the marking result: Detect the wireless backhaul link quality value between the abnormal base station and the adjacent base stations; If the bit error rate of the backhaul link exceeds 10 -4 and does not exceed 10 -3 and the transmission delay increases by 30 - 50 ms, then freeze the abnormal marking of this base station.

[0030] In the above embodiment, in the data collection link, a time period of 1 day, 7 days, or 30 days can be selected to collect the operating data of the base stations in the target area. The collection interval can be set to 10 minutes, 30 minutes, or 60 minutes. The collected operating data includes the received signal strength indication value, the reference signal received quality value, the handover success rate value, and the traffic load value. When calculating the coverage eigenvalue, the 10th percentile of the received signal strength indication value is selected, the coefficient of variation of the reference signal received quality value is used, the 24-hour moving average of the handover success rate value is taken, and the peak-to-valley difference of the traffic load value is used. Each index is linearly weighted and fused. The weighting coefficients can be dynamically adjusted according to the base station type, geographical environment, and time period. For example, the weighting coefficient of the traffic load value for urban base stations can be appropriately increased to 0.2, and the weighting coefficient of the received signal strength indication value for suburban base stations can be focused on to 0.5.

[0031] Data collection can be realized through the base station monitoring terminals deployed in the existing network. Such terminals can collect the operating data of each type of base station in real time and upload it to the core network data center through the transmission network. When calculating the coverage eigenvalue, data analysis software in the server cluster, such as the Pandas library of Python, can be used to perform statistical calculations on the time series data and achieve linear weighting through matrix operations. The original data of each index is stored in a distributed database such as HBase for subsequent model training and retrospective analysis.

[0032] When training the Isolation Forest model, it is first necessary to prepare the historical dataset. These data contain the operation data collected by the base stations in the target area in the past 1-6 months, including the Received Signal Strength Indicator (RSSI) value, the Reference Signal Received Quality (RSRQ) value, the handover success rate value, and the traffic load value. The data collection time interval is consistent with real-time detection and is set to 10-60 minutes to ensure that the model training data and the subsequent actual application data have the same temporal characteristics. The historical data needs to be cleaned and preprocessed to remove obvious outliers and missing data. At the same time, features with different dimensions are standardized so that each feature is in a similar numerical range. After the data preprocessing is completed, the coverage feature values of each base station are calculated according to the method described in S2, including the 10th percentile of the RSSI value, the coefficient of variation of the RSRQ value, the 24-hour moving average of the handover success rate value, and the peak-to-valley difference of the traffic load value. After these feature values are combined by linear weighting, the model input feature vector is formed. During the training process, the number of trees in the Isolation Forest is set to 100-200, and each tree is constructed by randomly selecting features and randomly dividing values. The depth of the tree is limited to an average of 8-15 layers to prevent overfitting. Unsupervised learning is used during training, and no manually labeled anomaly labels are required. The model identifies abnormal patterns by calculating the average path length of the samples in the forest. After the model training is completed, the validation set is used to evaluate its performance. The validation set contains known normal base station operation data and manually confirmed abnormal cases. By adjusting the anomaly score threshold of the Isolation Forest, the detection accuracy of the model on the validation set is optimized. The trained model parameters and structure are saved to the model library, including the feature selection rules and node division values of each tree. The model is retrained regularly using newly added historical data, and the update period is set to 1-3 months to maintain adaptability to network changes. During the model deployment phase, the trained model is integrated into the real-time detection system through the API interface, receives the input of the coverage feature values from the S2 step, and outputs the coverage anomaly probability value of each base station.

[0033] For model deployment, existing cloud computing platforms such as Alibaba Cloud Machine Learning API can be used to encapsulate the trained Isolation Forest model as an API interface, receive the input of the preprocessed coverage feature values, and output the coverage anomaly probability value in real time. For newly added base station operation data, it can be pushed to the model input end in real time through a message queue such as Kafka to realize the online calculation of the anomaly probability. When a typhoon warning is detected, the model can be triggered to preferentially call the training subset containing meteorological features to improve the detection sensitivity in bad weather.

[0034] When the coverage anomaly probability value is in the range greater than 0.7 and less than or equal to 0.9, the system automatically marks the corresponding base station location as an abnormal coverage area on an electronic map such as the Baidu Map API. After marking, spatial topology verification is performed, and the wireless backhaul link quality data of the abnormal base station and adjacent base stations is obtained through the Network Management System (NMS) to detect the backhaul link bit error rate, with the threshold set to 10 -4 -10 -3 and the transmission delay. If abnormal link quality is detected, the abnormal mark of the base station is frozen, and a trouble - shooting work order for the transport layer is generated.

[0035] The electronic map marking function can be implemented through front - end visualization components such as ECharts, which support different - colored icons to distinguish the anomaly levels. The detection of wireless backhaul link quality can call the performance statistics interface of existing network transmission devices such as Huawei OptiX RTN series microwave devices to obtain the bit error rate and delay data in real - time. The verification process is automatically triggered by the Business Process Management (BPM) system to ensure the accuracy of the abnormal mark and avoid misjudgment caused by transport layer faults.

[0036] This embodiment can comprehensively reflect the base station coverage status through multi - dimensional data collection and dynamic feature calculation; the spatial topology verification mechanism effectively distinguishes coverage anomalies from transport layer faults, reducing the ineffective investment of operation and maintenance resources. This method can provide an accurate positioning basis for base station coverage optimization, assist operation and maintenance personnel to quickly respond to network anomalies, improve the stability and service quality of the communication network, and solve the problem of difficult real - time identification of coverage anomalies in a massive heterogeneous data environment.

[0037] In one specific embodiment, in step S2, before calculating the coverage feature value, the operation data needs to be dimension - reduced, and the dimension - reduction process specifically includes: S210: Using the principal component analysis method to compress the received signal strength indication value, reference signal received quality value, handover success rate value, and traffic load value into a 2 - 4 - dimensional feature space; S220: Using the Piecewise Aggregate Approximation algorithm to compress the time - series data to 10% - 20% of the original length; In step S2, the weighting coefficients of the linear weighting are determined in the following manner: Establish a dynamic adjustment model for the weighting coefficients, with input parameters including base station type, geographical environment, and time period; Using the adaptive weighting coefficient algorithm, with the historical coverage anomaly detection accuracy as the objective function, periodically optimize the weighting coefficients of each feature; The weight coefficients of each feature satisfy the following ranges and the sum is 1.0: the weight coefficient of the 10th percentile of the received signal strength indication value is 0.3 - 0.5, the weight coefficient of the coefficient of variation of the reference signal received quality value is 0.2 - 0.3, the weight coefficient of the 24-hour moving average of the handover success rate value is 0.1 - 0.2, and the weight coefficient of the peak-to-valley difference of the traffic load value is 0.1 - 0.2.

[0038] In the above embodiments, the principal component analysis algorithm can be selected to reduce the dimension of the received signal strength indication value, the reference signal received quality value, the handover success rate value, and the traffic load value. The output dimension of the principal component analysis can be set to 2D, 3D, or 4D. The specific dimension is selected according to the deployment density of the base station. It is recommended to use 3 - 4D in high-density areas and 2 - 3D in low-density areas. The variance contribution rate of more than 90% of the original data is retained in the reduced-dimension feature vector.

[0039] The piecewise aggregate approximation algorithm can be selected to compress the time series data. The compression ratio can be set to 10%, 15%, or 20% of the original length. When the time window length is 60 minutes, 10% compression corresponds to 6-minute aggregation, 15% corresponds to 9-minute aggregation, and 20% corresponds to 12-minute aggregation. The aggregation method uses the arithmetic mean to reduce high-frequency noise interference.

[0040] The calculations of the principal component analysis and the piecewise aggregate approximation can be deployed on a distributed computing platform. The data input interface supports the Kafka or MQTT protocol, and the reduced-dimension results are output to an in-memory database for subsequent feature calculation calls, which can solve the problem of insufficient real-time performance caused by big data processing.

[0041] The input parameters of the weighted coefficient dynamic adjustment model can include the base station type, geographical environment, and time period. The base station types are divided into macro base stations and micro base stations, the geographical environment is divided into urban areas, suburban areas, and rural areas, and the time period is divided into peak hours (08:00 - 20:00) and off-peak hours (20:00 - 08:00). The model updates the weighted coefficients every 24 hours, and the historical detection accuracy data of the most recent 30 days is loaded as the objective function during the update.

[0042] The gradient descent algorithm can be selected to optimize the weighted coefficients. The value range of the weighted coefficient of the 10th percentile of the received signal strength indication value (RSSI) is 0.3, 0.4, or 0.5, the weighted coefficient of the coefficient of variation of the reference signal received quality value (RSRQ) is 0.2, 0.25, or 0.3, the weighted coefficient of the moving average of the handover success rate is 0.1, 0.15, or 0.2, and the weighted coefficient of the traffic peak-to-valley difference is 0.1, 0.15, or 0.2. During the optimization process, the sum of the weighted coefficients is constrained to be 1 to avoid overfitting.

[0043] The weighted coefficient adjustment module can be integrated into the policy engine of the base station network management system to read the base station operation status data in real time, adjust it dynamically, and write it into the configuration database. The adjustment result is pushed to the anomaly detection module through the API to ensure that the weighted coefficient matches the current network status.

[0044] In this embodiment, principal component analysis and piecewise aggregate approximation are used for dimensionality reduction to reduce the amount of data processing, improve the calculation efficiency, and retain key feature information at the same time. The dynamic weighted coefficient adjustment mechanism can adapt to the coverage anomaly detection requirements in different scenarios, improving the detection accuracy and adaptability. The overall solution can be integrated into the existing network management system without additional hardware investment, reducing the operation and maintenance complexity.

[0045] In one specific embodiment, in step S3, the isolation forest model is further updated, and the update method of the isolation forest model includes: S310. Periodically receive incremental operation data; S320. When the cumulative amount of incremental operation data reaches 5%-15% of the original training data volume, trigger the subtree update operation; S330. The subtree update operation includes: selecting the subtrees to be updated with a depth of 3-5 in the current isolation forest model, generating replacement subtrees based on the incremental operation data, and atomically replacing the subtrees to be updated with the replacement subtrees. Specifically: calculating the coverage difference degree of the subtree for the incremental operation data; selecting the top K subtrees with the highest coverage difference degree, where K is 10%-30% of the total number of current subtrees. Among them, the calculation method of the coverage difference degree is: , is an indicator function, which outputs 1 when the condition is satisfied and 0 otherwise. T is the anomaly judgment threshold, usually taking 0.5; D i is the coverage difference degree of the i-th subtree, n is the number of incremental operation data samples, is the predicted anomaly score value of the i-th subtree for the incremental operation data sample, and is the actual anomaly label of the incremental operation data sample, where the anomaly is 1 and the normal is 0; The atomic replacement of the replacement subtree includes: creating a replacement subtree version identifier; pointing the pointer of the subtree to be updated to the replacement subtree through an atomic transaction.

[0046] In the above embodiment, the incremental operation data reception period can be configured as 1 hour, 6 hours or 24 hours, and the specific period is adjusted according to the base station data generation frequency. The data interface can select the Kafka message queue or the distributed file storage system to support real-time streaming data access.

[0047] When the cumulative incremental operation data volume reaches 5%, 10% or 15% of the original training data volume, the sub-tree update operation is triggered. The original training data volume can be set to 10,000 to 100,000 samples, and the update is triggered when the incremental operation data reaches 500 to 1,500. The threshold judgment module can be deployed on the edge computing node to statistically calculate the scale of the incremental operation data in real time and trigger the update signal.

[0048] The incremental operation data can be stored in the time series database, and the data format includes time stamps, base station IDs, operation metrics, and anomaly labels. In the data preprocessing stage, linear interpolation is used for missing values to ensure the integrity of the incremental operation data input to the model.

[0049] The sub-trees with depths of 3, 4, or 5 in the current isolation forest model can be selected as candidates for update. The coverage difference threshold can be set to 0.25, 0.3, or 0.35, and the sub-trees with a coverage difference higher than the threshold enter the update list. The coverage difference reflects the prediction deviation of the sub-tree for the incremental data. The larger the value, the worse the adaptability of the sub-tree to the new data. The sub-tree with the highest coverage difference is preferentially selected for update.

[0050] K sub-trees with the highest difference can be selected for replacement according to 10%, 20%, or 30% of the total number of current sub-trees. The replacement sub-trees are retrained based on the incremental operation data, and the depth constraint of the original tree is retained during training. The atomic replacement process can create a version identifier, such as UUID, and ensure the atomicity of pointer switching through database transactions to avoid concurrent access conflicts.

[0051] The sub-tree update module can be deployed on the GPU server to accelerate the training process using CUDA (Compute Unified Device Architecture). After the replaced sub-tree is verified to be correct, it is synchronized to the online inference service to ensure seamless switching of model updates.

[0052] It should be noted that the object of action of the anomaly judgment threshold T is the prediction result of the isolation forest sub-tree for a single sample, and its purpose is to calculate the sub-tree coverage difference and evaluate the model performance, which is mainly used in the model update stage; the probability threshold is the overall output probability of the isolation forest model for the base station, which is used to judge whether the base station is marked as abnormal, such as greater than 0.7 and less than or equal to 0.9 in S4.

[0053] This embodiment reduces the computational overhead of full model retraining through the incremental operation data trigger mechanism and improves the update efficiency. The sub-tree difference screening ensures that the model iteration is optimized for the part with performance degradation. The atomic replacement guarantees service continuity and avoids interruption of services during the update process. The overall solution can adapt to the real-time anomaly detection requirements of large-scale base station groups.

[0054] In one specific embodiment, the training of the isolation forest model includes: The historical data includes a typhoon weather scenario dataset, and meteorological feature injection is performed during training: The wind speed values of 12 - 20 m / s and the rainfall intensity values of 30 - 50 mm / h are used as environmental feature dimensions; A meteorological feature judgment branch is added at the model splitting node.

[0055] In the above embodiment, when training the isolation forest model, the historical data also includes a typhoon weather scenario dataset. The typhoon weather scenario feature data includes wind speed values and rainfall intensity values. The wind speed values need to cover the typical wind speed intervals in typhoon weather, such as key thresholds like 12 m / s, 15 m / s, 20 m / s, etc., corresponding to the wind force intensities of tropical storms, severe tropical storms, and typhoons. Such data is used to characterize the interference intensity of typhoons on wireless signal propagation, such as antenna offsets caused by fallen trees and equipment vibrations. Rainfall intensity values: include rainfall levels such as 30 mm / h, 40 mm / h, 50 mm / h, etc., reflecting the impact of heavy precipitation accompanied by typhoons on signal penetration loss, such as the attenuation effect of rain on the millimeter wave band. The wind speed values and rainfall intensity values are used as environmental feature dimensions, and a meteorological feature judgment branch is added at the model splitting node, enabling the model to identify the impact of wind speed and rainfall intensity on base station coverage. During the training process, the cross - validation method can be used to optimize the model parameters, such as setting the number of sub - trees to 100 and the maximum depth of a single tree to 20 layers to improve the model's fitting ability for high - dimensional features. This embodiment can solve the problem of inaccurate traditional models under extreme weather conditions. It should be noted that the geographical environment refers to the long - term stable physical attributes of the base station deployment location, such as terrain type, building density, or vegetation coverage rate, etc., and does not include meteorological features. Meteorological features refer to the real - time collected weather state parameters, such as wind speed values, rainfall intensity, etc.

[0056] In one specific embodiment, in step S4, after marking the coverage abnormal area, a positioning compensation step is performed: Retrieve the user measurement report data within a range of 300 - 500 meters from the abnormal base station; When the proportion of sampling points with received signal strength indication values continuously lower than - 110 dBm in the user measurement report exceeds 20% and does not exceed 40%, add the corresponding geographical coordinates to the abnormal area boundary; After step S4, it further includes: S5. Perform spatial clustering analysis and spatial clustering verification on the marked coverage abnormal base stations, specifically: S510. Extract the geographical coordinates of the abnormal base stations and construct a spatial feature matrix, including longitude and latitude, base station type, and coverage feature values; S520. Use the DBSCAN algorithm to perform spatial clustering on the abnormal base stations, and set the parameters as follows: neighborhood radius ε, with a value range of 500 - 1500 meters; minimum number of samples MinPts, taking 3 - 5 base stations; S530: When there is a cluster containing ≥MinPts base stations in the clustering result, it is determined to be continuous area coverage attenuation. The specific verification rules are as follows: a. Calculate the average coverage anomaly probability value of all base stations in the cluster ,like 0.7, and the spatial distribution of base stations in the cluster is continuous and sheet-like, then a regional abnormal alarm is triggered; the spatial range of the cluster is output and marked as a continuous coverage attenuation area on the electronic map; S540: For isolated abnormal base stations that have not formed a valid cluster, retain the original single point abnormal mark.

[0057] In the above implementation, the user measurement report data within 300-500 meters of the abnormal base station is retrieved through positioning compensation, and the area where the received signal strength is continuously lower than -110dBm and the sampling points account for 20%-40% is selected to add the abnormal boundary, which can significantly improve the accuracy and efficiency of network optimization. For continuous lower than -110dBm, it can be understood as N consecutive sampling points with an interval of 10 minutes, N=12 corresponds to 2 hours. This method dynamically corrects the coverage abnormal boundary based on the actual user-side signal measurement data, avoiding the deviation of the traditional reliance on the base station theoretical model, and is particularly suitable for identifying transition areas where the signal is attenuated but not completely interrupted, and realizing early warning of weak signal blind spots. By setting a 20%-40% share threshold, temporary interference and continuous coverage problems can be effectively distinguished, while reducing misjudgment and ensuring that edge anomalies are not missed, providing operators with a high-reliability optimization target area. Combined with automated user measurement report analysis, small coverage holes in high-density urban environments can be quickly located, greatly reducing operation and maintenance costs compared to manual road testing. In addition, the identification of areas where signal strength is at a critical deterioration value (such as a percentage close to 40%) can also support preventive maintenance, proactively trigger antenna parameter adjustments or small base station deployment, thereby eliminating potential faults before users complain and improving the overall network service quality.

[0058] The construction of spatial feature matrix can extract the basic information of abnormal base stations from the base station management system. The longitude and latitude coordinates can be accurate to 6 decimal places, and the base station types can be divided into macro base stations, micro base stations and indoor distribution systems. The coverage feature values include indicators such as received signal strength and reference signal quality, and the value range is between 0 and 1. You can use the existing spatial database to store these feature data to support fast query and matrix construction.

[0059] The process of constructing the feature matrix will perform data standardization to ensure that features of different dimensions are comparable. An automatic update mechanism can be configured to update the feature matrix in real time when new abnormal base stations are added. Matrix data can be stored in an in-memory database to increase the reading speed during cluster analysis. After the construction is completed, a data verification report will be generated to ensure the integrity and accuracy of the matrix.

[0060] DBSCAN clustering analysis can be implemented using off-the-shelf spatial analysis algorithm libraries. The neighborhood radius ε can be set to 500 meters, 800 meters, or 1500 meters, and the specific value is adjusted according to the base station density. The minimum number of samples MinPts can be set to 3, 4, or 5 base stations to control the clustering sensitivity. When the algorithm is executed, the geographical distribution characteristics and signal coverage range of the base stations will be considered.

[0061] The clustering process will run on a distributed computing platform, supporting the rapid processing of large-scale base station data. An exception handling mechanism can be configured to automatically optimize parameters or switch algorithms when the clustering time exceeds the threshold. The clustering results will generate a visualization report showing the spatial distribution and clustering of the base stations. The entire process is usually completed within minutes, meeting the real-time requirements.

[0062] During the clustering result verification stage, the average coverage anomaly probability of the base stations within each cluster will be calculated, and the threshold is fixed at 0.7. The spatial continuity judgment will analyze the density and shape characteristics of the base station distribution to exclude misclustering of discrete distributions. Off-the-shelf spatial analysis tools can be selected for continuity judgment, and the result will be output in the form of a boolean value. The verification process will record detailed logs to support subsequent auditing and analysis.

[0063] When the cluster meets both the probability mean and continuity conditions, the system will generate a regional-level anomaly alert. The alert information includes key indicators such as the center coordinates, influence radius, and number of base stations of the cluster. A multi-level alert mechanism can be configured to trigger different levels of notifications according to the scale of the cluster. The alert information will be pushed to the operation and maintenance management platform in real time and highlighted on the electronic map.

[0064] The abnormal marking process will uniformly mark the clusters that pass the verification and display them in the form of a polygon area on the electronic map. The marking color can be set to orange or red, different from the single-point abnormal marking. Off-the-shelf GIS system interfaces can be selected to implement the area marking function, supporting multi-layer overlay display. The marking information will be synchronized to the operation and maintenance knowledge base to form a historical record.

[0065] For isolated base stations that do not form effective clusters, their original single-point abnormal markings will be retained. The system will periodically re-evaluate these isolated points and trigger reclustering when new abnormal base stations appear in the vicinity. An automatic cleaning mechanism can be configured to remove the corresponding markings when the isolation point anomaly is resolved. All marking changes will generate operation logs to ensure the entire process is traceable.

[0066] Through spatial clustering analysis, it is possible to effectively distinguish single-point anomalies and regional coverage problems, improving the accuracy of fault location. The clustering method based on the DBSCAN algorithm can adapt to different densities of base station distributions and identify abnormal regions with spatial continuity. The systematic verification rules ensure the reliability of alarm information and reduce false alarms. The overall solution can be integrated into the existing network management system to improve network operation and maintenance efficiency.

[0067] In one specific embodiment, in step S4, it further includes setting a dynamic threshold adjustment method for the coverage anomaly probability value, specifically including: Statistical standard deviation of the coverage anomaly probability values of all-region base stations within 24 hours before the current moment; When the standard deviation exceeds 0.15 and does not exceed 0.25, increase the probability threshold by 0.05 - 0.1; When the standard deviation is lower than 0.05, decrease the probability threshold by 0.03 - 0.08.

[0068] Among them, the dynamic threshold adjustment method further includes: Enable the load compensation algorithm during regional major events: When the base station traffic load value exceeds 70% and does not exceed 90% of the design capacity, generate a temporary coverage attenuation coefficient of 0.6 - 0.8; Multiply the coverage anomaly probability value by the attenuation coefficient and then compare it with the probability threshold.

[0069] In the above embodiment, the dynamic threshold adjustment mechanism can be implemented by the statistical analysis module deployed on the network management server. The standard deviation calculation window is fixed at 24 hours, and the calculation period can be set to 1 hour or 4 hours. The standard deviation thresholds are set to 0.15 and 0.25 as the upward adjustment interval, and 0.05 as the downward adjustment threshold. A ready-made time series database can be selected to store probability value data, supporting rolling window calculation.

[0070] When the standard deviation is in the range of 0.15 - 0.25, the upward adjustment amplitude of the probability threshold can be set to 0.05, 0.08, or 0.1. When the standard deviation is less than 0.05, the downward adjustment amplitude of the probability threshold can be set to 0.03, 0.05, or 0.08. The adjusted probability threshold will take effect immediately, and the version information will be recorded. Probability threshold change alarms can be configured to notify the operation and maintenance personnel for confirmation when the adjustment amplitude exceeds 0.1. It should be noted that the initial probability threshold is determined based on the historical data validation set. Since the probability threshold is a range interval value, when it is increased, the entire threshold interval is adjusted uniformly. For example, when it is increased by 0.05, the new probability threshold is greater than 0.75 and less than or equal to 0.95.

[0071] The threshold adjustment module can be deployed on the policy management server and interact with the anomaly detection engine in real time. The adjustment process takes into account the historical threshold change trend to avoid frequent fluctuations. The system retains the threshold records for the most recent 30 days to support retrospective analysis. Each adjustment generates an operation log recording the adjustment time, amplitude, and triggering conditions.

[0072] During major events, load compensation can be triggered through the event management platform. The traffic load monitoring period can be set to 5 minutes or 15 minutes, and the design capacity data is read from the base station configuration library. The load thresholds are set at 70% and 90%, and compensation is automatically enabled when the load is within this range. The load monitoring module of an off-the-shelf network performance management system can be selected.

[0073] The temporary coverage attenuation coefficient can be set to 0.6, 0.7, or 0.8, and the specific value is adjusted according to the event level. When the coefficient is applied, it will correct the coverage anomaly probability value in real time, and the corrected value is then compared with the current probability threshold. The validity period of the coefficient can be configured, and the original calculation method will be automatically restored after the event ends. The compensation process marks the data source for subsequent analysis.

[0074] The load compensation module can be deployed on the edge computing node of the core network to ensure low-latency processing. The system records all operations during the compensation period, including the coefficient values, the number of base stations affected, and the alarm changes. The compensation effect evaluation mechanism can be configured to generate a special report after the event ends. The entire process requires no manual intervention and realizes automated processing. It can solve the problem of a sharp increase in coverage false alarms in high-density user scenarios.

[0075] Dynamic threshold adjustment can adapt to the normal fluctuations of the network state and reduce false alarms caused by environmental factors. The adjustment strategy based on the standard deviation can maintain the stability of the alarm system and avoid overly sensitive or insensitive threshold settings. The load compensation mechanism can effectively distinguish real coverage problems and temporary capacity overloads during major events, improving the accuracy of anomaly determination. The overall solution can flexibly meet the requirements of different scenarios and enhance the intelligent level of network operation and maintenance.

[0076] In one specific embodiment, when collecting operation data, for 5G NR base stations, the collection of millimeter-wave band characteristics is increased to obtain the beamforming failure rate value and the millimeter-wave signal penetration loss compensation value; When calculating the coverage characteristic value, a millimeter-wave characteristic weighting coefficient factor is added, and the weighting coefficient factor is 0.2 - 0.4.

[0077] In the above embodiments, the millimeter-wave band feature collection can obtain beam management data through the northbound interface of the 5G base station. The beamforming failure rate can be statistically calculated per minute or every five minutes, and the statistical period can be set to 15 minutes or 30 minutes. The millimeter-wave signal penetration loss compensation value can be read from the measurement module of the base station radio frequency unit, and the compensation value range is usually between 10 - 30 dB. A network management protocol that supports the 3GPP standard can be selected to collect this data.

[0078] The acquisition system can be deployed at the edge computing node and communicate with the base station through the standard NM interface. The data acquisition frequency can be set to be synchronized with the conventional metrics to ensure time alignment. During the acquisition process, outliers such as sudden instantaneous failure rate peaks will be filtered. The acquired data will be stored in the time series database, and the tags include the base station ID, timestamp, and frequency band information. The system will regularly verify the data integrity to ensure the reliability of the acquisition process.

[0079] The setting of the millimeter-wave feature weighting coefficient can be implemented in the feature calculation engine. The weighting coefficient factor can be set to 0.2, 0.3, or 0.4, and the specific value is adjusted according to the network deployment environment. The weighting coefficient adjustment period can be set to 24 hours or triggered on demand, and the historical optimization effect will be referred to during the adjustment. Weighting coefficient constraint conditions can be configured to ensure that the millimeter-wave features do not overly affect the overall evaluation.

[0080] During the application process of the weighting coefficient, the millimeter-wave feature values will be normalized to make their dimensions consistent with other features. During the calculation, it will first be multiplied by the weighting coefficient factor and then added to other weighted features. A matrix operation library can be selected to accelerate the calculation process and support real-time update of the feature values. The system will record the parameters and effects of each weighting coefficient adjustment to form an optimization knowledge base. The abnormal probability value of the calculation result will be synchronized to the decision engine.

[0081] Increasing the millimeter-wave feature collection can more comprehensively reflect the coverage characteristics of 5G base stations, especially the propagation characteristics of high-frequency bands. The dynamic weighting coefficient setting can balance the influence of different frequency band features and improve the accuracy of anomaly detection. The overall solution is compatible with the management system of the existing 4G / 5G hybrid network, and the function upgrade can be achieved without modifying the hardware equipment. Through the fine-grained millimeter-wave feature analysis, the high-frequency band coverage problems can be more accurately identified. It can solve the coverage misjudgment problem caused by the environmental sensitivity of millimeter-wave base stations.

[0082] In one specific embodiment, the processing of the user measurement report data includes: When the abnormal base station is a 4G / 5G dual-mode base station, multi-mode data verification is performed: Compare the difference between the 4G reference signal received power value and the 5G synchronization signal reference signal received power value; If the difference exceeds 8 dB and does not exceed 12 dB and the duration ratio exceeds 30% and does not exceed 50%, then trigger the cooperative detection of neighboring base stations. The cooperative detection of neighboring base stations includes: a. Sending a cooperative request instruction to 1 - 3 co - frequency base stations adjacent to the target base station; b. Obtaining the beam scanning measurement data of the neighboring base stations. The beam scanning measurement data includes: the difference in user equipment reference signal received power and the signal delay spread value. Among them, the difference in user equipment reference signal received power is the difference in the reference signal received power measured by the same user equipment at the target base station and the neighboring base station. The calculation formula is: , RSRP target is the reference signal received power of the target base station, and RSRP neighbor is the reference signal received power measured by the neighboring base station; c. When the following conditions are met, confirm that the coverage anomaly is valid: the difference in user equipment reference signal received power between the target base station and the neighboring base station is greater than 6 dB and does not exceed 10 dB; and the proportion of sampling points where the signal delay spread value is greater than 100 nanoseconds and does not exceed 150 nanoseconds exceeds 25% and does not exceed 40%.

[0083] In the above - mentioned implementation, multi - mode data verification can be achieved through the measurement report processing module of the dual - mode base station. The 4G reference signal received power value can be extracted from the RRC measurement report, and the 5G synchronization signal reference signal received power value can be obtained from the NR measurement report. The threshold for difference comparison is set to 8 dB and 12 dB, and the threshold for the duration ratio is set to 30% and 50%. The data comparison function of an off - the - shelf multi - mode base station management system can be selected.

[0084] The difference calculation period can be set to 1 second or 5 seconds, and the duration statistics window can be set to 5 minutes or 10 minutes. The system will record the result of each comparison to form time - series data for trend analysis. When the conditions are met, a verification report will be automatically generated and the subsequent process will be triggered. The verification process will exclude short - term differences caused by normal scenarios such as device handover.

[0085] The cooperative detection of neighboring base stations can send the cooperative request instruction through the X2 interface or the Xn interface. The request instruction can be configured to include the target base station ID, the anomaly type, and the request data type. The number of neighboring base stations can be set to 1, 2, or 3, and the selection strategy is based on the base station topology relation table. A base - station - to - base - station communication protocol stack that supports the 3GPP standard can be selected.

[0086] After receiving the cooperative request, the neighboring base station will start the specified measurement task. The calculation of the difference in user equipment reference signal received power uses the formula , where the RSRP value is accurate to 0.1 dB. The signal delay spread value can be obtained from the channel estimation module, and the measurement accuracy is 1 nanosecond. The measurement data is sent back to the requester through an encrypted channel.

[0087] In the abnormal validity confirmation conditions, the reference signal received power difference threshold is set to 6 dB and 10 dB. The signal delay spread value threshold is set to 100 nanoseconds and 150 nanoseconds, and the abnormal sampling point ratio threshold is set to 25% and 40%. An off-the-shelf wireless signal analysis algorithm library can be selected for condition judgment. The confirmation process will synthesize the measurement results of multiple neighboring cells and adopt a voting mechanism to improve accuracy.

[0088] After the validity confirmation, a final abnormal diagnosis report will be generated, including the abnormal type, the affected range, and the confidence level assessment. The report will be synchronized to the network optimization platform and the operation and maintenance management system. For the confirmed abnormality, the system will recommend corresponding optimization measures, such as antenna adjustment or parameter optimization. The entire process is usually completed within minutes, meeting the real-time requirements.

[0089] Multi-mode data verification can effectively identify the coverage difference problem between different modes of dual-mode base stations, avoiding the limitations of single-mode detection. Neighboring cell collaborative detection improves the accuracy of abnormal determination through cross-verification of multi-dimensional measurement data. The overall solution makes full use of the existing cooperation mechanism between base stations and can achieve refined coverage analysis without adding hardware devices. Through standardized interfaces and processes, it ensures that the solution can be deployed and applied on a large scale. It can solve the positioning distortion problem caused by measurement standard differences in multi-mode base stations.

[0090] The number of devices and the processing scale described here are used to simplify the description of the present invention. Applications, modifications, and variations of the present invention will be obvious to those skilled in the art.

[0091] Although the embodiments of the present invention have been disclosed as above, it is not limited to the applications listed in the specification and the embodiments. It can be fully applied to various fields suitable for the present invention. For those familiar with the field, additional modifications can be easily made. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to specific details.

Claims

1. A base station coverage anomaly detection method based on big data analysis, characterized in that, It includes the following steps: S1. Collect the operation data of all base stations in the target area within a continuous time period. The operation data includes: received signal strength indication value, reference signal received quality value, handover success rate value, and traffic load value. The time period is 1 - 30 days, and the collection time interval is 10 - 60 minutes; S2. Calculate the coverage characteristic value of each base station according to the operation data. The coverage characteristic value is linearly weighted by the 10th percentile of the received signal strength indication value, the coefficient of variation of the reference signal received quality value, the 24 - hour moving average of the handover success rate value, and the peak - valley difference of the traffic load value; S3. Input the coverage characteristic value into the isolated forest model trained by historical data, and output the coverage anomaly probability value of each base station; S4. When the coverage anomaly probability value is within the range interval of the probability threshold, mark the corresponding base station position as a coverage anomaly area on the electronic map. Among them, the range interval of the probability threshold is greater than 0.7 and less than or equal to 0.

9.

2. The method for detecting abnormal base station coverage based on big data analysis according to claim 1, wherein In step S2, before calculating the coverage characteristic value, the operation data needs to be dimension - reduced. The dimension - reduction process specifically includes: S210. Use the principal component analysis method to compress the received signal strength indication value, the reference signal received quality value, the handover success rate value, and the traffic load value into a 2 - 4 - dimensional feature space; S220. Adopt the piecewise aggregate approximation algorithm to compress the time - series data to 10% - 20% of the original length; In step S2, the weighting coefficients of the linear weighting are determined in the following way: Establish a dynamic adjustment model of the weighting coefficients. The input parameters include base station type, geographical environment, and time period; Use the adaptive weighting coefficient algorithm, with the historical coverage anomaly detection accuracy as the objective function, and periodically optimize the weighting coefficients of each feature; The weighting coefficients of each feature satisfy the following range and the sum is 1.0: the weighting coefficient of the 10th percentile of the received signal strength indication value is 0.3 - 0.5, the weighting coefficient of the coefficient of variation of the reference signal received quality value is 0.2 - 0.3, the weighting coefficient of the 24 - hour moving average of the handover success rate value is 0.1 - 0.2, and the weighting coefficient of the peak - valley difference of the traffic load value is 0.1 - 0.

2.

3. The method for detecting abnormal base station coverage based on big data analysis according to claim 1, wherein, In step S3, it also includes updating the isolated forest model. The update method of the isolated forest model includes: S310. Periodically receive incremental operation data; S320. When the cumulative incremental operation data volume reaches 5% - 15% of the original training data volume, trigger the subtree update operation; S330. The subtree update operation includes: selecting the subtree to be updated with a depth of 3 - 5 in the current isolation forest model, generating a replacement subtree based on the incremental operation data, and atomically replacing the subtree to be updated with the replacement subtree. Specifically: calculating the coverage difference degree of the subtree for the incremental operation data; selecting the top K subtrees with the highest coverage difference degree, where K is 10% - 30% of the total number of current subtrees. Among them, the calculation method of the coverage difference degree is: , is an indicator function that outputs 1 when the condition is satisfied and 0 otherwise. T is the anomaly judgment threshold, and D i is the coverage difference degree of the i-th subtree, and n is the number of incremental operation data samples. is the predicted anomaly score value of the i-th subtree for the incremental operation data sample . is the incremental operation data sample 's actual anomaly label, where anomaly is 1 and normal is 0; The replacement subtree atomic replacement includes: creating a replacement subtree version identifier; pointing the pointer of the subtree to be updated to the replacement subtree through an atomic transaction.

4. The method for detecting abnormal base station coverage based on big data analysis according to claim 1, wherein In step S4, after marking the coverage anomaly area, perform a positioning compensation step: Retrieve the user measurement report data within a range of 300 - 500 meters from the abnormal base station; When the proportion of sampling points with the received signal strength indication value continuously lower than - 110 dBm in the user measurement report exceeds 20% and does not exceed 40%, add the corresponding geographical coordinates to the boundary of the abnormal area; After step S4, it further includes: S5. Perform spatial clustering analysis and spatial clustering verification on the marked abnormal base stations with coverage anomalies. Specifically: S510. Extract the geographical coordinates of the abnormal base stations and construct a spatial feature matrix, including longitude and latitude, base station type, and coverage eigenvalue; S520. Use the DBSCAN algorithm to perform spatial clustering on the abnormal base stations, and set the parameters as follows: neighborhood radius ε, with a value range of 500 - 1500 meters; minimum number of samples MinPts, taking 3 - 5 base stations; S530. When there is a cluster containing ≥ MinPts base stations in the clustering result, it is determined as continuous area coverage attenuation. The specific verification rules are as follows: a. Calculate the average value of the coverage anomaly probability values of all base stations within the cluster. , if ≥ 0.7, and the spatial distribution of the base stations within the cluster is continuous and sheet-like, then trigger a regional-level anomaly alarm; output the spatial range of the clustering cluster and mark it as a continuous coverage attenuation area on the electronic map. S540. For the isolated abnormal base stations that do not form effective clusters, retain the original single - point abnormal marks.

5. The base station coverage anomaly detection method based on big data analysis according to claim 1, characterized in that In step S4, it also includes setting a dynamic threshold adjustment method for the coverage anomaly probability value. Specifically, it includes: Statistical standard deviation of the coverage anomaly probability values of all base stations in the whole region within 24 hours before the current moment; When the standard deviation exceeds 0.15 and does not exceed 0.25, increase the probability threshold by 0.05 - 0.1; When the standard deviation is lower than 0.05, decrease the probability threshold by 0.03 - 0.

08.

6. The method for detecting abnormal base station coverage based on big data analysis according to claim 1, wherein When collecting operation data, increase the collection of millimeter - wave band characteristics for 5G NR base stations to obtain the beamforming failure rate value and the millimeter - wave signal penetration loss compensation value; When calculating the coverage eigenvalue, add a millimeter - wave feature weighting coefficient factor, and the weighting coefficient factor is 0.2 - 0.

4.

7. The method for detecting abnormal base station coverage based on big data analysis according to claim 4, wherein The processing of the user measurement report data includes: When the abnormal base station is a 4G / 5G dual - mode base station, perform multi - mode data verification: Compare the difference between the 4G reference signal received power value and the 5G synchronization signal reference signal received power value; If the difference exceeds 8 dB and does not exceed 12 dB, and the duration ratio exceeds 30% and does not exceed 50%, then trigger the collaborative detection of neighboring base stations. The collaborative detection of neighboring base stations includes: a. Send a collaborative request instruction to 1-3 co-frequency base stations adjacent to the target base station; b. Obtain the beam scanning measurement data of the neighboring base stations. The beam scanning measurement data includes: the difference in the received power of the user equipment reference signal and the signal delay spread value. Among them, the difference in the received power of the user equipment reference signal is the difference in the received power of the reference signal measured by the same user equipment at the target base station and the neighboring base station. The calculation formula is: , RSRP target is the received power of the reference signal of the target base station, and RSRP neighbor is the received power of the reference signal measured by the neighboring base station; c. When the following conditions are met, confirm that the coverage anomaly is valid: the difference in the received power of the user equipment reference signal between the target base station and the neighboring base station is greater than 6 dB and does not exceed 10 dB; and the proportion of sampling points where the signal delay spread value is greater than 100 nanoseconds and does not exceed 150 nanoseconds exceeds 25% and does not exceed 40%.

8. The method for detecting abnormal base station coverage based on big data analysis according to claim 1, characterized in that The training of the isolation forest model includes: The historical data also contains a typhoon weather scenario dataset, and meteorological feature injection is performed during training: Take the wind speed value of 12 - 20 m / s and the rainfall intensity value of 30 - 50 mm / h as environmental feature dimensions; Add a meteorological feature judgment branch at the model splitting node.

9. The base station coverage anomaly detection method based on big data analysis according to claim 5, characterized in that, The dynamic threshold adjustment method also includes: Enable the load compensation algorithm during regional major events: When the base station traffic load value exceeds 70% and does not exceed 90% of the designed capacity, generate a temporary coverage attenuation coefficient of 0.6 - 0.8; Multiply the coverage anomaly probability value by the attenuation coefficient and then compare it with the probability threshold.

10. The method for detecting abnormal base station coverage based on big data analysis according to claim 1, wherein Marking the coverage anomaly area includes: Perform spatial topology verification on the marking result: Detect the wireless backhaul link quality value between the abnormal base station and adjacent base stations; If the backhaul link bit error rate exceeds 10 -4 and does not exceed 10 -3 and the transmission delay increases by 30 - 50 ms, then freeze the abnormal flag of this base station.

Images