Wireless network fault diagnosis system and method based on big data analysis

Abstract

The application discloses a wireless network fault diagnosis system and method based on big data analysis, and relates to the technical field of wireless operation and maintenance. The wireless network fault diagnosis system and method based on big data analysis comprises the following steps: S1, collecting data and performing preprocessing; S2, issuing statistical granularity uniformity according to the same caliber back calculation analysis result, enabling original count transparent transmission, triggering time source verification and clock synchronization review, and solidifying cross-level contrast caliber; S3, switching the area statistical perspective according to the anti-dilution fidelity analysis result, enabling user equipment measurement report tracking collection, generating a switching event fine-grained log, and establishing evidence binding; S4, performing count details reporting, contrast difference convergence, verification treatment, and directional retest up-drilling treatment, and fixing the statistical perspective. The application solves the problem that, due to the difference between the network levels in statistics and aggregation mechanism, the same abnormality is inconsistent at different levels, which easily leads to the local fault being covered by the upper index, so that the fault positioning accuracy and stability are insufficient.

Description

Technical Field

[0001] This invention relates to the field of wireless operation and maintenance technology, specifically to a wireless network fault diagnosis system and method based on big data analysis. Background Technology

[0002] With the large-scale deployment of 5G mobile communication systems and subsequent evolution technologies, mobile internet and IoT services continue to grow. The coverage, service carrying capacity, and interference environment of wireless access networks exhibit dynamic and complex characteristics. Network operation and maintenance are gradually shifting from relying primarily on manual inspections and experience-based judgment to online monitoring, intelligent analysis, and closed-loop handling based on network management systems and big data platforms. Existing networks are typically divided into multi-level management objects such as cells, base stations, regions, and core networks according to network elements and management domains. Operational data is continuously collected under systems such as performance management, alarm management, configuration management, and measurement management. Through statistical aggregation, indicator calculation, event correlation, and time-series analysis, operational status views are formed for different levels to support fault detection, impact assessment, location analysis, and dispatching. In engineering practice, to improve diagnostic efficiency and stability, the operation and maintenance system performs time synchronization and data cleaning on multi-source asynchronous reported data, establishes reference baselines for historical operating status, manages network configuration evolution in a versioned manner, and combines statistical analysis, probabilistic modeling and machine learning methods to identify and interpret abnormal patterns, thereby realizing automated diagnosis and auxiliary decision-making for problems such as performance degradation, link anomalies and service quality fluctuations in complex wireless scenarios.

[0003] For example, the invention patent with announcement number CN115379484B discloses a method for assessing the health of wireless networks applicable to industry field networks. This method extracts industry terminal-level indicator data, analyzes multi-dimensional indicators such as traffic, coverage, connection anomalies, and antenna feeder anomalies, and scores the related items of each dimension according to priority based on a preset indicator scoring table. After completing the matching of industry terminal serial numbers in the network management system, it calculates the comprehensive score of each dimension and classifies the potential fault levels of industry terminals through an indicator mapping model, and executes corresponding fault handling solutions according to different fault levels. At the same time, this patent also continuously iterates and optimizes the indicator mapping model by combining field verification results, so that the model output results gradually approach the actual fault situation, thereby realizing the assessment and management of the health status of wireless networks in industry field networks.

[0004] For example, the invention patent with announcement number CN110149652B discloses a fault detection method and device for a vehicle-mounted mobile station. This method is based on the big data generated during the operation of the vehicle-mounted mobile station, collecting wireless network measurement reports, handover signaling data, and interface data between the terminal and the train control system, and establishing the correlation between fault detection parameters and train parameters. By constructing different types of correlation datasets, it analyzes and calculates fault detection parameters such as uplink level, uplink quality, handover failure rate, and network registration failure rate, and determines whether the vehicle-mounted mobile station has a fault based on the calculation results. This scheme also combines time series analysis, statistical averaging, and trend judgment to detect the performance changes of the vehicle-mounted mobile station under different operating conditions, thereby realizing the fault identification and judgment of the operating status of the vehicle-mounted mobile station.

[0005] In existing wireless network fault diagnosis technologies, analysis and evaluation are typically based on preset indicator systems to judge the operational status of terminals, cells, or local objects, which has certain applicability under a single level or a single statistical perspective. However, wireless networks themselves have multi-level structural characteristics, and different levels differ in data collection, statistical granularity, and aggregation methods. Existing methods often lack unified constraints on cross-level statistical caliber differences, leading to inconsistent manifestations of the same anomaly at different levels. Furthermore, upper-level or regional-level aggregated statistics, while reflecting the overall situation, can easily dilute local anomalies, making it difficult to effectively identify local faults. In addition, the diagnostic results lack sufficient constraints on temporal continuity and location stability, easily fluctuating under different analysis windows or statistical perspectives, affecting the accuracy and stability of fault location.

[0006] Therefore, in order to address the above problems, there is an urgent need for a wireless network fault diagnosis system and method based on big data analysis. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this invention provides a wireless network fault diagnosis system and method based on big data analysis. This solves the problem in existing technologies where, due to differences in statistical and aggregation mechanisms between network layers, the same anomaly manifests inconsistently at different layers, easily leading to local faults being masked by upper-layer indicators, thus resulting in insufficient fault location accuracy and stability.

[0008] To achieve the above objectives, the present invention is implemented through the following technical solution: a wireless network fault diagnosis method based on big data analysis, comprising: S1, collecting cell performance count data and wireless access bearer diagnostic data, acquiring historical operation comparison data and historical configuration structure data, and performing preprocessing; S2, performing back-calculation analysis on cell performance count data, wireless access bearer diagnostic data, and historical configuration structure data with the same caliber, issuing unified statistical granularity based on the back-calculation analysis results, enabling original count pass-through, triggering time source verification and clock synchronization review, and solidifying cross-level comparison caliber; S3, performing anti-dilution fidelity analysis based on historical operation comparison data and historical configuration structure data, switching regional statistical perspectives based on the anti-dilution fidelity analysis results, enabling user equipment measurement report tracking and collection, generating fine-grained logs of switching events, and establishing evidence binding; S4, performing joint stability issuance and adjudication analysis oriented towards caliber consistency, local fidelity, fault existence, and stability evidence, executing count detail reporting, comparison difference convergence, verification processing, and directional retest drilling processing, and fixing the statistical perspective.

[0009] Furthermore, the specific process of collecting cell performance count data and radio access bearer diagnostic data, and obtaining historical operation comparison data and historical configuration structure data is as follows: Collecting cell performance count data includes: access request count, access success count, dropped call count, handover request count, handover success count, and radio resource block occupancy rate; collecting radio access bearer diagnostic data includes: base station load statistics, base station backhaul link latency statistics, base station backhaul link packet loss statistics, base station alarm event sequence, baseline signal reception quality sequence, and interference-related measurement sequence; obtaining historical operation comparison data and historical configuration structure data, wherein the historical configuration structure data includes historical cell parameter configuration records, historical neighbor cell relationship records, historical cell affiliation area relationships, and historical effective timestamps.

[0010] Furthermore, the specific preprocessing process is as follows: Outlier removal and smoothing of cell performance count data and radio access bearer diagnostic data are performed using the sliding window Hampshire outlier detection algorithm and the quantile truncation smoothing algorithm; time alignment of cell performance count data and radio access bearer diagnostic data is performed using the cross-correlation-based time offset estimation algorithm and the unified time grid resampling algorithm; effective interval segmentation and binding of analysis windows are completed using the configuration effective interval mapping algorithm based on historical cell parameter configuration records, historical neighbor cell relationship records, and historical effective timestamps; and cell performance count data and radio access bearer diagnostic data are normalized to the zero-to-one interval using distribution normalization and linear normalization algorithms.

[0011] Furthermore, the specific process of performing backcalculation analysis on cell performance count data, radio access bearer diagnostic data, and historical configuration structure data using the same caliber is as follows: Based on historical cell parameter configuration records and historical effective timestamps, the cell performance count data and historical configuration structure data are used to calculate the drop rate, access success rate, and handover success rate within the same configuration effective interval, forming a count-derived index sequence with the radio resource block occupancy rate; For each cell under the same base station, a same-caliber backcalculation is performed to obtain the cell base station-side alignment sequence, which is then compared with the base station-side observation sequence composed of base station bearer load statistics, base station backhaul link delay statistics, and base station backhaul link packet loss statistics. The normalized L2 difference is then calculated to obtain the cell backcalculation base station deviation energy; Based on the historical cell affiliation region relationship and historical effective timestamps, backcalculation is performed on the cell indicators within the same configuration effective interval to form a regional backcalculation alignment index sequence, which is then compared with the regional base station-side observation index sequence to calculate the normalized L2 difference to obtain the cell backcalculation regional deviation energy; For the same area, calculate the heterogeneous fluctuation of the cell based on the drop rate and radio resource block occupancy rate, and obtain the aperture matching pass rate by statistically analyzing the proportion of windows that complete alignment within the same configuration effective interval; calculate the distribution comparable divergence of the baseline signal received quality sequence, interference-related measurement sequence, and drop rate; calculate the time alignment bending cost of the base station alarm event sequence and drop count; calculate the sum of the cell back-calculation base station deviation energy and the cell back-calculation area deviation energy, divide it by the cell heterogeneous fluctuation plus one, take the negative value of the result, and take the exponential function to obtain the deviation energy suppression factor; calculate the ratio of the aperture matching pass rate plus one to the distribution comparable divergence plus one, and take the natural logarithm to obtain the aperture logarithmic ratio; calculate the logarithmic ratio mapping function of the aperture logarithmic ratio to obtain the logarithmic ratio mapping factor; calculate the time alignment bending cost, take the negative value, and take the exponential function to obtain the alignment bending suppression factor; calculate the product of the deviation energy suppression factor, the logarithmic ratio mapping factor, and the alignment bending suppression factor to obtain the cross-level aperture comparable consistency discriminant value.

[0012] Furthermore, the specific process of issuing unified statistical granularity, enabling original count pass-through, triggering time source verification and clock synchronization review, and solidifying cross-level comparison calibers based on the back-calculation analysis results is as follows: Real-time comparison of cross-level caliber comparable consistency discrimination value and cross-level caliber comparable consistency discrimination threshold: When the cross-level caliber comparable consistency discrimination value is less than the cross-level caliber comparable consistency discrimination threshold, caliber and granularity correction processing is performed, and a unified analysis window length resampling task is issued to unify the reporting granularity of base stations and cells; within the same configuration effective interval, back-calculation is performed on cell performance counting data. The data is aggregated and compared with the observation data from the base station side. If they are still inconsistent, the raw count of performance management is transmitted and reported. At the same time, unified time source verification and clock synchronization review are triggered for the n base stations with the highest time alignment bending cost. After completion, the anti-dilution fidelity analysis is performed. When the cross-level caliber comparable consistency discrimination value is greater than or equal to the cross-level caliber comparable consistency discrimination threshold, the cell index back-calculation result under the configuration effective interval constraint is used as the unified reference caliber. The statistical granularity and time alignment method are locked, the alignment sequence after back-calculation using the statistical caliber is located and calculated, and the anti-dilution fidelity analysis is performed.

[0013] Furthermore, the specific process of de-dilution fidelity analysis based on historical operational comparison data and historical configuration structure data is as follows: Feature vectors are constructed based on cell performance count data, baseline signal reception quality sequences, and interference-related measurement sequences. These vectors are compared with historical operational comparison data within the same configuration effective interval. A cell anomaly score sequence is obtained using a density ratio-based relative likelihood anomaly detection algorithm, and the maximum and median cell anomaly scores are calculated. Based on historical cell affiliation regions and historical effective timestamps, numerical perturbation analysis is performed on the regional aggregation process within the same configuration effective interval. Aggregation sensitivity is obtained based on cell performance count data using a numerical difference calculation algorithm. The missing value is calculated by combining historical neighbor cell relationship records and anomaly score sequences to obtain the adjacent synchronization masking value; the difference between the maximum and median anomaly scores of a cell is calculated and divided by the maximum anomaly score plus one, the result is negative and then exponentially applied to obtain the dispersion suppression amount; the dispersion suppression amount is calculated by subtracting one from the dispersion suppression amount to obtain the anomaly salience enhancement amount; the aggregation sensitivity missing value is calculated, negative and then exponentially applied to obtain the aggregation sensitivity missing suppression factor; the adjacent synchronization masking value is calculated, negative and then exponentially applied to obtain the adjacent synchronization masking suppression factor; the product of the anomaly salience enhancement amount, the aggregation sensitivity missing suppression factor, and the adjacent synchronization masking suppression factor is calculated to obtain the anomaly de-dilution fidelity discriminant value.

[0014] Furthermore, the specific process of switching the regional statistical perspective, enabling user equipment measurement report tracking and collection, generating fine-grained logs of handover events, and establishing evidence binding based on the de-dilution fidelity analysis results is as follows: Real-time comparison of abnormal de-dilution fidelity discrimination values ​​and abnormal de-dilution fidelity discrimination value thresholds: When the abnormal de-dilution fidelity discrimination value is less than the abnormal de-dilution fidelity discrimination threshold, the current aggregation perspective is maintained, and the abnormal score sequence is input into the joint stability issuance and adjudication analysis in the form of the original sequence; When the abnormal de-dilution fidelity discrimination value is greater than or equal to the abnormal de-dilution fidelity discrimination threshold, de-dilution diagnostic enhancement processing is performed: In the positioning calculation link, the regional aggregation statistics are switched from mean-based statistics to upper quantile statistics, and upper quantile statistics are enabled for the drop rate, handover success rate, and radio resource block occupancy rate indicators; For the top k cells in terms of abnormal scores, user equipment measurement report tracking and collection and fine-grained log generation of handover events are enabled on the base station side, and the collection results are bound one-to-one with the cell performance count data in the corresponding time window; After completion, the joint stability issuance and adjudication analysis is initiated.

[0015] Furthermore, the specific process of conducting joint stability issuance adjudication analysis oriented towards caliber consistency, local fidelity, fault existence, and stability evidence is as follows: An observation sequence is constructed from cell performance counting data and radio access bearer diagnostic data; the log-likelihood ratio of the fault state relative to the normal state is calculated using a hidden Markov state model online forward probability recursion algorithm to obtain the fault existence log-likelihood gain value; independent component analysis decoupling algorithms are used to calculate the reconstruction error of the interference-related measurement sequence, radio resource block occupancy rate, call drop rate, and handover success rate to obtain the abnormal coupling undecoupling residual degree; and the call drop rate and baseline signal reception quality sequence are analyzed using a Behnken algorithm. The posterior probability of the running length for online variable point detection in Yess is used to obtain the consistency of the stable running length. The negative value of the log-likelihood gain after subtracting the fault existence is taken as the value of an exponential function to obtain the fault existence gain mapping quantity. The negative value of the fidelity discriminant value after subtracting the anomaly is taken as the value of an exponential function to obtain the fidelity gain mapping quantity. The negative value of the undecoupled residual degree of anomaly coupling is taken as the value of an exponential function to obtain the undecoupled residual suppression factor. The product of the cross-level comparable consistency discriminant value, the fault existence gain mapping quantity, the fidelity gain mapping quantity, the undecoupled residual suppression factor, and the stable running length consistency is calculated to obtain the level-consistent positioning stable issuance value.

[0016] Furthermore, the specific process of executing detailed count reporting, differential convergence, verification and handling, and directional retest drilling handling and fixing the statistical perspective is as follows: Real-time comparison of the hierarchical consistent positioning stable issuance value and the hierarchical consistent positioning stable issuance threshold: When the hierarchical consistent positioning stable issuance value is less than the hierarchical consistent positioning stable issuance threshold, the output of positioning conclusions is paused and the stability enhancement process is entered, and the original performance count details are reported; Cells with abnormal scores lower than the median of cell abnormal scores are selected as the control set, the differential deviation sequence is calculated and the candidate range is converged, a verification and handling action is issued, and the stable operating length determined by the change point detection is reached. The internal verification checks the consistency of the change direction of the disconnection count and the handover success count; when the hierarchical consistent positioning stable issuance value is greater than or equal to the hierarchical consistent positioning stable issuance threshold, a positioning is issued and an executable positioning handling chain is executed. For the top m cells in the cell anomaly score ranking, user equipment measurement report directional tracking collection and handover failure reason code extension reporting are enabled; the backhaul link latency and packet loss of the base station to which the positioning object belongs are retested and the consistency with the base station alarm event sequence is verified. If the backhaul is abnormal, the positioning object is drilled up to the base station backhaul link level, and the statistical perspective and alignment method used for positioning calculation are fixed until the cooling phase is entered.

[0017] The second aspect of this invention provides a wireless network fault diagnosis system based on big data analysis, comprising: a data acquisition and preprocessing module for acquiring cell performance count data and wireless access bearer diagnostic data, obtaining historical operation comparison data and historical configuration structure data, and performing preprocessing; and a cross-level caliber comparability and consistency discrimination module for performing caliber back-calculation analysis on cell performance count data, wireless access bearer diagnostic data, and historical configuration structure data, and issuing unified statistical granularity, enabling original count pass-through, triggering time source verification and clock synchronization review, and solidifying cross-level data based on the caliber back-calculation analysis results. The system includes: a reference caliber; a local anomaly anti-dilution fidelity discrimination module, used for anti-dilution fidelity analysis based on historical operational reference data and historical configuration structure data; a module for switching regional statistical perspectives, enabling user equipment measurement report tracking and collection, generating fine-grained logs of switching events, and establishing evidence binding based on the anti-dilution fidelity analysis results; and a hierarchical consistency positioning stability issuance module, used for joint stability issuance adjudication analysis oriented towards caliber consistency, local fidelity, fault existence, and stability evidence; a module for performing count detail reporting, reference difference convergence, verification processing, and directional retest drilling processing, and fixing the statistical perspective.

[0018] Beneficial effects The present invention has the following beneficial effects: (1) By introducing cross-level statistical caliber consistency constraints and unified alignment mechanisms in the process of multi-level network data analysis, this invention can establish a comparable analysis basis between different network levels, avoid cognitive bias caused by statistical caliber differences in different levels of the same anomaly, and thus achieve the effect of consistent cross-level fault judgment results. This effectively solves the problem of inconsistent anomaly performance caused by differences in statistical and aggregation mechanisms in the prior art.

[0019] (2) This invention analyzes the risk of local anomalies appearing from the perspective of upper-level statistics and introduces a de-dilution discrimination and enhancement mechanism, so that local faults can still be identified in regional or upper-level aggregate analysis, thereby achieving the effect of stable development of local anomalies and effectively solving the problem of local faults being masked by upper-level indicators in the prior art.

[0020] (3) By combining performance index analysis with wireless link status characterization, the present invention enables fault determination to no longer rely solely on a single statistical index or empirical threshold, thereby achieving a more comprehensive fault cause identification and effectively solving the problem of fault judgment relying on a single perspective and insufficient interpretability in the prior art.

[0021] (4) By introducing the joint adjudication and hierarchical disposal logic of multiple types of evidence in the location decision-making process, the present invention enables the location conclusion to enter the execution stage only when the evidence is sufficient and consistent, thereby achieving the effect of location results being executable and implementable, and effectively solving the problem that the diagnostic conclusions in the prior art are difficult to directly guide the disposal.

[0022] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0023] Figure 1 This is a flowchart of the wireless network fault diagnosis method based on big data analysis of the present invention; Figure 2 This is a structural diagram of the wireless network fault diagnosis system based on big data analysis according to the present invention; Figure 3 The heatmap showing the correlation between cross-level aperture discrimination parameters under different time windows in this invention; Figure 4 This is a flowchart of the abnormal de-dilution fidelity detection and enhancement process of the present invention. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] Please see Figures 1-4 This invention provides a technical solution: a wireless network fault diagnosis method based on big data analysis, comprising the following steps: S1, collecting cell performance count data and wireless access bearer diagnostic data, acquiring historical operation comparison data and historical configuration structure data, and preprocessing them; S2, performing back-calculation analysis on the cell performance count data, wireless access bearer diagnostic data, and historical configuration structure data, issuing unified statistical granularity based on the back-calculation analysis results, enabling original count pass-through, triggering time source verification and clock synchronization review, and solidifying cross-level comparison caliber; S3, performing anti-dilution fidelity analysis based on historical operation comparison data and historical configuration structure data, switching regional statistical perspectives based on the anti-dilution fidelity analysis results, enabling user equipment measurement report tracking and collection, generating fine-grained logs of switching events, and establishing evidence binding; S4, performing joint stability issuance and adjudication analysis oriented towards caliber consistency, local fidelity, fault existence, and stability evidence, executing count detail reporting, comparison difference convergence, verification processing, and directional retest drilling processing, and fixing the statistical perspective.

[0026] Specifically, the process of collecting cell performance count data and radio access bearer diagnostic data, and obtaining historical operation comparison data and historical configuration structure data is as follows: Collect cell performance count data, which includes: access request count, access success count, dropped call count, handover request count, handover success count, and radio resource block occupancy rate; used for subsequent cross-level caliber mapping, local anomaly quantification, and location stability issuance.

[0027] Wireless access bearer diagnostic data is collected, including: base station bearer load statistics, base station backhaul link latency statistics, base station backhaul link packet loss statistics, and base station alarm event sequences. These are used for subsequent cell-to-base station comparability verification and target drilling constraints. Baseline signal reception quality sequences and interference-related measurement sequences are also collected. The baseline signal reception quality sequence refers to the signal reception quality time series data formed under relatively stable network operation conditions, without significant alarms or sudden interference. It is used to characterize the normal wireless reception level and its natural fluctuation range, serving as a reference benchmark for performance degradation judgment. The interference-related measurement sequence refers to the wireless measurement time series data reflecting the intensity changes of co-channel interference, adjacent channel interference, or external electromagnetic disturbances. It is used to characterize the degree of impact of undesired signals on the wireless link transmission quality. This is used to subsequently align performance degradation with the physical characterization of the wireless link, avoiding reliance solely on aggregate counts.

[0028] Historical operational comparison data and historical configuration structure data are acquired. The historical configuration structure data includes historical cell parameter configuration records, historical neighbor cell relationship records, historical cell affiliation area relationships, and historical effective timestamps. Record segments in the historical configuration structure data that cover the timestamp of the current analysis window are used for subsequent alignment and cross-level mapping constraints according to the effective interval. During cross-level backcalculation and comparison, historical data is segmented and matched according to the effective interval of the configuration to avoid data mixing under different configuration conditions. Historical operational comparison data refers to cell performance count data and radio access bearer diagnostic data collected within the corresponding historical time period, including historical access success count, historical disconnection count, historical handover success count, and historical radio resource block occupancy rate. This is used to compare the current observation with the actual historical operational level to form a reference baseline for anomaly detection.

[0029] In this implementation plan, by uniformly collecting and organizing multi-source wireless network operation data and historical comparison data, a complete operational observation foundation covering both cell and base station levels can be constructed. This enables the analysis of the current operational status and historical actual operational levels to be compared, while forming a stable and reproducible data analysis environment under the constraints of configuration structure and effective time. This process provides a reliable data support foundation for subsequent consistency judgment of cross-level statistical calibers, quantitative identification of local anomalies, and stable issuance of positioning results, avoiding analysis biases introduced by dispersed data sources, inconsistent configuration conditions, or insufficient time alignment.

[0030] Specifically, the preprocessing process is as follows: Instantaneous spikes and outliers in cell performance count data and radio access bearer diagnostic data are removed and smoothed using a sliding window Hampshire outlier detection algorithm and a quantile truncation smoothing algorithm. This prevents short-term noise from being amplified into pseudo-hierarchical conflicts after cross-level aggregation. The sliding window length is determined based on the main sampling period of the cell performance count data and radio access bearer diagnostic data, and is taken to be between one-tenth and one-quarter of the analysis window length to ensure that anomaly detection has sufficient sensitivity to short-term disturbances without disrupting medium- and long-term trends. The quantile truncation smoothing algorithm... The truncation quantiles of the method are set within the range of 5% to 10% above and below to suppress the pulling effect of extreme outliers on the statistics. Using a time offset estimation algorithm based on cross-correlation and a unified time grid resampling algorithm, cell performance count data and radio access bearer diagnostic data are time-aligned within the same analysis window. The analysis window length is calculated from the statistical distribution of adjacent sampling intervals of the timestamp sequences of cell performance count data and radio access bearer diagnostic data. Specifically, the main peak sampling period of the cell and base station timestamp difference sequences is calculated separately, and the least common multiple of the two is taken as the analysis window length for statistical analysis. First, time grid alignment is used to ensure comparability of cross-level controls within the same time window. The time range is between 5 and 30 minutes. The time step for unified time grid resampling is no greater than one-quarter of the analysis window length to ensure that the window contains at least multiple valid sampling points, thereby maintaining the statistical stability of the time-aligned sequences and ensuring comparability of cross-level controls within the same time window. Second, short-term missing segments in the baseline signal reception quality sequence and interference-related measurement sequence are completed using an event sequence consistency-based missing segment identification algorithm and a piecewise cubic spline interpolation completion algorithm. The completed positions are then further analyzed. Row status labeling: Short-term missing segments refer to missing intervals whose continuous missing length does not exceed the length of a single analysis window. Missing intervals exceeding this length are not included in interpolation and are directly marked as unavailable, avoiding the smooth masking of local faults caused by missing segments during aggregation. By configuring the effective interval mapping algorithm, based on historical cell parameter configuration records, historical neighbor cell relationship records, and historical effective timestamps, the effective interval segmentation of cell performance count data and radio access bearer diagnostic data is aligned, and the corresponding effective interval identifier is bound to the analysis window to ensure that subsequent cross-level backcalculations and comparisons are only performed under the same configuration conditions.By employing distribution standardization and linear normalization algorithms, cell performance count data and radio access bearer diagnostic data are standardized and normalized to a zero-to-one interval. Standardization utilizes a robust method based on the statistical distribution within the historical configuration effective interval, while normalization employs a minimum-maximum value normalization method. The upper and lower bounds are determined by the historical statistical range within the corresponding configuration effective interval. This improves the numerical stability of subsequent calculations under different levels and dimensions, ensuring that all index parameters involved in subsequent discrimination and fusion calculations are dimensionless. It also ensures that nonlinear operations such as exponential, logarithmic, and logical function mappings are performed on a unified scale, avoiding numerical bias or computational instability caused by differences in dimensions or inconsistent scales.

[0031] In this implementation plan, this step systematically performs quality regulation and consistency processing on the operational data of multi-source wireless networks. This enables data from different acquisition sources, with different time granularities and statistical scales to form a continuous, stable, and comparable observation basis under a unified analysis window and consistent configuration. This significantly reduces the uncertainty caused by instantaneous noise, time misalignment, missing segments, and dimensional differences in subsequent analysis. After this processing, various performance and diagnostic data maintain consistency in temporal structure, configuration constraints, and numerical distribution, providing reliable and reproducible data quality assurance for cross-level statistical comparison, anomaly feature identification, and stable output of discrimination results.

[0032] Specifically, the process of performing a back-calculation analysis on cell performance count data, radio access bearer diagnostic data, and historical configuration structure data using the same caliber is as follows: Based on historical cell parameter configuration records and historical effective timestamps, the time axis is segmented according to parameter effective events. The continuous time interval between two adjacent configuration effective timestamps is recorded as the same configuration effective interval. Within the same configuration effective interval, the drop rate, access success rate, and handover success rate are calculated based on access request count, access success count, drop count, handover request count, and handover success count. These are then combined with the radio resource block occupancy rate. The same sequence of derived indicators is used to perform back-calculation on each cell under the same base station to obtain the cell base station-side aligned sequence. This sequence is then compared with the base station-side observation sequence, which consists of base station load statistics, base station backhaul link delay statistics, and base station backhaul link packet loss statistics. The normalized L2 difference is then used to calculate the cell back-calculated base station deviation energy. The normalized L2 difference refers to the process where, within the same configuration effective interval, the indicator sequences of each dimension involved in the calculation are first standardized on the same scale, and then normalized using the statistical amplitude or maximum observable amplitude under the historical stable state of each dimension as the normalization coefficient to ensure the homogeneity and comparability of indicators with different dimensions in energy calculation. Based on the cell affiliation area record, within the same configuration... A region is defined as a set of multiple base stations and their subordinate cells belonging to the same network management area identifier within the effective time interval. Cell performance count data within this region is back-calculated and aggregated based on historical cell affiliation and historical effective timestamps within the same configuration effective interval to obtain a region-level back-calculation aligned indicator sequence. This sequence is then compared with the regional base station-side observation indicator sequence aggregated within the region based on base station load statistics to calculate the cell back-calculation regional deviation energy. The call drop rate and radio resource block occupancy rate within the same region are calculated using a robust standard deviation algorithm to obtain intra-cell heterogeneous volatility. Consistency checks are performed within the same configuration effective interval based on historical configuration structure data. The caliber matching pass rate is obtained by statistically analyzing the proportion of windows that complete alignment within the same configuration effective interval. The caliber matching pass rate is the proportion of the number of analysis windows that meet the criteria of time alignment, consistent statistical granularity, and consistent indicator caliber within the configuration effective interval, out of the total number of analysis windows within the configuration effective interval. This is used to characterize the time coverage effectiveness of cross-level caliber constraints under the current configuration conditions. The baseline signal reception quality sequence, interference-related measurement sequence, and call drop rate are constructed into a distribution using a kernel density estimation probability distribution algorithm, and then the comparable divergence of the distribution is obtained using the Jensen-Shannon divergence calculation algorithm. The time alignment bending cost is obtained from the base station alarm event sequence and call drop count using a dynamic time warping alignment algorithm.

[0033] The sum of the cell-based base station back-calculation deviation energy and the cell-based regional back-calculation deviation energy is calculated and divided by the intra-cell heterogeneous volatility plus one. The result is then negative and taken as an exponential function to obtain the deviation energy suppression factor. The added term in the denominator serves as a numerical stability shift constant, used to avoid division by zero or over-amplification issues caused by extremely small intra-cell heterogeneous volatility. Simultaneously, it ensures that the deviation energy can still be smoothly constrained in low-volatility scenarios. By normalizing the deviation energy with intra-cell heterogeneous volatility, the deviation term maintains scale invariance in multi-cell concurrent volatility scenarios, avoiding... To prevent the suppression factor from erroneously converging due to natural fluctuations within the region, the ratio of the caliber matching pass rate plus one to the distribution comparable divergence plus one is calculated and its natural logarithm is taken to obtain the caliber logarithmic ratio. Adding one to both the numerator and denominator serves as a shift constant for the logarithmic operation, ensuring that the domain of the logarithmic operation is always positive and avoiding numerical divergence or logarithmic singularities when the caliber matching pass rate or distribution comparable divergence approaches zero. The use of the logarithmic ratio enhances the sensitivity to changes in the caliber matching pass rate and provides symmetric suppression of the nonlinear increase effect of the distribution comparable divergence. This improves the resolution for discriminating changes in caliber comparability; it calculates the logarithmic ratio of caliber using a logistic function mapping to obtain the logarithmic ratio mapping factor. The logistic function mapping employs a monotonically continuous function with upper and lower asymptotic bounds. This saturation characteristic limits the unilateral dominance of extreme logarithmic ratios on the discriminating value and ensures the mapping result stably falls within a finite interval, thereby improving the numerical stability and reproducibility of the calculation process across the analysis window. It also forms a differentiable and monotonically gated probabilistic output, facilitating multiplicative fusion with other suppression factors; and it negates the computation time alignment bending cost. The value is then used to perform an exponential function to obtain the alignment bending suppression factor. The exponential mapping is used to convert the linear change of the time alignment bending cost value into a nonlinear decay constraint, so that small time offsets have a limited impact on the discrimination results, while significant window misalignment or temporal deformation can be quickly suppressed. At the same time, it avoids scale imbalance caused by linear penalty in the high value range, so that cross-level pseudo-conflicts caused by time window misalignment are preferentially suppressed in the fusion stage. The product of the bias energy suppression factor, the log ratio mapping factor and the alignment bending suppression factor is calculated to obtain the cross-level comparable consistency discrimination value. The specific calculation formula is as follows: ; In the formula, This represents the cross-level caliber comparability and consistency judgment value, which serves as the triggering basis for caliber and granularity correction and provides a reliable caliber premise for subsequent stable issuance. It represents the energy of the cell back-calculation base station deviation, and characterizes the energy-type residual intensity of the cell-side back-calculation sequence and the base station-side observation sequence under the same statistical caliber. It represents the energy of the cell back-calculation area deviation, characterizing the dilution effect of the area aggregation layer on cell anomalies and the degree of perturbation of the consistency of area observations by cross-base station aggregation deviation; It represents the heterogeneous volatility within a cell, characterizing the structural dispersion level of the volatility of indicators across multiple cells within the same region; It represents the caliber matching pass rate, quantifies the window coverage of alignment completed within the effective period of the same configuration, and thus reflects the availability and stability of the current caliber constraints in the time dimension; It represents the distributional comparable divergence, quantifying the similarity deviation between the baseline signal received quality sequence, interference-related measurement sequence, and dropout rate at the probability distribution level; It represents the cost of time alignment bending, quantifies the cost of temporal distortion in the alignment process of alarm event sequences and disconnection counts, and reflects the risk of cross-source inconsistency caused by time source drift, incorrect window reporting, and event trigger lag. This represents a logic function mapping operator used to stably map logarithmic ratios to the interval between 0 and 1, converting caliber logarithmic ratios into comparable gating factors and preventing numerical overflow.

[0034] In this embodiment, Table 1 is a comparison table of cross-level statistical caliber consistency discrimination parameters, which records in detail the values ​​of key parameters involved in the cross-level caliber discrimination process under different time windows, including the cross-level caliber comparable consistency discrimination value, cell back-calculation base station deviation energy, cell back-calculation area deviation energy, time alignment bending cost, and caliber matching pass rate, used to characterize the changing characteristics of cross-level statistical caliber consistency under different operating states and alignment conditions. Specifically: in time window 1 (caliber consistency), the cell back-calculation base station deviation energy is 0.10, the cell back-calculation area deviation energy is 0.08, the time alignment bending cost is 0.05, the caliber matching pass rate is 0.92, and the corresponding cross-level caliber comparable consistency discrimination value is 0.526; in time window 2 (slight inconsistency), the deviation energies increase, the caliber matching pass rate decreases to 0.85, and the corresponding discrimination value decreases to 0.375; in time window 3 (significant inconsistency), the deviation energy and distribution differences are significant. The time alignment cost increases significantly, and the caliber matching pass rate drops to 0.55, while the discriminant value decreases to 0.120. In time window 4 (misalignment window or event lag), the time alignment cost increases to 0.60, while the discriminant value remains at a low level. In time window 5 (recovery after correction) and time window 6 (high consistency), the relevant deviation indicators gradually converge, the caliber matching pass rate rebounds, and the corresponding cross-level caliber comparable consistency discriminant values ​​recover to 0.440 and 0.561, respectively, thus reflecting the comparative relationship of caliber consistency status under different time windows.

[0035] Table 1. Comparison of Parameters for Cross-Level Statistical Calibration Consistency Judgment like Figure 3The figure shows a heatmap of the correlation between cross-level caliber discrimination parameters under different time windows. The graph uses time windows as the horizontal axis and key discrimination parameters as the vertical axis, visually reflecting the relative strength changes of each parameter under different time windows through color intensity. As shown in Table 1, within time windows of caliber consistency or recovery after correction, the deviation energy and time alignment bending cost are generally at a low level, while the comparable consistency discrimination values ​​for cross-level calibers show a higher distribution. In time windows with significant inconsistencies or window lag, the deviation energy and bending cost are significantly enhanced, and the corresponding consistency discrimination values ​​are significantly reduced. Overall, this heatmap visually demonstrates the changing structure and correlation of cross-level statistical caliber-related parameters under different operating conditions, and can be used to help understand the overall trend of cross-level caliber consistency discrimination results changing with time windows.

[0036] In this implementation plan, this step establishes a unified statistical comparison basis across cells, base stations, and regions under the constraint of the configured effective interval. This enables observations at different levels to be comparable in terms of consistent caliber and time structure. By comprehensively characterizing the bias intensity, regional dilution degree, internal fluctuation structure, distribution consistency, and time alignment reliability, the risk of cross-level statistical inconsistency is transformed into a quantifiable discrimination result. The resulting cross-level caliber comparable consistency discrimination value can stably reflect the reliability of the statistical caliber within the current analysis window, providing a clear trigger basis for subsequent caliber and granularity correction decisions, and establishing a reliable caliber consistency premise for the stable issuance of positioning results.

[0037] Specifically, the process of issuing unified statistical granularity based on the back-calculation analysis results, enabling raw count pass-through, triggering time source verification and clock synchronization review, and solidifying cross-level comparison calibers is as follows: Real-time comparison of cross-level caliber comparable consistency discrimination values ​​and cross-level caliber comparable consistency discrimination thresholds: When the cross-level caliber comparability and consistency discrimination value is less than the cross-level caliber comparability and consistency discrimination threshold, caliber and granularity correction is performed: a unified analysis window length resampling task is issued, and the unified analysis window length is determined based on the original reporting cycle configuration of cell and base station performance management data in the network management system, through the minimum common reporting cycle alignment rule, to adjust the performance management reporting granularity of base stations and cells to the same window length; based on historical cell parameter configuration records, historical neighbor cell relationship records, and historical effective timestamps, the cell performance count data within the analysis window is back-calculated and aggregated within the same configuration effective interval, and re-compared with the base station side observation data. If they are still inconsistent, the base station enables performance management original count transparent reporting to replace the secondary aggregation caliber. Here, original count transparent reporting refers to the base station side directly reporting the basic performance counts without secondary aggregation processing by the regional or network management layer. The basic performance counting object includes at least the original statistical fields corresponding to access request count, access success count, disconnection count, handover request count, handover success count, and wireless resource block occupancy rate, to ensure the consistency and traceability of statistical standards when comparing across levels and reduce the masking caused by upper-level standards. For base stations n before the occurrence of time alignment bend cost, n is the upper limit of the number of single batches processed for unified time source verification and clock synchronization review. It is determined by offline capacity accounting analysis based on the configuration value of the concurrency limit of time source verification task in the network management system and the average time consumption of single base station verification. The offline capacity accounting analysis uses the batch processing scale calculation rule under the fixed concurrency limit constraint to obtain n. The value of n is from 1 to 8. The unified time source verification process of base station alarm and performance count is enabled and clock synchronization review is triggered to reduce false conflicts caused by alarm and performance misalignment windows. After completion, it enters the de-dilution fidelity analysis.

[0038] When the cross-level caliber comparable consistency discrimination value is greater than or equal to the cross-level caliber comparable consistency discrimination threshold, the cell index back-calculation result under the configuration effective interval constraint is used as the unified reference caliber, the statistical granularity and time alignment method are locked, the alignment sequence after the statistical caliber back-calculation is located and calculated, the inconsistency caused by the difference in calibers at the source is eliminated, and the analysis is entered into anti-dilution fidelity analysis.

[0039] In this implementation plan, this step achieves dynamic diversion control of the statistical caliber credibility status by real-time monitoring and threshold determination of the cross-level caliber comparability discriminant value. This enables cross-level analysis to promptly enter the correction and calibration process when calibers are inconsistent, and to lock in a unified reference caliber to stably advance subsequent analysis when the calibers are credible. As a result, an adaptive caliber governance mechanism is formed during system operation, which avoids misjudgment caused by directly performing positioning calculations under caliber distortion conditions, and also prevents the result oscillation caused by repeatedly switching statistical perspectives when the calibers are consistent. This provides a stable, reliable and sustainable statistical foundation for subsequent anomaly de-dilution fidelity analysis.

[0040] Specifically, the process of de-dilution fidelity analysis based on historical operational control data and historical configuration structure data is as follows: Feature vectors are constructed based on cell performance count data, baseline signal reception quality sequences, and interference-related measurement sequences. Same-scale constraints are applied to features of different dimensions to ensure the identifiability of density estimation and reduce the risk of single-feature dominance. Cell performance count data corresponding to historical operational control data within the same configuration effective interval are compared. Configuration effective interval constraints are used to eliminate distribution drift interference introduced by parameter switching. A cell anomaly score sequence is obtained through a density ratio-based relative likelihood anomaly detection algorithm. The maximum value of the anomaly score sequence is taken as the maximum cell anomaly score, which is used to capture the strongest anomaly evidence within the window and serve as the peak representation for subsequent de-dilution triggering. The median cell anomaly score is obtained from the anomaly score sequence using a median calculation algorithm. The median provides a robust background level characterization and weakens the pulling effect of a few extreme anomalies on the baseline scale. Based on the historical cell affiliation region relationship and historical effective timestamps, under fixed regional mapping relationships and time conditions... Under the premise of consistency, the aggregation boundary is constrained, and numerical perturbation analysis is performed on the regional aggregation process within the same configuration effective interval. The numerical perturbation analysis simulates local anomaly changes by applying controlled-amplitude small perturbations to the performance count data of one or more cells. The perturbation amplitude is determined by the statistical fluctuation range of the corresponding index within the historical stable operating interval or a preset relative change ratio. The perturbation step size is set in a progressively increasing manner to characterize the response characteristics of the regional aggregation results to different perturbation intensities. The controllable perturbation simulates the transitivity of local cell changes to the regional aggregation results. Cell performance count data is used as input, and the input is kept only from the cell side to isolate the influence of the base station side aggregation caliber on the sensitivity assessment. The change amplitude of the regional aggregation results is calculated using a numerical difference algorithm to obtain the aggregation sensitivity missing degree. The adjacency synchronization masking degree is obtained by calculating the adjacency synchronization masking degree of historical neighbor cell relationship records and anomaly scoring sequences using a graph Laplace smoothing intensity calculation algorithm. The adjacency graph structure constraint is used to characterize the synchronous diffusion and consistency smoothing effect of anomalies between neighbor cells, making the risk of mutual masking of anomalies caused by topology correlation explicit.

[0041] The difference between the maximum and median anomaly scores of a cell is calculated and divided by the maximum score plus one. The result is then negative and exponentially applied to obtain the dispersion suppression amount. A peak dominance characterization is constructed using the comparison between the maximum and median scores, and exponential compression is used to limit the impact of extreme outliers on subsequent product links within a controllable range. The dispersion suppression amount is then subtracted from the difference to obtain the anomaly salience enhancement amount. This suppression amount is then inversely mapped to the enhancement amount, making the anomaly distribution more peak-like, thus maintaining visibility from a regional aggregation perspective and improving subsequent trigger sensitivity. Finally, the aggregation sensitivity missing value is negative and exponentially applied. The aggregation sensitivity deficiency suppression factor is obtained as a penalty term for the aggregation operator's passivation of local perturbations, prompting the overall discrimination to be automatically downweighted when aggregation is insensitive to local changes to avoid being masked by mean. The adjacency synchronization masking degree is calculated by taking a negative value and then applying an exponential function to obtain the adjacency synchronization masking suppression factor, which suppresses the pseudo-consistency caused by spatial adjacency synchronization fluctuations and reduces the interference of collective smoothing masking caused by co-directional fluctuations in neighboring regions on local anomaly discrimination. The product of the anomaly salience enhancement, the aggregation sensitivity deficiency suppression factor, and the adjacency synchronization masking suppression factor is calculated to obtain the anomaly de-dilution fidelity discrimination value. The specific calculation formula is as follows: ; In the formula, This represents the anomaly de-dilution fidelity discrimination value, which is used to drive whether to switch the regional statistical perspective and whether to enable tracking acquisition, so that local anomalies have stable development capabilities in the upper-level statistics. This represents the maximum anomaly score of the community, used to provide an upper bound anchor point for the strongest evidence of anomalies in the area, supporting a faithful characterization of the peak intensity of local faults. This represents the median of the community anomaly score, used to provide a robust baseline for the level of background anomalies within the region, reducing the bias of a small number of anomalies on the baseline estimate. It represents the degree of missing aggregation sensitivity, characterizing the degree of attenuation of the response of regional aggregation results to cell-level disturbances, thereby providing a structural constraint on statistical dilution risk; It represents the adjacency synchronization masking degree, which describes the synchronization trend caused by the topological propagation of neighboring regions or the same source disturbance, so that local anomalies are not disguised by adjacency consistency.

[0042] In this implementation plan, this step transforms the risk of local anomalies being weakened or masked during regional statistics into quantifiable criteria by systematically characterizing and structurally modeling the anomaly features at the cell level. This makes the anomaly detection capability no longer dependent on a single statistical result or empirical judgment. The resulting anomaly de-dilution fidelity discriminant value can comprehensively reflect the peak significance of local anomalies, the sensitivity of regional aggregation to local disturbances, and the masking effect brought about by neighboring cell topology synchronization. This effectively suppresses the masking of local anomalies by meanization and synchronization at the statistical aggregation level, enabling local faults to maintain stable and identifiable detection capabilities from the perspective of higher-level statistics. This provides a reliable triggering basis for whether to switch statistical perspectives and whether to enable targeted tracking and collection.

[0043] Specifically, the process of switching the regional statistical perspective based on the de-dilution fidelity analysis results, enabling user equipment measurement report tracking and collection, generating fine-grained logs of the switching event, and establishing evidence binding is as follows: Figure 4 The diagram shows the flowchart for anomaly dedilution fidelity detection and enhancement, which compares the anomaly dedilution fidelity detection value and the anomaly dedilution fidelity detection value threshold in real time. When the anomaly dedilution fidelity discriminant value is less than the anomaly dedilution fidelity discriminant threshold, the current aggregation perspective is maintained, and the anomaly score sequence is input into the joint stable issuance adjudication analysis in the form of the original sequence for participation in subsequent stability adjudication without triggering additional aggregation enhancement operations.

[0044] When the abnormal de-dilution fidelity discrimination value is greater than or equal to the abnormal de-dilution fidelity discrimination threshold, de-dilution diagnostic enhancement measures are performed to prevent local faults from being masked by the upper layer: In the positioning calculation link, the regional aggregation statistics are switched from mean-based statistics to upper quantile statistics, and upper quantile statistics are enabled for indicators such as call drop rate, handover success rate, and radio resource block occupancy rate; For the top k cells in the abnormal score ranking, user equipment measurement report tracking and fine-grained log generation for handover events are enabled on the base station side, where k is the upper limit of the number of parallel cells for directional tracking and fine-grained log generation, determined by offline collection budget accounting analysis based on the fixed throughput upper limit configuration value and the fixed log writing rate upper limit configuration value of the base station side measurement report collection interface. The offline collection budget accounting analysis uses the number of cells that can be carried under the throughput upper limit constraint to obtain k, where the value of k ranges from 1 to 12. The user equipment measurement report includes at least the reference signal received power, reference signal received quality, signal-to-noise ratio, and neighboring data. The measurement results and corresponding timestamp fields are included. The fine-grained log of the handover event includes at least the handover trigger reason, handover result status, failure reason code, and event occurrence timestamp fields. The collected measurement reports and event logs are time-stamped using a unified time source. The collected results are bound one-to-one with the cell performance count data within the corresponding time window. The one-to-one binding is completed through timestamp matching, with an allowed time alignment tolerance of sixty seconds. When the deviation between the timestamp of the user equipment measurement report or the timestamp of the fine-grained log of the handover event and the window start timestamp or window end timestamp of the analysis window to which the cell performance count data belongs does not exceed sixty seconds, it is determined to be a valid binding within the same analysis window. If it exceeds sixty seconds, it is determined to be cross-window data and the binding is rejected or marked as data to be reviewed, in order to be compatible with measurement report reporting jitter and event log triggering delay, while ensuring that the measurement data and performance counts are comparable within the same analysis window. After completion, the joint stability issuance adjudication analysis is performed.

[0045] In this implementation scheme, this step achieves adaptive control over whether the regional aggregation perspective needs to be enhanced by real-time determination of the anomaly de-dilution fidelity discriminant value. This enables the system to automatically introduce more sensitive statistical methods and directional acquisition means when the local anomaly development capability is insufficient, and to maintain the existing aggregation perspective to avoid unnecessary disturbances when the anomaly development capability is sufficient. Thus, a dynamic adjustment mechanism for the visibility of local faults is established in the overall diagnostic process. This prevents local faults from being excessively smoothed and masked in the upper-level statistics, and avoids frequent switching of statistical perspectives due to excessive enhancement. This provides a balanced and reliable anomaly development basis for subsequent joint and stable issuance of decisions.

[0046] Specifically, the process of conducting joint stability issuance and adjudication analysis oriented towards consistency of data, local fidelity, fault existence, and stability evidence is as follows: An observation sequence is constructed from cell performance counting data and radio access bearer diagnostic data. Timestamp consistency checks and missing segment state annotations are performed on the observation sequence to ensure the continuity of state recursion. The log-likelihood ratio of the fault state relative to the normal state is calculated using an online forward probability recursion algorithm based on a Hidden Markov State Model (HMM). The HMM includes at least two types of hidden states: normal and fault states. The number of states is based on the minimum discriminative requirement of the fault diagnosis scenario. The transmission probability is determined by constructing the statistical distribution or empirical estimation results of the observation sequence within the historical stable operating range. During the online recursion process, the stationarity of parameter updates is maintained. Forward probability recursion is used to accumulate and update the state evidence at each sampling point, suppressing excessive triggering of state handover by single-point noise, and normalizing the mapping to obtain the fault existence log-likelihood gain. For the cell-level interference-related measurement sequences, radio resource block occupancy rate, call drop rate, and handover success rate, the dominant components are separated using an independent component analysis decoupling algorithm, and the reconstruction error is calculated and normalized to obtain the abnormal coupling undecoupled residual degree. The independent components... The number of components used in the analysis is determined based on the input feature dimensions and their statistical independence constraints. The extracted dominant components are used to characterize the main explainable variation patterns, while the reconstruction error of the remaining components is used to characterize anomalous coupling structures that cannot be explained by the dominant components. Before decoupling, zero-mean and unit-variance constraints are applied to each input quantity to reduce the bias of dimensional differences on the separation matrix estimation. The reconstruction error is used to characterize the remaining structural quantities that cannot be explained by the dominant components. Anomalous couplings that are superimposed by multiple factors and difficult to separate are explicitly represented in the form of residuals and used as penalty evidence for stable issuance. The dropout rate and baseline signal reception quality sequences are analyzed using Bayesian methods. The stable runtime consistency is calculated using the posterior probability of the online change point detection runtime. The prior distribution and risk function used in Bayesian online change point detection are used to characterize the balance between state maintenance and state abrupt change. The parameters are determined by the statistical characteristics of historical runtime data to avoid oversensitivity to short-term random fluctuations. The confidence level of being in the same stable segment is quantified by the runtime posterior probability to avoid short-term fluctuations being misjudged as structural changes. The consistency of the change point positions of two sequences is used to characterize the synchronicity between performance degradation and wireless link characterization, thereby providing verifiable temporal consistency evidence for stable issuance.

[0047] The fault existence gain mapping quantity is obtained by taking the negative value of the log-likelihood gain after subtracting the fault's existence and then taking the exponential function value. This exponential saturation mapping compresses the evidence of fault existence into a finite interval, suppressing the excessive amplification of issuance decisions by single bursts of evidence, and ensuring that issuance strength increases smoothly with the enhancement of evidence. The fidelity gain mapping quantity is obtained by taking the negative value of the anomaly de-dilution fidelity discriminant and then taking the exponential function value. This transforms the visibility fidelity of local anomalies into a multiplicative constraint weight, automatically reducing issuance confidence when fidelity is insufficient, ensuring that the location conclusion is synchronized with the traceability of evidence. Finally, the undecoupled residual degree of anomaly coupling is calculated by taking the negative value of the exponential function value, resulting in an undecoupled residual suppression factor. This applies an exponential penalty to anomaly coupling residuals, reducing the undecoupled residuals... Explainable coupling residuals are considered as sources of structural uncertainty and discounted to reduce misattribution issuance caused by multi-factor coupling. The product of the cross-level comparable consistency discriminant value, fault presence gain mapping, fidelity gain mapping, undecoupled residual suppression factor, and stable operating length consistency is calculated to obtain the hierarchical consistent positioning stable issuance value. The range of the result is limited by exponential mapping and multiplicative superposition to avoid numerical overflow or scale imbalance caused by the superposition of multiple exponential functions. A hard-constraint adjudication structure is formed by product fusion; if any evidentiary dimension is insufficient, the overall issuance strength decreases. This mechanism ensures that consistency, fidelity visibility, fault verifiability, controllable coupling, and verifiable stability are simultaneously met before entering the issuance stage. The specific calculation formula is as follows: ; In the formula, This indicates a consistent and stable issuance value for the location at each level. It is used to drive whether the location conclusion enters the solidification process of the issuance and disposal chain, reducing the risk of disposal failure caused by frequent switching of location objects in adjacent analysis windows. It represents the cross-level caliber comparability and consistency discrimination value, which is used to provide a reliable basis for cross-level comparison calibers, so that the issuance of decisions is based on the same statistical caliber after comparability and alignment; It represents the discriminant value for anomalies dedilution, characterizing the credibility of local anomalies from the perspective of upper-level statistics, and avoiding the issuance bias caused by regional convergence masking but not being seen. The log-likelihood gain value indicates the presence of a fault and is used to provide the strength of probabilistic evidence of the existence of a fault, thus elevating the location from the level of indicator deterioration to a verifiable basis for the existence of a state. It represents the degree of undecoupled residuals in abnormal coupling, characterizes the degree of separability of the cause of the abnormality, and discounts the irremovable coupling residues to reduce the probability of mislocation in issuance. It indicates the consistency of stable running length, and the constraint that issuance is only valid when the time structure is stable and the evidence of running length is consistent, improves the consistency of the verification of issuance conclusions and the continuity across windows.

[0048] In this implementation plan, this step constructs a location issuance judgment basis oriented towards multi-evidence consistency by jointly characterizing the existence of faults, the degree of anomaly visibility fidelity, the separability of anomaly causes, and temporal stability. This makes the location decision no longer rely on a single indicator or instantaneous anomaly, but is based on the comprehensive conditions of cross-level credibility, local anomaly visibility, fault state verifiability, and temporal structure stability. The resulting hierarchical consistent location stable issuance value can synergistically constrain different evidence dimensions. It automatically reduces the issuance strength when the evidence is insufficient or there is structural uncertainty, and stably outputs the location conclusion when the evidence is sufficient and consistent. From a mechanism perspective, it effectively suppresses the risk of non-executable handling caused by frequent switching of location objects within adjacent analysis windows, and improves the reliability, continuity, and verifiability of fault location results.

[0049] Specifically, the process of performing detailed reporting of counts, comparison of differential convergence, verification and handling, and directional retesting and drilling up, and fixing the statistical perspective, is as follows: Real-time comparison of the stable issuance value of hierarchical consistent positioning and the stable issuance threshold of hierarchical consistent positioning: When the hierarchical consistent positioning stability issuance value is less than the hierarchical consistent positioning stability issuance threshold, the output of positioning conclusions is paused and the stability enhancement process is initiated: the candidate cell set obtained based on anomaly score sorting is switched to the original performance count detail reporting mode, the original count details of candidate cells are used to participate in positioning calculation, and the use of regional and base station hierarchical summary statistics is paused; within the same region, cells with anomaly scores lower than the median of cell anomaly scores are selected as the control set, the differential deviation sequence of candidate cells relative to the control set is calculated and the candidate range is converged, the differential deviation sequence is taken as the unified calculation caliber and time window benchmark under the configuration effective interval constraint, and the back-calculation alignment index sequence of the candidate cell and the back-calculation alignment index sequence of the control set are obtained by window-by-window difference within the same analysis window, which is used to characterize the net deviation of the candidate cell relative to the regional background level under the same caliber comparable condition, to avoid incomparability of deviation due to statistical caliber switching; a verification and handling action is issued to the converged candidate cells, and the consistency of the change direction of the disconnection count and the handover success count is verified within the stable operating length interval determined by change point detection. After completion, the hierarchical consistent positioning stability issuance value is recalculated and the next round of adjudication is initiated.

[0050] When the hierarchical consistent positioning stable issuance value is greater than or equal to the hierarchical consistent positioning stable issuance threshold, positioning is issued and an executable positioning handling chain is executed to avoid only outputting conclusions without practical application: the priority of positioning objects is determined based on the cell anomaly scoring ranking results, and for the top m cells, directional tracking collection of user equipment measurement reports and extended reporting of handover failure reason codes are enabled on the base station side. m is the upper limit of the number of cells processed in parallel in a single batch of the executable positioning handling chain, which is determined by offline handling capability calculation and analysis based on the configuration values ​​of the concurrent upper limit of the backhaul retest task and the concurrent upper limit of the reason code extended reporting task in the network management system. The force calculation analysis uses a parallel work order scale calculation rule under a fixed concurrency upper limit constraint to obtain m, where m ranges from 1 to 6. A directional retest of backhaul link latency and packet loss is performed on the base station to which the location object belongs, and the retest results are verified for consistency with the base station alarm event sequence. If the consistency verification backhaul is abnormal, the location object is drilled up from the cell level to the base station backhaul link level to avoid misjudging link-related problems as multi-cell wireless problems. The statistical perspective and alignment method used for location calculation are fixed, and the location object does not participate in statistical perspective switching until the current location processing is completed and the cooling-off phase begins, preventing the diagnostic conclusion from jumping within adjacent windows.

[0051] In this implementation plan, this step establishes a stability gating mechanism between the output of positioning conclusions and the execution of disposal by real-time determination of the stable issuance value of the hierarchical consistent positioning. This enables the system to proactively delay issuance and enter the enhanced verification process when there is insufficient evidence or the consistency standard is not met. Only when the evidence is sufficient and consistent will it enter the executable disposal stage. This avoids premature output of positioning conclusions under unstable conditions, which would lead to repeated corrections. It also ensures that once issued, there is a clear disposal object, disposal path, and execution boundary. From a mechanism perspective, this achieves the feasibility, traceability, and stable continuation of positioning conclusions across windows, effectively reducing the risks of frequent switching of positioning objects, hierarchical misjudgment, and ineffective consumption of disposal resources.

[0052] like Figure 2As shown, the second aspect of this invention provides a wireless network fault diagnosis system based on big data analysis, comprising: a data acquisition and preprocessing module, used to acquire cell performance count data and wireless access bearer diagnostic data, obtain historical operation comparison data and historical configuration structure data, and perform preprocessing; and a cross-level caliber comparability and consistency discrimination module, used to perform caliber back-calculation analysis on cell performance count data, wireless access bearer diagnostic data and historical configuration structure data, and, based on the caliber back-calculation analysis results, issue unified statistical granularity, enable original count pass-through, trigger time source verification and clock synchronization review, and solidify the cross-level caliber back-calculation analysis results. The system includes: a hierarchical comparison caliber module; a local anomaly anti-dilution fidelity discrimination module, used for anti-dilution fidelity analysis based on historical operational comparison data and historical configuration structure data, switching regional statistical perspectives based on the anti-dilution fidelity analysis results, enabling user equipment measurement report tracking and collection, generating fine-grained logs of switching events, and establishing evidence binding; and a hierarchical consistency positioning and stable issuance module, used for joint stable issuance adjudication analysis oriented towards caliber consistency, local fidelity, fault existence, and stability evidence, executing count detail reporting, comparison difference convergence, verification handling, and directional retest drilling handling, and fixing the statistical perspective.

[0053] In this implementation plan, through the collaborative operation of modules, a complete closed loop is formed, encompassing data collection, cross-level comparison, anomaly detection, and location adjudication. This ensures stable comparability between cells and base stations under unified statistical standards and time granularity constraints, significantly reducing the interference of hierarchical differences on anomaly determination. Furthermore, local anomalies remain visible and traceable in regional and upper-level statistics, preventing them from being masked by aggregation and smoothing processes. Simultaneously, through a multi-dimensional evidence joint adjudication and stable issuance mechanism, the location results are based on consistent standards, anomaly fidelity, verifiable faults, and temporal stability, thereby improving the accuracy, stability, and executability of fault location conclusions and supporting the continuity and consistency of subsequent handling processes.

[0054] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0055] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A wireless network fault diagnosis method based on big data analysis, characterized in that, Includes the following steps: S1, collect cell performance count data and wireless access bearer diagnostic data, obtain historical operation comparison data and historical configuration structure data, and perform preprocessing; S2, perform a back-calculation analysis on cell performance count data, radio access bearer diagnostic data, and historical configuration structure data using the same caliber. Based on the results of the back-calculation analysis, issue a unified statistical granularity, enable raw count pass-through, trigger time source verification and clock synchronization review, and solidify the cross-level comparison caliber. The specific process of performing the back-calculation analysis on cell performance count data, radio access bearer diagnostic data, and historical configuration structure data using the same caliber is as follows: Based on historical cell parameter configuration records and historical effective timestamps, cell performance counting data and historical configuration structure data are used to calculate the call drop rate, access success rate and handover success rate within the same configuration effective interval, which together with the radio resource block occupancy rate form a counting derived index sequence. For each cell under the same base station, a back-calculation with the same caliber is performed to obtain the cell base station-side alignment sequence. This sequence is then compared with the base station-side observation sequence, which consists of base station load statistics, base station backhaul link delay statistics, and base station backhaul link packet loss statistics, and the normalized L2 difference is calculated to obtain the cell back-calculation base station deviation energy. Based on the historical cell affiliation region relationship and historical effective timestamp, back-calculation is performed on the cell indicators within the same configuration effective interval to form a regional back-calculation alignment indicator sequence. The normalized L2 difference is calculated with the regional base station-side observation indicator sequence to obtain the cell back-calculation regional deviation energy. The heterogeneous fluctuation within the cell is calculated for the drop rate and radio resource block occupancy rate within the same region. The proportion of windows that complete alignment within the same configuration effective interval is statistically analyzed to obtain the caliber matching pass rate. The distribution comparability divergence is calculated for the baseline signal reception quality sequence, interference-related measurement sequence, and drop rate. The time alignment bending cost is calculated for the base station alarm event sequence and drop count. The sum of the cell back-calculation base station deviation energy and the cell back-calculation area deviation energy is calculated and divided by the sum of the heterogeneous fluctuations within the cell plus one. The result is then negative and an exponential function is applied to obtain the deviation energy suppression factor. Calculate the ratio of the caliber matching pass rate plus one to the distribution comparable divergence plus one, and take the natural logarithm to obtain the caliber logarithmic ratio. The logarithmic ratio of the caliber is calculated by a logical function mapping to obtain the logarithmic ratio mapping factor; The alignment bending cost is calculated by taking the negative value and then taking the exponential function to obtain the alignment bending suppression factor; The product of the deviation energy suppression factor, the log ratio mapping factor, and the alignment bending suppression factor is calculated to obtain the cross-level caliber comparability and consistency discrimination value. S3 performs anti-dilution fidelity analysis based on historical operation comparison data and historical configuration structure data. Based on the anti-dilution fidelity analysis results, it switches the regional statistical perspective, enables user equipment measurement report tracking and collection, generates fine-grained logs of switching events, and establishes evidence binding. The specific process of performing anti-dilution fidelity analysis based on historical operational comparison data and historical configuration structure data is as follows: Feature vectors are constructed based on cell performance counting data, baseline signal reception quality sequences, and interference-related measurement sequences. These vectors are compared with historical operational control data within the same configuration effective interval. Cell anomaly score sequences are obtained through a density ratio-based relative likelihood anomaly detection algorithm, and the maximum and median cell anomaly scores are calculated. Based on the historical cell affiliation and historical effective timestamp, numerical perturbation analysis is performed on the regional aggregation process within the same configuration effective interval. Based on cell performance count data, the aggregation sensitivity missing degree is obtained through numerical difference calculation algorithm. The adjacency synchronization masking degree is calculated by combining historical neighbor cell relationship records and abnormal scoring sequences. Calculate the difference between the maximum cell anomaly score and the median cell anomaly score, divide it by the sum of the maximum cell anomaly score plus one, take the negative value of the result and then take the exponential function to obtain the dispersion suppression amount; Calculate the difference suppression amount by subtracting the discrepancy amount to obtain the abnormal salience enhancement amount; calculate the aggregation sensitivity missing degree, take the negative value and then take the exponential function to obtain the aggregation sensitivity missing suppression factor; calculate the adjacent synchronization masking degree, take the negative value and then take the exponential function to obtain the adjacent synchronization masking suppression factor. The product of the anomalous salience enhancement, the polymerization sensitivity loss inhibitory factor, and the adjacent synchronous masking inhibitory factor is calculated to obtain the anomalous de-dilution fidelity discriminant value; S4, perform joint stability issuance and adjudication analysis based on evidence of caliber consistency, local fidelity, fault existence, and stability; execute detailed count reporting, differential convergence, verification processing, and directional re-testing drilling processing, and fix the statistical perspective; the specific process of performing joint stability issuance and adjudication analysis based on evidence of caliber consistency, local fidelity, fault existence, and stability is as follows: An observation sequence is constructed from cell performance counting data and wireless access bearer diagnostic data. The log-likelihood ratio of the fault state to the normal state is calculated using the online forward probability recursion algorithm of the Hidden Markov State Model to obtain the log-likelihood gain value of the fault presence. The reconstruction error is calculated using the independent component analysis decoupling algorithm for the interference-related measurement sequence, wireless resource block occupancy rate, call drop rate, and handover success rate to obtain the residual degree of abnormal coupling and non-decoupling. The stable operating length consistency degree is obtained by using the Bayesian online change point detection posterior probability of the call drop rate and baseline signal reception quality sequence. The fault existence gain mapping quantity is obtained by taking the negative value of the log-likelihood gain after calculating the fault existence, and then taking the difference of the exponential function. The fidelity gain mapping quantity is obtained by taking the negative value of the fidelity discriminant after subtracting the anomaly from the anti-dilution fidelity discriminant and then taking the difference of the exponential function. The undecoupled residual degree of anomalous coupling is calculated by taking the negative value and then taking the exponential function to obtain the undecoupled residual suppression factor; The product of the cross-level caliber comparable consistency discrimination value, fault presence gain mapping amount, fidelity gain mapping amount, undecoupling residual suppression factor and stable operating length consistency is calculated to obtain the level consistent positioning stable issuance value.

2. The wireless network fault diagnosis method based on big data analysis according to claim 1, characterized in that: The specific process for collecting cell performance count data and radio access bearer diagnostic data, and obtaining historical operational comparison data and historical configuration structure data is as follows: Collect cell performance count data, which includes: access request count, access success count, dropped connection count, handover request count, handover success count, and radio resource block occupancy rate. Collect wireless access bearer diagnostic data, which includes: base station bearer load statistics, base station backhaul link latency statistics, base station backhaul link packet loss statistics, base station alarm event sequence, baseline signal reception quality sequence, and interference-related measurement sequence. Obtain historical operation comparison data and historical configuration structure data. The historical configuration structure data includes historical cell parameter configuration records, historical neighbor cell relationship records, historical cell belonging area relationships, and historical effective timestamps.

3. The wireless network fault diagnosis method based on big data analysis according to claim 1, characterized in that: The specific process of preprocessing is as follows: Anomaly detection and smoothing algorithms based on the sliding window Hampshire algorithm and quantile truncation smoothing algorithm are used to remove and smooth out anomalies in cell performance count data and radio access bearer diagnostic data. Time alignment of cell performance count data and radio access bearer diagnostic data is performed using a cross-correlation-based time offset estimation algorithm and a unified time grid resampling algorithm. Based on historical cell parameter configuration records, historical neighbor cell relationship records, and historical effective timestamps, effective interval segmentation alignment and binding of analysis windows are completed using a configuration effective interval mapping algorithm. Distribution standardization and linear normalization algorithms are used to normalize cell performance count data and radio access bearer diagnostic data to the zero-to-one interval.

4. The wireless network fault diagnosis method based on big data analysis according to claim 1, characterized in that: The specific process of issuing unified statistical granularity based on the back-calculation analysis results, enabling raw count pass-through, triggering time source verification and clock synchronization review, and solidifying cross-level comparison calibers is as follows: Real-time comparison of cross-level caliber comparability and consistency criteria thresholds: When the cross-level caliber comparability discriminant value is less than the cross-level caliber comparability discriminant threshold, caliber and granularity correction processing is performed, and a unified analysis window length resampling task is issued to unify the reporting granularity of base stations and cells; within the same configuration effective interval, the cell performance count data is back-calculated and aggregated and re-compared with the base station side observation data. If they are still inconsistent, the performance management raw count transparent transmission and reporting is enabled; at the same time, unified time source verification and clock synchronization verification are triggered for the n base stations with the first time alignment bending cost. After completion, the anti-dilution fidelity analysis is entered. When the cross-level comparable consistency discrimination value is greater than or equal to the cross-level comparable consistency discrimination threshold, the cell index back-calculation result under the configuration effective interval constraint is used as the unified reference standard, the statistical granularity and time alignment method are locked, the alignment sequence after back-calculation using the statistical standard is located and calculated, and then the analysis is performed to prevent dilution and preserve fidelity.

5. The wireless network fault diagnosis method based on big data analysis according to claim 1, characterized in that: The specific process of switching the regional statistical perspective based on the de-dilution fidelity analysis results, enabling user equipment measurement report tracking and collection, generating fine-grained logs of the switching event, and establishing evidence binding is as follows: Real-time comparison of abnormal dedilution fidelity discriminant values ​​and abnormal dedilution fidelity discriminant thresholds: When the abnormal dedilution fidelity discriminant value is less than the abnormal dedilution fidelity discriminant threshold, the current aggregation perspective is maintained, and the abnormal score sequence is input into the joint stable issuance adjudication analysis in the form of the original sequence. When the abnormal de-dilution fidelity discrimination value is greater than or equal to the abnormal de-dilution fidelity discrimination threshold, the de-dilution diagnosis enhancement processing is performed: in the positioning calculation link, the regional aggregation statistics are switched from mean-based statistics to upper quantile statistics, and the upper quantile statistics method is enabled for the indicators of drop rate, handover success rate and wireless resource block occupancy rate. For the top k cells in the anomaly score ranking, enable user equipment measurement report tracking and collection and fine-grained log generation of handover events on the base station side, and bind the collection results one-to-one with the cell performance count data in the corresponding time window; Once completed, proceed to the joint stability issuance and analysis stage.

6. The wireless network fault diagnosis method based on big data analysis according to claim 1, characterized in that: The specific process of executing the count details report, comparing the difference convergence, verifying and handling the directional retest drilling and fixing the statistical perspective is as follows: Real-time comparison of stable issuance values ​​and stable issuance thresholds for consistent positioning at different levels: When the stable issuance value of hierarchical consistent positioning is less than the stable issuance threshold of hierarchical consistent positioning, the output of positioning conclusion is paused and the stability enhancement process is entered, and the original performance count details are reported. Cells with anomaly scores lower than the median of cell anomaly scores are selected as the control set. The differential deviation sequence is calculated and the candidate range is narrowed. A verification and handling action is issued. Within the stable operating length determined by the change point detection, the consistency of the change direction of the disconnection count and the handover success count is verified. When the stable issuance value of hierarchical consistent positioning is greater than or equal to the stable issuance threshold of hierarchical consistent positioning, positioning is issued and an executable positioning handling chain is executed. For the top m cells in cell anomaly score, user equipment measurement reports are used for directional tracking and collection and handover failure reason code extension reporting. For the base station to which the positioning object belongs, directional retesting of backhaul link latency and packet loss is performed and consistency verification with base station alarm event sequence is performed. If backhaul is abnormal, the positioning object is drilled up to the base station backhaul link level, and the statistical perspective and alignment method used for positioning calculation are fixed until the cooling phase is entered.

7. A wireless network fault diagnosis system based on big data analysis, employing the wireless network fault diagnosis method based on big data analysis as described in any one of claims 1-6, comprising: The data acquisition and preprocessing module is used to acquire cell performance count data and radio access bearer diagnostic data, obtain historical operation comparison data and historical configuration structure data, and perform preprocessing. The cross-level caliber comparability and consistency discrimination module is used to perform caliber back-calculation analysis on cell performance counting data, wireless access bearer diagnostic data and historical configuration structure data. Based on the caliber back-calculation analysis results, it issues unified statistical granularity, enables original count pass-through, triggers time source verification and clock synchronization review, and solidifies the cross-level comparison caliber. The local anomaly de-dilution fidelity discrimination module is used to perform de-dilution fidelity analysis based on historical operation comparison data and historical configuration structure data. Based on the de-dilution fidelity analysis results, it switches the regional statistical perspective, enables user equipment measurement report tracking and collection, generates fine-grained logs of switching events, and establishes evidence binding. The hierarchical consistency positioning and stable issuance module is used to perform joint stable issuance adjudication analysis based on evidence of caliber consistency, local fidelity, fault existence, and stability. It performs count detail reporting, comparison difference convergence, verification processing, and directional retest drilling processing, and fixes the statistical perspective.

Images