Multi-dimensional timing rate quality detection method for 5g-r control safety service

Abstract

This invention discloses a multi-dimensional time-series rate quality defect detection method for 5G-R train control safety services, including: acquiring hourly traffic statistics reports and network parameter configuration data of 5G-R network cells; splicing to generate a cell-level multi-dimensional time-series dataset; after clustering based on railway geographical scenarios, removing normal samples without rate quality defect risks through a two-level pruning mechanism; adopting a scenario-based rate anomaly identification algorithm guided by train control safety, and completing the quantification of rate anomaly degree and data cleaning through service security level weighted distance calculation, global relative density correction, and scenario adaptive neighbor point number dynamic matching; finally outputting the rate quality defect judgment result; this invention designs a complete detection process specifically for the 5G-R railway train control safety requirements, significantly reducing manual dependence, significantly improving the detection rate and calculation efficiency of hidden rate quality defects in low-density clusters, accurately capturing slight network rate degradation, and adapting to the large-scale rate quality defect analysis requirements of the entire 5G-R network.

Description

Technical Field

[0001] This invention relates to the fields of 5G mobile communication and railway intelligent operation and maintenance technology, specifically to the field of rate quality defect analysis of train control services in 5G-R (5G for Railway) railway mobile communication systems, and particularly to a multi-dimensional time-series rate quality defect detection method for 5G-R train control safety services. Background Technology

[0002] 5G-R, as the next-generation wireless communication system for China's railways, is gradually replacing the traditional GSM-R system and becoming a key technological foundation supporting core safety services such as the intelligent high-speed rail train control system (CTCS-3 / 4), automatic driving, and vehicle-to-ground collaboration. The speed performance of the 5G-R network directly affects the data transmission quality of train control safety services and railway operation safety. Train control services have strict requirements for data transmission rates; rate degradation can lead to delays in train control command transmission, data packet loss, and even communication interruptions, seriously threatening operational safety.

[0003] In the daily operation and maintenance of 5G-R networks and the management of poor speed quality, operation and maintenance personnel need to analyze massive amounts of network speed-related indicator data to identify potential speed anomalies that may affect train control services, pinpoint the root causes of speed quality issues, and provide data support for preventative network optimization. Among these indicators, average number of users, downlink traffic, channel quality indicator (CQI), and timing lead (TA) are the core characteristics that directly affect user speeds, and abnormal changes in these indicators usually indicate the occurrence of speed quality problems.

[0004] However, in actual operation and maintenance, the existing technology has the following prominent problems:

[0005] I. Traditional solutions cannot detect slight deterioration trends in train control operations;

[0006] Existing technologies primarily determine 5G-R network speed anomalies by setting fixed threshold values, i.e., based on the historical experience of optimization personnel. This approach has serious flaws in practical applications: it can only identify severe speed anomalies exceeding a global threshold, failing to capture slight deterioration trends in network speed indicators. This results in delayed detection of speed quality issues, hindering early warning and preventative optimization, and easily triggering subsequent serious security incidents such as train control command transmission delays and data interruptions.

[0007] Second, single-dimensional judgment has serious limitations;

[0008] Traditional solutions only make independent threshold judgments for single indicators, ignoring the strong correlation between 5G-R network speed-related indicators. For example, a sudden increase in the number of users, a decrease in CQI, and an increase in TA often occur in a coordinated manner, jointly leading to speed degradation. However, single-dimensional threshold solutions cannot identify such multi-indicator-linked speed anomalies, which are prone to missed or false judgments of poor speed quality and make it difficult to accurately locate the root cause of poor speed quality.

[0009] Third, it has poor adaptability to multiple scenarios and insufficient accuracy in anomaly recognition.

[0010] 5G-R routes cover various geographical scenarios, including stations, plains, bridges, and tunnels, with significant differences in the distribution of rate-related indicators across these different scenarios. For example, signal attenuation is high in tunnels, resulting in generally low CQI and naturally lower user speeds; while in plains, signal quality is good, leading to generally higher speeds. Existing fixed threshold schemes use a globally uniform standard, which cannot differentiate based on the local data distribution patterns in different scenarios, resulting in insufficient accuracy in identifying poor speed and quality, and making it difficult to support precise governance.

[0011] Fourth, it relies heavily on manual labor and cannot support large-scale analysis across the entire line.

[0012] The setting of threshold values ​​relies entirely on the historical experience of optimization personnel. There is no unified standard for setting the rate quality difference threshold under different scenarios, road sections, and vehicle speeds, which requires repeated manual adjustments, resulting in extremely low analysis efficiency. At the same time, the number of cells covered by 5G-R railways can reach hundreds, and the sample size can reach tens of thousands after the accumulation of hourly data. Existing technologies cannot meet the needs of large-scale rate quality difference analysis of 5G-R railways.

[0013] Fifth, anomaly detection methods based on statistical distribution cannot be adapted to the characteristics of 5G-R data;

[0014] Traditional anomaly detection schemes based on statistical methods and clustering algorithms usually require the assumption that the data follows a specific probability distribution. However, the rate-related data in the 5G-R high-speed scenario is highly dynamic and non-stationary, so this assumption often cannot be valid. At the same time, clustering algorithms can only output a binary judgment result of whether it is abnormal or not, and cannot quantify the degree of rate anomaly of each data point, making it difficult to adapt to the refined requirements of 5G-R rate quality difference analysis.

[0015] In summary, existing technologies cannot simultaneously address the multiple core technical challenges in 5G-R rate quality defect analysis, such as the inability to provide early warnings of train control safety rate risks, the concealment of low-density clusters, the missed detection of rate anomalies, and the insufficient efficiency of large-scale analysis across the entire railway line. Therefore, there is an urgent need for an intelligent rate quality defect detection method specifically tailored for the 5G-R railway scenario. Summary of the Invention

[0016] The purpose of this invention is to solve the technical problems of traditional fixed threshold schemes in the analysis of poor network speed in existing 5G-R networks, such as high dependence on manual intervention, inability to accurately identify slight degradation trends in train control service speeds, and poor adaptability to multiple scenarios. This invention provides a multi-dimensional time-series rate quality defect detection method for 5G-R train control security services.

[0017] To achieve the above objectives, the present invention provides the following technical solution: a multi-dimensional time-series rate quality defect detection method for 5G-R train control security services, characterized by comprising the following steps:

[0018] Data Acquisition: Hourly traffic statistics reports of target coverage cells along the 5G-R railway line are acquired, along with network parameter configuration data for the corresponding cells after each network parameter modification. The traffic statistics reports include core indicators for train control services, coverage quality indicators, and traffic volume indicators. Among these, core indicators for train control services include wireless connection rate, drop rate, and handover success rate; coverage quality indicators include average CQI and average TA; and traffic volume indicators include average number of users and downlink traffic.

[0019] Data splicing: Using the unique identifier of the cell as the association basis, the call statistics report data and network parameter configuration data are spliced ​​together to generate a cell-level multidimensional time series dataset;

[0020] Scene-aware clustering and two-level data pruning: Based on the railway geographical scene attributes, the multidimensional time series dataset is clustered by scene to generate sub-datasets corresponding to different railway scenes; then, through a two-level pruning mechanism of first-level pruning of railway operation and maintenance rules and second-level pruning of improved isolation forest, normal samples without rate quality difference risk in each sub-dataset are removed to generate a candidate sample set to be detected.

[0021] Scenario-based rate anomaly identification based on train control safety: For the candidate sample set to be detected, the rate anomaly degree of the sample points is optimized by calculating the weighted distance based on the train control business security level, the original local reachability density, the global relative density correction, and the quantification value of the rate anomaly degree.

[0022] Anomaly detection: Based on the obtained quantitative value of the rate anomaly degree, an adaptive dual-threshold detection mechanism, with a fixed threshold as the primary threshold and a proportional threshold as the secondary threshold, is used to detect anomalies.

[0023] Output results: Perform multi-dimensional correlation analysis and degradation level classification on abnormal samples to generate a rate quality poor report that includes abnormal cells, abnormal time, correlation indicators, and degradation level.

[0024] Furthermore, for the candidate sample set to be detected, the method for optimizing the calculation of the rate anomaly degree of the sample points through train control service security level weighted distance calculation, global relative density correction, and rate anomaly degree quantification calculation is as follows:

[0025] By combining the Analytic Hierarchy Process (AHP) with the experience of 5G-R railway operation and maintenance experts, weights are assigned to call volume statistics indicators to obtain the weighted Euclidean distance between sample points. Among them, the total weight of the core indicators of train control business is greater than the total weight of the coverage quality indicators, which is greater than the total weight of the call volume indicators, and the total weight of the core indicators of train control business is not less than 0.7.

[0026] The k-distance neighborhood of a sample point is calculated based on weighted Euclidean distance. And reachability distance, and then calculate the original local reachability density. ;

[0027] Global relative density correction: Calculate the global average reachable density Calculate the correction factor α(P) to obtain the corrected local reachability density lrd'(P);

[0028] ;

[0029] ;

[0030] Rate anomaly quantification: Based on the corrected local reachability density, calculate the rate anomaly quantification value of sample point P. The formula is:

[0031] .

[0032] Furthermore, the scene-aware clustering and two-level data pruning specifically include:

[0033] Based on the engineering parameter labels of the communities along the railway line, the multidimensional time series dataset is divided into four sub-datasets: station scene, plain line scene, bridge scene, and tunnel scene.

[0034] Level 1 pruning: Based on the qualified thresholds of wireless connection rate ≥99.5%, drop rate ≤0.1%, and handover success rate ≥99%, rule filtering is performed on the subset of data in each scenario to remove normal samples that meet all three conditions and have no manually initiated network-side configuration parameter change operations in the corresponding time period.

[0035] Secondary pruning: For the remaining samples after primary pruning, an improved isolation forest algorithm is used for rapid pre-screening to remove normal samples whose isolation depth is greater than the preset isolation depth threshold;

[0036] The above pruning operations were performed on the four types of scenario subsets respectively, and then merged to generate the final set of candidate samples to be detected.

[0037] Furthermore, based on the obtained quantitative value of rate anomaly, an anomaly point determination method using a dual-threshold adaptive determination mechanism, with a fixed threshold as the primary threshold and a proportional threshold as the secondary threshold, is employed:

[0038] The preset fixed anomaly judgment threshold is 1.5, and sample points with a rate anomaly degree quantification value ≥1.5 are marked as abnormal samples;

[0039] It supports dynamically adjusting the judgment criteria based on the abnormal point ratio threshold. When the abnormal point ratio judged by the fixed threshold deviates from the range of operation and maintenance experience, the sample points ranked in the top N% of the quantitative value of the abnormal rate are marked as abnormal samples.

[0040] Furthermore, for m-dimensional sample point P and sample point O, the weighted Euclidean distance is calculated as follows:

[0041] ;

[0042] Where m is the indicator dimension of the standardized sample set. , Let P and O be the numerical values ​​of sample points P and O on the i-th feature dimension, respectively. Let be the weight coefficient corresponding to the i-th dimension index, and satisfy . .

[0043] Furthermore, in the scenario-based rate anomaly identification based on train control safety, the process also includes: standardizing the candidate sample set to be detected, removing non-numerical fields, eliminating the dimensional differences of different indicators, obtaining a standardized multidimensional dataset, and performing deduplication preprocessing on the standardized multidimensional dataset to remove completely duplicate samples.

[0044] Furthermore, when performing deduplication preprocessing on the standardized multidimensional dataset and removing completely duplicated samples, a minimum reachable distance threshold is set for the case where there are 3 or more duplicate points in the k-distance neighborhood of the dataset. The minimum reachable distance threshold is the 1% quantile of the weighted Euclidean distance of all sample points in the scene subset.

[0045] Furthermore, in the scenario-based rate anomaly point identification based on train control safety guidance, it also includes: scenario-adaptive dynamic matching of the number of neighboring points, specifically including:

[0046] For each scene subset, based on the number of samples n and the coefficient of variation cv, the optimal number of nearest neighbors k for that scene is dynamically calculated using the following formula:

[0047] ;

[0048] in, The preset minimum number of neighboring points; The β value for scene adaptation coefficients increases sequentially for station scenes, plain line scenes, bridge scenes, and tunnel scenes; a maximum upper limit is also set for the k value. ;round[] represents the rounding function.

[0049] A non-transitory computer-readable storage medium storing a computer program that, when executed by a processor, implements the multi-dimensional timing rate quality defect detection method for 5G-R train control security services as described above.

[0050] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the multi-dimensional timing rate quality defect detection method for 5G-R train control security services as described above.

[0051] The beneficial technical effects of this invention are as follows:

[0052] (1) Significantly improve the detection rate of poor train control safety service speed;

[0053] By using weighted distance calculation based on the security level of train control services and global relative density correction, this invention can accurately capture the linked abnormal changes of key rate-related indicators such as average number of users, downlink traffic, CQI, and TA, and achieve early warning of rate degradation trends. Actual test data shows that for low-density clusters of hidden rate anomaly samples in tunnel scenarios (simultaneous occurrence of a surge in the number of users, a decrease in CQI, and a reduction in TA), traditional fixed threshold schemes cannot identify them (because the indicators do not exceed the fixed thresholds), while the method of this invention successfully detects them, significantly improving the rate quality defect detection rate.

[0054] (2) Significantly reduce reliance on manual labor and support large-scale rate quality analysis across the entire line;

[0055] Through scene-aware clustering and two-level pruning mechanisms, this invention can preemptively eliminate more than 90% of normal samples without rate quality defects, significantly reducing the computational load of subsequent anomaly detection algorithms. At the same time, the scene-adaptive dynamic matching function for the number of neighboring points enables automatic parameter adaptation in multiple scenarios, eliminating the need for manual parameter tuning for each scenario. It can support the automated rate quality defect analysis needs of thousands of cells and millions of samples per hour across the entire 5G-R line.

[0056] (3) Perfectly adapts to the multi-scenario rate data distribution characteristics of 5G-R;

[0057] This invention performs clustering processing according to four scenarios: stations, plains, bridges, and tunnels, ensuring that samples within the same subset have consistent rate-related data distribution characteristics. Through scene-adaptive dynamic matching of the number of neighboring points and differentiated settings of scene adaptation coefficients, each scenario can obtain optimal detection parameters, effectively avoiding the problem of decreased rate quality detection accuracy caused by the mixing of different data distribution characteristics.

[0058] (4) Multi-indicator linkage analysis to accurately locate the root cause of poor rate quality;

[0059] This invention employs multi-dimensional weighted distance calculation, comprehensively considering the correlation between wireless connection success rate, drop rate, handover success rate, average number of users, downlink traffic, average CQI, and average TA indicators. It can identify rate anomalies caused by the linkage of multiple indicators. For example, when a sudden increase in the number of users leads to a decrease in the rate per user, CQI degradation leads to a decrease in modulation and coding level, and abnormal changes in TA reflect abnormal user location distribution, these factors together lead to poor user rate quality. This invention can accurately locate the problem and overcome the limitations of traditional single-dimensional independent judgment.

[0060] (5) The output results are intuitive and usable;

[0061] The final output of the rate quality assessment results includes abnormal cell identifier, abnormal occurrence time, associated train control rate index, degradation level, and quantitative value of rate abnormality, which can directly provide data support and decision guidance for 5G-R network rate quality management, parameter optimization, and coverage adjustment. Attached Figure Description

[0062] Figure 1 This is a schematic diagram of the overall process of the intelligent detection method for multi-dimensional timing rate quality defects in 5G-R train control security services provided in the embodiments of this application. Detailed Implementation

[0063] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0064] Terminology Definition

[0065]

[0066] Example 1

[0067] Figure 1This is a flowchart illustrating the multi-dimensional timing rate quality defect detection method for 5G-R train control security services provided in this embodiment. The multi-dimensional timing rate quality defect detection method for 5G-R train control security services provided in this embodiment includes the following steps:

[0068] Step 101: Data Collection: Obtain hourly traffic statistics reports of target coverage cells along the 5G-R railway line, as well as network parameter configuration data of the corresponding cells after each network parameter modification;

[0069] Specifically, hourly traffic statistics reports of target coverage cells along the 5G-R network are extracted through the northbound interface, along with network parameter configuration data for the corresponding cells after each network parameter modification. The traffic statistics report data includes core indicators of train control services, coverage quality indicators, and traffic volume indicators. Among them, the core indicators of train control services include wireless connection rate, drop rate, and handover success rate; coverage quality indicators include average CQI and average TA; and traffic volume indicators include average number of users and downlink traffic. The average number of users, downlink traffic, average CQI, and average TA are core indicators related to user rate and together determine the 5G-R network user transmission rate.

[0070] Step 102, Data splicing: Using the unique identifier of the cell as the association basis, the call statistics report data and the network parameter configuration data are spliced ​​to generate a cell-level multidimensional time series dataset;

[0071] Specifically, using the cell name, which can uniquely identify the cell and has an extremely low modification frequency, as the core association key field, the traffic statistics report data and network parameter configuration data of the same cell at the same time dimension are concatenated by row dimension to construct a cell-level multidimensional time series dataset with cell identifier and time as dual indexes, including traffic statistics indicators (wireless connection rate, drop rate, handover success rate, average number of users, downlink traffic, average CQI, average TA) and network configuration parameter fields.

[0072] Step 103, Scene-aware clustering and two-level data pruning: Based on the railway geographical scene attributes, the multidimensional time series dataset is clustered according to the scene to generate sub-datasets corresponding to different railway scenes; then, through a two-level pruning mechanism of railway operation and maintenance rule first-level pruning and improved isolation forest second-level pruning, normal samples without rate quality difference risk in each sub-dataset are removed to generate a candidate sample set to be detected.

[0073] Specifically, the scene-aware clustering and two-level data pruning in step 103 include:

[0074] S1031, Scene-aware clustering: According to the engineering parameter labels of the communities along the railway line, the multidimensional time series dataset is divided into four types of scene sub-datasets: station scene, plain line scene, bridge scene, and tunnel scene.

[0075] S1032, Level 1 Pruning (Rule-based Pruning): Rule filtering is performed based on the qualified thresholds set in the "5G-R Railway Mobile Communication System Operation and Maintenance Regulations".

[0076] The specific acceptable thresholds are: wireless connection success rate ≥ 99.5%, drop rate ≤ 0.1%, and handover success rate ≥ 99%. Samples that simultaneously meet all three criteria and have no manually initiated network configuration parameter changes during the corresponding time period are directly judged as risk-free normal samples and are removed. Manual parameter changes can cause temporary fluctuations in network performance, and even if the current indicators are acceptable, there may be potential quality degradation risks. Therefore, these samples need to be retained for subsequent testing. Samples with no parameter changes and acceptable indicators have stable network conditions and no quality degradation risks.

[0077] S1033, Secondary Pruning (Rapid Pre-screening): For the remaining samples after primary pruning, an improved isolation forest algorithm is used for rapid pre-screening. The number of base classifiers in the isolation forest is set to 100, the maximum sample size per tree is set to 256, and the isolation depth threshold is set to 1.2 times the average isolation depth. Samples with an isolation depth greater than this threshold are considered normal samples and are discarded; only abnormal candidate samples with an isolation depth less than or equal to the threshold are retained.

[0078] The above pruning operations are performed on the four scenario subsets respectively, and then merged to generate the final candidate sample set to be detected. This step, through two-level pruning, can eliminate more than 90% of risk-free normal samples in advance.

[0079] Step 104, Scenario-based anomaly identification guided by train control safety:

[0080] S1041. Data Standardization Processing: The candidate sample set to be detected is standardized by removing non-numerical fields and eliminating the differences in the units of different indicators to obtain a standardized multidimensional dataset. The standardized multidimensional dataset is then deduplicated to remove completely duplicated samples. Specifically, for cases where there are 3 or more duplicate points in the k-distance neighborhood of a sample, a minimum reachable distance threshold is set for the duplicate points. This threshold is the 1st quantile of the weighted Euclidean distance of all sample points in the subset of the scene to avoid an infinite anomaly in the reachability density calculation.

[0081] The scene-adaptive dynamic matching method for the number of nearest neighbors k is as follows: For each scene subset, the optimal number of nearest neighbors k in that scene is dynamically calculated based on the number of samples n and the coefficient of variation cv. The calculation formula is as follows:

[0082] ;

[0083] In this embodiment, the minimum number of neighboring points 5; The scene adaptation coefficient β is set as follows: 0.2 for station scenes, 0.3 for plain line scenes, 0.4 for bridge scenes, and 0.5 for tunnel scenes; a maximum upper limit is also set for the k value. ;round[] represents the rounding function.

[0084] In addition, by combining the analytic hierarchy process with the experience of 5G-R railway operation and maintenance experts, weights were assigned to the call traffic statistics indicators, thereby obtaining the weighted Euclidean distance between sample points.

[0085] It should be noted that the total weight of the core indicators of train control business is greater than the total weight of the coverage quality indicators, which is greater than the total weight of the call volume indicators, and the total weight of the core indicators of train control business is not less than 0.7.

[0086] In this embodiment, the weights are allocated as follows: wireless connection success rate 0.25, drop rate 0.25, handover success rate 0.2, average CQI 0.1, average TA 0.1, average number of users 0.05, and downlink traffic 0.05;

[0087] For m-dimensional sample points P and O, the weighted Euclidean distance is calculated as follows:

[0088] ;

[0089] Where m is the indicator dimension of the standardized sample set. , Let P and O be the numerical values ​​of sample points P and O on the i-th feature dimension, respectively. Let be the weight coefficient corresponding to the i-th dimension index, and satisfy . ;

[0090] S1042. Calculation of k-distance neighborhood and reachability distance: For the target sample point P, sort its weighted Euclidean distances with all other sample points in ascending order, and take the distance between the k-th nearest neighbor and P as the k-distance of point P. All weighted Euclidean distances to point P are less than or equal to 1. The sample points constitute the k-distance neighborhood of point P. ;

[0091] For a sample point P and a sample point O within its k-distance neighborhood, the reachable distance of P relative to O is... for:

[0092] ;

[0093] The physical meaning of the reachability distance of P relative to O is: if P is within the k-distance neighborhood of O, then the k-distance of O is used as the reachability distance to avoid density calculation errors caused by excessively close distances.

[0094] S1043. Calculation of the original local reachability density: The original local reachability density of sample point P. It is the reciprocal of the average reachable distance from P to all sample points in its k-distance neighborhood, and the formula is:

[0095] ;

[0096] Local reachability density reflects the local data density around sample point P; the larger the value, the denser the data around that point.

[0097] S1044, Global Relative Density Correction: Calculate the global average reachability density of the scene subset containing sample point P. This refers to the average of the original locally reachable densities of all sample points in this scenario. A global relative density correction factor α(P) is introduced to correct the original locally reachable density, addressing the issue of missed detections caused by insignificant local density differences in low-density cluster scenarios.

[0098] ;

[0099] Corrected local reachable density for:

[0100] ;

[0101] The correction amplifies the density difference between abnormal and normal points in low-density clusters, thus improving the detection rate of hidden anomalies.

[0102] S1046. Calculation of the quantification value of the rate anomaly: Based on the corrected local reachability density, calculate the quantification value of the rate anomaly of sample point P. The formula is:

[0103] ;

[0104] The quantitative value of the rate anomaly of sample point P represents the degree of density difference between sample point P and its neighboring points: the closer the value is to 1, the more normal it is, and the larger the value is, the higher the degree of anomaly.

[0105] Step 105, Outlier Detection: A dual-threshold adaptive detection mechanism is adopted, using a fixed threshold as the primary threshold and a proportional threshold as a secondary threshold.

[0106] Main judgment rule: The preset fixed anomaly judgment threshold is 1.5. Sample points with a rate anomaly degree quantification value ≥1.5 are marked as abnormal samples. This threshold is determined based on the theoretical basis of the LOF algorithm. The closer the value is to 1, the more normal the sample is. A value greater than 1 indicates that there is a difference from the density of the neighborhood.

[0107] Dynamic adjustment mechanism: Supports dynamic adjustment of the judgment criteria through the abnormal point ratio threshold (default 5%). That is, when the abnormal point ratio judged by the fixed threshold deviates from the range of operation and maintenance experience, the top N% of sample points ranked by the quantitative value of the abnormal rate are marked as abnormal samples.

[0108] Priority: Fixed threshold has higher priority than proportional threshold. The proportional threshold is only used for fine-tuning in special scenarios.

[0109] Technical justification: The dual-threshold mechanism balances the scientific validity of the algorithm with the practicality of engineering. The fixed threshold ensures consistency of judgment standards across different scenarios; the proportional threshold can be flexibly adjusted according to the network load characteristics of different lines and time periods, avoiding misjudgments or omissions caused by special data distributions, and is more suitable for the complex and ever-changing operation and maintenance needs of the entire 5G-R railway line.

[0110] Step 106: Rate quality difference judgment and result output;

[0111] In this invention, the anomaly detection process and the rate quality defect determination process are completely unified: the samples marked as abnormal in step 105 are the samples with rate quality defects; the specific determination and output process is as follows:

[0112] 1. Abnormal sample extraction: Extract all samples marked as abnormal from the abnormal detection results in step 105 to establish a rate-quality poor sample set;

[0113] 2. Multi-dimensional correlation analysis: Each abnormal sample is correlated with the corresponding cell identifier, the time of the abnormality, and the correlation rate indicators (average number of users, downlink traffic, average CQI, average TA, etc.) to form a complete record of poor quality events.

[0114] 3. Deterioration level classification: based on the quantification value of the degree of rate anomaly. Poor quality events are classified into three levels:

[0115] Slight degradation: 1.5 ≤ Score < 2.0;

[0116] Moderate degradation: 2.0 ≤ Score < 3.0;

[0117] Severe deterioration: Score≥3.0;

[0118] The final output is a standardized 5G-R network rate quality assessment report, which includes: abnormal cell identifier, time of abnormal occurrence, associated control rate indicators, degradation level, and quantitative value of rate abnormality. This report provides data support and decision-making guidance for 5G-R network rate quality management, parameter optimization, and coverage adjustment.

[0119] As an example, in this embodiment, a set of simulation data extracted from a 5G-R network is used for verification. The indicators used are only for the calculation process and are not all the indicators used in this patent.

[0120] 1) Raw data

[0121] To demonstrate the detection of local anomalies, a set of data including normal samples, low-density clustered hidden rate anomaly samples, and obvious rate anomaly samples from tunnel and plain railway scenes are used as examples. The original data is shown in the table below:

[0122]

[0123] 2) Data standardization

[0124] The data is standardized using the following formula:

[0125] ;

[0126] Where x is the original data, For standardized data, This represents the global mean of the corresponding indicator. This represents the overall standard deviation of the corresponding indicator.

[0127] Taking the average number of users as an example, the global average Overall standard deviation The standardized value of sample P1 is calculated as follows:

[0128] ;

[0129] The standardized multidimensional sample set is shown in the table below:

[0130]

[0131] 3) Calculation of anomaly determination results

[0132] To objectively verify the improvement effect of this invention, the indicator weight allocation rules are first explained: The complete detection process includes 7 indicators, with the following weight allocation: average number of users 0.05, downlink traffic 0.05, average CQI 0.1, average TA 0.1, wireless connection success rate 0.25, drop rate 0.25, and handover success rate 0.2. This calculation example only simplifies and retains 4 core rate-related indicators (average number of users, downlink traffic, average CQI, and average TA). Therefore, it is necessary to normalize them according to the original weight ratios so that the sum of the weights satisfies the constraint condition of the weighted Euclidean distance formula (the sum of the weights is 1). The normalization calculation process is as follows:

[0133] The sum of the original weights for the four indicators = 0.05 + 0.05 + 0.1 + 0.1 = 0.3

[0134] Normalized weight of average number of users = 0.05 ÷ 0.3 ≈ 0.167

[0135] Downlink traffic normalization weight = 0.05 ÷ 0.3 ≈ 0.167

[0136] Average CQI normalized weight = 0.1 ÷ 0.3 ≈ 0.333

[0137] Average TA normalized weight = 0.1 ÷ 0.3 ≈ 0.333

[0138] After the four terms are normalized, the total weight is 1. Substitute this weighted Euclidean distance formula in step S1041 to calculate the weighted distance between any two sample points. This weighted distance is the basis for subsequent k-distance neighborhood delineation, reachability distance calculation, original local reachability density solution, global relative density correction, and rate anomaly quantification calculation, and runs through all anomaly detection calculation steps in step 104.

[0139] The general LOF algorithm was used as the benchmark for comparison. The general LOF uses standard equal weighting, and the weights of the four rate-related core indicators are all 0.25. The two algorithms use the same standardized samples and the same number of neighboring points k=3. The only difference is the weighting strategy and the density correction logic.

[0140] Following the complete calculation process of steps 104 and 105, the anomaly detection results of the two algorithms are shown in the table below:

[0141]

[0142] The raw data includes normal samples from tunnel and plains line scenarios, low-density clusters with concealed rate anomalies (P5), and obvious rate anomalies (P10). P5 represents a tunnel cell where traffic volume and channel quality simultaneously deteriorate: the number of users surges from an average of 12 to 38, downlink traffic surges from 1.1 GB to 8.5 GB, the average CQI drops from 10.0 to 6.1, and the average TA drops from 8.0 to 4.2. These interconnected changes in indicators suggest a severe rate quality deterioration in the cell—the surge in users leads to resource contention, the decrease in CQI results in a reduction in modulation and coding levels, and the change in TA reflects abnormal user location distribution, ultimately leading to a severe decrease in single-user rate. However, none of these indicators exceed traditional fixed thresholds, making them undetectable by conventional methods.

[0143] The calculation results show that:

[0144] P5 sample: Using the method of this invention, the quantification value of the rate anomaly is 2.18, which exceeds the fixed anomaly judgment threshold of 1.5, and is marked as an abnormal sample; the corresponding quality defect judgment result is that a moderate rate quality defect occurred in tunnel community A at 22:00 on 2025-09-01, which belongs to the low-density cluster hidden rate quality defect that affects the safety of train control and operation, which is consistent with the actual anomaly situation;

[0145] Normal samples (P1-P4, P6-P9): The quantitative values ​​of rate abnormality are all close to 1.0, which is lower than the fixed abnormality judgment threshold of 1.5, and are judged as normal samples without quality defects.

[0146] P10 sample: The quantification value of the rate anomaly is 2.51, which exceeds the fixed anomaly judgment threshold of 1.5, and it is marked as an abnormal sample; the corresponding quality defect judgment result is that the plain cell B experienced severe rate quality defects at 22:00 on 2025-09-01.

[0147] The verification results fully demonstrate that the present invention can effectively capture the low-density cluster hidden rate quality defects that cannot be identified by traditional fixed threshold schemes, and significantly improve the detection accuracy of train control safety-related rate anomalies.

[0148] The industrial applicability of this invention is as follows:

[0149] This invention is developed using Python and employs machine learning modules such as Scikit-learn for big data processing. It combines the characteristics of real-world 5G-R network data to meet the rate quality degradation analysis needs of high-speed 5G-R scenarios. This invention can be deployed on 5G-R network operation and maintenance platforms as the core algorithm module for detecting rate quality degradation in train control services. It can also be extended to optimization scenarios such as automatic multi-parameter processing and performance prediction in 5G-R networks, demonstrating broad industrial practical value.

[0150] Furthermore, the present invention adopts the following technical solution:

[0151] A non-transitory computer-readable storage medium storing a computer program that, when executed by a processor, implements the multi-dimensional timing rate quality defect detection method for 5G-R train control security services as described above.

[0152] Furthermore, the present invention adopts the following technical solution:

[0153] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the multi-dimensional timing rate quality defect detection method for 5G-R train control security services as described above.

[0154] From the above description of the embodiments, those skilled in the art will clearly understand that the facilities of the present invention can be implemented using software plus necessary general-purpose hardware platforms. Embodiments of the present invention can be implemented using existing processors, or by dedicated processors used for this or other purposes for suitable systems, or by hardwired systems. Embodiments of the present invention also include non-transitory computer-readable storage media, comprising machine-readable media for carrying or having machine-executable instructions or data structures stored thereon; such machine-readable media can be any available medium accessible by a general-purpose or special-purpose computer or other machine with a processor. For example, such machine-readable media can include RAM, ROM, EPROM, EEPROM, CD-ROM or other optical disc storage, disk storage or other magnetic storage devices, or any other medium that can be used to carry or store the required program code in the form of machine-executable instructions or data structures and is accessible by a general-purpose or special-purpose computer or other machine with a processor. When information is transmitted or provided to a machine via a network or other communication connection (hardwired, wireless, or a combination of hardwired and wireless), that connection is also considered a machine-readable medium.

[0155] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after such changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A multi-dimensional timing rate quality defect detection method for 5G-R train control security services, characterized in that, Includes the following steps: Data Acquisition: Hourly traffic statistics reports of target coverage cells along the 5G-R railway line are acquired, along with network parameter configuration data for the corresponding cells after each network parameter modification. The traffic statistics reports include core indicators for train control services, coverage quality indicators, and traffic volume indicators. Among these, core indicators for train control services include wireless connection rate, drop rate, and handover success rate; coverage quality indicators include average CQI and average TA; and traffic volume indicators include average number of users and downlink traffic. Data splicing: Using the unique identifier of the cell as the association basis, the traffic statistics report data and network parameter configuration data are spliced ​​together to generate a cell-level multidimensional time series dataset; Scene-aware clustering and two-level data pruning: Based on the railway geographical scene attributes, the multidimensional time series dataset is clustered by scene to generate sub-datasets corresponding to different railway scenes; then, through a two-level pruning mechanism of first-level pruning of railway operation and maintenance rules and second-level pruning of improved isolation forest, normal samples without rate quality difference risk in each sub-dataset are removed to generate a candidate sample set to be detected. Scenario-based rate anomaly identification based on train control safety: For the candidate sample set to be detected, the rate anomaly degree of the sample points is optimized by calculating the weighted distance based on the train control business security level, the original local reachability density, the global relative density correction, and the quantification value of the rate anomaly degree. Anomaly detection: Based on the obtained quantitative value of the rate anomaly degree, an adaptive dual-threshold detection mechanism, with a fixed threshold as the primary threshold and a proportional threshold as the secondary threshold, is used to detect anomalies. Output results: Perform multi-dimensional correlation analysis and degradation level classification on abnormal samples to generate a rate quality poor report that includes abnormal cells, abnormal time, correlation indicators, and degradation level.

2. The multi-dimensional timing rate quality defect detection method for 5G-R train control security services according to claim 1, characterized in that, For the candidate sample set to be detected, the method for optimizing the calculation of the rate anomaly degree of the sample points by weighted distance calculation based on train control service security level, global relative density correction, and quantification of rate anomaly degree is as follows: By combining the Analytic Hierarchy Process (AHP) with the experience of 5G-R railway operation and maintenance experts, weights are assigned to call volume statistics indicators to obtain the weighted Euclidean distance between sample points. Among them, the total weight of the core indicators of train control business is greater than the total weight of the coverage quality indicators, which is greater than the total weight of the call volume indicators, and the total weight of the core indicators of train control business is not less than 0.

7. The k-distance neighborhood of a sample point is calculated based on weighted Euclidean distance. And reachability distance, and then calculate the original local reachability density. ; Global relative density correction: Calculate the global average reachable density Calculate the correction factor α(P) to obtain the corrected local reachability density lrd'(P); ； ； Rate anomaly quantification: Based on the corrected local reachability density, calculate the rate anomaly quantification value of sample point P. The formula is: 。 3. The multi-dimensional timing rate quality defect detection method for 5G-R train control security services according to claim 1, characterized in that, The scene-aware clustering and two-level data pruning specifically include: Based on the engineering parameter labels of the communities along the railway line, the multidimensional time series dataset is divided into four sub-datasets: station scene, plain line scene, bridge scene, and tunnel scene. Level 1 pruning: Based on the qualified thresholds of wireless connection rate ≥99.5%, drop rate ≤0.1%, and handover success rate ≥99%, rule filtering is performed on the subset of data in each scenario to remove normal samples that meet all three conditions and have no manually initiated network-side configuration parameter change operations in the corresponding time period. Secondary pruning: For the remaining samples after primary pruning, an improved isolation forest algorithm is used for rapid pre-screening to remove normal samples whose isolation depth is greater than the preset isolation depth threshold; The above pruning operations were performed on the four types of scenario subsets respectively, and then merged to generate the final set of candidate samples to be detected.

4. The multi-dimensional timing rate quality defect detection method for 5G-R train control security services according to claim 1, characterized in that, Based on the obtained quantitative value of rate anomaly, the method for anomaly point determination using a dual-threshold adaptive determination mechanism, with a fixed threshold as the primary threshold and a proportional threshold as the secondary threshold, is as follows: The preset fixed anomaly judgment threshold is 1.5, and sample points with a rate anomaly degree quantification value ≥1.5 are marked as abnormal samples; It supports dynamically adjusting the judgment criteria based on the abnormal point ratio threshold. When the abnormal point ratio judged by the fixed threshold deviates from the range of operation and maintenance experience, the sample points ranked in the top N% of the quantitative value of the abnormal rate are marked as abnormal samples.

5. The multi-dimensional timing rate quality defect detection method for 5G-R train control security services according to claim 2, characterized in that, For m-dimensional sample points P and O, the weighted Euclidean distance is calculated as follows: ； Where m is the indicator dimension of the standardized sample set. , Let P and O be the numerical values ​​of sample points P and O on the i-th feature dimension, respectively. Let be the weight coefficient corresponding to the i-th dimension index, and satisfy . .

6. The multi-dimensional timing rate quality defect detection method for 5G-R train control security services according to claim 1, characterized in that, In the scenario-based rate anomaly identification based on train control safety, the following steps are also included: standardizing the candidate sample set to be detected, removing non-numerical fields, eliminating the dimensional differences of different indicators, obtaining a standardized multidimensional dataset, and performing deduplication preprocessing on the standardized multidimensional dataset to remove completely duplicate samples.

7. The multi-dimensional timing rate quality defect detection method for 5G-R train control security services according to claim 6, characterized in that, When performing deduplication preprocessing on the standardized multidimensional dataset and removing completely duplicated samples, a minimum reachable distance threshold is set for the case where there are 3 or more duplicate points in the k-distance neighborhood of the dataset. The minimum reachable distance threshold is the 1% quantile of the weighted Euclidean distance of all sample points in the sub-dataset of this scenario.

8. The multi-dimensional timing rate quality defect detection method for 5G-R train control security services according to claim 7, characterized in that, In the scenario-based rate anomaly point identification based on train control safety guidance, it also includes: scenario-adaptive dynamic matching of the number of neighboring points, specifically including: For each scene subset, based on the number of samples n and the coefficient of variation cv, the optimal number of nearest neighbors k for that scene is dynamically calculated using the following formula: ； in, The preset minimum number of neighboring points; The β value for scene adaptation coefficients increases sequentially for station scenes, plain line scenes, bridge scenes, and tunnel scenes; a maximum upper limit is also set for the k value. ;round[] represents the rounding function.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the multi-dimensional timing rate quality defect detection method for 5G-R train control security services as described in any one of claims 1 to 8.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the multi-dimensional timing rate quality defect detection method for 5G-R train control security services as described in any one of claims 1 to 8.

Images