Rru traffic anomaly detection method and device, electronic equipment, medium and program product

Abstract

The application relates to the technical field of wireless core networks, and provides an RRU traffic anomaly detection method and device, electronic equipment, a medium and a program product, which comprise the following steps: acquiring position data, basic parameters, coverage performance indexes and traffic data of a cell; determining the classification of the cell according to the position data, the basic parameters, the coverage performance indexes and the traffic data, wherein the cell comprises at least one RRU; performing regression fitting on the classification of the cell, user scheduling characteristics of the cell and characteristics of the RRU based on a preset regression prediction algorithm to obtain regression coefficients of the RRU; determining the fitted traffic of each cell according to the regression coefficients of the RRU; and determining abnormal RRUs according to the traffic data and the fitted traffic. The application considers the influence of cell coverage range and user behavior on the traffic data of the cell, and no longer observes the aggregated indexes of the cell granularity and the single RRU traffic indexes, but comprehensively considers the indexes of the cell granularity and the indexes of the RRU granularity, so that the identification precision and the fault discovery rate can be improved.

Description

Technical Field

[0001] This application relates to the field of wireless core network technology, and in particular to a method, device, electronic device, medium and program product for detecting abnormal RRU traffic. Background Technology

[0002] The passive indoor distribution system uses the remote radio unit (RRU) of a high-power distributed base station as the signal source. The radio frequency signal is split by a power divider and distributed to each antenna and leaky cable installed in various areas of the building via feeders, leaky cables, and other components. This enables wireless network coverage in areas such as building floors, zones, and tunnels, solving indoor wireless communication problems, achieving fixed-point deep coverage in weak signal and blind spots, ensuring the easy management, scalability, and high reliability of the 5G network, meeting the high-speed data transmission services of 5G base stations, helping macro base stations to share traffic load, and combining with edge computing and other technologies to meet the fragmented needs of vertical industry applications.

[0003] Because the operation and maintenance of RRUs are highly dependent on manual labor, data collection and statistics at the RRU granularity level are more difficult, and the user distribution and change models of indoor sites also vary significantly. Timely detection of RRU traffic anomalies and network failures has always been a challenge for operation and maintenance. Existing technologies observe aggregated indicators at the cell level and individual RRU traffic indicators, which is a relatively simple approach with a low failure detection rate. Summary of the Invention

[0004] This application provides a method, device, electronic equipment, medium, and program product for detecting abnormal RRU traffic, which solves the shortcomings of existing technologies that rely on a single method for observing aggregated indicators at the cell level and individual RRU traffic indicators, resulting in a low fault detection rate.

[0005] Firstly, this application provides a method for detecting RRU traffic anomalies, including:

[0006] The location data, basic parameters, coverage performance indicators, and traffic data of a cell are obtained. Based on the location data, basic parameters, coverage performance indicators, and traffic data, the classification of the cell is determined. The cell includes at least one RRU.

[0007] The regression coefficients of the RRU are obtained by performing regression fitting on the cell classification, the user scheduling characteristics of the cell, and the characteristics of the RRU based on a preset regression prediction algorithm.

[0008] Based on the regression coefficients of the RRU, the fitted traffic of each cell is determined;

[0009] Based on the traffic data and the fitted traffic, abnormal RRUs are identified.

[0010] In one embodiment, the regression fitting of the cell classification, the user scheduling characteristics of the cell, and the characteristics of the RRU based on a preset regression prediction algorithm to obtain the regression coefficients of the RRU includes:

[0011] The cell classification is used as the weight, the cell user scheduling features are used as the target data, and the RRU features are used as the information samples. These are substituted into the regression coefficient function of the regression prediction algorithm to construct a regression coefficient model corresponding to each cell classification and each RRU.

[0012] The regression coefficients of the RRU are obtained based on the regression coefficient model.

[0013] In one embodiment, the user scheduling characteristics of the cell include at least one or more of RRC connection count, traffic data, RSRP, and SINR, and the characteristics of the RRU include at least one or more of downlink data throughput, uplink data throughput, number of uplink PRBs, number of downlink PRBs, average noise, uplink data transmission duration, and downlink data transmission duration.

[0014] In one embodiment, acquiring cell location data, basic parameters, coverage performance indicators, and traffic data, and determining the cell classification based on the location data, basic parameters, coverage performance indicators, and traffic data, includes:

[0015] Obtain the cell's location data, basic parameters, coverage performance indicators, and traffic data;

[0016] Input the location data, the basic parameters, the coverage performance indicators, and the traffic data into a pre-trained classification model to obtain the classification of the cell output by the classification model;

[0017] The classification model is obtained by training a convolutional neural network using historical location data, historical basic parameters, historical coverage performance indicators, and historical traffic data.

[0018] In one embodiment, after acquiring the cell's location data, basic parameters, coverage performance metrics, and traffic data, the method further includes:

[0019] The traffic data is split based on a preset time granularity to obtain a traffic data unit corresponding to each time granularity.

[0020] The flow slope is determined based on the flow data units corresponding to two adjacent time granularities.

[0021] Based on the maximum and minimum values ​​of the traffic data, determine the peaks and troughs of the traffic data changes;

[0022] Correspondingly, the input of the location data, the basic parameters, the coverage performance indicators, and the traffic data to a pre-trained classification model to obtain the classification of the cell output by the classification model includes:

[0023] Input the location data, the basic parameters, the coverage performance index, the traffic slope, and the peaks and troughs of the variation into a pre-trained classification model to obtain the classification of the cell output by the classification model.

[0024] In one embodiment, the basic parameters include at least one or more of the following: rated power, reference signal transmit power, number of antennas, bandwidth, and subcarrier spacing. The coverage performance indicators include coverage volume and coverage density, wherein the coverage volume is the volume of the multidimensional indoor building covered by the cell, determined based on the floor spacing and the boundary information of the multidimensional indoor building.

[0025] In one embodiment, determining the fitted traffic of each cell based on the regression coefficients of the RRU includes:

[0026] Based on the regression coefficients of the RRUs, determine the fitted flow rate corresponding to each RRU;

[0027] The fitted flow rate of each cell is determined based on the fitted flow rate corresponding to all RRUs.

[0028] In one embodiment, determining the abnormal RRU based on the traffic data and the fitted traffic includes:

[0029] Based on the fitted flow rate, determine the mean and variance of the fitted flow rate;

[0030] Based on the flow data, the fitted flow mean, and the fitted flow variance, determine the expected deviation between the flow data and the fitted flow.

[0031] Based on the deviation from expectation and the preset expectation threshold, abnormal RRUs are determined.

[0032] Secondly, this application also provides an RRU flow anomaly detection device, comprising:

[0033] A classification module is used to acquire location data, basic parameters, coverage performance indicators and traffic data of a cell, and determine the classification of the cell based on the location data, the basic parameters, the coverage performance indicators and the traffic data, wherein the cell includes at least one RRU;

[0034] The regression module is used to perform regression fitting on the cell classification, the user scheduling characteristics of the cell, and the characteristics of the RRU based on a preset regression prediction algorithm, so as to obtain the regression coefficients of the RRU.

[0035] The determination module is used to determine the fitted traffic of each cell based on the regression coefficient of the RRU;

[0036] The detection module is used to determine abnormal RRUs based on the traffic data and the fitted traffic.

[0037] Thirdly, this application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement any of the RRU traffic anomaly detection methods described above.

[0038] Fourthly, this application provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the RRU traffic anomaly detection method as described above.

[0039] Fifthly, the present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the RRU traffic anomaly detection method as described above.

[0040] The RRU traffic anomaly detection method, apparatus, electronic device, medium, and program products provided in this application acquire cell location data, basic parameters, coverage performance indicators, and traffic data. Based on the location data, basic parameters, coverage performance indicators, and traffic data, the classification of the cell is determined, and each cell includes at least one RRU. A preset regression prediction algorithm is used to perform regression fitting on the cell classification, user scheduling characteristics of the cell, and characteristics of the RRU to obtain the regression coefficients of the RRU. Based on the regression coefficients of the RRU, the fitted traffic of each cell is determined. Based on the traffic data and the fitted traffic, abnormal RRUs are identified. This application classifies cells using location data, basic parameters, coverage performance indicators, and traffic data. Based on cell-level and RRU-level indicators, fitted traffic is obtained for each cell. Then, based on the traffic data and the fitted traffic, abnormal RRUs are identified. This approach considers the impact of cell coverage and user behavior on cell traffic data, and instead of observing through a combination of cell-level aggregate indicators and single RRU traffic indicators, it integrates cell-level and RRU-level indicators, thereby improving identification accuracy and fault detection rate. Attached Figure Description

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

[0042] Figure 1 This is a flowchart illustrating the RRU traffic anomaly detection method provided in this application;

[0043] Figure 2 This is a schematic block diagram of the RRU flow anomaly detection device provided in this application;

[0044] Figure 3 This is a schematic diagram of the structure of the electronic device provided in this application. Detailed Implementation

[0045] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions 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, 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.

[0046] Existing fault identification methods at the RRU granularity place more emphasis on trend prediction and pattern summarization in the application time-series prediction part. However, in addition to time-series changes, the more important influencing factors of communication network indicators are user behavior and coverage scenarios. This is especially true for passive base stations covering indoor areas, where the tidal statistics of access users and the perception of coverage characteristics are more important.

[0047] Currently, the coverage area of ​​passive indoor distribution systems is uncontrollable, and data collection is easy, resulting in low efficiency in identifying corresponding faults. When similar problems occur, it is difficult to associate the corresponding indicators and trace the problem. Furthermore, the various types of indicators on the network side have complex structural relationships, which greatly interferes with and affects problem identification.

[0048] To this end, this application prioritizes the classification of cells and RRUs in different coverage scenarios by combining geographical features such as cell configuration, cell distribution, and coverage type. Based on a joint regression prediction algorithm, it constructs regression coefficient models with centralized granularity (cell granularity) and decentralized granularity (RRU granularity). By classifying and predicting traffic fluctuations caused by users and geographical features, it identifies abnormal time points and abnormal RRU units, thereby achieving more accurate early detection and early warning of faults in passive RRU systems.

[0049] The technical solution of this application will be described in detail below.

[0050] Figure 1 This is a flowchart illustrating the RRU traffic anomaly detection method provided in this application, as shown below. Figure 1 As shown, this application provides a method for detecting RRU traffic anomalies, including:

[0051] Step S100: Obtain the location data, basic parameters, coverage performance indicators, and traffic data of the cell; determine the classification of the cell based on the location data, the basic parameters, the coverage performance indicators, and the traffic data; the cell includes at least one RRU.

[0052] Specifically, the cell is a 5G cell, which is a hypercell (with the same logical cell). Its traffic radio channel (TRP) can be independently scheduled with no capacity loss, effectively ensuring user experience.

[0053] The location data of a cell is its latitude and longitude, the basic parameters are its engineering parameter configurations, which are used to characterize the basic performance of the cell, and the coverage performance indicators are used to characterize the signal coverage of the cell.

[0054] The cell's location data, basic parameters, and coverage performance indicators are structured invariant data, while the traffic data is structured time-series data. It's clear that this application differs from existing feature classification methods by adding two dimensions: structured invariant data and structured time-series data. It also considers anomalies in indicators caused by periodicity and stationarity, thus completing the basic classification of cells in terms of coverage type and user model. Specifically, this application can classify cells into periodic cells, aperiodic cells, stationary cells, and non-stationary cells. Wireless scenarios are typical dynamic directed heterogeneous graph models. Signaling transmission between sites exhibits typical directionality, and user movement leads to changes in node structure and dynamic connections. Determining a structured model classification can effectively improve data application and training accuracy, enabling data augmentation.

[0055] Each cell in this application includes at least one RRU. The cell is treated as the master node, and the RRUs contained in the cell are treated as child nodes. That is, the RRUs are treated as a subset of the cell. The potential impact of different RRUs on the cell is analyzed to improve the accuracy of traffic anomaly detection.

[0056] Step S200: Based on a preset regression prediction algorithm, regression fitting is performed on the cell classification, the cell user scheduling characteristics, and the RRU characteristics to obtain the regression coefficients of the RRU;

[0057] Regression prediction algorithms are machine learning algorithms used to predict the numerical output of one or more continuous variables. They aim to establish a relationship model between input features and the output target to predict data. In a cell, user scheduling characteristics can be used as the dependent variable, and RRU characteristics can be used as independent variables. The user scheduling characteristics of a cell must at least include the cell's traffic data. Therefore, by fitting the relationship between the cell's user scheduling characteristics and the RRU characteristics, the cell's user scheduling characteristics can be predicted.

[0058] Step S300: Determine the fitted traffic of each cell based on the regression coefficients of the RRU. Specifically, the regression coefficients of the RRU can be used to characterize the relationship between the characteristics of the RRU and the user scheduling characteristics of the cell, and can be used to indicate the impact of the RRU characteristics on the cell's traffic data. By mapping the regression coefficients of the RRU to the characteristics of the RRU, the fitted traffic of each cell can be determined.

[0059] Step S400: Determine the abnormal RRU based on the traffic data and the fitted traffic.

[0060] Specifically, based on the deviation between the actual traffic data and the predicted fitted traffic, anomalies are identified by setting an expected threshold, thereby pinpointing the abnormal time point and the device ID of the abnormal RRU.

[0061] It is understood that this invention classifies cells using location data, basic parameters, coverage performance indicators, and traffic data. Based on cell-level indicators and RRU-level indicators, it obtains the fitted traffic for each cell. Then, based on the traffic data and the fitted traffic, it identifies abnormal RRUs. Taking into account the impact of cell coverage and user behavior on cell traffic data, it no longer observes using aggregated cell-level indicators and single RRU traffic indicators, but rather integrates cell-level indicators and RRU-level indicators, which can improve identification accuracy and fault detection rate.

[0062] Based on the above embodiments, as an optional embodiment, the step of acquiring cell location data, basic parameters, coverage performance indicators, and traffic data, and determining the cell classification based on the location data, the basic parameters, the coverage performance indicators, and the traffic data, includes:

[0063] Step S110: Obtain the cell's location data, basic parameters, coverage performance indicators, and traffic data;

[0064] The location data of a cell can be the latitude and longitude information of the cell. Specifically, the location data of a cell can be the latitude and longitude information of each indoor distributed antenna site or each antenna in the cell.

[0065] The basic parameters include at least one or more of the following: rated power, reference signal transmit power, number of antennas, bandwidth, and subcarrier spacing. The coverage performance indicators include coverage volume and coverage density, wherein the coverage volume is the volume of the multi-dimensional indoor buildings covered by the cell, which is determined based on the floor spacing and the boundary information of the multi-dimensional indoor buildings.

[0066] Step S120: Input the location data, the basic parameters, the coverage performance index, and the traffic data into the pre-trained classification model to obtain the classification of the cell output by the classification model;

[0067] The classification model is obtained by training a convolutional neural network using historical location data, historical basic parameters, historical coverage performance indicators, and historical traffic data.

[0068] Optionally, the classification model is a feedforward neural network consisting of convolutional layers, pooling layers, fully connected layers, and an output layer. The specific training process may include the following steps:

[0069] Input historical location data, historical basic parameters, historical coverage performance indicators, and historical traffic data into the convolutional layer to obtain the convolutional features output by the convolutional layer. The data types contained in the historical location data, historical basic parameters, historical coverage performance indicators, and historical traffic data correspond to the location data, basic parameters, coverage performance indicators, and traffic data, respectively. Therefore, the historical location data, historical basic parameters, and historical coverage performance indicators are structured invariant data, while the historical traffic data is structured time-series data.

[0070] Inputting convolutional features into a pooling layer yields a pooled feature sequence output by the pooling layer. Specifically, the pooling layer is used to downsample the output data of the convolutional layer, reducing data dimensionality while retaining important features. Feature extraction and feature combination can be performed by stacking multiple convolutional and pooling layers.

[0071] Pooled feature sequences are aggregated using fully connected layers. The aggregated features are then mapped to the output layer for classification prediction, and a loss function is calculated. The loss function measures the difference between the classification prediction result and the true label of the cell. Then, the gradient is calculated along the backward direction of the network using the backpropagation algorithm, and the weights of each layer, including the output layer, are updated based on the gradient. This gradual adjustment of the network parameters allows the network to achieve better performance on the training data.

[0072] It is known that the classification model is compatible with cell parameter configuration, hardware characteristics, coverage model dimensions, and trend flatness dimensions. It also takes into account the anomalies in indicators caused by periodicity and stability, and completes the basic classification of cells in terms of coverage type and user model.

[0073] The classification model can categorize cells into periodic cells, non-periodic cells, stable cells, and non-stable cells based on their location data, basic parameters, coverage performance indicators, and traffic data.

[0074] Understandably, this application uses deep learning to classify cell coverage scenarios at the level of coverage type and user model granularity, thus improving upon the shortcomings of traditional manual labeling and classification.

[0075] Based on the above embodiments, as an optional embodiment, after obtaining the cell's location data, basic parameters, coverage performance indicators, and traffic data, the method further includes:

[0076] Step S111: The traffic data is split based on a preset time granularity to obtain a traffic data unit corresponding to each time granularity. Specifically, the traffic data can be cell traffic data with a daily granularity. The preset time granularity is an hourly granularity, that is, the traffic data of 24 hours in a day is split into hourly traffic data units, and the traffic data unit is the traffic data of a cell in a certain hour.

[0077] Step S112: Determine the flow rate slope based on the flow data units corresponding to two adjacent time granularities; specifically, the flow rate slope can be calculated for each hour t. n Compared to the previous hour t n-1 The slope between the flow data units can also be calculated for each hour t. n With the next hour t n+1 The slope between the traffic data units.

[0078] Step S113: Based on the maximum and minimum values ​​of the traffic data, determine the peaks and troughs of the traffic data changes; specifically, the peaks and troughs of the cell traffic changes can be obtained by calculating traffic max - traffic min, thereby completing the full feature analysis of the cell.

[0079] Correspondingly, the input of the location data, the basic parameters, the coverage performance indicators, and the traffic data to a pre-trained classification model to obtain the classification of the cell output by the classification model includes:

[0080] Step S120: Input the location data, the basic parameters, the coverage performance index, the traffic slope, and the peaks and troughs of the change to the pre-trained classification model to obtain the classification of the cell output by the classification model.

[0081] Specifically, the structure and training steps of the classification model are the same as above, and will not be repeated here.

[0082] Understandably, the classification model provided in this application is compatible with cell parameter configuration, hardware features, coverage model dimensions, and trend flatness dimensions. It also takes into account the anomalies in indicators caused by periodicity and stability, and completes the basic classification of cells in terms of coverage type and user model, thus improving the problem of incomplete consideration in traditional manual annotation classification.

[0083] Based on the above embodiments, as an optional embodiment, the step of performing regression fitting on the cell classification, the user scheduling characteristics of the cell, and the characteristics of the RRU based on a preset regression prediction algorithm to obtain the regression coefficients of the RRU includes:

[0084] Step S210: The classification of the cell is used as the weight, the user scheduling feature of the cell is used as the target data, and the feature of the RRU is used as the information sample. These are substituted into the regression coefficient function of the regression prediction algorithm to construct a regression coefficient model corresponding to each classification of the cell and each RRU.

[0085] Traditional target management, Lasso algorithms, and ridge regression algorithms are unable to perform time-series-based prediction and regression evaluation of high-dimensional, multi-granular features. This application combines the Lasso algorithm in regression analysis with multiple dimensions, adding transfer learning to optimize the monotonicity and overfitting problems of the regression algorithm. Specifically, the Lasso algorithm (Least Absolute Shrinkage and Selection Operator) is a linear regression method for feature selection and regularization. It introduces an L1 regularization term into the ordinary least squares method, solving for model parameters by minimizing the sum of the loss function and the regularization term.

[0086] Optionally, the user scheduling characteristics of the cell include at least one or more of RRC connection count, traffic data, RSRP, and SINR, and the characteristics of the RRU include at least one or more of downlink data throughput, uplink data throughput, number of uplink PRBs, number of downlink PRBs, average noise, uplink data transmission duration, and downlink data transmission duration.

[0087] Specifically, this application decomposes the RRU as a subset of cells. Different RRUs may generate regression algorithms for cells, fusing cell classifications. Using cell classification as weights refers to constructing a cell classification matrix. The elements of the cell classification matrix represent the periodicity, aperiodicity, stationarity, and non-stationarity of cells, with the corresponding element taking a value of 1 and the non-corresponding element taking a value of 0. Multiplying different RRUs with the cell classification matrix yields the RRU classification coefficients.

[0088] The user scheduling characteristics of the cell include the number of RRC connections, the traffic data, RSRP and SINR, and the characteristics of RRU include downlink data throughput, uplink data throughput, number of uplink PRBs, number of downlink PRBs, average noise, uplink data transmission duration and downlink data transmission duration.

[0089] Specifically, the characteristics of the RRU are: the total downlink data throughput (MByte) transmitted by the TRP MAC layer of the Hyper Cell, the total uplink data throughput (MByte) received by the TRP MAC layer of the Hyper Cell, the average number of PRBs used in the downlink of the Hyper Cell, the average number of PRBs used in the uplink of the Hyper Cell, the average received interference noise per PRB in the uplink of the DU cell (dBm), the total duration of downlink data transmission (seconds) in the TRP MAC layer of the Hyper Cell, and the total duration of uplink data transmission (seconds) in the TRP MAC layer of the Hyper Cell.

[0090] After organizing the features of the dependent and independent variables, this application performs regression fitting based on the RRU classification coefficients and the trans-lasso algorithm. The trans-lasso algorithm is a method that combines transfer learning and the Lasso algorithm. In the traditional Lasso algorithm, each task learns a sparse model independently, while the trans-lasso algorithm achieves cross-task feature selection and parameter sharing by utilizing the similarity and correlation between different tasks.

[0091] The trans-lasso algorithm considers transfer learning in high-dimensional linear regression models, with the target model shown below:

[0092]

[0093] in, For the granularity of the community, For RRU granularity index, The target regression coefficient, For independently distributed random noise, such that

[0094] Observe K (preset integer) samples from the auxiliary model. The auxiliary model is shown below:

[0095]

[0096] Where, ω (k) It is the true regression coefficient of the kth study.

[0097] Let δ (k) =β-ω (k) Represents ω (k) The difference between ω and β (k) The sparsity of the difference between ω and β represents the information level of the k-th auxiliary study, assuming that the information auxiliary sample is a sample with a sufficient coefficient of difference, i.e., ω. (k) The difference between the sample and β is mostly zero. Let set A represent the information auxiliary sample, as shown below:

[0098] A = {1≤k≤K:||δ (k) ||0≤h};

[0099] Where h is a preset value.

[0100] For a vector Several norms are defined as follows: ||α||0 represents the number of non-zero elements in α.

[0101] Input target data (x) (0) Y (0) ) and information-assisted sample data {X (k) Y (k)} k∈A Output The calculation formula is as follows:

[0102]

[0103] in,

[0104] This application introduces the theoretical difference between the estimated value and the true value into the function γ. (k) =β-ω (k) -δ (k) To represent the difference between w+δ and the regression coefficients, the regression coefficient function is updated as follows:

[0105]

[0106]

[0107]

[0108] By substituting the user scheduling characteristics and RRU characteristics of each cell into the above regression coefficient function, the regression coefficient model corresponding to each cell and each RRU can be obtained.

[0109] Step S220: Obtain the regression coefficients of the RRU according to the regression coefficient model.

[0110] Regression coefficients of RRU It is expressed as follows:

[0111]

[0112] Understandably, this application constructs a regression coefficient model for the overall traffic of RRUs and the centralized and decentralized granularity of cells based on the Trans-Lasso algorithm, which improves the trend fluctuations and accuracy caused by single index or single node analysis. By combining multiple types of indicators, it considers the causal and logical relationships while taking into account the confounding effects of environmental variables, thereby improving the accuracy of fault prediction.

[0113] Based on the above embodiments, as an optional embodiment, determining the fitted traffic of each cell according to the regression coefficient of the RRU includes:

[0114] Step S310: Determine the fitted flow rate corresponding to each RRU based on the regression coefficient of the RRU; specifically, substitute the features of each RRU into the target model to determine the fitted flow rate corresponding to each RRU.

[0115] Step S320: Determine the fitted flow rate of each cell based on the fitted flow rate corresponding to all RRUs. The fitted flow rate of each cell is determined by summing the fitted flow rates corresponding to all RRUs.

[0116] It is understandable that this application obtains the best-fitting cell traffic time series by pre-classifying different cells and completing corresponding prediction trend curves and establishing corresponding regression coefficient models. By combining cell-level indicators and RRU-level indicators, the accuracy of identification and the fault detection rate can be improved.

[0117] Based on the above embodiments, as an optional embodiment, determining the abnormal RRU based on the traffic data and the fitted traffic includes:

[0118] Step S410: Determine the mean and variance of the fitted flow rate based on the fitted flow rate.

[0119] Step S420: Determine the expected deviation between the flow data and the fitted flow based on the flow data, the fitted flow mean, and the fitted flow variance;

[0120] Step S430: Based on the deviation from expectation and the preset expectation threshold, determine the abnormal RRU.

[0121] Specifically, this application utilizes Dropout to approximate the probability distribution of the model, replacing the expected bias and prediction variance of points to determine anomalous devices and anomalous times. In neural networks, the use of Dropout can be interpreted as a Bayesian approximation of the probabilistic model: a Gaussian process. During distribution prediction, Dropout in the generator network G remains enabled, and the prediction results are sampled under given conditions by repeating the experiment M times. The overall mean and variance are estimated by calculating the sample mean and variance, where the mean at time t is...

[0122]

[0123] Based on the mean, the variance of time t can be estimated as follows:

[0124]

[0125] The prediction result for time t is a c t Given the conditional probability distribution, we can assume that it approximately follows a Gaussian distribution:

[0126]

[0127] Where x is the input function, c is the judgment function, p is the noise distribution, and z is the sampling.

[0128] Anomalies at time t are scored by evaluating the degree of difference between the estimated observed values ​​and the true values, as well as the uncertainty of the model's prediction. For any time t, its predicted distribution can be calculated, and therefore the degree of deviation between its observed and true values ​​can be expressed as the expected value E(x). t -X t Anomaly at time t:

[0129]

[0130] Among them, A t It is the outlier score at time t, σ t It is the prediction variance at time t. The time series values ​​obtained from the training data are the boundary values. When performing anomaly detection on time series data, the input is the context conditions and data combination C, X, and the data generator G, and the output is the score A of all anomalies. t .

[0131] This application calculates the deviation from the expected value E based on the predicted mean and predicted variance of the predicted value and the actual value, and calculates the deviation anomaly based on a set threshold, which is set to τ using an automatic thresholding method. std =μ+2*σ, the time point exceeding the threshold is set as the abnormal time node, thereby locking the device ID of the abnormal time point and the abnormal RRU.

[0132] Understandably, this application establishes the concept of subsets for RRUs, sets the cell as the master node and the RRU as the child node, classifies the master nodes separately, establishes different scenario-based models, and then adds weight analysis to the composition of each master node. By analyzing the covariance of the two dimensions of weight and fit, it obtains the time points of deviation from the expected anomaly and the device points of covariance anomaly, thereby completing the anomaly detection and daily monitoring of RRU devices for which statistical data is difficult to obtain, and improving the technical defects of traditional single covariance judgment and lack of statistical data of prediction units.

[0133] The RRU flow anomaly detection device provided in this application is described below. The RRU flow anomaly detection device described below can be referred to in correspondence with the RRU flow anomaly detection method described above.

[0134] Figure 2 This is a schematic block diagram of the RRU flow anomaly detection device provided in this application, as follows: Figure 2 As shown, this application also provides an RRU flow anomaly detection device, comprising:

[0135] The classification module 210 is used to acquire the location data, basic parameters, coverage performance indicators and traffic data of the cell, and determine the classification of the cell based on the location data, the basic parameters, the coverage performance indicators and the traffic data, wherein the cell includes at least one RRU;

[0136] The regression module 220 is used to perform regression fitting on the cell classification, the cell user scheduling characteristics and the RRU characteristics based on a preset regression prediction algorithm to obtain the regression coefficients of the RRU.

[0137] The determination module 230 is used to determine the fitted traffic of each cell based on the regression coefficient of the RRU;

[0138] The detection module 240 is used to determine abnormal RRUs based on the traffic data and the fitted traffic.

[0139] It should be noted that the RRU traffic anomaly detection device provided by the present invention can execute the RRU traffic anomaly detection method described in any of the above embodiments during specific operation, and has the technical effects corresponding to the method. This embodiment will not elaborate on this.

[0140] Figure 3 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 3As shown, the electronic device may include: a processor 310, a communications interface 320, a memory 330, and a communication bus 340, wherein the processor 310, the communications interface 320, and the memory 330 communicate with each other via the communication bus 340. The processor 310 can call logical instructions in the memory 330 to execute an RRU traffic anomaly detection method, which includes:

[0141] The location data, basic parameters, coverage performance indicators, and traffic data of a cell are obtained. Based on the location data, basic parameters, coverage performance indicators, and traffic data, the classification of the cell is determined. The cell includes at least one RRU.

[0142] The regression coefficients of the RRU are obtained by performing regression fitting on the cell classification, the user scheduling characteristics of the cell, and the characteristics of the RRU based on a preset regression prediction algorithm.

[0143] Based on the regression coefficients of the RRU, the fitted traffic of each cell is determined;

[0144] Based on the traffic data and the fitted traffic, abnormal RRUs are identified.

[0145] Furthermore, the logical instructions in the aforementioned memory 330 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0146] On the other hand, this application also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer is able to execute the RRU traffic anomaly detection method provided by the above methods, the method including:

[0147] The location data, basic parameters, coverage performance indicators, and traffic data of a cell are obtained. Based on the location data, basic parameters, coverage performance indicators, and traffic data, the classification of the cell is determined. The cell includes at least one RRU.

[0148] The regression coefficients of the RRU are obtained by performing regression fitting on the cell classification, the user scheduling characteristics of the cell, and the characteristics of the RRU based on a preset regression prediction algorithm.

[0149] Based on the regression coefficients of the RRU, the fitted traffic of each cell is determined;

[0150] Based on the traffic data and the fitted traffic, abnormal RRUs are identified.

[0151] In another aspect, this application also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, is implemented to perform the RRU traffic anomaly detection method provided by the methods described above, the method comprising:

[0152] The location data, basic parameters, coverage performance indicators, and traffic data of a cell are obtained. Based on the location data, basic parameters, coverage performance indicators, and traffic data, the classification of the cell is determined. The cell includes at least one RRU.

[0153] The regression coefficients of the RRU are obtained by performing regression fitting on the cell classification, the user scheduling characteristics of the cell, and the characteristics of the RRU based on a preset regression prediction algorithm.

[0154] Based on the regression coefficients of the RRU, the fitted traffic of each cell is determined;

[0155] Based on the traffic data and the fitted traffic, abnormal RRUs are identified.

[0156] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0157] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0158] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for detecting abnormal RRU flow, characterized in that, include: The location data, basic parameters, coverage performance indicators, and traffic data of a cell are obtained. Based on the location data, basic parameters, coverage performance indicators, and traffic data, the classification of the cell is determined. The cell includes at least one RRU. The cell classification is used as the weight, the cell's user scheduling features are used as the target data, and the RRU's features are used as the information sample. These are substituted into the regression coefficient function of the regression prediction algorithm to construct a regression coefficient model corresponding to each cell classification and each RRU. The regression coefficient of the RRU is obtained based on the regression coefficient model. Based on the regression coefficients of the RRU, the fitted traffic of each cell is determined; Based on the traffic data and the fitted traffic, abnormal RRUs are identified.

2. The RRU flow anomaly detection method according to claim 1, characterized in that, The user scheduling characteristics of the cell include at least one or more of RRC connection count, traffic data, RSRP, and SINR, and the characteristics of the RRU include at least one or more of downlink data throughput, uplink data throughput, number of uplink PRBs, number of downlink PRBs, average noise, uplink data transmission duration, and downlink data transmission duration.

3. The RRU flow anomaly detection method according to claim 1, characterized in that, The process of acquiring cell location data, basic parameters, coverage performance indicators, and traffic data, and determining the cell classification based on the location data, basic parameters, coverage performance indicators, and traffic data, includes: Obtain the cell's location data, basic parameters, coverage performance indicators, and traffic data; Input the location data, the basic parameters, the coverage performance indicators, and the traffic data into a pre-trained classification model to obtain the classification of the cell output by the classification model; The classification model is obtained by training a convolutional neural network using historical location data, historical basic parameters, historical coverage performance indicators, and historical traffic data.

4. The RRU flow anomaly detection method according to claim 3, characterized in that, After acquiring the cell's location data, basic parameters, coverage performance indicators, and traffic data, the process also includes: The traffic data is split based on a preset time granularity to obtain a traffic data unit corresponding to each time granularity. The flow slope is determined based on the flow data units corresponding to two adjacent time granularities. Based on the maximum and minimum values ​​of the traffic data, determine the peaks and troughs of the traffic data changes; Correspondingly, the input of the location data, the basic parameters, the coverage performance indicators, and the traffic data to a pre-trained classification model to obtain the classification of the cell output by the classification model includes: Input the location data, the basic parameters, the coverage performance index, the traffic slope, and the peaks and troughs of the variation into a pre-trained classification model to obtain the classification of the cell output by the classification model.

5. The RRU flow anomaly detection method according to claim 1, 3, or 4, characterized in that, The basic parameters include at least one or more of the following: rated power, reference signal transmit power, number of antennas, bandwidth, and subcarrier spacing. The coverage performance indicators include coverage volume and coverage density, wherein the coverage volume is the volume of the multi-dimensional indoor buildings covered by the cell, which is determined based on the floor spacing and the boundary information of the multi-dimensional indoor buildings.

6. The RRU flow anomaly detection method according to claim 1, characterized in that, The step of determining the fitted traffic of each cell based on the regression coefficients of the RRU includes: Based on the regression coefficients of the RRUs, determine the fitted flow rate corresponding to each RRU; The fitted flow rate of each cell is determined based on the fitted flow rate corresponding to all RRUs.

7. The RRU flow anomaly detection method according to claim 1, characterized in that, The step of determining the abnormal RRU based on the traffic data and the fitted traffic includes: Based on the fitted flow rate, determine the mean and variance of the fitted flow rate; Based on the flow data, the fitted flow mean, and the fitted flow variance, determine the expected deviation between the flow data and the fitted flow. Based on the deviation from expectation and the preset expectation threshold, abnormal RRUs are determined.

8. A device for detecting abnormal RRU flow rates, characterized in that, include: A classification module is used to acquire location data, basic parameters, coverage performance indicators and traffic data of a cell, and determine the classification of the cell based on the location data, the basic parameters, the coverage performance indicators and the traffic data, wherein the cell includes at least one RRU; The regression module is used to take the cell classification as weight, the cell user scheduling features as target data, and the RRU features as information samples, and substitute them into the regression coefficient function of the regression prediction algorithm to construct a regression coefficient model corresponding to each cell classification and each RRU; and obtain the regression coefficient of the RRU according to the regression coefficient model. The determination module is used to determine the fitted traffic of each cell based on the regression coefficient of the RRU; The detection module is used to determine abnormal RRUs based on the traffic data and the fitted traffic.

9. 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 RRU traffic anomaly detection method as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the RRU traffic anomaly detection method as described in any one of claims 1 to 7.

11. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the RRU traffic anomaly detection method as described in any one of claims 1 to 7.

Images