Network element health assessment method, apparatus and device, and computer readable storage medium

Abstract

The invention relates to a network element health assessment method, device and equipment and a computer readable storage medium. According to the invention, the target features are selected according to the contribution degree of the network element instance sample to the prediction process, the influence of each feature data on the health degree of the network element instance is explained, the reason for the abnormal change of the health degree is conveniently positioned, and then the network element health probability is predicted based on the target features in the actual operation data. According to the invention, corresponding target features can be flexibly selected according to different network element categories or use scenes, so that the network element health assessment method can be better suitable for various network elements and scenes, and the flexibility and accuracy of the network element assessment method are improved.

Description

Technical Field

[0001] The present disclosure relates to the field of network detection technology, and in particular to a network element health assessment method, device, equipment and computer-readable storage medium. Background Art

[0002] Network elements play an important role in communication networks. They refer to devices with specific functions and independent operation capabilities. They are the basic units that constitute communication networks and are responsible for completing various communication tasks. In order to ensure the normal operation of communication networks, it is very important to evaluate the health of network elements and promptly detect abnormal conditions of network elements.

[0003] Normally, the staff will set the corresponding deduction threshold and deduction amount for the operation indicators of each dimension of the network element (such as equipment alarm, performance indicators, operation log, cutover operation, dialing dimension and business complaint volume), and set the weight factor of each dimension. Finally, the score value of the network element is evaluated by weighted sum calculation. The lower the score value, the higher the probability of abnormality of the network element, and timely maintenance is required. Another way is to predict the alarm sequence of the network element based on the hidden Markov prediction model, and issue an early warning when the prediction result indicates that the network element is about to enter a fault state.

[0004] However, in the existing network element health assessment methods, both the setting of the deduction criteria and the assumptions of the hidden Markov prediction model are relatively fixed, resulting in a lack of flexibility and accuracy in the network element health assessment methods. Summary of the invention

[0005] In order to solve the above technical problems, the present disclosure provides a network element health assessment method, device, equipment and computer-readable storage medium to improve the flexibility and accuracy of the network element assessment method.

[0006] In a first aspect, an embodiment of the present disclosure provides a network element health assessment method, including:

[0007] Obtain a sample data set of network element instances based on historical operation data of the network element;

[0008] For each network element instance sample in the network element instance sample data set, inputting the network element instance sample into a benchmark prediction model so that the benchmark prediction model predicts the health probability of the network element instance sample, wherein the network element instance sample includes multiple sample features;

[0009] Calculating the contribution of each of the sample features in the prediction process;

[0010] Determining a plurality of target features according to the contribution of each of the sample features;

[0011] Extracting the target features based on actual operation data of the network element to obtain a network element target feature data set;

[0012] The target feature data set is input into a trained network element health assessment model, so that the network element health assessment model predicts the health probability of the network element in real time to obtain a health assessment result of the network element.

[0013] In some embodiments, the acquiring of a network element instance sample data set based on historical operation data of the network element includes:

[0014] Obtaining an initial sample data set according to historical operation data, wherein the initial sample data set includes normal network element samples and abnormal network element samples, and the number of the abnormal network element samples is less than the number of the normal network element samples;

[0015] For each of the abnormal network element samples, generating at least one synthetic network element sample according to a plurality of adjacent samples of the abnormal network element sample, wherein the similarity between the synthetic network element sample and the abnormal network element sample is greater than a first preset threshold;

[0016] The normal network element samples, the abnormal network element samples and the synthetic network element samples are used as the network element instance sample data set, and the difference between the number of the abnormal network element samples and the number of the synthetic network element samples and the number of the normal network element samples is less than a second preset threshold.

[0017] In some embodiments, for each of the abnormal network element samples, generating at least one synthetic network element sample according to adjacent samples of the abnormal network element sample includes:

[0018] Calculating the sample distance between the abnormal network element sample and each other abnormal network element sample;

[0019] Determine a preset number of samples with the shortest sample distance as the adjacent samples;

[0020] For each of the adjacent samples, calculating a feature distance between the adjacent sample and each feature of the abnormal network element sample;

[0021] Multiply each of the feature distances by a preset random factor to obtain multiple synthetic features;

[0022] The synthetic network element sample is formed by the multiple synthetic features.

[0023] In some embodiments, after acquiring the network element instance sample data set based on the historical operation data of the network element, the method further includes:

[0024] Model training is performed based on the network element instance sample data set to obtain a benchmark prediction model.

[0025] In some embodiments, the prediction process includes a single prediction for each network element instance sample, and the calculating the contribution of each sample feature in the prediction process includes:

[0026] For each of the sample features, calculating the marginal contribution of the sample feature in each of the single predictions;

[0027] The cumulative value or average value of the marginal contribution in each of the single predictions is calculated to obtain the contribution of the sample feature in the prediction process.

[0028] In some embodiments, after determining a plurality of target features according to the contribution of each of the sample features, the method further includes:

[0029] Extracting the target features based on the historical operation data to obtain the network element feature sample data set;

[0030] The benchmark prediction model is updated and trained according to the network element feature sample data set to obtain a trained network element health assessment model.

[0031] In some embodiments, the method further comprises:

[0032] In the process of the network element health assessment model performing real-time prediction on the health probability of the network element, calculating the contribution of each of the target features in the real-time prediction process;

[0033] Based on the contribution of each of the target features, updating the target feature;

[0034] The health probability of the network element is predicted according to the updated target feature.

[0035] In a second aspect, an embodiment of the present disclosure provides a network element health assessment device, including:

[0036] A first acquisition module is used to acquire a network element instance sample data set based on historical operation data of the network element;

[0037] A first prediction module is used for inputting each network element instance sample in the network element instance sample data set into a benchmark prediction model so that the benchmark prediction model predicts the health probability of the network element instance sample, wherein the network element instance sample includes multiple sample features;

[0038] A calculation module, used to calculate the contribution of each of the sample features in the prediction process;

[0039] A determination module, used to determine a plurality of target features according to the contribution of each of the sample features;

[0040] A second acquisition module is used to extract the target features based on the actual operation data of the network element to obtain a network element target feature data set;

[0041] The second prediction module is used to input the target feature data set into the trained network element health assessment model, so that the network element health assessment model can make a real-time prediction on the health probability of the network element to obtain the health assessment result of the network element.

[0042] In a third aspect, an embodiment of the present disclosure provides an electronic device, including:

[0043] Memory;

[0044] Processor; and

[0045] Computer programs;

[0046] The computer program is stored in the memory and is configured to be executed by the processor to implement the method as described in the first aspect.

[0047] In a fourth aspect, an embodiment of the present disclosure provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the method described in the first aspect.

[0048] In a fifth aspect, an embodiment of the present disclosure further provides a computer program product, which includes a computer program or instructions, and when the computer program or instructions are executed by a processor, the network element health assessment method as described above is implemented.

[0049] The network element health assessment method, device, equipment and computer-readable storage medium provided by the embodiments of the present disclosure select target features according to the contribution of network element instance samples to the prediction process, explain the impact of each feature data on the health of the network element instance, facilitate the location of the cause of the health change, and then predict the network element health probability based on the target features in the actual operation data. The corresponding target features can be flexibly selected according to different network element categories or usage scenarios, so that the network element health assessment method can be better applied to various types of network elements and scenarios, thereby improving the flexibility and accuracy of the network element assessment method. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the present disclosure.

[0051] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0052] Figure 1 A flow chart of a network element health assessment method provided in an embodiment of the present disclosure;

[0053] Figure 2 A flow chart of a network element health assessment method provided by another embodiment of the present disclosure;

[0054] Figure 3 A schematic diagram of the structure of a network element health assessment device provided in an embodiment of the present disclosure;

[0055] Figure 4 A schematic diagram of the structure of an electronic device provided in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0056] In order to more clearly understand the above-mentioned objectives, features and advantages of the present disclosure, the scheme of the present disclosure will be further described below. It should be noted that the embodiments of the present disclosure and the features in the embodiments can be combined with each other without conflict.

[0057] In the following description, many specific details are set forth to facilitate a full understanding of the present disclosure, but the present disclosure may also be implemented in other ways different from those described herein; it is obvious that the embodiments in the specification are only part of the embodiments of the present disclosure, rather than all of the embodiments.

[0058] The traditional rule-based deduction method sets six dimensions: equipment alarm (A), performance index (B), operation log (C), cutover operation (D), dialing dimension (E) and business complaint volume (F). Based on different indicators in each dimension, the corresponding deduction threshold and deduction amount are set, and the weight factors α, β, ε, and ε of A, B, E, and F are set. Finally, the health W of the network element is evaluated by weighted sum calculation, and the calculation method is as follows:

[0059]

[0060] However, the above-mentioned index deduction values, index deduction thresholds, and dimension weights are all generated based on expert experience, which is subjective to a certain extent. There are personalized differences between different network elements, production environments, and user groups, which makes this method not universally interpretable. At the same time, the interval deduction method based on the index threshold can only roughly quantify the health deduction. For example, if the maximum number of concurrent online users of PGW (PDNGateWay) fluctuates by 5% month-on-month for two consecutive cycles, the health loss factor is 5 points added to the initial value of the health loss factor. Therefore, when the index change does not exceed the deduction threshold, the health deduction cannot be reflected, that is, the maximum number of concurrent online users of PGW fluctuates by 4.9% and 5.1% month-on-month for two consecutive cycles, respectively. If the month-on-month fluctuation does not reach more than 5% for two consecutive cycles, the health loss factor is 0 points added to the initial value of the health loss factor. At the same time, there are certain differences between expert experience for different operation and maintenance environments, and there are cases where this method is unavailable, which greatly increases the user's usage cost and reduces the availability of this method.

[0061] Another method is to predict the alarm sequence of network element objects based on the hidden Markov prediction model. This method only depends on the state at each moment, such as whether there is an alarm, and its corresponding observation object, such as an AMF (Access and Mobility management Function) network element. At a certain moment t, different instances of the same AMF network element express completely different states. For example, AMF network element 1 generates an alarm, and AMF network element 2 does not generate an alarm. However, at several moments after moment t, the states of the two network element instances are consistent. Therefore, the prediction results of this method are also consistent, that is, different input alarm sequences obtain the same prediction output. This inconsistency causes the method to fail.

[0062] In view of the above problems, the embodiments of the present disclosure provide a network element health assessment method, which is introduced below in conjunction with specific embodiments.

[0063] Figure 1 A flow chart of a network element health assessment method provided in an embodiment of the present disclosure. The method can be applied to a terminal device with a data processing function, such as a smart phone, a PDA, a tablet computer, a wearable device with a display screen, a desktop computer, a laptop computer, an all-in-one machine, a smart home device, etc. It is understandable that the network element health assessment method provided in an embodiment of the present disclosure can also be applied in other scenarios.

[0064] Below Figure 1 The network element health assessment method shown in the figure is introduced. The specific steps of the method are as follows:

[0065] S101. Acquire a network element instance sample data set based on historical operation data of the network element.

[0066] The historical operation data of a network element refers to the operation status and performance parameters recorded and saved by the network element device over a period of time.

[0067] Specifically, the historical operation data of the network element includes but is not limited to: performance data, such as central processing unit (CPU) utilization, memory occupancy, interface traffic, packet loss rate, delay, etc.; alarm data, such as the occurrence time, type, level and possible cause of the alarm; configuration data, such as hardware configuration, software version, interface configuration, routing configuration, etc.

[0068] The network element instance sample data set is extracted based on the historical operation data of the network element over a period of time. It includes the operating indicator values ​​of various operating indicators of the network element at different time points (that is, the sample feature values ​​of each sample feature), and also includes the network element status label, which is used to indicate whether the network element status at that time point is normal.

[0069] That is, a single network element instance sample includes a network element characteristic value of the network element at a certain time point and a network element state label at the time point.

[0070] S102. For each network element instance sample in the network element instance sample data set, input the network element instance sample into a benchmark prediction model so that the benchmark prediction model predicts the health probability of the network element instance sample, wherein the network element instance sample includes multiple sample features.

[0071] Optionally, model training is performed based on the network element instance sample data set to obtain a benchmark prediction model.

[0072] The reference prediction model is used to predict the probability that the network element state is normal according to the network element characteristic values ​​in the network element instance samples. The model is trained based on the network element instance samples, specifically, the model is trained based on the training set samples in the network element instance samples.

[0073] Model training is performed according to the network element instance sample data set to obtain a benchmark prediction model, including: for each network element instance sample, the network element characteristic value therein is input into the benchmark prediction model to be trained to obtain the prediction result output by the benchmark prediction model to be trained; the parameters of the benchmark prediction model to be trained are updated to guide the prediction result to be close to the network element status label.

[0074] Finally, when the error between the prediction result and the network element status label is controlled within a preset error range, a trained benchmark prediction model is obtained.

[0075] For example, the prediction result output by the baseline prediction model is the probability that the network element is in a normal state, and the value range is 0 to 1. The closer to 1, the greater the probability that the network element is in a normal state. The network element status label of the network element instance sample is 0 or 1, 0 indicates that the network element is abnormal, and 1 indicates that the network element is normal. In the process of training the baseline prediction model, for abnormal network element samples with a network element status label of 0, the prediction result output by the baseline prediction model is guided to be close to 0; for normal network element samples with a network element status label of 1, the prediction result output by the benchmark model is guided to be close to 1.

[0076] S103: Calculate the contribution of each sample feature in the prediction process.

[0077] When the baseline prediction model is making predictions, each sample feature has its own contribution. The higher the contribution, the greater the impact of the sample feature on the prediction result; the lower the contribution, the smaller the impact of the sample feature on the prediction result.

[0078] S104: Determine multiple target features according to the contribution of each of the sample features.

[0079] The sample features are sorted according to their contribution, and several sample features with the highest ranking results are selected as target features.

[0080] Optionally, a preset number of sample features with the highest ranking results are selected as target features; or, sample features whose contribution is greater than a preset contribution threshold are selected as target features; or, a preset number of sample features are selected as target features based on expert experience.

[0081] S105 . Extract the target features based on the actual operation data of the network element to obtain a network element target feature data set.

[0082] The health assessment of network elements is actually performed on the running network elements, so it is necessary to obtain the actual operation data of the network elements.

[0083] Similar to the historical operation data of the network element, the actual operation data of the network element refers to the operation data of the network element at the current time point. The operation index and the operation index value corresponding to the target feature are further selected to obtain the network element target feature data set.

[0084] S106: Input the target feature data set into a trained network element health assessment model, so that the network element health assessment model predicts the health probability of the network element in real time to obtain a health assessment result of the network element.

[0085] The disclosed embodiment obtains a network element instance sample data set based on the historical operation data of the network element; for each network element instance sample in the network element instance sample data set, the network element instance sample is input into a benchmark prediction model so that the benchmark prediction model predicts the health probability of the network element instance sample, and the network element instance sample includes multiple sample features; the contribution of each sample feature in the prediction process is calculated; multiple target features are determined according to the contribution of each sample feature; the target feature is extracted based on the actual operation data of the network element to obtain a network element target feature data set; the target feature data set is input into a trained network element health assessment model so that the network element health assessment model predicts the health probability of the network element in real time, selects the target feature according to the contribution of the network element instance sample to the prediction process, explains the influence of each feature data on the health of the network element instance, and conveniently locates the cause of the health change, and then predicts the health probability of the network element based on the target feature in the actual operation data, and can flexibly select the corresponding target feature according to different network element categories or usage scenarios, so that the network element health assessment method can be applied to various types of network elements and scenarios, thereby improving the flexibility of the network element assessment method.

[0086] At the same time, the benchmark prediction model or network element health assessment model in the embodiments of the present disclosure can choose an integrated model that is more suitable for the scenario. The selectable models include but are not limited to XGBoost, LightGBM, GBDT and random forest, among which the random forest is more intuitive in interpretability.

[0087] On the basis of the above embodiment, the method of obtaining a network element instance sample data set based on the historical operation data of the network element includes: obtaining an initial sample data set according to the historical operation data, the initial sample data set including normal network element samples and abnormal network element samples, and the number of abnormal network element samples is less than the number of normal network element samples; for each of the abnormal network element samples, generating at least one synthetic network element sample according to multiple adjacent samples of the abnormal network element sample, and the similarity between the synthetic network element sample and the abnormal network element sample is greater than a first preset threshold; taking the normal network element samples, the abnormal network element samples and the synthetic network element samples as the network element instance sample data set, and the difference between the number of the abnormal network element samples and the number of the synthetic network element samples and the number of the normal network element samples is less than a second preset threshold.

[0088] Specifically, for each of the abnormal network element samples, at least one synthetic network element sample is generated according to the adjacent samples of the abnormal network element sample, including: calculating the sample distance between the abnormal network element sample and each other abnormal network element sample; determining a preset number of samples with the shortest sample distance as the adjacent samples; for each of the adjacent samples, calculating the feature distance between the adjacent sample and each feature of the abnormal network element sample; multiplying each of the feature distances by a preset random factor to obtain multiple synthetic features; and the multiple synthetic features constitute the synthetic network element sample.

[0089] Usually in historical operation data, since network elements are in normal state most of the time, the number of abnormal network element samples in the corresponding initial sample data set is much less than the number of normal network element samples, which may lead to insufficient model prediction accuracy due to the small number of abnormal network element samples. Therefore, the embodiment of the present disclosure uses the existing abnormal network element samples to perform SMOTE (Synthetic Minority Oversampling Technique) algorithm data expansion to make the number of abnormal network element samples close to the number of normal network element samples.

[0090] Specifically, for each abnormal network element sample, the Euclidean distance between the feature of the abnormal network element sample and the feature of each other abnormal network element sample is calculated, and finally a preset number of samples with the shortest Euclidean distance are determined as adjacent samples.

[0091] Furthermore, each adjacent sample includes multiple sample features, such as the average system load, the number of authentication requests, etc. The feature distance between the adjacent sample and each feature of the abnormal network element sample is calculated, and each feature distance is multiplied by a random factor randomly generated between 0 and 1 to obtain multiple synthetic features. The sample composed of multiple synthetic features becomes a synthetic network element sample. That is, it is equivalent to randomly selecting a point on the line connecting the abnormal network element sample and its adjacent sample as a synthetic network element sample, so that the similarity between the synthetic network element sample and the corresponding abnormal network element sample is greater than the first preset threshold.

[0092] The above synthetic network element sample construction process is repeated until the difference between the number of abnormal network element samples and the number of synthetic network element samples and the number of normal network element samples is less than a second preset threshold. Usually, the number of synthetic network element samples is:

[0093] Y=N / 100×T

[0094] Where N is the sampling ratio, which is usually an integer such as 100, 200, etc., and is determined according to the number of training network element samples T.

[0095] The expanded abnormal network element samples include the original abnormal network element samples and the synthetic network element samples, and the number of the expanded abnormal network element samples and the number of normal network element samples reach about 1:1.

[0096] The disclosed embodiment expands the abnormal network element samples so that the number of the expanded abnormal network element samples is basically consistent with the number of normal network element samples, thereby solving the problem of insufficient model prediction accuracy caused by unbalanced sample labels, improving the accuracy of the baseline prediction model prediction and the accuracy of the contribution calculation, and thus improving the accuracy of the network element health assessment.

[0097] Figure 2 A flow chart of a network element health assessment method provided by another embodiment of the present disclosure is as follows: Figure 2 As shown, the method includes the following steps:

[0098] S201. Obtain a network element instance sample data set based on historical operation data of the network element.

[0099] Specifically, an initial sample data set is obtained based on historical operation data, and the initial sample data set includes normal network element samples and abnormal network element samples, and the number of the abnormal network element samples is less than the number of the normal network element samples; for each of the abnormal network element samples, at least one synthetic network element sample is generated based on multiple adjacent samples of the abnormal network element sample, and the similarity between the synthetic network element sample and the abnormal network element sample is greater than a first preset threshold; the normal network element samples, the abnormal network element samples and the synthetic network element samples are used as the network element instance sample data set, and the difference between the number of the abnormal network element samples and the number of the synthetic network element samples and the number of the normal network element samples is less than a second preset threshold.

[0100] S202. For each network element instance sample in the network element instance sample data set, input the network element instance sample into a benchmark prediction model so that the benchmark prediction model predicts the health probability of the network element instance sample, wherein the network element instance sample includes multiple sample features.

[0101] S203: For each of the sample features, calculate the marginal contribution of the sample feature in each of the single predictions.

[0102] S204: Calculate the cumulative value or average value of the marginal contribution in each of the single predictions to obtain the contribution of the sample feature in the prediction process.

[0103] SHAP (SHapley Additive exPlanation) is an additive explanation derived from game theory. It is the only feature interpretability measure with consistency and local accuracy in the integrated model interpretability tool. Its formula is:

[0104]

[0105] Among them, φ i represents the contribution of feature i, M is the set of all available data features; S is the set of features other than sample i that appear in a single prediction process, that is, is the number of elements in the feature set S; f x (S) is the prediction value of the benchmark prediction model for the sample when only feature set S is used, f x (S∪{i}) is the prediction value of the model for the sample when feature i is added to the feature set S.

[0106] Weight Item It is calculated based on the Shapley value distribution principle in combinatorial mathematics.

[0107] Among them, the definition

[0108] f x (S) = f(h x (z′))=E(f(x)|x s )

[0109] Where z′ represents the index of the feature set that does not appear in a single prediction, z′ i = 0 means feature i does not appear, z′ i =1 means feature i appears in a single prediction calculation. x (S∪{i})-f x (S) is the marginal contribution of feature i, which represents the change in the predicted value when feature i is added to a certain feature subset.

[0110] Therefore, the marginal contribution of a feature in a single prediction process can be represented, and the contribution of the entire data set can be represented by the accumulation or mean statistics of the single prediction results.

[0111] S205: Determine multiple target features according to the contribution of each of the sample features.

[0112] S206: Extract the target features based on the historical operation data to obtain the network element feature sample data set.

[0113] S207: Update and train the benchmark prediction model according to the network element feature sample data set to obtain a trained network element health assessment model.

[0114] After the target features have been determined, the target features can be used to extract features from the historical operation data, retaining only the target features to obtain a network element feature sample data set, and then the baseline prediction model is updated and trained again to obtain a trained network element health assessment model.

[0115] Optionally, feature importance can be ranked by data feature SHAP value, and feature selection can be performed based on expert experience to obtain target features, and model parameter tuning can be achieved by combining hyperparameter search cross-validation, including but not limited to grid search cross-validation and random search cross-validation.

[0116] Among them, grid search cross validation is an exhaustive search method that searches for the optimal model parameters by traversing every parameter combination in a given parameter grid. Random search cross validation is an improvement on grid search. Instead of traversing all possible parameter combinations, it randomly selects a part of parameter combinations for training.

[0117] S208: Extract the target features based on the actual operation data of the network element to obtain a network element target feature data set.

[0118] S209: Input the target feature data set into a trained network element health assessment model, so that the network element health assessment model predicts the health probability of the network element in real time to obtain a health assessment result of the network element.

[0119] Specifically, the network element health assessment model predicts the health probability of the network element in real time, and can output the failure probability p of the network element, and further map the normal probability (1-p) of the network element to the scoring interval of [0,100] to obtain the health assessment result of the network element. The calculation process is:

[0120] Health assessment result = (1-p) × 100

[0121] The disclosed embodiment solves the problem of insufficient model prediction accuracy caused by imbalanced sample labels by combining the SMOTE minority sample synthesis technology, maps the normal probability of the network element object into its health score, realizes real-time quantitative evaluation of network element health, and improves its sensitivity to quickly capture changes in network element health.

[0122] At the same time, the disclosed embodiment adopts SHAP feature interpretability technology to solve the problem of inconsistent feature importance measurement in classification tasks, adaptively optimize feature selection, improve model prediction accuracy, and solve the problem of scene feature differentiation.

[0123] Based on the above embodiments, the method also includes: in the process of the network element health assessment model performing real-time prediction on the health probability of the network element, calculating the contribution of each of the target features in the real-time prediction process; based on the contribution of each of the target features, updating the target features; and predicting the health probability of the network element based on the updated target features.

[0124] During the network element health assessment process, if there is a feature with a higher contribution than the existing target feature, it can be added to the target feature; or, if the existing target feature has a low contribution, it can be removed from the target feature.

[0125] Accordingly, when the target characteristics change, the network element health assessment model is also updated accordingly to make it more suitable for the current network element assessment application environment.

[0126] In addition, the data used by the model can also be optimized, such as increasing the amount of model training data through the accumulation of historical operation data, cleaning the data, adding abnormal network element samples and personalized data standards for specific scenarios through expert annotation, etc.

[0127] Based on the above method, the data and model can be further optimized to improve the accuracy of the network element health assessment method.

[0128] Figure 3 A schematic diagram of the structure of a network element health assessment device provided in an embodiment of the present disclosure. The network element health assessment device may be a terminal device with a data processing function as described in the above embodiment, or the network element health assessment device may be a component or assembly in the terminal device. The network element health assessment device provided in an embodiment of the present disclosure may execute the processing flow provided in an embodiment of the network element health assessment method, such as Figure 3As shown, the network element health assessment device 30 includes: a first acquisition module 31, a first prediction module 32, a calculation module 33, a determination module 34, a second acquisition module 35, and a second prediction module 36; wherein the first acquisition module 31 is used to acquire a network element instance sample data set based on the historical operation data of the network element; the first prediction module 32 is used to input the network element instance sample into a benchmark prediction model for each network element instance sample in the network element instance sample data set, so that the benchmark prediction model predicts the health probability of the network element instance sample, and the network element instance sample includes multiple sample features; the calculation module 33 is used to calculate the contribution of each of the sample features in the prediction process; the determination module 34 is used to determine multiple target features according to the contribution of each of the sample features; the second acquisition module 35 is used to extract the target feature based on the actual operation data of the network element to obtain the network element target feature data set; the second prediction module 36 is used to input the target feature data set into the trained network element health assessment model, so that the network element health assessment model predicts the health probability of the network element in real time to obtain the health assessment result of the network element.

[0129] Optionally, the first acquisition module 31 includes a first acquisition unit 311 and a sample generation unit 312. The first acquisition unit 311 is used to obtain an initial sample data set according to historical operation data, wherein the initial sample data set includes normal network element samples and abnormal network element samples, and the number of the abnormal network element samples is less than the number of the normal network element samples; the sample generation unit 312 is used to generate at least one synthetic network element sample for each abnormal network element sample according to multiple adjacent samples of the abnormal network element sample, and the similarity between the synthetic network element sample and the abnormal network element sample is greater than a first preset threshold; the normal network element samples, the abnormal network element samples and the synthetic network element samples are used as the network element instance sample data set, and the difference between the number of the abnormal network element samples and the number of the synthetic network element samples and the number of the normal network element samples is less than a second preset threshold.

[0130] Optionally, the sample generation unit 312 is specifically used to calculate the sample distance between the abnormal network element sample and each other abnormal network element sample; determine a preset number of samples with the shortest sample distance as the adjacent samples; for each of the adjacent samples, calculate the feature distance between the adjacent sample and each feature of the abnormal network element sample; multiply each of the feature distances by a preset random factor to obtain multiple synthetic features; and the multiple synthetic features constitute the synthetic network element sample.

[0131] Optionally, the network element health assessment device 30 further includes a first training module 37, which is used to perform model training according to the network element instance sample data set to obtain a benchmark prediction model.

[0132] Optionally, the prediction process includes a single prediction for each network element instance sample, and the calculation module 33 is specifically used to calculate, for each sample feature, the marginal contribution of the sample feature in each of the single predictions; calculate the cumulative value or average value of the marginal contribution in each of the single predictions to obtain the contribution of the sample feature in the prediction process.

[0133] Optionally, the network element health assessment device 30 also includes a second training module 38, which is used to extract the target features based on the historical operation data to obtain the network element feature sample data set; update and train the baseline prediction model according to the network element feature sample data set to obtain a trained network element health assessment model.

[0134] Optionally, the second prediction module 36 is also used to calculate the contribution of each of the target features in the real-time prediction process of the health probability of the network element by the network element health assessment model; update the target features based on the contribution of each of the target features; and predict the health probability of the network element based on the updated target features.

[0135] Figure 3 The network element health assessment device of the illustrated embodiment can be used to execute the technical solution of the above-mentioned method embodiment. Its implementation principle and technical effect are similar and will not be repeated here.

[0136] Figure 4 The electronic device provided in the embodiment of the present disclosure may be a terminal device with a data processing function as described in the above embodiment. The electronic device provided in the embodiment of the present disclosure may execute the processing flow provided in the embodiment of the network element health assessment method, such as Figure 4 As shown, the electronic device 40 includes: a memory 41, a processor 42, a computer program and a communication interface 43; wherein the computer program is stored in the memory 41 and is configured so that the processor 42 executes the network element health assessment method as described above.

[0137] In addition, an embodiment of the present disclosure further provides a computer-readable storage medium on which a computer program is stored. The computer program is executed by a processor to implement the network element health assessment method described in the above embodiment.

[0138] In addition, an embodiment of the present disclosure also provides a computer program product, which includes a computer program or instructions, and when the computer program or instructions are executed by a processor, the network element health assessment method as described above is implemented.

[0139] It should be noted that, in this article, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0140] The above description is only a specific embodiment of the present disclosure, so that those skilled in the art can understand or implement the present disclosure. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present disclosure. Therefore, the present disclosure will not be limited to the embodiments described herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A network element health assessment method, characterized in that: The method comprises: Obtain a sample data set of network element instances based on historical operation data of the network element; For each network element instance sample in the network element instance sample data set, inputting the network element instance sample into a benchmark prediction model so that the benchmark prediction model predicts the health probability of the network element instance sample, wherein the network element instance sample includes multiple sample features; Calculating the contribution of each of the sample features in the prediction process; Determining a plurality of target features according to the contribution of each of the sample features; Extracting the target features based on actual operation data of the network element to obtain a network element target feature data set; The target feature data set is input into a trained network element health assessment model, so that the network element health assessment model predicts the health probability of the network element in real time to obtain a health assessment result of the network element.

2. The method according to claim 1, characterized in that The obtaining of a network element instance sample data set based on the historical operation data of the network element includes: Obtaining an initial sample data set according to historical operation data, wherein the initial sample data set includes normal network element samples and abnormal network element samples, and the number of the abnormal network element samples is less than the number of the normal network element samples; For each of the abnormal network element samples, generating at least one synthetic network element sample according to a plurality of adjacent samples of the abnormal network element sample, wherein the similarity between the synthetic network element sample and the abnormal network element sample is greater than a first preset threshold; The normal network element samples, the abnormal network element samples and the synthetic network element samples are used as the network element instance sample data set, and the difference between the number of the abnormal network element samples and the number of the synthetic network element samples and the number of the normal network element samples is less than a second preset threshold.

3. The method according to claim 2, characterized in that The step of generating, for each of the abnormal network element samples, at least one synthetic network element sample according to adjacent samples of the abnormal network element sample comprises: Calculating the sample distance between the abnormal network element sample and each other abnormal network element sample; Determine a preset number of samples with the shortest sample distance as the adjacent samples; For each of the adjacent samples, calculating a feature distance between the adjacent sample and each feature of the abnormal network element sample; Multiply each of the feature distances by a preset random factor to obtain multiple synthetic features; The synthetic network element sample is formed by the multiple synthetic features.

4. The method according to claim 1, characterized in that: After acquiring the network element instance sample data set based on the historical operation data of the network element, the method further includes: Model training is performed based on the network element instance sample data set to obtain a benchmark prediction model.

5. The method according to claim 1, characterized in that: The prediction process includes a single prediction for each network element instance sample, and the calculation of the contribution of each sample feature in the prediction process includes: For each of the sample features, calculating the marginal contribution of the sample feature in each of the single predictions; The cumulative value or average value of the marginal contribution in each of the single predictions is calculated to obtain the contribution of the sample feature in the prediction process.

6. The method according to claim 1, characterized in that After determining a plurality of target features according to the contribution of each of the sample features, the method further includes: Extracting the target features based on the historical operation data to obtain the network element feature sample data set; The benchmark prediction model is updated and trained according to the network element feature sample data set to obtain a trained network element health assessment model.

7. The method according to claim 1, characterized in that The method further comprises: In the process of the network element health assessment model performing real-time prediction on the health probability of the network element, calculating the contribution of each of the target features in the real-time prediction process; Based on the contribution of each of the target features, updating the target feature; The health probability of the network element is predicted according to the updated target feature.

8. A network element health assessment device, characterized in that: include: A first acquisition module is used to acquire a network element instance sample data set based on historical operation data of the network element; A first prediction module is used for inputting each network element instance sample in the network element instance sample data set into a benchmark prediction model so that the benchmark prediction model predicts the health probability of the network element instance sample, wherein the network element instance sample includes multiple sample features; A calculation module, used to calculate the contribution of each of the sample features in the prediction process; A determination module, used to determine a plurality of target features according to the contribution of each of the sample features; A second acquisition module is used to extract the target features based on the actual operation data of the network element to obtain a network element target feature data set; The second prediction module is used to input the target feature data set into the trained network element health assessment model, so that the network element health assessment model predicts the health probability of the network element in real time to obtain the health assessment result of the network element.

9. An electronic device, characterized in that: include: Memory; processor; as well as Computer programs; The computer program is stored in the memory and is configured to be executed by the processor to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images