A method, device, equipment, medium and product for determining abnormal data monitoring parameters

Abstract

The present disclosure provides a method, apparatus, device, medium, and product for determining abnormal data monitoring parameters. The method comprises: training a graph attention model using multiple attribute network performance reference data at multiple historical sampling moments to obtain a performance data reconstruction model, obtaining the reference data reconstruction errors of the target attribute network performance at multiple historical sampling moments, determining the reconstruction errors of the original test data of the target attribute network performance at multiple test moments using the performance data reconstruction model, and then determining abnormal data monitoring parameters for the target attribute network performance data based on the two reconstruction errors. The method is highly usable and has relatively low investment costs, and is suitable for determining abnormal data detection parameters for multiple attribute network performance data.

Description

Technical Field

[0001] The present invention relates to the field of communication technology, and in particular to a method, device, equipment, medium and product for determining abnormal data monitoring parameters. Background Art

[0002] Currently, in order to ensure the reliability of communication systems and detect communication network anomalies as early as possible to avoid larger-scale failures, key performance indicators (KPIs) of the communication network can be monitored to promptly detect abnormal site fluctuations and perform alarm operations to ensure the availability of the network status.

[0003] In related technologies, threshold prediction models can be deployed at sites in each cell and subnet based on artificial intelligence technology. However, this requires collecting a large amount of supervisory data or self-supervisory data to train abnormal data monitoring models, which is not very practical and has high investment costs. Summary of the Invention

[0004] According to one aspect of the present disclosure, a method for determining abnormal data monitoring parameters is provided, comprising:

[0005] A graph attention model is trained based on network performance reference data with multiple attributes at multiple historical sampling moments to obtain a performance data reconstruction model.

[0006] Obtaining a reference data reconstruction error of the target attribute network performance at a plurality of the historical sampling moments;

[0007] Determine the reconstruction error of the original test data of the target attribute network performance at multiple test moments through the performance data reconstruction model;

[0008] Based on the reference data reconstruction error of the target attribute network performance at multiple historical sampling moments and the original test data reconstruction error of the target attribute network performance at multiple test moments, the abnormal data monitoring parameters of the target attribute network performance data are determined, and the abnormal data monitoring parameters of the target attribute network performance data meet the allowable limit of the reconstruction error.

[0009] According to another aspect of the present disclosure, there is provided a device for determining abnormal data monitoring parameters, comprising:

[0010] A training module is used to train a graph attention model based on multiple attribute network performance reference data at multiple historical sampling moments to obtain a performance data reconstruction model;

[0011] A determination module is used to obtain the reference data reconstruction error of the target attribute network performance at multiple historical sampling moments, determine the reconstruction error of the original test data of the target attribute network performance through the performance data reconstruction model, and determine the abnormal data monitoring parameters of the target attribute network performance data based on the reference data reconstruction error of the target attribute network performance at multiple historical sampling moments and the original test data of the target attribute network performance at multiple test moments, wherein the abnormal data monitoring parameters of the target attribute network performance data meet the allowable limit of the reconstruction error.

[0012] According to another aspect of the present disclosure, there is provided an electronic device, comprising:

[0013] processor; and,

[0014] Memory for storing programs,

[0015] The program includes instructions, which, when executed by the processor, cause the processor to perform the method according to the exemplary embodiment of the present disclosure.

[0016] According to another aspect of the present disclosure, a non-transitory computer-readable storage medium is provided, wherein the non-transitory computer-readable storage medium stores computer instructions, and the computer instructions are used to cause the computer to execute the method according to the exemplary embodiments of the present disclosure.

[0017] According to another aspect of the present disclosure, a computer program product is provided, including a computer program, which implements the method described in the exemplary embodiments of the present disclosure when executed by a processor.

[0018] One or more technical solutions provided in the exemplary embodiments of the present disclosure take into account that there may be certain intrinsic connections between network performance data of multiple attributes. A graph attention model can be trained based on multiple attribute network performance reference data at multiple historical sampling moments, fully exploring the deep connections between different attribute network performance reference data, so that the obtained performance data reconstruction model can not only be applied to the reconstruction of original data of network performance of multiple attributes, but also improve the reconstruction accuracy. The training data of the graph attention model is multiple attribute network performance reference data at multiple historical sampling moments. Therefore, the reconstruction error of the reference data of the target attribute network performance obtained at multiple historical sampling moments is relatively small, while the reconstruction error of the original test data of the target attribute network performance determined by the performance data reconstruction model at multiple test moments is relatively large.

[0019] Moreover, the exemplary embodiment of the present disclosure can determine the abnormal data monitoring parameters of the target attribute network performance data based on the reference data reconstruction error of the target attribute network performance at multiple historical sampling moments and the reconstruction error of the original test data of the target attribute network performance at multiple test moments. In essence, the reference data reconstruction error of the target attribute network performance at a large number of historical sampling moments is used as the normal data sequence of the target attribute network performance, and the reconstruction error of the original test data of the target attribute network performance at multiple test moments and the normal data sequence of the target attribute network performance are regarded as a set, thereby determining the reconstruction error that meets the allowable limit of the reconstruction error, and taking the reconstruction error that meets the allowable limit of the reconstruction error as the abnormal data monitoring parameter of the target attribute network performance. It can be seen that the exemplary embodiment of the present disclosure combines the normal data sequence of the target attribute network performance and the reconstruction error of the original test data at multiple test moments to determine the abnormal data monitoring parameters, so that the determined abnormal data monitoring parameters can be used to specifically analyze whether there is an abnormality in the original test data of the target attribute network performance and the time when the abnormality occurs.

[0020] In addition, in the process of training the graph attention model, the method of the exemplary embodiment of the present invention uses training data including multiple attribute network performance reference data at multiple historical sampling moments. The graph attention model can reconstruct multiple attribute network performance reference data at multiple historical sampling moments. Therefore, when training the graph attention model, there is no need to mark outliers or manually set thresholds for each attribute network performance reference data at multiple historical sampling moments, thereby reducing investment costs and having higher practicality. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Further details, features and advantages of the present disclosure are disclosed in the following description of exemplary embodiments in conjunction with the accompanying drawings, in which:

[0022] Figure 1 A schematic flow chart showing a method for determining abnormal data monitoring parameters according to an exemplary embodiment of the present disclosure is shown;

[0023] Figure 2 A schematic diagram of a module framework of an exemplary embodiment of the present disclosure is shown;

[0024] Figure 3 A schematic diagram of a graph attention model training process of an exemplary embodiment of the present disclosure is shown;

[0025] Figure 4 A schematic diagram of a process for obtaining reconstruction errors of original test data according to an exemplary embodiment of the present disclosure is shown;

[0026] Figure 5 A schematic diagram of a process for obtaining abnormal data monitoring parameters of target attribute network performance according to an exemplary embodiment of the present disclosure is shown;

[0027] Figure 6 A schematic block diagram of functional modules of an apparatus for determining abnormal data monitoring parameters according to an exemplary embodiment of the present disclosure is shown;

[0028] Figure 7 shows a schematic block diagram of a chip according to an exemplary embodiment of the present disclosure;

[0029] Figure 8 A structural block diagram of an exemplary electronic device that can be used to implement the embodiments of the present disclosure is shown. DETAILED DESCRIPTION

[0030] The following describes embodiments of the present disclosure in more detail with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as limited to the embodiments described herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are for illustrative purposes only and are not intended to limit the scope of protection of the present disclosure.

[0031] It should be understood that the various steps described in the method embodiments of the present disclosure may be performed in different orders and / or in parallel. In addition, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present disclosure is not limited in this respect.

[0032] The term "including" and its variations used in this document are open inclusions, that is, "including but not limited to". The term "based on" means "based at least in part on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one other embodiment"; the term "some embodiments" means "at least some embodiments". The relevant definitions of other terms will be given in the description below. It should be noted that the concepts of "first", "second", etc. mentioned in this disclosure are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.

[0033] It should be noted that the modifications of "one" and "plurality" mentioned in the present disclosure are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, they should be understood as "one or more".

[0034] Related technologies can use a given anomaly data monitoring parameter, such as an anomaly data monitoring threshold, which is constant. This parameter can then be used to monitor network key performance indicators (KPIs). However, this approach ignores the differences between sites, such as cells and subnets, and cannot effectively monitor site KPIs for anomalies. While experts can develop targeted anomaly monitoring strategies based on their experience, this is also time-consuming and labor-intensive, increasing operational costs.

[0035] In addition, supervised machine learning or self-supervision can be used to train anomaly data monitoring models, and then the trained anomaly data monitoring models can be used to predict whether there are anomalies in the network's KPIs, and to issue alerts immediately when abnormal fluctuations occur in key performance indicators, thereby ensuring the normal operation of cell, subnet and other site networks. However, when training anomaly data monitoring models through supervised machine learning or self-supervision, a large amount of supervisory data needs to be collected, and manual labeling of abnormal data is required. Therefore, the cost of the abnormal data monitoring methods of related technologies is high and the application value is relatively low.

[0036] In response to the above problems, an exemplary embodiment of the present disclosure provides a method for determining abnormal data monitoring parameters, which can be applied to a communication system. The technical solution can be deployed in a network management platform of the communication system, and the network management platform can perform intelligent operation and maintenance of the communication system. For example, the communication system may include a fifth generation (5th generation, 5G) or new wireless (new radio, NR system, long term evolution (long term evolution, LTE)) system, LTE time division duplex (time division duplex, TDD) system, etc. The technical solution provided by the exemplary embodiment of the present disclosure can also be applied to future communication systems, such as the sixth generation mobile communication system.

[0037] The technical solutions provided by the exemplary embodiments of the present disclosure can also be applied to device-to-device (D2D) communication, vehicle-to-everything (V2X) communication, machine-to-machine (M2M) communication, machine type communication (MTC), and Internet of Things (IoT) communication systems or other communication systems.

[0038] The technical solution of the exemplary embodiment of the present disclosure can obtain a performance data reconstruction model by training a graph attention model, and layout the performance data reconstruction model in a network management system, and then use the performance data reconstruction model as a bridge to obtain the original test data reconstruction error of the target attribute network performance at multiple test times and the reference data reconstruction error of the target attribute network performance at multiple said historical sampling times. Finally, using the reference data reconstruction error of the target attribute network performance at multiple said historical sampling times as a reference, the reference data reconstruction error of the target attribute network performance at multiple said historical sampling times and the original test data reconstruction error of the target attribute network performance at multiple test times are comprehensively considered, so that the abnormal data monitoring parameters finally obtained can be targeted and accurately monitor the original test data with relatively large reconstruction errors and the time nodes when data anomalies occur. Moreover, the method of the exemplary example of the present disclosure does not require abnormal data labeling, has high practicality, and has relatively low network operating costs.

[0039] The abnormal data monitoring parameters provided by the exemplary embodiments of the present disclosure can be applied to a server or a chip in the server. Figure 1 FIG. 1 is a flow chart showing a method for determining abnormal data monitoring parameters according to an exemplary embodiment of the present disclosure. Figure 1 As shown, the method for determining abnormal data monitoring parameters of the exemplary embodiment of the present disclosure includes:

[0040] Step 101: Train a graph attention model based on multiple attribute network performance reference data at multiple historical sampling moments to obtain a performance data reconstruction model.

[0041] In practical applications, the original historical data of the target network's various attribute network performances during the historical sampling period can be obtained, and then the original historical data of each attribute network performance can be preprocessed to remove outliers, thereby obtaining the reference data of the target network's various attribute network performances during the historical sampling period.

[0042] For example, the above historical sampling period may include multiple historical sampling moments, and each attribute network performance reference data at each historical sampling moment may refer to normal network performance data. For example, network performance reference data X = [x1, ..., x2] may be collected every 15 minutes for two consecutive weeks. C ], C represents the number of attributes of network performance reference data at the same historical sampling moment, x1 represents the first attribute network performance reference data, x C Represents the Cth attribute network performance reference data.

[0043] Considering that network performance data with multiple attributes can be collected by different sensors, and there is a certain intrinsic relationship between the network performance data collected by different sensors, it is possible to train a graph attention model based on the network performance reference data with multiple attributes at multiple historical sampling moments to obtain a performance data reconstruction model. It can be seen that in the exemplary embodiment of the present disclosure, when training the graph attention model, the training data used includes network performance reference data with multiple attributes at multiple historical sampling moments. The graph attention model can reconstruct network performance reference data with multiple attributes at multiple historical sampling moments. Therefore, when training the graph attention model, it is not necessary to mark outliers for each attribute network performance reference data at multiple historical sampling moments, thereby reducing the input cost and having high practicality.

[0044] Step 102: Obtain the reference data reconstruction error of the target attribute network performance at multiple historical sampling moments. Since the training data for the graph attention model is the reference data of multiple attribute network performance at multiple historical sampling moments, the reference data reconstruction error of the target attribute network performance at multiple historical sampling moments determined by the performance data reconstruction model is relatively small. These reference data reconstruction errors can be used as standard data reconstruction errors.

[0045] In practical applications, the target attribute network performance reference data of multiple historical sampling moments can be input into the performance data reconstruction model to obtain the reference data reconstruction results of the target attribute network performance at multiple historical sampling moments, and then the reference data reconstruction results G' based on the target attribute network performance at multiple historical sampling moments can be obtained. train and target attribute network performance reference data G at multiple historical sampling moments train , determine the reference data reconstruction error of the target attribute network performance at multiple historical sampling moments. For example, MSE can be used train =|G train -G' train Determine the mean square error (MSE) of the target attribute network performance at multiple historical sampling moments using the reference data reconstruction method. train .

[0046] Step 103: Determine the reconstruction error of the original test data of the target attribute network performance at multiple test moments through the performance data reconstruction model.

[0047] In practical applications, the original test data of the target attribute network performance at multiple test times can be directly input into the performance data reconstruction model to obtain the reconstruction results of the original test data of the target attribute network performance at multiple test times, and then the reconstruction results G' based on the original test data of the target attribute network performance at multiple test times can be obtained. test and the original test data G of the target attribute network performance testDetermine the reconstruction error of the original test data of the target attribute network performance at multiple test times.

[0048] Exemplarily, the above-mentioned original test data of the target attribute network performance at multiple test times may refer to the original test data collected from the target network without any data processing. Considering that the training data of the graph attention model is the reference data of multiple attribute network performances at multiple historical sampling times, the original test data of the target attribute network performance at multiple test times is used as the input of the performance data reconstruction model, and the reconstruction error of the reference data of the target attribute network performance at multiple historical sampling times finally determined may be relatively large.

[0049] For example, you can use MSE test =|G test -G' test Determine the mean square error (MSE) of the target attribute network performance at multiple historical sampling moments using the reference data reconstruction method. test , and MSE test May be greater than MSE train .

[0050] Step 104: Based on the reference data reconstruction error of the target attribute network performance at multiple historical sampling moments and the original test data reconstruction error of the target attribute network performance at multiple test moments, determine the abnormal data monitoring parameters of the target attribute network performance data, and the abnormal data monitoring parameters of the target attribute network performance data meet the reconstruction error allowable limit.

[0051] When determining the abnormal data monitoring parameters of the target attribute network performance data, the exemplary embodiment of the present disclosure essentially takes the reference data reconstruction error of the target attribute network performance at a large number of historical sampling moments as the normal data sequence of the target attribute network performance, regards the original test data reconstruction error of the target attribute network performance at multiple test moments and the normal data sequence of the target attribute network performance as a set, thereby determining the reconstruction error that meets the allowable limit of the reconstruction error, and uses the reconstruction error that meets the allowable limit of the reconstruction error as the abnormal data monitoring parameter of the target attribute network performance.

[0052] It can be seen that the exemplary embodiment of the present disclosure integrates the normal data sequence of the target attribute network performance and the reconstruction error of the original test data at multiple test times to determine the abnormal data monitoring parameters, so that the determined abnormal data monitoring parameters can be used to specifically analyze whether there is an abnormality in the original test data of the target attribute network performance and the time when the abnormality occurs.

[0053] As a possible implementation method, the method of the exemplary embodiment of the present disclosure may also include: determining the allowable distribution range of reference data of each attribute network performance based on the original historical data of each attribute network performance at multiple historical sampling moments, and then determining multiple attribute network performance reference data based on the allowable distribution range of reference data of multiple attribute network performances and the original data of multiple attribute network performances at multiple historical sampling moments.

[0054] In practical applications, the exemplary embodiment of the present disclosure can determine the mean of each attribute original data based on each attribute original data at multiple historical sampling moments. Then, based on the mean of each attribute raw data and each attribute raw data at multiple historical sampling moments, the standard deviation σ of each attribute raw data is determined. C i. Finally, based on the mean value and standard deviation of the original data of each attribute, the allowable distribution range of the reference data of the network performance of each attribute is determined.

[0055] When determining the reference data of multiple attribute network performances based on the allowable distribution range of the reference data of multiple attribute network performances and the original data of multiple attribute network performances at multiple historical sampling moments, it is possible to first determine whether the original data of a certain attribute is within the allowable distribution range of the reference data of the corresponding attribute network performance. If the original data of a certain attribute is within the allowable distribution range of the reference data of the corresponding attribute network performance, it means that the original data of this attribute is normal, and the original data of this attribute can be determined as the reference data of the corresponding attribute network performance. If the original data of a certain attribute is outside the allowable distribution range of the reference data of the corresponding attribute network performance, it means that the original data of this attribute is an extreme value and is abnormal data. The original data of the attribute network performance can be corrected to obtain the reference data of the corresponding attribute network performance.

[0056] For example, the exemplary embodiment of the present disclosure may set the reference data of each attribute network performance to allow the distribution range to be determined by OK. You can detect the mean of the original data of a certain attribute and the original data of the corresponding attribute Is the difference equal to According to the test results, determine whether the original data of this attribute corresponds to the reference data of the attribute network performance within the allowable distribution range. If the original data of a certain attribute is equal to the mean of the original data of the corresponding attribute The difference is equal to This indicates that the original data of this attribute is within the allowable distribution range of the reference data of the corresponding attribute network performance. The difference is not equal to This indicates that the original data of this attribute is outside the allowable distribution range of the reference data of the corresponding attribute network performance.

[0057] When the original historical data of a certain attribute network performance at the current historical sampling moment is outside the allowable range of the reference data of the corresponding attribute network performance, the original historical data of the certain attribute network performance at one or two historical sampling moments adjacent to the current historical sampling moment can be obtained to repair the original historical data of the certain attribute network performance at the current historical sampling moment.

[0058] Taking the connection rate, handover rate and user perceived rate as examples, Table 1 shows the raw data list of network performance in a certain historical sampling period. The historical sampling period may include Q historical sampling moments, denoted by t1, t2, ..., t Q express.

[0059] Table 1 List of raw data of network performance

[0060]

[0061] When the pth historical sampling time t is detected p The connection rate R p Abnormal, and the p-1th historical sampling moment t p-1 The connection rate R p-1 Normal, we can directly use the pth historical sampling time t p-1 The connection rate R p-1 Set to the pth historical sampling time t p The connection rate R p When the pth historical sampling time t is detected p The connection rate R p Abnormal, and the p+1th historical sampling time t p+1 The connection rate R p+1 Normal, we can directly use the p+1th historical sampling time t p+1 The connection rate R p+1 Set to the pth historical sampling time t p The connection rate R p .

[0062] When the pth historical sampling time t is detected p The connection rate R p Abnormal, and the p-1th historical sampling moment t p-1 The connection rate R p-1 and the p+1th historical sampling time t p+1 The connection rate R p+1 Normal, it can be directly based on the p-1th historical sampling time t p-1 The connection rate R p-1 and the p+1th historical sampling time t p+1 The connection rate Rp+1 The weighted sum of the results determines the p-th historical sampling time t p The connection rate R p .

[0063] In some embodiments, given that network performance reference data of the same attribute may have different dimensions and value ranges, data normalization can be used to improve the graph neural network model's learning of network performance reference data of different attributes. It should be understood that data normalization can be performed using methods such as z-score normalization and min-max normalization.

[0064] When performing data standardization, the mean of each attribute network performance reference data can be determined based on the network performance reference data of each attribute at multiple historical sampling moments, and then the mean of each attribute network performance reference data can be determined based on the network performance reference data of each attribute at multiple historical sampling moments. Determine the standard deviation of network performance reference data for each attribute Finally, based on the standard deviation of the network performance reference data for each attribute and the mean of the network performance reference data for each attribute For each attribute network performance reference data x j Perform data standardization. For example, The network performance reference data with the same attribute is scaled so that the network performance reference data with the same attribute X obeys the standard normal distribution X~N(0,1).

[0065] As a possible implementation, Figure 2 Schematic diagram of the module framework of the exemplary embodiment of the present disclosure is shown. Figure 2 As shown, the model framework 200 of the exemplary embodiment of the present disclosure may include a multi-scale feature extraction model 201 and a graph attention model 202. The multi-scale feature extraction model 201 may be functionally divided into a multi-scale decomposition unit 2011 and a feature fusion unit 2012.

[0066] Figure 3 FIG shows a flow chart of the graph attention model training process of an exemplary embodiment of the present disclosure. Figure 3 As shown, the exemplary embodiment of the present disclosure trains a graph attention model based on multiple attribute network performance reference data at multiple historical sampling moments to obtain a performance data reconstruction model, which may include:

[0067] Step 301: Input multiple attribute network performance reference data at multiple historical sampling moments into a multi-scale feature extraction model to determine a multi-scale reference fusion feature sequence for the multiple attribute network properties. It should be understood that the multi-scale feature extraction model can be a multi-scale feature extraction model that requires training, or it can be an extraction model composed of a simple algorithm that does not require training.

[0068] In order to achieve the purpose of amplifying the training data, the multi-scale reference fusion feature sequence of multiple attribute network performance of the exemplary embodiment of the present disclosure includes multiple multi-scale reference fusion feature slices. In this case, the method of the exemplary embodiment of the present disclosure may also include:

[0069] Multiple reference data slices are obtained from multiple attribute network performance reference data at multiple historical sampling moments. The multiple reference data slices are input into a multi-scale feature extraction model to determine multiple multi-scale reference fusion feature slices. Each reference data slice includes multiple attribute network performance reference data at at least two historical sampling moments.

[0070] In practical applications, multiple network performance reference data of various attributes can be obtained within a historical sampling period, including multiple network performance reference data of various attributes at multiple historical sampling moments. For this historical sampling period, a sliding window operation can be performed along the time dimension to obtain multiple reference data slices corresponding to the various network performance attributes.

[0071] For example, if the sliding window time length is L and the number of network performance attributes is C, M sliding window operations can be performed on the network performance reference data of multiple attributes within the historical sampling period to obtain M reference data slices. Then, the multiple reference data slices can be expressed as [X1,...,X M ],X∈R L×C , where R L×C represents a slice formed by the reference data of multiple attribute network performance in the same time period, X1 represents a slice formed by the reference data of multiple attribute network performance in the first time period, and X N It represents a slice formed by the reference data of the network performance of multiple attributes in the Nth time period. L represents the sliding window time length, which is usually an integer multiple of the sampling frequency.

[0072] When a sliding window operation is used to obtain multiple reference data slices, the network performance reference data contained in two adjacent reference data slices in the time dimension are partially identical. Therefore, when multiple reference data slices are input into the multi-scale feature extraction model and multiple multi-scale reference fusion feature slices are determined, the extraction accuracy of the multi-scale fusion feature can be improved.

[0073] Step 302: Input the multi-scale reference fusion feature sequence of multiple attribute network performance into the graph attention model to obtain the reconstructed data prediction results of the multiple attribute network performance. Here, each node feature of the graph attention model of the exemplary embodiment of the present disclosure is determined by the data attributes of the corresponding attribute network performance reference data.

[0074] When the number of network performance reference data attributes at each historical sampling moment is C, the number of nodes in the graph attention model is equal to C. Therefore, the graph attention model G can be expressed as: G = {v1,...,v C},v∈R L , where v1 represents the node feature of the first node, v C Represents the node characteristics of the Cth node.

[0075] For example, the node feature of the first node can be a multi-scale reference fusion feature sequence of the first attribute network performance, v2 represents the node feature of the second node, for example, the node feature of the second node can be a multi-scale reference fusion feature sequence of the second attribute network performance, ..., v C The node feature representing the Cth node, for example, the node feature of the Cth node can be a multi-scale reference fusion feature sequence of the Cth attribute network performance.

[0076] An exemplary embodiment of the present disclosure can obtain connection weights and attention coefficients between different node features based on multiple attribute network performance reference data at multiple historical sampling moments, and then perform multiple message transmissions on the multi-scale reference fusion feature sequences of each attribute network performance based on the connection weights and attention coefficients between different node features to obtain multi-scale deep reference feature sequences of multiple attribute network performances. Finally, the multi-scale deep reference feature sequences of multiple attribute network performances are fully connected through a fully connected layer to obtain reconstructed data prediction results of multiple attribute network performances with the same size of multi-scale reference fusion feature sequences of multiple attribute network performances.

[0077] Step 303: Based on the reconstructed data prediction results of the multiple attribute network performance and the multiple attribute network performance reference data at multiple historical sampling moments, determine the data reconstruction loss of the graph attention model. For example, the data reconstruction loss here can be expressed as mean square error (MSE) = |G-G'|.

[0078] When the data reconstruction loss of the graph attention model is less than or equal to the preset reconstruction loss, it means that the model parameters of the multi-scale feature extraction model and the model parameters of the graph attention model are appropriate, and step 304 can be executed. Otherwise, it means that the model parameters of the multi-scale feature extraction model and the model parameters of the graph attention model are not appropriate, and step 305 can be executed.

[0079] Step 304: Stop updating the model parameters of the graph attention model, obtain the performance data reconstruction model, and stop updating the model parameters of the multi-scale feature extraction model.

[0080] Step 305: Update the model parameters of the multi-scale feature extraction model and the model parameters of the graph attention model. The update method here can be a back-propagation algorithm.

[0081] In an optional manner, the exemplary embodiments of the present disclosure can learn the complex periodic contextual information present in the network performance reference data of multiple attributes at multiple historical sampling moments through a multi-scale feature extraction model, thereby improving the detection capability of subsequent graph neural network models. For example, the periodic variations in data between day and night, weekdays and weekends, quarters, years, and seasons present in the network performance reference data of multiple attributes at multiple historical sampling moments can be learned, providing the graph neural network model with training data with sufficient information, thereby improving the data reconstruction capability of the performance data reconstruction model after training.

[0082] The multi-scale decomposition unit can be used to decompose the reference data of each attribute network performance at each historical sampling moment using multiple key periods of each attribute network performance to obtain a multi-scale reference feature sequence for each attribute network performance; and the feature fusion unit is used to determine the multi-scale reference fusion feature sequence for each attribute network performance based on the multi-scale reference feature sequence input of each attribute network performance.

[0083] When the reference data of each attribute network performance at each historical sampling moment is decomposed using multiple key periods of each attribute network performance, the multi-scale reference feature sequence obtained for each attribute network performance can reflect the characteristic changes of each attribute network performance reference data in different key periods. Therefore, based on the multi-scale reference feature sequence of each attribute network performance, the multi-scale reference fusion feature sequence of each attribute network performance determined can actually reflect the feature fusion results of each attribute network performance in different key periods, thereby ensuring that the training data input into the graph attention model contains deeper features, so that the graph attention model can better reconstruct the performance data of each attribute network.

[0084] When the multi-scale reference fusion feature sequence for multiple attribute network performances includes multiple multi-scale reference fusion feature slices, each multi-scale reference fusion feature slice can be decomposed using multiple key periods for each attribute network performance to obtain a multi-scale reference feature sequence segment for each attribute network performance. Then, based on the multi-scale reference feature sequence segment for each attribute network performance, a multi-scale reference fusion feature sequence segment for each attribute network performance is determined. Considering that different multi-scale reference fusion feature slices are extracted from the same multi-scale reference fusion feature sequence, each multi-scale reference fusion feature slice for the same attribute network performance can be multi-scale decomposed using the same key period.

[0085] Exemplarily, the multi-scale decomposition unit of the exemplary embodiment of the present disclosure can also be used to: determine the spectrum information of each attribute network performance reference data based on the spectrum information of each attribute network performance reference data at multiple historical sampling moments; determine the key amplitude of each attribute network performance reference data based on the spectrum information of each attribute network performance reference data; and obtain multiple key periods of each attribute network performance based on the key amplitude of each attribute network performance reference data.

[0086] In practical applications, given the differences in network traffic volume between daytime and nighttime, as well as between weekends and weekdays, the network performance reference data for each attribute at multiple historical sampling moments has a relatively complex periodic distribution. In this case, a fast Fourier transform can be performed on the network performance reference data for the same attribute at multiple historical sampling moments to obtain the spectrum information F = FFT(X) for that attribute network performance reference data, where X represents the sequence of network performance reference data for the same attribute at multiple historical sampling moments, and FFT(·) represents the fast Fourier transform function.

[0087] Taking the spectrum information of a certain attribute network performance reference data as an example, the amplitude mean μ of the attribute network performance reference data can be determined based on multiple amplitudes of the attribute network performance reference data. F and amplitude standard deviation σ F , then based on the amplitude mean μ of the network performance reference data of this attribute F and amplitude standard deviation σ F When the amplitude of the attribute network performance reference data and the amplitude of the attribute network performance reference data |F|>μ F +3*σ F When , it indicates that the amplitude of the attribute network performance reference data is a candidate amplitude, and then all candidate amplitudes of the attribute network performance reference data are sorted in descending order, and the first k candidate amplitudes are determined as the key amplitudes of the attribute network performance reference data.

[0088] After obtaining the k key amplitudes of the attribute network performance reference data, the k key frequencies corresponding to the key amplitudes of the attribute network performance reference data can be obtained from the spectrum information of the attribute network performance reference data based on the k key amplitudes of the attribute network performance reference data, and then the k key periods [T1,...,T k ], T1 represents the first critical period of the network performance of this attribute, T k Indicates the kth critical period of network performance of this attribute.

[0089] Exemplarily, the multi-scale decomposition unit of the exemplary embodiment of the present disclosure is used to: determine the time length of each attribute network performance reference data based on each attribute network performance reference data at multiple historical sampling moments, determine each reference decomposition parameter of each attribute network performance reference data based on each key period of each attribute network performance and the time length of each attribute network performance reference data, decompose each attribute network performance reference data at multiple historical sampling moments based on the multiple decomposition parameters of each attribute network performance reference data, and obtain a multi-scale reference feature sequence for each attribute network performance.

[0090] Each decomposition parameter includes the length of the reference feature in the time dimension and the number of scales of the reference feature. The length of the reference feature in the time dimension is the same as the time length of the corresponding critical period, while the number of scales of the reference feature is determined by the time length of the corresponding attribute network performance reference data and the time length of the corresponding critical period. Here, the number of scales of the reference feature is negatively correlated with the time length of the corresponding critical period and positively correlated with the time length of the corresponding attribute network performance reference data.

[0091] When multiple reference data slices are obtained from multiple attribute network performance reference data at multiple historical sampling moments, each reference data slice of the corresponding attribute network performance reference data can be decomposed using multiple key periods of the same attribute network performance reference data, and the decomposition result of each reference data slice of each attribute network performance reference data can be regarded as a reference feature.

[0092] Assume that the time length of a reference data slice is L, and the k key periods of the attribute network performance are expressed as [T1,...,T k ], using the k key periods of network performance [T1,...,T k ] and the time length L of the reference data slice, each reference decomposition parameter of the attribute network performance reference data can be determined as represents the number of scales of the reference feature, and T represents the length of the reference feature of the attribute network performance in the time dimension, which can be equal to the time length of the key period corresponding to the attribute network performance. In this case, each reference data slice can be decomposed into k slices of size The feature tensor [X1,...,X k ], where X1 represents the reference feature corresponding to the first key feature, and its size is X k Represents the reference feature corresponding to the kth key feature, and its size is

[0093] In an optional manner, the feature fusion unit is used to extract features from the multi-scale reference feature sequence of each attribute network performance to obtain a multi-scale deep reference feature sequence, and based on the multi-scale deep reference feature sequence, determine the multi-scale reference fusion feature sequence of each attribute network performance.

[0094] For example, taking a reference data slice as an example, after the reference data slice is decomposed into k key cycles, k pieces of size are obtained. The feature tensor [X1,...,X k ], we can use convolution kernels of different sizes to filter k convolution kernels of different sizes. The feature tensor [X1,...,X k ] performs multi-layer feature extraction to obtain multi-scale deep reference feature sequence fragments, and then performs one-dimensional expansion and splicing based on the multi-scale deep reference feature sequence fragments to determine the multi-scale reference fusion feature sequence fragments of each attribute network.

[0095] Among them, the process of convolution extraction can be expressed as Represents the feature extraction result of the lth layer, Represents the feature extraction result of the l-1 layer, a represents the serial number of the reference data slice, and each attribute network multi-scale reference fusion feature slice X a Represents the multi-scale deep reference feature sequence fragment corresponding to the a-th reference data slice.

[0096] In an optional embodiment, the adjacency matrix of the graph attention model includes distances between pairwise attribute network performance reference data. The adjacency matrix can be determined by the k-nearest neighbor principle, i.e., the k-nearest neighbor algorithm. For example, the adjacency matrix can include connection weights between pairwise attribute network performance reference data.

[0097] For example, the Euclidean distance between a certain attribute network performance reference data and other attribute network performance reference data can be calculated, and the top k attribute network performance reference data with the Euclidean distance can be taken, and the connection weight between the attribute network performance reference data and the top k attribute network performance reference data can be set to 1.

[0098] For example, the method for determining the Euclidean distance between the two attribute network performance reference data may include: assuming that the i-th attribute network performance reference data is represented by x i , which may include the i-th attribute network performance reference data at one or more historical sampling moments, and the j-th attribute network performance reference data is represented by x j , which may include the jth attribute network performance reference data at one or more historical sampling moments, which can be obtained through d ij =||x i -x j ||2 Calculate the Euclidean distance d between the i-th attribute network performance reference data and the j-th attribute network performance reference data ij .

[0099] In order to better learn the topological relationship between each channel, the exemplary embodiment of the present disclosure introduces an attention mechanism to adaptively weight the node features. The degree matrix of the graph attention model of the exemplary embodiment of the present disclosure can be determined by the similarity between each node feature and the neighboring node features. For example: based on the adjacency matrix of the graph attention model, the k attribute network performance reference data adjacent to the i-th attribute network performance reference data can be obtained, and then the similarity coefficient between the i-th attribute network performance reference data and the k adjacent attribute network performance reference data can be calculated. Finally, the similarity coefficient between the i-th attribute network performance reference data and the k adjacent attribute network performance reference data is normalized so that it is mapped between 0 and 1, thereby obtaining the attention coefficient between the current node feature and the k adjacent node features.

[0100] Assuming that the j-th attribute network performance reference data is the neighbor node feature of the i-th attribute network performance reference data, it can be obtained through a ij =softmax(F(Wx i ,Wx j )) Determine the attention coefficient a between the i-th attribute network performance reference data and the j-th attribute network performance reference data ij .

[0101] F(x) represents a similarity function, which can be a cosine function or a tangent function, and is used to calculate the similarity between the features of two adjacent nodes, that is, to calculate the similarity between the network performance reference data of the i-th attribute and the network performance reference data of the j-th attribute. W1 represents a shared parameter that needs to be updated during the training process. Softmax(·) represents a normalized exponential function, which can be used to convert F(W1x i ,W1x j ) is mapped to between 0 and 1. Let w ij =F(W1x i ,W1x j ),but k represents the total number of attributes of the network performance reference data adjacent to the i-th attribute network performance reference data, w ir represents the similarity between the i-th attribute network performance reference data and the r-th attribute network performance reference data, w ij It represents the similarity between the i-th attribute network performance reference data and the j-th attribute network performance reference data.

[0102] After obtaining the adjacency matrix and attention coefficient matrix of the graph attention model, we can perform multiple message transfers on the multi-scale reference fusion feature sequence of network performance with multiple attributes to obtain a multi-scale deep reference feature sequence of network performance with multiple attributes. Each message transfer essentially aggregates and extracts the node features from the previous message transfer.

[0103] Taking the i-th attribute network performance reference data as an example, we can Indicates, l indicates the current number of message delivery, represents the extraction result of the i-th attribute network performance reference data after the l-1-th message transmission, represents the extraction result of the j-th attribute network performance reference data after the l-1-th message transmission, a ii represents the attention coefficient between the i-th attribute network performance reference data and the i-th attribute network performance reference data, a ij It represents the attention coefficient between the i-th attribute network performance reference data and the j-th attribute network performance reference data, and W2 represents the weight parameter, which needs to be updated during the training process.

[0104] Finally, the multi-scale deep reference feature sequences of multiple attribute network performances are fully connected through the fully connected layer, so as to obtain the reconstructed data prediction results G' of the multiple attribute network performances with the same size as the multi-scale reference fusion feature sequence G of the multiple attribute network performances.

[0105] The same size here means that when the multi-scale reference fusion feature sequence G of multiple attribute network performance includes the multi-scale reference fusion features of the attribute network performance corresponding to m historical sampling moments, the reconstructed data prediction result G' of the multiple attribute network performance includes the reconstructed data prediction results of the attribute network performance corresponding to m historical sampling moments. For example: the number of attributes of the multi-scale reference fusion feature sequence of network performance is C, and the reconstructed data prediction result G' of the multiple attribute network performance is {v'1,...,v' C}, v'1 represents the reconstructed data prediction result of the first attribute network performance, v' C Represents the prediction results of the reconstructed data of the C-th attribute network performance.

[0106] When the reconstructed data prediction results of the network performance of multiple attributes are obtained, the data reconstruction loss of the graph attention model can be determined based on the reconstructed data prediction results of the network performance of multiple attributes and the reference data of the network performance of multiple attributes at multiple historical sampling moments.

[0107] If the data reconstruction loss of the graph attention model is less than or equal to the preset reconstruction loss, the graph attention model is determined to be a performance data reconstruction model. Since the graph attention model has already been trained, when the data reconstruction loss of the graph attention model is less than or equal to the preset reconstruction loss, the reference data reconstruction error of the target attribute network performance at multiple historical sampling moments can also be determined based on the reconstructed data prediction results of the target attribute network performance and the target attribute network performance reference data at multiple historical sampling moments.

[0108] If the data reconstruction loss of the graph attention model is greater than the preset reconstruction loss, the model parameters of the multi-scale feature extraction model and the model parameters of the graph attention model are updated. When the exemplary embodiment of the present disclosure can be divided into a multi-scale decomposition unit and a feature fusion unit according to function, the model parameters of the multi-scale feature extraction model may refer to the model parameters involved in the feature aggregation unit, and the model parameters of the graph attention model may refer to the model parameters used in the message passing process, such as the shared parameter W1 and the weight parameter W2, and may also refer to the weight parameters of the fully connected layer.

[0109] As a possible implementation, Figure 4 FIG. 1 shows a schematic diagram of a process for obtaining reconstruction errors of original test data according to an exemplary embodiment of the present disclosure. Figure 4 As shown, the exemplary embodiment of the present disclosure determines the reconstruction error of the original test data of the target attribute network performance at multiple test times through the performance data reconstruction model, which may include:

[0110] Step 401: inputting the original test data of the target attribute network performance at multiple test moments into a multi-scale feature extraction model to determine a multi-scale original fusion feature sequence of the target attribute network performance.

[0111] In practical applications, the exemplary embodiment of the present disclosure can collect original test data of the target attribute network performance at multiple test times from the target network, first perform missing value supplementation and data standardization on the original test data of the target attribute network performance at multiple test times, and then input it into the multi-scale feature training model to determine the multi-scale original fusion feature sequence of the target attribute network performance.

[0112] Exemplarily, the missing value supplementation and data standardization processing of the reference data of network performance of multiple attributes can be referred to, and the missing value supplementation and data standardization processing can be performed on the original test data of the target attribute network performance at multiple test times. It should be understood that when the missing value supplementation is performed on the original test data of the target attribute network performance at multiple test times, there is no need to perform de-extreme value processing on the original test data of the target attribute network performance at multiple test times.

[0113] Exemplarily, when the multi-scale feature extraction model includes a multi-scale decomposition unit and a feature fusion unit, the multi-scale decomposition unit is used to decompose the original test data of the target attribute network performance at multiple test times using multiple key periods of the target attribute network performance to obtain a multi-scale original feature sequence of the target attribute network performance, and the feature fusion unit is used to determine the multi-scale original fusion feature sequence of the target attribute network performance based on the multi-scale original feature sequence of the target attribute network performance.

[0114] Considering that the original test data of the target attribute network performance at multiple test times and the reference data of multiple attribute network performance are all collected from the same target network, the multiple key periods of the target attribute network performance of the exemplary embodiment of the present disclosure can refer to the relevant description above and will not be repeated here.

[0115] Step 402: Input the multi-scale fusion feature sequence of the target attribute network into the performance data reconstruction model to determine the reconstruction results of the original test data of the target attribute network performance at multiple test moments.

[0116] In practical applications, multiple original data slices are obtained from the original test data of the target attribute network performance at multiple test times, the multiple original data slices are input into a multi-scale feature extraction model, and multiple multi-scale original fusion feature slices are determined, so that each multi-scale original fusion feature slice includes the original test data of the target attribute network performance at at least two test times, and then one of the multi-scale original fusion feature slices is input into the performance data reconstruction model to obtain the reconstruction result of the multi-scale original fusion feature slice, that is, the reconstruction result of the original test data of the target attribute network performance at multiple test times.

[0117] Exemplarily, the performance data reconstruction model can perform multiple message transmissions on the multi-scale fusion feature sequence of the target attribute network to obtain a multi-scale deep original feature sequence of the target attribute network performance, and then use a fully connected layer to fully connect the multi-scale deep original feature sequence of the target attribute network performance, thereby obtaining the reconstruction results of the original test data of the target attribute network performance at multiple test times. Here, the size of the reconstruction results of the original test data of the target attribute network performance at multiple test times is the same as the size of the original test data of the target attribute network performance at multiple test times, that is, when the original test data of the target attribute network performance at multiple test times includes the original test data of the target attribute network performance at m test times, the reconstruction results of the original test data of the target attribute network performance at multiple test times include the reconstruction results of the original test data of the target attribute network performance at m test times.

[0118] Step 403: Determine the target attribute network performance original test data reconstruction error based on the target attribute network performance original test data reconstruction results at multiple test times and the target attribute network performance original test data.

[0119] As a possible implementation, Figure 5 The following is a flow chart showing the acquisition process of abnormal data monitoring parameters of target attribute network performance according to an exemplary embodiment of the present disclosure. Figure 5 As shown, the exemplary embodiment of the present disclosure determines abnormal data monitoring parameters of the target attribute network performance based on the reconstruction errors of the target attribute network performance reference data at multiple historical sampling moments and the reconstruction errors of the original test data of the target attribute network performance at multiple test moments, which may include:

[0120] Step 501: determining a cumulative probability distribution of reconstruction errors based on reconstruction errors of target attribute network performance reference data at multiple historical sampling moments and reconstruction errors of original measured data of target attribute network performance;

[0121] In practical applications, the reconstruction errors of the target attribute network performance reference data at multiple historical sampling moments and the reconstruction errors of the original test data of the target attribute network performance at multiple test moments can be taken as a reconstruction error set. The probability density function is then used to estimate the probability distribution of the reconstruction error set, thereby obtaining the probability density distribution of the reconstruction error. Based on the probability density distribution of the reconstruction error, the cumulative probability distribution of the reconstruction error is determined.

[0122] For example, the mean square error (MSE) of the target attribute network performance at multiple historical sampling moments can be reconstructed using the reference data. train and the mean square error (MSE) of the target attribute network performance at multiple historical sampling moments of the reference data reconstruction testAs the reconstruction error set {ε test ,ε train}, then determine the probability density distribution of the reconstruction error based on the reconstruction error set, and then use the probability density distribution of the reconstruction error to find the integral, so as to obtain the cumulative probability density of the reconstruction error.

[0123] When the kernel function of the probability density function is a Gaussian kernel, the probability density function The form is expressed as: Among them, the kernel function φ(ε) of the Gaussian kernel can be expressed as Bandwidth h = 1.059*σ*N -0.2 , ε represents the element variable of the reconstruction error set, ε i represents the i-th element in the reconstruction error set, N represents the number of elements in the reconstruction error set, and σ represents the standard deviation of the elements in the reconstruction error set.

[0124] Step 502: Based on the cumulative probability distribution of the reconstruction error and the cumulative probability threshold, determine the abnormal data monitoring parameter for the target attribute network performance data. For example, if the cumulative probability of a reconstruction error in the reconstruction error set is greater than or equal to the cumulative probability threshold, it can be determined that the reconstruction error is within the reconstruction error tolerance limit, and the reconstruction error can be used as the abnormal data monitoring parameter for the target attribute network performance data. For example, the reconstruction error at the 0.3% quantile can be used as the reconstruction error tolerance limit or the abnormal data monitoring parameter for the target attribute network performance data.

[0125] Once the reconstruction error limit is determined, the abnormal reconstruction error in the reconstruction error set that exceeds the limit can be used to determine the abnormal original test data and the test time corresponding to the abnormal original test data. Furthermore, the greater the difference between the abnormal reconstruction error and the reconstruction error limit, the more severe the abnormality of the original test data.

[0126] One or more technical solutions provided in the exemplary embodiments of the present disclosure take into account that there may be certain intrinsic connections between network performance data of multiple attributes. A graph attention model can be trained based on multiple attribute network performance reference data at multiple historical sampling moments, fully exploring the deep connections between different attribute network performance reference data, so that the obtained performance data reconstruction model can not only be applied to the reconstruction of original data of network performance of multiple attributes, but also improve the reconstruction accuracy. The training data of the graph attention model is multiple attribute network performance reference data at multiple historical sampling moments. Therefore, the reconstruction error of the reference data of the target attribute network performance obtained at multiple historical sampling moments is relatively small, while the reconstruction error of the original test data of the target attribute network performance determined by the performance data reconstruction model at multiple test moments is relatively large.

[0127] Moreover, the exemplary embodiment of the present disclosure can determine the abnormal data monitoring parameters of the target attribute network performance data based on the reference data reconstruction error of the target attribute network performance at multiple historical sampling moments and the reconstruction error of the original test data of the target attribute network performance at multiple test moments. In essence, the reference data reconstruction error of the target attribute network performance at a large number of historical sampling moments is used as the normal data sequence of the target attribute network performance, and the original test data reconstruction error of the target attribute network performance at multiple test moments and the normal data sequence of the target attribute network performance are regarded as a set, so as to determine the data reconstruction error that meets the allowable limit of the reconstruction error, and use the data reconstruction error that meets the allowable limit of the reconstruction error as the abnormal data monitoring parameter of the target attribute network performance. It can be seen that the exemplary embodiment of the present disclosure combines the normal data sequence of the target attribute network performance and the reconstruction error of the original test data at multiple test moments to determine the abnormal data monitoring parameters, so that the determined abnormal data monitoring parameters can be used to specifically analyze whether the original test data of the target attribute network performance has an abnormality and the time when the abnormality occurs.

[0128] In addition, in the process of training the graph attention model, the method of the exemplary embodiment of the present invention uses training data including multiple attribute network performance reference data at multiple historical sampling moments. The graph attention model can reconstruct multiple attribute network performance reference data at multiple historical sampling moments. Therefore, when training the graph attention model, there is no need to mark outliers or manually set thresholds for each attribute network performance reference data at multiple historical sampling moments, thereby reducing investment costs and having higher practicality.

[0129] The above mainly introduces the solution provided by the embodiment of the present disclosure from the perspective of the server. It can be understood that in order to realize the above functions, the server of the exemplary embodiment of the present disclosure includes hardware structures and / or software modules corresponding to the execution of each function. It should be easy for those skilled in the art to realize that, in combination with the units and algorithm steps of each example described in the embodiment disclosed herein, the present disclosure can be implemented in the form of hardware or a combination of hardware and computer software. Whether a function is executed in the form of hardware or computer software driving hardware depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present disclosure.

[0130] The embodiments of the present disclosure can divide the server into functional units according to the above method examples. For example, each functional module can be divided according to each function, or two or more functions can be integrated into one processing module. The above integrated modules can be implemented in the form of hardware or in the form of software functional modules. It should be noted that the division of modules in the embodiments of the present disclosure is schematic and is only a logical functional division. In actual implementation, there may be other division methods.

[0131] In the case of dividing each functional module according to each function, an exemplary embodiment of the present disclosure provides a device for determining abnormal data monitoring parameters, which may be a server or a chip applied to a server. Figure 6 FIG. 1 shows a schematic block diagram of functional modules of a device for determining abnormal data monitoring parameters according to an exemplary embodiment of the present disclosure. Figure 6 As shown, the abnormal data monitoring parameter determination device 600 includes:

[0132] A training module 601 is configured to train a graph attention model based on multiple attribute network performance reference data at multiple historical sampling moments to obtain a performance data reconstruction model;

[0133] Determination module 602 is used to obtain the reference data reconstruction error of the target attribute network performance at multiple historical sampling moments, determine the reconstruction error of the original test data of the target attribute network performance through the performance data reconstruction model, and determine the abnormal data monitoring parameters of the target attribute network performance data based on the reference data reconstruction error of the target attribute network performance at multiple historical sampling moments and the original test data of the target attribute network performance at multiple test moments, and the abnormal data monitoring parameters of the target attribute network performance data meet the allowable limit of the reconstruction error.

[0134] As a possible implementation method, the device also includes a preprocessing module 603, which is used to determine the allowable distribution range of reference data of each attribute network performance based on the original historical data of each attribute network performance at multiple historical sampling moments; and determine multiple attribute network performance reference data based on the allowable distribution range of reference data of multiple attribute network performances and the original data of multiple attribute network performances at multiple historical sampling moments.

[0135] As a possible implementation method, the preprocessing module is used to determine that the original data of the attribute network performance is the reference data corresponding to the attribute network performance if the original data of the attribute network performance is within the allowable distribution range of the reference data corresponding to the attribute network performance; if the original data of the attribute network performance is outside the allowable distribution range of the reference data corresponding to the attribute network performance, correct the original data of the attribute network performance to obtain the reference data corresponding to the attribute network performance.

[0136] As a possible implementation, the training module 601 is used to input the multiple attribute network performance reference data at multiple historical sampling moments into a multi-scale feature extraction model to determine a multi-scale reference fusion feature sequence of the multiple attribute network performance;

[0137] Input a multi-scale reference fusion feature sequence of multiple attribute network performances into a graph attention model to obtain reconstruction data prediction results of the multiple attribute network performances; determine the data reconstruction loss of the graph attention model based on the reconstruction data prediction results of the multiple attribute network performances and the multiple attribute network performance reference data at multiple historical sampling moments; if the data reconstruction loss of the graph attention model is less than or equal to the preset reconstruction loss, determine that the graph attention model is a performance data reconstruction model; otherwise, update the model parameters of the multi-scale feature extraction model and the model parameters of the graph attention model.

[0138] As a possible implementation method, the multi-scale reference fusion feature sequence of the multi-attribute network performance includes: multiple multi-scale reference fusion feature slices, and the device also includes a preprocessing module 603, which is used to obtain multiple reference data slices from the multi-attribute network performance reference data of the multiple historical sampling moments, and each reference data slice includes at least two multi-attribute network performance reference data of the historical sampling moments; the multiple reference data slices are input into the multi-scale feature extraction model to determine multiple multi-scale reference fusion feature slices, and each multi-scale reference fusion feature slice includes the multi-scale reference fusion features of the multi-attribute network performance of at least two historical sampling moments.

[0139] As a possible implementation, the multi-scale feature extraction model includes a multi-scale decomposition unit and a feature fusion unit;

[0140] The multi-scale decomposition unit is used to decompose the reference data of each attribute network performance at each historical sampling moment using multiple key periods of each attribute network performance to obtain a multi-scale reference feature sequence of each attribute network performance;

[0141] The feature fusion unit is configured to determine a multi-scale reference fusion feature sequence for each attribute network performance based on the multi-scale reference feature sequence for each attribute network performance.

[0142] As a possible implementation manner, the multi-scale decomposition unit is further configured to:

[0143] Determining spectrum information of each attribute network performance reference data based on each attribute network performance reference data at multiple historical sampling moments;

[0144] determining a key amplitude of each attribute network performance reference data based on spectrum information of each attribute network performance reference data;

[0145] Based on the key amplitude of each attribute network performance reference data, multiple key periods of each attribute network performance are obtained.

[0146] As a possible implementation manner, the multi-scale decomposition unit is used to:

[0147] Determining the time length of each attribute network performance reference data based on the plurality of attribute network performance reference data at the historical sampling moments;

[0148] Determining each reference decomposition parameter of each attribute network performance reference data based on each key period of each attribute network performance and a time length of each attribute network performance reference data, wherein the reference decomposition parameter corresponds to the key period;

[0149] Each attribute network performance reference data at multiple historical sampling moments is decomposed based on multiple decomposition parameters of each attribute network performance reference data to obtain a multi-scale reference feature sequence for each attribute network performance.

[0150] As a possible implementation method, each of the decomposition parameters includes the length of the reference feature in the time dimension and the number of scales of the reference feature. The length of the reference feature in the time dimension is the same as the time length of the corresponding critical period, and the number of scales of the reference feature is determined by the time length of the corresponding attribute network performance reference data and the time length of the corresponding critical period.

[0151] As a possible implementation method, the feature fusion unit is used to extract features from the multi-scale reference feature sequence of each attribute network performance to obtain a multi-scale deep reference feature sequence, and based on the multi-scale deep reference feature sequence, determine the multi-scale reference fusion feature sequence of each attribute network performance.

[0152] As a possible implementation method, the determination module 602 is used to determine the reference data reconstruction error of the target attribute network performance at multiple historical sampling moments based on the reconstruction data prediction results of the target attribute network performance and the target attribute network performance reference data at multiple historical sampling moments if the data reconstruction loss of the graph attention model is less than or equal to the preset reconstruction loss.

[0153] As a possible implementation, the determination module 602 is configured to input the original test data of the target attribute network performance at multiple test times into the multi-scale feature extraction model to determine the multi-scale original fusion feature sequence of the target attribute network performance; input the multi-scale original fusion feature sequence of the target attribute network performance into the performance data reconstruction model to determine the reconstruction result of the original test data of the target attribute network performance at multiple test times;

[0154] Based on the reconstruction results of the original data to be measured of the target attribute network performance at multiple test moments and the original data to be measured of the target attribute network performance, a reconstruction error of the original data to be measured of the target attribute network performance is determined.

[0155] As a possible implementation method, the determination module 602 is used to determine the cumulative probability distribution of the reconstruction error based on the reconstruction error of the target attribute network performance reference data at multiple historical sampling moments and the reconstruction error of the original test data of the target attribute network performance; based on the probability distribution of the reconstruction error and the cumulative probability threshold, determine the abnormal data monitoring parameters of the target attribute network performance data.

[0156] Figure 7 FIG. 1 shows a schematic block diagram of a chip according to an exemplary embodiment of the present disclosure. Figure 7 As shown, the chip 700 includes one or more (including two) processors 701 and a communication interface 702. The communication interface 702 can support the server to execute the data sending and receiving steps in the above method, and the processor 701 can support the server to execute the data processing steps in the above method.

[0157] Optional, such as Figure 7 As shown, the chip 700 also includes a memory 703, which may include a read-only memory and a random access memory, and provides operation instructions and data to the processor. Part of the memory may also include a non-volatile random access memory (NVRAM).

[0158] In some embodiments, as Figure 7As shown, the processor 701 performs corresponding operations by calling the operation instructions stored in the memory (the operation instructions may be stored in the operating system). The processor 701 controls the processing operations of any one of the terminal devices, and the processor may also be called a central processing unit (CPU). The memory 703 may include a read-only memory and a random access memory, and provides instructions and data to the processor 701. A portion of the memory 703 may also include NVRAM. For example, in an application, the memory, the communication interface, and the memory are coupled together through a bus system, wherein the bus system may include a power bus, a control bus, and a status signal bus in addition to a data bus. However, for the sake of clarity, in Figure 7 Various buses are labeled as bus system 704 .

[0159] The methods disclosed in the above embodiments of the present disclosure can be applied to or implemented by a processor. The processor may be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method can be completed by hardware integrated logic circuits in the processor or by software instructions. The above processor may be a general-purpose processor, a digital signal processor (DSP), an ASIC, a field-programmable gate array (FPGA), or other programmable logic device, a discrete gate or transistor logic device, or a discrete hardware component. The methods, steps, and logic block diagrams disclosed in the embodiments of the present disclosure can be implemented or executed. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the methods disclosed in conjunction with the embodiments of the present disclosure can be directly implemented and executed by a hardware decoding processor, or by a combination of hardware and software modules in the decoding processor. The software module can be located in a storage medium well-known in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, etc. The storage medium is located in the memory, and the processor reads the information in the memory and, in conjunction with its hardware, completes the steps of the above method.

[0160] The exemplary embodiments of the present disclosure further provide an electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor. The memory stores a computer program executable by the at least one processor, the computer program being configured to cause the electronic device to perform a method according to an exemplary embodiment of the present disclosure when executed by the at least one processor.

[0161] Exemplary embodiments of the present disclosure further provide a non-transitory computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor of a computer, is used to cause the computer to perform a method according to an embodiment of the present disclosure.

[0162] Exemplary embodiments of the present disclosure further provide a computer program product, including a computer program, wherein when the computer program is executed by a processor of a computer, it is used to cause the computer to perform the method according to the embodiment of the present disclosure.

[0163] refer to Figure 8 , a block diagram of an electronic device 800 that can serve as a server or client of the present disclosure will now be described, which is an example of a hardware device that can be applied to various aspects of the present disclosure. The electronic device is intended to represent various forms of digital electronic computer devices, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or required herein.

[0164] like Figure 8 As shown, electronic device 800 includes a computing unit 801, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 802 or a computer program loaded from a storage unit 808 into a random access memory (RAM) 803. Various programs and data required for the operation of device 800 can also be stored in RAM 803. Computing unit 801, ROM 802, and RAM 803 are connected to each other via a bus 804. An input / output (I / O) interface 805 is also connected to bus 804.

[0165] like Figure 8As shown, multiple components within electronic device 800 are connected to I / O interface 805, including an input unit 806, an output unit 807, a storage unit 808, and a communication unit 809. Input unit 806 can be any type of device capable of inputting information into electronic device 800. Input unit 806 can receive input digital or character information and generate key input signals related to user settings and / or function control of the electronic device. Output unit 807 can be any type of device capable of presenting information and may include, but is not limited to, a display, a speaker, a video / audio output terminal, a vibrator, and / or a printer. Storage unit 808 may include, but is not limited to, a magnetic disk or an optical disk. Communication unit 809 allows electronic device 800 to exchange information / data with other devices via computer networks such as the Internet and / or various telecommunication networks and may include, but is not limited to, a modem, a network card, an infrared communication device, a wireless communication transceiver and / or a chipset, such as a Bluetooth™ device, a WiFi device, a WiMax device, a cellular communication device, and / or the like.

[0166] like Figure 8 As shown, the computing unit 801 can be various general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 801 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units that run machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 801 performs the various methods and processes described above. For example, in some embodiments, the method of the exemplary embodiments of the present disclosure may be implemented as a computer software program, which is tangibly included in a machine-readable medium, such as a storage unit 808. In some embodiments, part or all of the computer program can be loaded and / or installed on the electronic device 800 via the ROM 802 and / or the communication unit 809. In some embodiments, the computing unit 801 can be configured to perform the method of the exemplary embodiments of the present disclosure in any other appropriate manner (e.g., by means of firmware).

[0167] The program code for implementing the method of the present disclosure can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device so that when the program code is executed by the processor or controller, the functions / operations specified in the flow chart and / or block diagram are implemented. The program code can be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0168] In the context of the present disclosure, a machine-readable medium can be a tangible medium that can contain or store a program for use by or in conjunction with an instruction execution system, device or equipment. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium can include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0169] As used in this disclosure, the terms "machine-readable medium" and "computer-readable medium" refer to any computer program product, apparatus, and / or device (e.g., a magnetic disk, an optical disk, a memory, a programmable logic device (PLD)) for providing machine instructions and / or data to a programmable processor, including machine-readable media that receive machine instructions as machine-readable signals. The term "machine-readable signal" refers to any signal used to provide machine instructions and / or data to a programmable processor.

[0170] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0171] The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer having a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network (LAN), a wide area network (WAN), and the Internet.

[0172] Computer systems may include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The client and server relationship arises through computer programs running on the respective computers and having a client-server relationship to each other.

[0173] In the above embodiments, all or part of the embodiments can be implemented by software, hardware, firmware or any combination thereof. When implemented by software, all or part of the embodiments can be implemented in the form of a computer program product. The computer program product includes one or more computer programs or instructions. When the computer program or instructions are loaded and executed on a computer, the process or function described in the embodiment of the present disclosure is performed in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, a terminal, a user device or other programmable device. The computer program or instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer program or instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center via wired or wireless means. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium, such as a floppy disk, a hard disk, or a tape; it can also be an optical medium, such as a digital video disc (DVD); it can also be a semiconductor medium, such as a solid state drive (SSD).

[0174] Although the present disclosure has been described with reference to specific features and embodiments thereof, it will be apparent that various modifications and combinations may be made thereto without departing from the spirit and scope of the present disclosure. Accordingly, this specification and the drawings are merely illustrative of the present disclosure as defined by the appended claims and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present disclosure. Obviously, those skilled in the art may make various modifications and variations to the present disclosure without departing from the spirit and scope of the present disclosure. Thus, the present disclosure is intended to include such modifications and variations if they fall within the scope of the claims of the present disclosure and their equivalents.

Claims

1. A method for determining abnormal data monitoring parameters, characterized in that: include: A graph attention model is trained based on network performance reference data with multiple attributes at multiple historical sampling moments to obtain a performance data reconstruction model. Obtaining a reference data reconstruction error of the target attribute network performance at a plurality of the historical sampling moments; Determine the reconstruction error of the original test data of the target attribute network performance at multiple test moments by using the performance data reconstruction model; Determining abnormal data monitoring parameters of target attribute network performance data based on a reconstruction error of reference data of the target attribute network performance at a plurality of the historical sampling moments and a reconstruction error of original test data of the target attribute network performance at a plurality of test moments, wherein the abnormal data monitoring parameters of the target attribute network performance data meet an allowable reconstruction error limit; The method of training a graph attention model based on multiple attribute network performance reference data at multiple historical sampling moments to obtain a performance data reconstruction model includes: Inputting the multi-attribute network performance reference data at the plurality of historical sampling moments into a multi-scale feature extraction model to determine a multi-scale reference fusion feature sequence of the multi-attribute network performance; Inputting a multi-scale reference fusion feature sequence of the plurality of attribute network performances into a graph attention model to obtain reconstructed data prediction results of the plurality of attribute network performances; Determining the data reconstruction loss of the graph attention model based on the reconstruction data prediction results of the multiple attribute network performances and the multiple attribute network performance reference data at the multiple historical sampling moments; If the data reconstruction loss of the graph attention model is less than or equal to the preset reconstruction loss, the graph attention model is determined to be a performance data reconstruction model; otherwise, the model parameters of the multi-scale feature extraction model and the model parameters of the graph attention model are updated; The multi-scale feature extraction model includes a multi-scale decomposition unit and a feature fusion unit; the multi-scale decomposition unit is used to decompose the reference data of each attribute network performance at each historical sampling moment using multiple key periods of each attribute network performance to obtain a multi-scale reference feature sequence for each attribute network performance; the feature fusion unit is used to determine the multi-scale reference fusion feature sequence for each attribute network performance based on the multi-scale reference feature sequence of each attribute network performance.

2. The method according to claim 1, characterized in that The method further comprises: Based on the original historical data of each attribute network performance at multiple historical sampling moments, determining the allowable distribution range of the reference data of each attribute network performance; Based on the allowed distribution range of the reference data of the multiple attribute network performances and the original data of the multiple attribute network performances at multiple historical sampling moments, the multiple attribute network performance reference data are determined.

3. The method according to claim 2, characterized in that Determining the multiple attribute network performance reference data based on the allowable distribution range of the multiple attribute network performance reference data and the multiple attribute network performance original data at the multiple historical sampling moments includes: If the original data of the attribute network performance is within the allowable distribution range of the reference data corresponding to the attribute network performance, determining that the original data of the attribute network performance is the reference data corresponding to the attribute network performance; If the original data of the attribute network performance is outside the allowable distribution range of the reference data corresponding to the attribute network performance, the original data of the attribute network performance is corrected to obtain the reference data corresponding to the attribute network performance.

4. The method according to claim 1, wherein The multi-scale reference fusion feature sequence of the network performance of multiple attributes includes: multiple multi-scale reference fusion feature slices, and the multi-scale reference fusion feature sequence of the network performance of multiple attributes at multiple historical sampling moments is input into the multi-scale feature extraction model to determine the multi-scale reference fusion feature sequence of the network performance of multiple attributes, including: Acquire multiple reference data slices from the multiple attribute network performance reference data at the multiple historical sampling moments, each reference data slice including the multiple attribute network performance reference data at at least two historical sampling moments; Inputting a plurality of the reference data slices into the multi-scale feature extraction model, and determining a plurality of the multi-scale reference fusion feature slices, each of the multi-scale reference fusion feature slices including multi-scale reference fusion features of the network performance of multiple attributes at at least two of the historical sampling moments.

5. The method according to claim 1, wherein The multi-scale decomposition unit is further configured to: Determining spectrum information of each attribute network performance reference data based on each attribute network performance reference data at multiple historical sampling moments; determining a key amplitude of each attribute network performance reference data based on spectrum information of each attribute network performance reference data; Based on the key amplitude of each attribute network performance reference data, multiple key periods of each attribute network performance are obtained.

6. The method according to claim 1, characterized in that The multi-scale decomposition unit is used to: Determining the time length of each attribute network performance reference data based on the plurality of attribute network performance reference data at the historical sampling moments; Determining each reference decomposition parameter of each attribute network performance reference data based on each key period of each attribute network performance and a time length of each attribute network performance reference data, wherein the reference decomposition parameter corresponds to the key period; Each attribute network performance reference data at multiple historical sampling moments is decomposed based on multiple decomposition parameters of each attribute network performance reference data to obtain a multi-scale reference feature sequence for each attribute network performance.

7. The method according to claim 6, characterized in that Each of the decomposition parameters includes the length of the reference feature in the time dimension and the number of scales of the reference feature. The length of the reference feature in the time dimension is the same as the time length of the corresponding critical period, and the number of scales of the reference feature is determined by the time length of the corresponding attribute network performance reference data and the time length of the corresponding critical period.

8. The method according to claim 1, characterized in that The feature fusion unit is used to extract features from the multi-scale reference feature sequence of each attribute network performance to obtain a multi-scale deep reference feature sequence, and determine a multi-scale reference fusion feature sequence for each attribute network performance based on the multi-scale deep reference feature sequence.

9. The method according to claim 1, characterized in that The obtaining of the reference data reconstruction error of the target attribute network performance at the plurality of historical sampling moments includes: If the data reconstruction loss of the graph attention model is less than or equal to the preset reconstruction loss, based on the reconstructed data prediction results of the target attribute network performance and the reference data of the target attribute network performance at multiple historical sampling moments, the reference data reconstruction error of the target attribute network performance at multiple historical sampling moments is determined.

10. The method according to claim 1, characterized in that The determining, by the performance data reconstruction model, the original test data reconstruction error of the target attribute network performance at multiple test moments includes: Inputting the original test data of the target attribute network performance at multiple test moments into the multi-scale feature extraction model to determine the multi-scale original fusion feature sequence of the target attribute network performance; Inputting the multi-scale original fusion feature sequence of the target attribute network into the performance data reconstruction model to determine the reconstruction results of the original test data of the target attribute network performance at multiple test times; Based on the reconstruction results of the original data to be measured of the target attribute network performance at multiple test moments and the original data to be measured of the target attribute network performance, a reconstruction error of the original data to be measured of the target attribute network performance is determined.

11. The method according to any one of claims 1 to 10, characterized in that The determining of abnormal data monitoring parameters of the target attribute network performance based on the reconstruction errors of the target attribute network performance reference data at the plurality of historical sampling moments and the reconstruction errors of the original test data of the target attribute network performance at the plurality of test moments includes: Determining a cumulative probability distribution of reconstruction errors based on reconstruction errors of target attribute network performance reference data at multiple historical sampling moments and reconstruction errors of original measured data of the target attribute network performance; Based on the probability distribution of the reconstruction error and the cumulative probability threshold, abnormal data monitoring parameters of the target attribute network performance data are determined.

12. A device for determining abnormal data monitoring parameters, characterized in that: include: A training module is used to input the multi-attribute network performance reference data at multiple historical sampling moments into a multi-scale feature extraction model to determine a multi-scale reference fusion feature sequence of the multi-attribute network performance; Inputting a multi-scale reference fusion feature sequence of the plurality of attribute network performances into a graph attention model to obtain reconstructed data prediction results of the plurality of attribute network performances; Determine the data reconstruction loss of the graph attention model based on the reconstruction data prediction results of the multiple attribute network performances and the multiple attribute network performance reference data at the multiple historical sampling moments; if the data reconstruction loss of the graph attention model is less than or equal to the preset reconstruction loss, determine that the graph attention model is a performance data reconstruction model; otherwise, update the model parameters of the multi-scale feature extraction model and the model parameters of the graph attention model; The multi-scale feature extraction model includes a multi-scale decomposition unit and a feature fusion unit; the multi-scale decomposition unit is used to decompose the reference data of each attribute network performance at each historical sampling moment using multiple key periods of each attribute network performance to obtain a multi-scale reference feature sequence for each attribute network performance; The feature fusion unit is used to determine a multi-scale reference fusion feature sequence for each attribute network performance based on the multi-scale reference feature sequence of each attribute network performance; A determination module is used to obtain the reference data reconstruction error of the target attribute network performance at multiple historical sampling moments, determine the reconstruction error of the original test data of the target attribute network performance through the performance data reconstruction model, and determine the abnormal data monitoring parameters of the target attribute network performance data based on the reference data reconstruction error of the target attribute network performance at multiple historical sampling moments and the original test data of the target attribute network performance at multiple test moments, wherein the abnormal data monitoring parameters of the target attribute network performance data meet the allowable limit of the reconstruction error.

13. An electronic device, characterized in that: include: processor; as well as, Memory for storing programs, The program includes instructions, which, when executed by the processor, cause the processor to perform the method according to any one of claims 1 to 11.

14. A non-transitory computer-readable storage medium, characterized in that The non-transitory computer-readable storage medium stores computer instructions, and the computer instructions are used to cause the computer to execute the method according to any one of claims 1 to 11.

15. A computer program product, characterized in that The invention comprises a computer program, which implements the method according to any one of claims 1 to 11 when executed by a processor.

Images