A network quality evaluation method for computing power network

Abstract

The embodiment of the present invention provides a network quality evaluation method for a computing power network, which belongs to the field of communication technology, and specifically includes: step 1, obtaining different types of index data of the computing power network; step 2, dividing the index data into classification tasks and time series prediction tasks, using the classification model to predict the classification task, and obtaining the index result of each index data in the classification task, and using the time series prediction model to predict the time series prediction task, and obtaining the index result of each index data in the time series prediction task; step 3, using the calibration network based on the KAN network to calibrate and screen all the index results; step 4, aggregating the final output of the calibration network to obtain a quality score. Through the scheme of the present invention, the evaluation efficiency, accuracy and stability are improved.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of communication technology, and in particular to a method for evaluating the network quality of a computing network. Background Art

[0002] At present, with the rapid development of information technology, computing power has become a key driving force in the digital economy era, supporting the digital transformation and intelligent upgrading of all walks of life. As a new type of infrastructure that integrates computing power, storage power, and transportation power, computing power network realizes ubiquitous access and on-demand services of computing power through efficient integration and scheduling of distributed computing resources. However, with the continuous expansion of the scale of computing power network and the increasing complexity of application scenarios, how to scientifically and objectively evaluate its network quality and ensure the efficiency, stability and reliability of computing power services has become an urgent problem to be solved.

[0003] It can be seen that there is an urgent need for an accurate, efficient and stable network quality evaluation method for computing power networks. Summary of the invention

[0004] In view of this, an embodiment of the present invention provides a network quality evaluation method for a computing power network, which at least partially solves the problems of poor evaluation efficiency, accuracy and stability in the prior art.

[0005] An embodiment of the present invention provides a method for evaluating the network quality of a computing power network, comprising:

[0006] Step 1: Obtain different types of indicator data of the computing power network;

[0007] Step 2: divide the indicator data into a classification task and a time series prediction task, use the classification model to predict the classification task, and obtain the indicator result of each indicator data in the classification task; and use the time series prediction model to predict the time series prediction task, and obtain the indicator result of each indicator data in the time series prediction task;

[0008] Step 3: Use the calibration network based on the KAN network to calibrate and screen all indicator results;

[0009] Step 4: Aggregate the final output of the calibration network to obtain a quality score.

[0010] According to a specific implementation manner of an embodiment of the present invention, the types of indicator data include traffic classification, traffic prediction, error management and network security.

[0011] According to a specific implementation of the embodiment of the present invention, step 1 specifically includes:

[0012] Step 1.1, selecting indicator data of traffic classification types in the computing network, wherein the indicator data of the traffic classification types include indicator data corresponding to VPN and non-VPN traffic classification, malware detection, different protocol traffic classification and abnormal traffic detection in the computing network;

[0013] Step 1.2, selecting indicator data of the traffic prediction type in the computing power network, wherein the indicator data of the traffic prediction type includes indicator data corresponding to the traffic size prediction and the available bandwidth prediction in the computing power network;

[0014] Step 1.3, selecting indicator data of the error management type in the computing power network, wherein the indicator data of the error management type includes indicator data corresponding to fault detection, fault location and fault mitigation;

[0015] Step 1.4, selecting indicator data of security type in the computing power network, wherein the indicator data of security type includes indicator data corresponding to network intrusion detection, host intrusion detection and botnet detection.

[0016] According to a specific implementation of the embodiment of the present invention, after step 1, the method further includes:

[0017] Normalize the indicator data

[0018]

[0019] is the original data, is the normalized data, is the minimum value of the original data, is the maximum value of the original data.

[0020] According to a specific implementation of an embodiment of the present invention, the classification model includes a K nearest neighbor model, a support vector machine model, an identity authentication incremental support vector machine model, a stacked autoencoder model, an incremental support vector machine model, a stream sequence network model, and a sequence-image conversion model;

[0021] The time series prediction model includes a recurrent neural network model, a long short-term memory network model, an iTransformer model, a sequence-core fusion time series prediction model, a network traffic prediction model, a deep traffic prediction model, and an inverse discrete wavelet transform model.

[0022] According to a specific implementation of the embodiment of the present invention, the expression of the calibration network is:

[0023]

[0024] Among them, x is the input index result, e is the calibration network output, corresponding to the calibrated index result, a, b, c, d are the intermediate results of the calibration network, and d_model is the dimension size to be mapped. is a layer of the KAN network, q is the input dimension, p is the output dimension, Represents matrix transpose.

[0025] According to a specific implementation of the embodiment of the present invention, step 2 further includes:

[0026] The indicator data that needs to use time series data in the classification task is predicted using the time series prediction model to obtain the indicator results.

[0027] The network quality evaluation scheme of the computing power network in the embodiment of the present invention includes: step 1, obtaining different types of indicator data of the computing power network; step 2, dividing the indicator data into classification tasks and time series prediction tasks, using the classification model to predict the classification task, and obtaining the indicator result of each indicator data in the classification task, and using the time series prediction model to predict the time series prediction task, and obtaining the indicator result of each indicator data in the time series prediction task; step 3, using the calibration network based on the KAN network to calibrate and screen all indicator results; step 4, aggregating the final output of the calibration network to obtain a quality score.

[0028] The beneficial effects of the embodiments of the present invention are as follows: through the scheme of the present invention, by dividing the indicator data into classification tasks and time series prediction tasks, the most suitable model can be used for different types of data to improve the prediction accuracy; using the classification model and the time series prediction model to process specific tasks respectively can give full play to their respective advantages and improve the overall performance; the calibration mechanism of the KAN network can optimize the prediction results, reduce errors, and make the final output more reliable; the scheme can adapt to a variety of indicators and data sources, facilitate the subsequent addition of new indicators or task types, and enhance the flexibility of the system; by aggregating the results after calibration, a unified quality score can be formed to help decision makers quickly understand the indicator performance; the entire process can be automated, reduce manual intervention, improve work efficiency, and can update the indicator results in real time; using the transparent process provided by the calibration network, users can better understand how the model derives the final quality score and enhance the credibility of the system; through calibration and screening, outliers or noise data can be effectively removed to reduce the risk of overfitting of the model during training. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0030] Figure 1 A flowchart of a method for evaluating the network quality of a computing network provided by an embodiment of the present invention;

[0031] Figure 2 A system framework diagram of a network quality evaluation method for a computing power network provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0032] The embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0033] The following describes the embodiments of the present invention through specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work belong to the scope of protection of the present invention.

[0034] It should be noted that various aspects of the embodiments within the scope of the appended claims are described below. It should be apparent that the aspects described herein can be embodied in a wide variety of forms, and any specific structure and / or function described herein is merely illustrative. Based on the present invention, it should be understood by those skilled in the art that an aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number of aspects described herein can be used to implement the device and / or practice the method. In addition, other structures and / or functionalities other than one or more of the aspects described herein can be used to implement this device and / or practice this method.

[0035] It should also be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present invention. The illustrations only show components related to the present invention rather than being drawn according to the number, shape and size of components in actual implementation. In actual implementation, the form, quantity and proportion of each component may be changed arbitrarily, and the component layout may also be more complicated.

[0036] Additionally, in the following description, specific details are provided to facilitate a thorough understanding of the examples. However, it will be understood by those skilled in the art that the aspects described may be practiced without these specific details.

[0037] The network performance monitoring and evaluation of the computing power network is an important part of ensuring network smoothness and improving user satisfaction. Network performance will have a great impact on user experience. Specifically, it is reflected in the following aspects. (1) Network stability is the basis for ensuring user experience. If the network is interrupted or delayed seriously, it will cause a poor service experience and even lead to user loss. (2) Evaluating the network quality in advance and allocating devices with good network quality to users can improve the user experience. (3) By monitoring and evaluating network performance, security risks can be quickly discovered, nodes with malicious attacks or information leaks can be found, and measures can be taken in time to prevent and repair them, so as to improve the safety factor and protect the security of user data.

[0038] At present, the construction of computing power network is in a stage of rapid development, but it still faces many challenges. On the one hand, the uneven distribution and strong heterogeneity of computing power resources have brought difficulties to the unified management and scheduling of network resources; on the other hand, with the explosive growth of network traffic and the diversified needs of business, higher requirements are placed on the performance indicators of computing power network such as bandwidth, latency, and stability. Therefore, building a comprehensive, systematic, and quantifiable network quality evaluation system is of great significance to promoting the healthy development of computing power network, improving user experience, and promoting industrial innovation.

[0039] An embodiment of the present invention provides a method for evaluating the network quality of a computing power network, which can be applied to the quality evaluation process of a computing power network in an Internet scenario.

[0040] The method of the embodiment of the present invention aims to comprehensively and meticulously monitor and analyze the performance of the computing network by comprehensively using a variety of technical means and evaluation indicators, and provide a scientific basis for the optimization and improvement of the computing network. This method can not only effectively improve the service quality and user experience of the computing network, but also provide strong support for the development of the computing network industry and promote the high-quality development of the digital economy.

[0041] See also Figure 1 , is a flow chart of a method for evaluating the network quality of a computing network provided by an embodiment of the present invention. Figure 1 As shown, the method mainly comprises the following steps:

[0042] Step 1: Obtain different types of indicator data of the computing power network;

[0043] Furthermore, the types of indicator data include traffic classification, traffic prediction, error management and network security.

[0044] Furthermore, the step 1 specifically includes:

[0045] Step 1.1, selecting indicator data of traffic classification types in the computing network, wherein the indicator data of the traffic classification types include indicator data corresponding to VPN and non-VPN traffic classification, malware detection, different protocol traffic classification and abnormal traffic detection in the computing network;

[0046] Step 1.2, selecting indicator data of the traffic prediction type in the computing power network, wherein the indicator data of the traffic prediction type includes indicator data corresponding to the traffic size prediction and the available bandwidth prediction in the computing power network;

[0047] Step 1.3, selecting indicator data of the error management type in the computing power network, wherein the indicator data of the error management type includes indicator data corresponding to fault detection, fault location and fault mitigation;

[0048] Step 1.4, selecting indicator data of security type in the computing power network, wherein the indicator data of security type includes indicator data corresponding to network intrusion detection, host intrusion detection and botnet detection.

[0049] Furthermore, after step 1, the method further comprises:

[0050] Normalize the indicator data

[0051]

[0052] is the original data, is the normalized data, is the minimum value of the original data, is the maximum value of the original data.

[0053] In the specific implementation, the corresponding indicator data is collected and processed from the four aspects of traffic classification, traffic prediction, error management and network security. Traditional network quality evaluation methods generally focus on the construction of the evaluation system based on various performance indicators of the network, such as dividing bandwidth, load, latency, jitter, packet loss, etc. according to transmission, reliability and security. This requires strict collection of corresponding data and calculation of corresponding weights. Therefore, the four major aspects of traffic classification, traffic prediction, error management and network security are set. The number of indicators and specific indicators in the four aspects can be set according to the actual situation of the computing center, and they will not be listed one by one here.

[0054] Traffic classification is a key part of network management. Traffic classification covers a variety of applications such as QoS purposes, computing center service provider pricing, anomaly detection, etc. Traffic classification is the starting point of traffic engineering, and its purpose is to identify the traffic of different services or applications so that different processing strategies can be adopted. Its indicators can be selected from VPN and non-VPN traffic classification, malware detection, different protocol traffic classification, abnormal traffic detection, etc. The indicators can be expanded and adjusted according to actual needs.

[0055] Traffic prediction is an important task in network management and optimization. Its goal is to analyze and predict data traffic in the network and provide a basis for planning, scheduling and management of network resources. Considering the busy intermittent and regular nature of most networks on an hourly or daily scale, short-term traffic at the weekly level can be predicted to help mobile computing centers adjust the allocation of radio frequency resources in a timely manner. Therefore, traffic prediction has become an important task to reduce operating costs and improve user experience. Traffic prediction indicators can be selected from traffic size prediction, available bandwidth prediction, etc. The indicators can be expanded and adjusted according to actual needs.

[0056] Fault management refers to a set of tasks used to detect, isolate, and then correct network anomalies. In addition to common network failures, you must also deal with failures related to unreliable hardware, limited device life, connection failures, environmental conditions, etc. Therefore, in order to ensure QoS and high performance in communication systems and networks, fault management is essential to recover the system from failures. Fault management can choose indicators such as fault detection, fault location, and fault mitigation. The indicators can be expanded and adjusted according to actual needs.

[0057] Network security refers to protecting the computing center system from unauthorized access, leakage, interruption, modification or destruction, ensuring the integrity, confidentiality and availability of the computing center data. Ensuring network security is essential to maintaining corporate reputation, protecting user privacy and data security. Network security indicators can be selected from network intrusion detection, host intrusion detection, botnet detection, etc. The indicators can be expanded and adjusted according to actual needs.

[0058] After data collection, no matter what model is used, it needs to be normalized. This is a necessary condition to remove the difference in feature dimensions and make the model converge. The specific normalization method is:

[0059] (1)

[0060] is the original data, is the normalized data, is the minimum value of the original data, is the maximum value of the original data.

[0061] Step 2: divide the indicator data into a classification task and a time series prediction task, use the classification model to predict the classification task, and obtain the indicator result of each indicator data in the classification task; and use the time series prediction model to predict the time series prediction task, and obtain the indicator result of each indicator data in the time series prediction task;

[0062] Based on the above embodiment, step 2 further includes:

[0063] The indicator data that needs to use time series data in the classification task is predicted using the time series prediction model to obtain the indicator results.

[0064] Optionally, the classification model includes a K nearest neighbor model, a support vector machine model, an identity authentication incremental support vector machine model, a stacked autoencoder model, an incremental support vector machine model, a stream sequence network model, and a sequence-image conversion model;

[0065] The time series prediction model includes a recurrent neural network model, a long short-term memory network model, an iTransformer model, a sequence-core fusion time series prediction model, a network traffic prediction model, a deep traffic prediction model, and an inverse discrete wavelet transform model.

[0066] During specific implementation, the data is divided into classified data and time series data. The classified data is calculated according to the classification model, and the time series data is calculated according to the time series prediction model. The number of models can be set independently, and the specific model can be freely selected according to the specific situation of the computing power center.

[0067] In detail, the indicators introduced above can all be classified into classification tasks and time series prediction tasks. VPN and non-VPN traffic classification, malware detection, classification of traffic of different protocols, abnormal traffic detection, fault detection, fault location, network intrusion detection, host intrusion detection, and botnet detection can be regarded as classification tasks, and traffic size prediction, available bandwidth prediction, and fault mitigation can be regarded as time series prediction tasks. Classification tasks can be processed with classification models, and time series prediction tasks can be processed with time series models. Some classification tasks can still be processed with time series models because some classification tasks can use time series data. The specific data used can be collected according to the actual situation of the computing power center, and the specific model used can also be set according to the actual situation of the computing power center.

[0068] The classification models that can be used in the network include KNN, SVM, NB, C4.5, C5.0, RF, CNN, RNN, etc. The models proposed for classification tasks in the network include incremental support vector machine ISVM, identity authentication incremental support vector machine AISVM, cost-sensitive support vector machine CMSVM, stacked autoencoder SAE, multimodal deep learning traffic classification MIMETIC, stream sequence network FS-Net, sequence-image conversion model seq2image, etc. The time series models that can be used in the network include recurrent neural network RNN, long short-term memory network LSTM, ARIMA, iTransformer, sequence-core fusion time series prediction model SOFTS, etc. The models proposed for time series tasks in the network include network traffic prediction model LNTP, deep traffic prediction model DeepTP, deep spatiotemporal neural network STM, inverse discrete wavelet transform model iDWT, etc. The data sets targeted by various models are different. The specific data collection can be determined according to the actual situation of the computing power center, and the specific model can also be determined by the computing power center, which is also a feature of the intelligent evaluation framework. Each indicator can be calculated through multiple models to obtain multiple results. Which result is more accurate and which result is helpful for the final evaluation can be screened by the subsequent intelligent network, which also reflects the intelligent characteristics of this evaluation framework.

[0069] Step 3: Use the calibration network based on the KAN network to calibrate and screen all indicator results;

[0070] Furthermore, the expression of the calibration network is:

[0071]

[0072] Among them, x is the input index result, e is the calibration network output, corresponding to the calibrated index result, a, b, c, d are the intermediate results of the calibration network, and d_model is the dimension size to be mapped. is a layer of the KAN network, q is the input latitude, p is the output latitude, Represents matrix transpose.

[0073] In the specific implementation, the calculated values ​​of each indicator are calibrated and selected through the intelligent network. The results of various models are different, so they need to be calibrated. The values ​​obtained by each model are not necessarily helpful for the final evaluation. Therefore, an intelligent network is designed based on the KAN network to calibrate and screen the results of the classification model and the time series model.

[0074] The KAN network is a more explanatory neural network. The intelligent calibration network can be built based on the KAN network. First, the KAN network is introduced. The KAN network can be expressed as:

[0075] (2)

[0076] In the above formula, is the matrix function corresponding to the Lth KAN layer, and C is the input.

[0077] Each KAN layer can be expressed as:

[0078] (3)

[0079] With learnable parameters, is the input dimension of this layer, is the output dimension of this layer.

[0080] Assume there are L KAN layers in total, is the number of neurons in layer l, and (l, i) is used to define the i-th neuron in layer l. The activation value of this neuron is x(l, i). There will be An activation function, according to the characteristics of full connection, can give the symbolic representation of the activation function:

[0081] (4)

[0082] The input of the activation function is , the output is , so the nodes of the l+1 layer are represented by a matrix:

[0083] (5)

[0084] When L KAN layers are stacked, it can be expressed as , the calibrated index can be expressed as formula (2) through the KAN network.

[0085] The intelligent calibration network hopes to scale various model data so that the data can be adjusted intelligently and useful model data can be screened out. The self-attention mechanism can magnify or reduce the data so that the valuable parts are highlighted and the worthless parts are ignored. Based on this idea, the present invention designs a small self-calibration KAN network to calibrate the index values ​​calculated by each model. Multiple small networks are combined into an overall intelligent calibration network, which calculates the scores together with all the indicators, and finally updates the entire network with the feedback scores.

[0086] use represents a layer of the KAN network, x is the input, i is the input dimension, and j is the output dimension. Then the small self-calibrated KAN network can be expressed as:

[0087] (6)

[0088] x is the input of the small self-calibrated network, corresponding to the input index, e is the network output, corresponding to the calibrated index, a, b, c, d are the intermediate results of the network, and d_model is the dimension size that needs to be mapped.

[0089] The intelligent calibration network retains the pruning characteristics of the KAN network, that is, when the training is completed, the edge weight functions with too low weight influence can be removed. This makes it possible to delete some intermediate results in the later stage to achieve the purpose of screening, which also reflects the intelligence of the network.

[0090] Through the feedback of network quality evaluation results, the intelligent network is continuously updated so that the score can reflect the network status of the computing power center.

[0091] Step 4: Aggregate the final output of the calibration network to obtain a quality score.

[0092] In the specific implementation, the final score is output and the network quality is judged. The final score is calculated by a KAN network layer. The intelligent network will finally converge into a node. The node outputs the final score, which is between 0 and 1. The network quality is divided according to the score. When the score range is [0, 0.5], the network performance is considered to be poor, when the score range is [0.5, 0.7], the network performance is considered to be good, and when the score range is [0.7, 1], the network performance is considered to be excellent.

[0093] The network quality evaluation method of the computing power network provided in this embodiment can adopt the most appropriate model for different types of data to improve the prediction accuracy by dividing the indicator data into classification tasks and time series prediction tasks; using the classification model and the time series prediction model to process specific tasks respectively can give full play to their respective advantages and improve the overall performance; the calibration mechanism of the KAN network can optimize the prediction results, reduce errors, and make the final output more reliable; the scheme can adapt to a variety of indicators and data sources, facilitate the subsequent addition of new indicators or task types, and enhance the flexibility of the system; by aggregating the results after calibration, a unified quality score can be formed to help decision makers quickly understand the indicator performance; the entire process can be automated, reduce manual intervention, improve work efficiency, and can update the indicator results in real time; using the transparent process provided by the calibration network, users can better understand how the model derives the final quality score and enhance the credibility of the system; through calibration and screening, outliers or noise data can be effectively removed to reduce the risk of overfitting of the model during training.

[0094] It should be understood that various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof.

[0095] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A method for evaluating the network quality of a computing network, characterized in that: include: Step 1: Obtain different types of indicator data of the computing power network; Step 2: divide the indicator data into a classification task and a time series prediction task, use the classification model to predict the classification task, and obtain the indicator result of each indicator data in the classification task; and use the time series prediction model to predict the time series prediction task, and obtain the indicator result of each indicator data in the time series prediction task; Step 3: Use the calibration network based on the KAN network to calibrate and screen all indicator results; Step 4: Aggregate the final output of the calibration network to obtain a quality score; The expression of the calibration network is Among them, x is the input index result, e is the calibration network output, corresponding to the calibrated index result, a, b, c, d are the intermediate results of the calibration network, and d_model is the dimension size to be mapped. is a layer of the KAN network, q is the input latitude, p is the output latitude, Represents matrix transpose.

2. The method according to claim 1, characterized in that The types of indicator data include traffic classification, traffic prediction, error management and network security.

3. The method according to claim 2, characterized in that The step 1 specifically includes: Step 1.1, selecting indicator data of traffic classification types in the computing network, wherein the indicator data of the traffic classification types include indicator data corresponding to VPN and non-VPN traffic classification, malware detection, different protocol traffic classification and abnormal traffic detection in the computing network; Step 1.2, selecting indicator data of the traffic prediction type in the computing power network, wherein the indicator data of the traffic prediction type includes indicator data corresponding to the traffic size prediction and the available bandwidth prediction in the computing power network; Step 1.3, selecting indicator data of the error management type in the computing power network, wherein the indicator data of the error management type includes indicator data corresponding to fault detection, fault location and fault mitigation; Step 1.4, selecting indicator data of security type in the computing power network, wherein the indicator data of security type includes indicator data corresponding to network intrusion detection, host intrusion detection and botnet detection.

4. The method according to claim 1, characterized in that: After step 1, the method further comprises: Normalize the indicator data is the original data, is the normalized data, is the minimum value of the original data, is the maximum value of the original data.

5. The method according to claim 1, characterized in that The classification models include K nearest neighbor model, support vector machine model, identity authentication incremental support vector machine model, stacked autoencoder model, incremental support vector machine model, stream sequence network model, sequence-image conversion model; The time series prediction model includes a recurrent neural network model, a long short-term memory network model, an iTransformer model, a sequence-core fusion time series prediction model, a network traffic prediction model, a deep traffic prediction model, and an inverse discrete wavelet transform model.

6. The method according to claim 1, characterized in that The step 2 also includes: The indicator data that needs to use time series data in the classification task is predicted using the time series prediction model to obtain the indicator results.

Images