A bandwidth prediction method, device, equipment and medium for computing power network
By acquiring and splicing the bandwidth time domain and frequency domain information of the computing power network and extracting local and global features, the problem of low bandwidth prediction accuracy of computing power network is solved, and the accuracy of prediction bandwidth is improved.
Patent Information
- Application Number
- CN202411273542.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-12
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2044-09-12
AI Technical Summary
The prediction accuracy of computing power network bandwidth is low, making it difficult to effectively deal with burst traffic and network topology changes, resulting in insufficient prediction accuracy.
By obtaining the bandwidth value of the target computing power network, obtaining bandwidth time domain information and frequency domain information, splicing and linear projecting, local and global features are extracted, and prediction is carried out after fusion to improve the accuracy of the prediction bandwidth.
The information richness of the fused information is improved, the accuracy of the final feature is enhanced, and the accuracy of the predicted bandwidth value is improved.
Smart Images

Figure CN118802593B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of bandwidth prediction of computing power networks, and in particular to a bandwidth prediction method, device, equipment and medium for computing power networks. Background Art
[0002] With the continuous advancement of information technology, computing networks are becoming increasingly important as the infrastructure that supports a large number of traditional and new applications. These applications range from delay-sensitive services such as search engines, online recommendations, and instant messaging to computationally intensive tasks such as high-performance computing and distributed machine learning. These applications not only need to process large amounts of data, but also need to complete tasks within strict delay constraints, which places extremely high demands on the bandwidth prediction capabilities of computing networks.
[0003] In computing networks, bandwidth prediction is an important means to achieve efficient data transmission and resource management. By accurately predicting network bandwidth, data transmission strategies can be optimized, network congestion can be avoided, and the efficiency and stability of data transmission can be improved. At the same time, bandwidth prediction can also provide decision support for resource management, helping the system to dynamically adjust resource allocation according to network load conditions to meet the needs of different applications. However, computing network bandwidth prediction faces many challenges. First, network traffic is bursty and uncertain, which affects the accuracy of bandwidth prediction. Secondly, with the expansion of the scale of distributed systems and the increase in the complexity of applications, the network topology has become more and more complex, which has brought greater difficulty to bandwidth prediction. In addition, there are many different types of traffic in computing networks, such as data traffic, control traffic, etc., and the impact of these traffic on bandwidth is also different, which increases the difficulty of bandwidth prediction.
[0004] In order to meet these challenges, researchers have proposed a variety of computing power network bandwidth prediction methods. Among them, the traditional prediction method collects and analyzes historical network traffic data and establishes a linear model to predict future bandwidth changes. This method is simple and intuitive, but the prediction accuracy is low when faced with burst traffic and network topology changes. The machine learning-based prediction method uses machine learning algorithms to model and predict network traffic, which can better handle nonlinear and uncertainty problems. However, this method is difficult to capture the suddenness of bandwidth changes. It can be seen that the existing computing power network bandwidth prediction methods have the problem of low bandwidth prediction accuracy. Summary of the invention
[0005] The present application provides a bandwidth prediction method, device, equipment and medium for a computing power network, which can solve the problem of low bandwidth prediction accuracy.
[0006] In a first aspect, an embodiment of the present application provides a bandwidth prediction method for a computing power network, the bandwidth prediction method comprising:
[0007] Get the target computing power network T The bandwidth value at the moment; T The moment is the current moment;
[0008] Obtain bandwidth time domain information and bandwidth frequency domain information according to all bandwidth values, and concatenate the bandwidth time domain information and bandwidth frequency domain information to obtain fusion information; the bandwidth time domain information is used to describe the information of all bandwidth values in the time domain, and the bandwidth frequency domain information is used to describe the information of all bandwidth values in the frequency domain;
[0009] Linearly projecting the fused information to the hidden dimension to obtain the first projection and the second projection of the fused information, performing feature extraction on the first projection and the second projection to obtain local features, and performing feature extraction on the first projection and the second projection to obtain global features;
[0010] Fuse local features and global features to obtain the final features;
[0011] The final features are predicted to obtain the predicted bandwidth value of the target computing power network at the next moment of the current moment.
[0012] Optionally, bandwidth time domain information and bandwidth frequency domain information are obtained based on all bandwidth values, including:
[0013] Sort all bandwidth values according to all corresponding moments to obtain a bandwidth sequence;
[0014] Decompose the bandwidth sequence into multiple subsequences, and take the number of elements of the subsequence containing the most elements as the maximum number of elements;
[0015] For each subsequence, determine whether the number of elements in the subsequence is equal to the maximum number of elements. If so, take the subsequence as a frequency domain component. Otherwise, fill the subsequence with multiple 0 elements so that the number of elements in the subsequence is equal to the maximum number of elements, and take the filled subsequence as a frequency domain component.
[0016] All frequency domain components are concatenated into a matrix to obtain bandwidth frequency domain information;
[0017] Perform sliding average pooling on each subsequence to obtain the pooled subsequence corresponding to each subsequence;
[0018] For each pooling subsequence, determine whether the number of elements in the pooling subsequence is equal to the maximum number of elements. If so, use the pooling subsequence as a time domain component. Otherwise, fill the pooling subsequence with multiple 0 elements so that the number of elements in the pooling subsequence is equal to the maximum number of elements, and use the filled pooling subsequence as a time domain component.
[0019] All time domain components are concatenated into a matrix to obtain bandwidth time domain information.
[0020] Optionally, the bandwidth sequence is decomposed into multiple subsequences, including:
[0021] By formula:
[0022]
[0023] Get the subsequences;
[0024] in, Indicates The low-frequency components of the subsequences, Indicates The high frequency components of the subsequences, represents a low-pass filter, represents a high-pass filter, Indicates subsequences, , represents the number of subsequences, when hour, is the bandwidth sequence.
[0025] Optionally, performing feature extraction on the first projection and the second projection by downsampling to obtain local features includes:
[0026] Downsampling and activating the first projection to obtain a first downsampled projection vector, and processing the first downsampled projection vector using a state space model to obtain a first downsampled projection representation;
[0027] Downsampling and activation function processing are performed on the second projection to obtain a second downsampled projection vector;
[0028] The first downsampled projection representation is residually connected with the second downsampled projection vector to obtain the local feature.
[0029] Optionally, upsampling the first projection and the second projection is performed to extract features to obtain global features, including:
[0030] Performing upsampling and activation function processing on the first projection to obtain a first upsampled projection vector, and processing the first upsampled projection vector using a state space model to obtain a first upsampled projection representation;
[0031] Perform upsampling and activation function processing on the second projection to obtain a second upsampled projection vector;
[0032] The first upsampled projection representation and the second upsampled projection vector are residually connected to obtain the global feature.
[0033] Optionally, the local features and the global features are fused to obtain the final features, including:
[0034] The local features and the global features are spliced to obtain spliced features;
[0035] The fusion module is used to fuse the spliced features to obtain the final features.
[0036] Optionally, a fusion module is used to fuse the spliced features to obtain the final features, including:
[0037] By formula:
[0038]
[0039]
[0040]
[0041] Calculate the final features ;
[0042] in, Represents the splicing feature, represents the operation of the first fusion submodule, represents the operation of the second fusion submodule, is the input of the second fusion submodule and , represents a multi-layer perceptron, represents average pooling, represents the maximum pooling, represents the convolution operation, Represents the activation function.
[0043] In a second aspect, an embodiment of the present application provides a bandwidth prediction device for a computing network, including:
[0044] Get module, get the target computing power network T Multiple bandwidth values at a time; T The moment is the current moment;
[0045] A splicing module obtains bandwidth time domain information and bandwidth frequency domain information according to all bandwidth values, and splices the bandwidth time domain information and the bandwidth frequency domain information to obtain fusion information; the bandwidth time domain information is used to describe the information of all bandwidth values in the time domain, and the bandwidth frequency domain information is used to describe the information of all bandwidth values in the frequency domain;
[0046] A feature extraction module linearly projects the fused information to a hidden dimension to obtain a first projection and a second projection of the fused information, performs feature extraction on the first projection and the second projection to obtain local features, and performs feature extraction on the first projection and the second projection to obtain global features;
[0047] The fusion module fuses local features and global features to obtain the final features;
[0048] The prediction module predicts the final features and obtains the predicted bandwidth value of the target computing power network at the next moment of the current moment.
[0049] In a third aspect, an embodiment of the present application provides a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the bandwidth prediction method of the computing power network when executing the computer program.
[0050] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program, and when the computer program is executed by a processor, it implements the bandwidth prediction method of the computing power network mentioned above.
[0051] The above solution of the present application has the following beneficial effects:
[0052] In the embodiment of the present application, by obtaining the target computing power network T The bandwidth value at each moment is obtained, and then the bandwidth time domain information and bandwidth frequency domain information are obtained according to all bandwidth values, and the bandwidth time domain information and bandwidth frequency domain information are spliced to obtain fused information, and then the fused information is linearly projected to the hidden dimension to obtain the first projection and the second projection of the fused information, and the first projection and the second projection are downsampled and feature extraction is performed to obtain local features, and the first projection and the second projection are upsampled and feature extraction is performed to obtain global features, and then the local features and the global features are fused to obtain the final features, and finally the final features are predicted to obtain the predicted bandwidth value of the target computing power network at the next moment of the current moment. Among them, splicing the bandwidth time domain information and the bandwidth frequency domain information can improve the information richness in the fused information, extracting local features and global features from the fused information with high information richness, and can consider both local and global information, so that the accuracy of the final features is improved, and the accuracy of the predicted bandwidth value obtained by predicting the final features with high accuracy is improved.
[0053] Other beneficial effects of the present application will be described in detail in the subsequent specific implementation section. BRIEF DESCRIPTION OF THE DRAWINGS
[0054] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.
[0055] Figure 1 A flowchart of a bandwidth prediction method for a computing network provided in an embodiment of the present application;
[0056] Figure 2 A schematic diagram of the structure of a bandwidth prediction system provided in one embodiment of the present application;
[0057] Figure 3 A schematic diagram of the structure of a feature extraction fusion model provided in one embodiment of the present application;
[0058] Figure 4 A comparison curve diagram of the predicted bandwidth value and the actual bandwidth value provided in an embodiment of the present application;
[0059] Figure 5 A schematic diagram of the structure of a bandwidth prediction device for a computing network provided in an embodiment of the present application;
[0060] Figure 6 A schematic diagram of the structure of a terminal device provided in one embodiment of the present application. DETAILED DESCRIPTION
[0061] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present application.
[0062] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, wholes, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or combinations thereof.
[0063] It should also be understood that the term “and / or” used in the specification and appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.
[0064] As used in the specification and appended claims of this application, the term "if" can be interpreted as "when" or "uponce" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrase "if it is determined" or "if [described condition or event] is detected" can be interpreted as meaning "uponce it is determined" or "in response to determining" or "uponce [described condition or event] is detected" or "in response to detecting [described condition or event]", depending on the context.
[0065] In addition, in the description of the present application specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.
[0066] References to "one embodiment" or "some embodiments" etc. described in the specification of this application mean that one or more embodiments of the present application include specific features, structures or characteristics described in conjunction with the embodiment. Therefore, the statements "in one embodiment", "in some embodiments", "in some other embodiments", "in some other embodiments", etc. that appear in different places in this specification do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized in other ways.
[0067] In view of the problem of low accuracy in bandwidth prediction of existing computing power networks, an embodiment of the present application provides a bandwidth prediction method for a computing power network. The bandwidth prediction method obtains the bandwidth value of the target computing power network at T moments, then obtains bandwidth time domain information and bandwidth frequency domain information according to all bandwidth values, and splices the bandwidth time domain information and bandwidth frequency domain information to obtain fused information, and then linearly projects the fused information to the hidden dimension to obtain the first projection and the second projection of the fused information, performs down-sampling feature extraction on the first projection and the second projection to obtain local features, and performs up-sampling feature extraction on the first projection and the second projection to obtain global features, and then fuses the local features and the global features to obtain the final features, and finally predicts the final features to obtain the predicted bandwidth value of the target computing power network at the next moment of the current moment. Among them, splicing the bandwidth time domain information and the bandwidth frequency domain information can improve the information richness in the fused information, extracting local features and global features from the fused information with high information richness, and can consider both local and global information, so that the accuracy of the final features is improved, and the accuracy of the predicted bandwidth value obtained by predicting the final features with high accuracy is improved.
[0068] Next, the bandwidth prediction method of the computing power network provided in this application is exemplified.
[0069] like Figure 1 As shown, the bandwidth prediction method of the computing power network provided in this application includes the following steps:
[0070] Step 11: Get the target computing power network T The bandwidth value at a certain moment.
[0071] No. T The moment is the current moment.
[0072] It should be noted that the above-mentioned target computing network is the computing network that needs to perform bandwidth prediction, and the bandwidth value is the transmission bandwidth of the communication channel of the computing network. T The moments are the current moment and the previous moments, such as T is equal to 3, and the current moment is the 5th moment, then T The moments are the 3rd moment, the 4th moment, and the 5th moment respectively, and the next moment of the current moment is the 6th moment.
[0073] For example, you can use bandwidth detection tools such as perf to obtain the bandwidth value of the target computing power network.
[0074] Step 12: Obtain bandwidth time domain information and bandwidth frequency domain information according to all bandwidth values, and concatenate the bandwidth time domain information and the bandwidth frequency domain information to obtain fusion information.
[0075] The bandwidth time domain information is used to describe the information of all bandwidth values in the time domain, and the bandwidth frequency domain information is used to describe the information of all bandwidth values in the frequency domain.
[0076] In some embodiments of the present application, the above step of obtaining bandwidth time domain information and bandwidth frequency domain information according to all bandwidth values is specifically as follows:
[0077] The first step is to sort all bandwidth values according to all corresponding moments to obtain a bandwidth sequence.
[0078] In the second step, the bandwidth sequence is decomposed into multiple subsequences, and the number of elements in the subsequence containing the most elements is taken as the maximum number of elements.
[0079] Specifically, through the formula:
[0080]
[0081] Get the subsequences.
[0082] in, Indicates The low-frequency components of the subsequences, Indicates The high frequency components of the subsequences, represents a low-pass filter, represents a high-pass filter, Indicates subsequences, , represents the number of subsequences, when hour, is the bandwidth sequence.
[0083] It should be noted that after obtaining the low-frequency component and the high-frequency component of the subsequence, the low-frequency component and the high-frequency component are concatenated to obtain a complete subsequence.
[0084] The third step is to determine, for each subsequence, whether the number of elements in the subsequence is equal to the maximum number of elements. If so, the subsequence is used as a frequency domain component. Otherwise, multiple 0 elements are filled into the subsequence so that the number of elements in the subsequence is equal to the maximum number of elements, and the filled subsequence is used as a frequency domain component.
[0085] Exemplarily, the subsequence with the largest number of elements among all subsequences is the first subsequence, whose number of elements is 8, and the number of elements of the second subsequence is 5, then the second subsequence is padded with 3 0 elements so that the number of elements of the second subsequence is equal to 8.
[0086] The fourth step is to concatenate all frequency domain components into a matrix to obtain bandwidth frequency domain information.
[0087] The fifth step is to perform sliding average pooling on each subsequence to obtain the pooled subsequence corresponding to each subsequence.
[0088] In the sixth step, for each pooled subsequence, determine whether the number of elements in the pooled subsequence is equal to the maximum number of elements. If so, the pooled subsequence is used as a time domain component. Otherwise, multiple 0 elements are filled into the pooled subsequence so that the number of elements in the pooled subsequence is equal to the maximum number of elements, and the filled pooled subsequence is used as a time domain component.
[0089] In the seventh step, all time domain components are concatenated into a matrix to obtain bandwidth time domain information.
[0090] It should be noted that, since the dimensions of the bandwidth frequency domain information and the bandwidth time domain information are the same by filling 0 elements in the above process, the bandwidth frequency domain information and the bandwidth time domain information can be spliced to obtain fused information.
[0091] Exemplarily, the above steps can be run using computer software such as matlab, python, etc. to obtain fusion information.
[0092] It is worth mentioning that by obtaining the bandwidth time domain information and the bandwidth frequency domain information, the bandwidth information can be analyzed in both time domain and frequency domain dimensions, and the bandwidth time domain information and the bandwidth frequency domain information can be spliced to improve the information richness in the fused information.
[0093] Step 13, linearly projecting the fused information to the hidden dimension to obtain the first projection and the second projection of the fused information, performing feature extraction of downsampling on the first projection and the second projection to obtain local features, and performing feature extraction of upsampling on the first projection and the second projection to obtain global features.
[0094] It should be noted that when the fused information is linearly projected onto the hidden dimension, the first projection and the second projection are obtained by projecting onto two different hidden dimensions.
[0095] In some embodiments of the present application, the steps of performing feature extraction by downsampling the first projection and the second projection to obtain local features, and performing feature extraction by upsampling the first projection and the second projection to obtain global features are specifically:
[0096] In the first step, the first projection and the second projection are downsampled and feature extracted to obtain local features.
[0097] First, the first projection is downsampled and activated to obtain a first downsampled projection vector, and the first downsampled projection vector is processed using a state space model to obtain a first downsampled projection representation.
[0098] Then, the second projection is downsampled and activated to obtain a second downsampled projection vector.
[0099] Finally, the first downsampled projection representation is residually connected with the second downsampled projection vector to obtain the local features.
[0100] In the second step, up-sample the first projection and the second projection to extract features and obtain global features.
[0101] First, the first projection is upsampled and activated to obtain a first upsampled projection vector, and the first upsampled projection vector is processed using a state space model to obtain a first upsampled projection representation.
[0102] Then, the second projection is upsampled and activated to obtain a second upsampled projection vector.
[0103] Finally, the first upsampled projection representation and the second upsampled projection vector are residually connected to obtain the global feature.
[0104] Exemplarily, the above steps can be run using computer software such as matlab, python, etc. to obtain local features and global features.
[0105] It is worth mentioning that extracting local and global features from fused information with high information richness can consider both local and global information and provide accurate and comprehensive information for subsequent steps.
[0106] Step 14: fuse the local features and the global features to obtain the final features.
[0107] In some embodiments of the present application, the above step of fusing local features with global features to obtain final features specifically includes:
[0108] The first step is to concatenate local features and global features to obtain concatenated features.
[0109] In the second step, the fusion module is used to fuse the spliced features to obtain the final features.
[0110] Specifically, through the formula:
[0111]
[0112]
[0113]
[0114] Calculate the final features .
[0115] in, Represents the splicing feature, represents the operation of the first fusion submodule, represents the operation of the second fusion submodule, is the input of the second fusion submodule and , represents a multi-layer perceptron, represents average pooling, represents the maximum pooling, represents the convolution operation, Represents the activation function.
[0116] It can be understood that the fusion module includes a first fusion submodule and a second fusion submodule, and the above three formulas for calculating the final features are expressions of the fusion module including the first fusion submodule and the second fusion submodule.
[0117] For example, the above formula can be run using Matlab to obtain the final features.
[0118] It is worth mentioning that the final feature obtained by fusing local features and global features contains accurate and comprehensive information, and can accurately describe the bandwidth value of the target computing power network over a period of time.
[0119] Step 15, predict the final feature to obtain the predicted bandwidth value of the target computing power network at the next moment of the current moment.
[0120] In some embodiments of the present application, the final features can be predicted using prediction models such as a bidirectional long short-term memory network (BiLSTM, Bi-directional Long Short-Term Memory) to obtain a predicted bandwidth value. BiLSTM is composed of a forward long short-term memory network (LSTM, Long Short-Term Memory), a backward LSTM, and a final fully connected layer. BiLSTM uses sequential and reverse calculations on the sequence to obtain two different sets of hidden layer representations, and then obtains the final hidden layer representation through vector splicing. The core part of LSTM includes a forget gate, an input gate, and an output gate. When making a prediction, the feature matrix obtained by the feature extraction module is first expanded into a one-dimensional feature vector, and the bidirectional LSTM is composed of forward and backward LSTMs to fully utilize the information carried by the feature vector, and finally the final prediction result is output through a fully connected layer.
[0121] It should be noted that before directly using the prediction model to obtain the predicted bandwidth value at the current moment, multiple historical moments can be taken as the current moment to obtain the final features corresponding to each historical moment, and the predicted bandwidth value of each historical moment can be obtained using the final features corresponding to all historical moments. The state space model and prediction model mentioned in the above steps are trained according to the predicted bandwidth value and the actual bandwidth value of the historical moment to improve the performance of the state space model and the prediction model, so as to improve the accuracy of the predicted bandwidth value obtained at the current moment.
[0122] The specific training process is as follows:
[0123] First, through the formula:
[0124]
[0125]
[0126] Calculate the root mean square error and the mean absolute percentage error .
[0127] in, Indicates The predicted bandwidth value at each historical moment, Indicates The actual bandwidth value at a historical moment, Represents the number of historical moments.
[0128] Then determine whether the root mean square error is less than the first preset error value, and whether the absolute percentage mean error is less than the second preset error value. If both are less than, the training is completed. Otherwise, adjust the parameters of the state space model and the prediction model, and re-obtain the predicted bandwidth value at each historical moment, and calculate the root mean square error and the absolute percentage mean error.
[0129] For example, Matlab can be used to run a prediction model such as BiLSTM to predict the final features and obtain a predicted bandwidth value.
[0130] It is worth mentioning that splicing the bandwidth time domain information and the bandwidth frequency domain information can improve the information richness in the fused information, extract local features and global features from the fused information with high information richness, and consider both local and global information, so that the accuracy of the final features is improved. The accuracy of the predicted bandwidth value predicted using the final features with high accuracy is improved.
[0131] The bandwidth prediction method of the present application is exemplified below with reference to a specific example.
[0132] Bandwidth prediction systems such as Figure 2 As shown, the input bandwidth sequence passes through the pooling layer and the wavelet transform layer, that is, the step of decomposing into multiple sub-sequences in the above text, to obtain the bandwidth frequency domain information and bandwidth time domain information, and input them into the feature extraction and fusion layer to obtain the final features, and then through the BiLSTM (multiple long short-term memory networks) processing, the predicted bandwidth value is output through the linear layer.
[0133] The above feature extraction fusion layer is as follows Figure 3As shown, the feature extraction fusion layer includes a local feature extraction branch, a global feature extraction branch and a fusion module. The local feature extraction branch processes the first projection of the input through a linear layer and downsampling, and then sequentially inputs it into the activation function and the state space model to obtain a first down-sampled projection representation. At the same time, the second projection is processed through a linear layer, downsampling and an activation function in sequence to obtain a second down-sampled projection vector. The second down-sampled projection vector and the first down-sampled projection representation are processed through a residual connection, a linear layer and upsampling in sequence, and then input into the fusion module; the global feature extraction branch processes the first projection through a linear layer, upsampling, an activation function and a state space model in sequence to obtain a first up-sampled projection representation. At the same time, the second projection is processed through a linear layer, upsampling and an activation function in sequence to obtain a second up-sampled projection vector. The second up-sampled projection vector and the first up-sampled projection representation are processed through a residual connection, a linear layer and downsampling in sequence, and then input into the fusion module. The final feature is obtained by fusion of the fusion module.
[0134] The above bandwidth system was used to conduct experiments, and the bandwidth values of a computing network at multiple times were collected. All bandwidth values were divided into training sets and test sets. The size of the prediction time was changed, and the method of the present application was used to perform predictions. The comparison between the predicted value and the true value of the bandwidth value is shown in the figure below. Figure 4 As shown, Figure 4 The horizontal axis represents time in hours (h), and the vertical axis represents the traffic size (i.e., the bandwidth value, which is the normalized result and has no unit). The curve with a circular mark is the true value curve, and the curve with a triangular mark is the predicted value curve. The error of the prediction result is shown in Table 1.
[0135] Table 1
[0136]
[0137] Among them, MSE represents mean square error, MARE represents mean absolute relative error, the prediction time size is the prediction step size, and one prediction step size is one hour.
[0138] The prediction result errors obtained by the ablation experiment using the method of the present application are shown in Table 2.
[0139] Table 2
[0140]
[0141] Among them, the complete model is the complete process of the method of this application. No wavelet decomposition means that the steps of obtaining bandwidth time domain information and bandwidth frequency domain information in this application are removed. No Mamba means that other feature extraction fusion models are used to replace the step of using fusion information to obtain the final features in this application. No BiLSTM means that other prediction modules are used instead of BiLSTM for prediction.
[0142] It can be seen that the bandwidth prediction method of the present application can improve the accuracy of bandwidth prediction, and any missing step will have an impact on the final prediction result.
[0143] The bandwidth prediction device of the computing power network provided in this application is exemplified below.
[0144] like Figure 5 As shown, the embodiment of the present application provides a bandwidth prediction device for a computing power network, and the bandwidth prediction device 500 of the computing power network includes:
[0145] Acquisition module 501, obtains the target computing power network T Multiple bandwidth values at a time; T The moment is the current moment;
[0146] The splicing module 502 obtains bandwidth time domain information and bandwidth frequency domain information according to all bandwidth values, and splices the bandwidth time domain information and the bandwidth frequency domain information to obtain fusion information; the bandwidth time domain information is used to describe the information of all bandwidth values in the time domain, and the bandwidth frequency domain information is used to describe the information of all bandwidth values in the frequency domain;
[0147] The feature extraction module 503 linearly projects the fused information to the hidden dimension to obtain the first projection and the second projection of the fused information, performs feature extraction of downsampling on the first projection and the second projection to obtain local features, and performs feature extraction of upsampling on the first projection and the second projection to obtain global features;
[0148] A fusion module 504 fuses the local features with the global features to obtain the final features;
[0149] The prediction module 505 predicts the final feature and obtains the predicted bandwidth value of the target computing power network at the next moment of the current moment.
[0150] It should be noted that the information interaction, execution process, etc. between the above-mentioned devices / units are based on the same concept as the method embodiment of the present application. Their specific functions and technical effects can be found in the method embodiment part and will not be repeated here.
[0151] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.
[0152] like Figure 6 As shown, an embodiment of the present application provides a terminal device. The terminal device D10 of this embodiment includes: at least one processor D100 ( Figure 6 Only one processor is shown in the figure), a memory D101, and a computer program D102 stored in the memory D101 and executable on the at least one processor D100, wherein the processor D100 implements the steps of any of the above-mentioned method embodiments when executing the computer program D102.
[0153] Specifically, when the processor D100 executes the computer program D102, it obtains multiple bandwidth values of the target computing power network at multiple times, then obtains bandwidth time domain information and bandwidth frequency domain information according to all bandwidth values, and splices the bandwidth time domain information and bandwidth frequency domain information to obtain fusion information, and then linearly projects the fusion information to the hidden dimension to obtain the first projection and the second projection of the fusion information, performs down-sampling feature extraction on the first projection and the second projection to obtain local features, and performs up-sampling feature extraction on the first projection and the second projection to obtain global features, and then fuses the local features and the global features to obtain the final features, and finally predicts the final features to obtain the predicted bandwidth value of the target computing power network at the next moment of the current moment. Among them, splicing the bandwidth time domain information and the bandwidth frequency domain information can improve the information richness in the fusion information, extracting local features and global features from the fusion information with high information richness, and considering both local and global information, so that the accuracy of the final features is improved, and the accuracy of the predicted bandwidth value obtained by predicting the final features with high accuracy is improved.
[0154] The processor D100 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.
[0155] In some embodiments, the memory D101 may be an internal storage unit of the terminal device D10, such as a hard disk or memory of the terminal device D10. In other embodiments, the memory D101 may also be an external storage device of the terminal device D10, such as a plug-in hard disk, a smart memory card (SMC, SmartMedia Card), a secure digital (SD, Secure Digital) card, a flash card (Flash Card), etc. equipped on the terminal device D10. Further, the memory D101 may also include both an internal storage unit of the terminal device D10 and an external storage device. The memory D101 is used to store an operating system, an application program, a boot loader (BootLoader), data, and other programs, such as the program code of the computer program, etc. The memory D101 may also be used to temporarily store data that has been output or is to be output.
[0156] An embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in the above-mentioned method embodiments can be implemented.
[0157] An embodiment of the present application provides a computer program product. When the computer program product runs on a terminal device, the terminal device can implement the steps in the above-mentioned method embodiments when executing the computer program product.
[0158] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the processes in the above-mentioned embodiment method, which can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may at least include: any entity or device, recording medium, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electrical carrier signal, telecommunication signal and software distribution medium that can carry computer program code to the bandwidth prediction method device / terminal device of the computing power network. For example, a USB flash drive, a mobile hard disk, a disk or an optical disk.
[0159] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.
[0160] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software 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 this application.
[0161] The above is a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles described in the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.
Claims
1. A bandwidth prediction method for a computing network, characterized in that: include: Get the target computing power network T The bandwidth value at a certain moment; No. T The moment is the current moment; Obtain bandwidth time domain information and bandwidth frequency domain information according to all bandwidth values, and splice the bandwidth time domain information and the bandwidth frequency domain information to obtain fusion information; the bandwidth time domain information is used to describe information of all bandwidth values in the time domain, and the bandwidth frequency domain information is used to describe information of all bandwidth values in the frequency domain; Linearly projecting the fused information to a hidden dimension to obtain a first projection and a second projection of the fused information, performing downsampling feature extraction on the first projection and the second projection to obtain local features, and performing upsampling feature extraction on the first projection and the second projection to obtain global features; The local features and the global features are combined to obtain the final features; The final feature is predicted to obtain a predicted bandwidth value of the target computing power network at the next moment of the current moment.
2. The bandwidth prediction method according to claim 1, characterized in that: The obtaining of bandwidth time domain information and bandwidth frequency domain information according to all bandwidth values includes: Sort all bandwidth values according to all corresponding moments to obtain a bandwidth sequence; Decomposing the bandwidth sequence into multiple subsequences, and taking the number of elements of the subsequence containing the most elements as the maximum number of elements; For each of the subsequences, determining whether the number of elements in the subsequence is equal to the maximum number of elements; if so, taking the subsequence as a frequency domain component; otherwise, padding the subsequence with a plurality of 0 elements so that the number of elements in the subsequence is equal to the maximum number of elements, and taking the padded subsequence as a frequency domain component; All frequency domain components are concatenated into a matrix to obtain bandwidth frequency domain information; Perform sliding average pooling on each of the subsequences to obtain a pooled subsequence corresponding to each of the subsequences; For each of the pooling subsequences, determine whether the number of elements in the pooling subsequence is equal to the maximum number of elements; if so, use the pooling subsequence as a time domain component; otherwise, fill the pooling subsequence with multiple 0 elements so that the number of elements in the pooling subsequence is equal to the maximum number of elements, and use the filled pooling subsequence as a time domain component; All time domain components are concatenated into a matrix to obtain bandwidth time domain information.
3. The bandwidth prediction method according to claim 2, characterized in that: The step of decomposing the bandwidth sequence into a plurality of subsequences comprises: By formula: Get the subsequences; in, Indicates the The low-frequency components of the subsequences, Indicates the The high-frequency components of the subsequences, represents a low-pass filter, represents a high-pass filter, Indicates subsequences, , represents the number of subsequences, when hour, is the bandwidth sequence.
4. The bandwidth prediction method according to claim 1, characterized in that: The extracting features by downsampling the first projection and the second projection to obtain local features includes: Downsampling and activating the first projection to obtain a first downsampled projection vector, and processing the first downsampled projection vector using a state space model to obtain a first downsampled projection representation; Performing downsampling and activation function processing on the second projection to obtain a second downsampled projection vector; Perform a residual connection between the first down-sampled projection representation and the second down-sampled projection vector to obtain a local feature.
5. The bandwidth prediction method according to claim 1, characterized in that: The extracting features by upsampling the first projection and the second projection to obtain global features includes: Performing upsampling and activation function processing on the first projection to obtain a first upsampled projection vector, and processing the first upsampled projection vector using a state space model to obtain a first upsampled projection representation; Performing upsampling and activation function processing on the second projection to obtain a second upsampled projection vector; The first up-sampled projection representation and the second up-sampled projection vector are residually connected to obtain a global feature.
6. The bandwidth prediction method according to claim 1, characterized in that: The fusing of the local features and the global features to obtain the final features includes: Splicing the local feature and the global feature to obtain a spliced feature; The fusion module is used to fuse the spliced features to obtain the final features.
7. The bandwidth prediction method according to claim 6, characterized in that: The step of using a fusion module to fuse the spliced features to obtain final features includes: By formula: Calculate the final feature ; in, represents the splicing feature, represents the operation of the first fusion submodule, represents the operation of the second fusion submodule, is the input of the second fusion submodule and , represents a multi-layer perceptron, represents average pooling, represents the maximum pooling, represents the convolution operation, Represents the activation function.
8. A bandwidth prediction device for a computing network, characterized in that: include: Get module, get the target computing power network T Multiple bandwidth values at a time; No. T The moment is the current moment; A splicing module, which obtains bandwidth time domain information and bandwidth frequency domain information according to all bandwidth values, and splices the bandwidth time domain information and the bandwidth frequency domain information to obtain fusion information; the bandwidth time domain information is used to describe the information of all bandwidth values in the time domain, and the bandwidth frequency domain information is used to describe the information of all bandwidth values in the frequency domain; A feature extraction module, linearly projects the fused information to a hidden dimension to obtain a first projection and a second projection of the fused information, performs feature extraction of downsampling on the first projection and the second projection to obtain local features, and performs feature extraction of upsampling on the first projection and the second projection to obtain global features; A fusion module, which fuses the local features with the global features to obtain the final features; The prediction module predicts the final feature to obtain the predicted bandwidth value of the target computing power network at the next moment of the current moment.
9. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the bandwidth prediction method of the computing power network as described in any one of claims 1 to 7 is implemented.
10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the bandwidth prediction method of the computing power network as described in any one of claims 1 to 7 is implemented.
