Resource use state prediction method and device, equipment, storage medium and program product

Abstract

The invention relates to a resource use state prediction method and device, equipment, a storage medium and a program product. The method comprises the following steps: acquiring historical resource use data on a to-be-predicted cloud platform, and inputting the historical resource use data into a resource prediction model for resource use state prediction in a preset time period to obtain a prediction result; the resource prediction model comprises a time-frequency feature extraction network, a time sequence feature extraction network, a depth separable convolutional layer and a prediction network; the output of the time-frequency feature extraction network is connected with the input of the time sequence feature extraction network, the output of the time sequence feature extraction network is connected with the input of the depth separable convolutional layer, and the output of the depth separable convolutional layer is connected with the input of the prediction network. According to the method, the network has stronger time-frequency analysis capability when predicting the resource use data, so that the key frequency characteristics in the resource use data are more accurately captured, and the accuracy of the prediction result is higher.

Description

Technical Field

[0001] The present application relates to the field of cloud computing technology, and in particular to a method, device, equipment, storage medium and program product for predicting resource usage status. Background Art

[0002] Cloud computing is a computing model that provides on-demand computing resources and charges based on the resources used. Users can obtain and use these resources in remote data centers.

[0003] At present, in order to fully meet the actual needs of users, cloud computing service providers usually need to prepare sufficient resources to ensure that. If cloud resources cannot meet user needs in a timely manner, it may lead to longer service response time, system instability or frequent failures, affect user operation experience, cause user dissatisfaction, and even customer loss. If the resource preparation exceeds the actual demand, it is an unnecessary cost consumption for resource, power, and manpower maintenance. Based on this, it is very meaningful to predict the possible use of resources of the cloud platform during the calculation process of the cloud platform.

[0004] However, existing resource prediction methods suffer from inaccurate predictions. Summary of the invention

[0005] Based on this, it is necessary to provide a method, device, equipment, storage medium and program product for predicting resource usage status that can accurately predict resources on a cloud platform in response to the above technical problems.

[0006] In a first aspect, the present application provides a method for predicting resource usage status, comprising:

[0007] Obtain historical resource usage data on the cloud platform to be predicted;

[0008] The historical resource usage data is input into the resource prediction model to predict the resource usage status within a preset time period to obtain the prediction result; the resource prediction model includes a time-frequency feature extraction network, a time series feature extraction network, a depth-wise separable convolutional layer and a prediction network; the output of the time-frequency feature extraction network is connected to the input of the time series feature extraction network, the output of the time series feature extraction network is connected to the input of the depth-wise separable convolutional layer, and the output of the depth-wise separable convolutional layer is connected to the input of the prediction network.

[0009] In one embodiment, the historical resource usage data is input into the resource prediction model to predict the resource usage status within a preset time period to obtain a prediction result, including:

[0010] Input the historical resource usage data into the time-frequency feature extraction network to extract the time-frequency features and obtain the time-frequency features;

[0011] Input the historical resource usage data into the time series feature extraction network to extract the time series features and obtain the time series features;

[0012] Fuse the time-frequency features and time series features to obtain the target fusion features;

[0013] Input the target fusion features into the depth-separable convolutional layer for convolution processing to obtain the processed fusion features;

[0014] The processed fusion features are input into the prediction network for prediction to obtain the prediction results.

[0015] In one embodiment, the time-frequency feature extraction network includes a DWVD distribution transformation layer and a time-frequency feature extraction subnetwork. The historical resource usage data is input into the time-frequency feature extraction network to extract the time-frequency features to obtain the time-frequency features, including:

[0016] Input the historical resource usage data into the DWVD distribution transformation layer for time-frequency transformation to obtain the initial time-frequency features;

[0017] The initial time-frequency features are input into the time-frequency feature extraction subnetwork for multi-order feature extraction to obtain time-frequency features.

[0018] In one embodiment, the time-frequency feature extraction subnetwork includes N cascade-connected time-frequency feature extraction layers, and the initial time-frequency feature is input into the time-frequency feature extraction subnetwork for multi-order feature extraction to obtain the time-frequency feature, including:

[0019] The initial time-frequency features are input into N cascade-connected time-frequency feature extraction layers for step-by-step feature extraction to obtain the time-frequency features output by each time-frequency feature extraction layer.

[0020] In one embodiment, the time series feature extraction network includes M cascade-connected time series feature extraction layers, and the historical resource usage data is input into the time series feature extraction network to extract time series features to obtain time series features, including:

[0021] Input the historical resource usage data into the first time series feature extraction layer for feature extraction to obtain the first time series feature;

[0022] For the i-th time series feature extraction layer, the time series features output by the i-1-th time series feature extraction layer and the time-frequency features output by the i-1-th time-frequency feature extraction layer are fused to obtain the first fused features; i is greater than 1 and not greater than M, where M is an integer; N=M;

[0023] The first fusion feature is input to the i-th time series feature extraction layer for time series feature extraction, and the feature output by the M-th time series feature extraction layer is determined as the time series feature.

[0024] In one embodiment, the method further comprises:

[0025] Obtain sample resource usage data on the cloud platform to be predicted;

[0026] The sample resource usage data is input into the initial resource prediction model for training to obtain the resource prediction model; the initial resource prediction model includes an initial time-frequency feature extraction network, an initial time series feature extraction network, an initial depth-separable convolutional layer and an initial prediction network; the output of the initial time-frequency feature extraction network is connected to the input of the initial time series feature extraction network, the output of the initial time series feature extraction network is connected to the input of the initial depth-separable convolutional layer, and the output of the initial depth-separable convolutional layer is connected to the input of the initial prediction network.

[0027] In a second aspect, the present application also provides a resource usage status prediction device, including:

[0028] An acquisition module is used to obtain historical resource usage data on the cloud platform to be predicted;

[0029] The prediction module is used to input historical resource usage data into the resource prediction model to predict the resource usage status within a preset time period and obtain the prediction result; the resource prediction model includes a time-frequency feature extraction network, a time series feature extraction network, a depth-separable convolutional layer and a prediction network; the output of the time-frequency feature extraction network is connected to the input of the time series feature extraction network, the output of the time series feature extraction network is connected to the input of the depth-separable convolutional layer, and the output of the depth-separable convolutional layer is connected to the input of the prediction network.

[0030] In a third aspect, the present application further provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0031] Obtain historical resource usage data on the cloud platform to be predicted;

[0032] The historical resource usage data is input into the resource prediction model to predict the resource usage status within a preset time period to obtain the prediction result; the resource prediction model includes a time-frequency feature extraction network, a time series feature extraction network, a depth-wise separable convolutional layer and a prediction network; the output of the time-frequency feature extraction network is connected to the input of the time series feature extraction network, the output of the time series feature extraction network is connected to the input of the depth-wise separable convolutional layer, and the output of the depth-wise separable convolutional layer is connected to the input of the prediction network.

[0033] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the following steps are implemented:

[0034] Obtain historical resource usage data on the cloud platform to be predicted;

[0035] The historical resource usage data is input into the resource prediction model to predict the resource usage status within a preset time period to obtain the prediction result; the resource prediction model includes a time-frequency feature extraction network, a time series feature extraction network, a depth-wise separable convolutional layer and a prediction network; the output of the time-frequency feature extraction network is connected to the input of the time series feature extraction network, the output of the time series feature extraction network is connected to the input of the depth-wise separable convolutional layer, and the output of the depth-wise separable convolutional layer is connected to the input of the prediction network.

[0036] In a fifth aspect, the present application further provides a computer program product, including a computer program, which implements the following steps when executed by a processor:

[0037] Obtain historical resource usage data on the cloud platform to be predicted;

[0038] The historical resource usage data is input into the resource prediction model to predict the resource usage status within a preset time period to obtain the prediction result; the resource prediction model includes a time-frequency feature extraction network, a time series feature extraction network, a depth-wise separable convolutional layer and a prediction network; the output of the time-frequency feature extraction network is connected to the input of the time series feature extraction network, the output of the time series feature extraction network is connected to the input of the depth-wise separable convolutional layer, and the output of the depth-wise separable convolutional layer is connected to the input of the prediction network.

[0039] The above-mentioned resource usage status prediction method, device, equipment, storage medium and program product obtain the historical resource usage data on the cloud platform to be predicted, and input the historical resource usage data into the resource prediction model to predict the resource usage status within a preset time period to obtain the prediction result; the resource prediction model includes a time-frequency feature extraction network, a time series feature extraction network, a deep separable convolution layer and a prediction network; the output of the time-frequency feature extraction network is connected to the input of the time series feature extraction network, the output of the time series feature extraction network is connected to the input of the deep separable convolution layer, and the output of the deep separable convolution layer is connected to the input of the prediction network. The above-mentioned method predicts the resource usage status of the preset time period through the resource prediction model. The resource prediction model selects the time-frequency feature extraction network, the time series feature extraction network, the deep separable convolution layer and the prediction network. The time-frequency feature extraction network can more accurately reflect the instantaneous frequency and instantaneous amplitude changes of the resource usage data, and can enable the network to have a stronger time-frequency analysis ability when processing non-stationary and nonlinear resource usage data, so as to more accurately capture the key frequency features in the resource usage data, thereby making the predicted resource usage status more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the drawings required for use in the embodiments of the present application or related technical descriptions 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 related drawings can be obtained based on these drawings without paying creative work.

[0041] FIG1 is an application environment diagram of a method for predicting resource usage status in one embodiment;

[0042] Figure 2 A schematic diagram of a flow chart of a method for predicting resource usage status in one embodiment;

[0043] Figure 3 is a flow chart of a method for predicting resource usage status in another embodiment;

[0044] Figure 4 is a flow chart of a method for predicting resource usage status in another embodiment;

[0045] Figure 5 is a flow chart of a method for predicting resource usage status in another embodiment;

[0046] Figure 6 A schematic diagram of a resource prediction model in one embodiment;

[0047] Figure 7 is a flow chart of a method for predicting resource usage status in another embodiment;

[0048] Figure 8 is a flow chart of a method for predicting resource usage status in another embodiment;

[0049] FIG9 is a structural block diagram of a device for predicting resource usage status in one embodiment;

[0050] FIG. 10 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0051] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0052] A cloud platform is a set of servers and hardware in a data center that can provide cloud computing services to customers. Cloud computing is a computing model that provides on-demand computing resources and charges based on the resources used. Users can obtain and use these resources in remote data centers.

[0053] At present, in order to fully meet the actual needs of users, cloud computing service providers usually need to prepare sufficient resources to ensure. If cloud resources cannot meet user needs in a timely manner, it may lead to longer service response time, system instability or frequent failures, affecting user operating experience, causing user dissatisfaction and even customer loss. If resource preparation exceeds actual demand, it will be an unnecessary cost consumption of resources, electricity and manpower maintenance.

[0054] Therefore, if we can predict the utilization of resources and effectively schedule and control resources according to the prediction results, so that the supply and demand relationship of resources can be stabilized, we can effectively prevent the occurrence of resource shortages and ensure service quality and user satisfaction. Based on this, it is very meaningful to predict the possible use of resources of the cloud platform during the calculation process of the cloud platform. However, the existing resource prediction methods have the problem of inaccurate prediction. This application aims to solve this problem.

[0055] After the above introduction to the background technology of the resource usage status prediction method provided by the embodiment of the present application, the implementation environment involved in the resource usage status prediction method provided by the embodiment of the present application will be briefly described below. The resource usage status prediction method provided by the embodiment of the present application can be applied to the cloud platform as shown in Figure 1. The cloud platform includes a server 104, and the server 104 can be implemented with an independent server 104 or a server cluster composed of multiple servers 104. The data storage system 102 can store the data that the server 104 needs to process. The data storage system 102 can be integrated on the server 104, or it can be placed on the cloud or other network servers. Among them, the data storage system 102 can collect and pre-store historical resource usage data on the cloud platform to be predicted, and the server 104 can obtain the historical resource usage data on the cloud platform to be predicted from the data storage system 102, and predict the resource usage status within a preset time period based on the historical resource usage data to obtain the resource usage status within the preset time period.

[0056] In other possible implementations, the resource usage status prediction method provided in the embodiment of the present application can also be applied to terminals, which can be, but are not limited to, various personal computers, laptops, smart phones, tablet computers, IoT devices and portable wearable devices. IoT devices can be smart speakers, smart TVs, smart air conditioners, smart car-mounted devices, etc. Portable wearable devices can be smart watches, smart bracelets, head-mounted devices, etc.

[0057] After introducing the application scenarios of the method for predicting resource usage status provided by the embodiments of the present application, the following focuses on introducing the method for predicting resource usage status described in the present application.

[0058] In one embodiment, Figure 2As shown, a method for predicting resource usage status is provided, and the method is applied to the server in FIG1 as an example for explanation, including the following steps:

[0059] S201. Obtain historical resource usage data on the cloud platform to be predicted.

[0060] The historical resource usage data includes the CPU utilization, memory utilization, network traffic and other indicator data of the cloud platform to be predicted. It should be noted that the storage database of the cloud platform to be predicted can obtain and store the resource usage data on the cloud platform in real time, and the storage database of the cloud platform to be predicted can also obtain and store the resource usage data on the cloud platform periodically according to a preset time period.

[0061] Optionally, the historical resource usage data may be resource usage data within a preset historical time period, for example, three days of historical resource usage data on the cloud platform to be predicted, or one month of historical resource usage data on the cloud platform to be predicted.

[0062] In an embodiment of the present application, when it is necessary to predict the resource usage status of the cloud platform to be predicted within a preset future time period, historical resource usage data can be obtained from the storage database of the cloud platform to be predicted. Optionally, when it is necessary to obtain resource usage data within a historical time period, the corresponding historical time period can be entered in the storage database, the resource usage data corresponding to the historical time period can be queried, and the resource usage data corresponding to the historical time period can be obtained.

[0063] Furthermore, after obtaining the historical resource usage data of the cloud platform to be predicted, the historical resource usage data can also be normalized. The process is as follows:

[0064] 1) Record the acquired historical resource usage data as ;in, These are the resource usage data collected at different historical moments;

[0065] 2) Perform normalization to obtain normalized data , denoted as , where the normalized upper calculation formula is as follows (1):

[0066]

[0067] in for The minimum value in for The maximum value in .

[0068] Furthermore, in order to make the historical resource usage more accurately predict the resource usage status within the preset time period, the normalized data can be sliced ​​using the sliding window method to construct a new time series. The normalized data is divided into a window of h time periods, and (Th) sliding windows are obtained, which are: , , , ..., , and record the data corresponding to this (Th) sliding window as data set D.

[0069] S202: Input historical resource usage data into a resource prediction model to predict resource usage status within a preset time period to obtain a prediction result.

[0070] Among them, the resource prediction model includes a time-frequency feature extraction network, a time series feature extraction network, a depthwise separable convolutional layer and a prediction network. The output of the time-frequency feature extraction network is connected to the input of the time series feature extraction network, the output of the time series feature extraction network is connected to the input of the depthwise separable convolutional layer, and the output of the depthwise separable convolutional layer is connected to the input of the prediction network.

[0071] The prediction time period can be set according to specific business requirements, for example, 5 minutes, 30 minutes, 1 hour or 1 day in the future.

[0072] In the embodiment of the present application, after the historical resource usage data on the cloud platform to be predicted is obtained, the historical resource usage data can be directly input into the resource prediction model to predict the resource usage status within a preset time period, and the resource usage status within the preset time period can be obtained. The resource usage status can be an increase in resource usage rate or a decrease in resource usage rate.

[0073] For example, if the prediction result indicates that the resource utilization rate will increase within the preset time period, the allocation and use of resources can be adjusted in advance to avoid insufficient resources. If the prediction result indicates that the resource utilization rate will decrease within the preset time period, the use of resources can be appropriately reduced to save costs. Furthermore, the cloud platform can implement corresponding resource scheduling and allocation according to the prediction results. For example, for situations where it is necessary to increase resource utilization, the configuration of the cloud platform server can be improved in advance, or the number of cloud platform virtual machines can be increased. For situations where it is necessary to reduce resource utilization, the idle resources of the cloud platform can be suspended or unnecessary cloud platform servers can be shut down. In addition, after implementing the corresponding resource scheduling and allocation strategy according to the prediction results, it is also necessary to monitor and optimize the operation of the resources. If there is a difference between the prediction results and the actual operation, it is necessary to further optimize the corresponding resource scheduling and allocation strategy to ensure the efficient use of cloud platform resources.

[0074] The resource usage status prediction method provided in the embodiment of the present application obtains the historical resource usage data on the cloud platform to be predicted, and inputs the historical resource usage data into the resource prediction model to predict the resource usage status within a preset time period to obtain a prediction result; the resource prediction model includes a time-frequency feature extraction network, a time series feature extraction network, a depth-separable convolution layer and a prediction network; the output of the time-frequency feature extraction network is connected to the input of the time series feature extraction network, the output of the time series feature extraction network is connected to the input of the depth-separable convolution layer, and the output of the depth-separable convolution layer is connected to the input of the prediction network. The above method predicts the resource usage status of a preset time period through a resource prediction model. The resource prediction model selects a time-frequency feature extraction network, a time series feature extraction network, a depth-separable convolution layer and a prediction network. The time-frequency feature extraction network can more accurately reflect the instantaneous frequency and instantaneous amplitude changes of the resource usage data, and can enable the network to have a stronger time-frequency analysis capability when processing non-stationary and nonlinear resource usage data, thereby more accurately capturing the key frequency features in the resource usage data, thereby making the predicted resource usage status more accurate.

[0075] In one embodiment, Figure 2 Based on the embodiment shown, the process of obtaining the prediction result can be described as follows: Figure 3 As shown, the above S202 "inputting historical resource usage data into the resource prediction model to predict the resource usage status within a preset time period to obtain a prediction result" includes:

[0076] S301. Input historical resource usage data into a time-frequency feature extraction network to extract time-frequency features and obtain time-frequency features.

[0077] In the embodiment of the present application, after the historical resource usage data is acquired as described above, the historical resource usage data can be input into a time-frequency feature extraction network to extract time-frequency features to obtain time-frequency features corresponding to the historical resource usage data.

[0078] S302: Input the historical resource usage data into a time series feature extraction network to extract time series features and obtain time series features.

[0079] In the embodiment of the present application, after the historical resource usage data is acquired as described above, the historical resource usage data can be input into a time series feature extraction network for time series feature extraction to obtain time series features corresponding to the historical resource usage data.

[0080] S303: Fuse the time-frequency features and the time series features to obtain target fused features.

[0081] In the embodiment of the present application, after the time-frequency features and the time series features are obtained, the time-frequency features and the time series features can be fused to obtain the target fused features. Optionally, the obtained time-frequency features and the time series features can be input into the fusion network for fusion to obtain the fused target fused features.

[0082] S304: Input the target fusion feature into the depth-wise separable convolutional layer for convolution processing to obtain the processed fusion feature.

[0083] In the embodiment of the present application, after the target fusion features are obtained as described above, the target fusion features are input into the depthwise separable convolutional layer for convolution processing to obtain the processed fusion features.

[0084] S305: Input the processed fusion features into the prediction network for prediction to obtain a prediction result.

[0085] The prediction network may be a feed-forward neural network (FFN).

[0086] In the embodiment of the present application, after the processed fusion features are obtained, the processed fusion features can be input into the prediction network for prediction to obtain a prediction result. Optionally, after the processed fusion features are obtained, the processed fusion features can be input into the feedforward neural network FFN for prediction to obtain a prediction result.

[0087] The resource usage status prediction method provided in the embodiment of the present application adopts deep separable convolution to effectively reduce the number of network parameters and improve computing efficiency. In addition, FFN is used to fuse time series information of different features to further improve the accuracy of resource prediction.

[0088] In one embodiment, Figure 3 Based on the embodiment shown, the above-mentioned time-frequency feature extraction network includes a DWVD distribution transformation layer and a time-frequency feature extraction subnetwork, which can describe the process of obtaining time-frequency features based on historical resource usage data, such as Figure 4 As shown, the above S301 "inputting historical resource usage data into the time-frequency feature extraction network to extract time-frequency features to obtain time-frequency features" includes:

[0089] S401: Input historical resource usage data into the DWVD distribution transformation layer for time-frequency transformation to obtain initial time-frequency features.

[0090] Among them, the DWVD distribution transformation layer can be a discrete Wigner transform (Discrete Wigner-Ville Distribution, DWVD) network, in which the discrete Wigner transform DWVD, that is, discrete Wigner-Ville distribution, is a discrete version of the continuous Wigner-Ville distribution. It obtains the information of the signal in the time and frequency domain by calculating the instantaneous autocorrelation function of the signal. This distribution method has significant advantages in processing non-stationary signals. It can provide time-frequency representation of the signal and has a higher resolution.

[0091] In the embodiment of the present application, after the historical resource usage data is obtained, the historical resource usage data can be input into the DWVD distribution transformation layer for time-frequency transformation to obtain the initial time-frequency features. Optionally, for the historical resource usage data with a sampling period of NT, the historical resource usage data corresponding to any sliding window in the above-obtained data set D is obtained, for example, Perform discrete Wigner transform DWVD to obtain the initial time-frequency characteristics , initial time-frequency characteristics The determination process is shown in the following formula (2):

[0092]

[0093] in, represents the Dirac function, , n, m and k are integers between 0 and N-1 respectively.

[0094] S402: Input the initial time-frequency features into the time-frequency feature extraction subnetwork to perform multi-order feature extraction to obtain time-frequency features.

[0095] The time-frequency feature extraction subnetwork may be a plurality of cascade-connected time-frequency feature extraction layers, each of which can perform convolution on the initial video features and obtain corresponding time-frequency features.

[0096] In the embodiment of the present application, after the initial time-frequency features are obtained, the initial time-frequency features can be input into each cascade-connected time-frequency feature extraction layer in the time-frequency feature extraction subnetwork, and feature extraction is performed respectively to obtain time-frequency features. For example, the time-frequency feature extraction process of each time-frequency feature extraction subnetwork for the initial time-frequency features is as follows:

[0097]

[0098] in, Refers to the computational processing of 1D convolution. Representatives through The transformation process of the activation function.

[0099] Optionally, a method for acquiring time-frequency features is provided below, wherein the time-frequency feature extraction subnetwork includes N cascade-connected time-frequency feature extraction layers, that is, the above S402 "inputting the initial time-frequency features into the time-frequency feature extraction subnetwork for multi-order feature extraction to obtain time-frequency features" includes:

[0100] The initial time-frequency features are input into N cascade-connected time-frequency feature extraction layers for step-by-step feature extraction to obtain the time-frequency features output by each time-frequency feature extraction layer.

[0101] In the embodiment of the present application, after the initial time-frequency features are obtained as described above, the initial time-frequency features can be sequentially input into N cascade-connected time-frequency feature extraction layers for feature extraction step by step, to obtain the time-frequency features output by each time-frequency feature extraction layer. Optionally, in the case where the time-frequency feature extraction subnetwork includes three cascade-connected time-frequency feature extraction layers, the method for obtaining the time-frequency features output by each time-frequency feature extraction layer is as follows:

[0102]

[0103] in, is the first-order time-frequency feature, is the second-order time-frequency feature, It is the third-order time-frequency feature.

[0104] After introducing the process of N cascade-connected time-frequency feature extraction layers outputting each time-frequency feature, it is necessary to further explain the process of inputting historical resource usage data into M cascade-connected time series feature extraction layers to obtain corresponding time series features.

[0105] Further, a method for obtaining time series features is provided below. The time series feature extraction network includes M cascade-connected time series feature extraction layers. Figure 5 The above S302 “inputting historical resource usage data into a time series feature extraction network to extract time series features and obtain time series features” includes:

[0106] S501. Input historical resource usage data into a first time series feature extraction layer for feature extraction to obtain a first time series feature.

[0107] Among them, the first time series feature extraction layer can be a temporal convolutional network (Temporal Convolutional Network, TCN) network.

[0108] In the embodiment of the present application, after the historical resource usage data is obtained, the historical resource usage data can be input into the first time series feature extraction layer for feature extraction to obtain time series features. For example, the following formula (5) provides a method for obtaining the first time series feature:

[0109]

[0110] in, This is the first time series feature.

[0111] S502: For the i-th time series feature extraction layer, the time series features output by the i-1-th time series feature extraction layer and the time-frequency features output by the i-1-th time-frequency feature extraction layer are fused to obtain a first fused feature.

[0112] Wherein, i is greater than 1 and not greater than M, M is an integer, and N=M.

[0113] In the embodiment of the present application, the first time series feature is obtained After that, the first time series feature can be The first time-frequency feature extraction layer output by the first time-frequency feature extraction layer in the N cascade-connected time-frequency feature extraction layers , and fuse to get the first fusion feature ,in, , and then the first fusion feature Input to the second time series feature extraction layer in the M cascade-connected time series feature extraction layers for feature extraction to obtain the second time series feature , and then the second time series feature The second time-frequency feature extraction layer output by the second time-frequency feature extraction layer in the N cascade-connected time-frequency feature extraction layers , and fuse to get the second fusion feature ,in, , ..., for the i-th time series feature extraction layer, the time series features output by the i-1-th time series feature extraction layer in the M cascade-connected time series feature extraction layers and the time-frequency features output by the i-1-th time-frequency feature extraction layer in the N cascade-connected time-frequency feature extraction layers are fused to obtain the first fused feature, wherein, .

[0114] See also Figure 6 After the historical resource usage data is obtained, the historical resource usage data can be input into the time series feature extraction layer 1 for feature extraction to obtain the time series feature , the first time series feature The first time-frequency feature extracted from the time-frequency feature extraction layer 1 in the N cascade-connected time-frequency feature extraction layers , and fuse to get the first fusion feature , and then the first fusion feature Input to the time series feature extraction layer 2 in the M cascade-connected time series feature extraction layers for feature extraction to obtain the second time series feature , and then the second time series feature The second time-frequency feature extracted from the time-frequency feature extraction layer 2 in the N cascade-connected time-frequency feature extraction layers , and fuse to get the second fusion feature ,in, , ..., for the i-th time series feature extraction layer, the time series features output by the time series feature extraction layer i-1 in the M cascade-connected time series feature extraction layers and the time-frequency features output by the time-frequency feature extraction layer i-1 in the N cascade-connected time-frequency feature extraction layers are fused to obtain the first fusion feature ,in, .

[0115] It should be noted that, assuming that the time series of each time series feature extraction layer is , the convolution kernel is , the output expression of the lth neuron after the dilated convolution calculation is shown in the following formula (6):

[0116]

[0117] Among them, k is the convolution kernel size, d is the expansion factor, represents the i-th element in the convolution kernel, It is the time series element that is multiplied by the element in the convolution kernel.

[0118] S503: Input the first fusion feature to the i-th time series feature extraction layer to extract the time series feature, and determine the feature output by the M-th time series feature extraction layer as the time series feature.

[0119] In an embodiment of the present application, after the first fusion feature is obtained as mentioned above, the first fusion feature can be input into the i-th timing feature extraction layer for timing feature extraction to obtain the i-th timing feature, until all M cascade-connected timing feature extraction layers have performed timing feature extraction, and the feature output by the M-th timing feature extraction layer among the M cascade-connected timing feature extraction layers is determined as the timing feature.

[0120] Compared with other time-frequency analysis methods, such as short-time Fourier transform (STFT), the feature extraction method provided in the embodiments of the present application, DWVD, has no frequency leakage and resolution problems, and can better process non-stationary and nonlinear components in the signal. This feature enables the combination of DWVD in neural networks to improve the accuracy of signal processing.

[0121] In one embodiment, Figure 2-Figure 5 Based on the embodiment shown, Figure 7 As shown, the above method also includes:

[0122] S203: Obtain sample resource usage data on the cloud platform to be predicted.

[0123] The sample resource usage data includes the CPU utilization, memory utilization, network traffic and other indicator data of the cloud platform to be predicted. It should be noted that the storage database of the cloud platform to be predicted can obtain and store the sample resource usage data on the cloud platform in real time, and the storage database of the cloud platform to be predicted can also obtain and store the sample resource usage data on the cloud platform periodically according to a preset time period.

[0124] Optionally, the sample resource usage data may be resource usage data within a preset historical time period, for example, three days of historical resource usage data on the cloud platform to be predicted, or one month of historical resource usage data on the cloud platform to be predicted.

[0125] In an embodiment of the present application, when it is necessary to train a resource prediction model to predict the resource usage status within a preset time period, sample resource usage data can be first obtained from a storage database of the cloud platform to be predicted.

[0126] Furthermore, after obtaining the sample resource usage data of the cloud platform to be predicted, the historical resource usage data may be normalized. The normalization process is shown in the above formula (1), and the normalized sample resource usage data is obtained.

[0127] Furthermore, in order to make the sample resource usage adapt to the training requirements of the initial resource prediction model, the normalized data can be sliced ​​using a sliding window method to construct a new time series, and the normalized data is divided into a window of h time periods each, to obtain (Th) sliding windows.

[0128] S204: Input the sample resource usage data into the initial resource prediction model for training to obtain a resource prediction model.

[0129] Among them, the initial resource prediction model includes an initial time-frequency feature extraction network, an initial time series feature extraction network, an initial depth-separable convolutional layer and an initial prediction network; the output of the initial time-frequency feature extraction network is connected to the input of the initial time series feature extraction network, the output of the initial time series feature extraction network is connected to the input of the initial depth-separable convolutional layer, and the output of the initial depth-separable convolutional layer is connected to the input of the initial prediction network.

[0130] In the embodiment of the present application, after the sample resource usage data on the cloud platform to be predicted is obtained, the sample resource usage data can be input into the initial resource prediction model for training to obtain a resource prediction model.

[0131] It should be noted that in the initial resource prediction model stage, the parameters in the model are updated and optimized through back propagation, where the target loss function in the optimization direction uses the root mean square error, see the following formula (7):

[0132]

[0133] Among them, the root mean square error It is used to measure the deviation between the predicted value and the true label. Represents the i-th cloud resource utilization sample The corresponding predicted value is represents the true value corresponding to the actual i-th cloud resource utilization sample, and n represents the number of samples. During the training process, the parameters in the network are updated by the gradient descent method (SGD) so that the target loss function Optimize to the minimum, that is, the global optimal solution. At this time, the corresponding network model reaches convergence. At this time, the training can be stopped and the model parameters of the training stage can be saved.

[0134] The training method of the resource prediction model provided in this application provides a certain theoretical basis for subsequently predicting the resource usage status of a preset time period according to the resource prediction model.

[0135] In one embodiment, see Figure 8 , and also provides a resource usage status prediction method, including:

[0136] S10, obtaining sample resource usage data on the cloud platform to be predicted;

[0137] S11, inputting the sample resource usage data into the initial resource prediction model for training to obtain a resource prediction model;

[0138] S12, obtaining historical resource usage data on the cloud platform to be predicted;

[0139] S13, inputting the historical resource usage data into the time-frequency feature extraction network to extract the time-frequency features and obtain the time-frequency features;

[0140] S14, inputting the historical resource usage data into the time series feature extraction network to extract the time series features and obtain the time series features;

[0141] S15, fusing the time-frequency feature and the time series feature to obtain the target fusion feature;

[0142] S16, inputting the target fusion feature into the depth-separable convolutional layer for convolution processing to obtain the processed fusion feature;

[0143] S17: Input the processed fusion features into the prediction network for prediction to obtain the prediction result.

[0144] The above method predicts the resource usage status of a preset time period through a resource prediction model. The resource prediction model selects a time-frequency feature extraction network, a time series feature extraction network, a deep separable convolutional layer and a prediction network. The time-frequency feature extraction network can more accurately reflect the instantaneous frequency and instantaneous amplitude changes of the resource usage data, and can enable the network to have stronger time-frequency analysis capabilities when processing non-stationary and nonlinear resource usage data, thereby more accurately capturing the key frequency features in the resource usage data, thereby making the predicted resource usage status more accurate.

[0145] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0146] Based on the same inventive concept, the embodiment of the present application also provides a prediction device for resource usage status for implementing the above-mentioned prediction method for resource usage status. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above-mentioned method, so the specific limitations in the one or more resource usage status prediction device embodiments provided below can refer to the limitations of the resource usage status prediction method in the above text, and will not be repeated here.

[0147] In an exemplary embodiment, as shown in FIG9 , a prediction device for resource usage status is provided, comprising: an acquisition module 10 and a prediction module 11, wherein:

[0148] The acquisition module 10 is used to acquire the historical resource usage data on the cloud platform to be predicted.

[0149] The prediction module 11 is used to input historical resource usage data into the resource prediction model to predict the resource usage status within a preset time period and obtain a prediction result; the resource prediction model includes a time-frequency feature extraction network, a time series feature extraction network, a depth-separable convolutional layer and a prediction network; the output of the time-frequency feature extraction network is connected to the input of the time series feature extraction network, the output of the time series feature extraction network is connected to the input of the depth-separable convolutional layer, and the output of the depth-separable convolutional layer is connected to the input of the prediction network.

[0150] In an exemplary embodiment, the prediction module 11 includes: a first feature extraction unit, a second feature extraction unit, a fusion unit, a convolution processing unit and a prediction unit, wherein:

[0151] The first feature extraction unit is specifically used to input the historical resource usage data into the time-frequency feature extraction network to extract the time-frequency features and obtain the time-frequency features;

[0152] The second feature extraction unit is specifically used to input the historical resource usage data into the time series feature extraction network to extract the time series features and obtain the time series features;

[0153] A fusion unit is specifically used to fuse the time-frequency features and the time series features to obtain the target fusion features;

[0154] A convolution processing unit, specifically used to input the target fusion feature into the depth-separable convolution layer for convolution processing to obtain the processed fusion feature;

[0155] The prediction unit is specifically used to input the processed fusion features into the prediction network for prediction to obtain the prediction result.

[0156] In an exemplary embodiment, the time-frequency feature extraction network includes a DWVD distribution transformation layer and a time-frequency feature extraction subnetwork. The first feature extraction unit is further used to input the historical resource usage data into the DWVD distribution transformation layer for time-frequency transformation to obtain initial time-frequency features; and input the initial time-frequency features into the time-frequency feature extraction subnetwork for multi-order feature extraction to obtain time-frequency features.

[0157] In an exemplary embodiment, the time-frequency feature extraction subnetwork includes N cascade-connected time-frequency feature extraction layers, and the first feature extraction unit is specifically used to input the initial time-frequency features into the N cascade-connected time-frequency feature extraction layers for step-by-step feature extraction to obtain the time-frequency features output by each time-frequency feature extraction layer.

[0158] In an exemplary embodiment, the above-mentioned time series feature extraction network includes M cascade-connected time series feature extraction layers, and the above-mentioned second feature extraction unit is specifically used to input historical resource usage data into the first time series feature extraction layer for feature extraction to obtain a first time series feature; for the i-th time series feature extraction layer, the time series features output by the i-1-th time series feature extraction layer and the time-frequency features output by the i-1-th time-frequency feature extraction layer are fused to obtain a first fused feature; the i is greater than 1 and not greater than M, and M is an integer; the N=M; the first fused feature is input into the i-th time series feature extraction layer for time series feature extraction, and the feature output by the M-th time series feature extraction layer is determined as the time series feature.

[0159] In an exemplary embodiment, the above device further includes: a second acquisition module and a training module, wherein:

[0160] The second acquisition module is used to obtain sample resource usage data on the cloud platform to be predicted;

[0161] A training module is used to input sample resource usage data into an initial resource prediction model for training to obtain a resource prediction model; the initial resource prediction model includes an initial time-frequency feature extraction network, an initial time series feature extraction network, an initial depth-separable convolutional layer and an initial prediction network; the output of the initial time-frequency feature extraction network is connected to the input of the initial time series feature extraction network, the output of the initial time series feature extraction network is connected to the input of the initial depth-separable convolutional layer, and the output of the initial depth-separable convolutional layer is connected to the input of the initial prediction network.

[0162] Each module in the above-mentioned resource usage status prediction device can be implemented in whole or in part by software, hardware and a combination thereof. Each of the above-mentioned modules can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute the operations corresponding to each of the above modules.

[0163] In an exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be shown in FIG10. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. Among them, the processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store resource usage data of the cloud platform. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for predicting a resource usage state is implemented.

[0164] Those skilled in the art will understand that the structure shown in FIG. 10 is merely a block diagram of a partial structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different arrangement of components.

[0165] In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented:

[0166] Obtain historical resource usage data on the cloud platform to be predicted;

[0167] The historical resource usage data is input into the resource prediction model to predict the resource usage status within a preset time period to obtain the prediction result; the resource prediction model includes a time-frequency feature extraction network, a time series feature extraction network, a depth-wise separable convolutional layer and a prediction network; the output of the time-frequency feature extraction network is connected to the input of the time series feature extraction network, the output of the time series feature extraction network is connected to the input of the depth-wise separable convolutional layer, and the output of the depth-wise separable convolutional layer is connected to the input of the prediction network.

[0168] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0169] Input the historical resource usage data into the time-frequency feature extraction network to extract the time-frequency features and obtain the time-frequency features;

[0170] Input the historical resource usage data into the time series feature extraction network to extract the time series features and obtain the time series features;

[0171] Fuse the time-frequency features and time series features to obtain the target fusion features;

[0172] Input the target fusion features into the depth-separable convolutional layer for convolution processing to obtain the processed fusion features;

[0173] The processed fusion features are input into the prediction network for prediction to obtain the prediction results.

[0174] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0175] Input the historical resource usage data into the DWVD distribution transformation layer for time-frequency transformation to obtain the initial time-frequency features;

[0176] The initial time-frequency features are input into the time-frequency feature extraction subnetwork for multi-order feature extraction to obtain time-frequency features.

[0177] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0178] The initial time-frequency features are input into N cascade-connected time-frequency feature extraction layers for step-by-step feature extraction to obtain the time-frequency features output by each time-frequency feature extraction layer.

[0179] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0180] Input the historical resource usage data into the first time series feature extraction layer for feature extraction to obtain the first time series feature;

[0181] For the i-th time series feature extraction layer, the time series features output by the i-1-th time series feature extraction layer and the time-frequency features output by the i-1-th time-frequency feature extraction layer are fused to obtain the first fused features; i is greater than 1 and not greater than M, where M is an integer; N=M;

[0182] The first fusion feature is input to the i-th time series feature extraction layer for time series feature extraction, and the feature output by the M-th time series feature extraction layer is determined as the time series feature.

[0183] In one embodiment, when the processor executes the computer program, the processor further implements the following steps:

[0184] Obtain sample resource usage data on the cloud platform to be predicted;

[0185] The sample resource usage data is input into the initial resource prediction model for training to obtain the resource prediction model; the initial resource prediction model includes an initial time-frequency feature extraction network, an initial time series feature extraction network, an initial depth-separable convolutional layer and an initial prediction network; the output of the initial time-frequency feature extraction network is connected to the input of the initial time series feature extraction network, the output of the initial time series feature extraction network is connected to the input of the initial depth-separable convolutional layer, and the output of the initial depth-separable convolutional layer is connected to the input of the initial prediction network.

[0186] In one embodiment, a computer readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

[0187] Obtain historical resource usage data on the cloud platform to be predicted;

[0188] The historical resource usage data is input into the resource prediction model to predict the resource usage status within a preset time period to obtain the prediction result; the resource prediction model includes a time-frequency feature extraction network, a time series feature extraction network, a depth-wise separable convolutional layer and a prediction network; the output of the time-frequency feature extraction network is connected to the input of the time series feature extraction network, the output of the time series feature extraction network is connected to the input of the depth-wise separable convolutional layer, and the output of the depth-wise separable convolutional layer is connected to the input of the prediction network.

[0189] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0190] Input the historical resource usage data into the time-frequency feature extraction network to extract the time-frequency features and obtain the time-frequency features;

[0191] Input the historical resource usage data into the time series feature extraction network to extract the time series features and obtain the time series features;

[0192] Fuse the time-frequency features and time series features to obtain the target fusion features;

[0193] Input the target fusion features into the depth-separable convolutional layer for convolution processing to obtain the processed fusion features;

[0194] The processed fusion features are input into the prediction network for prediction to obtain the prediction results.

[0195] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0196] Input the historical resource usage data into the DWVD distribution transformation layer for time-frequency transformation to obtain the initial time-frequency features;

[0197] The initial time-frequency features are input into the time-frequency feature extraction subnetwork for multi-order feature extraction to obtain time-frequency features.

[0198] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0199] The initial time-frequency features are input into N cascade-connected time-frequency feature extraction layers for step-by-step feature extraction to obtain the time-frequency features output by each time-frequency feature extraction layer.

[0200] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0201] Input the historical resource usage data into the first time series feature extraction layer for feature extraction to obtain the first time series feature;

[0202] For the i-th time series feature extraction layer, the time series features output by the i-1-th time series feature extraction layer and the time-frequency features output by the i-1-th time-frequency feature extraction layer are fused to obtain the first fused features; i is greater than 1 and not greater than M, where M is an integer; N=M;

[0203] The first fusion feature is input to the i-th time series feature extraction layer for time series feature extraction, and the feature output by the M-th time series feature extraction layer is determined as the time series feature.

[0204] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0205] Obtain sample resource usage data on the cloud platform to be predicted;

[0206] The sample resource usage data is input into the initial resource prediction model for training to obtain the resource prediction model; the initial resource prediction model includes an initial time-frequency feature extraction network, an initial time series feature extraction network, an initial depth-separable convolutional layer and an initial prediction network; the output of the initial time-frequency feature extraction network is connected to the input of the initial time series feature extraction network, the output of the initial time series feature extraction network is connected to the input of the initial depth-separable convolutional layer, and the output of the initial depth-separable convolutional layer is connected to the input of the initial prediction network.

[0207] In one embodiment, a computer program product is provided, comprising a computer program, which, when executed by a processor, implements the following steps:

[0208] Obtain historical resource usage data on the cloud platform to be predicted;

[0209] The historical resource usage data is input into the resource prediction model to predict the resource usage status within a preset time period to obtain the prediction result; the resource prediction model includes a time-frequency feature extraction network, a time series feature extraction network, a depth-wise separable convolutional layer and a prediction network; the output of the time-frequency feature extraction network is connected to the input of the time series feature extraction network, the output of the time series feature extraction network is connected to the input of the depth-wise separable convolutional layer, and the output of the depth-wise separable convolutional layer is connected to the input of the prediction network.

[0210] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0211] Input the historical resource usage data into the time-frequency feature extraction network to extract the time-frequency features and obtain the time-frequency features;

[0212] Input the historical resource usage data into the time series feature extraction network to extract the time series features and obtain the time series features;

[0213] Fuse the time-frequency features and time series features to obtain the target fusion features;

[0214] Input the target fusion features into the depth-separable convolutional layer for convolution processing to obtain the processed fusion features;

[0215] The processed fusion features are input into the prediction network for prediction to obtain the prediction results.

[0216] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0217] Input the historical resource usage data into the DWVD distribution transformation layer for time-frequency transformation to obtain the initial time-frequency features;

[0218] The initial time-frequency features are input into the time-frequency feature extraction subnetwork for multi-order feature extraction to obtain time-frequency features.

[0219] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0220] The initial time-frequency features are input into N cascade-connected time-frequency feature extraction layers for step-by-step feature extraction to obtain the time-frequency features output by each time-frequency feature extraction layer.

[0221] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0222] Input the historical resource usage data into the first time series feature extraction layer for feature extraction to obtain the first time series feature;

[0223] For the i-th time series feature extraction layer, the time series features output by the i-1-th time series feature extraction layer and the time-frequency features output by the i-1-th time-frequency feature extraction layer are fused to obtain the first fused features; i is greater than 1 and not greater than M, where M is an integer; N=M;

[0224] The first fusion feature is input to the i-th time series feature extraction layer for time series feature extraction, and the feature output by the M-th time series feature extraction layer is determined as the time series feature.

[0225] In one embodiment, when the computer program is executed by a processor, the following steps are also implemented:

[0226] Obtain sample resource usage data on the cloud platform to be predicted;

[0227] The sample resource usage data is input into the initial resource prediction model for training to obtain the resource prediction model; the initial resource prediction model includes an initial time-frequency feature extraction network, an initial time series feature extraction network, an initial depth-separable convolutional layer and an initial prediction network; the output of the initial time-frequency feature extraction network is connected to the input of the initial time series feature extraction network, the output of the initial time series feature extraction network is connected to the input of the initial depth-separable convolutional layer, and the output of the initial depth-separable convolutional layer is connected to the input of the initial prediction network.

[0228] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., but are not limited to this.

[0229] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0230] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A method for predicting resource usage status, characterized in that: The method comprises: Obtain historical resource usage data on the cloud platform to be predicted; The historical resource usage data is input into a resource prediction model to predict the resource usage status within a preset time period to obtain a prediction result; the resource prediction model includes a time-frequency feature extraction network, a time series feature extraction network, a depth-separable convolutional layer and a prediction network; the output of the time-frequency feature extraction network is connected to the input of the time series feature extraction network, the output of the time series feature extraction network is connected to the input of the depth-separable convolutional layer, and the output of the depth-separable convolutional layer is connected to the input of the prediction network.

2. The method according to claim 1, characterized in that The inputting the historical resource usage data into the resource prediction model to predict the resource usage status within a preset time period to obtain a prediction result includes: Inputting the historical resource usage data into the time-frequency feature extraction network to extract time-frequency features to obtain time-frequency features; Inputting the historical resource usage data into the time series feature extraction network to extract time series features to obtain time series features; Fusing the time-frequency feature and the time series feature to obtain a target fusion feature; Inputting the target fusion feature into the depthwise separable convolutional layer for convolution processing to obtain a processed fusion feature; The processed fusion features are input into the prediction network for prediction to obtain the prediction result.

3. The method according to claim 2, characterized in that The time-frequency feature extraction network includes a DWVD distribution transformation layer and a time-frequency feature extraction subnetwork. The historical resource usage data is input into the time-frequency feature extraction network to extract time-frequency features to obtain time-frequency features, including: Inputting the historical resource usage data into the DWVD distribution transformation layer for time-frequency transformation to obtain initial time-frequency features; The initial time-frequency features are input into the time-frequency feature extraction subnetwork for multi-order feature extraction to obtain the time-frequency features.

4. The method according to claim 3, characterized in that The time-frequency feature extraction subnetwork includes N cascade-connected time-frequency feature extraction layers, and the initial time-frequency feature is input into the time-frequency feature extraction subnetwork for multi-order feature extraction to obtain the time-frequency feature, including: The initial time-frequency features are input into the N cascade-connected time-frequency feature extraction layers for feature extraction step by step, so as to obtain the time-frequency features output by each of the time-frequency feature extraction layers.

5. The method according to claim 4, characterized in that The time series feature extraction network includes M cascade-connected time series feature extraction layers, and the historical resource usage data is input into the time series feature extraction network to extract time series features to obtain time series features, including: Inputting the historical resource usage data into a first time series feature extraction layer for feature extraction to obtain a first time series feature; For the i-th time series feature extraction layer, the time series features output by the i-1-th time series feature extraction layer and the time-frequency features output by the i-1-th time-frequency feature extraction layer are fused to obtain a first fused feature; i is greater than 1 and not greater than M, M is an integer; N=M; The first fusion feature is input to the i-th time series feature extraction layer for time series feature extraction, and the feature output by the M-th time series feature extraction layer is determined as the time series feature.

6. The method according to any one of claims 1 to 5, characterized in that: The method further comprises: Obtaining sample resource usage data on the cloud platform to be predicted; The sample resource usage data is input into an initial resource prediction model for training to obtain the resource prediction model; the initial resource prediction model includes an initial time-frequency feature extraction network, an initial time series feature extraction network, an initial depth-separable convolutional layer and an initial prediction network; the output of the initial time-frequency feature extraction network is connected to the input of the initial time series feature extraction network, the output of the initial time series feature extraction network is connected to the input of the initial depth-separable convolutional layer, and the output of the initial depth-separable convolutional layer is connected to the input of the initial prediction network.

7. A resource usage status prediction device, characterized in that: The device comprises: An acquisition module is used to obtain historical resource usage data on the cloud platform to be predicted; A prediction module is used to input the historical resource usage data into a resource prediction model to predict the resource usage status within a preset time period to obtain a prediction result; the resource prediction model includes a time-frequency feature extraction network, a time series feature extraction network, a depth-separable convolutional layer and a prediction network; the output of the time-frequency feature extraction network is connected to the input of the time series feature extraction network, the output of the time series feature extraction network is connected to the input of the depth-separable convolutional layer, and the output of the depth-separable convolutional layer is connected to the input of the prediction network.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images