Workload prediction method for cloud-native applications and related devices
By performing tree-structured aggregation and correlation coefficient calculation on the operation and maintenance time-series data of cloud-native applications, and combining residual connection networks and generative adversarial networks, the challenges of resource contention and long-sequence prediction in cloud-native application scheduling are solved, achieving more accurate resource allocation and model stability.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- BEIJING UNIV OF POSTS & TELECOMM
- Filing Date
- 2024-03-25
- Publication Date
- 2026-05-19
AI Technical Summary
Existing cloud-native application management environments struggle to meet the diverse needs of cloud-native applications in their scheduling algorithms, leading to resource contention and waste. Traditional time series prediction methods incur high computational overhead on long time series and are difficult to accurately extract features.
By using tree-based aggregation processing of operation and maintenance time series data, correlation coefficients are calculated and information extraction networks based on residual connections are used for prediction. Combined with generative adversarial network optimization models, feature information of long-term series is extracted.
It improves the accuracy of cloud-native application scheduling, reduces resource waste, enhances model stability and convergence speed, and solves the problem of gradient explosion in long sequence states.
Smart Images
Figure CN118296413B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of cloud platform technology, and in particular to a prediction method and related equipment. Background Technology
[0002] Cloud-native application management environments provide runtime environments for various applications by loading massive amounts of templates. Existing task scheduling methods only consider the running status of each node at the time of application placement and score them based on current running status snapshots to select appropriate task placement strategies. Application node selection and resource usage are also specified once at application startup, failing to meet the dynamic adaptation requirements of changing application behavior (e.g., a sudden increase in business volume or an action triggering increased resource usage). Traditional scheduling algorithms typically rely on resource optimization theory. However, due to the diverse characteristics of cloud-native applications, these algorithms often struggle to fully consider all aspects of cloud-native application workloads, thus failing to meet the need for accurate scheduling of cloud-native applications. This leads to resource contention on the platform, impacting application runtime. Furthermore, scheduling algorithms that cannot meet elasticity requirements also cause resource redundancy and waste.
[0003] Workload forecasting for cloud-native applications presents several challenges in this scenario. First, the same application is often supported by multi-dimensional operational data. Furthermore, this operational data is time-series, and its characteristics will vary over time as it changes. Second, traditional time-series forecasting methods and their derivatives are mostly used for short time series; methods for long time series have high computational costs and struggle to accurately extract the sequence features of long-term operational data. Summary of the Invention
[0004] In view of this, the purpose of this application is to propose a prediction method and related equipment.
[0005] To achieve the above objectives, this application provides a prediction method, comprising:
[0006] Based on the internal data weights of the acquired operation and maintenance time series data, a tree-like aggregation process is performed on the operation and maintenance time series data to obtain an operation and maintenance time series data aggregation tree.
[0007] Calculate the correlation coefficient of each vector in the operation and maintenance time series data aggregation tree, and calculate the feature time series data vector corresponding to the operation and maintenance time series data aggregation tree based on the correlation coefficient;
[0008] The feature time series data vector is predicted using an information extraction network with residual connections to obtain the prediction result corresponding to the operation and maintenance time series data.
[0009] In one possible implementation, the step of performing tree-based aggregation processing on the acquired operation and maintenance time-series data based on the internal data weights to obtain an operation and maintenance time-series data aggregation tree includes:
[0010] Calculate the weights of the internal data;
[0011] Based on the internal data weights, a distance metric value is calculated between different operation and maintenance time series data.
[0012] By using the different types of the operation and maintenance time series data as vertices and the distance metric values as edges, an operation and maintenance feature description graph is constructed.
[0013] The edges corresponding to the distance metric values that are less than a preset threshold in the operation and maintenance feature description graph are removed to obtain the operation and maintenance time series data aggregation tree.
[0014] In one possible implementation, the internal data weights are calculated using the following formula:
[0015]
[0016] in, Indicates internal data weights. Indicates in The value of the operation and maintenance time series data at any given time. This indicates the amount of operation and maintenance time-series data.
[0017] In one possible implementation, the correlation coefficient is calculated using the following formula:
[0018]
[0019] in, Represents the correlation coefficient. This represents the first vector in the operation and maintenance time-series data aggregation tree. This represents the second vector in the operation and maintenance time-series data aggregation tree. This indicates the number of vectors in the operation and maintenance time-series data aggregation tree.
[0020] In one possible implementation, the step of calculating the feature time-series data vector corresponding to the operation and maintenance time-series data aggregation tree based on the correlation coefficient includes:
[0021] Each operation and maintenance time-series data in the operation and maintenance time-series data aggregation tree is standardized to obtain the processed operation and maintenance time-series data.
[0022] The processed operation and maintenance time-series data and the correlation coefficient are multiplied together to calculate the feature time-series data vector corresponding to the operation and maintenance time-series data aggregation tree.
[0023] In one possible implementation, the step of using an information extraction network with residual connections to predict the feature time-series data vector to obtain the prediction result corresponding to the operation and maintenance time-series data includes:
[0024] The feature time-series data vector is input into the residual connection information extraction network;
[0025] In the encoder of the information extraction network, the feature temporal data vector is processed using a self-attention mechanism. The processed feature temporal data vector is then subjected to dilated convolution and dimensionality reduction using a pooling layer. The residual connection is used to process the result of the initial self-attention mechanism processing of the feature temporal data vector and the result obtained after the final self-attention mechanism processing to obtain the residual connection result.
[0026] In the decoder of the information extraction network, the feature time-series data vector is processed using a self-attention mechanism. The processed feature time-series data vector is then subjected to dilated convolution and dimensionality reduction using a pooling layer. The residual connection results are then combined to predict the next time-series result of the operation and maintenance time-series data, thereby obtaining the prediction result corresponding to the operation and maintenance time-series data.
[0027] In one possible implementation, the method further includes:
[0028] The prediction results and the feature time-series data vectors are input into a generative adversarial network;
[0029] In the generative adversarial network, the parameters of the information extraction network are updated based on the prediction results and the feature temporal data vector.
[0030] Based on the same inventive concept, embodiments of this application also provide a prediction device, including:
[0031] The aggregation module is configured to perform tree-like aggregation processing on the acquired operation and maintenance time-series data based on the internal data weights of the acquired operation and maintenance time-series data to obtain an operation and maintenance time-series data aggregation tree;
[0032] The calculation module is configured to calculate the correlation coefficient of each vector in the operation and maintenance time series data aggregation tree, and calculate the feature time series data vector corresponding to the operation and maintenance time series data aggregation tree based on the correlation coefficient;
[0033] The prediction module is configured to use an information extraction network with residual connections to predict the feature time series data vector, thereby obtaining the prediction result corresponding to the operation and maintenance time series data.
[0034] Based on the same inventive concept, embodiments of this application also provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the prediction method as described in any of the above.
[0035] Based on the same inventive concept, embodiments of this application also provide a non-transitory computer-readable storage medium storing computer instructions for causing the computer to execute any of the prediction methods described above.
[0036] As can be seen from the above, the prediction method and related equipment provided in this application perform tree-like aggregation processing on the acquired operation and maintenance time-series data based on the internal data weights of the acquired operation and maintenance time-series data to obtain an operation and maintenance time-series data aggregation tree; calculate the correlation coefficient of each vector in the operation and maintenance time-series data aggregation tree, and calculate the feature time-series data vector corresponding to the operation and maintenance time-series data aggregation tree based on the correlation coefficient; predict the feature time-series data vector using an information extraction network with residual connections, and obtain the prediction result corresponding to the operation and maintenance time-series data. This application embodiment extracts deeper information within and between different sequences, and generates feature time-series data from multiple time-series data within a unified aggregation tree based on the correlation coefficient. A residual connection module is added to the main encoder, and the regular convolution is modified to a dilated convolution to extract more features from the time-series data. This effectively ensures that even with stacked network depths, the output layer can directly obtain more input information, thereby maintaining the stability of model training, increasing feature diversity, and accelerating the model's convergence speed. Furthermore, the introduction of generative adversarial networks (GANs) effectively solves the problem of gradient explosion in long sequence states by updating the gradient based on the optimal transmission distance for each iteration of the value function network and then pruning the weights. Attached Figure Description
[0037] To more clearly illustrate the technical solutions in this application or related technologies, the drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0038] Figure 1 This is a schematic diagram of the prediction method in an embodiment of this application;
[0039] Figure 2 This is a detailed flowchart illustrating the prediction method according to an embodiment of this application;
[0040] Figure 3 This is a schematic diagram of the generated operation and maintenance time-series data aggregation tree according to an embodiment of this application;
[0041] Figure 4 This is a schematic diagram of the predictive device structure according to an embodiment of this application;
[0042] Figure 5 This is a schematic diagram of the electronic device structure according to an embodiment of this application. Detailed Implementation
[0043] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with specific embodiments and the accompanying drawings.
[0044] It should be noted that, unless otherwise defined, the technical or scientific terms used in the embodiments of this application should have the ordinary meaning understood by one of ordinary skill in the art to which this application pertains. The terms "first," "second," and similar terms used in the embodiments of this application do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed after the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are only used to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.
[0045] As described in the background section, cloud-native application management environments provide runtime environments for various applications by loading massive amounts of templates. Existing task scheduling methods only consider the running status of each node at the time of application placement and score them based on current running status snapshots to select appropriate task placement strategies. Application node selection and resource usage are also specified only once at application startup, failing to meet the dynamic adaptation requirements of changing application behavior (e.g., a sudden increase in business volume or an action triggering increased resource usage). Traditional scheduling algorithms typically rely on resource optimization theory. However, due to the diverse characteristics of cloud-native applications, these algorithms often struggle to fully consider all aspects of cloud-native application workloads, thus failing to meet the need for accurate scheduling of cloud-native applications. This leads to resource contention on the platform, affecting application runtime. Furthermore, scheduling algorithms that cannot meet elasticity requirements also cause resource redundancy and waste.
[0046] Workload forecasting for cloud-native applications presents several challenges in this scenario. First, the same application is often supported by multi-dimensional operational data. Furthermore, this operational data is time-series, and its characteristics will vary over time as it changes. Second, traditional time-series forecasting methods and their derivatives are mostly used for short time series; methods for long time series have high computational costs and struggle to accurately extract the sequence features of long-term operational data.
[0047] Based on the above considerations, this application proposes a prediction method. The method involves performing tree-like aggregation on the acquired operation and maintenance time-series data based on its internal data weights, resulting in an operation and maintenance time-series data aggregation tree. The correlation coefficients of each vector in the aggregation tree are calculated, and a feature time-series data vector corresponding to the aggregation tree is calculated based on these correlation coefficients. The feature time-series data vector is then predicted using an information extraction network incorporating residual connections to obtain the prediction result corresponding to the operation and maintenance time-series data. This application extracts deeper information within and between different sequences, generating feature time-series data from multiple time-series data within a unified aggregation tree based on the correlation coefficients. A residual connection module is added to the main encoder, and the regular convolution is modified to a dilated convolution to extract more features from the time-series data. This effectively ensures that even with stacked network depths, the output layer can directly obtain more input information, thereby maintaining the stability of model training, increasing feature diversity, and accelerating model convergence. Furthermore, the introduction of generative adversarial networks (GANs) effectively solves the problem of gradient explosion in long sequence states by updating the gradient based on the optimal transmission distance for each iteration of the value function network and then pruning the weights.
[0048] The technical solutions of the embodiments of this application will be described in detail below through specific examples.
[0049] refer to Figure 1 The prediction method of this application includes the following steps:
[0050] Step S101: Based on the internal data weights of the acquired operation and maintenance time series data, perform tree aggregation processing on the operation and maintenance time series data to obtain an operation and maintenance time series data aggregation tree;
[0051] Step S102: Calculate the correlation coefficient of each vector in the operation and maintenance time series data aggregation tree, and calculate the feature time series data vector corresponding to the operation and maintenance time series data aggregation tree based on the correlation coefficient;
[0052] Step S103: The feature time series data vector is predicted using an information extraction network with residual connections to obtain the prediction result corresponding to the operation and maintenance time series data.
[0053] refer to Figure 2 This is a detailed flowchart illustrating the prediction method in an embodiment of this application.
[0054] like Figure 2 As shown in the diagram, firstly, an operational time-series data aggregation tree is constructed, i.e., the tree-shaped aggregation part of the operational data in the figure. Then, each operational time-series data aggregation tree is transformed into feature time-series data. Further, the feature time-series data is input into the Informer network, and the feature time-series data is processed using a multi-head sparse probabilistic self-attention mechanism. Then, the processed feature time-series data is processed using dilated convolutional kernel pooling layers. Combined with a residual module, the results of the initial self-attention mechanism processing of the feature time-series data vector and the results obtained after the final self-attention mechanism processing are further processed to obtain residual connection results. Then, the residual connection results are introduced into the decoder to obtain the final prediction result and output it. At the same time, the parameters of the Informer network are updated using a generative adversarial network (WGAN) based on the prediction result and the feature time-series data vector.
[0055] The following combination Figure 1 and Figure 2 The embodiments of this application are described in detail below.
[0056] For step S101, the operation and maintenance time series data is subjected to tree aggregation processing based on the internal data weights of the acquired operation and maintenance time series data to obtain an operation and maintenance time series data aggregation tree.
[0057] In some embodiments, the step of performing tree-like aggregation processing on the acquired operation and maintenance time-series data based on the internal data weights to obtain an operation and maintenance time-series data aggregation tree includes: calculating the internal data weights; calculating distance metrics between different operation and maintenance time-series data based on the internal data weights; constructing an operation and maintenance feature description graph by using the different types of the operation and maintenance time-series data as vertices and the distance metrics as edges; and removing the edges corresponding to the distance metrics in the operation and maintenance feature description graph that are less than a preset threshold to obtain the operation and maintenance time-series data aggregation tree.
[0058] In some embodiments, the internal data weights are calculated using the following formula:
[0059]
[0060] in, Indicates internal data weights. Indicates in The value of the operation and maintenance time series data at any given time. This indicates the amount of operation and maintenance time-series data.
[0061] In this embodiment, cloud-native heterogeneous task workloads are described by various time-series data (such as CPU, Mem, JobArrive, Network IO, DiskIO, etc.), and there are potential hierarchical and grouping correlations among these different types of time-series data. Therefore, tree-structured aggregation is performed based on the operational data to achieve dimensionality reduction. First, the data weights within the time-series operational data are... This can be represented using Euclidean distance, and the specific calculation can be found in the formula above.
[0062] Then, indicate the process. Distance metric between weighted time-series data of different operations and maintenance processes:
[0063]
[0064] in, and These represent two different types of operational time-series data. This represents the distance metric between two different types of operational time-series data.
[0065] refer to Figure 3 This is a schematic diagram of the generated operation and maintenance time-series data aggregation tree in an embodiment of this application.
[0066] like Figure 3 As shown, further, different types of operation and maintenance time-series data are used as vertices in the graph, and the weights of their edges are distance metrics among different types of operation and maintenance data, generating an operation and maintenance feature description graph for a specific cloud-native task. Based on the operation and maintenance feature description graph, edges with smaller weights (i.e., less relevant edges) are deleted according to their weights (the bottom 20% are removed after sorting), generating several independent tree diagrams from the operation and maintenance feature description graph. Several types of operation and maintenance feature indicators with strong relevance are aggregated into a tree structure for representation. It should be noted that in the embodiments of this application, the threshold for smaller weights is set to the value corresponding to the bottom 20% after sorting. The threshold given in the embodiments of this application is only exemplary, and those skilled in the art can adapt it according to their own needs. No related limitations are imposed in the embodiments of this application.
[0067] For step S102, the correlation coefficient of each vector in the operation and maintenance time series data aggregation tree is calculated, and the feature time series data vector corresponding to the operation and maintenance time series data aggregation tree is calculated based on the correlation coefficient.
[0068] In some embodiments, the correlation coefficient is calculated using the following formula:
[0069]
[0070] in, Represents the correlation coefficient. This represents the first vector in the operation and maintenance time-series data aggregation tree. This represents the second vector in the operation and maintenance time-series data aggregation tree. This indicates the number of vectors in the operation and maintenance time-series data aggregation tree.
[0071] In some embodiments, the step of calculating the feature time-series data vector corresponding to the operation and maintenance time-series data aggregation tree based on the correlation coefficient includes: standardizing each operation and maintenance time-series data in the operation and maintenance time-series data aggregation tree to obtain processed operation and maintenance time-series data; multiplying the processed operation and maintenance time-series data and the correlation coefficient to calculate the feature time-series data vector corresponding to the operation and maintenance time-series data aggregation tree.
[0072] In this embodiment, to further intensively acquire global context information and extract deeper information within and between different sequences, feature time-series data are generated from multiple time-series data within a unified aggregation tree based on correlation coefficients. Each operation and maintenance time-series data aggregation tree contains multiple highly correlated operation and maintenance time-series indicator data. Vectors are used. Represents each type of time-series operation and maintenance data, set This is used to represent the time-series data aggregation tree for operations and maintenance.
[0073]
[0074]
[0075] in, This indicates the number of time series data types within each aggregation tree. This indicates the quantity of operation and maintenance time-series data. This indicates the number of vectors in the operation and maintenance time-series data aggregation tree.
[0076] This application proposes a related layer To calculate the operation and maintenance time series data aggregation tree The correlation between the vectors is calculated using the formula above.
[0077] After calculating the correlation coefficient, the correlation coefficient between each node in the aggregation tree can be obtained, which is a measure of the similarity between various operation and maintenance indicators. Then, each operation and maintenance time-series data is standardized using the following formula:
[0078]
[0079] in, Indicates the first in the aggregate tree The class of Operation and maintenance time-series data, Indicates the first in the aggregate tree Minimum operational time-series data for the class. Indicates the first in the aggregate tree Maximum operational time-series data for the class.
[0080] Furthermore, using the correlation coefficient as weights, the standardized data is multiplied by the correlation coefficient to obtain a feature vector describing the workload information of cloud-native applications. Finally, each aggregation tree generates a feature vector, and the feature vectors from multiple aggregation trees form a set of feature time-series data vectors.
[0081] For step S103, the feature time series data vector is predicted using an information extraction network that introduces residual connections to obtain the prediction result corresponding to the operation and maintenance time series data.
[0082] In some embodiments, the step of using an information extraction network with residual connections to predict the feature time-series data vector to obtain the prediction result corresponding to the operation and maintenance time-series data includes: inputting the feature time-series data vector into the information extraction network with residual connections; in the encoder of the information extraction network, processing the feature time-series data vector using a self-attention mechanism, performing dilated convolution on the processed feature time-series data vector, and performing dimensionality reduction processing using a pooling layer; processing the result of the initial self-attention mechanism processing of the feature time-series data vector and the result obtained after the final self-attention mechanism processing using the residual connections to obtain the residual connection result; in the decoder of the information extraction network, processing the feature time-series data vector using a self-attention mechanism, performing dilated convolution on the processed feature time-series data vector, performing dimensionality reduction processing using a pooling layer, and combining the residual connection result to predict the next time-series result of the operation and maintenance time-series data to obtain the prediction result corresponding to the operation and maintenance time-series data.
[0083] In some embodiments, the method further includes: inputting the prediction result and the feature time-series data vector into a generative adversarial network; and updating the parameters of the information extraction network in the generative adversarial network based on the prediction result and the feature time-series data vector.
[0084] In the embodiments of this application, a residual connection module is added to the information extraction network, and the regular convolution is modified to dilated convolution to extract more features from the temporal data. The key method for adding residual connections is to introduce them into the main encoder. The core concept is to introduce a "direct mapping" approach, where the output of each layer contains not only the original information of that layer but also the new information after nonlinear transformation. This design ensures that even with stacked network depths, the output layer can directly obtain more input information, thereby maintaining the stability of model training. The effect is increased feature diversity and accelerated model convergence.
[0085] The distillation operation of the original Informer network directly reduces the dimensionality of the output of each multi-head attention mechanism layer, which will lose corresponding information and data to a certain extent. Therefore, the embodiments of this application improve the regular convolution in the distillation operation to dilated convolution and reduce the dimensionality through pooling layers, thereby reducing the amount of computation and further improving the efficiency of information extraction.
[0086] In some embodiments, the calculation process of dilated convolution is represented by the following formula:
[0087]
[0088] in, This represents the result of dilated convolution. Indicates size is convolution kernel, for Operation and maintenance time-series data at any given moment. This represents the filter in convolution. This represents the expansion coefficient.
[0089] Furthermore, WGAN is introduced during the Informer training process to update gradients based on the optimal transport distance (Wasserstein), addressing the problem of gradient explosion in long sequence states. WGAN, using a generator and a critic network model, updates the generator after n iterations of the critic. For each iteration of the critic, the gradient is updated based on the Wasserstein distance, and then the weights are pruned, effectively solving the gradient explosion problem in long sequence states while also effectively extracting non-periodic components.
[0090] The specific training process is as follows:
[0091] First, initialize the parameters:
[0092] Set the gradient penalty coefficient (λ), batch size (λ), etc. ), number of commenter iterations ( ), momentum value ( ), learning rate (α) and initial Critic parameters ( ) and initial Generator parameters ( ).
[0093] The following are the specific steps of the training loop:
[0094] In the Generator parameter ( If the convergence fails, perform the following steps:
[0095] First, Critic iteration:
[0096] For each Critic iteration (1 to 1) Second-rate):
[0097] A batch of real data and generated data is obtained by sampling real data and generator output.
[0098] Calculate the Critic loss for each pair of real and generated data ( ).
[0099] The Critic loss of all batches is averaged, and the Critic parameters are updated using the MADGRAD optimizer. ).
[0100] Then perform a Generator update:
[0101] Sample a batch of generator output.
[0102] Calculate the loss of the generator output ( ).
[0103] The Generator parameter (θ) is updated using the MADGRAD optimizer, with the goal of minimizing the loss of the generator output.
[0104] Finally, repeat the training loop until the generator parameters converge.
[0105] The above
[0106]
[0107]
[0108] in For length is Univariate time series, For its time The value. (To be) Split into lengths of of A time series, making the dataset in express The first in A time series, ( ) is the discriminator function used to evaluate the authenticity of the data. MADGRAD indicates model optimization based on the double-average form of AdaGrad.
[0109] As can be seen from the above embodiments, the prediction method described in this application performs tree-like aggregation processing on the acquired operation and maintenance time-series data based on the internal data weights, obtaining an operation and maintenance time-series data aggregation tree; calculates the correlation coefficient of each vector in the operation and maintenance time-series data aggregation tree, and calculates the feature time-series data vector corresponding to the operation and maintenance time-series data aggregation tree based on the correlation coefficient; and predicts the feature time-series data vector using an information extraction network with residual connections, obtaining the prediction result corresponding to the operation and maintenance time-series data. This application embodiment extracts deeper information within and between different sequences, and generates feature time-series data from multiple time-series data within a unified aggregation tree based on the correlation coefficient. A residual connection module is added to the main encoder, and the regular convolution is modified to a dilated convolution to extract more features from the time-series data. This effectively ensures that even with stacked network depths, the output layer can directly obtain more input information, thereby maintaining the stability of model training, increasing feature diversity, and accelerating the model's convergence speed. Furthermore, the introduction of generative adversarial networks (GANs) effectively solves the problem of gradient explosion in long sequence states by updating the gradient based on the optimal transmission distance for each iteration of the value function network and then pruning the weights.
[0110] It should be noted that the method in this embodiment can be executed by a single device, such as a computer or server. The method can also be applied in a distributed scenario, where multiple devices cooperate to complete the task. In such a distributed scenario, one of these devices may execute only one or more steps of the method in this embodiment, and the multiple devices will interact with each other to complete the method described.
[0111] It should be noted that the above description describes some embodiments of this application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in a different order than that shown in the above embodiments and still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
[0112] Based on the same inventive concept, corresponding to the methods of any of the above embodiments, this application also provides a prediction device.
[0113] refer to Figure 4 The prediction device includes:
[0114] The aggregation module 41 is configured to perform tree-like aggregation processing on the acquired operation and maintenance time-series data based on the internal data weights of the acquired operation and maintenance time-series data to obtain an operation and maintenance time-series data aggregation tree;
[0115] The calculation module 42 is configured to calculate the correlation coefficient of each vector in the operation and maintenance time series data aggregation tree, and calculate the feature time series data vector corresponding to the operation and maintenance time series data aggregation tree based on the correlation coefficient;
[0116] Prediction module 43 is configured to predict the feature time series data vector using an information extraction network with residual connections, and obtain the prediction result corresponding to the operation and maintenance time series data.
[0117] For ease of description, the above devices are described in terms of function, divided into various modules. Of course, in implementing this application, the functions of each module can be implemented in one or more software and / or hardware.
[0118] The apparatus of the above embodiments is used to implement the corresponding prediction method in any of the foregoing embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be repeated here.
[0119] Based on the same inventive concept, corresponding to the methods of any of the above embodiments, this application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the prediction method described in any of the above embodiments.
[0120] Figure 5 This embodiment illustrates a more specific hardware structure of an electronic device. The device may include a processor 1010, a memory 1020, an input / output interface 1030, a communication interface 1040, and a bus 1050. The processor 1010, memory 1020, input / output interface 1030, and communication interface 1040 are interconnected internally via the bus 1050.
[0121] The processor 1010 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this specification.
[0122] The memory 1020 can be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 1020 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented by software or firmware, the relevant program code is stored in the memory 1020 and is called and executed by the processor 1010.
[0123] The input / output interface 1030 is used to connect input / output modules to realize information input and output. Input / output modules can be configured as components within the device (not shown in the figure) or externally connected to the device to provide corresponding functions. Input devices may include keyboards, mice, touchscreens, microphones, various sensors, etc., while output devices may include displays, speakers, vibrators, indicator lights, etc.
[0124] The communication interface 1040 is used to connect a communication module (not shown in the figure) to enable communication between this device and other devices. The communication module can communicate via wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).
[0125] Bus 1050 includes a pathway for transmitting information between various components of the device, such as processor 1010, memory 1020, input / output interface 1030, and communication interface 1040.
[0126] It should be noted that although the above-described device only shows the processor 1010, memory 1020, input / output interface 1030, communication interface 1040, and bus 1050, in specific implementations, the device may also include other components necessary for normal operation. Furthermore, those skilled in the art will understand that the above-described device may only include the components necessary for implementing the embodiments of this specification, and not necessarily all the components shown in the figures.
[0127] The electronic devices described above are used to implement the corresponding prediction methods in any of the foregoing embodiments and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.
[0128] Based on the same inventive concept, corresponding to the methods of any of the above embodiments, this application also provides a non-transitory computer-readable storage medium storing computer instructions for causing the computer to perform the prediction method as described in any of the above embodiments.
[0129] The computer-readable medium of this embodiment includes permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by a computing device.
[0130] The computer instructions stored in the storage medium of the above embodiments are used to cause the computer to execute the prediction method as described in any of the above embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.
[0131] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of this application (including the claims) is limited to these examples; within the framework of this application, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of the embodiments of this application as described above, which are not provided in the details for the sake of brevity.
[0132] Additionally, to simplify the description and discussion, and to avoid obscuring the embodiments of this application, the well-known power / ground connections to integrated circuit (IC) chips and other components may or may not be shown in the provided drawings. Furthermore, the apparatus may be shown in block diagram form to avoid obscuring the embodiments of this application, and this also takes into account the fact that the details of the implementation of these block diagram apparatuses are highly dependent on the platform on which the embodiments of this application will be implemented (i.e., these details should be fully understood by those skilled in the art). While specific details (e.g., circuits) have been set forth to describe exemplary embodiments of this application, it will be apparent to those skilled in the art that the embodiments of this application can be implemented without these specific details or with variations thereof. Therefore, these descriptions should be considered illustrative rather than restrictive.
[0133] Although this application has been described in conjunction with specific embodiments thereof, many substitutions, modifications, and variations of these embodiments will be apparent to those skilled in the art from the foregoing description. For example, other memory architectures (e.g., dynamic RAM (DRAM)) may be used with the embodiments discussed.
[0134] The embodiments of this application are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the embodiments of this application should be included within the protection scope of this application.
Claims
1. A workload prediction method for cloud-native applications, characterized in that, include: Based on the internal data weights of the acquired cloud-native application operation and maintenance time series data, the operation and maintenance time series data is subjected to tree aggregation processing to obtain an operation and maintenance time series data aggregation tree. Calculate the correlation coefficient of each vector in the operation and maintenance time series data aggregation tree, and calculate the feature time series data vector corresponding to the operation and maintenance time series data aggregation tree based on the correlation coefficient; The feature time series data vector is predicted using an information extraction network with residual connections to obtain the workload prediction result corresponding to the operation and maintenance time series data. The step of performing tree-based aggregation processing on the acquired operation and maintenance time-series data based on its internal data weights to obtain an operation and maintenance time-series data aggregation tree includes: Calculate the weights of the internal data; Based on the internal data weights, a distance metric value is calculated between different operation and maintenance time series data. By using the different types of the operation and maintenance time series data as vertices and the distance metric values as edges, an operation and maintenance feature description graph is constructed. The edges corresponding to the distance metric values that are less than a preset threshold in the operation and maintenance feature description graph are removed to obtain the operation and maintenance time series data aggregation tree.
2. The method according to claim 1, characterized in that, The internal data weights are calculated using the following formula: in, Indicates internal data weights. Indicates in The value of the operation and maintenance time series data at any given time. This indicates the amount of operation and maintenance time-series data.
3. The method according to claim 1, characterized in that, The correlation coefficient is calculated using the following formula: in, Represents the correlation coefficient. This represents the first vector in the operation and maintenance time-series data aggregation tree. This represents the second vector in the operation and maintenance time-series data aggregation tree. This indicates the number of vectors in the operation and maintenance time-series data aggregation tree.
4. The method according to claim 1, characterized in that, The step of calculating the feature time-series data vector corresponding to the operation and maintenance time-series data aggregation tree based on the correlation coefficient includes: Each operation and maintenance time-series data in the operation and maintenance time-series data aggregation tree is standardized to obtain the processed operation and maintenance time-series data. The processed operation and maintenance time-series data and the correlation coefficient are multiplied together to calculate the feature time-series data vector corresponding to the operation and maintenance time-series data aggregation tree.
5. The method according to claim 1, characterized in that, The step of using an information extraction network with residual connections to predict the feature time-series data vector to obtain the workload prediction result corresponding to the operation and maintenance time-series data includes: The feature time-series data vector is input into the residual connection information extraction network; In the encoder of the information extraction network, the feature temporal data vector is processed using a self-attention mechanism. The processed feature temporal data vector is then subjected to dilated convolution and dimensionality reduction using a pooling layer. The residual connection is used to process the result of the initial self-attention mechanism processing of the feature temporal data vector and the result obtained after the final self-attention mechanism processing to obtain the residual connection result. In the decoder of the information extraction network, the feature time-series data vector is processed using a self-attention mechanism. The processed feature time-series data vector is then subjected to dilated convolution and dimensionality reduction using a pooling layer. The residual connection results are then combined to predict the next time-series result of the operation and maintenance time-series data, thereby obtaining the workload prediction result corresponding to the operation and maintenance time-series data.
6. The method according to claim 5, characterized in that, The method further includes: The workload prediction results and the feature time-series data vectors are input into a generative adversarial network; In the generative adversarial network, the parameters of the information extraction network are updated based on the workload prediction results and the feature time-series data vector.
7. A workload prediction device for cloud-native applications, characterized in that, include: The aggregation module is configured to perform tree-like aggregation processing on the acquired cloud-native application operation and maintenance time-series data based on the internal data weights of the acquired cloud-native application operation and maintenance time-series data to obtain an operation and maintenance time-series data aggregation tree. The calculation module is configured to calculate the correlation coefficient of each vector in the operation and maintenance time series data aggregation tree, and calculate the feature time series data vector corresponding to the operation and maintenance time series data aggregation tree based on the correlation coefficient; The prediction module is configured to use an information extraction network with residual connections to predict the feature time series data vector to obtain the workload prediction result corresponding to the operation and maintenance time series data. The aggregation module is further configured to: Calculate the weights of the internal data; Based on the internal data weights, a distance metric value is calculated between different operation and maintenance time series data. By using the different types of the operation and maintenance time series data as vertices and the distance metric values as edges, an operation and maintenance feature description graph is constructed. The edges corresponding to the distance metric values that are less than a preset threshold in the operation and maintenance feature description graph are removed to obtain the operation and maintenance time series data aggregation tree.
8. An electronic 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 program, it implements the method as described in any one of claims 1 to 6.
9. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to perform the method according to any one of claims 1 to 6.