Computing task processing method and real-time load prediction model training method

Abstract

The embodiment of the present application provides a computing task processing method and a real-time load prediction model training method, which is applied to a first computing network, including: for each second computing network in a plurality of second computing networks, the current conditional load time series and target computing task information of the second computing network are input into the real-time load prediction model, and the real-time load of the second computing network under the condition of adding the target computing task is predicted to obtain the predicted load of the second computing network; wherein the current conditional load time series is the current load time series of the second computing network under the conditions of the existing computing tasks and the existing computing resources; based on the predicted load of each second computing network, a target computing network is determined among the plurality of second computing networks; and the target computing task is sent to the target computing network for execution.

Description

Technical Field

[0001] The present application relates to the field of computers, and in particular to a computing task processing method and a real-time load prediction model training method. Background Art

[0002] In the process of scheduling computing power to the computing power network, due to the limited computing power of the computing power network, when too many computing power tasks are assigned, the load on the computing power network may be too heavy, which will lead to increased resource competition, increased computing power task execution time, and even computing power task execution failure.

[0003] In the related art, the load of the computing network is calculated based on the historical number of successful messages sent to the computing network, the target computing network is determined based on the calculated load, and the target computing task is sent to the target computing network for execution. In this method, the calculated load is proportional to the historical number of successful messages sent to the computing network. This situation easily leads to the selection of a computing network with a smaller number of historical successes as the target computing network, which may cause the subsequent failure of sending the target computing task to the target computing network, thereby failing to process the computing task. Summary of the invention

[0004] The embodiments of the present application provide a computing power task processing method and a real-time load prediction model training method, which are used to reduce the probability of failure in sending a target computing power task.

[0005] In a first aspect, an embodiment of the present application provides a computing task processing method, which is applied to a first computing network, and the method includes: For each of the multiple second computing power networks, the current condition load time series and target computing power task information of the second computing power network are input into the real-time load prediction model, and the real-time load of the second computing power network under the condition of adding the target computing power task is predicted to obtain the predicted load of the second computing power network; wherein the current condition load time series is the current load time series of the second computing power network under the conditions of the existing computing power task and the existing computing power resources; Determining a target computing network among the plurality of second computing networks based on the predicted load of each second computing network; Send the target computing power task to the target computing power network for execution.

[0006] In a second aspect, an embodiment of the present application provides a method for training a real-time load prediction model, the method comprising: Obtain a first granularity input sample, a second granularity input sample, and a third granularity input sample of the current sample; wherein the first granularity input sample includes: a first granularity historical load time series and a historical variable computing power task under the conditions of historical computing power tasks and historical computing power resources; the second granularity input sample includes: a second granularity historical load time series and the historical variable computing power task under the conditions of the historical computing power tasks and the historical computing power resources; the third granularity input sample includes: a third granularity historical load time series and the historical variable computing power task under the conditions of the historical computing power tasks and the historical computing power resources; the first granularity, the second granularity, and the third granularity are different from each other; Inputting the first granularity input sample, the second granularity input sample and the third granularity input sample into the real-time load prediction model to be trained, to obtain the predicted load output by the real-time load prediction model; Based on the loss value of the predicted load compared to the sample output load of the current sample, updating the model parameters of the real-time load prediction model; The above process is iterated until the loss value reaches a preset target.

[0007] In a third aspect, an embodiment of the present application provides a computing task processing device, the device comprising: A load prediction module is used to input the current conditional load time series and target computing task information of each second computing network in the plurality of second computing networks into a real-time load prediction model, predict the real-time load of the second computing network under the condition of adding the target computing task, and obtain the predicted load of the second computing network; wherein the current conditional load time series is the current load time series of the second computing network under the conditions of the existing computing task and the existing computing resources; A network determination module, configured to determine a target computing power network among the plurality of second computing power networks based on the predicted load of each second computing power network; The task sending module is used to send the target computing power task to the target computing power network for execution.

[0008] In a fourth aspect, an embodiment of the present application provides a training device for a real-time load prediction model, the device comprising: A sample acquisition module, used to acquire a first granularity input sample, a second granularity input sample and a third granularity input sample of a current sample; wherein the first granularity input sample includes: a first granularity historical load time series and a historical variable computing power task under the conditions of historical computing power tasks and historical computing power resources; the second granularity input sample includes: a second granularity historical load time series and the historical variable computing power task under the conditions of the historical computing power tasks and the historical computing power resources; the third granularity input sample includes: a third granularity historical load time series and the historical variable computing power task under the conditions of the historical computing power tasks and the historical computing power resources; the first granularity, the second granularity and the third granularity are different from each other; A sample input module, used for inputting the first granularity input sample, the second granularity input sample and the third granularity input sample into the real-time load prediction model to be trained, so as to obtain the predicted load output by the real-time load prediction model; A parameter updating module, configured to update a model parameter of the real-time load prediction model based on a loss value of the predicted load compared to a sample output load of a current sample; The iterative control module is used to control the above modules to iteratively perform the above processing until the loss value reaches a preset target.

[0009] In a fifth aspect, an embodiment of the present application provides an electronic device, including: a memory having a computer program stored thereon; A processor is used to execute the computer program to implement the method described in the first aspect or the second aspect of the embodiment of the present application.

[0010] In a sixth aspect, an embodiment of the present application provides a readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the processor enables the processor to implement the method described in the first aspect or the second aspect of the embodiment of the present application.

[0011] In an embodiment of the present application, for each second computing power network among the multiple second computing power networks, the current load time series of the second computing power network under the conditions of existing computing power tasks and existing computing power resources and the target computing power task information are input into the real-time load prediction model, and the real-time load of the second computing power network under the condition of adding the target computing power task is predicted to obtain the predicted load of the second computing power network. Then, based on the predicted load of each second computing power network, the target computing power network is determined among the multiple second computing power networks, and the target computing power task is sent to the target computing power network for execution. The real-time load prediction model can be used to predict the real-time load of the second computing power network when the target computing power task is sent to the second computing power network for execution, and a suitable target computing power network is selected based on the real-time load, and the target computing power task is sent to the suitable target computing power network for execution, thereby optimizing the computing power task allocation and reducing the probability of failure in computing power task sending. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate one or more embodiments of the present application or the technical solutions in the prior art, the drawings required for describing the embodiments will be briefly introduced below. Obviously, the drawings described below are only drawings of one or more embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0013] Figure 1 It is a schematic diagram of an application scenario of a computing task processing method provided by some embodiments of the present application; Figure 2 It is a flowchart of a computing task processing method provided by some embodiments of the present application; Figure 3 It is a schematic diagram of a training process of a real-time load prediction model provided in some embodiments of the present application; Figure 4 It is a module schematic diagram of a computing task processing device provided in some embodiments of the present application; Figure 5 It is a module schematic diagram of a training device for a real-time load prediction model provided in some embodiments of the present application; Figure 6 It is a schematic diagram of the structure of an electronic device provided in some embodiments of the present application. DETAILED DESCRIPTION

[0014] The embodiments of the present application provide a computing task processing method and a real-time load prediction model training method.

[0015] In order to enable those skilled in the art to better understand the technical solutions in this application, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this application.

[0016] The terms "first", "second", etc. in this application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the terms used in this way can be interchangeable where appropriate, so that the embodiments of the application can be implemented in an order other than those illustrated or described herein.

[0017] Please see Figure 1 , is a schematic diagram of an application scenario of the computing task processing method provided in the embodiment of the present application. Figure 1 As shown, in this application scenario, there are a first computing network 101 and multiple second computing networks 102. The first computing network 101 is any computing network in the computing network system that needs to send computing tasks to other computing networks for execution, and the second computing network 102 is other computing networks other than the first computing network 101 in the computing network system.

[0018] The computing task processing method provided in the embodiment of the present application can be performed as follows: Figure 1 The first computing power network 101 shown is executed so that the first computing power network 101 sends the target computing power task to a suitable second computing power network for execution based on the computing power task processing method provided in an embodiment of the present application, thereby reducing the probability of task sending failure. Figure 2 The following is a flow chart of a computing task processing method provided in an embodiment of the present application. Figure 2 , the computing task processing method provided in the embodiment of the present application is described in detail.

[0019] like Figure 2 As shown, the computing task processing method provided in the embodiment of the present application includes the following steps: S201, for each second computing power network among the multiple second computing power networks, input the current conditional load time series and target computing power task information of the second computing power network into the real-time load prediction model, predict the real-time load of the second computing power network under the condition of adding the target computing power task, and obtain the predicted load of the second computing power network.

[0020] In the embodiment of the present application, the target computing task is a computing task that needs to be sent to other computing networks for execution. It is a variable computing task that changes the existing computing tasks of the computing network. The target computing task information is, for example:

[0021] The current conditional load time series is the current load time series of the second computing power network under the conditions of the existing computing power tasks and existing computing power resources. The current load time series is a load information sequence at each time point within the specified time length up to the current time point. The current load time series of the second computing power network under the conditions of the existing computing power tasks and existing computing power resources is a conditional time series, which is, for example: …

[0022]

[0023] in, Indicates the current computing task information. Indicates the existing computing power resource information; Indicates the current computing task information of the second computing network , Existing computing power resource information Condition, at the current time point Load information.

[0024] The real-time load prediction model is a pre-trained conditional time series prediction model, which is used to predict the real-time load of the second computing network under the condition of adding the target computing task based on the current load time series of the second computing network under its existing computing tasks and existing computing resources and the target computing task information, and output the real-time load. The training method of the real-time load prediction model will be described in detail later.

[0025] The predicted load of the second computing power network is the real-time load of the second computing power network when the target computing power task is sent to the second computing power network for execution, which is, for example:

[0026] in, Adding computing task information of the target computing task to the second computing network, ; b2 is the network resource information when the second computing power network adds the target computing power task, which does not change before and after adding the target computing power task, bincreased=bexist. Indicates the target time point for sending the target computing task to the second computing network , the real-time load of the second computing power network when adding the target computing power task on the basis of the existing computing power tasks and existing computing power resources, that is, the real-time load of the second computing power network when the target computing power task is sent to the second computing power network for execution.

[0027] S202: Determine a target computing power network among the multiple second computing power networks based on the predicted load of each second computing power network.

[0028] In this embodiment of the present application, the target computing power network is a second computing power network used to execute the target computing power task.

[0029] In some embodiments, a second computing power network with the smallest predicted load among multiple second computing power networks may be determined as a target computing power network.

[0030] In some embodiments, the response time of sending the target computing power task to the second computing power network can be further considered, and the target computing power network can be determined among the multiple second computing power networks based on information of two dimensions: the predicted load of each second computing power network and the response time of sending the target computing power task to each second computing power network.

[0031] Specifically, for each second computing power network, a weighted sum of the predicted load of the second computing power network and the response time corresponding to the second computing power network can be calculated; based on the weighted sum corresponding to each second computing power network, a target computing power network is determined among the multiple second computing power networks.

[0032] For example, for each second computing network, the predicted load of the computing network can be The response time corresponding to the computing power network The weighted sum of is obtained to obtain the weighted sum corresponding to the computing power network:

[0033] Afterwards, the weighted sums corresponding to the second computing power networks may be compared, and the second computing power network with the smallest weighted sum may be determined as the target computing power network.

[0034] Among them, in calculating the predicted load The weight of the predicted load is the weighted sum of the response time RT. , weight of response time is the preset value.

[0035] Among them, the response time corresponding to the second computing power network , is the response time of sending the target computing task to the second computing network, which can be obtained based on the sum of the computing task ready time of sending the target computing task to the second computing network and the processing time of transmitting the target computing task to the computing network. The computing task ready time of sending the target computing task to the second computing network and the processing time of transmitting the target computing task to the computing network can be calculated based on the existing method, which will not be repeated here.

[0036] S203: Send the target computing power task to the target computing power network for execution.

[0037] For example, assuming the target computing network is the second computing network 2, the target computing task Send to the second computing power network 2 for execution.

[0038] The computing power task processing method provided in the embodiment of the present application inputs the current load time series of the second computing power network under the conditions of its existing computing power tasks and existing computing power resources and the target computing power task information into a real-time load prediction model for each second computing power network in a plurality of second computing power networks, predicts the real-time load of the second computing power network under the condition of adding the target computing power task, and obtains the predicted load of the second computing power network; based on the predicted load of each second computing power network, determines the target computing power network among the plurality of second computing power networks; sends the target computing power task to the target computing power network for execution, and can predict the real-time load of the second computing power network when the target computing power task is sent to the second computing power network for execution through the real-time load prediction model, selects a suitable target computing power network based on the real-time load, and sends the target computing power task to the suitable target computing power network for execution, thereby optimizing the computing power task allocation and reducing the probability of computing power task sending failure.

[0039] In the embodiment of the present application, the real-time load prediction model is a pre-trained conditional timing prediction model, which can be trained based on an appropriate method. Figure 3 is a flow chart of a training method for a real-time load prediction model provided in an embodiment of the present application. Figure 3 As shown, in some embodiments of the present application, the training method of the real-time load prediction model includes the following steps: S301, obtaining a first granularity input sample, a second granularity input sample, and a third granularity input sample of a current sample.

[0040] In the embodiment of the present application, the first particle size, the second particle size, and the third particle size are different from each other.

[0041] The first granularity input sample includes: a first granularity historical load time series and a historical variable computing task under the conditions of historical computing tasks and historical computing resources. The first granularity historical load time series under the conditions of historical computing tasks and historical computing resources is, for example: ,…

[0042] in, Indicates historical computing tasks , Historical computing resources Under the conditions, historical time point Load information.

[0043] The second granularity input sample includes: a second granularity historical load time series and a historical variable computing task under the conditions of historical computing tasks and historical computing resources. The second granularity historical load time series is obtained by clustering each load in the first granularity historical load time series.

[0044] For example, assuming that in the above first-granularity historical load time series, The values ​​of are the same, so they can be clustered into one category and assigned the cluster identifier CI of the category to which they belong, thereby obtaining the historical load time series of the second granularity: ,…

[0045] The third granularity input sample includes: a third granularity historical load time series and a historical variable computing power task under the conditions of historical computing power tasks and historical computing power resources. The third granularity historical load time series is obtained by mapping and transforming the second granularity historical load time series. In a specific implementation, the second granularity historical load time series can be mapped and transformed based on a preset mapping change function to obtain the third granularity historical load time series.

[0046] The mapping transformation function is, for example, a Fourier transformation function, and the historical load time series of the third granularity is, for example: ,… ) Among them, the historical variable computing tasks in the input samples of the above three granularities are the same.

[0047] S302: Input the first granularity input sample, the second granularity input sample and the third granularity input sample into a real-time load prediction model to be trained to obtain a predicted load output by the real-time load prediction model.

[0048] Among them, the real-time load prediction model to be trained has multiple levels of input layers to receive input samples of different granularities. In specific implementation, the first granularity input sample, the second granularity input sample and the third granularity input sample of the current sample can be respectively input into the first level input layer, the second level input layer and the third level input layer of the target prediction model, so that the target prediction model can predict the real-time load under the condition of adding the historical variable computing power task on the basis of the historical computing power task and historical computing power resources based on the three granularity input samples input into the three levels of input layers, and output the predicted load of the current sample.

[0049] S303: Update the parameters of the real-time load prediction model based on the loss value of the predicted load compared to the sample output load of the current sample.

[0050] The sample output load is the actual load of the current sample when the variable computing task is added. Based on the target loss function, the loss value of the predicted load and the sample output load can be calculated, and the parameters of the target prediction model can be updated based on the back propagation algorithm. After that, the above steps S301 to S303 can be iterated until the loss value reaches the preset target, thereby obtaining a trained real-time load prediction model.

[0051] In specific implementation, a first granularity sample set can be obtained; each sample in the first granularity sample set includes the first granularity input sample and the sample output load. Then, for each sample in the first granularity sample set, the historical load in its first granularity historical load time series is clustered to obtain a second granularity sample set; each sample in the second granularity sample set includes the second granularity input sample and the sample output load. Afterwards, for each sample in the second granularity sample set, the second granularity historical load time series is mapped and transformed to obtain a third granularity sample set; each sample in the third granularity sample set includes the third granularity input sample and the sample output load; Afterwards, the first granularity input sample, the second granularity input sample, and the third granularity input sample of the current sample can be obtained from the first granularity sample set, the second granularity sample set, and the third granularity input sample respectively, and the first granularity input sample, the second granularity input sample, and the third granularity input sample of the current sample are input into the real-time load prediction model to be trained to obtain the predicted load of the current sample. Afterwards, the parameters of the real-time load prediction model are updated based on the loss value of the predicted load of the current sample compared to the sample output load of the current sample.

[0052] Then, the first granularity input sample, the second granularity input sample, and the third granularity input sample of the next sample can be obtained from the first granularity sample set, the second granularity sample set, and the third granularity input sample respectively, and the first granularity input sample, the second granularity input sample, and the third granularity input sample of the next sample are input into the real-time load prediction model to be trained to obtain the predicted load of the next sample. Afterwards, the parameters of the real-time load prediction model are updated based on the loss value of the predicted load of the next sample compared to the sample output load of the next sample.

[0053] And so on, to train the real-time load prediction model.

[0054] In the embodiment of the present application, the first granularity sample set, the second granularity sample set, and the third granularity sample set may be a distributed elastic data set (Resilient Distributed Dataset, i.e., RDD), and the training of the real-time load prediction model is distributed training based on the distributed elastic data set. In specific implementation, the real-time load prediction model may be distributedly trained based on the three distributed elastic data sets to improve training efficiency.

[0055] The training method of the real-time load prediction model provided in the embodiment of the present application trains the real-time load prediction model based on input samples of different granularities, thereby facilitating the real-time load prediction model to learn deeper patterns and trends of the load time series, enabling the real-time load prediction model to output accurate prediction results for different inputs, thereby improving the accuracy of the real-time load prediction model prediction.

[0056] Of course, in addition to the training method provided in the above embodiment, other training methods may also be used to train and obtain the implementation load prediction model.

[0057] For example, the first granularity input sample of the current sample can be obtained, and the first granularity input sample includes: the first granularity historical load time series and the historical variable computing power task under the conditions of the historical computing power task and the historical computing power resource; the first granularity input sample is input into the real-time load prediction model to be trained, and the predicted load under the conditions of the historical variable computing power task is obtained on the basis of the historical computing power task; based on the loss value of the predicted load compared to the historical load, the model parameters of the real-time load prediction model are updated. Afterwards, the above process is iterated to train the real-time load prediction model.

[0058] The real-time load prediction model trained by the above method can also predict the real-time load of the second computing power network under the condition of adding the target computing power task, so as to facilitate determining the target computing power network suitable for the target computing power task based on the predicted real-time load, optimize the computing power allocation, and reduce the probability of failure in sending the computing power task.

[0059] In addition, Figure 2 Corresponding to the computing task processing method shown in FIG. 1 , the embodiment of the present application further provides a computing task processing device 400, such as Figure 4 As shown. Figure 4 As shown, the computing task processing device 400 includes: The load prediction module 401 is used to input the current condition load time series and target computing task information of each second computing network in the plurality of second computing networks into the real-time load prediction model, predict the real-time load of the second computing network under the condition of adding the target computing task, and obtain the predicted load of the second computing network; wherein the current condition load time series is the current load time series of the second computing network under the condition of the existing computing task and the existing computing resources; A network determination module 402, configured to determine a target computing power network among the plurality of second computing power networks based on the predicted load of each second computing power network; The task sending module 403 is used to send the target computing power task to the target computing power network for execution.

[0060] In some embodiments of the present application, the network determination module 402 is further configured to: Based on the predicted load of each second computing power network and the response time of sending the target computing power task to each second computing power network, a target computing power network is determined among the multiple second computing power networks.

[0061] In some embodiments of the present application, the network determination module 402 is specifically configured to: For each second computing power network, calculate a weighted sum of the predicted load of the second computing power network and the response time corresponding to the second computing power network; Based on the weighted sum corresponding to each second computing power network, a target computing power network is determined among the multiple second computing power networks.

[0062] In some embodiments of the present application, the computing task processing device 400 further includes: The first real-time load forecasting model training module is used to: Obtain a first granularity input sample, a second granularity input sample, and a third granularity input sample of the current sample; wherein the first granularity input sample includes: a first granularity historical load time series and a historical variable computing power task under the conditions of historical computing power tasks and historical computing power resources; the second granularity input sample includes: a second granularity historical load time series and the historical variable computing power task under the conditions of the historical computing power tasks and the historical computing power resources; the third granularity input sample includes: a third granularity historical load time series and the historical variable computing power task under the conditions of the historical computing power tasks and the historical computing power resources; the first granularity, the second granularity, and the third granularity are different from each other; Inputting the first granularity input sample, the second granularity input sample and the third granularity input sample into the real-time load prediction model to be trained, to obtain the predicted load output by the real-time load prediction model; Based on the loss value of the predicted load compared to the sample output load of the current sample, updating the model parameters of the real-time load prediction model; The above process is iterated until the loss value reaches a preset target.

[0063] In some embodiments of the present application, the training module of the first real-time load prediction model includes: A first sample set acquisition module, used to acquire a first granularity sample set; each sample in the first granularity sample set includes the first granularity input sample and the sample output load; A second sample set acquisition module is used to cluster the historical load in the first granularity historical load time series of each sample in the first granularity sample set to obtain a second granularity sample set; each sample in the second granularity sample set includes the second granularity input sample and the sample output load; A third sample set acquisition module is used to map and transform the second-granularity historical load time series of each sample in the second-granularity sample set to obtain a third-granularity sample set; each sample in the third-granularity sample set includes the third-granularity input sample and the sample output load; The training module of the first real-time load prediction model is specifically used for: A first granularity input sample, a second granularity input sample, and a third granularity input sample of a current sample are obtained from the first granularity sample set, the second granularity sample set, and the third granularity sample set, respectively.

[0064] In some embodiments of the present application, the first granularity sample set, the second granularity sample set, and the third granularity sample set are distributed elastic data sets, and the training of the real-time load prediction model is distributed training based on the distributed elastic data sets.

[0065] In some embodiments of the present application, the computing task processing device 400 further includes: The training device of the second real-time load prediction model is used for: Obtain a first-granularity input sample of the current sample; wherein the first-granularity input sample includes: a first-granularity historical load time series and a historical variable computing task under the conditions of historical computing tasks and historical computing resources; Inputting the first granularity input sample into the real-time load prediction model to be trained to obtain the predicted load output by the real-time load prediction model; Based on the loss value of the predicted load compared to the sample output load of the current sample, updating the model parameters of the real-time load prediction model; The above process is iterated until the loss value reaches a preset target.

[0066] The computing task processing device 400 provided in the embodiment of the present application can be used as the above Figure 2 The execution subject of the computing task processing method shown can realize Figure 2 The functions of the force task processing method shown in the figure are the same and will not be described again here.

[0067] In addition, Figure 3 Corresponding to the training method of the real-time load prediction model shown in FIG. 1 , the embodiment of the present application further provides a training device 500 for a real-time load prediction model, such as Figure 5 As shown. Figure 5 As shown, the training device 500 of the real-time load prediction model includes: The sample acquisition module 501 is used to acquire a first granularity input sample, a second granularity input sample and a third granularity input sample of the current sample; wherein the first granularity input sample includes: a first granularity historical load time series and a historical variable computing power task under the conditions of historical computing power tasks and historical computing power resources; the second granularity input sample includes: a second granularity historical load time series and a historical variable computing power task under the conditions of the historical computing power tasks and the historical computing power resources; the third granularity input sample includes: a third granularity historical load time series and a historical variable computing power task under the conditions of the historical computing power tasks and the historical computing power resources; the first granularity, the second granularity and the third granularity are different from each other; A sample input module 502, configured to input the first granularity input sample, the second granularity input sample and the third granularity input sample into the real-time load prediction model to be trained, to obtain a predicted load output by the real-time load prediction model; A parameter updating module 503, configured to update the model parameters of the real-time load prediction model based on the loss value of the predicted load compared to the sample output load of the current sample; The iterative control module 504 is used to control the above modules to iteratively perform the above processing until the loss value reaches a preset target.

[0068] In some embodiments of the present application, the real-time load prediction model training device 500 further includes: A first sample set acquisition module, used to acquire a first granularity sample set; each sample in the first granularity sample set includes the first granularity input sample and the sample output load; A second sample set acquisition module is used to cluster the historical load in the first granularity historical load time series of each sample in the first granularity sample set to obtain a second granularity sample set; each sample in the second granularity sample set includes the second granularity input sample and the sample output load; A third sample set acquisition module is used to map and transform the second-granularity historical load time series of each sample in the second-granularity sample set to obtain a third-granularity sample set; each sample in the third-granularity sample set includes the third-granularity input sample and the sample output load; The sample acquisition module is specifically used to respectively acquire the first granularity input sample, the second granularity input sample, and the third granularity input sample of the current sample from the first granularity sample set, the second granularity sample set, and the third granularity sample set.

[0069] In some embodiments of the present application, the first granularity sample set, the second granularity sample set, and the third granularity sample set are distributed elastic data sets, and the training of the real-time load prediction model is distributed training based on the distributed elastic data sets.

[0070] The training device 500 of the real-time load prediction model provided in the embodiment of the present application can be used as the above Figure 3 The execution subject of the training method of the real-time load prediction model shown can achieve Figure 3 The function of the training method of the real-time load prediction model shown in FIG. Since the principle is the same, it will not be repeated here.

[0071] In addition, based on the same inventive concept, an embodiment of the present application also provides an electronic device. Figure 6 6 is a schematic diagram of the structure of an electronic device 600 provided in some embodiments of the present application. Figure 6 As shown, the electronic device 600 includes a processor 601 and a memory 602 .

[0072] The memory 602 stores a computer program, and the processor 601 executes the computer program to implement: For each of the multiple second computing power networks, the current conditional load time series and the target computing power task information of the second computing power network are input into the real-time load prediction model, and the real-time load of the second computing power network under the condition of adding the target computing power task is predicted to obtain the predicted load of the second computing power network; wherein the current conditional load time series is the current load time series of the second computing power network under the conditions of the existing computing power task and the existing computing power resources; Determining a target computing network among the plurality of second computing networks based on the predicted load of each second computing network; Send the target computing power task to the target computing power network for execution.

[0073] In some embodiments, the memory 602 stores a computer program, and the processor 601 executes the computer program to specifically implement: Based on the predicted load of each second computing power network and the response time of sending the target computing power task to each second computing power network, a target computing power network is determined among the multiple second computing power networks.

[0074] In some embodiments, the memory 602 stores a computer program, and the processor 601 executes the computer program to specifically implement: For each second computing power network, calculate a weighted sum of the predicted load of the second computing power network and the response time corresponding to the second computing power network; Based on the weighted sum corresponding to each second computing power network, a target computing power network is determined among the multiple second computing power networks.

[0075] In some embodiments, the memory 602 stores a computer program, and the processor 601 executes the computer program to further implement: Obtain a first granularity input sample, a second granularity input sample, and a third granularity input sample of the current sample; wherein the first granularity input sample includes: a first granularity historical load time series and a historical variable computing power task under the conditions of historical computing power tasks and historical computing power resources; the second granularity input sample includes: a second granularity historical load time series and the historical variable computing power task under the conditions of the historical computing power tasks and the historical computing power resources; the third granularity input sample includes: a third granularity historical load time series and the historical variable computing power task under the conditions of the historical computing power tasks and the historical computing power resources; the first granularity, the second granularity, and the third granularity are different from each other; Inputting the first granularity input sample, the second granularity input sample and the third granularity input sample into the real-time load prediction model to be trained, to obtain the predicted load output by the real-time load prediction model; Based on the loss value of the predicted load compared to the sample output load of the current sample, updating the model parameters of the real-time load prediction model; The above process is iterated until the loss value reaches a preset target.

[0076] In some embodiments, the memory 602 stores a computer program, and the processor 601 executes the computer program to further implement: Acquire a first granularity sample set; each sample in the first granularity sample set includes the first granularity input sample and the sample output load; For each sample in the first granularity sample set, cluster the historical load in the first granularity historical load time series to obtain a second granularity sample set; each sample in the second granularity sample set includes the second granularity input sample and the sample output load; For each sample in the second granularity sample set, the second granularity historical load time series is mapped and transformed to obtain a third granularity sample set; each sample in the third granularity sample set includes the third granularity input sample and the sample output load; A first granularity input sample, a second granularity input sample, and a third granularity input sample of a current sample are obtained from the first granularity sample set, the second granularity sample set, and the third granularity sample set, respectively.

[0077] Among them, the first granularity sample set, the second granularity sample set, and the third granularity sample set are distributed elastic data sets, and the training of the real-time load prediction model is distributed training based on the distributed elastic data sets.

[0078] In some embodiments, the memory 602 stores a computer program, and the processor 601 executes the computer program to further implement: Obtain a first-granularity input sample of the current sample; wherein the first-granularity input sample includes: a first-granularity historical load time series and a historical variable computing task under the conditions of historical computing tasks and historical computing resources; Inputting the first granularity input sample into the real-time load prediction model to be trained to obtain the predicted load output by the real-time load prediction model; Based on the loss value of the predicted load compared to the sample output load of the current sample, updating the model parameters of the real-time load prediction model; The above process is iterated until the loss value reaches a preset target.

[0079] Or, if Figure 6 As shown, the electronic device 600 includes a processor 601 and a memory 602. The memory 602 stores a computer program, and the processor 601 executes the computer program to implement: Obtain a first granularity input sample, a second granularity input sample, and a third granularity input sample of the current sample; wherein the first granularity input sample includes: a first granularity historical load time series and a historical variable computing power task under the conditions of historical computing power tasks and historical computing power resources; the second granularity input sample includes: a second granularity historical load time series and the historical variable computing power task under the conditions of the historical computing power tasks and the historical computing power resources; the third granularity input sample includes: a third granularity historical load time series and the historical variable computing power task under the conditions of the historical computing power tasks and the historical computing power resources; the first granularity, the second granularity, and the third granularity are different from each other; Inputting the first granularity input sample, the second granularity input sample and the third granularity input sample into the real-time load prediction model to be trained, to obtain the predicted load output by the real-time load prediction model; Based on the loss value of the predicted load compared to the sample output load of the current sample, updating the model parameters of the real-time load prediction model; The above process is iterated until the loss value reaches a preset target.

[0080] The memory 602 stores a computer program, and the processor 601 executes the computer program to further implement: Acquire a first granularity sample set; each sample in the first granularity sample set includes the first granularity input sample and the sample output load; For each sample in the first granularity sample set, cluster the historical load in the first granularity historical load time series to obtain a second granularity sample set; each sample in the second granularity sample set includes the second granularity input sample and the sample output load; For each sample in the second granularity sample set, the second granularity historical load time series is mapped and transformed to obtain a third granularity sample set; each sample in the third granularity sample set includes the third granularity input sample and the sample output load; A first granularity input sample, a second granularity input sample, and a third granularity input sample of a current sample are obtained from the first granularity sample set, the second granularity sample set, and the third granularity sample set, respectively.

[0081] Among them, the first granularity sample set, the second granularity sample set, and the third granularity sample set are distributed elastic data sets, and the training of the real-time load prediction model is distributed training based on the distributed elastic data sets.

[0082] In addition, based on the same inventive concept, one or more embodiments of the present application further provide a storage medium for storing a computer program. In a specific embodiment, the storage medium may be a USB flash drive, an optical disk, a hard disk, etc. When the computer program stored in the storage medium is executed by a processor, it can achieve: For each of the multiple second computing power networks, the current condition load time series and target computing power task information of the second computing power network are input into the real-time load prediction model, and the real-time load of the second computing power network under the condition of adding the target computing power task is predicted to obtain the predicted load of the second computing power network; wherein the current condition load time series is the current load time series of the second computing power network under the conditions of the existing computing power task and the existing computing power resources; Determining a target computing network among the plurality of second computing networks based on the predicted load of each second computing network; Send the target computing power task to the target computing power network for execution.

[0083] In some embodiments, when the computer program stored in the storage medium is executed by the processor, it can specifically implement: Based on the predicted load of each second computing power network and the response time of sending the target computing power task to each second computing power network, a target computing power network is determined among the multiple second computing power networks.

[0084] In some embodiments, when the computer program stored in the storage medium is executed by the processor, it can specifically implement: For each second computing power network, calculate a weighted sum of the predicted load of the second computing power network and the response time corresponding to the second computing power network; Based on the weighted sum corresponding to each second computing power network, a target computing power network is determined among the multiple second computing power networks.

[0085] In some embodiments, when the computer program stored in the storage medium is executed by the processor, it can achieve: Obtain a first granularity input sample, a second granularity input sample, and a third granularity input sample of the current sample; wherein the first granularity input sample includes: a first granularity historical load time series and a historical variable computing power task under the conditions of historical computing power tasks and historical computing power resources; the second granularity input sample includes: a second granularity historical load time series and the historical variable computing power task under the conditions of the historical computing power tasks and the historical computing power resources; the third granularity input sample includes: a third granularity historical load time series and the historical variable computing power task under the conditions of the historical computing power tasks and the historical computing power resources; the first granularity, the second granularity, and the third granularity are different from each other; Inputting the first granularity input sample, the second granularity input sample and the third granularity input sample into the real-time load prediction model to be trained, to obtain the predicted load output by the real-time load prediction model; Based on the loss value of the predicted load compared to the sample output load of the current sample, updating the model parameters of the real-time load prediction model; The above process is iterated until the loss value reaches a preset target.

[0086] In some embodiments, when the computer program stored in the storage medium is executed by the processor, it can achieve: Acquire a first granularity sample set; each sample in the first granularity sample set includes the first granularity input sample and the sample output load; For each sample in the first granularity sample set, cluster the historical load in the first granularity historical load time series to obtain a second granularity sample set; each sample in the second granularity sample set includes the second granularity input sample and the sample output load; For each sample in the second granularity sample set, the second granularity historical load time series is mapped and transformed to obtain a third granularity sample set; each sample in the third granularity sample set includes the third granularity input sample and the sample output load; A first granularity input sample, a second granularity input sample, and a third granularity input sample of a current sample are obtained from the first granularity sample set, the second granularity sample set, and the third granularity sample set, respectively.

[0087] Among them, the first granularity sample set, the second granularity sample set, and the third granularity sample set are distributed elastic data sets, and the training of the real-time load prediction model is distributed training based on the distributed elastic data sets.

[0088] In some embodiments, when the computer program stored in the storage medium is executed by the processor, it can achieve: Obtain a first-granularity input sample of the current sample; wherein the first-granularity input sample includes: a first-granularity historical load time series and a historical variable computing task under the conditions of historical computing tasks and historical computing resources; Inputting the first granularity input sample into the real-time load prediction model to be trained to obtain the predicted load output by the real-time load prediction model; Based on the loss value of the predicted load compared to the sample output load of the current sample, updating the model parameters of the real-time load prediction model; The above process is iterated until the loss value reaches a preset target.

[0089] One or more embodiments of the present application further provide a storage medium for storing a computer program. In a specific embodiment, the storage medium may be a USB flash drive, an optical disk, a hard disk, etc. When the computer program stored in the storage medium is executed by a processor, it can achieve: Obtain a first granularity input sample, a second granularity input sample, and a third granularity input sample of the current sample; wherein the first granularity input sample includes: a first granularity historical load time series and a historical variable computing power task under the conditions of historical computing power tasks and historical computing power resources; the second granularity input sample includes: a second granularity historical load time series and the historical variable computing power task under the conditions of the historical computing power tasks and the historical computing power resources; the third granularity input sample includes: a third granularity historical load time series and the historical variable computing power task under the conditions of the historical computing power tasks and the historical computing power resources; the first granularity, the second granularity, and the third granularity are different from each other; Inputting the first granularity input sample, the second granularity input sample and the third granularity input sample into the real-time load prediction model to be trained, to obtain the predicted load output by the real-time load prediction model; Based on the loss value of the predicted load compared to the sample output load of the current sample, updating the model parameters of the real-time load prediction model; The above process is iterated until the loss value reaches a preset target.

[0090] In some embodiments, the storage medium may be a USB flash drive, an optical disk, a hard disk, etc. When the computer program stored in the storage medium is executed by the processor, it can achieve: Acquire a first granularity sample set; each sample in the first granularity sample set includes the first granularity input sample and the sample output load; For each sample in the first granularity sample set, cluster the historical load in the first granularity historical load time series to obtain a second granularity sample set; each sample in the second granularity sample set includes the second granularity input sample and the sample output load; For each sample in the second granularity sample set, the second granularity historical load time series is mapped and transformed to obtain a third granularity sample set; each sample in the third granularity sample set includes the third granularity input sample and the sample output load; A first granularity input sample, a second granularity input sample, and a third granularity input sample of a current sample are obtained from the first granularity sample set, the second granularity sample set, and the third granularity sample set, respectively.

[0091] Among them, the first granularity sample set, the second granularity sample set, and the third granularity sample set are distributed elastic data sets, and the training of the real-time load prediction model is distributed training based on the distributed elastic data sets.

[0092] The above describes specific embodiments of the present 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 an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0093] In short, the above description is only a preferred embodiment of the present application and is not intended to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

[0094] The systems, devices, modules or units described in the above embodiments may be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0095] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. 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 technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0096] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0097] Each embodiment in this application is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

Claims

1. A computing task processing method, characterized in that: Applied to the first computing power network, the method includes: For each of the multiple second computing power networks, the current conditional load time series and the target computing power task information of the second computing power network are input into the real-time load prediction model, and the real-time load of the second computing power network under the condition of adding the target computing power task is predicted to obtain the predicted load of the second computing power network; wherein the current conditional load time series is the current load time series of the second computing power network under the conditions of the existing computing power task and the existing computing power resources; Determining a target computing network among the plurality of second computing networks based on the predicted load of each second computing network; Send the target computing power task to the target computing power network for execution.

2. The method according to claim 1, characterized in that The step of determining a target computing power network from among the plurality of second computing power networks based on the predicted load of each second computing power network includes: Based on the predicted load of each second computing power network and the response time of sending the target computing power task to each second computing power network, a target computing power network is determined among the multiple second computing power networks.

3. The method according to claim 2, characterized in that The step of determining a target computing network among the plurality of second computing networks based on the predicted load of each second computing network and the response time of sending the target computing task to each second computing network includes: For each second computing power network, calculate a weighted sum of the predicted load of the second computing power network and the response time corresponding to the second computing power network; Based on the weighted sum corresponding to each second computing power network, a target computing power network is determined among the multiple second computing power networks.

4. The method according to claim 1, characterized in that: The real-time load prediction model is trained by the following method: Obtain a first granularity input sample, a second granularity input sample, and a third granularity input sample of the current sample; wherein the first granularity input sample includes: a first granularity historical load time series and a historical variable computing power task under the conditions of historical computing power tasks and historical computing power resources; the second granularity input sample includes: a second granularity historical load time series and the historical variable computing power task under the conditions of the historical computing power tasks and the historical computing power resources; the third granularity input sample includes: a third granularity historical load time series and the historical variable computing power task under the conditions of the historical computing power tasks and the historical computing power resources; the first granularity, the second granularity, and the third granularity are different from each other; Inputting the first granularity input sample, the second granularity input sample and the third granularity input sample into the real-time load prediction model to be trained, to obtain the predicted load output by the real-time load prediction model; Based on the loss value of the predicted load compared to the sample output load of the current sample, updating the model parameters of the real-time load prediction model; The above process is iterated until the loss value reaches a preset target.

5. The method according to claim 4, characterized in that The method further comprises: Acquire a first granularity sample set; each sample in the first granularity sample set includes the first granularity input sample and the sample output load; For each sample in the first granularity sample set, cluster the historical load in the first granularity historical load time series to obtain a second granularity sample set; each sample in the second granularity sample set includes the second granularity input sample and the sample output load; For each sample in the second granularity sample set, the second granularity historical load time series is mapped and transformed to obtain a third granularity sample set; each sample in the third granularity sample set includes the third granularity input sample and the sample output load; The obtaining of a first granularity input sample, a second granularity input sample, and a third granularity input sample of the current sample includes: A first granularity input sample, a second granularity input sample, and a third granularity input sample of a current sample are obtained from the first granularity sample set, the second granularity sample set, and the third granularity sample set, respectively.

6. The method according to claim 5, characterized in that The first granularity sample set, the second granularity sample set, and the third granularity sample set are distributed elastic data sets, and the training of the real-time load prediction model is distributed training based on the distributed elastic data sets.

7. The method according to claim 1, characterized in that The real-time load prediction model is trained by the following method: Obtain a first-granularity input sample of the current sample; wherein the first-granularity input sample includes: a first-granularity historical load time series and a historical variable computing task under the conditions of historical computing tasks and historical computing resources; Inputting the first granularity input sample into the real-time load prediction model to be trained to obtain the predicted load output by the real-time load prediction model; Based on the loss value of the predicted load compared to the sample output load of the current sample, updating the model parameters of the real-time load prediction model; The above process is iterated until the loss value reaches a preset target.

8. A method for training a real-time load prediction model, characterized in that: The method comprises: Obtain a first granularity input sample, a second granularity input sample, and a third granularity input sample of the current sample; wherein the first granularity input sample includes: a first granularity historical load time series and a historical variable computing power task under the conditions of historical computing power tasks and historical computing power resources; the second granularity input sample includes: a second granularity historical load time series and the historical variable computing power task under the conditions of the historical computing power tasks and the historical computing power resources; the third granularity input sample includes: a third granularity historical load time series and the historical variable computing power task under the conditions of the historical computing power tasks and the historical computing power resources; the first granularity, the second granularity, and the third granularity are different from each other; Inputting the first granularity input sample, the second granularity input sample and the third granularity input sample into the real-time load prediction model to be trained, to obtain the predicted load output by the real-time load prediction model; Based on the loss value of the predicted load compared to the sample output load of the current sample, updating the model parameters of the real-time load prediction model; The above process is iterated until the loss value reaches a preset target.

9. The method according to claim 8, characterized in that The method further comprises: Acquire a first granularity sample set; each sample in the first granularity sample set includes the first granularity input sample and the sample output load; For each sample in the first granularity sample set, cluster the historical load in the first granularity historical load time series to obtain a second granularity sample set; each sample in the second granularity sample set includes the second granularity input sample and the sample output load; For each sample in the second granularity sample set, the second granularity historical load time series is mapped and transformed to obtain a third granularity sample set; each sample in the third granularity sample set includes the third granularity input sample and the sample output load; The obtaining of a first granularity input sample, a second granularity input sample, and a third granularity input sample of the current sample includes: A first granularity input sample, a second granularity input sample, and a third granularity input sample of a current sample are obtained from the first granularity sample set, the second granularity sample set, and the third granularity sample set, respectively.

10. The method according to claim 8, characterized in that The first granularity sample set, the second granularity sample set, and the third granularity sample set are distributed elastic data sets, and the training of the real-time load prediction model is distributed training based on the distributed elastic data sets.

Images