Cloud platform resource scheduling method and device, equipment, storage medium and program product

Abstract

The invention relates to the technical field of computers, and provides a cloud platform resource scheduling method and device, equipment, a storage medium and a program product. The method comprises the steps of generating an initial resource scheduling strategy based on access request data of a cloud platform; on the basis of a trend prediction model, predicting a user service demand trend, and obtaining demand trend data in a future preset time period; inputting the initial resource scheduling strategy and the demand trend data into a hybrid prediction model to obtain a risk prediction result output by the hybrid prediction model; and adjusting the initial resource scheduling strategy based on a risk prediction result to obtain a target resource scheduling strategy. Through the above mode, the target resource scheduling strategy can dynamically adapt to the future user service demand trend, the possible abnormal risk is avoided while the future user service demand is satisfied, the flexible scheduling of the cloud platform resources is realized, and the scheduling effect is optimized.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular, to a cloud platform resource scheduling method, apparatus, device, storage medium, and program product. Background Art

[0002] As a breakthrough computing resource and service provision model, cloud computing has been widely applied in the network, enabling users to flexibly obtain and use the required computing resources according to their personal needs and release the computing resources when not needed. However, this on-demand service model has brought new problems in resource scheduling and allocation.

[0003] In a cloud computing environment, the core task of cloud platform resource scheduling is to effectively manage and allocate network resources to maximize the performance of the cloud platform resource scheduling system and at the same time meet the ever-changing user needs.

[0004] However, the existing cloud platform resource scheduling methods cannot effectively adapt to dynamic network traffic, resource load, and user needs, resulting in problems such as unbalanced resource utilization, degraded performance, and unstable service quality in the cloud platform. Therefore, the scheduling effect of the existing cloud platform resource scheduling methods is not good. Summary of the Invention

[0005] Embodiments of this application provide a cloud platform resource scheduling method, apparatus, device, storage medium, and program product to solve the technical problem of the poor scheduling effect of the existing cloud platform resource scheduling methods.

[0006] In a first aspect, embodiments of this application provide a cloud platform resource scheduling method, including: generating an initial resource scheduling policy based on the access request data of the cloud platform; predicting the trend of user service requirements based on a pre-trained trend prediction model to obtain demand trend data within a preset future time period; inputting the initial resource scheduling policy and the demand trend data into a pre-trained hybrid prediction model to obtain a risk prediction result output by the hybrid prediction model; the risk prediction result is the abnormal risk value that occurs in the cloud platform after resource scheduling according to the initial resource scheduling policy under the demand trend data; adjusting the initial resource scheduling policy based on the risk prediction result to obtain a target resource scheduling policy.

[0007] In an embodiment, the hybrid prediction model includes a convolutional neural network layer and a long short-term memory model connected in sequence; the convolutional neural network layer is used to extract features from the initial resource scheduling policy and the demand trend data to generate feature information; the long short-term memory model is used to simulate and predict the initial resource scheduling policy according to the demand trend data based on the feature information to generate a risk prediction result.

[0008] In one embodiment, before inputting the initial resource scheduling policy and demand trend data into a pre-trained hybrid prediction model to obtain the risk prediction result output by the hybrid prediction model, it further includes: constructing a training sample set and an initial hybrid prediction model; the initial hybrid prediction model includes an initial convolutional neural network layer and an initial long short-term memory model connected in sequence, the initial long short-term memory model includes at least one initial long short-term memory network layer, and each initial long short-term memory network layer includes multiple neurons; based on a swarm intelligence algorithm, optimizing the model structure of the initial hybrid prediction model to generate optimal hyperparameters; the model structure includes the number of initial long short-term memory network layers and the number of neurons in each initial long short-term memory network layer; based on the training sample set and the optimal hyperparameters, pre-training the initial hybrid prediction model to generate a hybrid prediction model.

[0009] In one embodiment, the swarm intelligence algorithm is the grey wolf algorithm; based on the swarm intelligence algorithm, optimizing the model structure of the initial hybrid prediction model to generate optimal hyperparameters includes: initializing the parameters of the grey wolf algorithm to generate an initial population; the initial population includes multiple grey wolf individuals; determining the position information of each grey wolf individual; determining the fitness value of each grey wolf individual, and based on the fitness value ranking, selecting multiple alpha wolves from all grey wolf individuals and saving the position information of all alpha wolves; determining a target ordinary grey wolf individual and mutating and updating the position information of the target ordinary grey wolf individual; the target ordinary grey wolf individual is randomly selected from all ordinary grey wolf individuals, and the ordinary grey wolf individuals are grey wolf individuals other than alpha wolves; based on the original grey wolf algorithm strategy, updating the position information of the remaining ordinary grey wolf individuals and updating the position of the initial population to obtain an updated population; the remaining ordinary grey wolf individuals are ordinary grey wolf individuals other than the target ordinary grey wolf individual; based on the updated population, determining whether the iteration termination condition is satisfied; if the iteration termination condition is not satisfied, returning to the step of determining the fitness value of each grey wolf individual until the iteration termination condition is satisfied, completing the optimization of the model structure of the initial hybrid prediction model, and generating optimal hyperparameters.

[0010] In one embodiment, generating an initial resource scheduling policy based on the access request data of the cloud platform includes: monitoring the resource access information of each computing node of the cloud platform to obtain access request data; performing conversion processing on the access request data to generate a resource demand description; generating an initial resource scheduling policy based on the resource demand description.

[0011] In one embodiment, predicting the user service demand trend based on a pre-trained trend prediction model to obtain demand trend data for a future preset time period includes: obtaining the historical access request data of the cloud platform; based on the trend prediction model, predicting the user service demand trend for a future preset time period according to the historical access request data to obtain demand trend data.

[0012] In a second aspect, an embodiment of the present application provides a cloud platform resource scheduling device, including: a scheduling policy generation module, configured to generate an initial resource scheduling policy based on access request data of the cloud platform; a trend prediction module, configured to predict the trend of user service requirements based on a pre-trained trend prediction model, and obtain demand trend data within a preset future time period; a risk prediction module, configured to input the initial resource scheduling policy and the demand trend data into a pre-trained hybrid prediction model, and obtain a risk prediction result output by the hybrid prediction model; the risk prediction result is an abnormal risk value that occurs in the cloud platform after resource scheduling according to the initial resource scheduling policy under the demand trend data; a scheduling policy adjustment module, configured to adjust the initial resource scheduling policy based on the risk prediction result to obtain a target resource scheduling policy.

[0013] In a third aspect, an embodiment of the present application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements any one of the cloud platform resource scheduling methods as described above.

[0014] In a fourth aspect, an embodiment of the present application provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements any one of the cloud platform resource scheduling methods as described above.

[0015] In a fifth aspect, an embodiment of the present application provides a computer program product, including a computer program. When the computer program is executed by a processor, it implements any one of the cloud platform resource scheduling methods as described above.

[0016] The cloud platform resource scheduling method, device, equipment, storage medium and program product provided by the embodiments of the present application first generate an initial resource scheduling policy according to the access request data of the cloud platform, then use a trend prediction model to predict the user service demand trend, obtain the demand trend data within a preset future time period, and input the initial resource scheduling policy and the demand trend data into a hybrid prediction model to output the risk prediction result of the cloud platform after resource scheduling according to the initial resource scheduling policy under the demand trend data. Finally, the initial resource scheduling policy is adjusted according to the risk prediction result to obtain the target resource scheduling policy. Through the above method, the consideration of the user service demand trend in the future time period is introduced in the process of cloud platform resource scheduling. The hybrid prediction model is used to predict the abnormal risks that may occur in the cloud platform after resource scheduling according to the initial resource scheduling policy under the demand trend data in the future time period, and the initial resource scheduling policy is adjusted according to the risk prediction result, so that the final target resource scheduling policy can dynamically adapt to the future user service demand trend, avoid the possible abnormal risks while meeting the future user service demands, thereby reducing the possibility of unstable cloud platform service quality, realizing the flexible scheduling of cloud platform resources, being beneficial to improving the performance of the cloud platform and the balance of cloud platform resource utilization, and optimizing the scheduling effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the present application or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0018] Figure 1 It is one of the schematic flowcharts of the cloud platform resource scheduling method provided by the embodiments of the present application.

[0019] Figure 2 It is the schematic system architecture diagram of the cloud platform resource scheduling system provided by the embodiments of the present application.

[0020] Figure 3 It is the second schematic flowchart of the cloud platform resource scheduling method provided by the embodiments of the present application.

[0021] Figure 4 It is the schematic flowchart of the hybrid prediction model training method provided by the embodiments of the present application.

[0022] Figure 5 It is the schematic structural diagram of the cloud platform resource scheduling device provided by the embodiments of the present application.

[0023] Figure 6 It is the schematic structural diagram of the electronic device provided by the embodiments of the present application. Detailed implementation manners

[0024] To make the objectives, technical solutions and advantages of this application clearer, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of this application. Apparently, the described embodiments are some but not all of the embodiments of this application. All other embodiments obtained by those of ordinary skill in the art without making creative efforts based on the embodiments in this application belong to the scope of protection of this application.

[0025] Please refer to Figure 1 , Figure 1 which is one of the schematic flowcharts of the cloud platform resource scheduling method provided by the embodiments of this application. As Figure 1 shown, in the embodiments of this application, the cloud platform resource scheduling method is applied to a cloud platform resource scheduling system. The cloud platform resource scheduling method includes steps S110 to S140. The specific steps are as follows: S110: Generate an initial resource scheduling policy based on the access request data of the cloud platform.

[0026] Please refer to Figure 2 , Figure 2 which is the schematic system architecture diagram of the cloud platform resource scheduling system provided by the embodiments of this application.

[0027] As Figure 2 shown, the embodiments of this application provide a cloud platform resource scheduling system based on user demand prediction. The cloud platform resource scheduling system can schedule and manage multiple distributed computing nodes of the cloud platform through an access platform to meet the business requirements of different users. Among them, the cloud platform resource scheduling system includes an access monitoring unit, a scheduling policy decision-making unit, an execution unit, a scheduling monitoring unit, and a prediction unit.

[0028] Optionally, the access monitoring unit is used to monitor the resource access situations of each computing node on the cloud platform in real time, collect and analyze the access request data of users for cloud computing resources, convert the access request data into a standard resource requirement description, and send the resource requirement description to the scheduling policy decision-making unit and the prediction unit.

[0029] Optionally, the scheduling policy decision-making unit is used to generate an initial resource scheduling policy according to the resource requirement description sent by the access monitoring unit, and send the generated initial resource scheduling policy to the execution unit and the prediction unit.

[0030] Optionally, the execution unit is used to call the open interface of the cloud platform according to the initial resource scheduling policy to perform operations of creating, configuring, starting, stopping, and releasing cloud computing resources.

[0031] Optionally, the scheduling monitoring unit is used to monitor the execution of the initial resource scheduling policy in real time and feedback the monitoring results to the prediction unit.

[0032] Optionally, the prediction unit is used to predict the trend of user service demand in the future for a period of time based on a pre-trained trend prediction model, and simulate and predict the initial resource scheduling policy generated by the scheduling policy decision unit according to the demand trend. It determines that after resource scheduling according to the initial resource scheduling policy under the predicted demand trend, the cloud platform may have an abnormal risk value, and takes the abnormal risk value as the risk prediction result and feedbacks it to the scheduling policy decision unit, so that the scheduling policy decision unit can dynamically adjust the initial resource scheduling policy according to the risk prediction result.

[0033] Please refer to Figure 3 , Figure 3 which is the second schematic flowchart of the cloud platform resource scheduling method provided by the embodiments of the present application.

[0034] As Figure 3 shown, in the embodiments of the present application, the cloud platform resource scheduling system can use the access monitoring unit to monitor the resource access information of each computing node of the cloud platform, obtain access request data, and perform conversion processing on the access request data to generate a resource demand description.

[0035] Optionally, in addition to monitoring the access request data received by the cloud platform in real time, the access monitoring unit can also obtain the historical access request data of the cloud platform by reading the cloud platform logs.

[0036] Optionally, the access monitoring unit can also preprocess the collected access request data. For example, it standardizes the format of the access request data to ensure that the access request data has a consistent structure for subsequent processing and analysis, and performs cleaning and denoising processing on the collected access request data to exclude outliers and invalid data to ensure the accuracy of subsequent prediction of the user service demand trend through these data.

[0037] Furthermore, the access monitoring unit sends the generated resource demand description to the scheduling policy decision unit; the scheduling policy decision unit generates an initial resource scheduling policy according to the resource demand description and the current operating status of each computing node of the cloud platform, and sends the generated initial resource scheduling policy to the execution unit and the prediction unit.

[0038] Optionally, to promptly respond to user service requirements and avoid service interruptions, the execution unit can directly perform cloud platform resource scheduling according to the initial resource scheduling policy, call the open interface of the cloud platform, perform operations such as creating, configuring, starting, stopping, and releasing cloud computing resources, and monitor the execution of the initial resource scheduling policy in real time through the scheduling monitoring unit, and feedback the monitoring results to the prediction unit.

[0039] Optionally, to ensure the reliability and accuracy of cloud platform resource scheduling, after receiving the initial resource scheduling policy, the execution unit does not immediately execute the initial resource scheduling policy, but waits for the scheduling policy decision unit to dynamically adjust the initial resource scheduling policy based on the risk prediction result and generate the target resource scheduling policy, and then execute the target resource scheduling policy.

[0040] S120: Based on the pre-trained trend prediction model, predict the trend of user service requirements to obtain the demand trend data within a preset future time period.

[0041] Specifically, the prediction unit can predict the trend of user service requirements within a preset future time period based on the historical access request data of the cloud platform collected by the access monitoring unit to obtain the demand trend data within a preset future time period.

[0042] Optionally, the prediction unit can perform modeling analysis and trend prediction based on the historical access request data collected by the access monitoring unit. Using the time series decomposition method, the historical access request data is decomposed into three categories: long-term change trend, periodic change trend, and random sudden change trend, and data fitting is performed on each of the three trends respectively. A normal distribution model with both mean and variance being constants is used for probability prediction, and then the demand trend data within a preset future time period is predicted.

[0043] Optionally, the prediction unit can also train a machine learning model (such as a convolutional neural network model, a transfer tensor model, etc.) based on the collected historical access request data to obtain a trend prediction model, and then complete the prediction of the demand trend data within a preset future time period according to the trend prediction model.

[0044] S130: Input the initial resource scheduling policy and the demand trend data into the pre-trained hybrid prediction model to obtain the risk prediction result output by the hybrid prediction model.

[0045] The risk prediction result is the abnormal risk value that appears on the cloud platform after resource scheduling according to the initial resource scheduling policy under the demand trend data.

[0046] Specifically, the prediction unit calls the pre-trained hybrid prediction model, and inputs the initial resource scheduling policy and demand trend data into the hybrid prediction model. The hybrid prediction model simulates and predicts the initial resource scheduling policy generated by the scheduling policy decision unit according to the predicted demand trend data, determines the possible abnormal risk value of the cloud platform after resource scheduling according to the initial resource scheduling policy under the predicted demand trend, and feeds back the abnormal risk value as the risk prediction result to the scheduling policy decision unit.

[0047] S140: Based on the risk prediction result, adjust the initial resource scheduling policy to obtain the target resource scheduling policy.

[0048] Specifically, after receiving the risk prediction result of the prediction unit, the scheduling policy decision unit can adjust the initial resource scheduling policy according to the risk prediction result to obtain the target resource scheduling policy, and schedule and manage the resources of each distributed computing node of the cloud platform according to the target resource scheduling policy.

[0049] The cloud platform resource scheduling method provided by the embodiments of this application first generates an initial resource scheduling policy according to the access request data of the cloud platform, then uses the trend prediction model to predict the user service demand trend, obtains the demand trend data within a preset future time period, and inputs the initial resource scheduling policy and the demand trend data into the hybrid prediction model to output the risk prediction result of the cloud platform after resource scheduling according to the initial resource scheduling policy under the demand trend data. Finally, the initial resource scheduling policy is adjusted according to the risk prediction result to obtain the target resource scheduling policy. Through the above method, the consideration of the user service demand trend in the future time period is introduced in the process of cloud platform resource scheduling. The hybrid prediction model is used to predict the possible abnormal risks of the cloud platform after resource scheduling according to the initial resource scheduling policy under the demand trend data in the future time period, and the initial resource scheduling policy is adjusted according to the risk prediction result, so that the final target resource scheduling policy can dynamically adapt to the future user service demand trend, avoid possible abnormal risks while meeting the future user service demand, thereby reducing the possibility of unstable cloud platform service quality, realizing flexible scheduling of cloud platform resources, being beneficial to improving the performance of the cloud platform and the balance of cloud platform resource utilization, and optimizing the scheduling effect.

[0050] In some embodiments, the hybrid prediction model includes a convolutional neural network layer and a long short-term memory model connected in sequence. The convolutional neural network layer is used to extract features from the initial resource scheduling policy and the demand trend data to generate feature information. The long short-term memory model is used to simulate and predict the initial resource scheduling policy according to the demand trend data based on the feature information to generate the risk prediction result.

[0051] In some embodiments, before inputting the initial resource scheduling policy and demand trend data into the pre-trained hybrid prediction model to obtain the risk prediction result output by the hybrid prediction model, the following steps are further included: constructing a training sample set and an initial hybrid prediction model; the initial hybrid prediction model includes an initial convolutional neural network layer and an initial long short-term memory model connected in sequence, the initial long short-term memory model includes at least one initial long short-term memory network layer, and each initial long short-term memory network layer includes multiple neurons; based on a swarm intelligence algorithm, optimizing the model structure of the initial hybrid prediction model to generate optimal hyperparameters; the model structure includes the number of layers of the initial long short-term memory network layer and the number of neurons in each initial long short-term memory network layer; based on the training sample set and the optimal hyperparameters, pre-training the initial hybrid prediction model to generate a hybrid prediction model.

[0052] Please continue to refer to Figure 3 , it can be understood that before using the hybrid prediction model for risk prediction, it is necessary to first construct an initial hybrid prediction model and pre-train the initial hybrid prediction model using a swarm intelligence algorithm to obtain a hybrid prediction model.

[0053] Please refer to Figure 4 , Figure 4 is a schematic flowchart of the hybrid prediction model training method provided by the embodiments of the present application.

[0054] Specifically, as Figure 4 shown, first construct a training sample set according to the pre-processed historical access request data, historical resource scheduling policies, and the execution situations corresponding to the historical resource scheduling policies; then construct an initial hybrid prediction model.

[0055] It should be noted that the initial hybrid prediction model in this embodiment includes an initial convolutional neural network layer and an initial long short-term memory model connected in sequence, the initial long short-term memory model includes at least one initial long short-term memory network layer, and each initial long short-term memory network layer includes multiple neurons.

[0056] The convolutional neural network (CNN) layer is a type of feedforward neural network (Feedforward Neural Networks) that contains convolutional calculations and has a deep structure. It is one of the typical structures of deep learning algorithms and can efficiently extract and learn relevant feature information from data automatically.

[0057] In the embodiments of the present application, the prediction unit can extract features from the collected real-time access request data, initial resource scheduling policy, and demand trend data through the CNN layer of the trained hybrid prediction model to generate feature information.

[0058] Considering that the embodiments of this application need to predict the execution of the initial resource scheduling strategy in a future period according to the demand trend data, therefore, a Long Short Term Memory (LSTM) model is added to the hybrid prediction model. The long short term memory model includes at least one long short term memory network layer (i.e., the LSTM layer), and the LSTM layer has good performance in dealing with time series problems. In the embodiments of this application, the CNN layer is fused with the LSTM model, and the CNN layer is set before the LSTM model to form a hybrid prediction model with a CNN-LSTM structure. After the execution unit executes the initial resource scheduling strategy, the current monitoring data, real-time access request data, initial resource scheduling strategy, and predicted demand trend data collected by the scheduling monitoring unit are used as inputs. These data will first enter the CNN layer, and after being processed by the convolutional layer and the pooling layer, feature information is generated. These generated feature information will be further input into the LSTM model for further processing; the LSTM model will simulate and predict the initial resource scheduling strategy according to the feature information and the demand trend data, and generate a risk prediction result.

[0059] In the actual application process, since the user's business requirements are changing rapidly, outputting the risk prediction result in a timely manner can ensure that the cloud platform resource scheduling system can adjust the resource scheduling strategy of the cloud platform in a timely manner according to the prediction result, thereby avoiding the problems of resource congestion and scheduling failures on the cloud platform.

[0060] Furthermore, in order to improve the simulation and prediction efficiency of the initial resource scheduling strategy, in the embodiments of this application, the model structure of the above-mentioned initial hybrid prediction model with a CNN-LSTM structure can be optimized through a swarm intelligence algorithm, optimizing the number of layers of the initial long short term memory network layer and the number of neurons in each initial long short term memory network layer in the initial hybrid prediction model to generate optimal hyperparameters.

[0061] Furthermore, based on the training sample set and the optimal hyperparameters, the initial hybrid prediction model is pre-trained to generate a hybrid prediction model.

[0062] In some embodiments, the swarm intelligence algorithm is the Grey Wolf Algorithm; based on the swarm intelligence algorithm, the model structure of the initial hybrid prediction model is optimized to generate optimal hyperparameters, including: initializing the parameters of the Grey Wolf Algorithm to generate an initial population; the initial population includes multiple grey wolf individuals; determining the position information of each grey wolf individual; determining the fitness value of each grey wolf individual, and based on the fitness value ranking, selecting multiple alpha wolves from all grey wolf individuals, and saving the position information of all alpha wolves; determining a target ordinary grey wolf individual and performing mutation update on the position information of the target ordinary grey wolf individual; the target ordinary grey wolf individual is randomly selected from all ordinary grey wolf individuals, and the ordinary grey wolf individuals are grey wolf individuals other than alpha wolves; based on the original Grey Wolf Algorithm strategy, updating the position information of the remaining ordinary grey wolf individuals and updating the positions of the initial population to obtain an updated population; the remaining ordinary grey wolf individuals are ordinary grey wolf individuals other than the target ordinary grey wolf individual; based on the updated population, determining whether the iteration termination condition is satisfied; if the iteration termination condition is not satisfied, then return to the step of determining the fitness value of each grey wolf individual until the iteration termination condition is satisfied, completing the optimization of the model structure of the initial hybrid prediction model and generating optimal hyperparameters.

[0063] Please continue to refer to Figure 4 , after constructing the initial hybrid prediction model, use the Grey Wolf Algorithm to optimize the model structure of the initial hybrid prediction model to generate optimal hyperparameters.

[0064] Specifically, initialize the parameters of the Grey Wolf Algorithm, set the population size M, the maximum number of iterations V, the search range (ib to ub), and the number of hyperparameters N of the problem to be optimized of the Grey Wolf Algorithm, so as to generate an initial population, and the initial population includes multiple grey wolf individuals.

[0065] Furthermore, determine the position information of each grey wolf individual.

[0066] It should be noted that the traditional Grey Wolf Algorithm has strong search ability and can effectively solve multi-objective optimization problems. However, the problem of the traditional Grey Wolf Algorithm is that: for high-dimensional problems, the Grey Wolf Algorithm may face the situation of decreasing convergence speed and falling into local optimum. In order to reduce the occurrence of the above problems, the embodiments of the present application have adaptively improved the traditional Grey Wolf Algorithm. Therefore, in the embodiments of the present application, the position information of each grey wolf individual can be determined based on the chaotic mapping function.

[0067] Furthermore, determine the fitness value of each grey wolf individual, and based on the fitness value ranking, select multiple alpha wolves from all grey wolf individuals, and save the position information of all alpha wolves.

[0068] Specifically, calculate the fitness value of each gray wolf individual. According to the fitness value ranking, determine the three gray wolf individuals with the highest fitness values as the leading wolves, and save the position information of the three leading wolves (the position information of the three leading wolves is respectively recorded as ) At this time, the gray wolf individuals in the initial population except the leading wolves are all regarded as ordinary gray wolf individuals to distinguish them from the leading wolves.

[0069] Furthermore, randomly select a target ordinary gray wolf individual from all ordinary gray wolf individuals, and update the position information of the target ordinary gray wolf individual through the following formula: ; ; ; Among them, represents the current iteration number; represents the position information of the target ordinary gray wolf individual after the th iteration; represents the maximum iteration number; represents the global optimal solution at the current iteration number, ; represents a random function; is the natural constant.

[0070] The calculation formula of is as follows: Among them, obeys the normal distribution N(0, ), obeys the standard normal distribution N(0, 1), The calculation formula of is as follows: Among them, is the standard gamma function, .

[0071] Furthermore, based on the original gray wolf algorithm strategy, update the position information of the remaining ordinary gray wolf individuals, and update the positions of the initial population to obtain the updated population.

[0072] Among them, the remaining ordinary gray wolf individuals are the ordinary gray wolf individuals except the target ordinary gray wolf individual.

[0073] Furthermore, calculate the fitness values of all gray wolf individuals in the updated population, and update the position information of the top three leading wolves according to the fitness values.

[0074] Furthermore, based on the updated population, determine whether the iteration termination condition is satisfied.

[0075] Specifically, it is determined whether the maximum number of iterations is reached. If the maximum number of iterations is reached, the iteration termination condition is satisfied; if the maximum number of iterations is not reached, the iteration termination condition is not satisfied.

[0076] If the iteration termination condition is not satisfied, return to the step of determining the fitness value of each gray wolf individual until the iteration termination condition is satisfied, complete the optimization of the model structure of the initial hybrid prediction model, and generate the optimal hyperparameters.

[0077] Furthermore, input the training sample data in the training sample set and the optimal hyperparameters into the pre-constructed initial hybrid prediction model with a CNN-LSTM structure for model pre-training to obtain a trained hybrid prediction model.

[0078] In some embodiments, an initial resource scheduling policy is generated based on the access request data of the cloud platform, including: monitoring the resource access information of each computing node of the cloud platform to obtain access request data; performing conversion processing on the access request data to generate a resource requirement description; and generating an initial resource scheduling policy based on the resource requirement description.

[0079] In some embodiments, based on a pre-trained trend prediction model, the user service demand trend is predicted to obtain demand trend data within a preset future time period, including: obtaining the historical access request data of the cloud platform; and predicting the user service demand trend within a preset future time period based on the trend prediction model according to the historical access request data to obtain demand trend data.

[0080] Before performing cloud platform resource scheduling, the cloud platform resource scheduling method provided by the embodiments of the present application first predicts the user service demand trend within a period of time in the future through the trend prediction model and real-time access request data. Then, when performing resource scheduling for the user's real-time service demand, the current resource scheduling policy can be simulated and predicted according to the service demand trend by using the hybrid prediction model to determine the possible abnormal risk value of the cloud platform after resource scheduling according to the current resource scheduling policy under the predicted demand trend, and the abnormal risk value is fed back to the scheduling policy decision unit as the risk prediction result, so that the scheduling policy decision unit can dynamically adjust the initial resource scheduling policy according to the risk prediction result, thereby enabling flexible scheduling of cloud platform resources based on the service demand trend, improving the balance of cloud platform resource utilization, and reducing the risk of cloud platform performance degradation and performance bottlenecks.

[0081] The embodiments of the present application also provide a cloud platform resource scheduling device. Please refer to Figure 5 , Figure 5It is a schematic structural diagram of a cloud platform resource scheduling device provided by an embodiment of the present application. In the embodiment of the present application, the cloud platform resource scheduling device includes a scheduling policy generation module 510, a trend prediction module 520, a risk prediction module 530, and a scheduling policy adjustment module 540.

[0082] The scheduling policy generation module 510 is configured to generate an initial resource scheduling policy based on the access request data of the cloud platform.

[0083] The trend prediction module 520 is configured to predict the trend of user service requirements based on a pre-trained trend prediction model, and obtain demand trend data within a preset future time period.

[0084] The risk prediction module 530 is configured to input the initial resource scheduling policy and the demand trend data into a pre-trained hybrid prediction model, and obtain a risk prediction result output by the hybrid prediction model.

[0085] The risk prediction result is the abnormal risk value that occurs in the cloud platform after resource scheduling according to the initial resource scheduling policy under the demand trend data.

[0086] The scheduling policy adjustment module 540 is configured to adjust the initial resource scheduling policy based on the risk prediction result to obtain a target resource scheduling policy.

[0087] In some embodiments, the hybrid prediction model includes a convolutional neural network layer and a long short-term memory model connected in sequence; the convolutional neural network layer is configured to extract features from the initial resource scheduling policy and the demand trend data to generate feature information; the long short-term memory model is configured to perform simulation prediction on the initial resource scheduling policy according to the demand trend data based on the feature information to generate a risk prediction result.

[0088] In some embodiments, the risk prediction module 530 is configured to construct a training sample set and an initial hybrid prediction model; the initial hybrid prediction model includes an initial convolutional neural network layer and an initial long short-term memory model connected in sequence, the initial long short-term memory model includes at least one initial long short-term memory network layer, and each initial long short-term memory network layer includes a plurality of neurons; based on a swarm intelligence algorithm, optimize the model structure of the initial hybrid prediction model to generate optimal hyperparameters; the model structure includes the number of layers of the initial long short-term memory network layer and the number of neurons in each initial long short-term memory network layer; based on the training sample set and the optimal hyperparameters, pre-train the initial hybrid prediction model to generate a hybrid prediction model.

[0089] In some embodiments, the swarm intelligence algorithm is a gray wolf algorithm.

[0090] A risk prediction module 530 is configured to initialize the parameters of the grey wolf algorithm, generate an initial population; the initial population includes multiple grey wolf individuals; determine the position information of each grey wolf individual; determine the fitness value of each grey wolf individual, and based on the fitness value ranking, select multiple alpha wolves from all grey wolf individuals, and save the position information of all alpha wolves; determine a target ordinary grey wolf individual, and perform mutation update on the position information of the target ordinary grey wolf individual; the target ordinary grey wolf individual is randomly selected from all ordinary grey wolf individuals, and the ordinary grey wolf individual is a grey wolf individual other than the alpha wolf; based on the original grey wolf algorithm strategy, update the position information of the remaining ordinary grey wolf individuals, and update the position of the initial population to obtain an updated population; the remaining ordinary grey wolf individuals are ordinary grey wolf individuals other than the target ordinary grey wolf individual; based on the updated population, determine whether the iteration termination condition is satisfied; if the iteration termination condition is not satisfied, return to the step of determining the fitness value of each grey wolf individual until the iteration termination condition is satisfied, complete the optimization of the model structure of the initial hybrid prediction model, and generate optimal hyperparameters.

[0091] In some embodiments, a scheduling policy generation module 510 is configured to monitor the resource access information of each computing node of the cloud platform to obtain access request data; perform conversion processing on the access request data to generate a resource requirement description; and generate an initial resource scheduling policy based on the resource requirement description.

[0092] In some embodiments, a trend prediction module 520 is configured to obtain the historical access request data of the cloud platform; and based on a trend prediction model, predict the user service demand trend within a preset future time period according to the historical access request data to obtain demand trend data.

[0093] An embodiment of the present application further provides an electronic device. Figure 6 is a schematic structural diagram of the electronic device provided by the embodiment of the present application, as Figure 6 shown. The electronic device may include: a processor 610, a communication interface 620, a memory 630, and a communication bus 640. Among them, the processor 610, the communication interface 620, and the memory 630 communicate with each other through the communication bus 640. The processor 610 may call the logic instructions in the memory 630 to execute the cloud platform resource scheduling method.

[0094] In addition, when the logical instructions in the above-mentioned memory 630 are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0095] The embodiments of the present application also provide a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the cloud platform resource scheduling method provided by the above-mentioned various methods.

[0096] The embodiments of the present application also provide a computer program product. The computer program product includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the cloud platform resource scheduling method provided by the above-mentioned various methods.

[0097] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated. The components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. A person of ordinary skill in the art can understand and implement it without creative labor.

[0098] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on such an understanding, the technical solution, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disks, optical discs, etc., and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments.

[0099] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A cloud platform resource scheduling method, characterized in that, Including: Generating an initial resource scheduling policy based on access request data of a cloud platform; Predicting the trend of user service requirements based on a pre-trained trend prediction model to obtain demand trend data within a preset future time period; Inputting the initial resource scheduling policy and the demand trend data into a pre-trained hybrid prediction model to obtain a risk prediction result output by the hybrid prediction model; the risk prediction result is an abnormal risk value that occurs in the cloud platform after resource scheduling according to the initial resource scheduling policy under the demand trend data; Adjusting the initial resource scheduling policy based on the risk prediction result to obtain a target resource scheduling policy.

2. The cloud platform resource scheduling method according to claim 1, wherein The hybrid prediction model includes a convolutional neural network layer and a long short-term memory model connected in sequence; The convolutional neural network layer is used to extract features from the initial resource scheduling policy and the demand trend data to generate feature information; The long short-term memory model is used to simulate and predict the initial resource scheduling policy according to the demand trend data based on the feature information to generate the risk prediction result.

3. The cloud platform resource scheduling method according to claim 1, wherein, Before inputting the initial resource scheduling policy and the demand trend data into the pre-trained hybrid prediction model to obtain the risk prediction result output by the hybrid prediction model, it further includes: Constructing a training sample set and an initial hybrid prediction model; the initial hybrid prediction model includes an initial convolutional neural network layer and an initial long short-term memory model connected in sequence, the initial long short-term memory model includes at least one initial long short-term memory network layer, and each initial long short-term memory network layer includes multiple neurons; Optimizing the model structure of the initial hybrid prediction model based on a swarm intelligence algorithm to generate optimal hyperparameters; the model structure includes the number of layers of the initial long short-term memory network layer and the number of neurons in each initial long short-term memory network layer; Pre-training the initial hybrid prediction model based on the training sample set and the optimal hyperparameters to generate the hybrid prediction model.

4. The cloud platform resource scheduling method according to claim 3, wherein, The swarm intelligence algorithm is a grey wolf algorithm; The optimizing the model structure of the initial hybrid prediction model based on a swarm intelligence algorithm to generate optimal hyperparameters includes: Initializing the parameters of the grey wolf algorithm to generate an initial population; the initial population includes multiple grey wolf individuals; Determining the position information of each grey wolf individual; Determining the fitness value of each grey wolf individual, and selecting multiple alpha wolves from all grey wolf individuals based on the fitness value ranking, and saving the position information of all alpha wolves; Determining a target ordinary grey wolf individual and mutating and updating the position information of the target ordinary grey wolf individual; the target ordinary grey wolf individual is randomly selected from all ordinary grey wolf individuals, and the ordinary grey wolf individual is a grey wolf individual other than the alpha wolf; Updating the position information of the remaining ordinary grey wolf individuals based on the original grey wolf algorithm strategy, and updating the position of the initial population to obtain an updated population; the remaining ordinary grey wolf individuals are ordinary grey wolf individuals other than the target ordinary grey wolf individual; Based on the updated population, determine whether the iteration termination condition is satisfied; If the iteration termination condition is not satisfied, return to the step of determining the fitness value of each gray wolf individual until the iteration termination condition is satisfied, complete the optimization of the model structure of the initial hybrid prediction model, and generate the optimal hyperparameters.

5. The cloud platform resource scheduling method according to claim 1, wherein Generating an initial resource scheduling policy based on the access request data of the cloud platform includes: Monitoring the resource access information of each computing node of the cloud platform to obtain the access request data; Performing transformation processing on the access request data to generate a resource requirement description; Generating the initial resource scheduling policy based on the resource requirement description.

6. The cloud platform resource scheduling method according to claim 1, wherein Predicting the user service demand trend based on the pre-trained trend prediction model to obtain the demand trend data within a preset future time period, including: Obtaining the historical access request data of the cloud platform; Based on the trend prediction model, predicting the user service demand trend within the preset future time period according to the historical access request data to obtain the demand trend data.

7. A cloud platform resource scheduling device, characterized in that, Including: A scheduling policy generation module for generating an initial resource scheduling policy based on the access request data of the cloud platform; A trend prediction module for predicting the user service demand trend based on the pre-trained trend prediction model to obtain the demand trend data within a preset future time period; A risk prediction module for inputting the initial resource scheduling policy and the demand trend data into the pre-trained hybrid prediction model to obtain the risk prediction result output by the hybrid prediction model; the risk prediction result is the abnormal risk value that appears on the cloud platform after resource scheduling according to the initial resource scheduling policy under the demand trend data; A scheduling policy adjustment module for adjusting the initial resource scheduling policy based on the risk prediction result to obtain the target resource scheduling policy.

8. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the cloud platform resource scheduling method according to any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the cloud platform resource scheduling method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the cloud platform resource scheduling method according to any one of claims 1 to 6.

Images