Cloud resource load prediction method and device, storage medium and electronic equipment

Abstract

The invention discloses a cloud resource load prediction method and device, a storage medium and electronic equipment, and relates to the technical field of computers, and the method comprises the steps: firstly obtaining to-be-predicted time series data of cloud resources; to-be-predicted time sequence data is input into the ARIMA model, the ARIMA model is used for load prediction of cloud resources, and a model coefficient of the ARIMA model is a load prediction result obtained by carrying out load prediction according to historical time sequence data of the cloud resources and a load prediction error between historical load information of the cloud resources; the ARIMA model is obtained by iteratively updating an initial model coefficient of the ARIMA model; and generating predicted load information of the to-be-predicted time series data according to an output result of the ARIMA model. Compared with the prior art, the prediction precision can be improved.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a cloud resource load prediction method, device, storage medium and electronic device. Background Art

[0002] On the big cloud operation and management platform, a large number of cloud service resources need to be monitored and managed, and user resources need to be expanded and scheduled. Cloud resource load can be periodically predicted through historical resource indicator data, and resource warnings can be issued based on the prediction results.

[0003] At present, cloud resource prediction is mainly carried out by building prediction models through neural networks. For example, a prediction model based on a long short-term memory (LSTM) network is used to normalize cloud resource data, reduce data dimensionality, select feature data, and obtain the prediction result of cloud resource load.

[0004] However, when the cloud resource load changes are nonlinear, the prediction model built using a neural network will fall into a local optimum, causing the prediction model to converge slowly, which in turn increases the model training cost and even affects the prediction accuracy and precision. Summary of the invention

[0005] In view of this, the present application provides a cloud resource load prediction method, device, storage medium and electronic device, the main purpose of which is to improve the current existing technology that when the cloud resource load changes nonlinearly, a prediction model constructed by a neural network is used for prediction, and the prediction model may fall into a local optimum, resulting in a slow convergence of the prediction model, which in turn leads to an increase in the model training cost and even affects the prediction accuracy and prediction precision.

[0006] In a first aspect, the present application provides a cloud resource load prediction method, comprising:

[0007] Obtain the time series data to be predicted of cloud resources;

[0008] The time series data to be predicted is input into an autoregressive moving average (ARIMA) model, the ARIMA model is used for load prediction of the cloud resources, and the model coefficients of the ARIMA model are obtained by iteratively updating the initial model coefficients of the ARIMA model based on the load prediction result of the load prediction based on the historical time series data of the cloud resources and the load prediction error between the historical load information of the cloud resources;

[0009] Based on the output result of the ARIMA model, the forecast load information of the time series data to be forecasted is generated.

[0010] In a second aspect, the present application provides a cloud resource load prediction device, comprising:

[0011] An acquisition module, configured to acquire time series data to be predicted of cloud resources;

[0012] An input module is configured to input the time series data to be predicted into an ARIMA model, wherein the ARIMA model is used for load prediction of the cloud resources, and the model coefficients of the ARIMA model are obtained by iteratively updating the initial model coefficients of the ARIMA model based on the load prediction result of load prediction based on the historical time series data of the cloud resources and the load prediction error between the historical load information of the cloud resources;

[0013] The generating module is configured to generate the forecast load information of the time series data to be forecasted according to the output result of the ARIMA model.

[0014] In a third aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the cloud resource load prediction method of the first aspect.

[0015] In a fourth aspect, the present application provides an electronic device, comprising a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, wherein the processor implements the cloud resource load prediction method of the first aspect when executing the computer program.

[0016] In a fifth aspect, the present application provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, the cloud resource load prediction method of the first aspect is implemented.

[0017] By means of the above technical scheme, the present application provides a cloud resource load prediction method, device, storage medium and electronic device, which first obtain the time series data to be predicted of the cloud resources; input the time series data to be predicted into the ARIMA model, the ARIMA model is used for load prediction of the cloud resources, and the model coefficients of the ARIMA model are obtained by iteratively updating the initial model coefficients of the ARIMA model based on the load prediction result of the load prediction of the historical time series data of the cloud resources and the load prediction error between the historical load information of the cloud resources; based on the output result of the ARIMA model, generate the predicted load information of the time series data to be predicted. Compared with the current existing technology, the present application uses the ARIMA model to predict the cloud resource load for the time series data to be predicted. The model coefficients of the ARIMA model used in the present application are obtained by iteratively updating the initial model coefficients based on the load prediction error, so that the ARIMA model has higher prediction accuracy and better adaptability and lower training cost, and can also avoid falling into local optimality and slow convergence. Since the ARIMA model can adaptively adjust the model parameters, the use of the ARIMA model containing the model coefficients for cloud resource load prediction can accurately predict the cloud resource load rate, improve the prediction accuracy, help to formulate effective resource allocation strategies and energy saving, and can also effectively improve the stability and reliability of cloud services.

[0018] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0020] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0021] Figure 1 A schematic diagram of a process of predicting cloud resource load provided by an embodiment of the present application is shown;

[0022] Figure 2 A schematic diagram of a process of predicting cloud resource load provided by an embodiment of the present application is shown;

[0023] Figure 3 A schematic diagram showing an example provided by an embodiment of the present application;

[0024] Figure 4 A schematic diagram showing an example provided by an embodiment of the present application;

[0025] Figure 5 A schematic diagram of the structure of a cloud resource load prediction device provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0026] The embodiments of the present application will be described in more detail below with reference to the accompanying drawings. It should be noted that the embodiments and features in the embodiments of the present application can be combined with each other without conflict.

[0027] Existing technical solutions mostly use statistical algorithms to predict resource loads, and the algorithm models include linear models, ARMA models, exponential smoothing models, etc. For the prediction model of cloud resource time series, including the Autoregressive Integrated Moving Average Model (ARIMA), etc., the prediction model is used to predict the resource load and schedule in advance. It is also possible to build a prediction model through multiple neural networks, such as the long short-term memory (LSTM) network, by processing and normalizing the data, reducing the data dimension, selecting feature data, and then training the model according to the model evaluation criteria to obtain a training model.

[0028] At present, the existing technologies cannot guarantee the accuracy of short-term and long-term predictions of traditional time series prediction models such as ARMA, ARIMA, and FARIMA for chaotic time series data. For models such as artificial neural network prediction to cope with the prediction of nonlinear cloud resource load changes, the algorithm model is prone to fall into local optimality and slow convergence, resulting in high model training costs and unstable prediction accuracy.

[0029] In order to improve the current existing technology, when the cloud resource load changes nonlinearly, the prediction model built by the neural network will fall into the local optimum, resulting in slow convergence of the prediction model, which in turn increases the model training cost and even affects the prediction accuracy and prediction precision. This embodiment provides a cloud resource load prediction method, such as Figure 1 As shown, the method includes:

[0030] Step 101: Obtain time series data to be predicted of cloud resources.

[0031] In the embodiments of the present application, cloud resources, i.e., cloud computing resources, refer to various computing resources and services that are provided through the Internet and can be accessed remotely. These resources allow users to obtain, configure, and manage computing power, storage space, network resources, application services, etc. on demand without directly managing the underlying hardware facilities. Users can choose appropriate cloud services according to their business needs to achieve rapid deployment, expansion, and optimization of their business. The core advantages of cloud resources are elastic scaling, pay-as-you-go, high availability, and easy management.

[0032] In some examples, the time series data to be predicted may be time series data that currently needs to be predicted.

[0033] Step 102: Input the time series data to be predicted into the ARIMA model.

[0034] Among them, the ARIMA model is used for load prediction of cloud resources. The model coefficients of the ARIMA model are obtained by iteratively updating the initial model coefficients of the ARIMA model based on the load prediction results of load prediction based on the historical time series data of cloud resources and the load prediction error between the historical load information of cloud resources.

[0035] For this embodiment, the Autoregressive Integrated Moving Average Model (ARIMA) is a statistical model for time series prediction and analysis, and is particularly suitable for non-stationary time series. The ARIMA model combines three components: autoregressive model (AR), difference (I) and moving average model (MA). In the embodiment of the present application, the initial model coefficients are the initial autoregressive coefficients, initial difference coefficients and initial moving average coefficients.

[0036] Correspondingly, the historical time series data may be resource usage of a periodic resource pool or a host, and the historical load information may be the historical load of cloud resources.

[0037] In some examples, the load prediction error may be a difference between the load prediction result and the historical load information, and correspondingly, the model coefficient may be a model coefficient corresponding to the model satisfying a preset convergence condition.

[0038] Step 103: Generate forecast load information of the time series data to be forecasted based on the output result of the ARIMA model.

[0039] Optionally, using an ARIMA model that includes model coefficients means using a trained ARIMA model for prediction, which can improve the accuracy of the prediction.

[0040] Compared with the current existing technology, this embodiment uses the ARIMA model to predict the cloud resource load for the time series data to be predicted. The model coefficients of the ARIMA model used in this application are obtained by iteratively updating the initial model coefficients based on the load prediction error, so that the ARIMA model has higher prediction accuracy and better adaptability and lower training cost, and can also avoid falling into local optimality and slow convergence. Since the ARIMA model can adaptively adjust the model parameters, using the ARIMA model containing the model coefficients for cloud resource load prediction can accurately predict the cloud resource load rate, improve the prediction accuracy, help to formulate effective resource allocation strategies and energy saving, and can also effectively improve the stability and reliability of cloud services.

[0041] In order to further illustrate the specific implementation process of the method of this embodiment, this embodiment provides the following Figure 2 The specific method shown includes:

[0042] Step 201: Obtain time series data to be predicted of cloud resources.

[0043] Optional, such as Figure 3 As shown in the figure, cloud resource load prediction can be completed by the data acquisition module, cloud resource prediction module, and resource management module. Among them, the data acquisition module regularly queries resource load data through the monitoring software on the underlying server and uploads it to the platform server; the cloud resource prediction module proposes an adaptive historical data weight method for multi-variable cloud resource central processing unit (CPU), memory, and disk load index prediction; the resource management module intelligently optimizes the resource allocation of cloud resources through the index prediction results and resource management strategies.

[0044] Step 202: Input the time series data to be predicted into the ARIMA model.

[0045] Among them, the ARIMA model is used for load prediction of cloud resources. The model coefficients of the ARIMA model are obtained by iteratively updating the initial model coefficients of the ARIMA model based on the load prediction results of load prediction based on the historical time series data of cloud resources and the load prediction error between the historical load information of cloud resources.

[0046] In an embodiment of the present application, the training process of the ARIMA model includes: obtaining historical time series data and historical load information; preprocessing the historical time series data, determining the initial model coefficients of the ARIMA model, and establishing the ARIMA model according to the initial model coefficients; performing load forecasting on the historical time series data based on the ARIMA model, and iteratively updating the initial model coefficients according to the load forecast error between the load forecast result and the historical load information to obtain the model coefficients that meet the preset convergence conditions.

[0047] Optionally, the training process of the ARIMA model also includes: performing differential processing on the historical time series data, and checking the randomness and stationarity of the historical time series data after the differential processing; classifying the historical time series data after the differential processing according to preset categories based on the verification results to obtain multiple historical interval data; determining the initial model coefficients of the ARIMA model corresponding to the multiple historical interval data, and establishing multiple ARIMA models according to the initial model coefficients.

[0048] In some examples, differencing time series data is an important step in time series analysis, mainly used to eliminate trends or seasonality in the data and make the series more stable, which is convenient for further analysis and modeling. The purpose of differencing is to meet the stationarity assumption in time series analysis and make subsequent statistical analysis more effective. The choice of several orders of differencing and whether to perform seasonal differencing is usually determined by analyzing the autocorrelation diagram and partial autocorrelation diagram of the data and performing unit root tests (such as ADF tests).

[0049] For this embodiment, in time series analysis, randomness and stationarity checks are very critical steps, which are essential to ensure the effectiveness of the model and the accuracy of prediction. Specifically, the randomness test is intended to verify whether the residuals in the time series (the difference between the model predicted value and the actual value) are randomly distributed, that is, there is no statistical correlation between the residuals. This shows that the model has fully captured the structural information in the data, and the remaining part is unpredictable random noise. The stationarity test is used to determine whether the time series is stationary, that is, whether the statistical characteristics of the sequence (such as mean, variance, covariance) remain unchanged over time. In time series analysis, the sequence is usually required to be stationary to ensure the stability of the model parameters. Through these tests, it can be determined whether the time series meets the prerequisites for analysis and modeling, and then a suitable model can be selected for further analysis and prediction.

[0050] As an optional method, the preset categories may include three types: linear long-term change type, cyclic fluctuation type, and irregular change type. Accordingly, each preset category contains historical interval data, and each preset category needs to establish a corresponding ARIMA model.

[0051] For example, analyze the changing trend within the historical data interval to obtain the corresponding periodic data The method based on interval and frequency variation will The data in the interval can be divided into three different types. The data in the interval may not have a trend of change. Through Fourier transform, the time domain signal can be decomposed into components of different frequencies to observe its spectrum characteristics. According to the spectrum characteristics, the transformation type of the time series data can be preliminarily determined. Cyclic fluctuation: There are peaks or valleys in the frequency, which may be biased towards cyclic fluctuations; nonlinear transformation: The frequency components change nonlinearly as the frequency increases or decreases.

[0052] Further, such as Figure 4 As shown, using each interval type To train separate ARIMA models. Each prediction model acts on data with a unique frequency range. Using separate prediction models for data with large random and periodic fluctuations can greatly improve the accuracy of predictions. For linear and nonlinear time series data, the p, d, and q parameters of the model must be different. For example, d is to make the data change from nonlinear to linear. Therefore, three models are created for three different types of data and predicted separately.

[0053] It should be noted that interval data is a period of historical time series data, which is predicted using the previous ARIMA optimization algorithm model using past historical data.

[0054] In some examples, the training process of the ARIMA model also includes: for the target historical interval data among multiple historical interval data, based on the target preset category corresponding to the target historical interval data, generating the target initial differential coefficient of the target ARIMA model corresponding to the target historical interval data; determining the target initial autoregressive coefficient and the target initial moving average coefficient of the target ARIMA model based on the verification result combined with the preset autocorrelation function and the preset partial autocorrelation function; and establishing the target ARIMA model according to the target initial differential coefficient, the target initial autoregressive coefficient and the target initial moving average coefficient.

[0055] In the ARIMA model, the autoregressive (AR) coefficient represents the influence of the observed values ​​at the past p time points on the current value in the model, and correspondingly, the moving average (MA) coefficient represents the influence of the past q error terms on the current error term in the model.

[0056] Exemplarily, determining p, d, and q can determine the ARIMA model, but since the time series data of resources varies greatly; therefore, the parameters of the model p and q are adaptively adjusted to improve the prediction accuracy. Specifically, the data acquisition module regularly uploads information such as the periodic resource pool, the resource usage of the host, and the overall load of the resource pool. The time series data is first preprocessed: by differentiating the time series data, the difference order d of the model is determined through stationarity and randomness tests. The order p of the autoregressive part and the order q of the moving average part are determined according to the autocorrelation function and the partial autocorrelation function. The d-order difference can convert nonlinear time series data into linearized data. p represents the order of autoregression (AR), q represents the order of moving average (MA), and d represents the order of difference. p, d, and q are the coefficients of the autoregressive term, the difference term, and the moving average term, respectively. The initial coefficients can determine the p and q coefficients through the autocorrelation function and the partial autocorrelation function.

[0057] ARIMA (p, d, q) is a systematic method for predictive modeling of stationary time series and non-stationary time series data. For non-stationary time series data, its mean and variance are unstable. Generally, it is first converted into a stationary time series through differential operations, and then the ARIMA model is established using the stationary time series data. The ARIMA model expression is shown in the following formula 1:

[0058]

[0059] In formula 1, x t To predict the data at time t, x t-1 is the historical data of the past t-1 period, φ0 is a constant, which is a benchmark value. Autoregression means that the past observations have an impact on the current observations. φ1...φ p The impact of each historical moment data on the current forecast: AR coefficient, ∈1…∈ t-p is the random error of noise in historical data. For nonlinear data, there is a prediction error. θ1...θ q Is the error coefficient: MA coefficient.

[0060] Optionally, the training process of the ARIMA model also includes: for the current load forecasting process, obtaining the regression weight corresponding to the target initial autoregressive coefficient and the moving average weight corresponding to the target initial moving average coefficient obtained in the previous load forecasting; and performing the current load forecast on the target historical interval data through the target ARIMA model based on the regression weight and the moving average weight.

[0061] In the embodiment of the present application, in each iteration of all training data, the prediction error is determined based on the difference between the actual value and the model prediction value. If the prediction error is obvious, it means that the model fails to better capture the data characteristics. Error feedback continuously adjusts the coefficients of AR and MA until a set of "optimal" AR and MA coefficient weights are found when the error converges. The expression of the relationship between the predicted time series and the weight is shown in the following formula 2:

[0062]

[0063] In formula 2, is the forecast data of the time series at time t+1, w i is the weight for the t-i+1 period. The useful quantity of the time series of the first n forecasts is used, and the weight coefficient is continuously adjusted according to the forecast error in each iteration.

[0064] It should be noted that Formula 2 only reflects the influence of past data and weights on the current prediction results. At the same time, the error between the prediction results and the actual results is compensated and the coefficients of AR and MA are adjusted.

[0065] In some examples, the training process of the ARIMA model also includes: determining the current load prediction error between the current load prediction result and the target historical load information corresponding to the target historical interval data; adjusting the regression weight and the moving average weight according to the relationship between the first load prediction error and the preset error reference value; if it is determined that the adjusted regression weight and moving average weight do not meet the preset convergence conditions, then performing the next load prediction for the target historical interval data through the target ARIMA model based on the adjusted regression weight and moving average weight.

[0066] Optionally, the preset error reference value can be set according to actual needs, and the specific numerical value of the error reference value is not specifically limited here.

[0067] Exemplarily, the forecast error of the observation model is used for error tracking, and the coefficients of the AR and MA parts are adjusted according to the direction and size of the error through instant compensation. The parameters of the ARIMA model are optimized by the gradient descent method, and the error reference value is set. If the forecast error of the model exceeds the error reference value, the coefficients of the AR and MA parts are increased to increase the flexibility and adaptability of the model. On the contrary, if the forecast error of the model is less than the error reference value, the coefficients of the AR and MA parts can be reduced to reduce the complexity of the model and the possibility of overfitting. The error can be obtained by comparing the predicted value with the actual time series.

[0068] Accordingly, the gradient of the AR and MA coefficients is calculated by the gradient descent method (steepest descent method) shown in the following formula 3, and then the parameters are updated in the opposite direction of the gradient. Formula 3 is as follows:

[0069]

[0070] In formula 3, w new Represents the weight of the model update, w old is the true value of the moment

[0071] W represents the AR and MA coefficients, k represents the learning rate, The gradient of the objective function with respect to the parameters.

[0072] The objective function is the coefficient of the AR part of the following gradient function. The gradient can be determined by formula 4, which is as follows:

[0073]

[0074] The coefficient and gradient of the MA part can be determined by formula 5, which is as follows:

[0075]

[0076] In Formula 4 and Formula 5, N represents the number of samples, y t represents the observed value, y rt Represents the predicted value of the model.

[0077] For this embodiment, the training process of the ARIMA model also includes: if it is determined that the adjusted regression weights and moving average weights meet the preset convergence conditions, the adjusted regression weights and moving average weights are determined as the target regression weights and target moving average weights of the target ARIMA model; and the target model coefficients of the target ARIMA model are determined based on the target regression weights and target moving average weights.

[0078] Exemplarily, the AR and MA coefficient weights are adjusted until a set of "optimal" weights is found to minimize the error. Assuming that the total training time series data is M, such weights are adjusted repeatedly (MN) times in one round of iteration. Then an initial learning rate is set, and the learning rate is decayed according to certain rules in each iteration, and the last set of weights obtained in this iteration is set as the initial weights for the next iteration. When the error convergence condition a is met or the maximum number of iterations is exceeded, the iteration stops, and then the "optimal" AR and MA coefficient weights are obtained. The gradient descent method is used with the minimum standard deviation of the prediction as the objective function, and the convergence threshold a is shown in Formula 6. When the threshold requirements are met, the iteration stops and the new AR and MA coefficients are used. Formula 6 is specifically shown as follows:

[0079]

[0080] Step 203: Perform load prediction on the time series data to be predicted based on the ARIMA model to obtain the CPU prediction value, memory prediction value, and disk prediction value of the cloud resources.

[0081] Optionally, step 203 may specifically include: determining category information corresponding to the time series data to be predicted in preset categories, and selecting an ARIMA model corresponding to the category information; performing load prediction on the time series data to be predicted by using the ARIMA model corresponding to the time series data to be predicted, and obtaining a CPU prediction value, a memory prediction value, and a disk prediction value of the cloud resources.

[0082] Among them, the model coefficients included in the ARIMA model corresponding to the category information are model coefficients.

[0083] Step 204: Generate predicted load information of the time series data to be predicted based on the CPU prediction value, the memory prediction value and the disk prediction value.

[0084] For example, in the embodiment of the present application, resources include memory, CPU, and disk. The percentage of future cpu, mem, and disk loads predicted by the improved ARIMA model sequence is unified into a "load" indicator Cload as the predicted load information of the cloud resource load. load As shown in the following formula 7:

[0085]

[0086] It should be noted that the cloud resource prediction in the embodiment of the present application evaluates the future workload by predicting the load information, so as to detect overload at an early stage and solve the possible system performance degradation, resulting in the crash of cloud service and other problems. load When it is close to 1, it means that the VM load is low. In this state, the performance of the VM is optimal. However, the resource utilization is too low. This will lead to a serious waste of resources. load Between 8 and 50, it means that the virtual machine is in a normal state or the virtual machine load is within a reasonable range. In this state, resources are used efficiently. load A high or full disk utilization indicates that the resource is overutilized and exceeds the capacity of the virtual machine.

[0087] Furthermore, the prediction result of the prediction module (prediction load information C load , CPU prediction value, memory prediction value and disk prediction value), and report to the resource configuration management module. The system tools can be used to schedule and manage different resource pools and adjust the actual resource allocation according to the load. This makes the cloud resource service stable and reduces ineffective energy consumption.

[0088] For the embodiment of the present application, on China Mobile's large cloud operation and management platform, it is necessary to monitor and manage a large number of cloud service resources, and at the same time, it is necessary to manage the user's resources by expansion scheduling and other management. Periodic predictions can be made through historical resource indicator data, and resource warnings can be made on this basis, as well as helping users to optimize the fine-grained allocation of resources, so that the cloud computing resources of the platform can be fully utilized. With the development and rise of cloud computing, the application and configuration of service resources have become more flexible and convenient, but the configuration and management of resources has become increasingly complex. With the continuous expansion of the scale of many enterprises and the continuous growth of business volume, the problems of high cost of cloud infrastructure services and low resource utilization have become more and more prominent, and all computing resources can be fully utilized. Resource management and scheduling is a challenging problem, and cloud data centers are composed of hundreds or thousands of resources. Most methods for accurate allocation, configuration and scheduling of cloud resources require prediction. Due to its dynamic characteristics, the effective prediction of cloud resource load remains an important issue.

[0089] The existing technologies mostly use statistical algorithms to predict resource loads, and the algorithm models include linear models, ARMA models, exponential smoothing models, etc. For the prediction models of cloud resource time series, including ARIMA models, etc., the prediction models are used to predict the load of resources and schedule them in advance. The prediction models are also constructed through neural networks, such as long short-term memory (LSTM) networks, by processing and normalizing the data, reducing the data dimension, selecting feature data, and then training the model according to the model evaluation criteria to obtain the training model. The existing technologies cannot guarantee the accuracy of short-term and long-term predictions of traditional time series prediction models such as ARMA, ARIMA, and FARIMA for chaotic time series data. For the prediction of nonlinear cloud resource load changes by artificial neural network prediction models, the algorithm model is prone to fall into local optimality and slow convergence, resulting in high model training costs and unstable prediction accuracy.

[0090] It should be noted that the real-time monitoring function of the cloud resource load in the embodiment of the present application has been widely promoted. As for the prediction of cloud load, it is possible to issue an early warning for high load of service resources, reasonably allocate cloud resources, and solve problems such as high cost of cloud infrastructure services and low resource utilization, which has a large application scenario. In the cloud product, configure monitoring item management in the console monitoring and choose to create server monitoring items. In the server monitoring items, you can select CPU, memory, and disk indicators. Check the alarm status and monitoring status view in the monitoring. The status of all collected monitoring items and devices in the physical space can be displayed. If you need to view the monitoring status of a specified device or a specified monitoring item, you can enter the corresponding search condition in the conditional search bar at the top of the page and click the query button to filter. It can also be combined with other machine learning methods, such as random forests, support vector machines, etc., to achieve cloud resource load prediction.

[0091] Compared with the current existing technology, this embodiment determines the initial model coefficients of the ARIMA model, and then iteratively updates the initial model coefficients based on the load prediction error to obtain model coefficients that meet the preset convergence conditions, so that the ARIMA model can adaptively adjust the model parameters, accurately predict the cloud resource load rate, improve the prediction accuracy, and effectively improve the stability and reliability of cloud services, which is helpful to formulate effective resource allocation strategies and energy saving, so that the ARIMA model has higher prediction accuracy and better adaptability. Different prediction models can be constructed by classifying historical time series data according to preset categories, avoiding the problem of inaccurate prediction due to large data fluctuations, while reducing the training time cost and improving prediction stability.

[0092] Further, as Figure 1 and Figure 2 The specific implementation of the method shown in the embodiment provides a cloud resource load prediction device, such as Figure 5 As shown, the device includes: an acquisition module 31, an input module 32, and a generation module 33.

[0093] An acquisition module 31 is configured to acquire the time series data to be predicted of the cloud resources;

[0094] An input module is configured to input the time series data to be predicted into an ARIMA model, wherein the ARIMA model is used for load prediction of the cloud resources, and the model coefficients of the ARIMA model are obtained by iteratively updating the initial model coefficients of the ARIMA model based on the load prediction result of load prediction based on the historical time series data of the cloud resources and the load prediction error between the historical load information of the cloud resources;

[0095] The generating module is configured to generate the forecast load information of the time series data to be forecasted according to the output result of the ARIMA model.

[0096] In some examples of this embodiment, the training process of the ARIMA model includes: acquiring the historical time series data and the historical load information; preprocessing the historical time series data to determine the initial model coefficients of the ARIMA model, and establishing the ARIMA model according to the initial model coefficients; performing load forecasting on the historical time series data based on the ARIMA model, and iteratively updating the initial model coefficients according to the load forecast error between the load forecast result and the historical load information to obtain model coefficients that meet preset convergence conditions.

[0097] In some examples of this embodiment, the training process of the ARIMA model also includes: performing differential processing on the historical time series data, and checking the randomness and stationarity of the historical time series data after the differential processing; classifying the historical time series data after the differential processing according to preset categories based on the verification results to obtain multiple historical interval data; determining the initial model coefficients of the ARIMA models corresponding to the multiple historical interval data, and establishing multiple ARIMA models according to the initial model coefficients.

[0098] In some examples of this embodiment, the training process of the ARIMA model also includes: performing differential processing on the historical time series data, and checking the randomness and stationarity of the historical time series data after the differential processing; classifying the historical time series data after the differential processing according to preset categories based on the verification results to obtain multiple historical interval data; determining the initial model coefficients of the ARIMA models corresponding to the multiple historical interval data, and establishing multiple ARIMA models according to the initial model coefficients.

[0099] In some examples of this embodiment, the training process of the ARIMA model also includes: for the current load forecasting process, obtaining the regression weight corresponding to the target initial autoregressive coefficient and the moving average weight corresponding to the target initial moving average coefficient obtained in the previous load forecasting; and performing the current load forecast on the target historical interval data through the target ARIMA model based on the regression weight and the moving average weight.

[0100] In some examples of this embodiment, the training process of the ARIMA model also includes: determining a current load prediction error between a current load prediction result and the target historical load information corresponding to the target historical interval data; adjusting the regression weight and the moving average weight according to a size relationship between the first load prediction error and a preset error reference value; if it is determined that the adjusted regression weight and moving average weight do not meet the preset convergence conditions, performing the next load prediction for the target historical interval data through the target ARIMA model based on the adjusted regression weight and moving average weight.

[0101] In some examples of this embodiment, the training process of the ARIMA model also includes: if it is determined that the adjusted regression weights and moving average weights meet the preset convergence conditions, the adjusted regression weights and moving average weights are determined as the target regression weights and target moving average weights of the target ARIMA model; and the target model coefficients of the target ARIMA model are determined based on the target regression weights and the target moving average weights.

[0102] In some examples of this embodiment, the generation module 33 is specifically configured to perform load prediction on the time series data to be predicted based on the ARIMA model to obtain the CPU prediction value, memory prediction value and disk prediction value of the cloud resources; based on the CPU prediction value, the memory prediction value and the disk prediction value, generate predicted load information of the time series data to be predicted.

[0103] In some examples of the present embodiment, the generation module 33 is further specifically configured to determine the category information corresponding to the time series data to be predicted in the preset category, select the ARIMA model corresponding to the category information, wherein the model coefficients contained in the ARIMA model corresponding to the category information are the model coefficients; load prediction is performed on the time series data to be predicted through the ARIMA model corresponding to the time series data to be predicted, and the central processing unit prediction value, memory prediction value and disk prediction value of the cloud resources are obtained.

[0104] It should be noted that for other corresponding descriptions of the functional units involved in the cloud resource load prediction device provided in this embodiment, please refer to Figure 1 and Figure 2 The corresponding description in will not be repeated here.

[0105] Based on the above Figure 1 to Figure 2 The method shown in the embodiment also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the above-mentioned Figure 1 to Figure 2 The method shown.

[0106] Based on this understanding, the technical solution of the present application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, USB flash drive, mobile hard disk, etc.), including a number of instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods of various implementation scenarios of the present application.

[0107] Based on the above Figure 1 to Figure 2 The method shown, and Figure 5 In order to achieve the above-mentioned purpose, the embodiment of the present application also provides an electronic device, such as a personal computer, a server, a laptop computer, a smart phone, an intelligent robot or other intelligent terminal, which includes a storage medium and a processor; the storage medium is used to store a computer program; the processor is used to execute the computer program to achieve the above-mentioned Figure 1 to Figure 2 The method shown.

[0108] Optionally, the above-mentioned physical device may also include a user interface, a network interface, a camera, a radio frequency (RF) circuit, a sensor, an audio circuit, a WI-FI module, etc. The user interface may include a display, an input unit such as a keyboard, etc., and the optional user interface may also include a USB interface, a card reader interface, etc. The network interface may optionally include a standard wired interface, a wireless interface (such as a WI-FI interface), etc.

[0109] Those skilled in the art will appreciate that the above-mentioned physical device structure provided in this embodiment does not constitute a limitation on the physical device, and may include more or fewer components, or a combination of certain components, or different arrangements of components.

[0110] The storage medium may also include an operating system and a network communication module. The operating system is a program that manages the hardware and software resources of the above-mentioned physical device, and supports the operation of the information processing program and other software and / or programs. The network communication module is used to realize the communication between the components inside the storage medium, and the communication with other hardware and software in the information processing physical device.

[0111] Based on the above Figure 1 to Figure 2 The present application also provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the above-mentioned Figure 1 to Figure 2 The method shown, the method implemented when the computer program is executed by the processor can refer to the various embodiments of the present application, and will not be described in detail here.

[0112] Through the description of the above implementation methods, the technical personnel in this field can clearly understand that the present application can be implemented by means of software plus the necessary general hardware platform, or by hardware. Compared with the current prior art, this embodiment determines the initial model coefficients of the ARIMA model, and then iteratively updates the initial model coefficients based on the load prediction error to obtain the model coefficients that meet the preset convergence conditions, so that the ARIMA model can adaptively adjust the model parameters, accurately predict the cloud resource load rate, improve the prediction accuracy, and effectively improve the stability and reliability of cloud services, which is helpful to formulate effective resource allocation strategies and energy saving, so that the ARIMA model has higher prediction accuracy and better adaptability. Different prediction models can be constructed by classifying historical time series data according to preset categories to avoid the problem of prediction inaccuracy caused by large fluctuations in data, while reducing the training time cost and improving prediction stability.

[0113] It should be noted that, in this article, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article 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, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0114] The above is only a specific implementation of the present application, so that those skilled in the art can understand or implement the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments described herein, but will conform to the widest scope consistent with the principles and novel features applied for herein.

Claims

1. A cloud resource load prediction method, characterized in that: include: Obtain the time series data to be predicted of cloud resources; The time series data to be predicted is input into a differential autoregressive moving average (ARIMA) model, the ARIMA model is used for load prediction of the cloud resources, and the model coefficients of the ARIMA model are obtained by iteratively updating the initial model coefficients of the ARIMA model based on the load prediction result of the load prediction based on the historical time series data of the cloud resources and the load prediction error between the historical load information of the cloud resources; Based on the output result of the ARIMA model, the forecast load information of the time series data to be forecasted is generated.

2. The method according to claim 1, characterized in that The training process of the ARIMA model includes: Acquiring the historical time series data and the historical load information; Preprocessing the historical time series data, determining the initial model coefficients, and establishing the ARIMA model according to the initial model coefficients; The load is predicted for the historical time series data based on the ARIMA model, and the initial model coefficients are iteratively updated according to the load prediction error between the load prediction result and the historical load information to obtain the model coefficients that meet the preset convergence conditions.

3. The method according to claim 2, characterized in that The preprocessing of the historical time series data, determining the initial model coefficients of the ARIMA model, and establishing the ARIMA model according to the initial model coefficients includes: Performing differential processing on the historical time series data, and performing randomness and stationarity verification on the differentially processed historical time series data; Based on the verification result, the historical time series data after the difference processing is classified according to preset categories to obtain multiple historical interval data; Initial model coefficients of the ARIMA models corresponding to the multiple historical interval data are determined, and multiple ARIMA models are established according to the initial model coefficients.

4. The method according to claim 3, characterized in that The determining of initial model coefficients of the ARIMA models corresponding to the plurality of historical interval data respectively, and establishing a plurality of ARIMA models according to the initial model coefficients respectively, comprises: For target historical interval data among the plurality of historical interval data, generating target initial differential coefficients of a target ARIMA model corresponding to the target historical interval data according to a target preset category corresponding to the target historical interval data; Determine the target initial autoregressive coefficient and the target initial moving average coefficient of the target ARIMA model based on the verification result combined with the preset autocorrelation function and the preset partial autocorrelation function; A target ARIMA model is established according to the target initial difference coefficient, the target initial autoregressive coefficient and the target initial moving average coefficient.

5. The method according to claim 4, characterized in that The method of performing load forecasting on the historical time series data based on the ARIMA model, and iteratively updating the initial model coefficients according to the load forecast error between the load forecast result and the historical load information to obtain the model coefficients that meet the preset convergence conditions, includes: For the current load forecasting process, obtaining the regression weight corresponding to the target initial autoregressive coefficient and the moving average weight corresponding to the target initial moving average coefficient obtained in the previous load forecasting process; The current load forecast is performed on the target historical interval data through the target ARIMA model according to the regression weight and the moving average weight.

6. The method according to claim 5, characterized in that The method further comprises: Determine a current load prediction error between a current load prediction result and target historical load information corresponding to the target historical interval data; According to the magnitude relationship between the first load prediction error and the preset error reference value, the regression weight and the moving average weight are adjusted; If it is determined that the adjusted regression weight and moving average weight do not meet the preset convergence condition, the next load forecast is performed on the target historical interval data through the target ARIMA model based on the adjusted regression weight and moving average weight.

7. The method according to claim 6, characterized in that The method further comprises: If it is determined that the adjusted regression weight and moving average weight meet the preset convergence condition, the adjusted regression weight and moving average weight are determined as the target regression weight and target moving average weight of the target ARIMA model; The target model coefficients of the target ARIMA model are determined based on the target regression weight and the target moving average weight.

8. The method according to any one of claims 1 to 7, characterized in that Generate the forecast load information of the time series data to be forecasted according to the output result of the ARIMA model, including: Perform load forecasting on the time series data to be forecasted according to the ARIMA model to obtain a CPU forecast value, a memory forecast value, and a disk forecast value of the cloud resources; Based on the central processing unit prediction value, the memory prediction value and the disk prediction value, the prediction load information of the time series data to be predicted is generated.

9. The method according to claim 8, characterized in that The load forecasting of the time series data to be forecasted is performed based on the ARIMA model including the model coefficients to obtain the central processing unit forecast value, memory forecast value and disk forecast value of the cloud resources, including: Determine the category information corresponding to the time series data to be predicted in the preset category, and select the ARIMA model corresponding to the category information, wherein the model coefficients contained in the ARIMA model corresponding to the category information are model coefficients; The load of the time series data to be predicted is predicted by using the ARIMA model corresponding to the time series data to be predicted, so as to obtain the CPU prediction value, memory prediction value and disk prediction value of the cloud resources.

10. A cloud resource load prediction device, characterized in that: include: An acquisition module, configured to acquire time series data to be predicted of cloud resources; An input module is configured to input the time series data to be predicted into an ARIMA model, wherein the ARIMA model is used for load prediction of the cloud resources, and the model coefficients of the ARIMA model are obtained by iteratively updating the initial model coefficients of the ARIMA model based on the load prediction result of load prediction based on the historical time series data of the cloud resources and the load prediction error between the historical load information of the cloud resources; The generating module is configured to generate the forecast load information of the time series data to be forecasted according to the output result of the ARIMA model.

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

12. An electronic device comprising a storage medium, a processor, and a computer program stored in the storage medium and executable on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 9 is implemented.

13. A computer program product, comprising a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 9 is implemented.

Images