A cloud server cluster load prediction method, system, terminal, and storage medium

Abstract

The present application relates to a method, system, terminal, and storage medium for load prediction of a cloud server cluster. The method includes: obtaining task load data of a cloud server cluster in a cloud data center; using the S-MTF algorithm to convert the task load data from a multivariate time series into a supervised learning sequence; inputting the converted task load data into a trained esDNN model based on a convolutional-gated recurrent unit, and predicting the load status of the cloud data center within a preset future time through the esDNN network model. The embodiments of the present application solve the problems that existing prediction methods are difficult to handle task load prediction with high dimensions, high variability, and multiple variables, as well as inaccurate cloud server task load prediction, overly complex prediction methods, long training time, and vanishing gradients of long time series.

Description

Technical Field

[0001] This application belongs to the technical field of cloud computing, and particularly relates to a method, a system, a terminal and a storage medium for predicting the load of a cloud server cluster. Background Art

[0002] As an important part of IT technology, cloud computing technology brings many benefits to both service providers and customers, and is considered a successful operation model in the IT industry. However, cloud computing also faces many challenges, one of which is the very low efficiency of resource allocation for dynamic task loads. Predicting the task load of cloud servers can not only facilitate cloud service providers to better handle high-load problems and improve the stability of cloud servers, but also enable users of cloud servers to always maintain a stable state when using cloud services. Therefore, the problem of predicting the task load of cloud servers has become a decisive issue in determining how to select a resource allocation scheme.

[0003] Currently, the mainstream methods for predicting the task load of cloud servers mainly include:

[0004] (1) Logistic regression prediction: Logistic regression prediction, also known as Logisitic regression prediction, is often applied to fields such as univariate or multivariate time series prediction. Its essence is a binary classification problem and can be used to represent the possibility of something happening. Logistic regression has the advantages of simple implementation, very small computational cost during classification, high speed, and low storage resources. However, when the feature space is very large, the prediction effect of Logistic regression is not very good, and it is prone to underfitting during the prediction process, resulting in low accuracy.

[0005] (2) Recurrent Neural Network (RNN): The recurrent neural network is a type of neural network. When processing data, this network not only considers the input of the current state but also the previous information, so it is very sensitive to data with sequential characteristics. However, due to its own structural reasons, it often experiences gradient disappearance or gradient explosion during training. Therefore, it cannot handle some very long time series.

[0006] (3) Long Short-Term Memory (LSTM): Long Short-Term Memory is a variant of the recurrent neural network and is often applied to training multivariate time series prediction models. It can solve problems such as gradient disappearance and gradient explosion, as well as how to better analyze and predict time series. However, the amount of sample data required by Long Short-Term Memory is very large. If the amount of sample data is insufficient, it may cause problems such as inaccurate model prediction. In addition, due to its own internal structural reasons, the training time required by Long Short-Term Memory is very long. Summary of the Invention

[0007] This application provides a method, a system, a terminal, and a storage medium for predicting the load of a cloud server cluster, aiming to solve at least one of the above technical problems in the prior art to a certain extent.

[0008] To solve the above problems, this application provides the following technical solutions:

[0009] A method for predicting the load of a cloud server cluster includes:

[0010] Obtaining the task load data of the cloud server cluster in the cloud data center;

[0011] Using the S-MTF algorithm to transform the task load data from a multivariate time series into a supervised learning sequence;

[0012] Inputting the transformed task load data into a trained esDNN model based on a convolutional-gated recurrent unit, and predicting the load status of the cloud data center within a preset future time through the esDNN network model.

[0013] The technical solutions adopted in the embodiments of this application further include: The obtained task load data includes a timestamp, a machine number, a CPU utilization rate, and a memory occupancy size.

[0014] The technical solutions adopted in the embodiments of this application further include: After obtaining the task load data of the cloud server cluster in the cloud data center, it further includes:

[0015] Performing data cleaning and data normalization on the task load data;

[0016] The data cleaning is specifically: deleting the redundant items containing null data in the task load data, and then classifying the task load data according to the time series, and calculating the average value of each parameter with the same timestamp by using the grouping function;

[0017] The data normalization is specifically: using MinMaxScaler to transform each data, and scaling each data into a decimal number between 0 and 1. The operation formula of MinMaxScaler is:

[0018]

[0019] X scaled =X std *(X max -X min )+X min

[0020] In the above formula, X represents the set of data to be processed, X std represents the intermediate value for converting the value of the set X into a standardized value, Xmin and X max are the minimum and maximum data in the set, respectively, and X scaled is the data after the final normalization process.

[0021] The technical solution adopted in the embodiment of this application further includes: The specific process of converting the task load data from a multivariate time series into a supervised learning sequence by using the S-MTF algorithm is as follows:

[0022] Simultaneously obtain the time series data E(t) at the current time t, the time series data E(t-1) at the previous time, and the time series data E(t+1) at the next time;

[0023] Recombine E(t) with E(t-1) and E(t+1) respectively to obtain the time series recombination data L(i-1), C(i), and F(i+1) in the intermediate process of conversion;

[0024] Concatenate the three data of L(i-1), C(i), and F(i+1) to obtain the supervised learning sequences S(n), S(n-1), and S(n+1) corresponding to the current time, the previous time, and the next time respectively.

[0025] The technical solution adopted in the embodiment of this application further includes: The first layer of the esDNN model is a 1DCNN model, and the 1D CNN model includes an input layer, a convolutional layer, a pooling layer, a non-linear layer, and a fully connected layer; The second layer of the esDNN model is a GRU layer; The GRU includes an update gate, a reset gate, a candidate hidden layer, and an output gate, and the calculation formula for each gating unit is:

[0026] z t =σ(W z ·[h t-1 , x t )

[0027] r t =σ(W r ·[h t-1 , x t )

[0028] y′ t =tanh(W·[r t *h t-1 , x t )

[0029] y t =(1-z t )*h t-1 +z t *y′ t

[0030] In the above formula, zt For the update gate, r t For the reset gate, y t For the candidate hidden layer, y t For the output gate; σ represents the activation function, W, W z 、W h are matrices propagated within the GRU cell respectively; h t represents the output value of the current GRU cell, h t-1 represents the output value of the previous GRU cell; x t represents the value to be predicted transmitted at the current moment, and the matrix W is transformed from x t transformed.

[0031] The technical solution adopted in the embodiment of the present application further includes: The activation function of the esDNN model is:

[0032] Multiply ReLU and Sigmoid as the activation function Swish of the esDNN model:

[0033] f(x) = x · sigmoid(βx)

[0034] In the above formula, β is a constant or a trainable parameter.

[0035] The technical solution adopted in the embodiment of the present application further includes: The prediction of the load status of the cloud data center by the esDNN network model within a preset future time further includes:

[0036] Based on the change trend of the load status of the cloud data center within a preset future time period, use the auto-scaling mechanism to adjust the server scheduling strategy of the cloud data center and adjust the number of machines in the cloud server cluster.

[0037] Another technical solution adopted in the embodiment of the present application is: A cloud server cluster load prediction system, including:

[0038] Data acquisition module: used to acquire the task load data of the cloud server cluster in the cloud data center;

[0039] Data conversion module: used to convert the task load data from a multivariate time series into a supervised learning sequence using the S-MTF algorithm;

[0040] Load prediction module: used to input the converted task load data into the trained esDNN model based on convolutional-gated recurrent unit, and predict the load status of the cloud data center within a preset future time through the esDNN network model.

[0041] Another technical solution adopted in the embodiments of the present application is: a terminal, the terminal includes a processor and a memory coupled to the processor, wherein,

[0042] the memory stores program instructions for implementing the cloud server cluster load prediction method;

[0043] the processor is configured to execute the program instructions stored in the memory to control the cloud server cluster load prediction.

[0044] Another technical solution adopted in the embodiments of the present application is: a storage medium storing program instructions executable by a processor, the program instructions being used to execute the cloud server cluster load prediction method.

[0045] Compared with the prior art, the beneficial effects generated by the embodiments of the present application are as follows: the cloud server cluster load prediction method of the embodiments of the present application uses the sliding window method to convert the multivariate time series into a supervised learning sequence, accurately predicts the task load status of the cloud server cluster based on the convolutional-gated recurrent unit in deep learning, and uses the automatic scaling mechanism to dynamically adjust the server scheduling strategy of the cloud data center according to the future cloud server task load change trend within a certain period of time, solving the problems that existing prediction methods are difficult to cope with high-dimensional, highly variable, and multivariate task load predictions, as well as inaccurate cloud server task load predictions, overly complex prediction methods, long training time, and long time series gradient disappearance, optimizing the system performance and reducing energy consumption. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 is a flowchart of the cloud server cluster load prediction method according to the embodiments of the present application;

[0047] Figure 2 is a schematic diagram of the data conversion method according to the embodiments of the present application;

[0048] Figure 3 is a schematic diagram of the basic structure of GRU according to the embodiments of the present application;

[0049] Figure 4 is a schematic diagram of the structure of the cloud server cluster load prediction system according to the embodiments of the present application;

[0050] Figure 5 is a schematic diagram of the terminal structure according to the embodiments of the present application;

[0051] Figure 6 is a schematic diagram of the structure of the storage medium according to the embodiments of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0052] To make the objectives, technical solutions and advantages of this application more clear and understandable, the following further details this application in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not used to limit this application.

[0053] Please refer to Figure 1 , which is a flowchart of the cloud server cluster load prediction method according to the embodiments of this application. The cloud server cluster load prediction method according to the embodiments of this application includes the following steps:

[0054] S1: Obtain the task load data of the cloud server cluster from the cloud data center;

[0055] In this step, the cloud data center includes cloud servers, which provide cloud services to users as the provider. During the service provision process, the load data of the cloud server cluster is collected at regular time intervals and saved. The task load data refers to the task load from the running cloud server cluster collected by a specific server or a built-in program at regular intervals. The obtained task load data includes information such as time stamp time_stamp, machine number machine_id, CPU utilization rate cpu_util_percent, and memory occupancy size mem_util_percent.

[0056] S2: Preprocess the task load data;

[0057] In this step, the preprocessing of the task load data respectively includes data cleaning and data normalization processing. Among them, data cleaning is specifically as follows: First, delete the redundant items containing null data in the task load data to avoid the negative impact of redundant items on the prediction data. Then, classify the task load data according to the time series, and use the grouping function (groupby) to calculate the average value of each parameter with the same time stamp.

[0058] The data normalization processing is specifically as follows: Normalization is a data processing method for reducing dimensions. Normalization can not only improve the convergence speed of the model, but also improve the prediction accuracy. In the embodiments of this application, the normalization method is specifically as follows: Use MinMaxScaler to transform each data, and scale each data to a decimal between 0 and 1. The operation of MinMaxScaler is based on the min-max scaling method, and the specific formula is as follows:

[0059]

[0060] X scaled =X std *(X max -X min )+X min (2) In the above formula, X represents the set of data to be processed, and X std represents the intermediate value for converting the values of set X into standardized values, and X min and X max are the minimum and maximum data in the set respectively, and X scaled is the data after the final normalization process.

[0061] In other embodiments of the present application, a normalization method of changing the dimensional expression to a non-dimensional expression can also be adopted, that is, converting the data into a scalar.

[0062] S3: Use the S-MTF algorithm (Sliding Window for Multivariate Time Series Fore-cast) to convert the task load data from a multivariate time series into a supervised learning sequence;

[0063] In this step, a time series refers to a sequence formed by arranging the values of the same statistical indicator in the order of their occurrence time. Supervised learning is to train an optimal model with the feature and label information contained in the existing training samples, and then use this model to map all inputs to the corresponding outputs and make a simple judgment on the outputs, so that the model has the ability to predict and classify unknown data. In the embodiments of the present application, the S-MTF algorithm is used to convert the multivariate time series prediction problem into a prediction problem based on supervised learning. By using the preprocessed task load data as the input of the transfer function and reconstructing the time-related sequence data into a supervised learning sequence. The S-MTF algorithm contains all the data at the previous moment at any time, which has a linear relationship with time. At the same time, the S-MTF also contains future labels, so it can be applied to any time-related data set.

[0064] As Figure 2 shown, it is a schematic diagram of the data conversion method. The conversion process mainly includes time series splitting, time series recombination, and recombination sequence merging. Figure 2 On the left is the multivariate time series E(t) to be converted in the task load data. When the S-MTF algorithm processes the time series corresponding to time t, it will simultaneously obtain the time series data E(t) at the current moment and the time series data within two adjacent time periods, that is, the time series data E(t - 1) at the previous moment and the time series data E(t + 1) at the next moment. Then, E(t) is recombined with E(t - 1) and E(t + 1) respectively to obtain the time series recombination data L(i - 1), C(i), and F(i + 1) in the intermediate process of conversion. Finally, the three data L(i - 1), C(i), and F(i + 1) are spliced to obtain respectively as Figure 2The supervised learning sequences S(n), S(n - 1) and S(n + 1) corresponding to the three moments shown on the right. It should be noted that since there is no state of the previous moment at t = 0, and there will be no new time series after the last state recording ends, the actual length of the supervised learning sequence S(n) obtained will not be the same as that of the multivariate time series E(t), but will be shorter than the multivariate time series E(t) by several tuple lengths, and its specific length depends on the selected step size.

[0065] S4: Input the transformed task load data into the trained esDNN (Efficient Supervised learning-based Deep Neural Network, an efficient deep neural network algorithm based on supervised learning) model based on convolutional-gated recurrent unit, and predict the load status of the cloud data center in the future for a period of time through the esDNN model;

[0066] In this step, the network structure of the esDNN model includes two layers. The first layer is a CNN model (Convolutional Neural Networks), and the CNN model is based on the feedforward neural network model and consists of an input layer, a convolutional layer, a pooling layer, a non-linear layer, and a fully connected layer. Preferably, since 1DCNN (one-dimensional convolutional neural network) can extract features from local raw time series data and then establish a short-term correlation model between local time series data and subsequent trends, the 1D CNN is used as the first layer network structure of the esDNN in the embodiments of the present application.

[0067] The second layer of the esDNN model is GRU (Gated Recurrent Unit), and the basic structure of the GRU is as Figure 3 shown, including an update gate, a reset gate, a candidate hidden layer, and an output gate. In the embodiments of the present application, the GRU that combines the forget gate and the input gate into an "update gate" can effectively avoid problems such as gradient explosion and gradient disappearance existing in traditional algorithms (such as the BPTT algorithm). Among them, the calculation formulas of each gating unit are as follows:

[0068] z t = σ(W z ·[h t-1 , x t ) (3)

[0069] r t = σ(W r ·[h t-1 , x t ) (4)

[0070] y′t = tanh(W · [r t * h t-1 , x t ) (5)

[0071] y t = (1 - z t ) * h t-1 + z t * y' t (6)

[0072] In the above formula, z t is the update gate, r t is the reset gate, yt is the candidate hidden layer, y t is the output gate. σ represents the activation function, W, W z , W h are the matrices propagated within the GRU cell respectively. h t represents the output value of the current GRU cell, h t-1 represents the output value of the previous GRU cell. x t represents the value to be predicted transmitted at the current moment, and the matrix W is transformed from x t .

[0073] During model training, in the embodiments of the present application, by multiplying ReLU and Sigmoid, the activation function Swish of the esDNN model is obtained. The activation function Swish uses the same value for gating, that is, the so-called self-gating. The advantage of self-gating is that it only requires a simple scalar input, which can simplify the gating mechanism, while traditional gating requires multiple scalar inputs. This feature enables the activation function Swish to easily replace those activation functions that take a single scalar as input without changing the hidden capacity or the number of parameters. The formula of the activation function Swish is as follows:

[0074] f(x) = x · Sigmoid(βx) (7)

[0075] In the above formula, β is a constant or a trainable parameter.

[0076] S5: Based on the changing trend of the load status of the cloud data center within a certain period of the future, use the auto-scaling mechanism to adjust the server scheduling policy of the cloud data center and adjust the number of machines in the cloud server cluster;

[0077] In this step, auto-scaling means Auto-Scaling. The auto-scaling mechanism can dynamically adjust the number of active machines in the system according to the system state. The objects of auto-scaling include various resources such as computing, storage, and network resources. For cloud servers, the auto-scaling mechanism can specifically adjust the number of machines in the cluster, that is, turn off some servers when the overall system utilization rate is low, or turn on more servers when the system utilization rate is too high. By leveraging the advantages of the auto-scaling mechanism, system performance can be optimized and energy consumption can be reduced.

[0078] The goal of auto-scaling is to improve resource utilization and reduce the number of active machines on the premise of sufficient and accurate prediction. Therefore, the prerequisite for auto-scaling is to have sufficiently accurate task load prediction as support. Currently, common auto-scaling methods include threshold-based rules (such as static thresholds), which are implemented through horizontal scheduling, such as increasing the number of virtual machines. This method is not applicable to task load scheduling with large variability. As a deep learning prediction algorithm with high prediction accuracy, esDNN can accurately predict the load status in the next period of time and adjust the server scheduling strategy according to the change trend of the load status in the next period of time. The specific adjustment strategy is as follows: Use the ratio of the average number of active machines in at least two previous time periods to the total number of machines as the trigger threshold for the auto-scaling mechanism, use the CPU utilization rate as the input of the auto-scaling mechanism, and the output is the percentage of the number of active machines in the cloud data center in the current state. The calculation formula for this trigger threshold is:

[0079]

[0080] In the above formula, M(t) represents the number of active machines within the time interval t, m represents the number of time periods used for prediction before, and i represents the index value. In the embodiment of the present application, it is preferably set that m = 5.

[0081] S6: Optimize the cloud data center through an optimizer;

[0082] In this step, the optimizer is a supplementary strategy for cloud server cluster scheduling, which is used to turn on or off some low-load machines after predicting the server load size, so as to reduce the number of servers that need to be turned on in the entire cloud data center.

[0083] Based on the above, the cloud server cluster load prediction method of the embodiments of the present application uses a sliding window method to convert a multivariate time series into a supervised learning sequence, accurately predicts the task load status of the cloud server cluster based on the convolutional-gated recurrent unit in deep learning, and uses an automatic scaling mechanism to dynamically adjust the server scheduling strategy of the cloud data center according to the change trend of the cloud server task load in a certain future time period, solving the problems that existing prediction methods are difficult to cope with task load prediction with high dimensions, high variability, and multiple variables, as well as inaccurate cloud server task load prediction, overly complex prediction methods, long training time, and vanishing gradients of long time series, optimizing system performance and reducing energy consumption.

[0084] To verify the feasibility and effectiveness of the embodiments of the present application, experiments are conducted using the task load dataset cluster-trace-v2018 of the cloud server cluster from Alibaba and the task load dataset clusterdata-2011-2 of the cloud server cluster from Google. For the source data obtained from the Alibaba dataset, reference variables that often appear with null values are removed, and null values of reference variables that occasionally appear with null values are assigned 0. For the source data obtained from the Google dataset, a new column of data is added to the task_usage (task resource usage table), and this new data is a 1-second CPU usage sample randomly selected within the 5-minute usage reporting period related to the task. And the S-MTF algorithm, an algorithm widely used in current load prediction methods, is used as a comparison algorithm. The experimental results show that the esDNN algorithm proposed by the present invention is superior to existing methods in cloud server task load prediction.

[0085] Please refer to Figure 4 , which is a schematic structural diagram of the cloud server cluster load prediction system of the embodiments of the present application. The cloud server cluster load prediction system 40 of the embodiments of the present application includes:

[0086] A data acquisition module 41: used to acquire the task load data of the cloud server cluster in the cloud data center;

[0087] A data conversion module 42: used to convert the task load data from a multivariate time series into a supervised learning sequence using the S-MTF algorithm;

[0088] A load prediction module 43: used to input the converted task load data into a trained esDNN model based on a convolutional-gated recurrent unit, and predict the load status of the cloud data center within a preset future time through the esDNN network model.

[0089] Please refer to Figure 5 , which is a schematic structural diagram of the terminal of the embodiments of the present application. The terminal 50 includes a processor 51 and a memory 52 coupled to the processor 51.

[0090] The memory 52 stores program instructions for implementing the above-mentioned cloud server cluster load prediction method.

[0091] The processor 51 is configured to execute the program instructions stored in the memory 52 to control the cloud server cluster load prediction.

[0092] Among them, the processor 51 can also be referred to as a CPU (Central Processing Unit). The processor 51 may be an integrated circuit chip with signal processing capabilities. The processor 51 can also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc.

[0093] Please refer to Figure 6 , which is a schematic structural diagram of the storage medium according to the embodiment of the present application. The storage medium according to the embodiment of the present application stores a program file 61 capable of implementing all the above methods. Among them, the program file 61 can be stored in the above storage medium in the form of a software product, including several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor to execute all or part of the steps of the methods according to various embodiments of the present invention. The aforementioned storage medium includes: various media such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disc that can store program codes, or a terminal device such as a computer, a server, a mobile phone, or a tablet.

[0094] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined in the present application 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 these embodiments shown in the present application, but will be accorded the widest scope consistent with the principles and novel features disclosed in the present application.

Claims

1. A method for predicting the load of a cloud server cluster, characterized in that, it includes: Obtain the task load data of the cloud server cluster in the cloud data center; Use the S-MTF algorithm to transform the task load data from a multivariate time series into a supervised learning sequence; Input the transformed task load data into a trained esDNN model based on a convolutional-gated recurrent unit, and predict the load status of the cloud data center within a preset future time through the esDNN network model; where: The specific process of using the S-MTF algorithm to transform the task load data from a multivariate time series into a supervised learning sequence is as follows: First, simultaneously obtain the time series data E(t) at the current time t, the time series data E(t - 1) at the previous time, and the time series data E(t + 1) at the next time; Then, recombine E(t) with E(t - 1) and E(t + 1) respectively to obtain the time series recombination data L(i - 1), C(i), and F(i + 1) in the intermediate process of transformation; Finally, splice the three data of L(i - 1), C(i), and F(i + 1) to obtain the supervised learning sequences S(n), S(n - 1), and S(n + 1) corresponding to the current time, the previous time, and the next time respectively.

2. The method for predicting the load of a cloud server cluster according to claim 1, characterized in that, The obtained task load data includes a timestamp, a machine number, a CPU utilization rate, and a memory occupancy size.

3. The method for predicting the load of a cloud server cluster according to claim 2, characterized in that, After obtaining the task load data of the cloud server cluster in the cloud data center, it further includes: Perform data cleaning and data normalization on the task load data; The data cleaning is specifically: delete the redundant items containing null data in the task load data, and then classify the task load data according to the time series, and use the grouping function to calculate the average value of each parameter with the same timestamp; The data normalization is specifically: use MinMaxScaler to transform each data, and scale each data to a decimal between 0 and 1. The MinMaxScaler operation formula is: X scaled = X std *(X max - X min ) + X min In the above formula, X represents the set of data to be processed, X std represents the intermediate value for converting the values of the set X into standardized values, X min and X max are the minimum and maximum data in the set respectively, X scaled is the data after the final normalization process.

4. The method for predicting the load of a cloud server cluster according to any one of claims 1 to 3, characterized in that, The first layer of the esDNN model is a 1D CNN model, and the 1D CNN model includes an input layer, a convolutional layer, a pooling layer, a non-linear layer, and a fully connected layer; the second layer of the esDNN model is a GRU layer; the GRU includes an update gate, a reset gate, a candidate hidden layer, and an output gate. The calculation formulas for each gating unit are: z t = σ(W z · [h t-1 , x t ) r t = σ(W r · [h t-1 , x t ) y′ t =tanh(W·[r t *h t-， x t ) y t = (1 - z t ) * h t-1 + z t * y' t In the above formula, z t is the update gate, r t is the reset gate, y' t is the candidate hidden layer, y t is the output gate; σ represents the activation function, W, W z , W h are the matrices propagated within the GRU cell respectively; h t represents the output value of the current GRU cell, h t-1 represents the output value of the previous GRU cell; x t represents the value to be predicted transmitted at the current moment, and the matrix W is transformed from x t .

5. The method for predicting the load of a cloud server cluster according to claim 4, characterized in that, The activation function of the esDNN model is: Multiply ReLU and Sigmoid as the activation function Swish of the esDNN model: f(x) = x · sigmoid(βx) In the above formula, β is a constant or a trainable parameter.

6. The cloud server cluster load prediction method according to claim 5, wherein, the prediction of the load status of the cloud data center within a preset future time by the esDNN network model further includes: Based on the change trend of the load status of the cloud data center within a preset future time period, the server scheduling strategy of the cloud data center is adjusted by using an auto-scaling mechanism to adjust the number of machines in the cloud server cluster.

7. A cloud server cluster load prediction system using the cloud server cluster load prediction method according to claim 1, wherein, it includes: A data acquisition module: used to acquire the task load data of the cloud server cluster in the cloud data center; A data conversion module: used to convert the task load data from a multivariate time series into a supervised learning sequence by using the S-MTF algorithm; A load prediction module: used to input the converted task load data into a trained esDNN model based on a convolutional-gated recurrent unit, and predict the load status of the cloud data center within a preset future time through the esDNN network model.

8. A terminal, wherein, the terminal includes a processor and a memory coupled to the processor, wherein, the memory stores program instructions for implementing the cloud server cluster load prediction method according to any one of claims 1-6; the processor is used to execute the program instructions stored in the memory to control the cloud server cluster load prediction.

9. A storage medium, wherein, it stores program instructions that can be run by a processor, and the program instructions are used to execute the cloud server cluster load prediction method according to any one of claims 1 to 6.

Images