Cloud platform resource prediction method based on WT-IWOA-GRU model

Abstract

The invention discloses a cloud platform resource prediction method based on a WT-IWOA-GRU model, relates to the technical field of cloud platform resource prediction, and solves the technical problem that the quantity of calculation, storage, network and other resources required by a cloud platform in a period of time in the future is difficult to predict in the prediction of cloud platform resources. According to the method, cloud platform resource time sequence data is preprocessed, a cloud platform resource prediction model based on a GRU neural network is established, hyper-parameters of the cloud platform resource prediction model are optimized by using an improved whale optimization algorithm, and future cloud platform resources are predicted by using the optimized cloud platform resource prediction model; the method can improve the accuracy and timeliness of cloud platform resource prediction, and can be suitable for cloud platform resource time sequence prediction with complex characteristics.

Description

Technical Field

[0001] The present invention belongs to the field of cloud platform resource scheduling, relates to cloud platform resource prediction technology, and specifically is a cloud platform resource prediction method based on the WT-IWOA-GRU model. Background Art

[0002] Currently, cloud computing has become the mainstream technology in the IT field, and its elastic resources and on-demand service model have attracted wide attention. With the increase in cloud computing applications, the reasonable utilization and optimized management of resources have become crucial. The resources of the cloud platform include various resources such as computing, storage, and network. How to effectively predict and manage the usage of these resources to improve resource utilization and reduce costs is one of the important challenges faced by current cloud computing. Under the cloud computing environment, the uncertainty of user demands is relatively large, and sudden changes in resource demands may occur at any time. Traditional resource management methods often cannot meet such dynamic changing demands.

[0003] The WT-IWOA-GRU model is a model based on wavelet transform (WT), improved whale optimization algorithm (IWOA), and gated recurrent unit (GRU), and is a model that can be applied to cloud platform resource prediction; currently, most cloud platform resource prediction methods are difficult to predict the amounts of resources such as computing, storage, and network required by the cloud platform in a future period of time, so that cloud service providers can better plan and adjust resource allocation, avoid waste of cloud platform resources, and reduce costs.

[0004] Therefore, the present invention discloses a cloud platform resource prediction method based on the WT-IWOA-GRU model to solve the above technical problems. Summary of the Invention

[0005] The present invention aims to solve at least one of the technical problems existing in the prior art; for this purpose, the present invention proposes a cloud platform resource prediction method based on the WT-IWOA-GRU model to solve the technical problem that in the prediction of cloud platform resources, it is difficult to predict the amounts of resources such as computing, storage, and network required by the cloud platform in a future period of time. The present invention preprocesses the time series data of cloud platform resources and establishes a cloud platform resource prediction model based on the GRU neural network, optimizes the hyperparameters of the cloud platform resource prediction model by using the improved whale optimization algorithm, and uses the optimized cloud platform resource prediction model to predict future cloud platform resources to solve the above problem.

[0006] To achieve the above object, a first aspect of the present invention provides a cloud platform resource prediction method based on the WT-IWOA-GRU model, including:

[0007] S1: Collect cloud platform resource time series data;

[0008] S2: Preprocess the cloud platform resource time series data; wherein, the preprocessing includes wavelet transform and data normalization;

[0009] S3: Establish a cloud platform resource prediction model based on the GRU neural network according to the data obtained in S2;

[0010] S4: Optimize the hyperparameters of the cloud platform resource prediction model using the improved whale optimization algorithm;

[0011] S5: Predict future cloud platform resources using the optimized cloud platform resource prediction model.

[0012] Preferably, the collection of cloud platform resource time series data includes CPU utilization rate, database response time, network bandwidth, and free memory.

[0013] Preferably, the preprocessing of the cloud platform resource time series data includes:

[0014] Use wavelet transform to extract features from the original cloud platform resource data. The specific formula for wavelet transform is:

[0015]

[0016] where x(t) is the original cloud platform resource sequence, W(a,b) is the wavelet coefficient under the scale parameter a and the translation parameter b, and ψ a,b (t) is the scaled and translated version of the wavelet function under the scale parameter a and the translation parameter b;

[0017] Select the Daubechies wavelet as the wavelet function and set the decomposition level to 3 for wavelet decomposition;

[0018] Perform data normalization on the subsequences after wavelet decomposition. The specific formula for data normalization is:

[0019]

[0020] where X is the cloud platform resource sequence, X min is the minimum value in the sequence, X max is the maximum value in the sequence, and X normalized is the normalized data.

[0021] Preferably, the establishment of the cloud platform resource prediction model based on the GRU neural network includes:

[0022] Let the cloud platform resource input sequence be \(x\) t =(x 1 ,x 2 ,x 3 …x d ); where \(d\) is the time window, and \(t\) is the cloud platform resource value at the \(t\)-th moment in the subsequence.

[0023] Calculate the reset gate \(r\) t , update gate \(z\) t and candidate hidden state . Based on the reset gate \(r\) t , update gate \(z\) t and candidate hidden state calculate the new hidden state \(h\) t , and the specific calculation formula is:[[]]

[0024]

[0025] Output \(h\) t as the final network prediction result.

[0026] Preferably, the calculation of the reset gate \(r\) t , update gate \(z\) t and candidate hidden state includes:[[]]

[0027] Calculate the reset gate \(r\) t through formula (3), and the specific calculation formula is:[[]]

[0028] r t =σ(W r ·[h t-1 ,x t )(3);

[0029] where σ represents the sigmoid activation function, W r is the weight matrix related to the reset gate, and [h t-1 ,x t represents the vector formed by concatenating the hidden state \(h\) t-1 at the previous time step with the input \(x\) t at the current time step;

[0030] Calculate the update gate \(z\) t through formula (4), and the specific calculation formula is:[[]]

[0031] z t =σ(W z ·[h t-1 ,x t )(4);

[0032] Among them, W z is the weight matrix related to the update gate;

[0033] The candidate hidden state is calculated through formula (5) The specific calculation formula is:

[0034]

[0035] Among them, tanh represents the hyperbolic tangent activation function, ⊙ represents element-wise multiplication, and W h is the weight matrix related to the candidate hidden state.

[0036] Preferably, the hyperparameters of the cloud platform resource prediction model are optimized by using the improved whale optimization algorithm, including:

[0037] A1: Mark the time window and the number of hidden layer nodes of the cloud platform resource prediction model as hyperparameters, and perform initialization settings on the hyperparameters; among them, the initialization settings include initializing the hyperparameter range, the number of whales, the learning rate, and the number of iterations;

[0038] A2: Initialize the whale population, establish and train the cloud platform resource prediction model according to the positions of the initialized whale population; set the mean square error MSE of the whale population positions as the fitness function, and calculate the fitness value of each whale according to the fitness function;

[0039] A3: Update the positions of the whale individuals through the operations of surrounding the prey, bubble net attack operation, and searching for the prey, establish and train the cloud platform resource prediction model based on the positions, and update the fitness values of the whales;

[0040] A4: Determine whether the maximum number of iterations is reached; if yes, decode the position of the optimal whale individual, and use the hyperparameters corresponding to the position of the optimal whale individual as the best hyperparameters of the cloud platform resource model; if not, jump to A3.

[0041] Preferably, the initialization of the whale population includes:

[0042] Randomly generate a specified number of whales through formula (7), and the specific calculation formula is:

[0043] X i = D max + d × (D max - D min )(7);

[0044] Among them, X i is the position of the i-th whale, D max is the maximum value of the hyperparameter range, D min is the minimum value of the hyperparameter range, and d is a random number between [0, 1].

[0045] Preferably, the operation of surrounding the prey includes:

[0046] Taking t as the current iteration number, the operation of surrounding the prey is carried out through formula (8), and the specific formula is:

[0047]

[0048] where T max is the maximum number of iterations, A and C are coefficient vectors, X(t) is the current whale position, X(t + 1) is the obtained whale position, X * (t) is the current optimal whale position, r 1 and r 2 are random vectors within the range of [0, 1], and a represents linearly decaying to 0 during the whale search process.

[0049] Preferably, the bubble net attack operation includes:

[0050] The bubble net attack operation is carried out through formula (9), and the specific formula is:

[0051]

[0052] where b is a constant representing the shape of the helix; I is a random vector between [-1, 1], p is a random number between [0, 1], and are the adaptive weights of the whale positions.

[0053] Preferably, the operation of searching for the prey includes:

[0054] The operation of searching for the prey is carried out through formula (10), and the specific formula is:

[0055]

[0056] where X rand represents a random position vector in the current whale group.

[0057] Compared with the prior art, the beneficial effects of the present invention are:

[0058] 1. The present invention pre - processes the time - series data of cloud - platform resources and establishes a cloud - platform resource prediction model based on the GRU neural network. It uses an improved whale optimization algorithm to optimize the hyperparameters of the cloud - platform resource prediction model, and uses the optimized cloud - platform resource prediction model to predict future cloud - platform resources, solving the technical problem that it is difficult to predict the amounts of computing, storage, network and other resources required by the cloud platform in a future period of time. The present invention can improve the accuracy and timeliness of cloud - platform resource prediction, and is applicable to the time - series prediction of cloud - platform resources with complex characteristics.

[0059] 2. The time - series of cloud - platform resources has change characteristics such as periodicity, suddenness and randomness. Wavelet transform can analyze the cloud - platform time - series data in both the time domain and the frequency domain. It can not only extract the overall change trend of the cloud - platform resource time - series, but also retain the local characteristics of the cloud - platform resource time - series. This method uses wavelet transform to perform multi - scale analysis on the cloud - platform resource sequence, decomposes the original cloud - platform resource sequence into several high - frequency and low - frequency sequences, and then uses a neural network to learn the change trend and volatility characteristics of the cloud - platform resource data, constructing a cloud - platform resource prediction model based on wavelet decomposition and reconstruction.

[0060] 3. The time - series of cloud - platform resources has long - term correlation. Traditional methods are essentially linear combinations of historical data and historical noise, and are only applicable to processing non - stationary sequences. It is difficult to obtain an accurate prediction result for the cloud - platform resource sequence. This method uses a GRU neural network. By introducing the mechanism of update gates and reset gates, this model can more effectively capture the long - term dependence relationship of the cloud - platform resource time - series and obtain a more accurate long - sequence prediction result. At the same time, the GRU neural network avoids the problems of gradient disappearance and gradient explosion of traditional neural networks, further improving the model prediction performance and being more applicable to the time - series prediction of cloud - platform resources with complex characteristics.

[0061] 4. Aiming at the problem that the hyperparameters of the neural network have a great influence on the prediction result and are difficult to tune, this method uses the whale optimization algorithm to optimize the hyperparameters of the GRU model, which can avoid the model falling into local optima and further improve the prediction accuracy of the model. At the same time, based on the existing whale optimization algorithm, an adaptive weight is introduced to further improve the search ability and convergence speed of the algorithm. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following - described drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0063] Figure 1 It is a schematic diagram of the operation steps of the present invention;

[0064] Figure 2 It is a schematic diagram of the wavelet transform of the present invention;

[0065] Figure 3 It is a structural diagram of the GRU unit of the present invention;

[0066] Figure 4 It is a schematic diagram of the operation steps for optimizing the cloud platform resource prediction model of the present invention;

[0067] Figure 5 It is a schematic diagram of the prediction result in the test set of the present invention;

[0068] Figure 6 It is a schematic diagram of the LSTM prediction result of the present invention;

[0069] Figure 7 It is a schematic diagram of the prediction result of the SVM model of the present invention. Specific embodiments

[0070] Next, the technical solution of the present invention will be clearly and completely described in conjunction with the embodiments. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative work shall fall within the protection scope of the present invention.

[0071] Please refer to Figure 1 - Figure 2 , the first aspect embodiment of the present invention provides a cloud platform resource prediction method based on the WT-IWOA-GRU model, including:

[0072] S1: Collect cloud platform resource time series data;

[0073] S2: Preprocess the cloud platform resource time series data; wherein, the preprocessing includes wavelet transform and data normalization;

[0074] S3: According to the data obtained in S2, establish a cloud platform resource prediction model based on the GRU neural network;

[0075] S4: Use the improved whale optimization algorithm to optimize the hyperparameters of the cloud platform resource prediction model;

[0076] S5: Use the optimized cloud platform resource prediction model to predict future cloud platform resources.

[0077] In this embodiment, an independently built OpenStack cloud platform is adopted, and the data is the CPU utilization rate, which is collected at one-hour intervals. The prediction results of this method in the test set of the resource time series data of the OpenStack cloud platform are as Figure 5 shown. The root mean square error RMSE, mean absolute error, MAE, and mean absolute percentage error MAPE are respectively used as evaluation indicators, as shown in formulas (11), (12), and (13). N is the number of data samples, y predictive is the predicted value, and y true is the actual value. RMSE, MAE, and MAPE are used as evaluation indicators for the prediction results. The smaller the evaluation indicator, the smaller the prediction error of the model and the higher the prediction accuracy. Figure 6 and Figure 7 are respectively the prediction results of the LSTM and SVM models. The comparison of the prediction performance of this method with the LSTM and SVM models is shown in Table 1. It can be seen that compared with the single method LSTM and the traditional machine learning method SVM, the values of each index of this method are lower, indicating that this method has higher prediction accuracy and can more effectively predict the change trend of cloud platform resources.

[0078]

[0079] Table 1 Comparison of prediction performance of different models

[0080]

[0081] In this application, the time series data of cloud platform resources is collected, including CPU utilization rate, database response time, network bandwidth, and free memory.

[0082] In this application, the time series data of cloud platform resources is preprocessed, including:

[0083] Feature extraction is performed on the original cloud platform resource data using wavelet transform. The specific formula for wavelet transform is:

[0084]

[0085] where x(t) is the original cloud platform resource sequence, W(a,b) is the wavelet coefficient under the scale parameter a and the translation parameter b, and ψ a,b (t) is the scaled and translated version of the wavelet function under the scale parameter a and the translation parameter b;

[0086] Daubechies wavelet is selected as the wavelet function, and the decomposition level is set to 3 for wavelet decomposition;

[0087] Data normalization is performed on the subsequences after wavelet decomposition. The specific formula for data normalization is:

[0088]

[0089] Among them, X is the cloud platform resource sequence, and X min is the minimum value in the sequence, and X max is the maximum value in the sequence, and X normalized is the data after normalization.

[0090] It should be noted that wavelet transform can decompose the cloud platform resource time series into subsequences of different scales and frequencies, so as to extract the local features and global trends in the cloud platform resource time series.

[0091] It should be noted that the selection of the Daubechies wavelet as the wavelet function is obtained based on the discrete characteristics of the cloud platform resource time series data.

[0092] It should be noted that after wavelet transform, an approximation component and three detail components are obtained. Among them, the approximation component represents the overall change trend in the cloud platform resource time series, and the detail components represent the local features in the cloud platform resource time series; subsequently, each component is modeled and predicted separately, and then wavelet reconstruction is performed to obtain the final cloud platform resource prediction result.

[0093] It should be noted that data normalization of the subsequences after wavelet decomposition can eliminate the influence of different dimensions, accelerate the convergence speed of the prediction model, and improve the performance of the model.

[0094] It should be noted that data normalization can map each subsequence after wavelet decomposition to the range of [0, 1].

[0095] Please refer to Figure 3 , in this application, a cloud platform resource prediction model based on the GRU neural network is established, including:

[0096] Let the cloud platform resource input sequence be x t =(x 1 , x 2 , x 3 …x d ); among them, d is the time window, and t is the cloud platform resource value at the t-th moment in the subsequence;

[0097] Perform the calculation of the reset gate r t , update gate z t and candidate hidden state . Based on the reset gate r t , update gate z t and candidate hidden state , calculate the new hidden state h t , and the specific calculation formula is:

[0098]

[0099] Take h t as the final network prediction result for output.

[0100] It should be noted that a cloud platform resource prediction model based on the GRU neural network is established. The GRU neural network can capture long-term dependencies and deeply mine the long-term information in the cloud platform resource sequence.

[0101] In this application, the reset gate r t , update gate z t and candidate hidden state are calculated, including:

[0102] The reset gate r t is calculated through formula (3), and the specific calculation formula is:

[0103] r t = σ(W r · [h t-1 , x t )(3);

[0104] where σ represents the sigmoid activation function, W r is the weight matrix related to the reset gate, and [h t-1 , x t represents the vector formed by concatenating the hidden state h t-1 at the previous time step and the input x t at the current time step;

[0105] The update gate z t is calculated through formula (4), and the specific calculation formula is:

[0106] z t = σ(W z · [h t-1 , x t )(4);

[0107] where W z is the weight matrix related to the update gate;

[0108] The candidate hidden state is calculated through formula (5), and the specific calculation formula is:

[0109]

[0110] where tanh represents the hyperbolic tangent activation function, ⊙ represents element-wise multiplication, and W h is the weight matrix related to the candidate hidden state.

[0111] It should be noted that h t is the hidden state h at the previous time step t-1 and the candidate hidden state is a weighted sum, and the weights are controlled by the update gate z t The update gate determines how much information of cloud platform resources from the previous time step is retained and how much new information of cloud platform resources is introduced. The cloud platform resource prediction model based on the GRU neural network outputs h t as the final network prediction result.

[0112] Please refer to Figure 4 In this application, the improved whale optimization algorithm is used to optimize the hyperparameters of the cloud platform resource prediction model, including:

[0113] A1: Mark the time window and the number of hidden layer nodes of the cloud platform resource prediction model as hyperparameters and perform initialization settings on the hyperparameters; among them, the initialization settings include initializing the hyperparameter range, the number of whales, the learning rate, and the number of iterations;

[0114] A2: Initialize the whale population, establish and train the cloud platform resource prediction model according to the positions of the initialized whale population; set the mean square error MSE of the whale population positions as the fitness function, and calculate the fitness value of each whale according to the fitness function;

[0115] A3: Update the positions of whale individuals through the encircling prey operation, the bubble net attack operation, and the searching for prey operation, establish and train the cloud platform resource prediction model based on the positions, and update the fitness values of the whales;

[0116] A4: Judge whether the maximum number of iterations is reached; if yes, decode the position of the optimal whale individual, and use the hyperparameters corresponding to the position of the optimal whale individual as the best hyperparameters of the cloud platform model; if not, jump to A3.

[0117] It should be noted that the initialization settings are obtained through manual settings.

[0118] It should be noted that the specific method for optimizing the hyperparameters of the cloud platform resource prediction model using the improved whale optimization algorithm is as follows: First, determine the hyperparameters to be optimized in the cloud platform resource prediction model as the time window and the number of hidden layer nodes, and perform parameter initialization, including the number of iterations, the number of whales, the learning rate, the hyperparameter range, etc.; then perform population initialization, randomly generate the positions of the whale population, and each position corresponds to a set of hyperparameter combinations. Based on the position of the whale, establish and train the cloud platform resource prediction model; then set the mean square error MSE of the position as the fitness function, and calculate the fitness value of each whale; start the algorithm iteration, update the position of the whale through the operations of surrounding prey, bubble net attack, and searching for prey in the whale optimization algorithm, establish and train the cloud platform resource prediction model based on this position, and update the fitness value of the whale; after the algorithm iteration ends, output the position of the optimal whale individual, and use the corresponding hyperparameters as the best hyperparameters of the cloud platform resource prediction model.

[0119] In this application, the initialization of the whale population includes:

[0120] Calculate a specified number of whales randomly generated through formula (7). The specific calculation formula is:

[0121] X i = D max + d × (D max - D min )(7);

[0122] Among them, X i is the position of the i-th whale, D max is the maximum value of the hyperparameter range, D min is the minimum value of the hyperparameter range, and d is a random number between [0, 1].

[0123] In this application, the operation of surrounding prey includes:

[0124] Take t as the current number of iterations, and perform the operation of surrounding prey through formula (8). The specific formula is:

[0125]

[0126] Among them, T max is the maximum number of iterations, A and C are coefficient vectors, X(t) is the current whale position, X(t + 1) is the obtained whale position, X * (t) is the current optimal whale position, r 1 and r 2 are random vectors within the range of [0, 1], and a represents a linear decay to 0 during the whale search process.

[0127] In this application, the bubble net attack operation includes:

[0128] Perform the bubble net attack operation through formula (9), and the specific formula is as follows:

[0129]

[0130] Among them, b is a constant representing the shape of the helix; I is a random vector between [-1, 1], p is a random number between [0, 1], and are the adaptive weights of the whale position.

[0131] It should be noted that and are the adaptive weights of the whale position. By introducing this adaptive weight, the convergence speed of the algorithm can be further improved.

[0132] In this application, the operation of searching for prey includes:

[0133] Perform the operation of searching for prey through formula (10), and the specific formula is as follows:

[0134]

[0135] Among them, X rand represents the random position vector in the current whale group.

[0136] Some of the data in the above formula are calculated by removing the dimension and taking their numerical values. The formula is the one closest to the actual situation obtained through software simulation of a large amount of collected data; the preset parameters and preset thresholds in the formula are set by those skilled in the art according to the actual situation or obtained through simulation of a large amount of data.

[0137] The working principle of the present invention:

[0138] Collect the time series data of cloud platform resources, perform preprocessing of wavelet transform and data normalization on the time series data of cloud platform resources to provide data support for subsequent analysis; then establish a cloud platform resource prediction model based on the GRU neural network. There are parameter errors in the established cloud platform resource prediction model. Therefore, the present invention then uses the improved whale optimization algorithm to optimize the hyperparameters of the cloud platform resource prediction model, and finally uses the optimized cloud platform resource prediction model to predict future cloud platform resources.

[0139] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical method of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A cloud platform resource prediction method based on the WT-IWOA-GRU model, characterized by: S1: Collect cloud platform resource time series data; S2: Preprocessing the cloud platform resource time series data; the preprocessing includes wavelet transform and data normalization; S3: Based on the data obtained in S2, a cloud platform resource prediction model based on the GRU neural network is established; S4: Optimize the hyperparameters of the cloud platform resource prediction model using the improved whale optimization algorithm; S5: Use the optimized cloud platform resource prediction model to predict future cloud platform resources.

2. The cloud platform resource prediction method based on the WT-IWOA-GRU model according to claim 1, characterized in that: The collected cloud platform resource time series data includes CPU utilization, database response time, network bandwidth, and free memory.

3. The cloud platform resource prediction method based on the WT-IWOA-GRU model according to claim 1, characterized in that: The preprocessing of the cloud platform resource time series data includes: Wavelet transform is used to extract features from the original cloud platform resource data. The specific formula of wavelet transform is: Where x(t) is the original cloud platform resource sequence, W(a,b) is the wavelet coefficient under scale parameter a and translation parameter b, ψ a,b (t) is the scaled and translated version of the wavelet function with scale parameter a and translation parameter b; Select Daubechies wavelet as the wavelet function and set the decomposition level to 3 for wavelet decomposition; The subsequence after wavelet decomposition is normalized. The specific formula for data normalization is: Among them, X is the cloud platform resource sequence, X min is the minimum value in the sequence, X max is the maximum value in the sequence, X normalized is the normalized data.

4. The cloud platform resource prediction method based on the WT-IWOA-GRU model according to claim 1, characterized in that: The cloud platform resource prediction model based on the GRU neural network is established, including: Let the cloud platform resource input sequence be x t =(x1,x2,x3…x d ), where d is the time window, and t is the cloud platform resource value at the tth moment in the subsequence; Reset gate t , update gate z t and candidate hidden states The calculation is based on the reset gate r t , update gate z t and candidate hidden states Calculate the new hidden state h t , the specific calculation formula is: h t Output as the final network prediction result.

5. The cloud platform resource prediction method based on the WT-IWOA-GRU model according to claim 4, characterized in that: The reset gate r t , update gate z t and candidate hidden states Calculations include: The reset gate r is calculated by formula (3) t , the specific calculation formula is: r t =σ(W r ·[h t-1 ,x t ])(3); Among them, σ represents the sigmoid activation function, W r is the weight matrix associated with the reset gate, [h t-1 ,x t ] means to change the hidden state h of the previous time step t-1 With the input x at the current time step t The concatenated vector; The update gate z is calculated by formula (4): t , the specific calculation formula is: z t =σ(W z ·[h t-1 ,x t ])(4); Among them, W z is the weight matrix associated with the update gate; The candidate hidden state is calculated by formula (5) The specific calculation formula is: Among them, tanh represents the hyperbolic tangent activation function, ⊙ represents element-by-element multiplication, and W h is the weight matrix associated with the candidate hidden states.

6. The cloud platform resource prediction method based on the WT-IWOA-GRU model according to claim 1, characterized in that: The method of optimizing the hyper parameters of the cloud platform resource prediction model using the improved whale optimization algorithm includes: A1: Mark the time window and number of hidden layer nodes of the cloud platform resource prediction model as hyperparameters, and initialize the hyperparameters; the initialization settings include the initialization hyperparameter range, number of whales, learning rate, and number of iterations; A2: Initialize the whale population, establish and train the cloud platform resource prediction model based on the location of the whale population after initialization; set the mean square error (MSE) of the whale population location as the fitness function, and calculate the fitness value of each whale based on the fitness function; A3: Update the location of individual whales through the operations of surrounding prey, attacking with bubble nets, and searching for prey, establish and train a cloud platform resource prediction model based on the location, and update the fitness value of the whales; A4: Determine whether the maximum number of iterations has been reached; if yes, decode the position of the optimal whale individual, and use the hyperparameters corresponding to the position of the optimal whale individual as the optimal hyperparameters of the cloud platform resource model; if no, jump to A3.

7. The cloud platform resource prediction method based on the WT-IWOA-GRU model according to claim 6, characterized in that: Initializing the whale population includes: The specified number of whales is randomly generated by formula (7). The specific calculation formula is: X i =D max +d×(D max -D min )(7); Among them, X i is the position of the ith whale, D max is the maximum value of the hyperparameter range, D min is the minimum value of the hyperparameter range, and d is a random number between [0,1].

8. The cloud platform resource prediction method based on the WT-IWOA-GRU model according to claim 6, characterized in that: The operation of encircling the prey includes: Take t as the current iteration number and use formula (8) to perform the prey encirclement operation. The specific formula is: Among them, T max is the maximum number of iterations, A and C are coefficient vectors, X(t) is the current whale position, X(t+1) is the obtained whale position, X * (t) is the current optimal whale position, r1 and r2 are random vectors in the range [0,1], and a represents linear decay to 0 during the whale search process.

9. The cloud platform resource prediction method based on the WT-IWOA-GRU model according to claim 8, characterized in that: The Bubble Network attack operation includes: The bubble net attack operation is performed through formula (9), and the specific formula is: Where b is a constant representing the shape of the spiral; I is a random vector between [-1,1], and p is a random number between [0,1]. and is the adaptive weight of the whale position.

10. The cloud platform resource prediction method based on the WT-IWOA-GRU model according to claim 9, characterized in that: The prey searching operation includes: The prey search operation is performed using formula (10), and the specific formula is: Among them, X rand Represents a random position vector in the current whale group.

Images