Cloud platform load prediction method and system
By extending multi-dimensional decomposition and adaptive hybrid social optimization algorithm to optimize the BiGRU-BLS model, the problem that traditional cloud platform load prediction methods are difficult to capture dynamic load data is solved, and prediction accuracy and model optimization performance are improved.
Patent Information
- Application Number
- CN202411119366.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-15
- Publication Date
- 2024-11-19
AI Technical Summary
Traditional cloud platform load prediction methods are difficult to capture the potential connections in dynamic and volatile load data, resulting in the impact of model prediction accuracy and the inability to predict future cloud platform load conditions quickly and accurately.
Extended multidimensional decomposition (EMD) is used to decompose multidimensional load data into stationary modal components, and the key parameters of BiGRU-BLS model are optimized using adaptive hybrid social optimization (AJSO) algorithm, and the optimized model is carried out by inputting modal components. Inverse normalization to obtain prediction results.
The accuracy of cloud platform load prediction and the convergence speed of the model are improved, the search strategy is dynamically adjusted to adapt to different environmental conditions, and the optimization accuracy is enhanced.
Smart Images

Figure CN118982050A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of cloud platform load prediction, and more specifically to a cloud platform load prediction method and system. Background Art
[0002] With the widespread application of cloud computing, more and more enterprises and organizations deploy their businesses and applications on cloud platforms, which leads to huge challenges in resource management and scheduling for cloud service providers. Effective load forecasting can identify changes in resource demand in advance, so as to automatically expand resources during peak periods and shrink resources during trough periods, thus achieving elastic resource management.
[0003] Traditional cloud platform load forecasting methods, such as time series analysis and regression analysis, usually assume that the machine load data recorded in the cloud platform has a certain fixed pattern or regularity. In the increasingly complex cloud platform service environment, traditional load forecasting methods also find it difficult to capture the potential connections in dynamic and volatile load data, and are unable to quickly and accurately predict future cloud platform load conditions, resulting in a significant impact on the model prediction accuracy.
[0004] Therefore, how to accurately predict the load of a cloud platform is an urgent problem that technical personnel in this field need to solve. Summary of the invention
[0005] In view of this, the present invention provides a cloud platform load prediction method and system, which uses EMD to decompose multi-dimensional load data of the cloud platform into several relatively stable modal components; uses AJSO to optimize the key parameters in the BiGRU-BLS prediction model to improve the prediction accuracy of the model; inputs the modal components into the AJSO-BiGRU-BLS model, and inversely normalizes the obtained prediction values to obtain the final prediction results.
[0006] In order to achieve the above object, the present invention adopts the following technical solution:
[0007] A cloud platform load prediction method, comprising:
[0008] Acquire multidimensional load data, and establish a sample set according to the multidimensional load data;
[0009] Decomposing the multidimensional load data in the sample set to obtain a number of modal components;
[0010] Pre-establishing a cloud platform load prediction model, and optimizing the cloud platform load prediction model through the AJSO algorithm;
[0011] The plurality of modal components are input into the optimized cloud platform load prediction model to obtain a cloud platform load data prediction value, and the cloud platform load data prediction value is inversely normalized to obtain a prediction result.
[0012] Preferably, before decomposing the multidimensional load data in the sample set, the method further includes: processing abnormal values of the multidimensional load data in the sample set by an interpolation method.
[0013] Preferably, after obtaining the plurality of modal components, it also includes: normalizing the plurality of modal components, and then dividing them into a training set and a test set using a sliding window method, training the cloud platform load prediction model through the training set, and testing the trained cloud platform load prediction model through the test set.
[0014] Preferably, the normalization process specifically includes: performing Min-Max normalization on a number of the modal components, and the specific formula is:
[0015]
[0016] In the formula, X norm is the normalized data value, X is the original data value, and X min is the minimum value of the data, X max is the maximum value of the data value.
[0017] Preferably, the sliding window method specifically includes:
[0018]
[0019] Among them, n is the fixed window size, [x t-n ,x t-n+1 ,…,x t-1 ] is the tnth load value combination, [x t ] is the load value at the tth moment, X and Y are the input data and real data of the model respectively.
[0020] Preferably, decomposing the multi-dimensional load data in the sample set specifically includes:
[0021] S1: Calculate the local maximum and minimum of the original signal y(t) to obtain the upper and lower envelopes m max (t), m min (t); the original signal is one-dimensional load data in the multi-dimensional load data;
[0022] S2: Calculate a local average line based on the upper and lower envelopes and interpolation method as the intrinsic mode function. The specific formula is:
[0023]
[0024] Among them, m max (t) is the upper envelope, m min (t) is the lower envelope, m(t) is the eigenmode function;
[0025] Subtracting the intrinsic mode function from the original signal can get the residual function. The specific formula is:
[0026] y 1 (t) = y(t) - m(t);
[0027] Among them, m(t) is the intrinsic mode function, y(t) is the original signal, and y 1 (t) is the residual function;
[0028] S3: Determine whether the remaining signal meets the conditions of the intrinsic mode function. The specific formula is:
[0029]
[0030] Among them, S D is the current screening value, y(t) is the original signal, y 1 (t) is the residual function, α is the screening threshold, if S D <α, the final intrinsic mode function is obtained, otherwise return to continue executing S2;
[0031] S4: Determine whether the current residual function is a constant or a monotonic function. If the condition is not met, return to continue executing S2. If the condition is met, save the current residual function, and the current residual function is the modal component.
[0032] Preferably, the establishing of the cloud platform load prediction model specifically includes:
[0033] Construct a bidirectional gated recurrent unit module. The bidirectional gated recurrent unit uses the forward gated recurrent unit and the reverse gated recurrent unit to jointly determine the output result. The hidden layer state value of the bidirectional gated recurrent unit at time t is obtained by the forward hidden layer state and the reverse hidden layer state The weighted summation is as follows:
[0034]
[0035] Among them, GRU(·) is a gated recurrent unit, is the forward hidden layer state at time r, is the reverse hidden layer state at time r, w r and v r are the forward weight parameter and the reverse weight parameter, respectively, b r is the bias parameter, hr is the hidden layer state value of the bidirectional gated recurrent unit at time r;
[0036] Build a width learning system module, build a mapping from input data D to n feature nodes, and the feature T of the kth node k The specific formula is as follows:
[0037]
[0038] Where D is the output data of the bidirectional gated recurrent unit module, and is the weight matrix and bias term randomly generated for the kth feature node, φ(·) is the activation function, k=1,...,n; n feature nodes are combined into a matrix T n =(T 1 ,…,T n ), and according to T n Generate the mth enhancement node. The specific formula is as follows:
[0039]
[0040] in, and is a randomly generated weight matrix and bias term, ξ(·) is an activation function, j=1,…,m; all enhanced nodes form a matrix H m =(H 1 ,...,H m ); let Y = (y 1 ,…,y n ) T , the mapping feature T n and H m Merge and output the prediction results. The specific formula is as follows:
[0041] Y=(T n ,H m )W m ;
[0042] Among them, Y is the predicted value, W m is the weight matrix of the output layer.
[0043] Preferably, the optimizing the cloud platform load prediction model by using the AJSO algorithm specifically includes:
[0044] Initialize the cloud platform load prediction model parameter range and the parameters in the AJSO algorithm;
[0045] Randomly initialize the position of the jellyfish, and set the position of each jellyfish to be z = [z α ,z β ,zγ ],z α is the number of feature layer node groups, z β is the number of nodes in the enhancement layer, z γ is the number of nodes in the group;
[0046] Calculate the fitness value of the jellyfish population as fit i , the specific calculation formula is:
[0047]
[0048] Among them, y i and are the true value and the predicted value of the model corresponding to the i-th time point, respectively, and N is the number of samples;
[0049] Record the fitness value fit corresponding to each group of jellyfish individuals i , find the best jellyfish location to save;
[0050] The calculation time control function c(t) is used to determine the movement mode of the jellyfish individual. The calculation formula is:
[0051]
[0052] In the formula, t is the number of iterations, iter is the upper limit of the number of iterations, and ε is a random number in (0,1);
[0053] When c(t)≥0.5, the jellyfish position is updated and iterated in an active motion mode, otherwise the jellyfish moves in a passive manner. The calculation formula for the jellyfish active motion is:
[0054] z i (t+1)=z i (t)+ε×σ×(U v -L v );
[0055] In the formula, z i (t) is the position of the i-th jellyfish at the t-th iteration, U v is the upper limit of the jellyfish search space, L v is the lower limit of the jellyfish search space, σ is the motion coefficient of the jellyfish individual;
[0056] The passive motion of jellyfish is calculated as:
[0057] V=z i (t+1)-z i (t);
[0058] Among them, z i (t+1) is the position of the i-th jellyfish at the t+1-th iteration;
[0059] If the number of iterations or the fitness value corresponding to the jellyfish is less than the set threshold, the optimization is stopped, the optimal parameter combination is output, and the optimal parameter combination is input into the cloud platform load prediction model to update the model parameters.
[0060] A cloud platform load prediction system, comprising:
[0061] A sample set establishment module, which acquires multi-dimensional load data and establishes a sample set according to the multi-dimensional load data;
[0062] A modal component acquisition module decomposes the multi-dimensional load data in the sample set to obtain a number of modal components;
[0063] A model building module, which pre-builds a cloud platform load prediction model and optimizes the cloud platform load prediction model through an AJSO algorithm;
[0064] The prediction module inputs the plurality of modal components into the optimized cloud platform load prediction model, obtains a cloud platform load data prediction value, performs inverse normalization processing on the cloud platform load data prediction value, and obtains a prediction result.
[0065] It can be seen from the above technical solutions that, compared with the prior art, the present invention discloses a cloud platform load prediction method and system, which has the following effects:
[0066] (1) The present invention uses EMD to decompose the high-dimensional cloud platform load data into several stationary modal components, which reduces the volatility of the data and improves the accuracy of the prediction model.
[0067] (2) The present invention adopts the AJSO algorithm. Compared with other optimization algorithms, the AJSO algorithm can dynamically adjust the search strategy and parameters to adapt to different optimization problems and environmental conditions without manual adjustment of parameters; and enhances its performance in terms of convergence speed and final optimization accuracy through an adaptive mechanism.
[0068] (3) The AJSO algorithm is used to optimize the key parameters of the BiGRU-BLS model, further improving the accuracy of the prediction model. BRIEF DESCRIPTION OF THE DRAWINGS
[0069] In order to more clearly illustrate the embodiments of the present invention 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, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.
[0070] Figure 1 A diagram of the method steps provided by the present invention;
[0071] Figure 2 Flowchart of key parameters of the BiGRU-BLS model optimized by the AJSO algorithm provided by the present invention. DETAILED DESCRIPTION
[0072] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0073] The embodiment of the present invention discloses a cloud platform load prediction method, such as Figure 1 As shown, including:
[0074] Acquire multidimensional load data, and establish a sample set according to the multidimensional load data;
[0075] Decompose the multidimensional load data in the sample set to obtain several modal components;
[0076] Establish a cloud platform load prediction model in advance and optimize it through the AJSO algorithm;
[0077] Several modal components are input into the optimized cloud platform load prediction model to obtain the cloud platform load data prediction value, and the cloud platform load data prediction value is inverse normalized to obtain the prediction result.
[0078] In a specific embodiment, the multi-dimensional load data is obtained by recording the load data of the machine in the container cloud platform within a time period T, and the data mainly includes timestamp, CPU utilization, memory utilization and I / O utilization;
[0079] In a specific embodiment, before decomposing the multidimensional load data in the sample set, the method further includes: processing the abnormal values of the multidimensional load data in the sample set by an interpolation method.
[0080] The CPU utilization, memory utilization, and I / O utilization in the multi-dimensional load data that are not in the range of [0, 99] are processed using the interpolation method to normalize the values to between [0, 1].
[0081] In a specific embodiment, after obtaining a number of modal components, it also includes: normalizing the several modal components, and then using a sliding window method to divide them into a training set and a test set, training the cloud platform load prediction model through the training set, and testing the trained cloud platform load prediction model through the test set.
[0082] In a specific embodiment, the normalization process specifically includes: performing Min-Max normalization on a number of modal components, and the specific formula is:
[0083]
[0084] Where, X norm is the normalized data value, X is the original data value, and X min is the minimum value of the data, X max is the maximum value of the data value.
[0085] In a specific embodiment, the sliding window method specifically includes:
[0086]
[0087] Among them, n is the fixed window size, [x t-n ,x t-n+1 ,…,x t-1 ] is the tnth load value combination, [x t ] is the load value at the tth moment, X and Y are the input data and real data of the model respectively. Among them, X is the training set and Y is the test set.
[0088] In a specific embodiment, when decomposing the multidimensional load data in the sample set, EMD is used for decomposition to obtain multiple stable modal components, specifically including:
[0089] S1: Calculate the local maximum and minimum of the original signal y(t) to obtain the upper and lower envelopes m max (t), m min (t); the original signal is one-dimensional load data in the multi-dimensional load data;
[0090] S2: Calculate a local average line based on the upper and lower envelopes and interpolation method as the intrinsic mode function. The specific formula is:
[0091]
[0092] Among them, m max (t) is the upper envelope, m min (t) is the lower envelope, m(t) is the eigenmode function;
[0093] Subtracting the intrinsic mode function from the original signal can get the residual function. The specific formula is:
[0094] y 1 (t) = y(t) - m(t);
[0095] Among them, m(t) is the intrinsic mode function, y(t) is the original signal, and y 1(t) is the residual function;
[0096] S3: Determine whether the remaining signal meets the conditions of the intrinsic mode function. The specific formula is:
[0097]
[0098] Among them, S D is the current screening value, y(t) is the original signal, y 1 (t) is the residual function, α is the screening threshold, if S D <α, the final intrinsic mode function is obtained, otherwise return to continue executing S2;
[0099] S4: Determine whether the current residual function is a constant or a monotonic function. If the condition is not met, return to continue executing S2. If the condition is met, save the current residual function, and the current residual function is the modal component.
[0100] In a specific embodiment, establishing a cloud platform load prediction model includes a bidirectional gated recurrent unit and a width learning system, specifically:
[0101] Construct a bidirectional gated recurrent unit module. The bidirectional gated recurrent unit uses the forward gated recurrent unit and the reverse gated recurrent unit to jointly determine the output result. The hidden layer state value of the bidirectional gated recurrent unit at time t is obtained by the forward hidden layer state and the reverse hidden layer state The weighted summation is as follows:
[0102]
[0103] Among them, GRU(·) is a gated recurrent unit, is the forward hidden layer state at time r, is the reverse hidden layer state at time r, w r and v r are the forward weight parameter and the reverse weight parameter, respectively, b r is the bias parameter, h r is the hidden layer state value of the bidirectional gated recurrent unit at time r;
[0104] Build a width learning system module, build a mapping from input data D to n feature nodes, and the feature T of the kth node k The specific formula is as follows:
[0105]
[0106] Where D is the output data of the bidirectional gated recurrent unit module, and is the weight matrix and bias term randomly generated for the kth feature node, φ(·) is the activation function, k=1,...,n; n feature nodes are combined into a matrix T n =(T 1 ,…,T n ), and according to T n Generate the mth enhancement node. The specific formula is as follows:
[0107]
[0108] in, and is a randomly generated weight matrix and bias term, ξ(·) is an activation function, j=1,...,m; all enhanced nodes form a matrix H m =(H 1 ,…,H m ); let Y = (y 1 ,…,y n ) T , the mapping feature T n and H m Merge and output the prediction results. The specific formula is as follows:
[0109] Y=(T n ,H m )W m ;
[0110] Among them, Y is the predicted value, W m is the weight matrix of the output layer.
[0111] In a specific embodiment, Figure 2 As shown in the figure, the AJSO algorithm is used to optimize the number of feature layer node groups, the number of enhancement layer node groups, and the number of nodes within the group in the BiGRU-BLS model (cloud platform load prediction model). The optimization of the cloud platform load prediction model by the AJSO algorithm specifically includes:
[0112] Initialize parameters such as the number of jellyfish populations, and determine the search range of key parameters such as the number of feature layer node groups, the number of enhancement layer node groups, and the number of nodes within a group;
[0113] Randomly initialize the position of the jellyfish, and set the position of each jellyfish to be z = [z α ,z β ,z γ ],z α is the number of feature layer node groups, z β is the number of nodes in the enhancement layer, z γ is the number of nodes in the group;
[0114] Calculate the fitness value of the jellyfish population as fit i, the specific calculation formula is:
[0115]
[0116] Among them, y i and are the true value and the predicted value of the model corresponding to the i-th time point, respectively, and N is the number of samples;
[0117] Record the fitness value fit corresponding to each group of jellyfish individuals i , find the best jellyfish location to save;
[0118] The calculation time control function c(t) is used to determine the movement mode of the jellyfish individual. The calculation formula is:
[0119]
[0120] In the formula, t is the number of iterations, iter is the upper limit of the number of iterations, and ε is a random number in (0,1);
[0121] When c(t)≥0.5, the jellyfish position is updated and iterated in an active motion mode, otherwise the jellyfish moves in a passive manner. The calculation formula for the jellyfish active motion is:
[0122] z i (t+1)=z i (t)+ε×σ×(U v -L v );
[0123] In the formula, z i (t) is the position of the i-th jellyfish at the t-th iteration, U v is the upper limit of the jellyfish search space, L v is the lower limit of the jellyfish search space, σ is the motion coefficient of the jellyfish individual;
[0124] The passive motion of jellyfish is calculated as:
[0125] V=z i (t+1)-z i (t);
[0126] Among them, z i (t+1) is the position of the i-th jellyfish at the t+1-th iteration;
[0127] If the number of iterations or the fitness value corresponding to the jellyfish is less than the set threshold, the optimization is stopped, the optimal parameter combination is output, and the optimal parameter combination is input into the cloud platform load prediction model to update the model parameters.
[0128] A cloud platform load prediction system, comprising:
[0129] A sample set establishment module obtains multi-dimensional load data and establishes a sample set according to the multi-dimensional load data;
[0130] The modal component acquisition module decomposes the multi-dimensional load data in the sample set to obtain several modal components;
[0131] Model building module, which pre-builds the cloud platform load prediction model and optimizes the cloud platform load prediction model through the AJSO algorithm;
[0132] The prediction module inputs several modal components into the optimized cloud platform load prediction model, obtains the cloud platform load data prediction value, performs inverse normalization processing on the cloud platform load data prediction value, and obtains the prediction result.
[0133] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.
[0134] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A cloud platform load prediction method, characterized in that: include: Acquire multidimensional load data, and establish a sample set according to the multidimensional load data; Decomposing the multidimensional load data in the sample set to obtain a number of modal components; Pre-establishing a cloud platform load prediction model, and optimizing the cloud platform load prediction model through the AJSO algorithm; The plurality of modal components are input into the optimized cloud platform load prediction model to obtain a cloud platform load data prediction value, and the cloud platform load data prediction value is inversely normalized to obtain a prediction result.
2. A cloud platform load prediction method according to claim 1, characterized in that: Before decomposing the multidimensional load data in the sample set, the method further includes: processing abnormal values of the multidimensional load data in the sample set by an interpolation method.
3. A cloud platform load prediction method according to claim 2, characterized in that: After the abnormal values of the multidimensional load data in the sample set are processed by the interpolation method, and before the multidimensional load data in the sample set is decomposed, it also includes: normalizing the multidimensional load data after the interpolation method, and then using the sliding window method to divide it into a training set and a test set, training the cloud platform load prediction model through the training set, and testing the trained cloud platform load prediction model through the test set.
4. A cloud platform load prediction method according to claim 3, characterized in that: The normalization of the multi-dimensional load data after the interpolation method processing specifically includes adopting Min-Max normalization, and the specific formula is: Where, X norm is the normalized data value, X origin is the original data value, X min is the minimum value of the data, X max is the maximum value of the data value.
5. A cloud platform load prediction method according to claim 3, characterized in that: The sliding window method specifically includes: Among them, n is the fixed window size, [x t-n ,x t-n+1 ,…,x t-1 ] is the tnth load value combination, [x t ] is the load value at the tth moment, X and Y are the input data and real data of the model respectively.
6. A cloud platform load prediction method according to claim 1, characterized in that: Decomposing the multidimensional load data in the sample set specifically includes: S1: Calculate the local maximum and minimum of the original signal y(t) to obtain the upper and lower envelopes m max (t), m min (t); the original signal is one-dimensional load data in the multi-dimensional load data; S2: Calculate a local average line based on the upper and lower envelopes and interpolation method as the intrinsic mode function. The specific formula is: Among them, m max (t) is the upper envelope, m min (t) is the lower envelope, m(t) is the eigenmode function; Subtracting the intrinsic mode function from the original signal can get the residual function. The specific formula is: y1(t)=y(t)-m(t); Among them, m(t) is the intrinsic mode function, y(t) is the original signal, and y1(t) is the residual function; S3: Determine whether the remaining signal meets the conditions of the intrinsic mode function. The specific formula is: Among them, S D is the current screening value, y(t) is the original signal, y1(t) is the residual function, α is the screening threshold, if S D <α, the final intrinsic mode function is obtained, otherwise return to continue executing S2; S4: Determine whether the current residual function is a constant or a monotonic function. If the condition is not met, return to continue executing S2. If the condition is met, save the current residual function, and the current residual function is the modal component.
7. A cloud platform load prediction method according to claim 1, characterized in that: The establishment of the cloud platform load prediction model specifically includes: Construct a bidirectional gated recurrent unit module. The bidirectional gated recurrent unit uses the forward gated recurrent unit and the reverse gated recurrent unit to jointly determine the output result. The hidden layer state value of the bidirectional gated recurrent unit at time t is obtained by the forward hidden layer state and the reverse hidden layer state The weighted summation is as follows: Among them, GRU(·) is a gated recurrent unit, is the forward hidden layer state at time r, is the reverse hidden layer state at time r, w r and v r are the forward weight parameter and the reverse weight parameter, respectively, b r is the bias parameter, h r is the hidden layer state value of the bidirectional gated recurrent unit at time r; Build a width learning system module, build a mapping from input data D to n feature nodes, and the feature T of the kth node k The specific formula is as follows: Where D is the output data of the bidirectional gated recurrent unit module, and The weight matrix and bias term randomly generated for the kth feature node, is the activation function, k=1,...,n; n feature nodes are combined into a matrix T n =(T1,…,T n ), and according to T n Generate the mth enhancement node. The specific formula is as follows: in, and is a randomly generated weight matrix and bias term, ξ(·) is an activation function, j=1,…,m; all enhanced nodes form a matrix H m =(H1,…,H m ); Let Y = (y1,…,y n ) T , the mapping feature T n and H m Merge and output the prediction results. The specific formula is as follows: Y=(T n ,H m )W m ; Among them, Y is the predicted value, W m is the weight matrix of the output layer.
8. A cloud platform load prediction method according to claim 1, characterized in that: The optimization of the cloud platform load prediction model by the AJSO algorithm specifically includes: Initialize the cloud platform load prediction model parameter range and the parameters in the AJSO algorithm; Randomly initialize the position of the jellyfish, and set the position of each jellyfish to be z = [z α ,z β ,z γ ],z α is the number of feature layer node groups, z β is the number of nodes in the enhancement layer, z γ is the number of nodes in the group; Calculate the fitness value of the jellyfish population as fit i , the specific calculation formula is: Among them, y i and are the true value and the predicted value of the model corresponding to the i-th time point, respectively, and N is the number of samples; Record the fitness value fit corresponding to each group of jellyfish individuals i , find the best jellyfish location to save; The calculation time control function c(t) is used to determine the movement mode of the jellyfish individual. The calculation formula is: In the formula, t is the number of iterations, iter is the upper limit of the number of iterations, and ε is a random number in (0,1); When c(t)≥0.5, the jellyfish position is updated and iterated in an active motion mode, otherwise the jellyfish moves in a passive manner. The calculation formula for the jellyfish active motion is: z i (t+1)=z i (t)+ε×σ×(U v -L v ); In the formula, z i (t) is the position of the i-th jellyfish at the t-th iteration, U v is the upper limit of the jellyfish search space, L v is the lower limit of the jellyfish search space, σ is the motion coefficient of the jellyfish individual; The passive motion of jellyfish is calculated as: V=z i (t+1)-z i (t); Among them, z i (t+1) is the position of the i-th jellyfish at the t+1-th iteration; If the number of iterations or the fitness value corresponding to the jellyfish is less than the set threshold, the optimization is stopped, the optimal parameter combination is output, and the optimal parameter combination is input into the cloud platform load prediction model to update the model parameters.
9. A cloud platform load prediction system, applying a cloud platform load prediction method according to any one of claims 1 to 8, characterized in that: include: A sample set establishment module, which acquires multi-dimensional load data and establishes a sample set according to the multi-dimensional load data; A modal component acquisition module decomposes the multi-dimensional load data in the sample set to obtain a number of modal components; A model building module, which pre-builds a cloud platform load prediction model and optimizes the cloud platform load prediction model through an AJSO algorithm; The prediction module inputs the plurality of modal components into the optimized cloud platform load prediction model, obtains a cloud platform load data prediction value, performs inverse normalization processing on the cloud platform load data prediction value, and obtains a prediction result.
Citation Information
Patent Citations
Shaft system thermal error modeling method and thermal error compensation system based on SLSTM neural network
CN111240268A
System load prediction method integrating isolated forest and long-term and short-term memory network
CN111738520A
Short-term load prediction method and device based on neural network combination model, and medium
CN113239624A
VMD-BiGRU-based short-term load prediction method
CN113902206A
Cloud production scheduling method based on BLS and LSTM and system thereof
CN113988630A
