Server performance prediction method fusing multi-time scale features
Through the combination of empirical modal decomposition method and two-way long and short-term memory network, the problem of insufficient server performance prediction accuracy and stability in the existing technology is solved, and the accurate prediction and optimization of the server's operating status is realized, and the fault repair cost is reduced. , improves the user experience.
Patent Information
- Application Number
- CN202411347112.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-26
- Publication Date
- 2025-01-10
AI Technical Summary
The existing server performance prediction methods have limited prediction accuracy and generalization capabilities in complex and variable actual operating environments, and require a lot of prior knowledge and manual adjustment of parameters, resulting in cumbersome prediction process and error-prone.
Empirical modal decomposition (EMD) is used to decompose the historical data of the server's operating state by multi-time scale feature, and a two-way long and short-term memory network (BiLSTM) prediction model is built, and multi-time scale features are fused to improve prediction accuracy and stability.
By fully understanding the operating status of the server, the accuracy and stability of the prediction model are improved, real-time operating status information is provided, and optimization strategies are helped to avoid system crashes and data loss, reducing the cost of fault repair and resource waste, and increasing the cost of User experience.
Smart Images
Figure CN119292888A_ABST
Abstract
Description
Technical Field
[0001] The present invention discloses a server performance prediction method integrating multi-time scale features, and relates to the field of server performance prediction. Background Art
[0002] With the rapid development of the information society, servers are core devices for data storage, processing and transmission. The stability and efficiency of their performance are crucial to the operation of the entire network system. However, the performance of servers is affected by many factors, such as hardware configuration, software environment, workload, etc., which makes server performance prediction a complex and challenging task. However, existing server performance prediction methods are often based on empirical formulas or statistical models. These methods may have certain prediction capabilities in simple scenarios, but their prediction accuracy and generalization capabilities are often limited when faced with complex and changeable actual operating environments. In addition, existing methods usually require a lot of prior knowledge and manual adjustment of parameters, which makes the prediction process cumbersome and error-prone. Summary of the invention
[0003] In view of the problems of the prior art, the present invention provides a server performance prediction method integrating multi-time scale features, which can fully capture the multi-time scale feature information of server performance data, comprehensively understand the operating status of the server from both global and local perspectives, improve the accuracy and stability of the prediction model, ensure the safe and stable operation of the server, and facilitate operation and maintenance personnel to obtain real-time server operating status information and formulate optimization strategies.
[0004] The specific scheme proposed by the present invention is:
[0005] The present invention provides a server performance prediction method integrating multi-time scale features, comprising:
[0006] Step 1: Collect historical data of server operation status;
[0007] Step 2: Preprocess the historical data of the server operation status.
[0008] Step 3: Use the empirical mode decomposition method EMD to perform multi-time scale feature decomposition on the preprocessed data to obtain features of different time scales;
[0009] Step 4: Construct a bidirectional long short-term memory network BiLSTM. BiLSTM consists of two LSTM networks, one for forward processing of the input sequence and the other for backward processing of the input sequence.
[0010] Step 5: Based on the bidirectional long short-term memory network BiLSTM, a BiLSTM prediction model that integrates multi-time scale features is constructed, the multi-time scale feature data is input into the BiLSTM prediction model for prediction, and the prediction results are integrated to obtain the final server performance prediction result.
[0011] Furthermore, in step 2 of the server performance prediction method integrating multi-time scale features, data preprocessing is performed, including:
[0012] The historical data of the server operation status is filtered, outliers are removed, and normalized. The normalization process is as follows:
[0013]
[0014] Among them, x represents the historical data of the server operation status, x' represents the normalized data, and x max Indicates the maximum value of the data, x min Indicates the minimum value of the data.
[0015] Furthermore, in step 3 of the server performance prediction method integrating multi-time scale features, the empirical mode decomposition method EMD is used to perform multi-time scale feature decomposition on the preprocessed data, including:
[0016] ① Obtain the maximum and minimum points of the preprocessed data X(t), fit the extreme points by curve interpolation method, and obtain the upper envelope X of this set of data max (t) and the lower envelope X min (t),
[0017] ②Calculate the average value of the upper and lower envelopes:
[0018]
[0019] ③ Compare the preprocessed data X(t) with the average envelope m 1 (t) is subtracted to obtain the remaining data d 1 (t),
[0020] ④ For d 1 (t) The data continues to be processed using steps ①-③ until the screening threshold SD is less than the threshold, and the first intrinsic mode function IMF is obtained. 1 ,
[0021] ⑤ Calculate the difference between the preprocessed data X(t) and IMF1 to obtain the first-order residual r 1 (t), and then according to steps ①-④, 1 (t) is processed and repeated N times to obtain N intrinsic mode functions IMF1 -IMF N and the final residual r N (t), then the expression of the preprocessed data after EMD decomposition is:
[0022]
[0023] According to formula (3), the characteristics of different time scales are obtained.
[0024] Furthermore, in step 4 of the server performance prediction method integrating multi-time scale features, the final hidden layer output h is determined by weighted summation through the bidirectional long short-term memory network BiLSTM. i :
[0025]
[0026] Among them, α and β are learnable parameters that can be output through the hidden layer h i A BiLSTM prediction model is established between the target output y.
[0027] Furthermore, in step 5 of the server performance prediction method integrating multi-time scale features, the multi-time scale features are used to divide the training set and the test set, and the training set is used to predict each IMF 1-n Train the corresponding BiLSTM separately 1-n Prediction model, the output of each BiLSTM is the estimated value of the target performance result y Finally, n estimated results about the target server performance are obtained, and weighted fusion is performed to obtain the final prediction output result, and the root mean square error RMSE is used as the evaluation indicator.
[0028] Furthermore, the evaluation index in step 5 of the server performance prediction method integrating multi-time scale features is as follows:
[0029]
[0030] Where N represents the number of prediction samples, y i Represents the output true value, Represents the predicted value.
[0031] Furthermore, in step 5 of the server performance prediction method integrating multi-time scale features, the weight of each BiLSTM sub-model during weighted fusion is determined according to the RMSE error value predicted during model training. The specific weight calculation method is as follows:
[0032]
[0033] Among them, w irepresents the weight of the i-th BiLSTM sub-model, RMSE i Represents the prediction RMSE error of the i-th BiLSTM sub-model, and the final prediction output of the entire model As shown below:
[0034]
[0035] in Represents the predicted output value of the i-th BiLSTM sub-model.
[0036] The present invention also provides a server performance prediction device integrating multi-time scale features, including a collection module, a preprocessing module, a feature decomposition module,
[0037] Step 1: Collect historical data of server operation status;
[0038] Step 2: Preprocess the historical data of the server operation status.
[0039] Step 3: Use the empirical mode decomposition method EMD to perform multi-time scale feature decomposition on the preprocessed data to obtain features of different time scales;
[0040] Step 4: Construct a bidirectional long short-term memory network BiLSTM. BiLSTM consists of two LSTM networks, one for forward processing of the input sequence and the other for backward processing of the input sequence.
[0041] Step 5: Based on the bidirectional long short-term memory network BiLSTM, a BiLSTM prediction model that integrates multi-time scale features is constructed, the multi-time scale feature data is input into the BiLSTM prediction model for prediction, and the prediction results are integrated to obtain the final server performance prediction result.
[0042] The present invention also provides an electronic device for server performance prediction integrating multi-time scale features, comprising: at least one memory and at least one processor;
[0043] The at least one memory is used to store a machine-readable program;
[0044] The at least one processor is used to call the machine-readable program to execute the server performance prediction method integrating multi-time scale features.
[0045] The present invention also provides a computer-readable medium, on which computer instructions are stored. When the computer instructions are executed by a processor, the processor executes the server performance prediction method that integrates multiple time scale features.
[0046] The benefits of the present invention are:
[0047] The present invention is used to realize accurate prediction of the server operation status. By using EMD to perform multi-time scale information decomposition on the server operation data, the prediction model can comprehensively learn the server operation status characteristics, improve the prediction accuracy of the model, and by establishing a prediction model based on BiLSTM, it can fully learn the dependencies between the server performance data, thereby obtaining a prediction model with excellent performance, providing powerful decision support for operation and maintenance personnel, avoiding the risks of system crash or data loss caused by insufficient performance, and reducing the additional costs caused by fault repair and data recovery. At the same time, it can also help operation and maintenance personnel to plan the purchase and deployment of servers more reasonably, avoiding waste of resources and waste of costs. Through accurate prediction and optimization of server performance, it can ensure that users can obtain better response speed and stability when using network services, and improve user satisfaction and loyalty. BRIEF DESCRIPTION OF THE DRAWINGS
[0048] Figure 1 It is a schematic flow chart of the method of the present invention.
[0049] Figure 2 This is a diagram of the long short-term memory network structure.
[0050] Figure 3 It is a diagram of the bidirectional long short-term memory network structure.
[0051] Figure 4 It is a schematic diagram of the output flow of the bidirectional long short-term memory network prediction model that integrates multi-time scale features. DETAILED DESCRIPTION
[0052] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments so that those skilled in the art can better understand the present invention and implement it, but the embodiments are not intended to limit the present invention.
[0053] Example 1
[0054] Combination Figure 1 The present invention provides a server performance prediction method integrating multi-time scale features, comprising:
[0055] Step 1: Collect historical data of the server's operating status. The historical data may include but is not limited to server hardware and software information such as CPU usage, memory usage, and disk usage.
[0056] Step 2: Preprocess the historical data of the server operation status, such as filtering information, removing outliers, and normalizing. The normalization process is as follows:
[0057]
[0058] Among them, x represents the historical data of the server operation status, x' represents the normalized data, and x max Indicates the maximum value of the data, x min Indicates the minimum value of the data.
[0059] Step 3: Use the empirical mode decomposition method (EMD) to decompose the preprocessed data into multiple time scales to obtain features at different time scales.
[0060] The empirical mode decomposition method EMD is used to perform multi-time scale feature decomposition on the preprocessed data, and multiple intrinsic mode functions IMF and a residual can be obtained. Different IMFs correspond to characteristics of different time scales.
[0061] The specific process may include:
[0062] ① Obtain the maximum and minimum points of the preprocessed data X(t), fit the extreme points by curve interpolation method, and obtain the upper envelope X of this set of data max (t) and the lower envelope X min (t),
[0063] ②Calculate the average value of the upper and lower envelopes:
[0064]
[0065] ③ Compare the preprocessed data X(t) with the average envelope m 1 (t) is subtracted to obtain the remaining data d 1 (t),
[0066] ④ For d 1 (t) The data is processed using steps ①-③ until the screening threshold SD is less than the threshold, which is usually 0.2-0.3, and the first intrinsic mode function IMF is obtained. 1 , you can refer to the formula:
[0067]
[0068] ⑤ Calculate the difference between the preprocessed data X(t) and IMF1 to obtain the first-order residual r 1 (t), and then according to steps ①-④, 1 (t) is processed and repeated N times to obtain N intrinsic mode functions IMF 1 -IMF N and the final residual r N (t), then the expression of the preprocessed data after EMD decomposition is:
[0069]
[0070] According to formula (3), the characteristics of different time scales are obtained.
[0071] Step 4: Construct a bidirectional long short-term memory network BiLSTM. BiLSTM consists of two LSTM networks, one for forward processing of the input sequence and the other for backward processing of the input sequence.
[0072] The long short-term memory network LSTM is mainly composed of a forget gate, an input gate, and an output gate. The gate control logic is introduced to achieve selective memory and forgetting of information. The forget gate determines which information needs to be discarded from the cell state, that is, how much information to retain:
[0073] f t =σ(W f ·[h t-1 , x t ]+b f ) Formula (4)
[0074] The input gate determines what new information needs to be stored in the cell state. It consists of two parts: the Sigmoid layer and the tanh layer. The Sigmoid layer determines what value needs to be updated, using the probability i t Represents; the tanh layer creates a candidate value vector Will be added to the cell state and used to update subsequent cell states:
[0075] i t =σ(W i ·[h t-1 , x t ]+b i ) Formula (5)
[0076]
[0077] Update cell status:
[0078]
[0079] The output gate controls what information is output from the cell state, and the output cell state C is determined by the Sigmoid layer. t Then the cell state is passed through the tanh layer to make its value between -1 and 1, and then multiplied by the output of the Sigmoid layer to get the final output h i :
[0080] σ t =σ(W o ·[h t-1 , x t ]+b o ) Formula (8)
[0081] ht =o t ☉tanh(C t ) Formula (9)
[0082] BiLSTM consists of two LSTM networks, one that processes the input sequence forward and the other that processes it backward. The two LSTM networks are trained independently, and then their outputs are combined at some point, usually concatenated or summed, to produce the final output. This structure allows the network to consider both forward and backward related information of the input sequence when producing the output, such as Figure 3 As shown, the use of BiLSTM For forward propagation, use Backpropagation is performed to learn the possible relationship between past and future data to solve the long-term forgetting problem of the model.
[0083] The final hidden layer output h is determined by weighted summation through the bidirectional long short-term memory network BiLSTM i :
[0084]
[0085] Among them, α and β are learnable parameters that can be output through the hidden layer h i A BiLSTM prediction model is established between the target output y.
[0086] Step 5: Based on the bidirectional long short-term memory network BiLSTM, a BiLSTM prediction model that integrates multi-time scale features is constructed, the multi-time scale feature data is input into the BiLSTM prediction model for prediction, and the prediction results are integrated to obtain the final server performance prediction result.
[0087] The multi-time scale features are used to divide the training set and the test set. The training set is used for each IMF 1-n Train the corresponding BiLSTM separately 1-n Prediction model, the output of each BiLSTM is the estimated value of the target performance result y Finally, n estimated results about the target server performance are obtained, and weighted fusion is performed to obtain the final prediction output result, and the root mean square error RMSE is used as the evaluation indicator.
[0088] The evaluation index can be further expressed as follows:
[0089]
[0090] Where N represents the number of prediction samples, y i Represents the output true value, Represents the predicted value.
[0091] The weight of each BiLSTM sub-model during weighted fusion is determined according to the RMSE error value predicted during model training. The specific weight calculation method is as follows:
[0092]
[0093] Among them, w i represents the weight of the i-th BiLSTM sub-model, RMSE i Represents the prediction RMSE error of the i-th BiLSTM sub-model, and the final prediction output of the entire model As shown below:
[0094]
[0095] in Represents the predicted output value of the i-th BiLSTM sub-model. The test set is then used to test the performance of the established prediction model, and the root mean square error RMSE is also used as the evaluation indicator of the model performance.
[0096] The method of the present invention improves the accuracy and stability of server performance prediction, can fully capture the multi-time scale feature information of server performance data, comprehensively understand the operating status of the server from both global and local perspectives, improve the accuracy and stability of the prediction model, ensure the safe and stable operation of the server, and facilitate operation and maintenance personnel to obtain real-time server operating status information and formulate optimization strategies.
[0097] Example 2
[0098] The present invention also provides a server performance prediction device integrating multi-time scale features, including a collection module, a preprocessing module, a feature decomposition module,
[0099] Step 1: Collect historical data of server operation status;
[0100] Step 2: Preprocess the historical data of the server operation status.
[0101] Step 3: Use the empirical mode decomposition method EMD to perform multi-time scale feature decomposition on the preprocessed data to obtain features of different time scales;
[0102] Step 4: Construct a bidirectional long short-term memory network BiLSTM. BiLSTM consists of two LSTM networks, one for forward processing of the input sequence and the other for backward processing of the input sequence.
[0103] Step 5: Based on the bidirectional long short-term memory network BiLSTM, a BiLSTM prediction model that integrates multi-time scale features is constructed, the multi-time scale feature data is input into the BiLSTM prediction model for prediction, and the prediction results are integrated to obtain the final server performance prediction result.
[0104] The information interaction, execution process and other contents between the modules of the above-mentioned device are based on the same concept as the embodiment of the method of the present invention. For specific contents, please refer to the description in the embodiment of the method of the present invention, and no further description is given here.
[0105] Similarly, the device of the present invention is used to achieve accurate prediction of the server operation status. By using EMD to perform multi-time scale information decomposition on the server operation data, the prediction model can fully learn the server operation status characteristics and improve the prediction accuracy of the model. By establishing a prediction model based on BiLSTM, the dependencies between server performance data can be fully learned, thereby obtaining a prediction model with excellent performance, providing powerful decision-making support for operation and maintenance personnel, avoiding risks such as system crashes or data loss due to insufficient performance, and reducing additional costs caused by fault repair and data recovery. At the same time, it can also help operation and maintenance personnel to plan the purchase and deployment of servers more reasonably, avoiding waste of resources and costs. Through accurate prediction and optimization of server performance, it can ensure that users can obtain better response speed and stability when using network services, thereby improving user satisfaction and loyalty.
[0106] Example 3
[0107] The present invention also provides an electronic device for server performance prediction integrating multi-time scale features, comprising: at least one memory and at least one processor;
[0108] The at least one memory is used to store a machine-readable program;
[0109] The at least one processor is used to call the machine-readable program to execute the server performance prediction method integrating multi-time scale features.
[0110] The information interaction of the processor in the above-mentioned device, the execution process of the readable program, etc., are based on the same concept as the embodiment of the method of the present invention. The specific contents can be found in the description of the embodiment of the method of the present invention and will not be repeated here.
[0111] Similarly, the device of the present invention is used to achieve accurate prediction of the server operation status. By using EMD to perform multi-time scale information decomposition on the server operation data, the prediction model can fully learn the server operation status characteristics and improve the prediction accuracy of the model. By establishing a prediction model based on BiLSTM, the dependencies between server performance data can be fully learned, thereby obtaining a prediction model with excellent performance, providing powerful decision support for operation and maintenance personnel, avoiding risks such as system crashes or data loss due to insufficient performance, and reducing additional costs caused by fault repair and data recovery. At the same time, it can also help operation and maintenance personnel to plan the purchase and deployment of servers more reasonably, avoiding waste of resources and costs. Through accurate prediction and optimization of server performance, it can ensure that users can obtain better response speed and stability when using network services, thereby improving user satisfaction and loyalty.
[0112] Example 4
[0113] The present invention also provides a computer-readable medium, on which a computer instruction is stored, and when the computer instruction is executed by a processor, the processor executes the server performance prediction method integrating multi-time scale features. Specifically, a system or device equipped with a storage medium can be provided, on which a software program code implementing the functions of any of the above embodiments is stored, and a computer (or CPU or MPU) of the system or device reads and executes the program code stored in the storage medium.
[0114] In this case, the program code itself read from the storage medium can realize the function of any one of the above-mentioned embodiments, and thus the program code and the storage medium storing the program code constitute a part of the present invention.
[0115] The storage medium embodiments for providing the program code include a floppy disk, a hard disk, a magneto-optical disk, an optical disk (such as CD-ROM, CD-R, CD-RW, DVD-ROM, DVD-RAM, DVD-RW, DVD+RW), a magnetic tape, a non-volatile memory card, and a ROM. Alternatively, the program code can be downloaded from a server computer by a communication network.
[0116] In addition, it should be clear that the functions of any of the above embodiments can be implemented not only by executing the program code read by the computer, but also by enabling an operating system operating on the computer to complete part or all of the actual operations based on instructions from the program code.
[0117] In addition, it can be understood that the program code read from the storage medium is written to a memory provided in an expansion board inserted into the computer or written to a memory provided in an expansion unit connected to the computer, and then based on the instructions of the program code, a CPU installed on the expansion board or the expansion unit is enabled to perform part or all of the actual operations, thereby realizing the functions of any of the above-mentioned embodiments.
[0118] The above-described embodiments are only preferred embodiments for fully illustrating the present invention, and the protection scope of the present invention is not limited thereto. Equivalent substitutions or changes made by those skilled in the art based on the present invention are within the protection scope of the present invention. The protection scope of the present invention shall be subject to the claims.
Claims
1. A server performance prediction method integrating multi-time scale features, characterized by include: Step 1: Collect historical data of server operation status; Step 2: Preprocess the historical data of the server operation status. Step 3: Use the empirical mode decomposition method EMD to perform multi-time scale feature decomposition on the preprocessed data to obtain features of different time scales; Step 4: Construct a bidirectional long short-term memory network BiLSTM. BiLSTM consists of two LSTM networks, one for forward processing of the input sequence and the other for backward processing of the input sequence. Step 5: Based on the bidirectional long short-term memory network BiLSTM, a BiLSTM prediction model that integrates multi-time scale features is constructed, the multi-time scale feature data is input into the BiLSTM prediction model for prediction, and the prediction results are integrated to obtain the final server performance prediction result.
2. According to the server performance prediction method integrating multi-time scale features according to claim 1, it is characterized by In step 2, data preprocessing is performed, including: The historical data of the server operation status is filtered, outliers are removed, and normalized. The normalization process is as follows: Among them, x represents the historical data of the server operation status, x' represents the normalized data, and x max Indicates the maximum value of the data, x min Indicates the minimum value of the data.
3. The server performance prediction method integrating multi-time scale features according to claim 1 is characterized in that In step 3, the empirical mode decomposition method EMD is used to perform multi-time scale feature decomposition on the preprocessed data, including: ① Obtain the maximum and minimum points of the preprocessed data X(t), fit the extreme points by curve interpolation method, and obtain the upper envelope X of this set of data max (t) and the lower envelope X min (t), ②Calculate the average value of the upper and lower envelopes: ③ Subtract the preprocessed data X(t) from the average envelope m1(t) to obtain the remaining data d1(t), ④ Continue to process the d1(t) data using steps ①-③ until the screening threshold SD is less than the threshold, and then obtain the first intrinsic mode function IMF1. ⑤ Calculate the difference between the preprocessed data X(t) and IMF1 to obtain the first-order residual r1(t). Process r1(t) according to steps ①-④. After repeating N times, N intrinsic mode functions IMF1-IMF are obtained. N and the final residual r N (t), then the expression of the preprocessed data after EMD decomposition is: According to formula (3), the characteristics of different time scales are obtained.
4. According to the server performance prediction method integrating multi-time scale features of claim 1, it is characterized in that in step 4, the final hidden layer output h is determined by weighted summation through a bidirectional long short-term memory network BiLSTM i : Among them, α and β are learnable parameters that can be output through the hidden layer h i A BiLSTM prediction model is established between the target output y.
5. The server performance prediction method integrating multi-time scale features according to claim 1 is characterized in that In step 5, the multi-time scale features are used to divide the training set and the test set. The training set is used to calculate each IMF. 1-n Train the corresponding BiLSTM separately 1-n Prediction model, the output of each BiLSTM is the estimated value of the target performance result y Finally, n estimated results about the target server performance are obtained, and weighted fusion is performed to obtain the final prediction output result, and the root mean square error RMSE is used as the evaluation indicator.
6. The server performance prediction method integrating multi-time scale features according to claim 5 is characterized in that The evaluation indicators in step 5 are as follows: Where N represents the number of prediction samples, y i Represents the output true value, Represents the predicted value.
7. The server performance prediction method integrating multi-time scale features according to claim 6 is characterized by: The weight of each BiLSTM sub-model during weighted fusion in step 5 is determined according to the RMSE error value predicted during model training. The specific weight calculation method is as follows: Among them, w i represents the weight of the i-th BiLSTM sub-model, RMSE i Represents the prediction RMSE error of the i-th BiLSTM sub-model, and the final prediction output of the entire model As shown below: in Represents the predicted output value of the i-th BiLSTM sub-model.
8. A server performance prediction device integrating multi-time scale features, characterized in that Including collection module, preprocessing module, feature decomposition module, Step 1: Collect historical data of server operation status; Step 2: Preprocess the historical data of the server operation status. Step 3: Use the empirical mode decomposition method EMD to perform multi-time scale feature decomposition on the preprocessed data to obtain features of different time scales; Step 4: Construct a bidirectional long short-term memory network BiLSTM. BiLSTM consists of two LSTM networks, one for forward processing of the input sequence and the other for backward processing of the input sequence. Step 5: Based on the bidirectional long short-term memory network BiLSTM, a BiLSTM prediction model that integrates multi-time scale features is constructed, the multi-time scale feature data is input into the BiLSTM prediction model for prediction, and the prediction results are integrated to obtain the final server performance prediction result.
9. An electronic device for predicting server performance by integrating multi-time scale features, characterized in that include: at least one memory and at least one processor; The at least one memory is used to store a machine-readable program; The at least one processor is used to call the machine-readable program to execute a server performance prediction method integrating multi-time scale features as described in any one of claims 1 to 7.
10. A computer readable medium, characterized in that The computer-readable medium stores computer instructions, and when the computer instructions are executed by a processor, the processor executes the server performance prediction method integrating multi-time scale features as described in any one of claims 1 to 7.