A deep learning-based system access volume prediction method, medium and device

Abstract

The application relates to a deep learning-based system access volume prediction method, medium and equipment. After loading and processing historical data, the related parameters of a genetic algorithm are set, the genetic algorithm is used to optimize a variational mode decomposition algorithm, the processed historical data is decomposed by the optimized variational mode decomposition algorithm, the optimal decomposition result of the current historical data is obtained, the multiple intrinsic mode components obtained by the variational mode decomposition algorithm are normalized, each function after processing is input into a multi-scale CNN neural network, features are extracted, the features are input into a GRU network as parameters, and the model is trained until stable; historical data before a time to be predicted is decomposed by the optimized variational mode decomposition algorithm, the decomposition result is input into the trained model, and the result output by the model is used as a system access volume prediction value. The application can accurately and quickly predict the access volume of the next moment through the current moment access volume.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electric digital data processing, and in particular to a deep learning-based system access volume prediction method based on genetic algorithm-optimized variational modal decomposition parameters, a multi-scale convolutional neural network (CNN) and a gated recurrent unit (GRU), a medium and equipment. BACKGROUND

[0002] With the rapid development of the Internet and the improvement of people's economic level, Internet services are deeply welcomed by people due to their real-time, interactive, convenient and other characteristics. Today, Internet services have become an indispensable part of people's daily life. Correspondingly, the widespread use of the Internet has also promoted the rapid development of Internet technology.

[0003] Due to the popularity of Internet devices, users are accessing service providers at all times to obtain the information they need. The user's access is not always stable and unchanging, but fluctuates over time with certain regularity. Therefore, it is crucial to ensure that user access requests can be quickly and efficiently returned during access peak and access valley periods. Therefore, the system of the service provider must be able to quickly and efficiently process and return user requests at any time.

[0004] The ability of the system to quickly and efficiently process and return user requests is directly related to the resources of the system, and the access volume of the system is also directly related to the available resources of the system. An increase in the user's request volume will increase the resources consumed by the system and reduce the available resources. If the available resources are insufficient, the system's ability to process requests will decrease, and the user will not be able to immediately obtain the response to the request. Therefore, it is necessary to ensure that the available resources of the system are always sufficient. However, if the resources of the system remain unchanged during the access peak and access valley periods, it will result in resource waste or excessive system load and insufficient resources, which will further prevent the request from being processed normally.

[0005] In order to ensure the normal operation of the system and the full and reasonable use of system resources, it is a good way to make corresponding adjustments by predicting the change in system access volume in the next stage. Therefore, access volume prediction of the system is of great value. If the historical access volume cannot be correctly and quickly calculated and the access volume cannot be predicted, it will be unfavorable for the full and reasonable use of system resources. SUMMARY

[0006] The present application solves the problems in the prior art and provides an optimized deep learning-based system access volume prediction method, medium and equipment.

[0007] The technical solution adopted by the present application is a deep learning-based system access volume prediction method, which comprises the following steps:

[0008] Step 1: load historical data and process;

[0009] Step 2: based on historical data, set the correlation parameters of genetic algorithm;

[0010] Step 3: adopt genetic algorithm to optimize variational mode decomposition algorithm, and use the optimized variational mode decomposition algorithm to perform mode decomposition on the processed historical data to obtain the optimal decomposition result of the current historical data;

[0011] Step 4: normalize the multiple intrinsic mode components obtained by the variational mode decomposition algorithm;

[0012] Step 5: input each processed function into a multi-scale CNN neural network, extract features, and input the features as parameters into a GRU network, and train the model until it is stable;

[0013] Step 6: use the optimized variational mode decomposition algorithm to perform variational mode decomposition on the historical data before the time to be predicted, input the decomposition result into the trained model in step 5, and add the output results of the model to obtain the predicted value of the system access volume.

[0014] Preferably, in step 1, the number of times the monitored API is called per day is recorded, and the access volume is increased by one each time the API is called. The daily access volume is stored daily as historical data. The historical data of the access volume is read as a data set for training the model. If there is missing data for one or more days, the historical data of the associated date is used to complete the missing data according to a preset rule.

[0015] Preferably, in step 2, the correlation parameters of the genetic algorithm include the number of iterations of the genetic algorithm, the size of the population, the crossover probability in the population, the mutation probability, the variation range of the parameter a of the variational mode algorithm, and the variation range of K.

[0016] Preferably, step 3 includes the following steps:

[0017] Step 3.1: initialize the population;

[0018] Step 3.2: select the population;

[0019] Step 3.3: perform crossover operation on the population;

[0020] Step 3.4: perform mutation operation on the population;

[0021] Step 3.5: update the fitness of the individual, select the individual with the best fitness in the iteration process, and use the parameters of this individual as the parameters a and K in the variational mode decomposition algorithm to decompose the historical data to obtain K intrinsic mode functions.

[0022] Preferably, in steps 3.1 and 3.2, when initializing the population, randomly generate chromosomes, any chromosome represents an individual in the population, and all randomly generated chromosomes correspond to the entire population;

[0023] Each chromosome is composed of individual fitness and quadratic penalty factor a of the variational mode decomposition algorithm, and the number K of eigenmode functions; individuals are repeatedly generated and the fitness of each individual is calculated until the number of individuals meets the set size of the population, wherein the fitness function of the individual is the sample entropy of the individual;

[0024] The selection operation is performed on the population, and the roulette method is used to select individuals from the population to enter the next generation, the reciprocal of the fitness value of each individual is calculated, and the probability of each individual entering the next generation is the ratio of the fitness of the individual to the total fitness of the entire population; repeat until the number of selected individuals is equal to the set population size.

[0025] Preferably, in steps 3.3 and 3.4,

[0026] Two individuals are randomly selected for crossover operation, and whether to perform crossover is determined by the crossover probability; after determining the two individuals that need to be crossed, the corresponding position variables of the individuals are randomly selected for crossover operation, that is, the parameters at the same position in the two individuals are crossed using the arithmetic crossover method, and the arithmetic crossover linearly combines the parameters at the same position in the two individuals to generate a new individual;

[0027] The mutation operation is performed on the population, and whether the individual is mutated is determined by the mutation probability; the parameters in a certain individual are randomly mutated, and the value of the parameter at the position is re-assigned to a value in the parameter change range.

[0028] Preferably, in step 4, after normalization, each function is divided into two parts, one part is the input of the model, and the other part is the output of the model.

[0029] Preferably, in step 5, a one-dimensional convolution kernel of several different sizes is used to perform convolution at several scales on the input data, different size convolution kernels extract different feature values, and an equal-width convolution method is used to pad zeros at both ends of the features, so that the extracted feature dimensions are the same; the GRU network is composed of two layers of GRU units, two layers of DroupOut layers, and one layer of fully connected layers.

[0030] A computer-readable storage medium having a deep learning-based system access volume prediction program stored thereon, the program being executed by a processor to implement the deep learning-based system access volume prediction method.

[0031] A computer device comprises a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the program to access the system access amount prediction method based on deep learning.

[0032] The application provides an optimized system access amount prediction method based on deep learning, a medium and a device, after loading and processing historical data, setting the correlation parameters of the genetic algorithm, optimizing the variational mode decomposition algorithm by using the genetic algorithm, and using the optimized variational mode decomposition algorithm to perform mode decomposition on the processed historical data to obtain the optimal decomposition result of the current historical data, normalizing the multiple intrinsic mode components obtained by the variational mode decomposition algorithm, inputting each function after processing into a multi-scale CNN neural network, inputting the features into a GRU network as parameters after feature extraction, and training the model until it is stable; using the optimized variational mode decomposition algorithm to perform variational mode decomposition on the historical data before the time to be predicted, inputting the decomposition result into the trained model, and taking the result output by the model as the system access amount prediction value.

[0033] The application considers that the system access amount has regularity in a time period, so the data of the system access amount in the past is recorded and processed, the regularity between the historical data is found out by using the optimized variational mode decomposition algorithm, the model is constructed according to the regularity, and the future time is predicted, so that the access amount of the next time can be accurately and quickly predicted through the access amount of the current time. BRIEF DESCRIPTION OF DRAWINGS

[0034] Figure 1 The figure is a flowchart of the application;

[0035] Figure 2 The figure is the main flow of the variational mode decomposition algorithm optimized by the genetic algorithm in the application;

[0036] Figure 3 The figure is the structure diagram of the multi-scale CNN network in the application. DETAILED DESCRIPTION

[0037] The application will be further described in detail in combination with the embodiments, but the protection scope of the application is not limited thereto.

[0038] The application relates to a system access amount prediction method based on deep learning.

[0039] Step 1: loading historical data and processing;

[0040] In the step 1, the number of times of daily calling of the monitored API is recorded, the access amount is added by one each time the API is called, the daily access amount is stored by day as historical data; the historical data of the access amount is read as a data set of the training model, and if there is missing data in any day or days, the historical data of the associated date is used to complete the missing data according to a preset rule.

[0041] In the application, the monitored API refers to the API interface monitored by the system, and the total access amount is recorded in the disk of the computer every day.

[0042] In the application, the missing data is generally caused by system failure and the like, so that the system does not run and there is no access data, at this time, the preset rule is generally to sum the data of the three days before and after the missing data time (date) and the six days in total, and take the average value as the missing data value.

[0043] Step 2: setting the associated parameters of the genetic algorithm based on the historical data;

[0044] In the step 2, the associated parameters of the genetic algorithm include the number of iterations of the genetic algorithm, the size of the population, the crossover probability in the population, the mutation probability, the change range of the parameter alpha of the variational mode algorithm and the change range of K.

[0045] In the embodiment of the application, the number of iterations is set to 100, the population size is 30, the crossover probability in the population is set to 0.8, the mutation probability is set to 0.2, the change range of the parameter alpha of the variational mode algorithm is set to [500, 2000], and the change range of K is set to [3, 10].

[0046] Step 3: optimizing the variational mode decomposition algorithm by using the genetic algorithm, and performing mode decomposition on the processed historical data by using the optimized variational mode decomposition algorithm to obtain the optimal decomposition result of the current historical data;

[0047] The step 3 includes the following steps:

[0048] Step 3.1: initializing the population, randomly assigning the quadratic penalty factor alpha and the number K of the intrinsic mode function of each individual in the population, calculating the fitness of each individual, and taking the sample entropy as the fitness function of the individual;

[0049] Step 3.2: performing a selection operation on the population, iterating all individuals, and obtaining a new population based on the optimal solution;

[0050] In the step 3.1 and the step 3.2, when initializing the population, a chromosome is randomly generated, any chromosome represents an individual in the population, and all randomly generated chromosomes correspond to the entire population.

[0051] Each chromosome is composed of individual fitness and quadratic penalty factor alpha of the variational mode decomposition algorithm, and the number of eigenmode functions K; the individual is repeatedly generated and the fitness of each individual is calculated until the number of individuals meets the set size of the population, wherein the fitness function of the individual is the sample entropy of the individual; the alpha and K values in each individual are taken as the parameters of the variational mode decomposition algorithm, the corresponding set of eigenmode functions is obtained through the variational mode decomposition algorithm, then the sample entropy of each function in the set is calculated, and the smallest sample entropy in the set is selected as the fitness of the current individual;

[0052] The selection operation is performed on the population, the roulette method is used to select individuals from the population to enter the next generation, the reciprocal of the fitness value of each individual is calculated, and the probability of each individual entering the next generation is the ratio of the fitness of the individual to the total fitness of the entire population; the operation is repeated until the number of selected individuals is equal to the set population size. The probability of each individual being selected is calculated, the cumulative probability of the individual is calculated, all individuals are traversed until the cumulative probability is greater than a random value, then the current individual is selected as an individual in the new population, the operation is repeated thirty times until thirty individuals are selected, and the fitness of the selected individuals is calculated.

[0053] Step 3.3: performing the crossover operation on the population;

[0054] Step 3.4: performing the mutation operation on the population, generating a random value, and if the random value is less than the mutation probability, randomly selecting the individual alpha or K to perform the mutation operation;

[0055] In the step 3.3 and the step 3.4,

[0056] Two individuals are randomly selected to perform the crossover operation, and whether to perform the crossover is determined by the crossover probability; after the two individuals to be crossed are determined, the corresponding position variables of the individuals are randomly selected to perform the crossover operation, that is, the parameters at the same positions in the two individuals are crossed using the arithmetic crossover method to generate new individuals;

[0057] The mutation operation is performed on the population, whether the individual is mutated is determined by the mutation probability, and the parameters in the individual are randomly mutated; when the mutation is performed, the parameters at the positions are revalued as values in the parameter change range.

[0058] Step 3.5: updating the fitness of the individual, selecting the individual with the best fitness in the iteration process, taking the parameters of the individual as the parameters alpha and K in the variational mode decomposition algorithm, and decomposing the historical data to obtain K eigenmode functions.

[0059] In the application, the alpha and K values in each individual are taken as the parameters of the variational mode decomposition algorithm, and other parameter settings include that the noise tolerance is set to 0, the initialization center frequency is set to 0, the convergence condition is set to 1e-7, and the direct current component is set to 0.

[0060] In the application, roulette algorithm is used for selection operation on all individuals in the population, the fitness of each individual is inverted, and then the probability of selection of each individual is calculated

[0061]

[0062] Wherein, p(x i ) is the probability of selection of each individual, and f(x i ) is the fitness value of the individual

[0063] The cumulative probability of the individual is obtained

[0064]

[0065] Wherein, q(x i ) is the cumulative probability of the individual, and p(x j ) is the probability of selection of the individual

[0066] A number between 0 and 1 is randomly generated, and all individuals are traversed until the cumulative probability is greater than the random value, then the current individual is selected as the individual in the new population, and the process is repeated 30 times until 30 individuals are selected, and the fitness of the selected individuals is calculated.

[0067] In the application, the cross operation of the population refers to thirty iterations, in each iteration, two individuals in the population are randomly selected, a number between 0 and 1 is randomly generated, if the number is less than 0.8, the cross operation is performed, and the cross operation is randomly selected to be alpha or K, the same cross position of the two individuals to be cross operated is arithmetically cross operated, and the obtained new value is taken as the value of the individual.

[0068] In the application, the mutation operation is performed on the population, the mutation probability is 0.1, a value between 0 and 1 is randomly generated, and the mutation operation is performed if the value is less than 0.1, and in the mutation operation, a parameter in the individual is randomly assigned a random value in the parameter range.

[0069] In the application, the fitness of all individuals after mutation is updated, and the individual with the best fitness in this iteration is selected, after all iterations are completed, the global optimal individual parameter is obtained as the parameters alpha and K in the variational modal decomposition algorithm, and the data set is decomposed to obtain K intrinsic modal functions.

[0070] Step 4: normalizing the multiple intrinsic modal components obtained by the variational modal decomposition algorithm

[0071] In step 4, after normalization, each function is divided into two parts, one part is the input of the model, and the other part is the output of the model.

[0072] In the application, normalization is to limit the range of function value to [0, 1], which can accelerate the speed of subsequent solution and improve the accuracy of results.

[0073] In the application, the formula of normalization is,

[0074]

[0075] wherein, X norm is the normalized result, X is the original data, X min is the minimum value in the original data, X max is the maximum value in the original data, after normalization processing, the value of the intrinsic modal function obtained by decomposition is in the range of [0, 1].

[0076] In the application, after normalization is completed, each function is divided into two parts, one part is the input of the model, and the other part is the output result of the model, and the data set is divided into the following results with thirty days as the boundary,

[0077]

[0078]

[0079] wherein: X is the input data set, Y represents the output result; each row in X is the input data of the model, representing the system access volume in the previous 30 days, and each row in Y is the output data corresponding to X, representing the value to be predicted.

[0080] Step 5: each processed function is respectively input into a multi-scale CNN neural network, and after extracting the features, the features are respectively input into a GRU network as parameters, and the model is trained until stable;

[0081] In step 5, the input data is convolved with a plurality of one-dimensional convolution kernels of different sizes, different size convolution kernels extract different feature values, and the features are zero-padded at both ends in the form of equal-width convolution, so that the extracted feature dimensions are the same; the GRU network is composed of two layers of GRU units, two layers of DroupOut layers and one layer of full connection layers.

[0082] In the application, the input data is convolved with n one-dimensional convolution kernels of different sizes, different size convolution kernels can extract different feature values, and the effective features in the multi-scale extracted signals; because different size convolution kernels are used to convolve the input data, the dimensions of the features are also different, so the features are zero-padded at both ends in the form of equal-width convolution to ensure that the extracted feature dimensions are the same.

[0083] In the present application, the convolution layer in the network adopts ReLU as the activation function, that is, f(x)=max(0,x), which can increase the nonlinearity of the neural network model.

[0084] In the present application, the GRU network uses the tanh activation function and the Adam optimizer, and the tanh function is

[0085] Step 6: using the optimized variational modal decomposition algorithm, the historical data (such as 30 days) before the time to be predicted are decomposed by variational modal decomposition, and the decomposed results are input into the model trained in step 5, and the result of adding the output results of the model is taken as the system access quantity prediction value.

[0086] A computer readable storage medium having a deep learning-based system access quantity prediction program stored thereon, the program being executed by a processor to implement the deep learning-based system access quantity prediction method.

[0087] In order to realize the above-mentioned embodiments, a computer readable storage medium having a deep learning-based system access quantity prediction program stored thereon is provided, the program being executed by a processor to implement the deep learning-based system access quantity prediction method, mainly solving the problems of large amount of calculation and low system performance in the prior art.

[0088] A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor executes the program based on the deep learning-based system access quantity prediction method.

[0089] In order to realize the above-mentioned embodiments, the present application also relates to a computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor executes the program based on the deep learning-based system access quantity prediction method.

[0090] Those skilled in the art will appreciate that embodiments of the present application can be provided as methods, systems or computer program products. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product embodied on one or more computer usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer usable program code.

[0091] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0092] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0093] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0094] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0095] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for predicting system traffic based on deep learning, characterized in that: The method comprises the following steps: Step 1: load historical data and process; Record the number of times the monitored API is called each day, and the access volume is increased by one each time the API is called, and the daily access volume is stored daily as historical data; read the historical data of the access volume as the data set of the training model, if there is missing data in any day or more days, then the historical data of the associated date is used to complete the missing data according to the preset rule; Step 2: based on the historical data, set the associated parameters of the genetic algorithm; the associated parameters of the genetic algorithm include the number of iterations of the genetic algorithm, the size of the population, the crossover probability in the population, the mutation probability, the change range of the parameter α of the variational mode algorithm and the change range of K; Step 3: use the genetic algorithm to optimize the variational mode decomposition algorithm, and use the optimized variational mode decomposition algorithm to perform mode decomposition on the processed historical data to obtain the optimal decomposition result of the current historical data; the step 3 comprises the following steps: Step 3.1: initialize the population; Step 3.2: select the population; Step 3.3: perform crossover operation on the population; Step 3.4: perform mutation operation on the population; Step 3.5: update the fitness of the individual, and select the individual with the best fitness in the iteration process, and use the parameters of the individual as the parameters α and K in the variational mode decomposition algorithm to decompose the historical data to obtain K intrinsic mode functions; Step 4: normalize the multiple intrinsic mode components obtained by the variational mode decomposition algorithm; Step 5: input each processed function into a multi-scale CNN neural network respectively, extract the features, input the features into a GRU network as parameters, and train the model until it is stable; Step 6: use the optimized variational mode decomposition algorithm to perform variational mode decomposition on the historical data before the time to be predicted, input the decomposition result into the model trained in step 5, and add the results output by the model to obtain the predicted value of the system access volume. 2.The deep learning-based system access volume prediction method of claim 1, wherein: In the step 3.1 and the step 3.2, when initializing the population, randomly generate chromosomes, and each chromosome represents an individual in the population, and all randomly generated chromosomes correspond to the entire population; Each chromosome is composed of individual fitness and quadratic penalty factor α of the variational mode decomposition algorithm and intrinsic mode function number K value; repeat to generate individuals and calculate the fitness of each individual until the number of individuals meets the set size of the population, wherein the fitness function of the individual is the sample entropy of the individual; Select the individual into the next generation from the population by using the roulette method, calculate the reciprocal of the fitness value of each individual, and the probability of each individual entering the next generation is the ratio of the fitness of the individual to the total fitness of the entire population; repeat until the number of selected individuals is equal to the set population size. 3.The deep learning-based system access volume prediction method of claim 1, wherein: In the step 3.3 and the step 3.4, Two individuals are randomly selected for crossover operation, and whether to perform crossover is determined by a crossover probability; after determining the two individuals that need to be crossed, a corresponding position variable of the individual is randomly selected for crossover operation, that is, the parameters at the same position in the two individuals are crossed using an arithmetic crossover method, the arithmetic crossover linearly combines the parameters at the same position in the two individuals to generate a new individual; The population is subjected to mutation operation, and whether an individual is mutated is determined by a mutation probability; a parameter in a certain individual is randomly mutated, and the parameter at the position is revalued as a value in a parameter change range. 4.The deep learning-based system access volume prediction method of claim 1, wherein: In step 4, after normalization, each function is divided into two parts, one part is the input of the model, and the other part is the output of the model. 5.The deep learning-based system access volume prediction method of claim 1, wherein: In step 5, the input data is convolved at several scales by using several one-dimensional convolution kernels of different sizes, different feature values are extracted by using the convolution kernels of different sizes, and the features are zero-padded at both ends in an equal-width convolution manner, so that the extracted feature dimensions are the same; the GRU network is composed of two GRU units, two DroupOut layers and one fully connected layer.

6. A computer-readable storage medium, characterized in that, The system access volume prediction program based on deep learning is stored on the storage medium, and the program is executed by the processor to implement the system access volume prediction method based on deep learning according to any one of claims 1-5.

7. A computer device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the program to implement the system access volume prediction method based on deep learning according to any one of claims 1-5.

Images