Service volume data prediction method and device for network service platform

Abstract

The invention discloses a service volume data prediction method and device oriented to a network service platform. The method comprises the following steps: extracting multivariable time sequence features based on an encoder DSTCN; introducing Self-attention, extracting multivariable time sequence features output by the encoder DSTCN, fusing time step features and variable features, and adding weights to the time step features and the variable features which are output by the encoder DSTCN and extracted from a network service platform service volume data set; a decoder BIGRU is constructed by using a BIGRU network model, and a prediction result is obtained by fusing the Self-attention output features; and taking the network service platform service volume data set as model input, carrying out model training and parameter tuning on a model by using particle swarm optimization (PSO), and determining the optimal number of layers of encoders and decoders to obtain a network service platform service volume prediction result. The device comprises a processor and a memory.

Description

Technical Field

[0001] The present invention relates to the field of network service platform prediction, and in particular to a method and device for predicting service volume data for a network service platform. Background Art

[0002] Business volume is an important indicator for measuring the development and operation of the network business platform, and is also the basis for the platform's future strategic planning. The business volume of the network business platform is closely related to factors such as user demand, policy adjustments, and public events, and has certain volatility and intermittency. For example, during holidays, the public service demand is low, while when notices are issued or important events occur, the platform visits and processing volume may surge.

[0003] Through accurate business volume forecasting, the network business platform can optimize resource allocation, refine management operations, and adjust operating strategies in a timely manner, thereby improving user experience.

[0004] Traffic volume data is a kind of multivariate time series data. Current traditional models, including linear models such as the Autoregressive Integrated Moving Average (ARIMA) model, have been used for time series forecasting tasks, but these models ignore the nonlinear characteristics of the data. Actual data generally do not meet the assumptions of linearity and smoothness, so the accuracy of linear models in practical applications is low. Machine learning-based forecasting methods are outstanding in processing time series data, especially nonlinear problems, such as support vector machines (SVR) and artificial neural networks (ANN). However, although this method takes into account nonlinear characteristics and has a positive effect on improving forecasting performance, it is difficult to represent the correlation between different time series data and the dependency between different time steps, and as the data scale and data attributes grow, the model's forecasting performance is poor.

[0005] Deep learning prediction methods have been applied to time series prediction, including: Temporal Convolutional Network (TCN), Long Short-Term Memory (LSTM). Such methods show high accuracy. Because they rely on complex network structures, they can extract deeper features in a more abstract way, which is sufficient to capture the long-term dependencies between different time steps and the correlations between different time series, and have stronger modeling capabilities.

[0006] Since then, a large number of time prediction methods based on deep learning have been used for multivariate time series prediction tasks. However, the business volume data prediction models for network service platforms are relatively rare, and most of them are single models. Since the business volume of network service platforms is closely related to factors such as user demand, policy adjustments, and public events, it has certain volatility and intermittency, and the prediction accuracy of a single model is low. Summary of the invention

[0007] The present invention provides a method and device for predicting traffic volume data for a network service platform. The present invention can effectively extract the nonlinear characteristics and time volatility of multivariate time series data in traffic volume data, effectively improve the accuracy and reliability of traffic volume data prediction, thereby helping the network service platform to optimize resource allocation, refine management operations, and timely adjust operation strategies to improve user experience. See the following description for details:

[0008] In a first aspect, a method for predicting traffic volume data for a network service platform is provided, the method comprising:

[0009] Using the Pearson correlation coefficient, a correlation coefficient matrix between the traffic volume to be predicted and all characteristic variables in the network traffic platform traffic data set is constructed to evaluate the linear relationship between the input variables and determine the input variables of the DABnet model.

[0010] Extract multivariate time series features based on encoder DSTCN; introduce self-attention to extract multivariate time series features output by encoder DSTCN, fuse time step features and variable features, and add weight to time step features and variable features extracted from network service platform traffic data set output by encoder DSTCN;

[0011] The decoder BIGRU is constructed using the BIGRU network model, and the Self-attention output features are integrated to obtain the prediction results. The network service platform traffic data set is used as the model input, and the particle swarm optimization PSO is used to train the model and tune the parameters to determine the optimal number of encoder and decoder layers to obtain the network service platform traffic prediction results.

[0012] Wherein, the input variables for determining the DABnet model are:

[0013] Use the Pearson correlation coefficient to perform correlation analysis on the preprocessed data, and mine the correlation between the target business volume to be predicted and the characteristic variables of the network business platform business volume data set according to the results of the Pearson correlation coefficient analysis;

[0014] According to the correlation value, the features with correlation higher than the threshold are screened, the Pearson correlation coefficient between multiple features is calculated, the linear relationship between the input variables is evaluated, and the features with negative correlation with the business volume to be predicted are eliminated to reduce data redundancy.

[0015] The method of extracting multivariate time series features based on encoder DSTCN is as follows: the network service platform data set after data preprocessing is converted into a one-dimensional matrix with time series, and input into the DSTCN network model;

[0016] Among them, the calculation method of DSTCN network is:

[0017]

[0018] D(T)=D1(T)×D2(T)

[0019] Y(T)=relu[D(T)+X(T)]

[0020] Y′(T)=w t Y(T)+b t

[0021] Where X is a one-dimensional time series, X = x0, x1, x2, …, x t , T is the total time, d represents the expansion factor, k is the filter size, D1(T) and D2(T) are the dilated convolution outputs of the two channels respectively, D(T) is the output information merging result, X(T) is the input one-dimensional time series, Y(T) is the result of fusion with the original input features using the activation function relu, Y′(T) is the output of the model after convolution, W t is a learnable parameter, b t is the offset, X t-d·i is the element in the input sequence corresponding to the dilated convolution kernel operation.

[0022] Among them, the BIGRU network model is composed of two independent GRU units, one processes data forward according to the time series, and the other processes data reversely according to the time series; through this bidirectional structure, the BIGRU network model can simultaneously capture the forward and backward information of the sequence data.

[0023] In a second aspect, a traffic volume data prediction device for a network service platform comprises: a processor and a memory, wherein the memory stores program instructions, and the processor calls the program instructions stored in the memory to enable the device to execute any one of the methods described in the first aspect.

[0024] In a third aspect, a computer-readable storage medium stores a computer program, wherein the computer program includes program instructions, and when the program instructions are executed by a processor, the processor executes any one of the methods described in the first aspect.

[0025] The beneficial effects of the technical solution provided by the present invention are:

[0026] 1. The present invention uses the heterogeneous encoding and decoding structure as the basic framework DABnet, and integrates the improved time convolution network DSTCN, the bidirectional gated recurrent neural network BIGRU, and the time step feature and variable feature attention mechanism Self-attention. Among them, DSTCN is used as an encoder to extract multivariate time series features, so as to better capture the long-distance dependency of feature data; then Self-attention is used to extract and fuse time step features and variable features, and the weight of the features is increased to enhance the model's understanding of nonlinear features and time volatility features; finally, BIGRU is used as a decoder to further fuse features to obtain more accurate network service platform data volume prediction results;

[0027] 2. A network of the present invention is an encoder composed of multiple layers of serially connected DSTCNs. By adjusting the model structure, number of information channels, and activation function of TCN, an improved TCN model with better model performance is obtained to better capture the long-distance dependency of feature data;

[0028] 3. Another network of the present invention is a decoder composed of BIGRU and Self-attention. Self-attention can add weight to features and enhance the model's understanding of nonlinear features and time volatility features. BIGRU further integrates features when facing longer sequence inputs to obtain more accurate network service platform data volume prediction results, timely optimize resource allocation, and timely adjust operation strategies, thereby improving user experience;

[0029] 4. The present invention uses root mean square error (RMSE), mean square error (MSE), mean absolute error (MAE), symmetric mean absolute percentage error (SMAPE), standardized absolute error (NAE), determination coefficient (R 2 ) and comprehensive evaluation index (IA) are used to evaluate the effect of the neural network framework. Experiments show that the invention has good generalization performance, high prediction accuracy, small error, and is effectively applicable to the task of predicting the traffic volume data of the network service platform;

[0030] 5. Through accurate prediction of the business volume data of the network business platform, the network business platform can optimize resource allocation, refine management operations, and adjust operation strategies in a timely manner, thereby improving user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 A schematic diagram of a business volume data forecast for a network business platform;

[0032] Figure 2 A flowchart of a method for predicting business volume data for a network business platform;

[0033] Figure 3 This is a schematic diagram of the DABnet structure;

[0034] Figure 4 It is a schematic diagram of the structure of the gated recurrent neural network GRU;

[0035] Figure 5 It is a schematic diagram of the structure of the bidirectional gated recurrent neural network BIGRU;

[0036] Figure 6 It is a structural diagram of DSTCN;

[0037] Figure 7 This is a schematic diagram of the structure of Self-attention; DETAILED DESCRIPTION

[0038] In order to make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention are described in further detail below.

[0039] Example 1

[0040] The embodiment of the present invention provides a method for predicting traffic volume data for a network service platform, the method comprising the following steps:

[0041] 101: Collect and analyze historical data of different business volumes handled daily by the network business platform, perform data preprocessing, and establish a network business platform business volume data set with a time series relationship;

[0042] 102: Using the Pearson correlation coefficient, a correlation coefficient matrix is ​​constructed between the traffic volume to be predicted and all characteristic variables in the network traffic platform traffic data set, the linear relationship between the input variables is evaluated, and the input variables of the DABnet model are determined;

[0043] Among them, the DABnet (Heterogeneous Codec Network) model is the overall model, which consists of the encoder DSTCN dual-channel single-layer temporal convolutional network), the intermediate layer self-attention (self-attention mechanism), and the decoder BIGRU (bidirectional recurrent neural network).

[0044] 103: Use the DSTCN network model to build an encoder DSTCN, and extract multivariate time series features based on the encoder DSTCN;

[0045] 104: Self-attention is introduced to extract the multivariate time series features output by the encoder DSTCN, and to fuse the time step features and variable features, thereby adding weight to the time step features and variable features extracted from the network service platform traffic data set output by the encoder DSTCN, and enhancing the model's understanding of nonlinear features and time volatility features;

[0046] 105: Use the BIGRU network model to build a decoder BIGRU, and further integrate the Self-attention output features to obtain the prediction results;

[0047] 106: Using the network service platform traffic data set as the model input, the particle swarm optimization PSO is used to train the DABnet model and tune its parameters to determine the optimal number of encoder and decoder layers to obtain better network service platform traffic prediction results, thereby helping the network service platform to optimize resource allocation, refine management operations, and adjust operation strategies in a timely manner to improve user experience.

[0048] Example 2

[0049] The following is a further introduction to the solution in Example 1 by combining a specific example and a calculation formula, as described below:

[0050] 201: Obtain historical data of different business volumes handled daily by the network business platform for ten years, perform data analysis and preprocessing, and establish a network business platform business volume data set with a time series relationship;

[0051] The collected historical data of different business volumes handled daily include: daily completed business volume, daily uncompleted business volume, daily completed business volume growth rate and daily uncompleted business volume growth rate of different businesses. The collected data is saved in csv file format, and Python is used to pre-process the outliers and missing values ​​in the original data to avoid the missing values ​​affecting the prediction results.

[0052] Linear interpolation is used to fill in the missing data in the data set. The calculation formula is:

[0053]

[0054] In the formula, x b is the value of the missing part, x a For x b The previously known value, x c For x b The following known values, a, b, c, are natural numbers greater than or equal to 1.

[0055] After completing the preprocessing of the network service platform traffic volume data, a network service platform traffic volume data set with a time series relationship is established for predictive modeling and training.

[0056] 202: Using the Pearson correlation coefficient, a correlation coefficient matrix is ​​constructed between the traffic volume to be predicted and all characteristic variables in the network traffic platform traffic data set, the linear relationship between the input variables is evaluated, and the input characteristics of the DABnet model are determined;

[0057] Among them, the Pearson correlation coefficient method is used to perform correlation analysis on the preprocessed data, and the correlation between the target business volume to be predicted and the characteristic variables of the network business platform business volume data set is mined. According to the results of the Pearson correlation coefficient analysis, the correlation values ​​are sorted to screen out the features with correlations higher than the threshold. The Pearson correlation coefficients between multiple features are calculated, the linear relationship between the input variables is evaluated, and the features that are negatively correlated with the business volume to be predicted are eliminated to reduce data redundancy.

[0058] The calculation formula of Pearson correlation coefficient is:

[0059]

[0060] Where COV(X,Y) is the covariance between features X and Y, σ X and σ Y They are the standard deviations of features X and Y. Finally, the maximum and minimum normalization process is used, and the normalization interval is 0 to 1 to eliminate the dimensional influence between features to solve the comparability between data indicators.

[0061] Among them, the maximum and minimum normalization formula is:

[0062]

[0063] Y=Y inorm *(mx-mi)+mi

[0064] In the formula, x imax is the maximum value of the i-th feature of the input sample, x imin is the minimum value of the i-th feature of the input sample, X inorm is the normalized value of the i-th feature of the input sample, Y is the final result, mx is the maximum value of the specified interval, and mi is the minimum value of the specified interval.

[0065] In the specific implementation, the embodiment of the present invention selects the traffic data of a certain network service platform in the United States as the model input data for illustration.

[0066] 203: Use the DSTCN network model to build an encoder and extract multivariate time series features based on the encoder;

[0067] Among them, the DSTCN network model is as follows Figure 6 As shown in the figure, the DSTCN network model can extract features from the input multivariate time series, thereby capturing the long-range dependencies of the time series and embedding the key information of the input data. The DSTCN network model with different hole convolution sizes can focus on learning the corresponding tasks, thereby fully extracting the features of the data.

[0068] The information operation channel of the residual layer of the DSTCN network model is changed from 1 to 2. The input features enter the two channels for calculation respectively, and then the features of the dual-channel outputs are fused. After being fused with the original input features, convolution is further used to fuse the features.

[0069] The specific steps are as follows: the network service platform data set after data preprocessing is converted into a one-dimensional matrix with time series and input into the DSTCN network model;

[0070] Among them, the calculation method of DSTCN network is:

[0071]

[0072] D(T)=D1(T)×D2(T)

[0073] Y(T)=relu[D(T)+X(T)]

[0074] Y′(T)=w t Y(T)+b t

[0075] Where X is a one-dimensional time series, X = x0, x1, x2, …, x t , T is the total time, d represents the expansion factor, k is the filter size, D1(T) and D2(T) are the dilated convolution outputs of the two channels respectively, D(T) is the output information merging result, X(T) is the input one-dimensional time series, Y(T) is the result of fusion with the original input features using the activation function relu, Y′(T) is the output of the model after convolution, W t is a learnable parameter, b t is the offset, X t-d·i is the element in the input sequence corresponding to the dilated convolution kernel operation.

[0076] 204: Self-attention is introduced to extract the output features of DSTCN and fuse the time step features and variable features, thereby increasing the weight of the important features of DSTCN output and enhancing the model's understanding of nonlinear features and time volatility features;

[0077] Among them, Self-attention can help the model focus on and weight the input parts at different time steps and input features at each time step to better capture spatiotemporal characteristics.

[0078] Self-attention uses Tanh activation function and Sigmoid activation function for learning attention weights, and obtains two attention factor matrices about time step features and variable features respectively. The two attention factor matrices are normalized with softmax function to obtain the attention weight matrix A. tf , the attention weight matrix A tf With the time series matrix x n Combined with the time series matrix x n Add together and get

[0079] Among them, the Tanh activation function calculation method is:

[0080]

[0081] Sigmoid activation function calculation method:

[0082]

[0083] Among them, the calculation method of Self-attention is:

[0084] Assume that a one-dimensional time series X = {x0, x1, x2..., x t}, at each time step x n The input is [batchsize, timestep, features]:

[0085]

[0086] Where W t1 , W t2 , W f1 , W f2 is a learnable parameter, b t1 , b t2 , b f1 , b f2 is the offset, σ is defined as the sigmoid activation function, δ is defined as the tanh activation function, A t1 is the temporal attention factor using σ, A t2 is the temporal attention factor using δ, A f1 is the temporal attention factor using σ, A f2 is the temporal attention factor using δ, Atf is the normalized spatiotemporal attention factor, represents the input features after spatiotemporal attention, and softmax is the Gaussian normalization of the spatial dimension. is defined as the Hadamard product.

[0087] 205: Use the BIGRU network model to build a decoder to further extract and fuse the output features of Self-attention to obtain the prediction results;

[0088] Among them, the BIGRU network model is as follows Figure 5 As shown in the figure, the BIGRU network model consists of two independent GRU units, one for forward processing of data in the time series and the other for reverse processing of data in the time series. Through this bidirectional structure, the BIGRU network model can simultaneously capture the forward and backward information of the sequence data, thereby better understanding and predicting the patterns in the sequence and avoiding the problems of gradient vanishing and gradient exploding.

[0089] The specific steps are as follows: Input into the BIGRU network model;

[0090] Among them, the calculation method of the GRU network is:

[0091] f t =σ(w f ·[h t-1 ,x t ]+b f )

[0092] i t =σ(w i ·[h t-1 ,x t ]+b i )

[0093] c t =tanh(w c ·[h t-1 ,x t ]+b c )

[0094] h t =(1-f t )·h t-1 +f t ·c t

[0095] In the formula, f t represents the forget gate, i t represents the input gate, h t represents the output value of the neuron at time t, σ is defined as the sigmoid activation function, tanh is the tanh activation function, and wf , w i , w c is a learnable parameter, [h t-1 ,x t ] represents the input value of the previous moment and the output value information of this moment, b f , b i , b c is the offset.

[0096] The BIGRU network model generates the final output by combining the hidden states in both the forward GRU and reverse GRU directions.

[0097] Among them, the calculation method of the BIGRU network is:

[0098]

[0099] In the formula, H t represents the output value of the neuron at time t, represents the positive output value of the neuron at time t, Represents the negative output value of the neuron at time t.

[0100] 206: Using the network service platform traffic data set as the model input, the particle swarm optimization PSO is used to tune the model training parameters of the proposed model and determine the optimal number of encoder and decoder layers to obtain better network service platform traffic prediction results;

[0101] The encoder and decoder structures must have a certain depth to capture the irregularities between time series data and its context. Since the superposition of DSTCN network layers and BIGRU network layers is only effective within a certain number of layers, the performance of the decoder may be impaired once this range is exceeded. The optimal number of layers for the encoder and decoder is determined by particle swarm optimization (PSO) to optimize the performance of the model.

[0102] The idea of ​​PSO comes from the study of the foraging behavior of bird flocks. Bird flocks share collective information to find the optimal destination. It has the advantages of fast convergence speed, few parameters, and simple and easy-to-implement algorithm.

[0103] PSO is initialized as a group of random particles (random solutions). Then the optimal solution is found through selection. In each selection, the particles update themselves by tracking two "extreme values" (pbest, gbest). After finding these two optimal values, the particles update their speed and position using the following formula.

[0104] Among them, the calculation method of PSO network is:

[0105] v i =vi +c1·rand()·(pbest i -x i )+c2·rand()·(gbest i -x i )

[0106] x i =x i +v i

[0107] v i =w·v i +c1·rand()·(pbest i -x i )+c2·rand()·(gbest i -x i )

[0108] Where i = 1, 2, ..., N, N is the total number of particles in the group, v i is the velocity of the particle, pbest i and gbest i is the extreme value of the i-th particle, rand() is a random number between (0,1), x i is the current position of the particle, c1 and c2 are learning factors, usually c1=c2=2, and w is the inertia factor.

[0109] When outputting the predicted value of business volume, the root mean square error (RMSE), mean square error (MSE), mean absolute error (MAE), symmetric mean absolute percentage error (SMAPE), standardized absolute error (NAE), determination coefficient (R 2 ) and the comprehensive assessment index (IA) as the accuracy index to evaluate the difference between the true value and the predicted value.

[0110]

[0111]

[0112] In the formula, y i is the true value, is the predicted value, is the mean of the true values, ε=le-8.

[0113] Obtain better network business platform business volume forecast results, thereby helping the network business platform to optimize resource allocation, refine management operations, and adjust operation strategies in a timely manner to improve user experience.

[0114] Example 3

[0115] A traffic volume data prediction device for a network service platform, the device comprising: a processor and a memory, wherein program instructions are stored in the memory, and the processor calls the program instructions stored in the memory to enable the device to execute the following method steps in embodiment 1:

[0116] Using the Pearson correlation coefficient, a correlation coefficient matrix between the traffic volume to be predicted and all characteristic variables in the network traffic platform traffic data set is constructed to evaluate the linear relationship between the input variables and determine the input variables of the DABnet model.

[0117] Extract multivariate time series features based on encoder DSTCN; introduce self-attention to extract multivariate time series features output by encoder DSTCN, fuse time step features and variable features, and add weight to time step features and variable features extracted from network service platform traffic data set output by encoder DSTCN;

[0118] The decoder BIGRU is constructed using the BIGRU network model, and the Self-attention output features are integrated to obtain the prediction results. The network service platform traffic data set is used as the model input, and the particle swarm optimization PSO is used to train the model and tune the parameters to determine the optimal number of encoder and decoder layers to obtain the network service platform traffic prediction results.

[0119] Among them, the input variables of the DABnet model are determined as:

[0120] Use the Pearson correlation coefficient to perform correlation analysis on the preprocessed data, and mine the correlation between the target business volume to be predicted and the characteristic variables of the network business platform business volume data set according to the results of the Pearson correlation coefficient analysis;

[0121] According to the correlation value, the features with correlation higher than the threshold are screened, the Pearson correlation coefficient between multiple features is calculated, the linear relationship between the input variables is evaluated, and the features with negative correlation with the business volume to be predicted are eliminated to reduce data redundancy.

[0122] Among them, the multivariate time series feature extraction based on the encoder DSTCN is as follows: the network service platform data set after data preprocessing is converted into a one-dimensional matrix with time series and input into the DSTCN network model;

[0123] Among them, the calculation method of DSTCN network is:

[0124]

[0125] D(T)=D1(T)×D2(T)

[0126] Y(T)=relu[D(T)+X(T)]

[0127] Y′(T)=w t Y(T)+b t

[0128] Where X is a one-dimensional time series, X = x0, x1, x2, …, x t , T is the total time, d represents the expansion factor, k is the filter size, D1(T) and D2(T) are the dilated convolution outputs of the two channels respectively, D(T) is the output information merging result, X(T) is the input one-dimensional time series, Y(T) is the result of fusion with the original input features using the activation function relu, Y′(T) is the output of the model after convolution, W t is a learnable parameter, b t is the offset, X t-d·i is the element in the input sequence corresponding to the dilated convolution kernel operation.

[0129] Among them, the BIGRU network model consists of two independent GRU units, one processes data forward according to the time series, and the other processes data reversely according to the time series; through this bidirectional structure, the BIGRU network model can simultaneously capture the forward and backward information of the sequence data.

[0130] It should be pointed out here that the device description in the above embodiment corresponds to the method description in the embodiment, and the embodiment of the present invention will not be described in detail here.

[0131] The execution subjects of the above-mentioned processor and memory can be devices with computing functions such as computers, single-chip microcomputers, and microcontrollers. In specific implementation, the embodiments of the present invention do not limit the execution subjects and are selected according to the needs of actual applications.

[0132] The data signal is transmitted between the memory and the processor via a bus, which is not described in detail in the embodiment of the present invention.

[0133] Based on the same inventive concept, an embodiment of the present invention further provides a computer-readable storage medium, the storage medium includes a stored program, and when the program is running, the device where the storage medium is located is controlled to execute the method steps in the above embodiment.

[0134] The computer-readable storage medium includes but is not limited to a flash memory, a hard disk, a solid-state drive, and the like.

[0135] It should be pointed out here that the description of the readable storage medium in the above embodiment corresponds to the description of the method in the embodiment, and the embodiment of the present invention will not be described in detail here.

[0136] In the above embodiments, all or part of the embodiments may be implemented by software, hardware, firmware or any combination thereof. When implemented by software, all or part of the embodiments may be implemented in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions according to the embodiments of the present invention are generated.

[0137] The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. Computer instructions may be stored in a computer-readable storage medium or transmitted via a computer-readable storage medium. The computer-readable storage medium may be any available medium that can be accessed by the computer or a data storage device such as a server or a data center that includes one or more available media. The available medium may be a magnetic medium or a semiconductor medium, etc.

[0138] Unless otherwise specified, the models of the components in the embodiments of the present invention are not limited, and any device that can perform the above functions may be used.

[0139] Those skilled in the art will appreciate that the accompanying drawing is only a schematic diagram of a preferred embodiment, and the serial numbers of the embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0140] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modification, equivalent substitution, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention. Equivalent substitution, improvement, etc. shall be included in the protection scope of the present invention.

Claims

1. A method for predicting traffic data for a network service platform, characterized in that: The method comprises: Using the Pearson correlation coefficient, a correlation coefficient matrix between the traffic volume to be predicted and all characteristic variables in the network traffic platform traffic data set is constructed to evaluate the linear relationship between the input variables and determine the input variables of the DABnet model. Extract multivariate time series features based on encoder DSTCN; introduce self-attention to extract multivariate time series features output by encoder DSTCN, fuse time step features and variable features, and add weight to time step features and variable features extracted from network service platform traffic data set output by encoder DSTCN; The decoder BIGRU is constructed using the BIGRU network model, and the Self-attention output features are integrated to obtain the prediction results. The network service platform traffic data set is used as the model input, and the particle swarm optimization PSO is used to train the model and tune the parameters to determine the optimal number of encoder and decoder layers to obtain the network service platform traffic prediction results.

2. A method for predicting traffic volume data for a network service platform according to claim 1, characterized in that: The input variables for determining the DABnet model are: Use the Pearson correlation coefficient to perform correlation analysis on the preprocessed data, and mine the correlation between the target business volume to be predicted and the characteristic variables of the network business platform business volume data set according to the results of the Pearson correlation coefficient analysis; According to the correlation value, the features with correlation higher than the threshold are screened, the Pearson correlation coefficient between multiple features is calculated, the linear relationship between the input variables is evaluated, and the features with negative correlation with the business volume to be predicted are eliminated to reduce data redundancy.

3. The method for predicting traffic volume data for a network service platform according to claim 1, characterized in that: The encoder DSTCN-based extraction of multivariate time series features is as follows: the network service platform data set after data preprocessing is converted into a one-dimensional matrix with time series, and input into the DSTCN network model; Among them, the calculation method of DSTCN network is: D(T)=D1(T)×D2(T) Y(T)=relu[D(T)+X(T)] Y(T)=wtY(T)+b Where X is a one-dimensional time series, X = x0, x1, x2, …, x t , T is the total time, d represents the expansion factor, k is the filter size, D1(T) and D2(T) are the dilated convolution outputs of the two channels respectively, D(T) is the output information merging result, X(T) is the input one-dimensional time series, Y(T) is the result of fusion with the original input features using the activation function relu, Y′(T) is the output of the model after convolution, W t is a learnable parameter, b t is the offset, X t-d i is the element in the input sequence corresponding to the dilated convolution kernel operation.

4. The method for predicting traffic volume data for a network service platform according to claim 1, characterized in that: The BIGRU network model consists of two independent GRU units, one that processes data forward in a time series, and the other that processes data reversely in a time series; through this bidirectional structure, the BIGRU network model can simultaneously capture the forward and backward information of the sequence data.

5. A traffic data prediction device for a network service platform, characterized in that: The device comprises: a processor and a memory, wherein program instructions are stored in the memory, and the processor calls the program instructions stored in the memory to enable the device to execute the method according to any one of claims 1 to 4.

6. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, wherein the computer program includes program instructions, and when the program instructions are executed by a processor, the processor is enabled to perform the method according to any one of claims 1 to 4.

Images