A cellular traffic prediction method and system based on multi-layer meta-learning model

Abstract

The present invention relates to the field of communication technology, and specifically discloses a cellular traffic prediction method and system based on a multi-layer meta-learning model. Step S01: Construct a multi-layer meta-learning model, wherein: the first layer is a main model, which is used to generate weights according to the frequency domain information of the cellular network traffic load; the second layer is a weight distribution model, which is used to map the corresponding sub-model according to the weights; the third layer is a sub-model, which is used to predict the traffic load of the next time period; Step S02: Collect cellular network traffic load data, pre-process the cellular network traffic load data, and use it to train the main model and the sub-model to obtain a trained multi-layer meta-learning model; Step S03: Use the trained multi-layer meta-learning model to predict the cellular network traffic load, calculate the traffic load prediction value, and obtain predicted load information.

Description

Technical Field

[0001] The present invention relates to the field of communication technology, and in particular to a cellular traffic prediction method and system based on a multi-layer meta-learning model. Background Art

[0002] With the rapid development of mobile communication technologies and applications such as the Internet of Things (IoT), cloud computing, and virtual / augmented reality, traffic in mobile networks is growing exponentially. To meet growing user demand and reduce operating expenses (OpEx), cellular network traffic load forecasting plays a crucial role in mobile networks, as many mobile applications rely on real-time or near-real-time traffic analysis within the radio access network. Traffic forecasting methods fall into two main categories: time series models based on historical traffic statistics (such as ARIMA and exponential smoothing) are limited in their ability to effectively capture the spatiotemporal correlations between traffic loads across multiple base stations and the impact of sudden events. Machine learning-based forecasting frameworks (such as LSTM and CNN) can extract local spatiotemporal features, but they lack the ability to collaboratively model multidimensional data across a regional base station cluster (such as user mobility trajectories, application type distribution, and network slicing requirements). They also suffer from high computational complexity and poor real-time performance. Furthermore, existing solutions generally lack the ability to adaptively model dynamic network topology (such as base station sleep / active state transitions), resulting in a significant decrease in forecasting accuracy in dynamic scenarios. Therefore, if the future traffic loads of multiple base stations can be accurately predicted, a sleep strategy can be implemented for the corresponding base stations based on the future network traffic load to reduce energy consumption, and network resources can be optimized before improving the service experience of mobile users, which becomes a key technical challenge to enhance the intelligence level of mobile networks. Summary of the Invention

[0003] To address these issues, the first objective of this invention is to provide a cellular traffic prediction method based on a multi-layer meta-learning model. By constructing a multi-layer meta-learning model, using multiple sub-models for parallel prediction, and fusing the prediction results, the overall model accuracy is improved. A correction mechanism based on prediction errors is also incorporated, making the entire model more robust.

[0004] The second object of the present invention is to provide a cellular traffic prediction system based on a multi-layer meta-learning model.

[0005] To achieve the first objective, the first technical solution of the present invention is: a cellular traffic prediction method based on a multi-layer meta-learning model, comprising:

[0006] Step S01: Construct a multi-layer meta-learning model, wherein: the first layer is a main model, which is used to generate weights based on the frequency domain information of the cellular network traffic load; the second layer is a weight distribution model, which is used to map the corresponding sub-models according to the weights; and the third layer is a sub-model, which is used to predict the traffic load of the next time period;

[0007] Step S02: collecting cellular network traffic load data, preprocessing the cellular network traffic load data, and using the data to train the main model and the sub-model to obtain a trained multi-layer meta-learning model;

[0008] Step S03: Use the trained multi-layer meta-learning model to predict the cellular network traffic load, calculate the traffic load prediction value, and obtain predicted load information.

[0009] Preferably, in step S01, the number of the sub-models is multiple, and the sub-model is composed of multiple GRU neural networks.

[0010] Preferably, in step S01, the main model includes a DNN network, and the DNN network is used to reduce the dimension of the frequency domain information of the cellular network traffic load to obtain low-dimensional samples, decode the low-dimensional samples to obtain decoded samples, calculate the reconstruction error between the decoded samples and the frequency domain information, and obtain output samples, and the output samples include the reconstruction error and the low-dimensional samples.

[0011] Preferably, in step S01, the main model also includes a GMM network, which is used to perform density estimation on the output samples of the DNN network, and obtain weights based on the density estimation, wherein the weights include the weights of the frequency domain information of the cellular network traffic load in the corresponding sub-model.

[0012] Using multiple sub-models for parallel prediction provides a foundation for improving the overall accuracy of subsequent models. Using soft weights, the prediction results of multiple prediction sub-models are integrated, rather than using the prediction results of only one sub-model, which improves model accuracy.

[0013] Preferably, in step S02, the training of the main model is performed according to a constraint function, and the constraint function is:

[0014] ;

[0015] ;

[0016] in, is the encoder parameter, is the decoder parameter, is the reconstruction error, For output samples, For the Output sample of the DNN network with cells The degree of abnormality, For the The DNN network of the cell output The degree of abnormality, is a minimum value, is the mixing probability, is the mean, is the variance, exp() represents the natural exponential function, and are the constant coefficients, is the number of input samples, Output No. vectors, For the Output sample of the DNN network with cells No. The degree of abnormality of a vector.

[0017] Preferably, in step S02, training the sub-model includes using different types of cellular network traffic load data for training, and the cellular network traffic load data includes real-time video data and non-video data.

[0018] Preferably, in step S02, the preprocessing of the cellular network traffic load data includes: normalizing the collected cellular network traffic load data using the minimum-maximum normalization method to obtain a traffic load vector, and processing the traffic load vector using the fast Fourier transform method to generate frequency domain information of the cellular network traffic load, and the frequency domain information is used to train the main model and the sub-model.

[0019] Preferably, the system further includes step S04 of determining a prediction error between the predicted cellular network traffic load at the previous moment and the actual cellular network traffic load. When the prediction error is greater than or equal to an error threshold, the main model is retrained, the weights are updated, and the predicted load information is re-output based on the weights. The addition of a correction mechanism based on the prediction error makes the entire model more robust.

[0020] To achieve the second objective, the second technical solution of the present invention is: a cellular traffic prediction system based on a multi-layer meta-learning model, comprising:

[0021] Model construction module: used to build a multi-layer meta-learning model, where the first layer is a main model, used to generate weights based on the frequency domain information and load information of the cellular network traffic load; the second layer is a weight allocation model, used to map the corresponding sub-models according to the weights; the third layer is a sub-model, used to predict the traffic load of the next time period;

[0022] Model training module: used for collecting cellular network traffic load data, preprocessing the cellular network traffic load data, and training the main model and the sub-model to obtain a trained multi-layer meta-learning model;

[0023] Result output module: used to use the trained multi-layer meta-learning model to predict the cellular network traffic load, calculate the traffic load prediction value, and obtain and output the predicted load information.

[0024] Preferably, it further includes a prediction error and correction module: configured to determine whether to output the predicted load information based on the prediction error between the predicted load information at the previous moment and the actual load;

[0025] When the prediction error at the current moment is greater than or equal to the error threshold, the main model is retrained to update the weights and the predicted load information is re-output.

[0026] Beneficial effects of the above technical solution:

[0027] The cellular traffic prediction method and system based on the multi-layer meta-learning model provided by the present invention achieves a significant improvement in the accuracy and robustness of cellular network traffic load prediction. The present invention constructs multiple sub-models to parallelly process the frequency domain information, time series traffic data, real load information and other features of the cellular network traffic load, thereby effectively predicting the traffic load. Compared with the limitation of a single model relying on local feature extraction, the present invention utilizes the complementarity between sub-models to provide a more comprehensive feature expression basis for the subsequent fusion stage. Improve prediction accuracy. A dynamic weight allocation strategy is adopted to perform weighted fusion of the output results of multiple models according to the prediction confidence of each sub-model in different spatiotemporal scenarios to avoid the error accumulation caused by the deviation of a single model. The introduction of prediction error and correction mechanism makes the entire model more robust. The sub-model weights are adjusted by real-time monitoring of the prediction deviation of each sub-model, effectively suppressing the impact of abnormal traffic fluctuations on the overall prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0029] Figure 1 A flow chart of a cellular traffic prediction method based on a multi-layer meta-learning model provided by one embodiment of the present invention;

[0030] Figure 2 A diagram showing the structure of a multi-layer meta-learning model according to an embodiment of the present invention;

[0031] Figure 3 A workflow diagram based on a multi-layer meta-learning model provided for one embodiment of the present invention;

[0032] Figure 4 A diagram showing the structure of a main model in a multi-layer meta-learning model provided by one embodiment of the present invention;

[0033] Figure 5 A diagram showing the relationship between R2 and K value for a multi-layer meta-learning model provided by one embodiment of the present invention;

[0034] Figure 6 A relationship diagram between MAE and K value of a multi-layer meta-learning model provided for one embodiment of the present invention. DETAILED DESCRIPTION

[0035] The following further describes the implementation methods of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than an exhaustive list of all embodiments. It should be noted that the embodiments and features in the embodiments of the present application can be combined with each other unless there is a conflict.

[0036] The terms "first," "second," and the like, if any, in the specification and claims are used to distinguish similar objects and are not necessarily used to describe a particular order or sequential sequence. It should be understood that the terms used in this manner are interchangeable where appropriate so that the embodiments described herein can be implemented in sequences other than those illustrated or described herein. In addition, the terms "including" and "having," and any variations thereof, are intended to cover non-exclusive inclusions, e.g., a process, method, system, product, or apparatus that includes a series of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, product, or apparatus.

[0037] It should be understood that the term "and / or" as used herein is merely a description of the relationship between associated objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A exists alone, A and B exist simultaneously, or B exists alone. Furthermore, the character " / " in this document generally indicates that the associated objects are in an "or" relationship.

[0038] Example 1

[0039] The cellular traffic prediction method based on the multi-layer meta-learning model provided by the present invention is as follows: Figure 1 As shown, the method includes: step S01: constructing a multi-layer meta-learning model, wherein: the first layer is a main model, which is used to generate weights based on the frequency domain information of the cellular network traffic load; the second layer is a weight allocation model, which is used to map the corresponding sub-models according to the weights; and the third layer is a sub-model, which is used to predict the traffic load of the next time period; step S02: collecting cellular network traffic load data, pre-processing the cellular network traffic load data, and using it to train the main model and the sub-model to obtain a trained multi-layer meta-learning model; step S03: using the trained multi-layer meta-learning model to predict the cellular network traffic load, calculate the traffic load prediction value, and obtain predicted load information. In step S04, the prediction error between the predicted cellular network traffic load at the previous moment and the actual cellular network traffic load is determined. When the prediction error is greater than or equal to the error threshold, the main model is retrained, the weights are updated, and the predicted load information is re-output based on the weights.

[0040] By building a multi-layer meta-learning model, using multiple sub-models for parallel prediction, and fusing the prediction results, the overall model accuracy is improved. A correction mechanism based on prediction errors is added to make the entire model more robust.

[0041] 1. Building a multi-layer meta-learning model

[0042] The structure of the multi-layer meta-learning model is as follows Figure 2 As shown, the multi-layer meta-learning model of the present invention is divided into three layers. The first-layer main model mainly generates weights through the frequency domain information of the cellular network traffic load. Since the frequency domain information does not contain time scalars, it can be processed by the DNN network, and the soft classification weights are generated by the GMM simulated by the neural network. The second layer is mapped to different sub-models through weights, and the third layer predicts the traffic load of the next time period through different types of sub-models. The sub-model needs to learn the cellular network traffic load data, and a multi-layer prediction model based on the GRU network is used as a learner. Finally, the predicted load information is subtracted from the actual load information to obtain the prediction error. When the prediction error is too large, the sub-model weight will be re-assigned. Figure 2 FFT in it is fast Fourier transform, DNN is deep neural network, GMM is Gaussian mixture model composed of multi-layer neural networks, GRU module is gated recurrent unit module which is an improved recurrent neural network, Live Video is a real-time video dataset type, and Non Video is a non-video dataset type.

[0043] The workflow based on the multi-layer meta-learning model is as follows Figure 3 As shown. It includes the following steps:

[0044] Traffic data preprocessing

[0045] In order to speed up the training process of the traffic load prediction model for each cell network, the elements of the initial traffic load vector of each cell are normalized to the range of [0, 1] by the min-max normalization method:

[0046] (1)

[0047] in, For the Initial network traffic load of each cell, is the initial network traffic load vector The value of the largest element in is the initial network traffic load vector The value of the smallest element in is the tth element in the initial network traffic load vector, After normalization The network traffic load vector of each cell.

[0048] Use FFT (Fast Fourier Transform) method to generate frequency domain information of different cellular network traffic loads. The normalized network traffic load vector is used to construct a discrete periodic signal.

[0049] (2)

[0050] in, Cells after periodic processing To facilitate the analysis of cellular network traffic load, the entire time span of the dataset is divided into continuous equal time intervals. Indicates the In this embodiment, preferably T=168 (one week is 168 hours).

[0051] (3)

[0052] in, and is the imaginary unit, is the coefficient of the frequency component, is the fast Fourier transform process.

[0053] After calculation, the five frequency components of this embodiment are (when hour), (when hour), (when hour), (when time), and (when ), the five frequency components correspond to the Sinusoidal signals with cellular periods of one week, one day, 12 hours, 8 hours, and 6 hours are used to represent the frequency domain information of the normalized network traffic load vector of each cellular network and are used for subsequent model training and prediction.

[0054] Sub-model pre-training

[0055] The multi-layer meta-learning model provided by the present invention includes multiple sub-models, each of which is composed of K GRU neural networks. Each sub-model is in parallel relationship, and each sub-model is trained using different types of network traffic load data. Figure 2 As shown in , sub-model 1 is trained on real-time video data; sub-model k is trained on non-video data.

[0056] The forward propagation process of a single GRU neural network training is as follows

[0057] (4)

[0058] (5)

[0059] (6)

[0060] (7)

[0061] in, To input information, is the Sigmoid function, For Hadamard, Activation for the tanh function, For the update gate (control how much of the previous moment's information enters the current state), Reset gate (controls how much of the previous state is written into the current candidate state) middle), 、 、 Different gates correspond to different learning parameters, is the input state at time t-1, is the output state at time t.

[0062] Main model pre-training

[0063] The main model is trained using the frequency domain information of the cellular network traffic load. The main model consists of a DNN network (deep neural network) and a GMM network (Gaussian mixture model). The main model structure is as follows: Figure 4shown. Figure 4 in For the The frequency domain information vector of each cell.

[0064] The DNN network consists of a deep autoencoder, frequency domain information After the encoder reduces the dimension, the low-dimensional samples are obtained , Then decode the sample to get .

[0065] (8)

[0066] (9)

[0067] in, is the encoder parameter, For The process of encoding the encoder for the parameters, is the decoder parameter, For The process of decoding for the decoder of parameters.

[0068] Output of the DNN network Contains two variables, including: For low-dimensional samples, is the reconstruction error ( and reconstruction error between ).

[0069] (10)

[0070] (11)

[0071] in, is a 2-dimensional feature, is the Euclidean distance, is the cosine similarity.

[0072] (12)

[0073] (13)

[0074] The GMM network (Gaussian mixture model) is a clustering algorithm that performs density estimation by predicting the mixture membership of each sample. Therefore The output of the multi-layer neural network with parameters, It is a K-dimensional vector, and K is equal to the number of sub-models.

[0075] Given N data samples, ∀1≤k≤K, the parameters in GMM are as follows:

[0076] (14)

[0077] (15)

[0078] (16)

[0079] in, is the mixing probability, is the mean, is the variance, For the input samples In the Density estimation under Gaussian mixture model components.

[0080] The constraint function during the entire main model training process is:

[0081] (17)

[0082] (18)

[0083] in, is the reconstruction error, For the The DNN network of the cell output The degree of abnormality, is a minimum value (applied in the training process of Gaussian mixture model to prevent the values ​​on the diagonal of the covariance matrix from becoming 0, which makes the matrix irreversible). and are two constant coefficients respectively.

[0084] Finally, the output of the GMM network is the fit coefficient , It is also the output of the entire main model. Assume there are K sub-models:

[0085] (19)

[0086] For the The frequency domain information vector of cells Sub-model weights in sub-model 1, sub-model 2, ..., sub-model k (1≤k≤K).

[0087] Traffic Forecast

[0088] After the main model and sub-model are trained, the output value of sub-model k (1≤k≤K) is used to predict the first The traffic load of each cell.

[0089] Assuming there are K sub-models, first The frequency domain information vector of each cell is input into the main model to obtain the sub-model weight Then, the traffic information at time t is input into K sub-models at the same time and parallel calculation is performed to obtain the first Traffic load prediction value of each cell at time t+1 .

[0090] (20)

[0091] Will and Multiply them together to get the predicted load information and traffic load prediction value. .

[0092] (twenty one)

[0093] Prediction error and correction

[0094] The distance between the actual load information and the predicted load information is calculated, and the sub-model weights are adjusted.

[0095] (twenty two)

[0096] in, is the mean absolute error (forecast error), is the actual cellular network traffic load value at time t+1, is the predicted value of cellular network traffic load at time t+1.

[0097] If the prediction error ≥Error threshold ( It can be set according to actual needs). The main model will be retrained and the fit coefficient will be updated. Then re-predict the cellular network traffic load and calculate the prediction error.

[0098] If the prediction error <Error threshold , then the cellular network traffic load prediction is completed, the prediction results are output, and the cellular traffic prediction based on the multi-layer meta-learning model is completed.

[0099] The present invention also provides a cellular network traffic load prediction system based on a multi-layer meta-learning model, comprising:

[0100] Model construction module: used to build a multi-layer meta-learning model, where the first layer is a main model, used to generate weights based on the frequency domain information and load information of the cellular network traffic load; the second layer is a weight allocation model, used to map the corresponding sub-models according to the weights; the third layer is a sub-model, used to predict the traffic load of the next time period;

[0101] Model training module: used for collecting cellular network traffic load data, preprocessing the cellular network traffic load data, and training the main model and the sub-model to obtain a trained multi-layer meta-learning model;

[0102] Result output module: used to use the trained multi-layer meta-learning model to predict the cellular network traffic load, calculate the cellular network traffic load prediction value, and obtain and output the predicted load information.

[0103] It also includes a prediction error and correction module: used to determine whether to output the predicted load information based on the prediction error between the predicted load information at the previous moment and the actual load;

[0104] When the prediction error at the current moment is greater than or equal to the error threshold, the main model is retrained to update the weights and the predicted load information is re-output.

[0105] Simulation results of cellular traffic prediction method based on multi-layer meta-learning model

[0106] 1. Experimental environment and parameter settings

[0107] The experimental environment runs on Windows 10 (64-bit), an Intel Core i7-8750H CPU, 32GB of RAM, and a 932GB mechanical hard drive. Python 3.6.0 is used as the development language, and the machine learning environment includes numpy 1.19.5, tensorflow 1.2.0, pandas 1.1.5, torch 1.7.1, and Scikit-learn 1.0.

[0108] The simulation used the BDC mobile network dataset, provided by Telecom Italia's "Big Data Challenge" project. The dataset contains approximately 3 million mobile traffic records generated in Milan between November 1, 2013, and January 1, 2014. Specifically, the city of Milan is divided into 10,000 grids, each measuring 235m x 235m. Each traffic record contains a timestamp, grid ID, and mobile traffic load (i.e., data volume).

[0109] Table 1 Parameter settings

[0110]

[0111] 2. Simulation

[0112] To evaluate the prediction effect, two indicators, mean absolute error (MAE) and coefficient of determination (R2), were used. MAE reflects the closeness between the predicted value and the true value, while R2 measures the fit between the predicted value and the true value.

[0113] ;

[0114] in, Let mean square error (prediction error), is the actual traffic load value at time t+1, is the traffic load prediction value at time t+1.

[0115] ;

[0116] Where R2 is the coefficient of determination, is the actual traffic load value at time t+1, is the traffic load prediction value at time t+1. is the average of the actual traffic load values.

[0117] In the simulation, 16 tests were conducted based on the BDC mobile network dataset. The test results are as follows: Figure 5 and Figure 6 ,in, Figure 5 The horizontal axis is the number of sub-models K, and the vertical axis is the initial draw R2 value. Figure 6 The horizontal axis represents the number of sub-models, K, and the vertical axis represents the initial average MAE value. As can be seen, the greater the number of sub-models, the smaller the model error, the closer the predicted values ​​are to the real data, and the better the model performance.

[0118] The model provided by the present invention is compared with other models in the prior art. When the number of training samples is unified to 840, the performance comparison is shown in Table 2 below.

[0119] Table 2 Simulation results of MAK and R2

[0120]

[0121] The cellular traffic prediction method and system based on the multi-layer meta-learning model provided by the present invention achieves a significant improvement in the prediction accuracy and robustness of cellular network traffic load. The present invention effectively performs traffic prediction by constructing multiple sub-models to parallelly process the frequency domain information, time series traffic data, real load information and other features of the cellular network traffic load. Compared with the limitation of a single model relying on local feature extraction, the present invention utilizes the complementarity between sub-models to provide a more comprehensive feature expression basis for the subsequent fusion stage. Improve prediction accuracy. A dynamic weight allocation strategy is adopted to perform weighted fusion of the output results of multiple models according to the prediction confidence of each sub-model in different spatiotemporal scenarios to avoid the error accumulation caused by the deviation of a single model. The introduction of prediction error and correction mechanism makes the entire model more robust. The sub-model weights are adjusted by real-time monitoring of the prediction deviation of each sub-model, effectively suppressing the impact of abnormal traffic fluctuations on the overall prediction.

[0122] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not limitations on the implementation methods of the present invention. For ordinary technicians in the relevant field, other different forms of changes or modifications can be made based on the above description. It is impossible to list all the implementation methods here. Any obvious changes or modifications derived from the technical solution of the present invention are still within the scope of protection of the present invention.

Claims

1. A cellular traffic prediction method based on a multi-layer meta-learning model, characterized in that: include: Step S01: Construct a multi-layer meta-learning model, wherein: the first layer is a main model, which is used to generate weights based on the frequency domain information of the cellular network traffic load; the second layer is a weight distribution model, which is used to map the corresponding sub-models according to the weights; and the third layer is a sub-model, which is used to predict the traffic load of the next time period; Step S02: collecting cellular network traffic load data, preprocessing the cellular network traffic load data, and using the data to train the main model and the sub-model to obtain a trained multi-layer meta-learning model; Step S03: using the trained multi-layer meta-learning model to predict the cellular network traffic load, calculate the traffic load prediction value, and obtain predicted load information; In step S02, the training of the main model is performed according to a constraint function, which is: Among them, θ e is the encoder parameter, θ d is the decoder parameter, z r is the reconstruction error, z p is the output sample, E(z p ) is the output sample z of the DNN network of the pth cell p The degree of abnormality, is a minimum value, is the mixing probability, is the mean, is the variance, exp() represents the natural exponential function, η1 and η2 are constant coefficients, N is the number of input samples, z pi Output z p The i-th vector, E(z pi ) is the output sample z of the DNN network of the pth cell p The abnormality degree of the i-th vector.

2. The cellular traffic prediction method based on the multi-layer meta-learning model according to claim 1, characterized in that In step S01, there are multiple sub-models, and each sub-model is composed of multiple GRU neural networks.

3. The cellular traffic prediction method based on the multi-layer meta-learning model according to claim 1, characterized in that In step S01, the main model includes a DNN network, which is used to reduce the dimension of the frequency domain information of the cellular network traffic load to obtain low-dimensional samples, decode the low-dimensional samples to obtain decoded samples, calculate the reconstruction error between the decoded samples and the frequency domain information, and obtain output samples, wherein the output samples include the reconstruction error and the low-dimensional samples.

4. The cellular traffic prediction method based on a multi-layer meta-learning model according to claim 3, characterized in that In step S01, the main model also includes a GMM network, which is used to perform density estimation on the output samples of the DNN network and obtain weights based on the density estimation, wherein the weights include the weights of the frequency domain information of the cellular network traffic load in the corresponding sub-model.

5. The cellular traffic prediction method based on a multi-layer meta-learning model according to claim 1, characterized in that In step S02, training the sub-model includes using different types of cellular network traffic load data for training, and the cellular network traffic load data includes real-time video data and non-video data.

6. The cellular traffic prediction method based on a multi-layer meta-learning model according to claim 1, characterized in that In step S02, the preprocessing of the cellular network traffic load data includes: normalizing the collected cellular network traffic load data using the minimum-maximum normalization method to obtain a traffic load vector, and processing the traffic load vector using the fast Fourier transform method to generate frequency domain information of the cellular network traffic load, and the frequency domain information is used to train the main model and the sub-model.

7. The cellular traffic prediction method based on a multi-layer meta-learning model according to claim 1, characterized in that: The method also includes step S04, determining the prediction error between the predicted cellular network traffic load at the previous moment and the actual cellular network traffic load; when the prediction error is greater than or equal to the error threshold, retraining the main model, updating the weights, and re-outputting the predicted load information based on the weights.

8. A cellular traffic prediction system based on a multi-layer meta-learning model, characterized in that: include: Model construction module: used to build a multi-layer meta-learning model, where the first layer is a main model, used to generate weights based on the frequency domain information and load information of the cellular network traffic load; the second layer is a weight allocation model, used to map the corresponding sub-models according to the weights; the third layer is a sub-model, used to predict the traffic load of the next time period; Model training module: used to collect cellular network traffic load data, pre-process the cellular network traffic load data, and train the main model and the sub-model to obtain a trained multi-layer meta-learning model; the training of the main model is performed according to a constraint function, and the constraint function is: Among them, θ e is the encoder parameter, θ d is the decoder parameter, z r is the reconstruction error, z p is the output sample, E(z p ) is the output sample z of the DNN network of the pth cell p The degree of abnormality, is a minimum value, is the mixing probability, is the mean, is the variance, exp() represents the natural exponential function, η1 and η2 are constant coefficients, N is the number of input samples, z pi Output z p The i-th vector, E(z pi ) is the output sample z of the DNN network of the pth cell p The abnormality degree of the i-th vector; Result output module: used to use the trained multi-layer meta-learning model to predict cellular network traffic load, calculate traffic load prediction values, and obtain and output predicted load information.

9. The cellular traffic prediction system based on the multi-layer meta-learning model according to claim 8, characterized in that It also includes a prediction error and correction module: used to determine whether to output the predicted load information based on the prediction error between the predicted load information at the previous moment and the actual load; When the prediction error at the current moment is greater than or equal to the error threshold, the main model is retrained to update the weights and the predicted load information is re-output.

Images