Network traffic prediction method and device, electronic equipment and storage medium

Abstract

The invention provides a network traffic prediction method and device, electronic equipment and a storage medium, relates to the technical field of wireless networks, and aims to acquire multi-modal fusion features influencing network traffic according to multi-modal data, realize comprehensive collection of features influencing network traffic in a to-be-predicted area, and improve the prediction efficiency. And the accuracy of subsequently obtaining the business volume density map and predicting the business volume data can be improved. And performing kernel density estimation on the multi-modal fusion features according to the kernel function and the bandwidth parameters to obtain a traffic density map, thereby realizing accurate calculation of the traffic density of the to-be-predicted region. Through the service volume prediction model, feature extraction and feature analysis are automatically performed on the service volume density map and the multi-modal fusion features, the predicted service volume data are obtained, and the efficiency of obtaining the predicted service volume data is improved. According to the invention, the accuracy of obtaining the predicted service volume is improved, and the efficiency of obtaining the predicted service volume is improved according to the service volume prediction model.

Description

Technical Field

[0001] The present invention relates to the field of wireless network technology, and in particular to a method, device, electronic device and storage medium for predicting network traffic. Background Art

[0002] With the rapid development of wireless communication technology and the continuous increase in the number of mobile Internet users, the wireless network environment has become more and more complex, and the fluctuation and unpredictability of traffic volume have also increased. In this environment, traditional wireless network traffic forecasting technologies often fail to meet the requirements of real-time and accuracy. These traditional technologies usually rely on a single data source (such as historical traffic data) or a simple statistical model, which makes it difficult to accurately capture and analyze the complex dynamics of traffic changes.

[0003] In summary, the existing prediction of network traffic is not accurate. Summary of the invention

[0004] The present invention provides a network traffic prediction method, device, electronic device and storage medium, which are used to solve the defect that the prediction of network traffic in the prior art is not accurate, and improve the accuracy of network traffic prediction.

[0005] The present invention provides a method for predicting network traffic, comprising: extracting and fusing features of multimodal data that affects network traffic in a predicted area to obtain multimodal fusion features; determining a kernel function and a bandwidth parameter based on the distribution form of the multimodal fusion features; performing kernel density estimation on the multimodal fusion features based on the kernel function and the bandwidth parameters to obtain a traffic density map; inputting the traffic density map and the multimodal fusion features into a traffic prediction model to obtain predicted traffic data output by the traffic prediction model, wherein the traffic prediction model is obtained by training based on a preset deep learning model based on sample traffic density maps, sample multimodal fusion features and labels of sample predicted traffic data.

[0006] According to the network traffic prediction method provided by the present invention, multimodal data is obtained based on the following steps: based on satellite images and urban monitoring systems, the geographical layout of the area to be predicted, the building density of the area to be predicted and the traffic flow of the area to be predicted are obtained to obtain video data of the area to be predicted; based on the urban noise monitoring system, the environmental noise level of the area to be predicted at different times and places is obtained to obtain audio data of the area to be predicted; based on the public data of the public platform, text data affecting the communication needs of the area to be predicted is obtained; based on the video data, audio data and text data, multimodal data is obtained.

[0007] According to the network traffic prediction method provided by the present invention, feature extraction and feature fusion are performed on multimodal data that affect the network traffic in the prediction area to obtain multimodal fusion features, including: preprocessing the multimodal data to obtain preprocessed video data, preprocessed audio data and preprocessed text data, and the preprocessing includes unified formatting processing, data cleaning and standardization processing; based on the image feature processing module of the contrastive language-image pre-training CLIP model, feature extraction is performed on the preprocessed video data to obtain image space features; based on the text feature processing module of the CLIP model, semantic features and context features are extracted from the preprocessed text data to obtain event semantic features; based on the multi-layer perception model, feature extraction of noise intensity is performed on the preprocessed audio data to obtain noise features; image space features, event semantic features and noise features are fused to obtain multimodal fusion features.

[0008] According to the network traffic prediction method provided by the present invention, based on the distribution form of multimodal fusion features, a kernel function and a bandwidth parameter are determined, including: when the distribution form is a normal distribution, the kernel function is determined to be a Gaussian kernel function, and the bandwidth parameter of the Gaussian kernel function is determined based on grid search and cross validation; when the distribution form is a concentrated distribution, the kernel function is determined to be an Epanechnikov kernel function, and the bandwidth parameter of the Epanechnikov kernel function is determined based on grid search and cross validation; when the distribution form is a long-tail distribution, the kernel function is determined to be a hyperbolic tangent kernel function, and the bandwidth parameter of the hyperbolic tangent kernel function is determined based on grid search and cross validation.

[0009] According to the network traffic prediction method provided by the present invention, the multimodal fusion feature includes multiple feature vectors, and kernel density estimation is performed on the multimodal fusion feature based on the kernel function and the bandwidth parameter to obtain a traffic density map, including: taking the kernel function as the center and the bandwidth parameter as the radius, obtaining a probability density estimation value of each feature vector; and obtaining a traffic density map based on the probability density estimation values ​​of all feature vectors.

[0010] According to the network traffic prediction method provided by the present invention, the traffic prediction model is determined based on the following steps: obtaining labels of sample predicted traffic data according to a sample traffic heat map, a sample traffic prediction value and a sample traffic peak time period; marking the sample traffic density map and the sample multimodal fusion features based on the labels of the sample predicted traffic data to obtain labeled training samples; training, testing and verifying a preset deep learning model based on the training samples until the error output by the preset deep learning model is less than the set error, thereby obtaining a traffic prediction model.

[0011] According to the network traffic prediction method provided by the present invention, the bandwidth parameter of the Gaussian kernel function is determined based on grid search and cross validation, including: obtaining multiple initial bandwidth parameters based on the numerical range of multimodal fusion features; taking each initial bandwidth parameter as a grid, and performing grid search and cross validation on each grid according to the multimodal fusion features and the Gaussian kernel function to obtain an evaluation value of each initial bandwidth parameter; and taking the initial bandwidth parameter with the highest evaluation value as the bandwidth parameter.

[0012] The present invention also provides a network traffic prediction device, comprising: a feature extraction module, used to extract and fuse features of multimodal data that affects the network traffic in the prediction area to obtain multimodal fusion features; a determination module, used to determine the kernel function and bandwidth parameters based on the distribution form of the multimodal fusion features; a traffic density map acquisition module, used to perform kernel density estimation on the multimodal fusion features based on the kernel function and bandwidth parameters to obtain a traffic density map; a prediction module, used to input the traffic density map and the multimodal fusion features into a traffic prediction model, and obtain predicted traffic data output by the traffic prediction model, wherein the traffic prediction model is obtained by label training based on a preset deep learning model and based on sample traffic density maps, sample multimodal fusion features and sample predicted traffic data.

[0013] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, any of the above-mentioned network traffic prediction methods is implemented.

[0014] The present invention also provides a non-transitory computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the method for predicting network traffic as described above is implemented.

[0015] The network traffic prediction method, device, electronic device and storage medium provided by the present invention obtain multimodal fusion features that affect network traffic based on multimodal data, and realize comprehensive collection of features that affect network traffic in the area to be predicted, which is conducive to improving the accuracy of subsequent acquisition of traffic density maps and predicted traffic data. Kernel density estimation is performed on multimodal fusion features according to kernel functions and bandwidth parameters to obtain traffic density maps, thereby realizing accurate calculation of traffic density in the area to be predicted. Through the traffic prediction model, feature extraction and feature analysis are automatically performed on traffic density maps and multimodal fusion features to obtain predicted traffic data, thereby improving the efficiency of obtaining predicted traffic data. The present invention improves the accuracy of obtaining predicted traffic based on multimodal fusion features and kernel density estimation, and improves the efficiency of obtaining predicted traffic based on the traffic prediction model. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0017] Figure 1 This is one of the flow charts of the network traffic prediction method provided by the present invention.

[0018] Figure 2 This is the second flow chart of the network traffic prediction method provided by the present invention.

[0019] Figure 3 It is a structural schematic diagram of the network traffic prediction device provided by the present invention.

[0020] Figure 4 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0021] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0022] Combine the following Figure 1-Figure 4 The present invention describes a method, device and electronic device for predicting network traffic.

[0023] Figure 1 FIG. 1 is one of the flow charts of the network traffic prediction method provided by the present invention. Figure 1 As shown, the network traffic prediction method includes S100 to S400, and each step is specifically as follows.

[0024] S100: extracting and fusing features of multimodal data that affects network traffic in the predicted area to obtain multimodal fusion features.

[0025] Multimodal data is data with multiple modes, such as video data, audio data, and text data. These multimodal data can capture different types of factors affecting regional network traffic, such as video data (capturing factors affecting urban layout and traffic flow), audio data (capturing factors affecting urban noise levels), and text data (capturing factors affecting events and activity information).

[0026] The data source of each modality reflects the influencing factors directly related to the network traffic. For example, video data can reflect the density and activity of the area; audio data reflects the intensity of urban activities through noise levels; and text data provides information about events that may affect communication needs. These factors will affect the wireless network traffic in the area. Therefore, by extracting and fusing features of multimodal data, multimodal fusion features are obtained, which can then more comprehensively predict the traffic.

[0027] S200: Determine a kernel function and a bandwidth parameter based on the distribution form of the multimodal fusion feature.

[0028] Multimodal fusion features include multiple feature vectors, each of which constitutes a feature point. In the kernel density estimation process, the kernel function is used to calculate the density near the feature point. The bandwidth parameter defines the smoothness of the kernel density estimation, that is, the sensitivity to the local area. The larger the bandwidth parameter, the smoother the estimation result, which is suitable for the analysis of macro trends; the smaller the bandwidth parameter, the more delicate the estimation result, which is more suitable for capturing micro details.

[0029] According to the distribution form and value range of the multimodal fusion features of the area to be predicted, the kernel function and bandwidth parameters are determined to ensure the accuracy of kernel density estimation of the multimodal fusion features according to the kernel function and bandwidth parameters.

[0030] S300: performing kernel density estimation on the multimodal fusion features based on the kernel function and the bandwidth parameter to obtain a traffic density map.

[0031] The multimodal fusion feature includes multiple feature vectors. Kernel density estimation is performed on the multimodal fusion feature based on the kernel function and the bandwidth parameter to obtain a traffic density map. Specifically, the probability density estimation value of each feature vector is obtained with the kernel function as the center and the bandwidth parameter as the radius; the traffic density map is obtained based on the probability density estimation values ​​of all feature vectors.

[0032] According to the kernel function and bandwidth parameters, the density of each feature vector in the multimodal fusion feature is estimated to obtain the probability density estimate of each feature vector. According to all the probability density estimates of all feature vectors, the traffic density map of the area to be predicted is obtained.

[0033] The present invention calculates the traffic density map of the multimodal fusion feature based on kernel density estimation, and can effectively capture the local and global features of the multimodal fusion feature.

[0034] S400: Inputting the traffic density map and the multimodal fusion features into a traffic prediction model to obtain predicted traffic data output by the traffic prediction model.

[0035] Among them, the business volume prediction model is obtained by training based on the sample business volume density map, sample multimodal fusion features and label of sample predicted business volume data on the basis of the preset deep learning model.

[0036] The business volume prediction model is determined based on the following steps: according to the sample business volume heat map, the sample business volume prediction value and the sample business peak time period, the label of the sample predicted business volume data is obtained; based on the label of the sample predicted business volume data, the sample business volume density map and the sample multimodal fusion features are marked to obtain the training samples with labels; based on the training samples, the preset deep learning model is trained, tested and verified until the error of the preset deep learning model output is less than the set error, and the business volume prediction model is obtained.

[0037] Determine the sample traffic density map based on the existing historical traffic density data of the area to be predicted. Determine the sample multimodal fusion features based on the existing historical multimodal fusion features of the area to be predicted. Determine the sample traffic heat map based on the existing historical traffic heat map of the area to be predicted. Determine the sample traffic forecast value based on the existing historical traffic values ​​of the area to be predicted. Determine the sample traffic peak time period based on the existing historical traffic peak time period of the area to be predicted.

[0038] According to the sample business volume heat map, sample business volume prediction value and sample business peak time period, obtain the label of the sample predicted business volume data. According to the label of the sample predicted business volume data, mark the sample business volume density map and the sample multimodal fusion features to obtain the training samples with labels. Divide the training samples into training set, test set and validation set according to the preset ratio. Construct a preset deep learning model and train the preset deep learning model according to the training set. Test the trained preset deep learning model according to the test set. Verify the tested preset deep learning model according to the validation set. When the error output by the preset deep learning model is less than the set error, it is determined that the preset deep learning model has been verified and the business volume prediction model is obtained.

[0039] The traffic density map and multimodal fusion features are input into the traffic volume prediction model. The traffic volume prediction model extracts and analyzes the traffic density map and multimodal fusion features to obtain the predicted traffic volume data of the area to be predicted in the future time period. The feature analysis includes time series analysis.

[0040] Furthermore, the predicted traffic volume data is displayed, including the traffic volume heat map, traffic volume prediction value and traffic peak time period. The network configuration is updated according to the traffic volume data.

[0041] The network traffic prediction method provided by the present invention obtains multimodal fusion features that affect network traffic based on multimodal data, realizes comprehensive collection of features that affect network traffic in the area to be predicted, and is conducive to improving the accuracy of subsequent acquisition of traffic density maps and predicted traffic data. Kernel density estimation is performed on multimodal fusion features according to kernel functions and bandwidth parameters to obtain traffic density maps, thereby realizing accurate calculation of traffic density in the area to be predicted. Through the traffic prediction model, feature extraction and feature analysis are automatically performed on traffic density maps and multimodal fusion features to obtain predicted traffic data, thereby improving the efficiency of obtaining predicted traffic data. The present invention improves the accuracy of obtaining predicted traffic based on multimodal fusion features and kernel density estimation, and improves the efficiency of obtaining predicted traffic based on the traffic prediction model.

[0042] Based on the above embodiment, the multimodal data is acquired based on S110 to S140 , and each step is specifically as follows.

[0043] S110: Based on satellite images and the city monitoring system, the geographical layout of the area to be predicted, the building density of the area to be predicted, and the traffic flow of the area to be predicted are obtained to obtain video data of the area to be predicted.

[0044] S120: Based on the urban noise monitoring system, the environmental noise levels of the area to be predicted at different times and locations are obtained to obtain audio data of the area to be predicted.

[0045] S130: Based on the public data of the public platform, obtain text data that affects the communication demand of the area to be predicted.

[0046] S140: Obtain multimodal data based on the video data, the audio data, and the text data.

[0047] The geographical layout of the area to be predicted, the building density of the area to be predicted, and the traffic flow of the area to be predicted are captured from the satellite images of the area to be predicted and the urban monitoring system to obtain the video data of the area to be predicted. The video data can reflect the density and activity of the network traffic volume in the area to be predicted. The environmental noise level of the area to be predicted at different times and locations is obtained from the urban noise monitoring system of the area to be predicted to obtain the audio data of the area to be predicted. The audio data reflects the intensity of the activities in the area to be predicted (such as the city) through the noise level. The text data that affects the communication needs of the area to be predicted is obtained from the public data of the public platform of the area to be predicted. The public platform includes social media platforms, news websites, public organization platforms, etc. The text data includes information about large-scale events, traffic accidents or other events that may affect communication needs. According to the above video data, audio data and text data, multimodal data is obtained.

[0048] Furthermore, the multimodal data also includes other multimodal data such as historical traffic data, time tags, meteorological information, etc. of the area to be predicted. For example, other multimodal data are obtained from network operators and meteorological departments.

[0049] The multimodal data of the present invention includes video data, audio data and text data, which greatly expands the range of characteristics for obtaining network traffic that affects the area to be predicted. It can comprehensively consider factors such as the geographical distribution, social activities, and environmental changes of the area to be predicted, which is conducive to improving the accuracy of obtaining traffic density maps and predicting traffic data.

[0050] Based on the above embodiment, feature extraction and feature fusion are performed on the multimodal data that affects the network traffic in the predicted area to obtain multimodal fusion features, including S150 to S190, and each step is specifically as follows.

[0051] S150: Preprocessing the multimodal data to obtain preprocessed video data, preprocessed audio data, and preprocessed text data, wherein the preprocessing includes unified formatting processing, data cleaning, and standardization processing.

[0052] S160: Based on the image feature processing module of the contrastive language-image pre-training CLIP model, feature extraction is performed on the pre-processed video data to obtain image spatial features.

[0053] S170: A text feature processing module based on the CLIP model extracts semantic features and context features from the preprocessed text data to obtain event semantic features.

[0054] S180: Extract the noise intensity feature of the preprocessed audio data based on the multi-layer perception model to obtain the noise feature.

[0055] S190: Fusing image spatial features, event semantic features, and noise features to obtain multimodal fusion features.

[0056] like Figure 2 As shown, all collected multimodal data are uniformly formatted, cleaned and standardized to improve the quality of multimodal data.

[0057] The Contrastive Language-Image Pretraining (CLIP) model is a deep learning model for processing image and text information simultaneously. The main goal of the CLIP model is to embed images and text into the same feature space through contrastive learning, so that similar images and texts are close to each other in this space, while unrelated images and texts are far away. The CLIP model consists of a visual encoder (image feature processing module) for processing images and a text encoder (text feature processing module) for processing text. These two encoders convert the preprocessed video data and preprocessed text data into feature vectors (including image space features and event semantic features), respectively.

[0058] The noise intensity feature of the preprocessed audio data is extracted according to the Multi-Layer Perceptron (MLP) model to obtain the noise feature.

[0059] The image spatial features, event semantic features and noise features are input into the fusion layer of the neural network to perform data fusion and obtain multimodal fusion features. The multimodal fusion features include multiple feature vectors.

[0060] The present invention obtains image space features and event semantic features through the CLIP model, thereby improving the accuracy of feature extraction of preprocessed video data and preprocessed text data. By fusing image space features, event semantic features and noise features, multimodal fusion features are obtained, ensuring that the features of data of all modalities are evenly and effectively fused, which is conducive to improving the accuracy of subsequent acquisition of traffic density maps and predicted traffic data.

[0061] Based on the above embodiment, the kernel function and bandwidth parameters are determined based on the distribution form of the multimodal fusion features, including S210 to S230, and each step is specifically as follows.

[0062] S210: When the distribution form is a normal distribution, determining that the kernel function is a Gaussian kernel function, and determining a bandwidth parameter of the Gaussian kernel function based on grid search and cross validation.

[0063] S220: When the distribution form is concentrated distribution, determine that the kernel function is an Epanechnikov kernel function, and determine a bandwidth parameter of the Epanechnikov kernel function based on grid search and cross validation.

[0064] S230: When the distribution form is a long-tail distribution, determine that the kernel function is a hyperbolic tangent kernel function, and determine a bandwidth parameter of the hyperbolic tangent kernel function based on grid search and cross validation.

[0065] Based on grid search and cross-validation, the bandwidth parameter of the Gaussian kernel function is determined. Specifically, based on the numerical range of the multimodal fusion feature, multiple initial bandwidth parameters are obtained; each initial bandwidth parameter is taken as a grid, and grid search and cross-validation are performed on each grid according to the multimodal fusion feature and the Gaussian kernel function to obtain an evaluation value of each initial bandwidth parameter; the initial bandwidth parameter with the highest evaluation value is used as the bandwidth parameter.

[0066] When the distribution form is normal distribution, the kernel function is a Gaussian kernel function, and the calculation formula of the Gaussian kernel function is as follows.

[0067] ; in, is the kernel function, is the feature vector of multimodal fusion features.

[0068] The multiple feature vectors of the multimodal fusion features are divided into multiple subsets, each subset is used as a validation set one by one, and the remaining subsets are used as training sets.

[0069] According to the numerical range of the multimodal fusion feature, multiple initial bandwidth parameters of the Gaussian kernel function are obtained. The calculation formula of the initial bandwidth parameter of the Gaussian kernel function is as follows.

[0070] ; in, is the initial bandwidth parameter, is the sample standard deviation of the training set (calculated based on the numerical range of the training set), is the number of samples in the training set.

[0071] Each initial bandwidth parameter is regarded as a grid. According to the validation set, training set and Gaussian kernel function, grid search and cross-validation are performed on each grid to obtain the evaluation value of each initial bandwidth parameter. The initial bandwidth parameter with the highest evaluation value is taken as the bandwidth parameter.

[0072] The grid search and cross validation include: performing kernel density estimation based on the initial bandwidth parameter and Gaussian kernel function corresponding to the training set, and calculating the evaluation value of the initial bandwidth parameter based on the validation set. The training set and validation set are transformed to calculate the evaluation value of each initial bandwidth parameter. The initial bandwidth parameter with the highest evaluation value is used as the bandwidth parameter of the Gaussian kernel function.

[0073] When the distribution form is concentrated distribution, the kernel function is the Epanechnikov kernel function, and the calculation formula of the Epanechnikov kernel function is as follows.

[0074] ; in, is the kernel function, is the feature vector of multimodal fusion features.

[0075] According to the numerical range of multimodal fusion features, multiple initial bandwidth parameters of the Epanechnikov kernel function are obtained. Each initial bandwidth parameter is taken as a grid, and grid search and cross-validation are performed on each grid according to the validation set, training set and Epanechnikov kernel function to obtain the evaluation value of each initial bandwidth parameter; the initial bandwidth parameter with the highest evaluation value is used as the bandwidth parameter of the Epanechnikov kernel function.

[0076] When the distribution is long-tailed, the kernel function is the hyperbolic tangent kernel function. The calculation formula of the hyperbolic tangent kernel function is as follows.

[0077] ; in, is the kernel function, is the feature vector of multimodal fusion features.

[0078] According to the numerical range of the multimodal fusion feature, multiple initial bandwidth parameters of the hyperbolic tangent kernel function are obtained. Each initial bandwidth parameter is taken as a grid, and grid search and cross-validation are performed on each grid according to the validation set, training set and hyperbolic tangent kernel function to obtain the evaluation value of each initial bandwidth parameter; the initial bandwidth parameter with the highest evaluation value is used as the bandwidth parameter of the hyperbolic tangent kernel function.

[0079] The present invention determines the kernel function and bandwidth parameters according to the distribution form and numerical range of the multimodal fusion features, realizes the targeted kernel density estimation of the multimodal fusion features with different distribution forms and numerical ranges, and improves the accuracy of the kernel density estimation of the multimodal fusion features.

[0080] The kernel density estimation adopted in the present invention can dynamically adjust the kernel function and bandwidth parameters. In conjunction with the real-time data update mechanism, the business volume prediction model can quickly respond to changes in the environment and the market, and provide business volume predictions in real time, greatly improving the practical value and timeliness of the predictions.

[0081] The present invention can accurately and timely predict the traffic volume, so that wireless network operators can more scientifically plan network resources, such as base station deployment and spectrum allocation, thereby reducing operating costs and improving resource utilization efficiency while ensuring service quality.

[0082] The present invention is not only applicable to conventional wireless network traffic forecasting, but can also be extended to various scenarios such as emergency communication management, large-scale public event support, etc. For example, during large-scale events such as music festivals or sports events, by accurately predicting the traffic volume in specific time periods and areas, network operators can adjust and optimize network configuration in advance to ensure the continuity and stability of communication services.

[0083] The present invention can significantly improve the service experience of end users through more accurate and timely network service adjustments. In particular, during peak traffic hours and important public events, the optimized network resource configuration can effectively avoid network congestion and improve service quality.

[0084] The detailed traffic volume analysis and prediction results provided by the present invention can provide strong data support for network planners and decision makers, helping them to make more scientific and reasonable decisions, especially in the face of rapidly changing market and technical environments.

[0085] The network traffic prediction device provided by the present invention is described below. The network traffic prediction device described below and the network traffic prediction method described above can be referenced to each other.

[0086] like Figure 3 As shown, a network traffic prediction device includes: a feature extraction module 301, which is used to extract and fuse features of multimodal data that affects the network traffic in the prediction area to obtain multimodal fusion features.

[0087] The determination module 302 is used to determine the kernel function and bandwidth parameters based on the distribution form of the multimodal fusion features.

[0088] The traffic density map acquisition module 303 is used to perform kernel density estimation on the multimodal fusion features based on the kernel function and the bandwidth parameter to acquire the traffic density map.

[0089] The prediction module 304 is used to input the business volume density map and the multimodal fusion features into the business volume prediction model to obtain the predicted business volume data output by the business volume prediction model, wherein the business volume prediction model is obtained by label training based on the sample business volume density map, the sample multimodal fusion features and the sample predicted business volume data on the basis of a preset deep learning model.

[0090] The network traffic prediction device provided by the present invention obtains multimodal fusion features that affect network traffic based on multimodal data, realizes comprehensive collection of features that affect network traffic in the area to be predicted, and is conducive to improving the accuracy of subsequent acquisition of traffic density maps and predicted traffic data. Kernel density estimation is performed on multimodal fusion features according to kernel functions and bandwidth parameters to obtain traffic density maps, thereby realizing accurate calculation of traffic density in the area to be predicted. Through the traffic prediction model, feature extraction and feature analysis are automatically performed on traffic density maps and multimodal fusion features to obtain predicted traffic data, thereby improving the efficiency of obtaining predicted traffic data. The present invention improves the accuracy of obtaining predicted traffic based on multimodal fusion features and kernel density estimation, and improves the efficiency of obtaining predicted traffic based on the traffic prediction model.

[0091] In one embodiment, the feature extraction module 301 is also used to: based on satellite images and urban monitoring systems, obtain the geographical layout of the area to be predicted, the building density of the area to be predicted and the traffic flow of the area to be predicted, and obtain video data of the area to be predicted; based on the urban noise monitoring system, obtain the environmental noise level of the area to be predicted at different times and places, and obtain audio data of the area to be predicted; based on public data on a public platform, obtain text data that affects the communication needs of the area to be predicted; based on video data, audio data and text data, obtain multimodal data.

[0092] In one embodiment, the feature extraction module 301 is used to: preprocess the multimodal data to obtain preprocessed video data, preprocessed audio data and preprocessed text data, wherein the preprocessing includes unified formatting, data cleaning and standardization; based on the image feature processing module of the contrastive language-image pre-training CLIP model, perform feature extraction on the preprocessed video data to obtain image space features; based on the text feature processing module of the CLIP model, perform semantic feature and context feature extraction on the preprocessed text data to obtain event semantic features; based on the multi-layer perception model, perform feature extraction of noise intensity on the preprocessed audio data to obtain noise features; and fuse the image space features, event semantic features and noise features to obtain multimodal fusion features.

[0093] In one embodiment, the determination module 302 is used to: when the distribution form is a normal distribution, determine that the kernel function is a Gaussian kernel function, and determine the bandwidth parameter of the Gaussian kernel function based on grid search and cross validation; when the distribution form is a concentrated distribution, determine that the kernel function is an Epanechnikov kernel function, and determine the bandwidth parameter of the Epanechnikov kernel function based on grid search and cross validation; when the distribution form is a long-tail distribution, determine that the kernel function is a hyperbolic tangent kernel function, and determine the bandwidth parameter of the hyperbolic tangent kernel function based on grid search and cross validation.

[0094] In one embodiment, the multimodal fusion feature includes multiple feature vectors, and the traffic density map acquisition module 303 is used to: obtain the probability density estimate of each feature vector with the kernel function as the center and the bandwidth parameter as the radius; and obtain the traffic density map based on the probability density estimates of all feature vectors.

[0095] In one embodiment, the prediction module 304 is also used to: obtain labels of sample predicted business volume data based on the sample business volume heat map, sample business volume prediction values ​​and sample business peak time period; based on the labels of the sample predicted business volume data, mark the sample business volume density map and sample multimodal fusion features to obtain labeled training samples; train, test and verify the preset deep learning model based on the training samples until the error output by the preset deep learning model is less than the set error, thereby obtaining a business volume prediction model.

[0096] In one embodiment, the determination module 302 is used to: obtain multiple initial bandwidth parameters based on the numerical range of the multimodal fusion feature; take each initial bandwidth parameter as a grid, and perform grid search and cross-validation on each grid according to the multimodal fusion feature and the Gaussian kernel function to obtain an evaluation value of each initial bandwidth parameter; and use the initial bandwidth parameter with the highest evaluation value as the bandwidth parameter.

[0097] Figure 4 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 4As shown, the electronic device may include: a processor 410, a communication interface 420, a memory 430 and a communication bus 440, wherein the processor 410, the communication interface 420 and the memory 430 communicate with each other through the communication bus 440. The processor 410 may call the logic instructions in the memory 430 to execute the network traffic prediction method, which includes: extracting and fusing features of multimodal data affecting the network traffic in the prediction area to obtain multimodal fusion features; determining the kernel function and bandwidth parameters based on the distribution form of the multimodal fusion features; performing kernel density estimation on the multimodal fusion features based on the kernel function and the bandwidth parameters to obtain a traffic density map; inputting the traffic density map and the multimodal fusion features into the traffic prediction model to obtain the predicted traffic data output by the traffic prediction model, wherein the traffic prediction model is obtained based on the sample traffic density map, the sample multimodal fusion features and the label training of the sample predicted traffic data on the basis of the preset deep learning model.

[0098] In addition, the logic instructions in the above-mentioned memory 430 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0099] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the network traffic prediction method provided by the above-mentioned methods, the method comprising: extracting and fusing features of multimodal data that affects the network traffic in the predicted area to obtain multimodal fusion features; determining kernel functions and bandwidth parameters based on the distribution form of the multimodal fusion features; performing kernel density estimation on the multimodal fusion features based on the kernel functions and bandwidth parameters to obtain a traffic density map; inputting the traffic density map and the multimodal fusion features into a traffic prediction model to obtain predicted traffic data output by the traffic prediction model, wherein the traffic prediction model is obtained by label training based on a sample traffic density map, sample multimodal fusion features and sample predicted traffic data on the basis of a preset deep learning model.

[0100] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0101] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0102] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting network traffic, characterized in that: include: Perform feature extraction and feature fusion on the multimodal data that affects the network traffic in the prediction area to obtain multimodal fusion features; Determining a kernel function and a bandwidth parameter based on the distribution form of the multimodal fusion feature; Performing kernel density estimation on the multimodal fusion feature based on the kernel function and the bandwidth parameter to obtain a traffic density map; The traffic density map and the multimodal fusion features are input into a traffic prediction model to obtain predicted traffic data output by the traffic prediction model, wherein the traffic prediction model is trained based on a preset deep learning model based on sample traffic density maps, sample multimodal fusion features and labels of sample predicted traffic data.

2. The method for predicting network traffic according to claim 1, characterized in that: The multimodal data is obtained based on the following steps: Based on satellite images and urban monitoring systems, the geographical layout of the area to be predicted, the building density of the area to be predicted, and the traffic flow of the area to be predicted are obtained to obtain video data of the area to be predicted; Based on the urban noise monitoring system, the environmental noise levels of the area to be predicted at different times and locations are obtained to obtain the audio data of the area to be predicted; Based on the public data of the public platform, obtaining text data affecting the communication demand of the area to be predicted; The multimodal data is obtained based on the video data, the audio data and the text data.

3. The method for predicting network traffic according to claim 2, characterized in that: The step of extracting and fusing the multimodal data that affects the network traffic in the area to be predicted to obtain multimodal fusion features includes: Preprocessing the multimodal data to obtain preprocessed video data, preprocessed audio data, and preprocessed text data, wherein the preprocessing includes unified formatting, data cleaning, and standardization; An image feature processing module based on a contrastive language-image pre-trained CLIP model performs feature extraction on the pre-processed video data to obtain image spatial features; Based on the text feature processing module of the CLIP model, semantic features and context features are extracted from the preprocessed text data to obtain event semantic features; Extracting noise intensity features from the preprocessed audio data based on a multi-layer perception model to obtain noise features; The image spatial feature, the event semantic feature and the noise feature are fused to obtain the multimodal fusion feature.

4. The method for predicting network traffic according to claim 1, characterized in that: The step of determining a kernel function and a bandwidth parameter based on the distribution form of the multimodal fusion feature includes: When the distribution form is a normal distribution, determining that the kernel function is a Gaussian kernel function, and determining a bandwidth parameter of the Gaussian kernel function based on grid search and cross validation; When the distribution form is concentrated distribution, determining that the kernel function is an Epanechnikov kernel function, and determining a bandwidth parameter of the Epanechnikov kernel function based on the grid search and the cross validation; When the distribution form is a long-tail distribution, the kernel function is determined to be a hyperbolic tangent kernel function, and a bandwidth parameter of the hyperbolic tangent kernel function is determined based on the grid search and the cross validation.

5. The method for predicting network traffic according to claim 1, characterized in that: The multimodal fusion feature includes a plurality of feature vectors, and performing kernel density estimation on the multimodal fusion feature based on the kernel function and the bandwidth parameter to obtain a traffic density map includes: Taking the kernel function as the center and the bandwidth parameter as the radius, obtaining a probability density estimate of each of the feature vectors; The traffic density map is obtained based on the probability density estimation values ​​of all the feature vectors.

6. The method for predicting network traffic according to claim 1, characterized in that: The traffic volume prediction model is determined based on the following steps: According to the sample traffic heat map, the sample traffic forecast value and the sample traffic peak time period, obtain the label of the sample forecast traffic data; Based on the label of the sample predicted traffic volume data, the sample traffic volume density map and the sample multimodal fusion feature are marked to obtain a training sample carrying the label; The preset deep learning model is trained, tested and verified based on the training samples until the error output by the preset deep learning model is less than the set error, thereby obtaining the business volume prediction model.

7. The method for predicting network traffic according to claim 4, characterized in that: The step of determining the bandwidth parameter of the Gaussian kernel function based on grid search and cross validation includes: Based on the numerical range of the multimodal fusion feature, obtaining a plurality of initial bandwidth parameters; Taking each of the initial bandwidth parameters as a grid, and performing the grid search and the cross validation on each of the grids according to the multimodal fusion feature and the Gaussian kernel function, to obtain an evaluation value of each of the initial bandwidth parameters; The initial bandwidth parameter with the highest evaluation value is used as the bandwidth parameter.

8. A network traffic prediction device, characterized in that: include: A feature extraction module is used to extract and fuse the multimodal data that affects the network traffic in the predicted area to obtain multimodal fusion features; A determination module, used to determine a kernel function and a bandwidth parameter based on the distribution form of the multimodal fusion feature; A traffic density map acquisition module, used to perform kernel density estimation on the multimodal fusion feature based on the kernel function and the bandwidth parameter to acquire a traffic density map; A prediction module is used to input the business volume density map and the multimodal fusion features into a business volume prediction model to obtain the predicted business volume data output by the business volume prediction model, wherein the business volume prediction model is obtained by label training based on sample business volume density maps, sample multimodal fusion features and sample predicted business volume data on the basis of a preset deep learning model.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the method for predicting network traffic according to any one of claims 1 to 7 is implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for predicting network traffic according to any one of claims 1 to 7 is implemented.

Images