Mobile network flow processing method and device based on general prediction model

Abstract

The invention provides a mobile network traffic processing method and device based on a general prediction model, and the method comprises the steps: segmenting target mobile network traffic corresponding to a target prediction task based on a number element segmentation unit of the general prediction model, and obtaining a mobile number element sequence; based on a space-time mask mechanism of the general prediction model and a task type corresponding to the target prediction task, performing mask processing on the mobile number element sequence to obtain a mask sequence; and based on a prediction task processing layer of the general prediction model, performing task processing on the target environment data corresponding to the mask sequence and the moving number element sequence to obtain target prediction data corresponding to the target prediction task. According to the method, the problems of task splitting, insufficient model generalization, high engineering deployment complexity and the like of a current mobile traffic prediction method are solved; multiple types of mobile traffic prediction tasks can be processed at the same time, and excellent migration ability and task adaptation ability are achieved; and the generalization performance and robustness of the model under multiple cities and multiple tasks are improved.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a method and device for processing mobile network traffic based on a universal prediction model. Background Art

[0002] With the continuous development of mobile communication technology, mobile networks are carrying an ever-increasing amount of data traffic and service demands, covering a wide range of scenarios, from commuting to industrial production, daily life, and entertainment consumption. In this context, accurate prediction of mobile network traffic has become a key means of improving network efficiency and service quality. Through forward-looking predictions of mobile traffic, network operators can proactively identify potential changes in network load, enabling proactive resource scheduling, rational infrastructure deployment, and optimized service policy configuration, ultimately improving network performance and user experience.

[0003] However, most current mainstream mobile traffic prediction methods adopt a "task customization + data customization" approach, designing independent model structures for different tasks or data types. While this "specialization" strategy can achieve good results for specific tasks, its generalization and cross-task adaptability are limited, making it difficult to meet the requirements of multi-task parallelization and rapid deployment in real-world network environments. Furthermore, different tasks or data types often require the design of multiple modules, significantly increasing design complexity and system deployment costs. Therefore, an effective solution is urgently needed to address these issues. Summary of the Invention

[0004] To solve the above problems, the present invention provides a method and device for processing mobile network traffic based on a universal prediction model.

[0005] The present invention provides a mobile network traffic processing method based on a universal prediction model, comprising the following steps.

[0006] The data segmentation unit based on the general prediction model segments the target mobile network traffic corresponding to the target prediction task to obtain a mobile data sequence; Based on the spatiotemporal masking mechanism of the universal prediction model and the task type corresponding to the target prediction task, masking the mobile digital sequence to obtain a mask sequence; Based on the prediction task processing layer of the universal prediction model, task processing is performed on the target environment data corresponding to the mask sequence and the mobile digital element sequence to obtain target prediction data corresponding to the target prediction task.

[0007] According to a mobile network traffic processing method based on a universal prediction model provided by the present invention, the prediction task processing layer includes a noise addition layer, a denoising network and an adaptive condition generation layer; The prediction task processing layer based on the universal prediction model performs task processing on the target environment data corresponding to the mask sequence and the mobile bit sequence to obtain target prediction data corresponding to the target prediction task, including: Based on the noise adding layer, performing noise processing on the masked part of the mask sequence to obtain a noisy masked part; Based on the adaptive condition generation layer, condition generation and feature extraction are performed on the target environment data to obtain environmental condition features; Adaptively adjusting the residual, scale, and offset parameters in the denoising network based on the environmental condition characteristics; Based on the adjusted denoising network and the unmasked part of the mask sequence, the masked part after the noise is added is reconstructed to obtain target prediction data corresponding to the target prediction task.

[0008] According to a mobile network traffic processing method based on a universal prediction model provided by the present invention, the prediction task processing layer further includes a perception alignment mechanism; Before performing condition generation and feature extraction on the target environment data based on the adaptive condition generation layer to obtain the environment condition features, the method further includes: Based on the perception alignment mechanism, context-aware alignment is performed on the target environment data and the target mobile network traffic to obtain aligned target environment data; The step of performing condition generation and feature extraction on the target environment data based on the adaptive condition generation layer to obtain environment condition features includes: Based on the adaptive condition generation layer, condition generation and feature extraction are performed on the aligned target environment data to obtain environment condition features.

[0009] According to a mobile network traffic processing method based on a universal prediction model provided by the present invention, the target environment data includes traffic tokens and a dynamic point of interest sequence; The prediction task processing layer based on the universal prediction model performs task processing on the target environment data corresponding to the mask sequence and the mobile bit sequence to obtain the target prediction data corresponding to the target prediction task, further comprising: Acquiring initial environment data corresponding to the mobile digital sequence, the initial environment data including mobile user data and point of interest distribution data; Extracting static POI features from the POI distribution data and fusing the static POI features with dynamic time tags to obtain dynamic POI features, wherein the dynamic time tags are obtained by projecting timestamps into time embeddings using a multi-layer perceptron network; The mobile user data is segmented to obtain the traffic token, and the dynamic interest point features are segmented to obtain the dynamic interest point sequence.

[0010] According to a mobile network traffic processing method based on a universal prediction model provided by the present invention, the data segmentation unit based on the universal prediction model segments the target mobile network traffic corresponding to the target prediction task to obtain a mobile data sequence, and further includes: Constructing a first training set based on first sample mobile network traffic of different task types and different data sources, wherein the first training set includes multiple sample task sets, and the sample task sets include first sample tasks and first sample mobile network traffic corresponding to the first sample tasks; The first training set is used to train the denoising network in the prediction task processing layer, and the loss function for training the denoising network is a reconstruction error loss function.

[0011] According to a mobile network traffic processing method based on a universal prediction model provided by the present invention, the data segmentation unit based on the universal prediction model segments the target mobile network traffic corresponding to the target prediction task to obtain a mobile data sequence, and further includes: Based on different task types and second sample mobile network traffic from different data sources and their corresponding sample target environment data, a second training set is constructed, wherein the second training set includes multiple positive sample task sets and multiple negative sample task sets, wherein the positive sample task set includes the second sample task and the positive sample corresponding to the second sample task, and the positive sample includes the second sample mobile network traffic and the sample target environment data in the same spatiotemporal block; the negative sample task set includes the third sample task and the negative sample corresponding to the third sample task, and the negative sample includes the second sample mobile network traffic and the sample target environment data in the same spatiotemporal block; Based on each of the negative samples and each of the positive samples, a perceptual alignment mechanism in the prediction task processing layer is trained, wherein a goal of training the perceptual alignment mechanism is to minimize contrastive learning loss; Based on each of the positive sample task sets and each of the negative sample task sets, the universal prediction model is fine-tuned, and the loss function for fine-tuning the universal prediction model is a comprehensive loss function, which includes a reconstruction error loss function and an alignment loss function.

[0012] According to a mobile network traffic processing method based on a universal prediction model provided by the present invention, the data element segmentation unit based on the universal prediction model segments the target mobile network traffic corresponding to the target prediction task to obtain a mobile data element sequence, including: Based on the digital element segmentation unit, a time-space word segmentation method is adopted to segment the target mobile network traffic according to the minimum time-space unit to obtain multiple mobile digital elements with uniform time dimension, spatial section length and spatial section width. The multiple mobile digital elements are sequenced in chronological order to form the mobile digital element sequence.

[0013] According to a mobile network traffic processing method based on a universal prediction model provided by the present invention, the spatiotemporal masking mechanism based on the universal prediction model and the task type corresponding to the target prediction task is used to mask the mobile data sequence to obtain a mask sequence, including: When the task type is a short-term prediction task, masking the mobile bit sequence based on the short-term mask corresponding to the short-term prediction task in the spatiotemporal mask mechanism to obtain the mask sequence; When the task type is a long-term prediction task, masking the mobile digital sequence based on the long-term mask corresponding to the long-term prediction task in the spatiotemporal mask mechanism to obtain the mask sequence; When the task type is a data generation task, the mobile bit sequence is masked based on the generation mask corresponding to the data generation task in the spatiotemporal mask mechanism to obtain the mask sequence, and each bit of the generation mask is 0.

[0014] The present invention also provides a mobile network traffic processing device based on a universal prediction model, comprising the following modules: A segmentation module is configured as a data segmentation unit based on a general prediction model, which segments the target mobile network traffic corresponding to the target prediction task to obtain a mobile data sequence; a masking module configured to perform masking on the mobile digital sequence based on the spatiotemporal masking mechanism of the universal prediction model and the task type corresponding to the target prediction task to obtain a masked sequence; The prediction module is configured to perform task processing on the target environment data corresponding to the mask sequence and the mobile digital element sequence based on the prediction task processing layer of the universal prediction model to obtain target prediction data corresponding to the target prediction task.

[0015] 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. When the processor executes the computer program, the method for processing mobile network traffic based on a general prediction model as described above is implemented.

[0016] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for processing mobile network traffic based on any of the general prediction models described above is implemented.

[0017] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements any of the above-mentioned methods for processing mobile network traffic based on the universal prediction model.

[0018] The present invention provides a mobile network traffic processing method and device based on a universal prediction model. The method and device segment the target mobile network traffic corresponding to the target prediction task through a digital segmentation unit based on the universal prediction model to obtain a mobile digital sequence; based on the spatiotemporal masking mechanism of the universal prediction model and the task type corresponding to the target prediction task, the mobile digital sequence is masked to obtain a mask sequence; based on the prediction task processing layer of the universal prediction model, the target environment data corresponding to the mask sequence and the mobile digital sequence are task processed to obtain target prediction data corresponding to the target prediction task. The present invention solves the outstanding problems of current mobile traffic prediction methods in terms of task splitting, insufficient model generalization, and high complexity of engineering deployment; the present invention can simultaneously process multiple types of mobile traffic prediction tasks and has excellent migration and task adaptation capabilities; by introducing a spatiotemporal masking strategy to achieve task differentiation and adaptation, as well as the ability to model the relationship between urban environment and traffic changes, the generalization performance and robustness of the model in multiple cities and multiple tasks are significantly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction is given below to 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 any creative work.

[0020] Figure 1 This is one of the flow charts of the mobile network traffic processing method based on the universal prediction model provided by the present invention.

[0021] Figure 2 This is the second flow chart of the mobile network traffic processing method based on the universal prediction model provided by the present invention.

[0022] Figure 3 It is a flowchart of the training of the general prediction model provided by the present invention.

[0023] Figure 4 This is one of the effect diagrams of mobile network traffic processing provided by the present invention; Figure 5 This is the second effect diagram of the mobile network traffic processing provided by the present invention; Figure 6 It is a structural diagram of a mobile network traffic processing device based on a universal prediction model provided by the present invention.

[0024] Figure 7 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0025] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0026] The following combination Figure 1-Figure 7 The present invention describes a method and device for processing mobile network traffic based on a universal prediction model.

[0027] First, the relevant contents of the present invention are briefly described.

[0028] On the one hand, mobile traffic data is inherently heterogeneous, coming from diverse sources and sampling granularities. This includes user-level data collected at millisecond intervals, such as Measurement Reports (MRs), as well as base station performance statistics with a 15-minute granularity, such as Product Manager (PM) data. This lacks a unified structural representation. Furthermore, cities vary significantly in geography, population distribution, and communication needs, resulting in traffic data exhibiting complex spatiotemporal variations. These factors often result in insufficient generalization of existing models in cross-city scenarios, limiting their portability for multi-regional deployments.

[0029] On the other hand, different network optimization tasks place different emphasis on traffic forecasting. For example, short-term forecasting is often used for wireless resource scheduling, requiring real-time awareness of traffic fluctuations and often relying on autoregressive modeling or event-driven mechanisms. Long-term forecasting, on the other hand, is often used for base station deployment or capacity planning, focusing more on traffic periodicity and long-term evolution trends, often employing techniques such as time series decomposition.

[0030] Due to the above differences, current models often need to design multiple sub-modules for different tasks, which significantly increases the complexity of model design and system deployment costs.

[0031] Although in recent years there have been studies exploring the use of new methods such as Transformers and diffusion models for mobile traffic modeling, related work has focused on single-task optimization, lacking a unified modeling framework system design. Existing research attempts to convert mobile traffic into a natural language-like format to transfer language model capabilities, but such methods are highly dependent on artificially constructed prompts and have limitations in semantic understanding and generalization ability. Overall, current mobile traffic prediction models face two major bottlenecks: one is the limited generalization ability, making it difficult to transfer and use across different cities; the other is the lack of task adaptation ability, lacking a unified modeling capability to support multiple tasks.

[0032] In recent years, large-scale foundation models have made breakthroughs in natural language processing and computer vision, demonstrating strong data modeling capabilities, generalization abilities, and zero-shot / small-shot learning abilities. Multiple professional fields such as medicine, urban transportation, and finance have begun to build foundation models tailored to specific data to support complex needs such as multi-task collaboration and cross-scenario transfer. Mobile networks naturally have the potential to develop a unified modeling paradigm, as they aggregate vast amounts of traffic, user behavior, and spatial geographic data. However, so far there has been no general prediction model specifically optimized for mobile network tasks.

[0033] Therefore, it is necessary to build a mobile network traffic prediction universal model with unified feature extraction capability and multi-task adaptation capability, which is crucial for promoting the evolution of intelligent wireless networks. Such a model should be able to fully exploit the common features of large-scale mobile traffic data, support multiple prediction tasks (short-term prediction, long-term trend, data generation), have good cross-city generalization performance and engineering deployment efficiency, and provide unified technical support for network planning and optimization in multiple scenarios.

[0034] Figure 1 is one of the flowcharts of the mobile network traffic processing method based on the universal prediction model provided by the application, as shown in Figure 1 The method comprises steps 101 and 103.

[0035] Step 101: Based on the universal prediction model, the number element segmentation unit divides the target mobile network traffic corresponding to the target prediction task to obtain a mobile number element sequence.

[0036] Specifically, the universal prediction model refers to a universal model of mobile traffic prediction (UoMo) for wireless network optimization. The target mobile network traffic refers to the mobile network data being processed, i.e., mobile traffic data. The mobile number element sequence refers to a mobile number element sorted in chronological order.

[0037] In practical applications, mobile network traffic has different variation characteristics in time and space. In order to uniformly model mobile network traffic in different scenarios, see Figure 2 , Figure 2 This is the second flow chart of the mobile network traffic processing method based on the universal prediction model provided by the present invention: the digital element segmentation unit in the universal prediction model can perform digital element segmentation on the target mobile network traffic to obtain unified mobile digital elements, and then obtain a mobile digital element sequence.

[0038] It should be noted that after obtaining the moving digit sequence, see Figure 2 , the mobile digital sequence can be input into the digital encoder to obtain the encoded mobile digital sequence, which is then input into the prediction task processing layer (diffusion model) of the general prediction model.

[0039] Step 102: Based on the spatiotemporal masking mechanism of the general prediction model and the task type corresponding to the target prediction task, mask processing is performed on the mobile digital sequence to obtain a mask sequence.

[0040] Specifically, see Figure 2 ,The spatiotemporal masking mechanism is a unit for performing multi-scale spatiotemporal masking.

[0041] In order to achieve unified modeling of different prediction tasks, a task-oriented spatiotemporal masking mechanism is constructed in the general prediction model.

[0042] The spatiotemporal masking mechanism can mask the mobile element sequence based on the task type corresponding to the target prediction task, thereby obtaining the corresponding mask sequence.

[0043] Through this spatiotemporal masking mechanism, the universal prediction model can handle multiple tasks using a unified input structure without modifying the main architecture or introducing additional branches. Furthermore, the spatiotemporal masking mechanism not only supports target selection during task training but also provides explicit guidance for task selection during inference.

[0044] Step 103: Based on the prediction task processing layer of the general prediction model, task processing is performed on the target environment data corresponding to the mask sequence and the mobile element sequence to obtain target prediction data corresponding to the target prediction task.

[0045] Specifically, the prediction task processing layer may include a diffusion model, and the diffusion model includes a noise addition layer and a denoising network.

[0046] In actual applications, after obtaining the target environment data corresponding to the mask sequence and the moving element sequence, the prediction task processing layer of the general prediction model performs prediction based on the target environment data corresponding to the mask sequence and the moving element sequence, thereby obtaining the target prediction data corresponding to the target prediction task.

[0047] In addition, after obtaining the target prediction data, the optimization interface can be called based on the target prediction task and target prediction data to improve network operation efficiency and service quality.

[0048] The mobile network traffic processing method based on the universal prediction model provided by the present invention divides the target mobile network traffic corresponding to the target prediction task through a digital segmentation unit based on the universal prediction model to obtain a mobile digital sequence; based on the spatiotemporal mask mechanism of the universal prediction model and the task type corresponding to the target prediction task, the mobile digital sequence is masked to obtain a mask sequence; based on the prediction task processing layer of the universal prediction model, the target environment data corresponding to the mask sequence and the mobile digital sequence are task processed to obtain the target prediction data corresponding to the target prediction task. The present invention solves the outstanding problems of current mobile traffic prediction methods in terms of task splitting, insufficient model generalization, and high complexity of engineering deployment; the present invention can process multiple types of mobile traffic prediction tasks at the same time, and has excellent migration and task adaptation capabilities; by introducing a spatiotemporal mask strategy to achieve task differentiation and adaptation, as well as the ability to model the relationship between urban environment and traffic changes, the generalization performance and robustness of the model in multiple cities and multiple tasks are significantly improved.

[0049] Optionally, the data element segmentation unit based on the general prediction model segments the target mobile network traffic corresponding to the target prediction task to obtain a mobile data element sequence, including: Based on the digital element segmentation unit, a time-space word segmentation method is adopted to segment the target mobile network traffic according to the minimum time-space unit to obtain multiple mobile digital elements with uniform time dimension, spatial section length and spatial section width. The multiple mobile digital elements are sequenced in chronological order to form the mobile digital element sequence.

[0050] In practical applications, see Figure 2 The data segmentation unit uses time-space segmentation technology to segment the original differentiated (time*space) target mobile traffic data into unified mobile data according to the smallest time-space unit. This ensures the accuracy of segmentation.

[0051] Specifically, if the time dimension length of the target mobile traffic data is , the length of the space section is , width is , then the moving element sequence can be obtained by element segmentation using the following formula: in, represents the smallest space-time unit, , , The digital segmentation unit will convert the original The target mobile network traffic of the dimension is divided into The smallest space-time element with the same dimensions, that is, A sequence of moving numbers.

[0052] Optionally, the spatiotemporal masking mechanism based on the universal prediction model and the task type corresponding to the target prediction task performs masking on the mobile digital sequence to obtain a mask sequence, including: When the task type is a short-term prediction task, masking the mobile bit sequence based on the short-term mask corresponding to the short-term prediction task in the spatiotemporal mask mechanism to obtain the mask sequence; When the task type is a long-term prediction task, masking the mobile digital sequence based on the long-term mask corresponding to the long-term prediction task in the spatiotemporal mask mechanism to obtain the mask sequence; When the task type is a data generation task, the mobile bit sequence is masked based on the generation mask corresponding to the data generation task in the spatiotemporal mask mechanism to obtain the mask sequence, and each bit of the generation mask is 0.

[0053] In practical applications, see Figure 2 ,Mobile traffic prediction tasks mainly include short-term prediction tasks, long-term prediction tasks and distribution generation tasks.,That is, the task types corresponding to the target prediction task include short-term prediction, long-term prediction and distribution generation.,Different types of prediction tasks have significant differences in input requirements,,target length and training strategies.,If the time dimension length of the mobile data sequence is, , the moving number sequence is , then the mask mechanism of different prediction tasks can be defined as follows: For short-term prediction tasks, the mask should enable the model to access data only in the past time period, so the short-term mask Can be set to: in, l is the window length of past observations, and .

[0054] For long-term prediction tasks, the model input window length is shorter, but the prediction target sequence is longer, so the long-term mask Can be set to: For distribution generation tasks, the model does not rely on historical data, but is completely synthetically generated based on the environmental context, so the generated mask Use a full mask, as follows: In addition to the three task-specific masks mentioned above, see Figure 2 , a random mask is also set in the spatiotemporal mask mechanism to help the model capture and understand the various complex correlations in mobile traffic data. for: in, Indicates from space ,time and spatial dimensions The masking operation allows the model to capture complex relationships across time and space. This masking mechanism not only allows the model to capture the characteristics of long-term data series during training, but also enhances its adaptability to different tasks, such as short-term and long-term forecasting.

[0055] Optionally, the prediction task processing layer includes a noise addition layer, a denoising network, and an adaptive condition generation layer; the prediction task processing layer based on the universal prediction model performs task processing on the target environment data corresponding to the mask sequence and the mobile digital element sequence to obtain target prediction data corresponding to the target prediction task, including: Based on the noise adding layer, performing noise processing on the masked part of the mask sequence to obtain a noisy masked part; Based on the adaptive condition generation layer, condition generation and feature extraction are performed on the target environment data to obtain environmental condition features; Adaptively adjusting the residual, scale, and offset parameters in the denoising network based on the environmental condition characteristics; Based on the adjusted denoising network and the unmasked part of the mask sequence, the masked part after the noise is added is reconstructed to obtain target prediction data corresponding to the target prediction task.

[0056] In practical applications, see Figure 2 After completing the multi-task masking process, the mask sequence is divided into two parts: the masked part that needs to be reconstructed (masked target data) and the unmasked observation part (unmasked part). The masked part is reconstructed by the model, while the unmasked part is input into the denoising network as a condition. and the unmasked part The process is shown in the following formula: in, Corresponding to the four masking strategies, represents the element-wise product. Subsequently, the unmasked part is input into the denoising network as a condition, and the masked part Then, noise is added in the forward process, and the noise adding process is as follows: in, is the preset weight, It's noise. is the standard normal distribution, is the masked part after adding noise. Then The data is input into the multi-layer Transformer-based denoising network for further feature extraction. It uses the self-attention mechanism to capture the complex dependencies in spatiotemporal data. To fully capture the dependencies between environmental data and mobile traffic, the adaptive conditional generation layer uses an adaptive conditional adjustment method to conditionally generate and extract features from the target environmental data, obtaining environmental condition features. This method uses the target environmental data to adjust the residual, scale, and offset parameters in the Transformer in the denoising network, thereby improving the effectiveness and computational efficiency of the denoising network. The specific formula is as follows: in, and denote the linear layer and the attention layer respectively, are the residual, scale, and offset parameters respectively.

[0057] Furthermore, the denoising network is reconstructed based on the unmasked part of the mask sequence and the masked part after noise addition to obtain the target prediction data corresponding to the target prediction task.

[0058] Optionally, the prediction task processing layer further includes a perception alignment mechanism; and before the adaptive condition generation layer generates conditions and extracts features from the target environment data to obtain the environment condition features, the method further includes: Based on the perception alignment mechanism, context-aware alignment is performed on the target environment data and the target mobile network traffic to obtain aligned target environment data; The step of performing condition generation and feature extraction on the target environment data based on the adaptive condition generation layer to obtain environment condition features includes: Based on the adaptive condition generation layer, condition generation and feature extraction are performed on the aligned target environment data to obtain environment condition features.

[0059] In practical applications, see Figure 2 After obtaining the target environment data and target mobile network traffic, we need to perform context-aware alignment on the target environment data and the target mobile network traffic through the perception alignment mechanism to obtain the aligned target environment data. Based on this aligned target environment data, we can generate environmental condition features. This ensures data processing consistency.

[0060] Optionally, the target environment data includes a traffic token and a dynamic point of interest sequence; The prediction task processing layer based on the universal prediction model performs task processing on the target environment data corresponding to the mask sequence and the mobile bit sequence to obtain the target prediction data corresponding to the target prediction task, further comprising: Acquiring initial environment data corresponding to the mobile digital sequence, the initial environment data including mobile user data and point of interest distribution data; Extracting static POI features from the POI distribution data and fusing the static POI features with dynamic time tags to obtain dynamic POI features, wherein the dynamic time tags are obtained by projecting timestamps into time embeddings using a multi-layer perceptron network; The mobile user data is segmented to obtain the traffic token, and the dynamic interest point features are segmented to obtain the dynamic interest point sequence.

[0061] In practical applications, mobile network traffic is also affected by environmental data. Therefore, we can first go back to the initial environmental data corresponding to the mobile data sequence, namely human dynamics (mobile user data) and point of interest (POI) information (POI distribution data).

[0062] See also Figure 2 ,“mobile users” (mobile user data) refers to the number of users accessing the network, which can fully characterize the human dynamics in mobile networks. Similar to mobile network traffic, mobile user data is essentially a spatiotemporal series and can be expressed as Use the data processing method (digital segmentation) of mobile network traffic to process mobile user data and generate traffic tokens, thereby converting mobile user data into Input the prediction task processing layer for processing.

[0063] The distribution data of points of interest reflects the static distribution of urban layout and can be expressed as Although the distribution of POIs is static, the impact of different types of POIs on human behavior can change over time, leading to changes in mobile traffic. The following dynamic POI conversion scheme can be used for POI distribution data.

[0064] First, extract the static features of the interest point distribution data , expressed using the following formula: in, is the Sigmoid activation function, and are the weight and bias parameters of the Multilayer Perceptron (MLP) network.

[0065] Then, use an MLP network The timestamp is projected into a time embedding, and then the static POI features are fused with the time index (dynamic time label) to obtain the dynamic point of interest feature fusion process as follows: in, represents vector concatenation, and is a learnable parameter, is the dynamic point of interest feature. In this way, a spatiotemporal dynamic POI representation (dynamic point of interest feature) can be obtained. The final dynamic interest point features will be calculated using the same tokenization method (digit segmentation) as the traffic data to obtain a dynamic interest point sequence. .

[0066] Optionally, before the data element segmentation unit based on the universal prediction model segments the target mobile network traffic corresponding to the target prediction task to obtain the mobile data element sequence, it further includes: Constructing a first training set based on first sample mobile network traffic of different task types and different data sources, wherein the first training set includes multiple sample task sets, and the sample task sets include first sample tasks and first sample mobile network traffic corresponding to the first sample tasks; The first training set is used to train the denoising network in the prediction task processing layer, and the loss function for training the denoising network is a reconstruction error loss function.

[0067] See also Figure 2 and Figure 3 , Figure 3This is a schematic diagram of the process for training a universal prediction model provided by the present invention: First, a first training set is obtained. Then, the universal prediction model is pre-trained (model pre-training) based on multiple sample task sets in the first training set. This primarily involves self-supervised masked training of the denoising network. Therefore, the first sample mobile network traffic is segmented to obtain the corresponding mobile data, which is then used to obtain a sample mobile data sequence. This sample mobile data sequence is then spatiotemporally masked based on the type of the first sample task, and the denoising network is then pre-trained.

[0068] It should be noted that the goal of the denoising network is to minimize the reconstruction error of the masked part, as shown below: in, The reconstruction error loss function emphasizes the reconstruction accuracy of the masked part. Through this process, the model can learn the optimal reconstruction results and improve the prediction ability of mobile traffic data.

[0069] Optionally, before the data element segmentation unit based on the universal prediction model segments the target mobile network traffic corresponding to the target prediction task to obtain the mobile data element sequence, it further includes: Based on different task types and second sample mobile network traffic from different data sources and their corresponding sample target environment data, a second training set is constructed, wherein the second training set includes multiple positive sample task sets and multiple negative sample task sets, wherein the positive sample task set includes the second sample task and the positive sample corresponding to the second sample task, and the positive sample includes the second sample mobile network traffic and the sample target environment data in the same spatiotemporal block; the negative sample task set includes the third sample task and the negative sample corresponding to the third sample task, and the negative sample includes the second sample mobile network traffic and the sample target environment data in the same spatiotemporal block; Based on each of the negative samples and each of the positive samples, a perceptual alignment mechanism in the prediction task processing layer is trained, wherein a goal of training the perceptual alignment mechanism is to minimize contrastive learning loss; Based on each of the positive sample task sets and each of the negative sample task sets, the universal prediction model is fine-tuned, and the loss function for fine-tuning the universal prediction model is a comprehensive loss function, which includes a reconstruction error loss function and an alignment loss function.

[0070] See also Figure 2 and Figure 3 ,It is also necessary to fine-tune the general prediction model based on environmental data,,i.e., user distribution (mobile users) and POI information (static POI,distribution).

[0071] In practical applications, the second training set is obtained first.

[0072] See also Figure 2 In order to establish a connection between mobile traffic and environmental characteristics, a contrastive learning algorithm can be used. The traffic data (second sample mobile network traffic) and environmental feature data (sample target environment data) in the same spatiotemporal block are taken as positive samples and recorded as ; The two data from different spatiotemporal blocks (the second sample mobile network traffic and the sample target environment data) are taken as negative samples and recorded as .

[0073] Furthermore, the goal is set to maximize the mutual information between traffic characteristics and environmental characteristics , the density ratio can be used to preserve mutual information. The maximization process is equivalent to minimizing the InfoNCE (contrastive learning) loss, as follows: in, Represents the samples of the entire batch.

[0074] Then fine-tune and achieve alignment between mobile traffic and environmental features by optimizing the mean square error (MSE) of positive and negative samples. The loss function is as follows: in, L It is a comprehensive loss function, which contains two parts: Part 1 Is the traffic characteristic e And the model prediction results The error between represents the accuracy of the model in reconstructing the masked part, that is, the reconstruction error loss function; The second part It is a constraint on negative samples, which aims to minimize the error between the flow feature e and the negative samples e' from different spatiotemporal blocks, thereby helping the model learn better feature distinction, that is, the alignment loss function.

[0075] Is a regularization parameter that controls the weight of the error of negative samples in the total loss. Ultimately, the goal of the model is to optimize the reconstruction accuracy of traffic features and the ability to distinguish different samples by minimizing the loss function. m is a masking mask that indicates the focus of the model during training, ensuring that the model trains only on unmasked traffic features. During fine-tuning, we partially freeze the main parameters of the pre-trained model, including the attention layer, linear layer, and MLP network, to maintain the model's ability to learn the spatiotemporal characteristics of traffic. The main update is to update the parameters of the adaptability condition and output decoder layer. By updating only these components, we can reduce the time and computational overhead of fine-tuning.

[0076] See also Figure 2 ,The fine-tuning module mainly includes the following steps: (1) Linear layer and activation layer: Part of the network structure in the fine-tuning process includes linear layer and activation layer. The linear layer maps the input data to a new feature space, while the activation layer is used to increase the nonlinear expression ability of the model, helping the network to extract more complex patterns from the data.

[0077] (2) Splicing operation: During the fine-tuning process, the input traffic data features and environmental features (POI data and mobile user data) are fused through the splicing operation. The splicing operation combines features from different data sources into a unified input, ensuring that the model can learn features from different sources simultaneously.

[0078] (3) Partially freeze the main parameters of the pre-trained model: To avoid overfitting and reduce computational overhead, the main parameters of the pre-trained model are partially frozen during fine-tuning. Only the parameters of the adaptive conditioning layer and the output decoder layer are updated, thereby improving the performance of the model on specific tasks while maintaining the spatiotemporal features learned by the pre-trained model.

[0079] The following describes the application process of the UoMo model in a typical mobile network traffic prediction and optimization scenario, combining the actual deployment process of the UoMo model to further illustrate the mobile network traffic processing method based on the universal prediction model provided by the present invention.

[0080] The implementation of the present invention is based on city-level wireless network traffic data, with the goal of completing short-term prediction, long-term trend prediction and simulation generation tasks through a unified model, and applying the results to scenarios such as wireless resource scheduling, base station deployment and energy-saving control.

[0081] In practical applications, the UoMo model can be deployed on the operator's edge computing platform or cloud-based inference system, receiving real-time or periodic uploaded traffic data and environmental context information. The model training process can be completed offline, and the inference phase only calls the prediction model with fixed parameters. The implementation steps are as follows: First, the data preprocessing stage involves collecting traffic data from the mobile network management system. The data types may include MR data reported by the user side and PM data exported by the network element side, corresponding to millisecond and minute sampling granularities respectively. The system slices the data of different granularities uniformly and constructs fixed-length input windows. Each window sequence contains The observation points of time steps are recorded as .

[0082] Secondly, the system extracts the regional environmental feature information corresponding to each sequence. The environmental feature sources include mobile user data and POI data. This information is encoded to form a context embedding vector , which is input into the prompt module and diffusion model as a conditional control variable. Mobile user data can provide information about user activity within a region, reflecting the impact of human dynamics on traffic changes. Point of Interest (POI) data, on the other hand, reflects the functional layout of different urban areas (such as restaurants, commercial areas, and residential areas), helping the model understand traffic demand in different types of areas. After processing, these data are combined with traffic data as model input to capture the impact of environmental changes on network traffic.

[0083] The UoMo model will then perform short-term predictions, long-term predictions, and generation tasks based on these input data. Short-term prediction tasks use historical traffic data and environmental context to predict traffic for the next few steps, and are often used for wireless resource scheduling and load balancing; long-term prediction tasks predict long-term traffic trends based on longer historical data, and are usually used for base station deployment and network planning; generation tasks simulate traffic distribution in new or unobserved areas to assist in network expansion and new site construction. Finally, the system calls the optimization interface based on the model output results and the tasks. For example, in the short-term prediction scenario, the model output is used to update the load balancing strategy of the resource scheduling module; in the long-term prediction scenario, the model results are input into the deployment simulation system to evaluate regional capacity bottlenecks; and in the generation task, the output data can be used to build a digital twin region to simulate and compare new regional site construction plans. See Figure 4 and Figure 5 , Figure 4 This is one of the effect diagrams of mobile network traffic processing provided by the present invention. Figure 5 This is the second effect diagram of the mobile network traffic processing provided by the present invention. Figure 4 and Figure 5 The results of the UoMo model on short-term prediction, long-term prediction, and generation tasks on the traffic datasets of City A and City B are shown, where the horizontal axis is the time series points and the vertical axis is the mobile network traffic.

[0084] It should be noted that in the above process, the UoMo model only needs to be trained once, and there is no need to adjust the structure or switch the model for each task during the inference process, which greatly reduces the system deployment and management costs and has practical engineering feasibility.

[0085] The mobile network traffic processing method based on a universal prediction model proposed in this paper aims to address prominent challenges of current mobile traffic prediction methods, including task fragmentation, insufficient model generalization, and high engineering deployment complexity. In actual network operations, traffic prediction tasks exhibit diverse tasks (e.g., short-term prediction, long-term prediction, and data generation), data heterogeneity (e.g., varying sampling frequencies and spatial granularity), and semantic differences between cities.

[0086] Traditional methods often customize models for each type of task or data, which not only increases the cost of training and deployment, but also makes it difficult to support dynamic, large-scale, and multi-scenario wireless network optimization needs. To this end, the present invention provides a mobile network traffic processing method based on a universal prediction model, and constructs a unified modeling framework UoMo, which can simultaneously handle multiple types of mobile traffic prediction tasks and has excellent migration and task adaptation capabilities. The model is based on the Transformer structure, combined with the diffusion modeling mechanism, and introduces a spatiotemporal masking strategy to achieve task differentiation and adaptation, and enhances the modeling ability of the relationship between urban environment and traffic changes through a context-aware contrast learning mechanism. The model backbone adopts the Transformer-based Diffusion Architecture, abandoning the traditional U-Net structure, and enhancing its ability to model the temporal and spatial relationships under non-Euclidean structures. At the same time, the present invention designs a unified training process, combining the contrast loss term and the variational lower bound optimization mechanism to significantly improve the generalization performance and robustness of the model in multiple cities and multiple tasks.

[0087] The mobile network traffic processing device based on the universal prediction model provided by the present invention is described below. The mobile network traffic processing device based on the universal prediction model described below and the mobile network traffic processing method based on the universal prediction model described above can refer to each other.

[0088] Figure 6 Schematic diagram of the structure of the mobile network traffic processing device based on the universal prediction model provided by the present invention. Figure 6 As shown, the device includes: The segmentation module 601 is configured as a data segmentation unit based on a general prediction model, which segments the target mobile network traffic corresponding to the target prediction task to obtain a mobile data sequence; a masking module 602 configured to perform masking on the mobile bit sequence based on the spatiotemporal masking mechanism of the general prediction model and the task type corresponding to the target prediction task to obtain a masked sequence; The prediction module 603 is configured to perform task processing on the target environment data corresponding to the mask sequence and the mobile element sequence based on the prediction task processing layer of the general prediction model to obtain target prediction data corresponding to the target prediction task.

[0089] The mobile network traffic processing device based on a universal prediction model provided by the present invention divides the target mobile network traffic corresponding to the target prediction task through a digital segmentation unit based on the universal prediction model to obtain a mobile digital sequence; based on the spatiotemporal mask mechanism of the universal prediction model and the task type corresponding to the target prediction task, the mobile digital sequence is masked to obtain a mask sequence; based on the prediction task processing layer of the universal prediction model, the mask sequence and the target environment data corresponding to the mobile digital sequence are task processed to obtain the target prediction data corresponding to the target prediction task, which solves the outstanding problems of current mobile traffic prediction methods in terms of task splitting, insufficient model generalization and high engineering deployment complexity; the present invention can simultaneously process multiple types of mobile traffic prediction tasks and has excellent migration and task adaptation capabilities; by introducing a spatiotemporal mask strategy to achieve task differentiation and adaptation, as well as the ability to model the relationship between urban environment and traffic changes, the generalization performance and robustness of the model in multiple cities and multiple tasks are significantly improved.

[0090] Optionally, the prediction task processing layer includes a noise addition layer, a denoising network and an adaptive condition generation layer; The prediction module 603 is further configured to: Based on the noise adding layer, performing noise processing on the masked part of the mask sequence to obtain a noisy masked part; Based on the adaptive condition generation layer, condition generation and feature extraction are performed on the target environment data to obtain environmental condition features; Adaptively adjusting the residual, scale, and offset parameters in the denoising network based on the environmental condition characteristics; Based on the adjusted denoising network and the unmasked part of the mask sequence, the masked part after the noise is added is reconstructed to obtain target prediction data corresponding to the target prediction task.

[0091] Optionally, the prediction task processing layer further includes a perception alignment mechanism; The prediction module 603 is further configured to: Based on the perception alignment mechanism, context-aware alignment is performed on the target environment data and the target mobile network traffic to obtain aligned target environment data; Based on the adaptive condition generation layer, condition generation and feature extraction are performed on the aligned target environment data to obtain environment condition features.

[0092] Optionally, the target environment data includes a traffic token and a dynamic point of interest sequence; The apparatus further includes a processing module configured to: Acquiring initial environment data corresponding to the mobile digital sequence, the initial environment data including mobile user data and point of interest distribution data; Extracting static POI features from the POI distribution data and fusing the static POI features with dynamic time tags to obtain dynamic POI features, wherein the dynamic time tags are obtained by projecting timestamps into time embeddings using a multi-layer perceptron network; The mobile user data is segmented to obtain the traffic token, and the dynamic interest point features are segmented to obtain the dynamic interest point sequence.

[0093] Optionally, the apparatus further includes a first training module configured to: Constructing a first training set based on first sample mobile network traffic of different task types and different data sources, wherein the first training set includes multiple sample task sets, and the sample task sets include first sample tasks and first sample mobile network traffic corresponding to the first sample tasks; The first training set is used to train the denoising network in the prediction task processing layer, and the loss function for training the denoising network is a reconstruction error loss function.

[0094] Optionally, the apparatus further includes a second training module configured to: Based on different task types and second sample mobile network traffic from different data sources and their corresponding sample target environment data, a second training set is constructed, wherein the second training set includes multiple positive sample task sets and multiple negative sample task sets, wherein the positive sample task set includes the second sample task and the positive sample corresponding to the second sample task, and the positive sample includes the second sample mobile network traffic and the sample target environment data in the same spatiotemporal block; the negative sample task set includes the third sample task and the negative sample corresponding to the third sample task, and the negative sample includes the second sample mobile network traffic and the sample target environment data in the same spatiotemporal block; Based on each of the negative samples and each of the positive samples, a perceptual alignment mechanism in the prediction task processing layer is trained, wherein a goal of training the perceptual alignment mechanism is to minimize contrastive learning loss; Based on each of the positive sample task sets and each of the negative sample task sets, the universal prediction model is fine-tuned, and the loss function for fine-tuning the universal prediction model is a comprehensive loss function, which includes a reconstruction error loss function and an alignment loss function.

[0095] Optionally, the segmentation module 601 is further configured to: Based on the digital element segmentation unit, a time-space word segmentation method is adopted to segment the target mobile network traffic according to the minimum time-space unit to obtain multiple mobile digital elements with uniform time dimension, spatial section length and spatial section width. The multiple mobile digital elements are sequenced in chronological order to form the mobile digital element sequence.

[0096] Optionally, the mask module 602 is further configured to: When the task type is a short-term prediction task, masking the mobile bit sequence based on the short-term mask corresponding to the short-term prediction task in the spatiotemporal mask mechanism to obtain the mask sequence; When the task type is a long-term prediction task, masking the mobile digital sequence based on the long-term mask corresponding to the long-term prediction task in the spatiotemporal mask mechanism to obtain the mask sequence; When the task type is a data generation task, the mobile bit sequence is masked based on the generation mask corresponding to the data generation task in the spatiotemporal mask mechanism to obtain the mask sequence, and each bit of the generation mask is 0.

[0097] Figure 7 An example of a physical structure diagram of an electronic device is shown below. Figure 7 As shown, the electronic device may include: a processor 710, a communications interface 720, a memory 730, and a communication bus 740, wherein the processor 710, the communications interface 720, and the memory 730 communicate with each other via the communication bus 740. The processor 710 may call logic instructions in the memory 730 to execute a mobile network traffic processing method based on a universal prediction model. The method includes: a data segmentation unit based on the universal prediction model segments the target mobile network traffic corresponding to the target prediction task to obtain a mobile data sequence; based on the spatiotemporal mask mechanism of the universal prediction model and the task type corresponding to the target prediction task, masking the mobile data sequence to obtain a mask sequence; and a prediction task processing layer based on the universal prediction model performs task processing on the mask sequence and target environment data corresponding to the mobile data sequence to obtain target prediction data corresponding to the target prediction task.

[0098] Furthermore, the logic instructions in the aforementioned memory 730 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product, stored in a storage medium, includes instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0099] On the other hand, the present invention also provides a computer program product, which includes a computer program, which can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the mobile network traffic processing method based on the general prediction model provided by the above methods, the method including: based on the digital element segmentation unit of the general prediction model, dividing the target mobile network traffic corresponding to the target prediction task to obtain a mobile digital element sequence; based on the spatiotemporal mask mechanism of the general prediction model and the task type corresponding to the target prediction task, masking the mobile digital element sequence to obtain a mask sequence; based on the prediction task processing layer of the general prediction model, task processing is performed on the target environment data corresponding to the mask sequence and the mobile digital element sequence to obtain the target prediction data corresponding to the target prediction task.

[0100] 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 mobile network traffic processing method based on the general prediction model provided by the above-mentioned methods, the method comprising: based on a digital element segmentation unit of the general prediction model, dividing the target mobile network traffic corresponding to the target prediction task to obtain a mobile digital element sequence; based on the spatiotemporal mask mechanism of the general prediction model and the task type corresponding to the target prediction task, masking the mobile digital element sequence to obtain a mask sequence; based on the prediction task processing layer of the general prediction model, task processing is performed on the target environment data corresponding to the mask sequence and the mobile digital element sequence to obtain target prediction data corresponding to the target prediction task.

[0101] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0102] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0103] 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 various embodiments of the present invention.

Claims

1. A method for processing mobile network traffic based on a universal prediction model, characterized in that: include: The data segmentation unit based on the general prediction model segments the target mobile network traffic corresponding to the target prediction task to obtain a mobile data sequence; Based on the spatiotemporal masking mechanism of the universal prediction model and the task type corresponding to the target prediction task, masking the mobile digital sequence to obtain a mask sequence; Based on the prediction task processing layer of the universal prediction model, task processing is performed on the target environment data corresponding to the mask sequence and the mobile digital element sequence to obtain target prediction data corresponding to the target prediction task.

2. The method for processing mobile network traffic based on a universal prediction model according to claim 1, characterized in that: The prediction task processing layer includes a noise addition layer, a denoising network and an adaptive condition generation layer; The prediction task processing layer based on the universal prediction model performs task processing on the target environment data corresponding to the mask sequence and the mobile bit sequence to obtain target prediction data corresponding to the target prediction task, including: Based on the noise adding layer, performing noise processing on the masked part of the mask sequence to obtain a noisy masked part; Based on the adaptive condition generation layer, condition generation and feature extraction are performed on the target environment data to obtain environmental condition features; Adaptively adjusting the residual, scale, and offset parameters in the denoising network based on the environmental condition characteristics; Based on the adjusted denoising network and the unmasked part of the mask sequence, the masked part after the noise is added is reconstructed to obtain target prediction data corresponding to the target prediction task.

3. The method for processing mobile network traffic based on a universal prediction model according to claim 2, characterized in that: The prediction task processing layer also includes a perception alignment mechanism; Before performing condition generation and feature extraction on the target environment data based on the adaptive condition generation layer to obtain the environment condition features, the method further includes: Based on the perception alignment mechanism, context-aware alignment is performed on the target environment data and the target mobile network traffic to obtain aligned target environment data; The step of performing condition generation and feature extraction on the target environment data based on the adaptive condition generation layer to obtain environment condition features includes: Based on the adaptive condition generation layer, condition generation and feature extraction are performed on the aligned target environment data to obtain environment condition features.

4. The method for processing mobile network traffic based on a universal prediction model according to claim 1, wherein: The target environment data includes a traffic token and a dynamic point of interest sequence; The prediction task processing layer based on the universal prediction model performs task processing on the target environment data corresponding to the mask sequence and the mobile bit sequence to obtain the target prediction data corresponding to the target prediction task, further comprising: Acquiring initial environment data corresponding to the mobile digital sequence, the initial environment data including mobile user data and point of interest distribution data; Extracting static POI features from the POI distribution data and fusing the static POI features with dynamic time tags to obtain dynamic POI features, wherein the dynamic time tags are obtained by projecting timestamps into time embeddings using a multi-layer perceptron network; The mobile user data is segmented to obtain the traffic token, and the dynamic interest point features are segmented to obtain the dynamic interest point sequence.

5. The method for processing mobile network traffic based on a universal prediction model according to claim 1, wherein: The data segmentation unit based on the universal prediction model, before segmenting the target mobile network traffic corresponding to the target prediction task to obtain the mobile data sequence, further includes: Constructing a first training set based on first sample mobile network traffic of different task types and different data sources, wherein the first training set includes multiple sample task sets, and the sample task sets include first sample tasks and first sample mobile network traffic corresponding to the first sample tasks; The first training set is used to train the denoising network in the prediction task processing layer, and the loss function for training the denoising network is a reconstruction error loss function.

6. The method for processing mobile network traffic based on a universal prediction model according to claim 1, wherein: The data segmentation unit based on the universal prediction model, before segmenting the target mobile network traffic corresponding to the target prediction task to obtain the mobile data sequence, further includes: Based on different task types and second sample mobile network traffic from different data sources and their corresponding sample target environment data, a second training set is constructed, wherein the second training set includes multiple positive sample task sets and multiple negative sample task sets, wherein the positive sample task set includes the second sample task and the positive sample corresponding to the second sample task, and the positive sample includes the second sample mobile network traffic and the sample target environment data in the same spatiotemporal block; the negative sample task set includes the third sample task and the negative sample corresponding to the third sample task, and the negative sample includes the second sample mobile network traffic and the sample target environment data in the same spatiotemporal block; Based on each of the negative samples and each of the positive samples, a perceptual alignment mechanism in the prediction task processing layer is trained, wherein a goal of training the perceptual alignment mechanism is to minimize contrastive learning loss; Based on each of the positive sample task sets and each of the negative sample task sets, the universal prediction model is fine-tuned, and the loss function for fine-tuning the universal prediction model is a comprehensive loss function, which includes a reconstruction error loss function and an alignment loss function.

7. The method for processing mobile network traffic based on a universal prediction model according to claim 1, wherein: The data element segmentation unit based on the general prediction model segments the target mobile network traffic corresponding to the target prediction task to obtain a mobile data element sequence, including: Based on the digital element segmentation unit, a time-space word segmentation method is adopted to segment the target mobile network traffic according to the minimum time-space unit to obtain multiple mobile digital elements with uniform time dimension, spatial section length and spatial section width. The multiple mobile digital elements are sequenced in chronological order to form the mobile digital element sequence.

8. The method for processing mobile network traffic based on a universal prediction model according to any one of claims 1 to 7, characterized in that: The spatiotemporal masking mechanism based on the universal prediction model and the task type corresponding to the target prediction task performs masking on the mobile digital sequence to obtain a mask sequence, including: When the task type is a short-term prediction task, masking the mobile bit sequence based on the short-term mask corresponding to the short-term prediction task in the spatiotemporal mask mechanism to obtain the mask sequence; When the task type is a long-term prediction task, masking the mobile digital sequence based on the long-term mask corresponding to the long-term prediction task in the spatiotemporal mask mechanism to obtain the mask sequence; When the task type is a data generation task, the mobile bit sequence is masked based on the generation mask corresponding to the data generation task in the spatiotemporal mask mechanism to obtain the mask sequence, and each bit of the generation mask is 0.

9. A mobile network traffic processing device based on a universal prediction model, characterized in that: include: A segmentation module is configured as a data segmentation unit based on a general prediction model, which segments the target mobile network traffic corresponding to the target prediction task to obtain a mobile data sequence; a masking module configured to perform masking on the mobile digital sequence based on the spatiotemporal masking mechanism of the universal prediction model and the task type corresponding to the target prediction task to obtain a masked sequence; The prediction module is configured to perform task processing on the target environment data corresponding to the mask sequence and the mobile digital element sequence based on the prediction task processing layer of the universal prediction model to obtain target prediction data corresponding to the target prediction task.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the method for processing mobile network traffic based on a universal prediction model according to any one of claims 1 to 8 is implemented.

Images