Method and system for service demand prediction and mapping migration based on multi-granularity spatiotemporal graph

Abstract

This invention provides a method and system for service demand prediction and mapping migration based on multi-granularity spatiotemporal graphs, relating to the field of wireless communication technology. The method includes: collecting service behavior data of terminals in the network at multiple time granularities; processing and aggregating the service behavior data of each terminal in both time and space dimensions to obtain fused time features and fused spatial features; fusing the fused time features and fused spatial features to obtain fused spatiotemporal features; predicting service demands; and mapping and migrating these features to obtain communication demands. The method and system provided by this invention, by collecting service behavior data of multiple terminals in the network at multiple time granularities, mines the complex temporal and spatial correlation features of service behaviors, achieves high-precision service behavior prediction, and maps the prediction results to network communication demands, guiding precise scheduling of network resources, improving network capacity and efficiency, and realizing full-process automation from data perception and intelligent prediction to demand mapping.

Description

Technical Field

[0001] This invention relates to the field of wireless communication technology, and in particular to a service demand prediction and mapping migration method and system based on multi-granularity spatiotemporal graphs. Background Technology

[0002] With the rapid development of the Industrial Internet and intelligent manufacturing, massive numbers of wireless terminal devices, such as sensors, actuators, and controllers, are deployed in industrial automation scenarios. These devices are connected through industrial wireless networks, carrying various types of business data, such as production monitoring, equipment control, and environmental monitoring. To ensure the reliability and real-time performance of critical business operations, the key lies in the precise, on-demand allocation of network resources, such as time slots and bandwidth—that is, precise resource scheduling.

[0003] Achieving precise resource scheduling requires accurate prediction of service behaviors and their demand for network resources. However, existing technologies suffer from several problems, such as a single prediction dimension, resulting in low accuracy in predicting service behaviors and failing to effectively support efficient and precise resource scheduling in industrial wireless networks.

[0004] Therefore, how to improve the accuracy and efficiency of industrial wireless network resource scheduling has become a technical problem that the industry urgently needs to solve. Summary of the Invention

[0005] This invention provides a service demand prediction and mapping migration method and system based on multi-granularity spatiotemporal graphs, which solves the shortcomings of existing technologies in predicting service behavior with low accuracy and failing to effectively support efficient and accurate resource scheduling of industrial wireless networks, thereby improving the accuracy and efficiency of resource scheduling in industrial wireless networks.

[0006] This invention provides a method for business demand prediction and mapping migration based on multi-granularity spatiotemporal graphs, including: Collect business behavior data of multiple terminals in the target network at different time granularities; The business behavior data of each terminal at each time granularity is processed to obtain single time granularity feature vectors at different time granularities. Feature aggregation is performed on all the single-temporal-granularity feature vectors of each terminal to obtain fused temporal features; Spatial features are aggregated for the business behavior data of each terminal to obtain fused spatial features; The fused temporal features and the fused spatial features are fused to obtain fused spatiotemporal features, and communication requirements are obtained based on the fused spatiotemporal features.

[0007] In some embodiments, processing the business behavior data at each time granularity of each terminal to obtain single-time-granularity feature vectors at different time granularities includes: The business behavior data of each terminal at each time granularity are input into the preprocessing model to obtain the single time granularity feature vector output by the preprocessing model. The preprocessing model includes convolutional layers and pooling layers; the convolutional layers are used to process the business behavior data to obtain feature maps; and the pooling layers are used to compress the feature maps to obtain single-temporal-granularity feature vectors.

[0008] In some embodiments, the step of aggregating all the single-temporal-granularity feature vectors of each terminal to obtain fused temporal features includes: Perform a linear transformation and concatenate all the single-time-granularity feature vectors of different time granularities for each terminal to obtain the concatenated vector; Based on the concatenated vector, determine the temporal attention weights; Based on the time attention weights, all the single-time granularity feature vectors of each terminal are weighted and aggregated to determine the fused time features.

[0009] In some embodiments, the spatial feature aggregation of the business behavior data of each terminal to obtain fused spatial features includes: A spatial graph is constructed using each terminal as a node and the business relationships between terminals as edges. Based on the spatial graph, spatial attention weights are calculated between the target terminal and its neighboring terminals; wherein, the target terminal refers to any one of the terminals. Based on the spatial attention weights, the business behavior data of the target terminal's neighboring terminals are weighted and aggregated to determine the fused spatial features of the target terminal.

[0010] In some embodiments, calculating the spatial attention weights between the target terminal and its neighboring terminals based on the spatial map includes: The fusion time features of each node in the spatial graph are stacked to obtain a stacking matrix; Based on the stacking matrix, the attention coefficient between the target terminal and its neighboring terminals in the spatial graph is calculated. The attention coefficients are normalized to obtain the spatial attention weights.

[0011] In some embodiments, fusing the fused temporal features and the fused spatial features to obtain fused spatiotemporal features, and obtaining communication requirements based on the fused spatiotemporal features, includes: The fused temporal features and the fused spatial features are spliced ​​together or fused through a fusion network to obtain the fused spatiotemporal features; Predict business requirements based on the fused spatiotemporal features; Based on a preset mapping function, the business requirements are mapped and migrated to obtain the communication requirements.

[0012] This invention provides a business demand prediction and mapping migration system based on multi-granularity spatiotemporal graphs, comprising: The data acquisition module is used to collect business behavior data of multiple terminals in the target network at different time granularities; The time feature extraction module is used to process the business behavior data of each terminal at each time granularity to obtain single time granularity feature vectors at different time granularities. The time-related modeling module is used to perform feature aggregation on all the single-time-granularity feature vectors of each terminal to obtain fused time features; The spatial feature aggregation module is used to aggregate spatial features of the business behavior data of each terminal to obtain fused spatial features; The business requirement mapping and migration module is used to fuse the fused temporal features and the fused spatial features to obtain fused spatiotemporal features, and to obtain communication requirements based on the fused spatiotemporal features.

[0013] The present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the aforementioned service demand prediction and mapping migration method based on multi-granularity spatiotemporal graphs.

[0014] The present invention provides a non-transitory computer-readable storage medium storing a computer program thereon, wherein the computer program, when executed by a processor, implements the aforementioned business demand prediction and mapping migration method based on multi-granularity spatiotemporal graphs.

[0015] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the aforementioned business demand prediction and mapping migration method based on multi-granularity spatiotemporal graphs.

[0016] The present invention provides a service demand prediction and mapping migration method and system based on multi-granularity spatiotemporal graphs. By continuously collecting service behavior data of multiple terminals in the network at multiple different time granularities, and simultaneously modeling the multi-time granularity correlation of a single terminal and the spatial correlation of multiple terminals, it can more comprehensively and profoundly capture the complex dynamic patterns of service behavior, significantly improving the accuracy and reliability of prediction. By deeply mining the complex correlation features of service behavior in the time and space dimensions, it achieves high-precision service behavior prediction and intelligently maps the prediction results to specific network communication needs. This enables network resource scheduling to shift from "extensive reservation" to "refined on-demand allocation," thereby guiding precise scheduling of network resources, avoiding resource waste and shortage, and significantly improving network capacity and efficiency. It realizes full-process automation from data perception and intelligent prediction to demand mapping, reduces the complexity and error rate of manual configuration, and promotes the intelligent transformation of industrial wireless networks. Attached Figure Description

[0017] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0018] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the business demand prediction and mapping migration method based on multi-granularity spatiotemporal graphs provided by the present invention.

[0020] Figure 2 This is a schematic diagram of the multi-terminal, multi-time granularity business data acquisition tensor of the business demand prediction and mapping migration method based on multi-granularity spatiotemporal graphs provided by the present invention.

[0021] Figure 3 This is a method architecture diagram of the business demand prediction and mapping migration method based on multi-granularity spatiotemporal graphs provided by the present invention.

[0022] Figure 4 This is a schematic diagram of the business demand prediction and mapping migration system based on multi-granularity spatiotemporal graphs provided by the present invention.

[0023] Figure 5 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0024] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0025] It should be noted that the terms "first," "second," etc., used in this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that comprises a series of steps, units, or modules is not necessarily limited to those explicitly listed, but may include other steps, units, or modules not explicitly listed or inherent to such processes, methods, products, or devices.

[0026] Figure 1 This is a flowchart illustrating the business demand prediction and mapping migration method based on multi-granularity spatiotemporal graphs provided by the present invention, as shown below. Figure 1 As shown, the method includes steps 110, 120, 130, 140 and 150.

[0027] Step 110: Collect business behavior data of multiple terminals in the target network at different time granularities.

[0028] Specifically, the execution entity of the service demand prediction and mapping migration method based on multi-granularity spatiotemporal graphs provided in this embodiment of the invention is a service demand prediction and mapping migration system based on multi-granularity spatiotemporal graphs. This system can be implemented in software, such as a service demand prediction and mapping migration program based on multi-granularity spatiotemporal graphs running on a computer; or it can be implemented in hardware, such as a computer or server that executes the service demand prediction and mapping migration method based on multi-granularity spatiotemporal graphs.

[0029] The prerequisite for achieving efficient and precise resource scheduling in industrial wireless networks is the accurate prediction of service behavior and its demand for network resources. However, existing technologies have the following shortcomings: Single prediction dimension: Traditional prediction methods often only focus on historical data at a single time scale, such as seconds or minutes, ignoring the periodic and trend-related characteristics of business behavior at different time granularities, such as hours or days, resulting in low prediction accuracy.

[0030] Ignoring spatial dependencies: In industrial scenarios where multiple terminals work collaboratively, the business behaviors of different terminals often have spatial correlations and influences, such as the triggering of one sensor causing the action of another actuator. Existing methods typically predict each terminal as an independent entity, failing to effectively model this spatial dependency.

[0031] The disconnect between business and network requirements: Predicted business behaviors, such as packet period and packet length, are application-layer metrics and cannot be directly used for network-layer resource scheduling. Current technologies lack an effective mechanism to intelligently and accurately map abstract business behaviors to specific network communication requirements, such as latency, transmission rate, and packet loss rate.

[0032] To address the shortcomings of the existing technologies, this invention provides a service demand prediction and mapping migration method based on multi-granularity spatiotemporal graphs. This method can be widely applied to industrial IoT and wireless network resource scheduling technologies, and also involves data prediction technologies such as artificial intelligence and graph neural networks. It can intelligently and accurately map abstract business behaviors into specific network communication needs.

[0033] The target network can be an industrial IoT network, a wireless sensor network, or another wireless communication network that requires deterministic latency.

[0034] Terminals refer to various devices with data transmission and reception functions deployed in a network, such as temperature sensors, pressure sensors, vibration monitors, robotic arm controllers, valve actuators, etc., deployed on production lines.

[0035] Business behavior data is primarily used to characterize the application-layer business characteristics of a terminal. In this embodiment of the invention, business behavior data includes, but is not limited to, the data packet sending cycle and data packet length. The data packet sending cycle refers to the time interval between two consecutive data packet transmissions by the terminal, reflecting the frequency of the service. The data packet length refers to the size of the data packet sent by the terminal, such as the number of bytes, reflecting the service load.

[0036] Time granularity refers to the time scale or time resolution for sampling, observing, or aggregating statistical data on continuous business behavior. It specifies the smallest time unit or time window size used for data acquisition or feature extraction. Different time granularities represent discrete representations of business behavior data at different levels of granularity in the time dimension. Time granularity includes, but is not limited to, millisecond, second, and minute levels.

[0037] In this embodiment of the invention, service behavior data from multiple terminals in the target network at different time granularities are continuously collected. Furthermore, the collected service behavior data is represented as a three-dimensional tensor.

[0038] For example, continuously collect service data from N sensor nodes in the network over a past period. The time granularity M=3, set to 1 second, 1 minute, and 1 hour respectively. The collected service behavior data includes the data packet transmission cycle and data packet length of each node at each time point, forming three datasets with different granularities. Collect historical service behavior data from N terminals in the network at M different time granularities.

[0039] Then, the collected data can be represented as a three-dimensional tensor D. Figure 2 This is a schematic diagram of the multi-terminal, multi-temporal granularity business data acquisition tensor of the business demand prediction and mapping migration method based on multi-granularity spatiotemporal graphs provided by the present invention, as shown in the figure. Figure 2 As shown, its formula is: Where N is the number of terminals; T is the total number of time steps; and F is the dimension of business behavior characteristics. For example, F=2 represents the data packet length and the sending period, respectively.

[0040] Step 120: Process the business behavior data of each terminal at each time granularity to obtain single time granularity feature vectors at different time granularities.

[0041] Specifically, a single-time-granularity feature vector refers to a numerical vector obtained after feature extraction of business behavior data of a specific terminal at a specific time granularity, such as milliseconds, seconds, or minutes, which is used to characterize the business behavior pattern of the terminal at that time granularity.

[0042] Since raw data typically contains a large amount of redundant information and has high dimensionality, directly using it for subsequent cross-granularity or cross-spatial correlation analysis is not only computationally intensive but also makes it difficult to capture deep-seated hidden patterns. Therefore, in this embodiment of the invention, for each time granularity of each terminal, a linear convolutional neural network is used to process the historical business behavior data at the corresponding granularity to extract the single-time granularity feature vector at that time granularity.

[0043] Step 130: Perform feature aggregation on all the single-time granularity feature vectors of each terminal to obtain fused time features.

[0044] Specifically, a time-granularity graph is constructed, with nodes representing different time granularities and edges representing the relationships between granularities. Each node represents a specific time granularity (e.g., seconds, minutes, hours). A graph attention network is used to model this graph, learning and aggregating the correlation information between features of different time granularities to form a time feature representation that integrates the correlations between multiple time granularities—that is, a fused time feature. This fused time feature not only includes the current business status but also adaptively incorporates historical trends or instantaneous fluctuation information that is most valuable for current prediction.

[0045] Step 140: Aggregate spatial features of the business behavior data of each terminal to obtain fused spatial features.

[0046] Specifically, terminals in industrial networks typically do not operate in isolation; they often have collaborative, triggering, or cascading relationships. In this embodiment of the invention, such cross-device dependencies are captured by aggregating the spatial features of the business behavior data of each terminal.

[0047] First, a spatial graph is constructed with different terminals as nodes and the business relationships between terminals as edges. Then, a spatial feature aggregation module, such as another graph attention network or graph convolutional network, is used to model this spatial graph, learn and aggregate the spatial dependency information between the business behaviors of different terminals, calculate the spatial attention weight between neighboring terminals and the target terminal, and based on this weight, the target terminal aggregates the feature information of all its neighboring terminals to generate fused spatial features, forming a spatial feature representation that incorporates spatial relationships, i.e., fused spatial features.

[0048] Step 150: Fuse the fused temporal features and the fused spatial features to obtain fused spatiotemporal features, and obtain communication requirements based on the fused spatiotemporal features.

[0049] Specifically, in this embodiment of the invention, the fused temporal features and fused spatial features are further fused to construct a complete fused spatiotemporal feature containing multi-dimensional spatiotemporal information. This is typically achieved through feature concatenation or through a fusion neural network layer.

[0050] Then, based on this comprehensive spatiotemporal feature, predictive models, such as Multi-Layer Perceptron (MLP), are used to infer the application layer business behavior of the terminal in the next moment, i.e., business requirements, such as data packet sending cycle and data packet length.

[0051] Finally, a mapping model between business behavior and network resource demand is established. This involves taking the predicted business layer parameters (cycle time and packet length) as input and intelligently mapping and transferring them into specific network communication demand indicators, including latency, transmission rate, and packet loss rate, through pre-defined rules, formulas, or neural networks. This transformation allows the prediction results to be directly input into the Time Division Multiple Access (TDMA) scheduler, enabling refined and on-demand allocation of network resources. This significantly improves network bandwidth utilization efficiency while ensuring service quality.

[0052] Figure 3This is a method architecture diagram of the business demand prediction and mapping migration method based on multi-granularity spatiotemporal graphs provided by the present invention, such as... Figure 3 As shown in the embodiments of the present invention, the service demand prediction and mapping migration method based on multi-granularity spatiotemporal graphs first uses linear convolution to extract features of data at different time granularities in the time dimension; then, it innovatively employs graph attention networks to analyze the correlation between features at different time granularities, thereby capturing the intrinsic connection between long-term trends and short-term fluctuations. In the spatial dimension, a spatial feature aggregation module, or a graph neural network, is used to model the mutual influence and dependency relationships of service behaviors among multiple terminal devices; simultaneously considering the "correlation between multiple time granularities of a single terminal" and the "spatial correlation between multiple terminals," a three-dimensional "spatiotemporal feature model" is constructed, thereby greatly improving the accuracy and reliability of service behavior prediction. Furthermore, the predicted abstract service behaviors can be intelligently migrated and converted into quantitative indicators that the network layer can directly understand and execute, such as latency, transmission rate, and packet loss rate. This "mapping migration" mechanism breaks down the barriers between the application layer and the network layer, enabling the prediction results to be directly used to guide the precise scheduling of network resources such as TDMA, truly realizing a closed loop from "intelligent perception" to "intelligent decision-making."

[0053] The service demand prediction and mapping migration method based on multi-granularity spatiotemporal graphs provided in this invention continuously collects service behavior data from multiple terminals in the network at multiple different time granularities. Simultaneously, it models the multi-time granularity correlation of a single terminal and the spatial correlation of multiple terminals, enabling a more comprehensive and profound capture of the complex dynamic patterns of service behavior, significantly improving the accuracy and reliability of prediction. By deeply mining the complex correlation features of service behavior in the time and space dimensions, it achieves high-precision service behavior prediction and intelligently maps the prediction results to specific network communication needs. This allows network resource scheduling to shift from "extensive reservation" to "refined on-demand allocation," guiding precise scheduling of network resources, avoiding resource waste and shortages, and significantly improving network capacity and efficiency. It achieves full-process automation from data perception and intelligent prediction to demand mapping, reducing the complexity and error rate of manual configuration and promoting the intelligent transformation of industrial wireless networks.

[0054] In some embodiments, processing the business behavior data at each time granularity of each terminal to obtain single-time-granularity feature vectors at different time granularities includes: The business behavior data of each terminal at each time granularity are input into the preprocessing model to obtain the single time granularity feature vector output by the preprocessing model. The preprocessing model includes convolutional layers and pooling layers; the convolutional layers are used to process the business behavior data to obtain feature maps; and the pooling layers are used to compress the feature maps to obtain single-temporal-granularity feature vectors.

[0055] Specifically, a feature map refers to the intermediate data representation obtained after the original business behavior data sequence has undergone convolution operations and nonlinear activation processing in a one-dimensional convolutional layer. It is essentially a two-dimensional matrix.

[0056] A global average pooling layer is an operational layer or computational module used for feature dimensionality reduction and compression in neural networks. In this embodiment of the invention, a two-dimensional feature map containing temporal information is compressed into a one-dimensional vector through a global average pooling layer. This preserves the global statistical features at that time granularity, significantly reduces the number of parameters required for subsequent computation, and endows the model with the ability to adapt to changes in the length of the input sequence.

[0057] In this embodiment of the invention, for any terminal, the system uses its historical business behavior data at a specific time granularity as an input sequence, processes it using a preprocessing model, and obtains a single-time-granularity feature vector. The preprocessing model includes convolutional layers and pooling layers; the convolutional layers are used to process the business behavior data to obtain a feature map; the pooling layers are used to compress the feature map to obtain a single-time-granularity feature vector.

[0058] In this embodiment of the invention, a one-dimensional convolutional layer is used to process the feature map. The feature map output by the convolutional layer still retains the time dimension. In order to obtain a fixed-length compact vector that can represent the overall business state at this time granularity, the feature map is further compressed through a global average pooling layer to obtain a single-time-granularity feature vector.

[0059] For example, for each terminal node's 1-second, 1-minute, and 1-hour granular data, a one-dimensional convolutional layer, Conv1D, is used for processing. Taking the 1-minute granularity as an example, assuming the data from the past 60 minutes is taken as input, the input data sequence would be: .

[0060] Set the Conv1D convolution kernel size k=5, and the number of output channels... .

[0061] The convolution operation slides along the time dimension, outputting a shape of... The feature map.

[0062] Subsequently, a global average pooling layer is used to compress the features from 56 time steps into a 16-dimensional vector.

[0063] This process can be represented as: ;in, This represents a single-time-granularity feature vector at the minute level; GAP stands for global average pooling.

[0064] Similarly, we can obtain and ;in, It is a single-time-granularity feature vector at the second level; It is a single-time-granularity feature vector at the hour level.

[0065] The business demand prediction and mapping migration method based on multi-granularity spatiotemporal graphs provided in this invention extracts high-dimensional features that reflect business change patterns within their respective time windows from original data sequences at different time scales such as seconds and minutes by utilizing the local perception characteristics of convolution operations in the time dimension. At the same time, it achieves dimensionality reduction and noise reduction of the data through a global average pooling layer, providing highly abstract and pure feature inputs for subsequent cross-granularity correlation analysis.

[0066] In some embodiments, the step of aggregating all the single-temporal-granularity feature vectors of each terminal to obtain fused temporal features includes: Perform a linear transformation and concatenate all the single-time-granularity feature vectors of different time granularities for each terminal to obtain the concatenated vector; Based on the concatenated vector, determine the temporal attention weights; Based on the time attention weights, all the single-time granularity feature vectors of each terminal are weighted and aggregated to determine the fused time features.

[0067] Specifically, in this embodiment of the invention, a time granularity graph is constructed using each time granularity of any terminal as nodes and the relationships between granularities as edges. A graph attention network is then used to model this time granularity graph, and the time attention weights between nodes of different time granularities are calculated. For example, a graph containing three nodes, each representing 1 second, 1 minute, and 1 hour respectively, is constructed; the weight relationships between the features of these three nodes are learned through a graph attention network (GAT).

[0068] First, the single-time-granularity feature vectors of each terminal at different time granularities are linearly transformed and concatenated to obtain the concatenated vector. For any terminal, assuming its time granularity set includes second-level, grade-level, and hour-level, for any two time granularity nodes in this set... and First, their corresponding single-temporal-granularity feature vectors are passed through a shared learnable weight matrix. Perform a linear transformation to obtain the transformed features, and then concatenate the two transformed feature vectors to form a concatenated vector.

[0069] Next, based on the concatenated vector, temporal attention weights are determined to quantify the temporal granularity. For time granularity The importance of the nodes. In this embodiment of the invention, a graph attention network can be used to model the temporal granularity graph and calculate the temporal attention coefficients between nodes of different temporal granularities. It should be noted that, in order to make the coefficients between different granularities comparable and satisfy the characteristics of probability distribution, the original attention coefficients are normalized. In this embodiment of the invention, the Softmax function is used to normalize the attention coefficients of the nodes involved. All neighboring nodes The original attention coefficients are normalized to obtain the final temporal attention weights.

[0070] Finally, based on the temporal attention weights, the fused temporal features are determined. That is, the transformed features at each temporal granularity are weighted and aggregated using the calculated temporal attention weights.

[0071] Assuming for a single terminal Given three granularity feature vectors, determine the granularity feature vectors of each granularity. Stacked into a matrix This serves as the input to GAT. Assuming the GAT layer maps 16-dimensional features to 8 dimensions, then the weight matrix... Attention vector Let's take calculating the attention of "second-level" nodes to "minute-level" nodes as an example: First, perform a linear transformation and concatenation on the single-time-granularity feature vectors: in, It is a second-level time granularity feature vector; This is a feature vector with a time granularity of minutes. and All The vectors, after concatenation, are ; It is a non-linear activation function; The unnormalized attention coefficient represents the importance score of the attention of "second-level" nodes to "minute-level" nodes.

[0072] Then, normalizing the attention coefficient yields the attention weights: in, The raw importance score for the attention of "second-level" nodes to "minute-level" nodes; The raw importance score for the attention of "second-level" nodes to "hour-level" nodes; The attention weights are normalized from the "second-level" nodes to the "minute-level" nodes.

[0073] Finally, the single-temporal-granularity feature vectors are weighted and aggregated based on the obtained attention weights: in, These are nonlinear activation functions, such as the Rectified Linear Unit (ReLU) and the Hyperbolic Tangent (Tanh). Hourly time-granularity feature vectors; The attention weights are normalized from the "second-level" nodes to the "minute-level" nodes; The attention weights are normalized from the "hour-level" nodes to the "minute-level" nodes; This is the weight matrix.

[0074] Finally, the three updated node features are aggregated, such as by summation, to obtain the final fused time features of the terminal. .

[0075] The business demand prediction and mapping migration method based on multi-granularity spatiotemporal graphs provided in this invention innovatively adopts graph attention networks in the time dimension to model the correlation between these features of different time granularities, thereby capturing the intrinsic connection between long-term trends and short-term fluctuations and generating an adaptive fused time feature vector that integrates multi-granularity information, thus significantly improving the accuracy, sensitivity and noise resistance of business demand prediction.

[0076] In some embodiments, the spatial feature aggregation of the business behavior data of each terminal to obtain fused spatial features includes: A spatial graph is constructed using each terminal as a node and the business relationships between terminals as edges. Based on the spatial graph, spatial attention weights are calculated between the target terminal and its neighboring terminals; wherein, the target terminal refers to any one of the terminals. Based on the spatial attention weights, the business behavior data of the target terminal's neighboring terminals are weighted and aggregated to determine the fused spatial features of the target terminal.

[0077] Specifically, in industrial IoT scenarios, terminal devices often do not operate in isolation; they are interconnected by complex collaborative logic or physical coupling. For example, data from sensor A triggers the action of actuator B. In this embodiment of the invention, by aggregating spatial features of the business behavior data of each terminal, a fused spatial feature is obtained that incorporates the spatial collaborative correlation between the terminals. This explicitly models the interdependence between devices, thereby utilizing information from neighboring nodes to help improve the prediction accuracy of the target node.

[0078] First, each terminal in the network, such as a sensor or controller, is treated as a graph node; a spatial graph is constructed using the service relationships between terminals as edges. In this context, business relationships refer to the interdependencies, mutual influences, or mutual constraints between different terminals in a network at the levels of business logic, physical environment, or data interaction. It forms the basis for constructing a spatial graph to characterize the collaborative features between devices. In this embodiment of the invention, business relationships include, but are not limited to, logical relationships, physical proximity, and communication relationships.

[0079] Then, after constructing the spatial graph, a graph attention network (GAT) is used to model the graph, learning and aggregating spatial dependency information between different terminal service behaviors. Any target terminal in the graph is denoted as a node. , and its adjacent terminals are denoted as nodes. It automatically calculates the spatial attention weights between them. These spatial attention weights reflect the attention of neighboring terminals. The service status of the target terminal The extent of the impact.

[0080] Finally, based on the calculated spatial attention weights, the information of each neighboring node is aggregated to update the feature representation of the target node. The feature vectors of all neighboring terminals are multiplied by their corresponding spatial attention weights, and the products are summed. The aggregated result constitutes the fused spatial features of the target terminal.

[0081] Assuming there are N sensor nodes in the network, a graph attention network is used to model the spatial graph. The spatial attention weights between the target terminal and its neighboring terminals are calculated. Neighbor features can be aggregated through weighted summation. New features It is a weighted sum of the features of all its neighbors, with the weights being the calculated attention coefficients, as shown in the following formula: in, The weight matrix is ​​a learnable weight matrix; These are nodes The input feature vector; For normalized nodes Its neighboring nodes Attention weights; It is a non-linear activation function, such as the ReLU function.

[0082] Ultimately, the new features of all nodes constitute a spatial feature matrix that integrates spatial correlations. .

[0083] The business demand prediction and mapping migration method based on multi-granularity spatiotemporal graphs provided in this invention model the mutual influence and dependency of business behaviors among multiple terminal devices, and considers both the "correlation between multiple time granularities of a single terminal" and the "spatial correlation between multiple terminals" to construct a three-dimensional "spatiotemporal feature model", thereby greatly improving the accuracy and reliability of business behavior prediction.

[0084] In some embodiments, calculating the spatial attention weights between the target terminal and its neighboring terminals based on the spatial map includes: The fusion time features of each node in the spatial graph are stacked to obtain a stacking matrix; Based on the stacking matrix, the attention coefficient between the target terminal and its neighboring terminals in the spatial graph is calculated. The attention coefficients are normalized to obtain the spatial attention weights.

[0085] Specifically, in this embodiment of the invention, the fusion time features of each node in the spatial graph are stacked to obtain a stacking matrix; based on this stacking matrix, for each edge existing in the spatial graph, i.e., the target terminal... Its adjacent terminal The attention coefficient is calculated for the connections between them.

[0086] First, the target terminal is indexed from the stacking matrix. eigenvectors and adjacent terminals eigenvectors Then, to enhance the expressive power of the features, they are typically mapped to a high-dimensional space using a shared, learnable linear mapping matrix. Subsequently, a self-attention mechanism is employed to calculate the correlation. For example, the two transformed feature vectors are concatenated and then dot-producted with a predefined attention weight vector. The result is then processed by a non-linear activation function to obtain a scalar value. .in, This refers to the unnormalized attention coefficient, which represents the attention ratio between adjacent terminals in the current environment. For the target terminal The original importance score.

[0087] Because different nodes have different numbers of neighbors, and the original coefficients The numerical range is not limited, but normalization is required to ensure that it possesses the properties of a probability distribution and guarantees the numerical stability of the calculation. In this embodiment of the invention, the Softmax function is used to normalize the target terminal. All neighboring nodes, including their own attention coefficients, are normalized.

[0088] Assuming there is in the network There are N sensing nodes, each with a fused temporal feature vector of dimension d. The input is the fused temporal feature of each node. These N feature vectors are stacked to construct a stacked matrix of dimension N×d. ,in, It is a dimension of time characteristics, for example For a certain layer of GAT, the node The feature update process is as follows: For nodes any neighboring node The attention coefficient between them is calculated as follows: in, and These are nodes and The input feature vector; This is the learnable weight matrix for this layer, used to transfer features from... Dimension mapping to Dimensions, for example ; It is a learnable weight vector of a single-layer feedforward neural network; It is a non-linear activation function.

[0089] Next, the attention weights are normalized. The attention coefficients are normalized using the Softmax function to obtain the node values. To all his neighbors Attention weights: in, It is a node The index of all neighboring nodes is a "circular variable" that represents each member in the set.

[0090] The service demand prediction and mapping migration method based on multi-granularity spatiotemporal graphs provided in this invention uses graph attention networks to model the spatial graph, calculates the spatial attention weights between the target terminal and its neighboring terminals, quantifies the mutual dependence strength between different terminals in the network in terms of service behavior, and significantly improves the accuracy and robustness of prediction in multi-device collaborative scenarios.

[0091] In some embodiments, fusing the fused temporal features and the fused spatial features to obtain fused spatiotemporal features, and obtaining communication requirements based on the fused spatiotemporal features, includes: The fused temporal features and the fused spatial features are spliced ​​together or fused through a fusion network to obtain the fused spatiotemporal features; Predict business requirements based on the fused spatiotemporal features; Based on a preset mapping function, the business requirements are mapped and migrated to obtain the communication requirements.

[0092] Specifically, after capturing multi-granular historical patterns in the time dimension and multi-terminal collaborative relationships in the spatial dimension for feature extraction, it is necessary to organically combine these two sets of features to obtain a comprehensive understanding of the network status. In this embodiment of the invention, the fused temporal features and fused spatial features are concatenated or combined through a more complex fusion network such as a gating mechanism. The fused feature vector is then fed into a fully connected layer or a recurrent neural network to ultimately predict the service requirements, namely the next data packet period and packet length for each node.

[0093] For example, for the terminal Its temporal characteristics and spatial features Perform feature concatenation to obtain .

[0094] The fused spatiotemporal features are fed into a multilayer perceptron for prediction. A two-layer MLP is designed with 16-dimensional input and 2-dimensional output, namely packet length and period.

[0095] Among them, the weight matrix , Bias vector: ; This is the activation function.

[0096] Final output .

[0097] Traditional traffic prediction often stops at predicting packet length or traffic volume, failing to directly guide the underlying network's time slot allocation. In this embodiment of the invention, a mapping migration mechanism is introduced, utilizing a preset mapping function to transform the predicted application-layer service requirements into communication requirements understandable to the network layer. Communication requirements refer to quantifiable constraints derived from upper-layer service behavior characteristics, such as data packet sending patterns, that the network layer needs to allocate and schedule resources. In this embodiment, communication requirements include, but are not limited to, latency requirements, transmission rate requirements, and packet loss rate requirements.

[0098] In this embodiment of the invention, communication requirements are obtained by mapping and migrating business requirements based on a preset mapping function. For example: It can be an empirical formula, such as... ,in .

[0099] The precise calculation is as follows: Unit: bps (bits per second).

[0100] This can be obtained by looking up a table based on business priority. For example, critical business is <0.01%, and ordinary business is <0.1%.

[0101] Through mapping and migration, the abstract prediction result of "period 10ms (milliseconds), packet length 64 bytes" is accurately transformed into the communication requirements of "latency <5ms (milliseconds), rate 51.2kbps (kilobits per second), packet loss rate <0.01%".

[0102] The output interface module sends these specific network requirement parameters to the TDMA resource scheduler. Based on these precise requirements, the scheduler allocates the most suitable communication time slots and bandwidth to each terminal, thereby achieving efficient and deterministic utilization of network resources.

[0103] The service demand prediction and mapping migration method based on multi-granularity spatiotemporal graphs provided in this invention transforms the originally abstract service behavior prediction results into specific latency limits, bandwidth requirements, and reliability levels through the mapping migration process. This provides key and reliable input for deterministic scheduling technologies such as TDMA and is a core supporting technology for realizing high-reliability, low-latency communication in the Industrial Internet of Things.

[0104] The apparatus provided in the embodiments of the present invention will be described below. The apparatus described below can be referred to in correspondence with the method described above.

[0105] Figure 4 This is a schematic diagram of the business demand prediction and mapping migration system based on multi-granularity spatiotemporal graphs provided by the present invention, as shown below. Figure 4 As shown, the device includes a data acquisition module 410, a time feature extraction module 420, a time correlation modeling module 430, a spatial feature aggregation module 440, and a business requirement mapping and migration module 450 connected in sequence.

[0106] Data acquisition module 410 is used to collect business behavior data of multiple terminals in the target network at different time granularities; The time feature extraction module 420 is used to process the business behavior data of each terminal at each time granularity to obtain single time granularity feature vectors at different time granularities. The time-related modeling module 430 is used to perform feature aggregation on all the single-time-granularity feature vectors of each terminal to obtain fused time features; The spatial feature aggregation module 440 is used to aggregate spatial features of the business behavior data of each terminal to obtain fused spatial features; The business requirement mapping and migration module 450 is used to fuse the fused temporal features and the fused spatial features to obtain fused spatiotemporal features, and to obtain communication requirements based on the fused spatiotemporal features.

[0107] The service demand prediction and mapping migration system based on multi-granularity spatiotemporal graphs provided in this invention continuously collects service behavior data from multiple terminals in the network at multiple different time granularities. Simultaneously, it models the multi-time granularity correlation of a single terminal and the spatial correlation of multiple terminals, enabling a more comprehensive and profound capture of the complex dynamic patterns of service behavior, significantly improving the accuracy and reliability of prediction. By deeply mining the complex correlation features of service behavior in the time and space dimensions, it achieves high-precision service behavior prediction and intelligently maps the prediction results to specific network communication needs. This allows network resource scheduling to shift from "extensive reservation" to "refined on-demand allocation," guiding precise scheduling of network resources, avoiding resource waste and shortages, and significantly improving network capacity and efficiency. It achieves full-process automation from data perception and intelligent prediction to demand mapping, reducing the complexity and error rate of manual configuration and promoting the intelligent transformation of industrial wireless networks.

[0108] Figure 5 This is a schematic diagram of the structure of the electronic device provided by the present invention, such as... Figure 5As shown, the electronic device may include: a processor 510, a communications interface 520, a memory 530, and a communications bus 540, wherein the processor 510, the communications interface 520, and the memory 530 communicate with each other via the communications bus 540. The processor 510 can call logical commands stored in the memory 530 to execute the methods described in the above embodiments, for example: The system continuously collects service behavior data from multiple terminals in the target network at different time granularities; processes the service behavior data of each terminal at each time granularity to obtain single-time-granularity feature vectors; aggregates the single-time-granularity feature vectors of each terminal to obtain fused time features; aggregates the service behavior data of each terminal to obtain fused spatial features that incorporate the spatial collaborative correlation between terminals; fuses the fused time features and the fused spatial features to obtain fused spatiotemporal features; predicts service requirements based on the fused spatiotemporal features; and maps and migrates the service requirements to obtain communication requirements.

[0109] Furthermore, the logical instructions in the aforementioned memory can be implemented as software functional units and sold or used as independent products, and can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0110] The processor in the electronic device provided in this embodiment of the invention can call logical instructions in the memory to implement the above method. Its specific implementation method is the same as the aforementioned method implementation method and can achieve the same beneficial effects, which will not be repeated here.

[0111] This invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the methods provided in the above embodiments.

[0112] The specific implementation method is the same as the aforementioned method implementation method and can achieve the same beneficial effects, so it will not be repeated here.

[0113] This invention provides a computer program product, including a computer program that, when executed by a processor, implements the method described above.

[0114] The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0115] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part 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, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0116] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for business demand prediction and mapping migration based on multi-granularity spatiotemporal graphs, characterized in that, include: Collect business behavior data of multiple terminals in the target network at different time granularities; The business behavior data of each terminal at each time granularity is processed to obtain single time granularity feature vectors at different time granularities. Feature aggregation is performed on all the single-temporal-granularity feature vectors of each terminal to obtain fused temporal features; Spatial features are aggregated for the business behavior data of each terminal to obtain fused spatial features; The fused temporal features and the fused spatial features are fused to obtain fused spatiotemporal features, and communication requirements are obtained based on the fused spatiotemporal features.

2. The business demand prediction and mapping migration method based on multi-granularity spatiotemporal graphs according to claim 1, characterized in that, The process of processing the business behavior data at each time granularity of each terminal to obtain single-time-granularity feature vectors at different time granularities includes: The business behavior data of each terminal at each time granularity are input into the preprocessing model to obtain the single time granularity feature vector output by the preprocessing model. The preprocessing model includes convolutional layers and pooling layers; the convolutional layers are used to process the business behavior data to obtain feature maps; and the pooling layers are used to compress the feature maps to obtain single-temporal-granularity feature vectors.

3. The business demand prediction and mapping migration method based on multi-granularity spatiotemporal graphs according to claim 1, characterized in that, The step of aggregating all the single-time-granularity feature vectors of each terminal to obtain fused time features includes: Perform a linear transformation and concatenate all the single-time-granularity feature vectors of different time granularities for each terminal to obtain the concatenated vector; Based on the concatenated vector, determine the temporal attention weights; Based on the time attention weights, all the single-time granularity feature vectors of each terminal are weighted and aggregated to determine the fused time features.

4. The business demand prediction and mapping migration method based on multi-granularity spatiotemporal graphs according to claim 1, characterized in that, The step of aggregating spatial features of the business behavior data of each terminal to obtain fused spatial features includes: A spatial graph is constructed using each terminal as a node and the business relationships between terminals as edges. Based on the spatial graph, spatial attention weights are calculated between the target terminal and its neighboring terminals; wherein, the target terminal refers to any one of the terminals. Based on the spatial attention weights, the business behavior data of the target terminal's neighboring terminals are weighted and aggregated to determine the fused spatial features of the target terminal.

5. The business demand prediction and mapping migration method based on multi-granularity spatiotemporal graphs according to claim 4, characterized in that, The step of calculating the spatial attention weights between the target terminal and its neighboring terminals based on the spatial graph includes: The fusion time features of each node in the spatial graph are stacked to obtain a stacking matrix; Based on the stacking matrix, the attention coefficient between the target terminal and its neighboring terminals in the spatial graph is calculated. The attention coefficients are normalized to obtain the spatial attention weights.

6. The business demand prediction and mapping migration method based on multi-granularity spatiotemporal graphs according to claim 1, characterized in that, The process of fusing the fused temporal features and the fused spatial features to obtain fused spatiotemporal features, and obtaining communication requirements based on the fused spatiotemporal features, includes: The fused temporal features and fused spatial features are spliced ​​together or fused through a fusion network to obtain the fused spatiotemporal features; Predict business requirements based on the fused spatiotemporal features; Based on a preset mapping function, the business requirements are mapped and migrated to obtain the communication requirements.

7. A business demand prediction and mapping migration system based on multi-granularity spatiotemporal graphs, characterized in that, include: The data acquisition module is used to collect business behavior data of multiple terminals in the target network at different time granularities; The time feature extraction module is used to process the business behavior data of each terminal at each time granularity to obtain single time granularity feature vectors at different time granularities. The time-related modeling module is used to perform feature aggregation on all the single-time-granularity feature vectors of each terminal to obtain fused time features; The spatial feature aggregation module is used to aggregate spatial features of the business behavior data of each terminal to obtain fused spatial features; The business requirement mapping and migration module is used to fuse the fused temporal features and the fused spatial features to obtain fused spatiotemporal features, and to obtain communication requirements based on the fused spatiotemporal features.

8. 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, it implements the business demand prediction and mapping migration method based on multi-granularity spatiotemporal graphs as described in any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the business demand prediction and mapping migration method based on multi-granularity spatiotemporal graphs as described in any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the business demand prediction and mapping migration method based on multi-granularity spatiotemporal graphs as described in any one of claims 1 to 6.

Images