A method, device and equipment for actively predicting hot spots in ultra-dense networking CDS considering spatiotemporal information

Abstract

The present invention provides a method, device and equipment for actively predicting hot spots in ultra-dense networking CDS that considers spatiotemporal information, wherein the prediction method includes: obtaining a spatiotemporal record map of the area to be predicted, the spatiotemporal record map represents a topological structure map of nodes corresponding to multiple different users in the area to be predicted and the relationship between the nodes, the spatiotemporal record map contains multiple subgraphs, and each subgraph corresponds to a different time period; updating the topological structure of the spatiotemporal record map to obtain an updated spatiotemporal record map that is close to the real scene; inputting the updated spatiotemporal record map into a regional prediction model for prediction processing to obtain a target hot spot area in the area to be predicted. The solution provided by the present invention can accurately predict the target hot spot area under the ultra-dense network architecture, thereby providing guidance for content caching and load balancing of edge servers under the ultra-dense network architecture.

Description

Technical Field

[0001] The present invention relates to the technical field of computer information processing, and in particular to a method, device and equipment for actively predicting hot spots in ultra-dense networking CDS considering spatiotemporal information. Background Art

[0002] With the substantial growth of low-latency scenarios such as cloud gaming and autonomous driving, traditional network architectures are unable to meet their needs, and ultra-dense networking (UDN) has emerged. In the UDN architecture, user requests are quickly processed and responded to through a large number of small base stations and cache nodes distributed at the edge of the network. However, due to the high-density deployment of UDN and the significant reduction in the service range of edge nodes, higher requirements are placed on the intelligent scheduling and load balancing capabilities of content distribution services (CDS). Understanding the distribution of CDS is extremely important to ensure server load balancing and continuous and stable low-latency responses for user requests in a high-density UDN environment.

[0003] In the coverage area of ​​each small cell (small base station) in ultra-dense networking, there are many types of user requests (for example, games, videos, etc.). Different symbols are used to represent the main request types of each small cell. When a user makes a request, the intelligent scheduling algorithm will route the request to the edge server closest to the user. However, due to the limited coverage of small cells and frequent changes in user locations, the effectiveness of cached content is often unsustainable, especially in crowded areas and during peak request periods. A large number of service requests lead to a sharp increase in server load and resource allocation difficulties. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a method, device and equipment for actively predicting CDS hotspots in ultra-dense networking that takes into account spatiotemporal information, so as to accurately predict target hotspot areas in crowded areas and during peak service request periods, thereby providing guidance for content caching and load balancing of edge servers under an ultra-dense network architecture.

[0005] In order to solve the above technical problems, an embodiment of the present invention provides an ultra-dense network CDS hotspot active prediction method considering spatiotemporal information, including:

[0006] Acquire a spatiotemporal record graph of the area to be predicted, wherein the spatiotemporal record graph represents a topological structure graph of nodes corresponding to multiple different users in the area to be predicted and the relationships between the nodes, wherein the spatiotemporal record graph includes multiple subgraphs, and each subgraph corresponds to a different time period;

[0007] Updating the topological structure of the space-time record graph to obtain an updated space-time record graph that is close to the actual scene;

[0008] The updated spatiotemporal record graph is input into the regional prediction model for prediction processing to obtain the target hot spot area in the area to be predicted. The regional prediction model is obtained by training the nodes, the connecting edges between nodes and the node features in the historical spatiotemporal record graph based on a preset graph neural network model and a preset selection state space model.

[0009] In one embodiment, obtaining a spatiotemporal record map of a region to be predicted includes:

[0010] Determining a plurality of different users within the area to be predicted;

[0011] The space-time record graph is determined based on the geographic location information and CDS usage intensity information corresponding to the multiple different users, the user correspondence represents the nodes in the space-time record graph, and the relationship between different users represents the connection edges between different nodes in the topological structure graph corresponding to the space-time record graph.

[0012] In one embodiment, determining the spatiotemporal recording graph according to the time information and location information corresponding to the plurality of different users includes:

[0013] Adding the plurality of different users as nodes to a preset graph, and adding connecting edges between the nodes in the preset graph according to the relationships between the plurality of different users;

[0014] Based on a preset coding algorithm, the location information of the user corresponding to each node in each subgraph and the CDS usage intensity information are encoded to obtain multiple subgraphs with node characteristics, and the overall graph composed of multiple subgraphs with node characteristics is determined as the space-time recording graph.

[0015] In one embodiment, the topological structure of the space-time recording graph is updated to obtain an updated space-time recording graph that is close to the real scene, including:

[0016] Determine the number of connection edges to be modified in the topological structure graph corresponding to the space-time record graph;

[0017] According to the similarity scores between any two nodes in each subgraph of the spatiotemporal record graph, the connecting edges between the nodes in each subgraph are modified to obtain each updated subgraph, and the updated spatiotemporal record graph is obtained based on each updated subgraph.

[0018] In one embodiment, the updated spatiotemporal record graph is input into a regional prediction model for prediction to obtain a target hotspot area in the area to be predicted, including:

[0019] Input the updated spatiotemporal record graph into the first graph convolution loop module of the regional prediction model, perform a first convolution loop processing on the nodes and connecting edges in the updated spatiotemporal record graph, so as to obtain a first feature corresponding to each node in each subgraph of the multiple subgraphs of the updated spatiotemporal record graph;

[0020] Inputting the first features corresponding to the nodes of each subgraph in the multiple subgraphs of the updated spatiotemporal record graph into the second graph convolution loop module of the regional prediction model for a second convolution loop processing to obtain the embedding vector corresponding to each subgraph in the multiple subgraphs of the updated spatiotemporal record graph;

[0021] Inputting the embedding vector corresponding to each sub-graph in the multiple sub-graphs of the updated spatiotemporal record graph into the classification prediction module of the regional prediction model for prediction processing to obtain a prediction result for each sub-graph in the multiple sub-graphs of the updated spatiotemporal record graph;

[0022] According to the prediction result of each sub-graph, a target hot spot area in the area to be predicted is determined.

[0023] In one embodiment, the updated spatiotemporal record graph is input into the first graph convolution loop module of the regional prediction model to perform a first convolution loop processing on the nodes and connecting edges in the updated spatiotemporal record graph to obtain the first features corresponding to each node in each subgraph of the multiple subgraphs of the updated spatiotemporal record graph, including:

[0024] Extract features from each node in each subgraph to obtain aggregate information between nodes in each subgraph;

[0025] According to the feature vector of the current node and the feature vector of its related nodes in each subgraph, a correlation matrix between the current node and its related nodes in each subgraph is obtained;

[0026] The aggregation information and the correlation matrix are integrated to obtain a first feature corresponding to each node of each subgraph.

[0027] In one embodiment, the first feature corresponding to each node of each subgraph in the multiple subgraphs of the updated spatiotemporal recording graph is input into the second graph convolution loop module of the regional prediction model for a second convolution loop processing to obtain an embedding vector corresponding to each subgraph in the multiple subgraphs of the updated spatiotemporal recording graph, including:

[0028] Based on the first features corresponding to the nodes of each subgraph in the multiple subgraphs of the updated spatiotemporal record graph, aggregation processing is performed on the connection edges in each subgraph of the updated spatiotemporal record graph to obtain the embedding vector corresponding to each subgraph.

[0029] In one embodiment, the embedding vector corresponding to each sub-graph in the multiple sub-graphs of the updated spatiotemporal record graph is input into the prediction module of the regional prediction model for prediction processing to obtain the prediction result of each sub-graph in the multiple sub-graphs of the updated spatiotemporal record graph, including:

[0030] According to the embedding vector corresponding to each sub-graph, the prediction result corresponding to each sub-graph is determined.

[0031] An embodiment of the present invention further provides an ultra-dense network CDS hotspot active prediction device considering spatiotemporal information, comprising:

[0032] An acquisition module is used to acquire a spatiotemporal record diagram of the area to be predicted, wherein the spatiotemporal record diagram represents a topological structure diagram of nodes corresponding to multiple different users in the area to be predicted and the relationship between the nodes, wherein the spatiotemporal record diagram contains multiple sub-graphs, and each sub-graph corresponds to a different time period;

[0033] A processing module, used for updating the topological structure of the space-time recording graph to obtain an updated space-time recording graph that is close to the real scene;

[0034] A prediction module is used to input the updated spatiotemporal record graph into a regional prediction model for prediction processing to obtain a target hot spot area in the area to be predicted. The regional prediction model is obtained by training the nodes, connecting edges between nodes, and node features in the historical spatiotemporal record graph based on a preset graph neural network model and a preset selection state space model.

[0035] An embodiment of the present invention further provides a computing device, comprising:

[0036] A memory for storing one or more programs;

[0037] One or more processors are used to execute the one or more programs to implement the method described in the above embodiment.

[0038] The above solution of the present invention includes at least the following beneficial effects:

[0039] The above-mentioned scheme of the present invention provides an ultra-dense networking CDS hotspot active prediction method, device and equipment that considers spatiotemporal information, wherein the method obtains a spatiotemporal record map of the area to be predicted, the spatiotemporal record map represents the nodes corresponding to multiple different users in the area to be predicted and the topological structure map of the relationship between the nodes, the spatiotemporal record map contains multiple sub-graphs, and each sub-graph corresponds to a different time period; and the topological structure of the spatiotemporal record map is updated to obtain an updated spatiotemporal record map that is close to the real scene; the updated spatiotemporal record map is further input into the regional prediction model for prediction processing to obtain the target hotspot area in the area to be predicted; here, the regional prediction model is obtained by training the nodes, the connecting edges between the nodes and the node features in the historical spatiotemporal record map based on a preset graph neural network model and a preset selection state space model, so as to accurately predict the target hotspot area in areas with dense traffic and during peak service request periods, thereby providing guidance for content caching and load balancing of edge servers under the ultra-dense network architecture. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 is a flow chart of an active prediction method for ultra-dense networking CDS hotspots considering spatiotemporal information provided by an embodiment of the present invention;

[0041] Figure 2 is an architecture diagram of a regional prediction model provided by an optional embodiment of the present invention;

[0042] Figure 3 is a schematic diagram of training time corresponding to different algorithms on a first data set provided by an optional embodiment of the present invention;

[0043] Figure 4 is a schematic diagram of training time corresponding to different algorithms on a first data set provided by an optional embodiment of the present invention;

[0044] Figure 5 It is a module block diagram of an ultra-dense networking CDS hotspot active prediction device considering spatiotemporal information provided by an embodiment of the present invention;

[0045] Figure 6 is a schematic block diagram of an electronic device provided by an embodiment of the present invention; and

[0046] Figure 7 is a schematic block diagram of a computing device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0047] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0048] In the following description, certain specific details are set forth for the purpose of illustrating the various disclosed embodiments to provide a thorough understanding of the various disclosed embodiments. However, those skilled in the relevant art will recognize that the embodiments may be practiced without one or more of these specific details. In other cases, well-known devices, structures, and techniques associated with the present application may not be shown or described in detail to avoid unnecessarily obscuring the description of the embodiments.

[0049] References throughout the specification to "one embodiment" or "an embodiment" indicate that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment. Thus, the appearances of "in one embodiment" or "in an embodiment" in various places throughout the specification are not necessarily all referring to the same embodiment. Furthermore, the particular features, structures, or characteristics may be combined in any manner in one or more embodiments.

[0050] In the following description, in order to clearly show the structure and working mode of the present invention, many directional words will be used for description, but the words "front", "back", "left", "right", "outside", "inside", "outward", "inward", "up", "down", etc. should be understood as convenient terms and should not be understood as restrictive terms.

[0051] like Figure 1 As shown, an embodiment of the present invention provides an ultra-dense network CDS hotspot active prediction method 10 considering spatiotemporal information, comprising:

[0052] Step 11, obtaining a spatiotemporal record map of the area to be predicted, the spatiotemporal record map represents a topological structure map of nodes corresponding to multiple different users in the area to be predicted and the relationship between the nodes, the spatiotemporal record map contains multiple sub-graphs, and each sub-graph corresponds to a different time period;

[0053] Step 12, updating the topological structure of the space-time record graph to obtain an updated space-time record graph that is close to the real scene;

[0054] Step 13, input the updated spatiotemporal record graph into the regional prediction model for prediction processing to obtain the target hot spot area in the area to be predicted. The regional prediction model is obtained by training the nodes, the connecting edges between nodes and the node features in the historical spatiotemporal record graph based on a preset graph neural network model and a preset selection state space model.

[0055] In this embodiment, the area to be predicted is an area covered by a small base station under an ultra-dense network architecture. A spatiotemporal record map of the current area to be predicted can be constructed based on all users distributed in the area and related information of the users. Each sub-map in the spatiotemporal record map corresponds to a sub-area of ​​the area to be predicted. Here, the related information of the user may include the user's own time attribute information, the user's geographical location information, and CDS usage intensity information.

[0056] Each subgraph in the spatiotemporal record graph contains nodes representing each user and corresponding to each user, as well as connecting edges between nodes representing the relationship between users, wherein each node contains the time attribute information, geographical location information and CDS usage intensity information of the current user;

[0057] Here, after determining the spatiotemporal record graph of the area to be predicted, the topological structure of each subgraph of the spatiotemporal record graph is updated based on the adaptive data enhancement strategy of the graph structure to truly simulate the existence of the real area, and then obtain an updated spatiotemporal record graph close to the real scene to ensure the accuracy of subsequent predictions;

[0058] Furthermore, based on the trained regional prediction model, the updated spatiotemporal record map is predicted and processed; the subsequent embodiments will elaborate on the specific architecture and training process of the regional prediction model;

[0059] Here, the graph convolution loop module in the trained regional prediction model is used to capture and dynamically process the spatial features corresponding to the nodes and the relationship features between the nodes (the relationship features can be represented by the edge features), and the classification prediction model in the trained regional prediction model is used to deeply explore the time dependency between different subgraphs and different nodes, and predict the target subgraph in the multiple subgraphs corresponding to the spatiotemporal record graph, and then determine the sub-region corresponding to the target subgraph as the target hotspot area;

[0060] In the above-mentioned embodiment of the present invention, the initially generated spatiotemporal record graph is updated to simulate a more realistic CDS usage in a real-life scenario, thereby ensuring the accuracy of subsequent target hotspot area predictions; further, the graph convolutional neural network and the selection state space model technology are used in the regional prediction model to capture the spatial characteristics of the nodes in the updated spatiotemporal record graph and the relationship characteristics between the nodes, thereby improving the prediction effect of the model and further ensuring the accuracy of the predicted target hotspot areas, so that the target hotspot areas can be quickly and accurately predicted in crowded areas and during peak service request periods, providing guidance for content caching and load balancing of edge servers under ultra-dense network architectures.

[0061] In an optional embodiment of the present invention, the above step 11 may include:

[0062] Step 111, determining a plurality of different users in the area to be predicted;

[0063] Step 112, determine a spatiotemporal record graph based on the geographic location information and CDS usage intensity information corresponding to multiple different users, where the user correspondence represents the nodes in the spatiotemporal record graph, and the relationship between different users represents the connection edges between different nodes in the topological structure graph corresponding to the spatiotemporal record graph.

[0064] In this embodiment, relevant information of multiple users in the area to be predicted can be collected within a preset time period, and a spatiotemporal record graph can be constructed and determined based on the geographical location information and CDS usage intensity information corresponding to multiple different users, and the geographical location information of each user and the CDS usage intensity information corresponding to the user at the current geographical location are used as the spatial features of the current node, and the connecting edges between the current node and other different nodes are used as the edge features of the current node, thereby obtaining the overall spatiotemporal record graph corresponding to the area to be predicted, so as to facilitate subsequent processing and prediction based on the spatial features and edge features of the spatiotemporal record graph; here, the CDS usage intensity information corresponding to different users represents the total number of CDSs in the area to be predicted, and this number is actually the number of connecting edges in the spatiotemporal record graph corresponding to the area to be predicted;

[0065] Furthermore, the preset time period can be segmented according to preset time intervals, and the record graph corresponding to the user within the corresponding time can be determined as a subgraph of the overall space-time record graph to achieve the segmentation of the space-time record graph. In subsequent processing, each subgraph is processed separately. Since the number of nodes in the subgraph is greatly reduced, the subsequent calculation time and resource consumption will be reduced.

[0066] In an optional embodiment of the present invention, the above step 112 may include:

[0067] Step 1121, adding multiple different users as nodes to a preset graph, and adding connection edges between the nodes in the preset graph according to the relationships between the multiple different users;

[0068] Step 1122, based on a preset coding algorithm, encode the user's location information corresponding to each node in each subgraph and the CDS usage intensity information to obtain multiple subgraphs with node characteristics, and determine the overall graph composed of multiple subgraphs with node characteristics as a spatiotemporal recording graph.

[0069] In this embodiment, the preset graph is a blank graph. According to the geographical location information corresponding to the multiple different users, multiple different users are added as nodes to the preset graph, and connecting edges are added between the corresponding nodes according to the relationship between the users (such as whether two adjacent users are friends), to obtain an initial record graph;

[0070] In an ultra-dense networking environment, the deployment of small base stations reduces the service range and leads to finer regional divisions. In order to ensure the accuracy of subsequent target hotspot area predictions, the location information of the user corresponding to the node and the CDS usage intensity information can be encoded according to a preset coding algorithm, and the result of the encoding processing can be used as the node feature of the current node; the above operation can not only reflect the impact of the CDS usage intensity information in the surrounding area of ​​the area to be predicted on the area to be predicted, but also reflect the location information of the area to be predicted to a certain extent; at the same time, based on the location information of the user corresponding to the node in the current area to be predicted and the CDS usage intensity information in the surrounding area of ​​the current area to be predicted, encoding and processing as embedded node features can achieve accurate description of the key information of the user corresponding to the node in the current area to be predicted, thereby ensuring the accuracy of the subsequent regional prediction model prediction.

[0071] In an implementable example of the present invention, the preset encoding algorithm may include: Geohash encoding and time encoding, encoding based on the geographical location information of the node and the time period corresponding to the CDS usage intensity information to generate high-dimensional node features, which facilitates the subsequent regional prediction model to extract and process the node features;

[0072] Here, a 7-bit Geohash code can be used to express the geographical location information of the current node. Since the geographical location of the current node is further subdivided into 32 sub-location areas with each additional bit of the Geohash code, the length of the first part of the node code is set to 32 bits, and the node is obtained through Geohash code conversion. Here, the first node feature corresponding to the first 32 bits of the node can be specifically expressed as:

[0073]

[0074] in, represents the first node feature of node i in the mth child position; Represents the node number, n is a positive integer; Indicates the number of the sub-location area where the node is located. S is the node The CDS usage intensity information of the sub-region corresponding to the sub-graph is set to the situation of successful use of the service, such as downloading music, etc. Since the Geohash encoding characters are mapped to decimal numbers, ;

[0075] Here, 16-bit time embedding coding can be used to express the node The second node feature corresponding to the second part (the time used for time embedding coding is the time attribute information of the node corresponding to the user itself), then the node The second feature corresponding to the last 16 bits can be expressed as:

[0076]

[0077] in, Represents the second node feature of node i in the mth subgraph;

[0078] Furthermore, the feature vector corresponding to node i in subgraph m It can be expressed as:

[0079]

[0080] In this embodiment, the node feature representation method based on CDS using intensity information and time embedding in the surrounding area can achieve accurate description of the key feature information of the user corresponding to the node; so that in the subsequent regional prediction model processing, the regional prediction model can capture the correlation between the subgraphs in the spatiotemporal record graph and the time correlation of the node corresponding to the user, thereby ensuring the accuracy of the target hot spot area prediction.

[0081] In an optional embodiment of the present invention, the above step 112 may further include:

[0082] Step 1123, based on the geographical location information of the nodes, confirm the edge features of each node in each subgraph, and use the edge features of the current node as the relationship features between different nodes.

[0083] In this embodiment, the The haversine function calculates the distance between different nodes to characterize the edge characteristics of the current node, thereby effectively ensuring the accuracy of the distance calculation between nodes, which is important for hot spots with extremely small spatial granularity.

[0084]

[0085] in: and Is a node The longitude and latitude of Representation Node The longitude and latitude of Representation Node and nodes The edge features connecting the edges, This is the distance formula for calculating two longitude and latitude points; it should be known that for two nodes with connecting edges in a subgraph, their corresponding edge features should be the same.

[0086] In an optional embodiment of the present invention, the above step 12 may include:

[0087] Step 121, determining the number of connection edges to be modified in the topological structure graph corresponding to the spatiotemporal record graph;

[0088] Step 122, based on the similarity scores between any two nodes in each subgraph of the spatiotemporal record graph, modify the connection edges between the nodes in each subgraph to obtain each updated subgraph, and obtain an updated spatiotemporal record graph based on each updated subgraph.

[0089] During service peak hours, excessive load on edge servers can cause CDS failures, leading to service failures. (here, Represents the set of nodes in the topological structure of subgraph m, E represents the set of connected edges in the topological structure of subgraph m), and proposes an algorithm for adaptive enhancement of graph data to modify and update the topological structure in each subgraph to obtain an updated spatiotemporal record graph that is closer to the real scene, thereby ensuring the accuracy of subsequent predictions.

[0090] When performing an update operation, first determine the number of connected edges to be modified in the topological structure graph corresponding to each subgraph. Specifically, the number of connected edges to be modified can be calculated by the following formula:

[0091]

[0092] in, To modify the number of connected edges of an edge, is the modification ratio parameter set, representing the percentage of the modified edge. Indicates the number of edges in each subgraph.

[0093] When two users in the predicted area differ greatly in the spatiotemporal dimension, the similarity score between the corresponding nodes will be significantly reduced, which means that there is potential instability when distributing services, resulting in service interruption. Therefore, in order to ensure the stability of service use, the topological structure in the subgraph of the spatiotemporal record graph needs to be changed according to the similarity score between nodes.

[0094] Here, for a given user content distribution service in a service process, the similarity score between each current node and each of the other remaining nodes in each subgraph is determined in turn:

[0095]

[0096] in, and Is a node and The characteristic vector of for norm;

[0097] Then, based on the similarity score The topological structure of each subgraph is modified and updated according to the value of ; here, the modification and update includes: deleting some connection edges in the subgraph and / or adding some connection edges in the subgraph; specifically:

[0098] The calculation formula for deleting connected edges is as follows:

[0099]

[0100] in, Indicates deleting a node and Connecting Edges The probability of deletion;

[0101] Furthermore, a cumulative distribution function (CDF) is constructed according to the deletion probability and sampling is performed on the distribution to select the node pairs whose connection edges are to be deleted, so as to delete the connection edges between them;

[0102] The probability of adding a connecting edge is calculated as follows:

[0103]

[0104] in, Indicates that at the node and Add connecting edges between The probability of adding

[0105] Similar to the deletion process, a cumulative distribution function (CDF) is constructed according to the deletion probability and a certain number of node pairs are randomly sampled from the subgraph for connection to add connecting edges between nodes while ensuring that the total number of edges in the region remains unchanged.

[0106] In this embodiment, the subgraph topology is modified to make better use of user information in the area to be predicted, and to enhance the edge features corresponding to the nodes in the spatiotemporal record graph, and the information provided by the data is enhanced. The purpose of data enhancement is to make better use of the information provided by the data, so that the features corresponding to the nodes captured when the regional prediction model is input are more complete, thereby ensuring the accuracy of the prediction.

[0107] like Figure 2As shown, in an optional embodiment of the present invention, the regional prediction model may include a first graph convolution loop module, a second graph convolution loop module and a classification prediction module; here, the first graph convolution loop module may be a dynamic model based on the SPGCN graph convolution model, and the second graph convolution loop module may be a dynamic model based on the SEGCN graph convolution model; the updated spatiotemporal record graph is processed by the first graph convolution loop module and the second graph convolution loop module to respectively capture the spatial characteristics and social dependencies of the user's corresponding nodes;

[0108] The classification prediction module can be a Mamba selection state space model; the Mamba selection state space model is used to deeply mine the results processed by the first graph convolution loop module and the second graph convolution loop module to determine the potential time dependency between the subgraphs corresponding to the sub-regions in the predicted area and the nodes corresponding to the users. The regional prediction model is trained based on the nodes, the connecting edges between the nodes, and the node features in the historical spatiotemporal record graph. It should be known that the training process of the regional prediction model is exactly the same as the specific process of the prediction processing described below.

[0109] Specifically, in an optional embodiment of the present invention, the above step 13 may include:

[0110] Step 131, inputting the updated spatiotemporal record graph into the first graph convolution loop module of the regional prediction model, performing a first convolution loop processing on the nodes and connecting edges in the updated spatiotemporal record graph, so as to obtain a first feature corresponding to each node of each subgraph in a plurality of subgraphs of the updated spatiotemporal record graph;

[0111] In this embodiment, the nodes and connection points in the updated spatiotemporal record graph are processed, specifically, the nodes and connection edges in each subgraph of the updated spatiotemporal record graph are processed by convolution loop; here, considering that the behavioral relationship between the users corresponding to the nodes directly affects their distribution, and the differentiated neighbor information aggregation plays an important role in understanding the different social behaviors of the users corresponding to the nodes, which cannot be achieved by using only graph convolution; therefore, the first graph convolution loop module gives the graph convolution model the ability to dynamically adjust weights and correlation aggregation;

[0112] Specifically, the above step 131 may include:

[0113] Step 1311, extracting features from each node in each subgraph to obtain aggregation information between nodes in each subgraph;

[0114] Step 1312, obtaining a correlation matrix between the current node and its related nodes in each subgraph according to the feature vector of the current node and its related nodes in each subgraph;

[0115] Step 1313 , integrating the aggregation information and the correlation matrix to obtain the first feature corresponding to each node of each subgraph.

[0116] Here, feature extraction is performed on each node in each subgraph. Specifically, feature vectors corresponding to the nodes in the subgraph may be preliminarily aggregated to generate a node representation with richer information, namely, first aggregated information, so that each node in each subgraph can learn partial information of the nodes related to it.

[0117] When performing preliminary aggregation, specifically: you can set the first learnable parameter matrix Dynamically aggregate the feature vectors of all nodes ( is the number of nodes, The information of node feature dimension is propagated as follows:

[0118]

[0119] in, Represents the feature vector of node i in the l+1th layer (equal to ) Indicates The feature vector of layer node i, Representation Node and nodes The edge weights between To avoid bias caused by large differences in the social dependencies of users corresponding to different nodes, is the ReLU activation function, Representation Node The set of neighbor nodes (that is, the set of nodes that have connecting edges with node i)

[0120] Here, the node and nodes The edge weight between The calculation of is as follows:

[0121]

[0122] Through the above operations, each node can learn some information about the nodes related to it. However, the behavior pattern of the user corresponding to the node cannot be well captured only by the above operations; graph attention is a good way to understand the social attributes of users in the graph convolution model. However, since the computational complexity of graph attention increases quadratically with the number of nodes, it is not applicable to the spatiotemporal record graph composed of a large number of widely distributed CDS user records; therefore, in this embodiment, the correlation calculation between nodes is used instead of the attention mechanism to understand the behavior pattern of the user corresponding to the node, and to capture the periodicity of the characteristics of the user corresponding to the node;

[0123] The correlation matrix is ​​calculated as follows:

[0124] ;

[0125] ;

[0126]

[0127]

[0128] in, represents the query vector, represents the key vector, represents a value vector, is the set learnable parameter, is the Fourier transform, A correlation matrix representing the behavior patterns between users corresponding to nodes;

[0129] Furthermore, the correlation matrix between nodes is aggregated to obtain the first feature corresponding to each node in the subgraph. Specifically, the specific calculation process of the aggregation process is as follows:

[0130]

[0131]

[0132]

[0133]

[0134] in, Representation Node The correlation between Before obtaining The feature vector with the highest correlation with node i, It represents the result of normalizing the correlation value of each selected vector. is the value vector (R) calculated above, is the learnable projection matrix.

[0135] In an optional embodiment of the present invention, the above step 13 further includes:

[0136] Step 132, input the first feature corresponding to each node of each subgraph in the multiple subgraphs of the updated spatiotemporal record graph into the second graph convolution loop module of the regional prediction model for a second convolution loop processing to obtain the embedding vector corresponding to each subgraph in the multiple subgraphs of the updated spatiotemporal record graph.

[0137] Although the first feature of a node contains location information, the implicit first feature can only express the correlation of surrounding locations. In order to more effectively capture these complex spatial information and the location patterns of the nodes corresponding to the users, it is necessary to aggregate the connecting edges between the nodes in the subgraph of the updated spatiotemporal record graph to obtain more effective and accurate features and ensure the accuracy of subsequent predictions.

[0138] In an optional embodiment of the present invention, the above step 132 may specifically include:

[0139] Step 1321, based on the first features corresponding to the nodes of each subgraph in the multiple subgraphs of the updated spatiotemporal record graph, aggregate the connection edges in each subgraph of the updated spatiotemporal record graph to obtain the embedding vector corresponding to each subgraph.

[0140] Here, when the connection edges between nodes in the subgraph are aggregated, the edge features corresponding to the connection edges between nodes are aggregated. The specific aggregation process is as follows:

[0141]

[0142] in, represents the maximum pooling, represents the feature vector of node i in the l+1th layer, represents a linear layer;

[0143] Furthermore, the feature vectors of all nodes in the subgraph are averaged to obtain the embedding vector of each subgraph. ;

[0144]

[0145] In an optional embodiment of the present invention, the above step 13 further includes:

[0146] Step 133, input the embedding vector corresponding to each sub-graph in the multiple sub-graphs of the updated spatiotemporal record graph into the classification prediction module of the regional prediction model for prediction processing to obtain the prediction result of each sub-graph in the multiple sub-graphs of the updated spatiotemporal record graph.

[0147] For all users corresponding to the nodes, despite location and social differences, there are behavioral similarities in service usage patterns, such as watching videos and playing online games. This similarity is particularly evident in terms of time, such as video watching and game activity during peak hours in the morning and evening. Therefore, capturing and analyzing this global temporality is also important for predicting future high-frequency service usage areas.

[0148] Since traditional graph convolutional networks have limited performance in processing time series data or capturing time sensitivity, they are better at modeling spatial dependencies, but are insufficient in processing dynamic changes in time and capturing user behavior patterns in different time windows. Here, Mamba can be used to select the state space model as the classification prediction module to capture the temporal correlation of subgraphs in different time windows, thereby identifying the potential change trends of CDS records and enhancing the prediction ability of the regional prediction model by integrating this time information.

[0149] Specifically, in an optional embodiment of the present invention, the above step 133 may include:

[0150] Step 1331, determining the prediction result corresponding to each sub-graph according to the embedding vector corresponding to each sub-graph.

[0151] Here, according to the embedding vector corresponding to each sub-graph, the prediction result corresponding to the current sub-graph is determined. The specific process is as follows:

[0152] First, determine the hidden state corresponding to the sub-image through the following calculation formula:

[0153]

[0154]

[0155]

[0156] in, ( is the hidden state dimension), which represents the hidden state of the selection state space at time t, They are respectively represented as the selective parameters of training, is the first output result at time t; Represents the final output, is the output at a certain moment, but a certain area only needs to output one result, so For each moment The final result is obtained by adding up the average.

[0157] Further, It is input into the fully connected layer and sigmoid function of the classification prediction module to obtain the final prediction result corresponding to the current subgraph. The calculation process is as follows:

[0158]

[0159]

[0160] in, To represent the prediction result of the hot spot prediction probability of any sub-graph, W represents the weight matrix, represents the bias; z represents the result of the linear layer. Here, it should be noted that in actual application, the output should be a set of prediction results of each sub-graph corresponding to the entire spatiotemporal record graph. .

[0161] In an optional embodiment of the present invention, the above step 13 further includes:

[0162] Step 134: determine the target hot spot area in the area to be predicted according to the prediction result of each sub-image.

[0163] Here, the prediction result corresponding to each sub-image can be compared with the set probability threshold according to the actual test needs. When the prediction result of the corresponding sub-image is greater than the set probability threshold, it is determined that the current sub-image is the target hot spot area.

[0164] Preferably, here, the performance of the regional prediction model can be measured by the BCELoss loss function, and the calculation process is as follows:

[0165]

[0166] in, Indicates the number of output prediction results, represents the true value, represents the predicted value, Represents a small positive number to avoid the value of 0 in the logarithmic function.

[0167] Through the collaboration of various modules in the regional prediction model, the characteristic information of each node in the spatiotemporal record graph can be fully captured, thereby ensuring the accuracy of the prediction results; here, the regional prediction model can not only be used to predict hot spots, but also has wide applicability for general spatiotemporal graph prediction tasks.

[0168] The prediction method provided in the above embodiment will be specifically described below with a specific comparative example:

[0169] Here, the comparative experiment settings, evaluation indicators, and data sets are described in detail, and the results of the prediction methods provided by the above embodiments and the baseline algorithm for the target hotspot area are analyzed and compared; the specific process is as follows:

[0170] Step 21, data description.

[0171] The Gowalla first dataset is a location-based social dataset, which contains friend relationship data and check-in data, where users share their locations by checking in. The dataset contains 6,442,890 check-ins by users from February 2009 to October 2010.

[0172] The Foursquare second dataset is a social network dataset that records users' check-in data at different locations. The dataset contains 1,021,970 check-ins and 27,098,490 social relationships; each user has an ID and a geospatial location.

[0173] Step 22, experimental setup.

[0174] Step 221, data processing:

[0175] The number of time periods used for training is set to 9, that is, the data of 9 time periods are used to predict a period of time in the future. Therefore, each time window t is represented as 30 days in the first Gowalla dataset and the second Foursquare dataset. In order to focus on the social patterns of users, the three most frequent locations of users are selected in the first Gowalla dataset, and in the second Foursquare dataset, three large areas are divided, that is, areas with similar geographical locations are grouped into one category.

[0176] Table 1 Network information

[0177]

[0178] Step 23, evaluation indicators and baseline algorithms.

[0179] Accuracy, Precision, Recall, F1 score and AUC are used to evaluate the performance of the algorithm. The baseline algorithms can be: GCN (a method that uses the normalized form of the Laplacian matrix to model the graph structure, which can effectively capture the local feature information of the nodes in the graph and is widely used in graph classification tasks), GAT (a method that uses the self-attention mechanism to dynamically calculate the weights between the nodes and their neighbors in the graph, which has a good effect when processing heterogeneous graphs and irregular graph structures) or AGDN (an efficient graph convolutional network, which uses two scalable weighting coefficient mechanisms to capture multi-hop information), etc.

[0180] Step 24, evaluation.

[0181] As shown in Table 2, the prediction method of the present invention performs well in multiple evaluation indicators and shows significant performance advantages. Specifically, the indicator values ​​of the prediction method of the present invention in different categories and different data sets are mostly better than other baseline models. From the numerical analysis, its performance is significantly better than other models at a significance level of 0.01, indicating that it captures the information in the data more comprehensively. In particular, the prediction method of the present invention has an average accuracy of 0.92 on two data sets, an average F1 score of 0.75, and an average AUC of 0.85, which fully demonstrates its ability to comprehensively extract spatiotemporal and social features.

[0182] In contrast, GCN and GAT have small performance differences in most indicators and achieve good classification accuracy in some scenarios. However, these models perform poorly on comprehensive indicators such as recall and F1 score. This phenomenon shows that the method of extracting features based solely on the structural information of the graph (such as the social network characteristics of users) is not sufficient to fully characterize the full picture features required for hotspot prediction. The spatiotemporal attributes of user behavior play a key role in improving model performance, so models that ignore temporal and spatial dynamic factors have limited applicability in complex scenarios.

[0183] On the first Gowalla dataset, although the prediction method of the present invention is lower than AGDN in terms of accuracy, it is significantly better than other baseline models in other indicators, and its AUC reaches 0.83. This shows that the model has strong predictive ability. In the second FourSquare dataset, due to the reduction of data, the prediction difficulty becomes more difficult, but due to the comprehensiveness of the prediction method of the present invention in capturing information in the graph, the Recall and AUC reach 0.85 and 0.88 respectively. This reflects the excellent performance of the model when the amount of data is small.

[0184] Table 2 Performance comparison of Gowalla and Foursquare datasets

[0185]

[0186] In terms of model training time, the experiment fixed the number of training rounds, and the average training time was as follows: Figure 5 As shown in the figure: Since the structure of models such as GCN is relatively simple, the training time is the lowest; the basic structure of SGAT adopts the attention mechanism, and the training time is close to that of GAT; AGDN adds a multi-step propagation mechanism on the basis of the attention mechanism, so the training time is slightly increased; in addition, T-EEGCN introduces LSTM on the basis of graph convolution, which increases the time cost. Although the structure of GCC-M is more complex, its core components, the autocorrelation mechanism and Mamba, do not significantly increase the additional training time cost.

[0187] As shown in Table 3, the prediction method of the present invention significantly outperforms other models in both datasets (Gowalla and Foursquare), especially in comprehensive indicators such as accuracy (Acc), F1 score and AUC. In the first dataset of Gowalla, the average accuracy of the prediction method of the present invention reached 0.94, which is significantly higher than other models; its F1 score (0.73) and AUC (0.84) both indicate its comprehensive ability to capture hotspot features. In the second dataset of Foursquare, the prediction method of the present invention also performed outstandingly, with accuracy (0.90) and AUC (0.88) both leading other models, especially the F1 score of 0.78, which further demonstrated its effective integration of spatiotemporal features and its adaptability to multi-category scenarios.

[0188] In contrast, GCN and GAT have relatively balanced performance on both datasets, but generally lack in F1 score and recall rate in Gowalla dataset (e.g., GCN’s F1 is only 0.50). This shows that the structural information of the dependency graph alone cannot fully capture the features. AGDN achieved suboptimal performance in Gowalla dataset (e.g., precision rate of 0.83) by modeling long-range dependencies, but its overall indicators did not reach the optimal level, especially in Foursquare dataset.

[0189] Table 3 Average performance comparison between Gowalla and Foursquare

[0190]

[0191] The present invention provides a method, device and equipment for actively predicting hot spots in ultra-dense networking CDS that considers spatiotemporal information, wherein the prediction method includes: obtaining a spatiotemporal record map of the area to be predicted, the spatiotemporal record map represents a topological structure map of nodes corresponding to multiple different users in the area to be predicted and the relationships between the nodes, the spatiotemporal record map contains multiple sub-graphs, and each sub-graph corresponds to a different time period; updating the topological structure of the spatiotemporal record map to obtain an updated spatiotemporal record map that is close to the actual scene; inputting the updated spatiotemporal record map into a regional prediction model for prediction processing to obtain a target hot spot area in the area to be predicted. The solution provided by the present invention can accurately predict the target hot spot area under an ultra-dense network architecture, thereby providing guidance for content caching and load balancing of edge servers under the ultra-dense network architecture.

[0192] like Figure 5 As shown, an embodiment of the present invention further provides an ultra-dense network CDS hotspot active prediction device 20 considering spatiotemporal information, comprising:

[0193] An acquisition module 21 acquires a spatiotemporal record map of the area to be predicted, wherein the spatiotemporal record map represents a topological structure map of nodes corresponding to multiple different users in the area to be predicted and the relationship between the nodes, and the spatiotemporal record map contains multiple sub-maps, and each sub-map corresponds to a different time period;

[0194] The processing module 22 is used to update the topological structure of the space-time record graph to obtain an updated space-time record graph that is close to the real scene;

[0195] The prediction module 23 is used to input the updated spatiotemporal record graph into the regional prediction model for prediction processing to obtain the target hot spot area in the area to be predicted. The regional prediction model is obtained by training the nodes, the connecting edges between nodes and the node features in the historical spatiotemporal record graph based on the preset graph neural network model and the preset selection state space model.

[0196] It should be noted that this device is a device corresponding to the above-mentioned ultra-dense networking CDS hotspot active prediction method 10 considering spatiotemporal information. All implementation methods in the above-mentioned method embodiment are applicable to this embodiment and can achieve the same technical effect.

[0197] like Figure 6 As shown, an embodiment of the present invention further provides an electronic device 50, including: a memory 51 for storing one or more computer programs; one or more processors 52 for executing one or more computer programs, and when the computer program is executed by the processor, the prediction method 10 of the hot spot area under the ultra-dense network as described above is executed. All implementations in the above method embodiment are applicable to this embodiment, and can also achieve the same technical effect. The electronic device 50 is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workbenches, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices and other similar computing devices. The components shown in the present invention, their connections and relationships, and their functions are only examples, and are not intended to limit the implementation of the present invention described and / or required in the present invention.

[0198] like Figure 7As shown, the electronic device 50 is a computing device, or a computer system, which may include a CPU 501 (computing unit), which can perform various appropriate actions and processes according to a computer program stored in a ROM 502 (read-only memory) or a computer program loaded from a storage unit 508 into a random access RAM 503 (memory). In the RAM 503, various programs and data required for the operation of the device 500 may also be stored. The CPU 501, the ROM 502, and the RAM 503 are connected to each other via a bus 504. An I / O interface 505 (input / output interface) is also connected to the bus 504.

[0199] A number of components in the electronic device 500 are connected to the I / O interface 505, including: an input unit 506, such as a keyboard, a mouse, etc.; an output unit 507, such as various types of displays, speakers, etc.; a storage unit 508, such as a disk, an optical disk, etc.; and a communication unit 509, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 509 allows the device 500 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0200] CPU 501 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of CPU 501 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. CPU 501 performs the various methods and processes described above. For example, in some embodiments, the prediction method 10 for hotspot areas under ultra-dense networks may be implemented as a computer software program, which is tangibly contained in a computer-readable storage medium, such as a storage unit 508. In some embodiments, part or all of the computer program may be loaded and / or installed on the device 500 via ROM 502 and / or communication unit 509. When the computer program is loaded into RAM 503 and executed by CPU 501, one or more steps of the prediction method 10 for hotspot areas under ultra-dense networks described above may be performed. Alternatively, in other embodiments, the CPU 501 may be configured to execute the method 10 for predicting hotspot areas in an ultra-dense network in any other appropriate manner (eg, by means of firmware).

[0201] The embodiment of the present invention further provides a computer-readable storage medium storing instructions, which, when executed on a computer, enables the computer to execute the above-mentioned method 10 for predicting hotspot areas in an ultra-dense network. All implementations in the above-mentioned method embodiment are applicable to this embodiment and can achieve the same technical effect.

[0202] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0203] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0204] In the embodiments provided by the present invention, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of units is only a logical function division, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0205] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0206] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0207] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it 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 the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for 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 of various embodiments of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, ROM, RAM, disk or optical disk, etc., which can store program codes.

[0208] In addition, it should be pointed out that in the apparatus and method of the present invention, it is obvious that each component or each step can be decomposed and / or recombined. These decompositions and / or recombinations should be regarded as equivalent schemes of the present invention. Moreover, the steps of performing the above series of processing can naturally be performed in chronological order according to the order of description, but it is not necessary to perform them in chronological order, and some steps can be performed in parallel or independently of each other. For those of ordinary skill in the art, it is understandable that all or any steps or components of the method and apparatus of the present invention can be implemented in hardware, firmware, software or a combination thereof in any computing device (including processors, storage media, etc.) or a network of computing devices, which can be achieved by those of ordinary skill in the art using their basic programming skills after reading the description of the present invention.

[0209] Therefore, the purpose of the present invention can also be achieved by running a program or a group of programs on any computing device. The computing device can be a well-known general-purpose device. Therefore, the purpose of the present invention can also be achieved by simply providing a program product containing a program code for implementing a method or device. That is to say, such a program product also constitutes the present invention, and a storage medium storing such a program product also constitutes the present invention. Obviously, the storage medium can be any well-known storage medium or any storage medium developed in the future. It should also be pointed out that in the device and method of the present invention, it is obvious that each component or each step can be decomposed and / or recombined. These decompositions and / or recombinations should be regarded as equivalent schemes of the present invention. In addition, the steps of performing the above-mentioned series of processing can naturally be performed in chronological order according to the order of description, but it is not necessary to perform them in chronological order. Some steps can be performed in parallel or independently of each other.

[0210] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. An active prediction method for CDS hotspots in ultra-dense networking considering spatiotemporal information, characterized in that: include: Acquire a spatiotemporal record graph of the area to be predicted, wherein the spatiotemporal record graph represents a topological structure graph of nodes corresponding to multiple different users in the area to be predicted and the relationships between the nodes, wherein the spatiotemporal record graph includes multiple subgraphs, and each subgraph corresponds to a different time period; Updating the topological structure of the space-time record graph to obtain an updated space-time record graph that is close to the actual scene; Inputting the updated spatiotemporal record graph into a regional prediction model for prediction processing to obtain a target hotspot area in the area to be predicted, wherein the regional prediction model is obtained by training the nodes, the connecting edges between the nodes, and the node features in the historical spatiotemporal record graph based on a preset graph neural network model and a preset selection state space model; Wherein, obtaining the spatiotemporal record map of the area to be predicted includes: Determining a plurality of different users within the area to be predicted; The space-time record graph is determined based on the geographic location information and CDS usage intensity information corresponding to the multiple different users, the user correspondence represents the nodes in the space-time record graph, and the relationship between different users represents the connection edges between different nodes in the topological structure graph corresponding to the space-time record graph.

2. The method for actively predicting hot spots in ultra-dense networking CDS considering spatiotemporal information according to claim 1, characterized in that: Determining the time-space recording graph according to the time information and location information corresponding to the multiple different users includes: Adding the plurality of different users as nodes to a preset graph, and adding connecting edges between the nodes in the preset graph according to the relationships between the plurality of different users; Based on a preset coding algorithm, the location information of the user corresponding to each node in each subgraph and the CDS usage intensity information are encoded to obtain multiple subgraphs with node characteristics, and the overall graph composed of the multiple subgraphs with node characteristics is determined as the space-time recording graph.

3. The method for actively predicting hot spots in ultra-dense networking CDS considering spatiotemporal information according to claim 1, characterized in that: The topological structure of the space-time record graph is updated to obtain an updated space-time record graph that is close to the actual scene, including: Determine the number of connection edges to be modified in the topological structure graph corresponding to the space-time record graph; According to the similarity scores between any two nodes in each subgraph of the spatiotemporal record graph, the connecting edges between the nodes in each subgraph are modified to obtain each updated subgraph, and the updated spatiotemporal record graph is obtained based on each updated subgraph.

4. The method for actively predicting hot spots in ultra-dense networking CDS considering spatiotemporal information according to claim 1, characterized in that: Inputting the updated spatiotemporal record map into a regional prediction model for prediction to obtain a target hotspot area in the area to be predicted, including: Inputting the updated spatiotemporal record graph into the first graph convolution module of the regional prediction model, performing a first convolution process on the nodes and connecting edges in the updated spatiotemporal record graph, so as to obtain a first feature corresponding to each node in each subgraph of the multiple subgraphs of the updated spatiotemporal record graph; Inputting the first features corresponding to the nodes of each subgraph in the multiple subgraphs of the updated spatiotemporal record graph into the second graph convolution module of the regional prediction model for a second convolution process to obtain an embedding vector corresponding to each subgraph in the multiple subgraphs of the updated spatiotemporal record graph; Inputting the embedding vector corresponding to each sub-graph in the multiple sub-graphs of the updated spatiotemporal record graph into the classification prediction module of the regional prediction model for prediction processing to obtain a prediction result for each sub-graph in the multiple sub-graphs of the updated spatiotemporal record graph; According to the prediction result of each sub-graph, a target hot spot area in the area to be predicted is determined.

5. The method for actively predicting hot spots in ultra-dense networking CDS considering spatiotemporal information according to claim 4, characterized in that: Inputting the updated spatiotemporal record graph into the first graph convolution loop module of the regional prediction model, performing a first convolution loop processing on the nodes and connecting edges in the updated spatiotemporal record graph to obtain a first feature corresponding to each node in each subgraph of the multiple subgraphs of the updated spatiotemporal record graph, including: Extract features from each node in each subgraph to obtain aggregate information between nodes in each subgraph; According to the feature vector of the current node and the feature vector of its related nodes in each subgraph, a correlation matrix between the current node and its related nodes in each subgraph is obtained; The aggregation information and the correlation matrix are integrated to obtain a first feature corresponding to each node of each subgraph.

6. The method for actively predicting hot spots in ultra-dense networking CDS considering spatiotemporal information according to claim 4, characterized in that: Inputting the first feature corresponding to each node of each subgraph in the multiple subgraphs of the updated spatiotemporal record graph into the second graph convolution loop module of the regional prediction model for a second convolution loop processing to obtain an embedding vector corresponding to each subgraph in the multiple subgraphs of the updated spatiotemporal record graph, including: Based on the first features corresponding to the nodes of each subgraph in the multiple subgraphs of the updated spatiotemporal record graph, aggregation processing is performed on the connection edges in each subgraph of the updated spatiotemporal record graph to obtain the embedding vector corresponding to each subgraph.

7. The method for actively predicting hot spots in ultra-dense network CDS considering spatiotemporal information according to claim 4, characterized in that In, Inputting the embedding vector corresponding to each sub-graph in the multiple sub-graphs of the updated spatiotemporal record graph into the prediction module of the regional prediction model for prediction processing to obtain the prediction result of each sub-graph in the multiple sub-graphs of the updated spatiotemporal record graph, including: According to the embedding vector corresponding to each sub-graph, the prediction result corresponding to each sub-graph is determined.

8. An ultra-dense network CDS hotspot active prediction device considering spatiotemporal information, characterized in that: include: An acquisition module is used to acquire a spatiotemporal record diagram of the area to be predicted, wherein the spatiotemporal record diagram represents a topological structure diagram of nodes corresponding to multiple different users in the area to be predicted and the relationship between the nodes, wherein the spatiotemporal record diagram contains multiple sub-graphs, and each sub-graph corresponds to a different time period; A processing module, used for updating the topological structure of the space-time recording graph to obtain an updated space-time recording graph that is close to the real scene; A prediction module, used for inputting the updated spatiotemporal record graph into a regional prediction model for prediction processing to obtain a target hot spot area in the area to be predicted, wherein the regional prediction model is obtained by training the nodes, the connecting edges between the nodes, and the node features in the historical spatiotemporal record graph based on a preset graph neural network model and a preset selection state space model; Wherein, obtaining the spatiotemporal record map of the area to be predicted includes: Determining a plurality of different users within the area to be predicted; The space-time record graph is determined based on the geographic location information and CDS usage intensity information corresponding to the multiple different users, the user correspondence represents the nodes in the space-time record graph, and the relationship between different users represents the connection edges between different nodes in the topological structure graph corresponding to the space-time record graph.

9. A computing device, characterized in that include: A memory for storing one or more programs; One or more processors, configured to execute the one or more programs to implement the method according to any one of claims 1 to 7.

Images