Crowd gathering prediction method, apparatus and system

Abstract

The present application relates to the field of communications. Provided are a crowd gathering prediction method, apparatus and system. The method comprises: acquiring historical user data and user switching data of a plurality of stations, wherein each station is a base station or a base station cell, the historical user data comprises the number of users of any one of the plurality of stations at any one of a plurality of historical moments, and the user switching data comprises the frequency of switching from one station to another station among the plurality of stations at any one of the plurality of historical moments; and then, on the basis of the historical user data and the user switching data, using a prediction model to predict whether there is an abnormal increase in the number of users of the plurality of stations at a future moment, so as to infer whether crowd gathering occurs in the area covered by the stations. In this way, multi-base-station joint crowd gathering prediction is realized on the basis of the correlation between the number of access users of a station and crowd gathering, thereby improving the accuracy of predicting crowd gathering conditions.

Description

Crowd gathering prediction method, device and system

[0001] This application claims priority to the Chinese patent application filed with the State Intellectual Property Office on September 28, 2023, with application number 202311291079.3 and application name “Crowd Gathering Prediction Method, Device and System”, the entire contents of which are incorporated by reference into this application. Technical Field

[0002] The present application relates to the field of communications, and in particular to a method, device, and system for predicting crowd gatherings. Background Art

[0003] In public space management, it's often necessary to predict crowd gatherings and formulate appropriate management measures based on these situations. Current crowd gathering prediction solutions include those based on non-visual technologies. These predict the future number of users accessing a designated access point (e.g., base station, wireless access point, etc.) based on traffic data, such as the number of users accessing the site. This number is used to represent the crowd gathering situation within the coverage area of ​​the designated access point.

[0004] However, the above-mentioned solution based on non-visual technology can only predict the crowd gathering situation in the covered area based on the data of a single access site. The data of the access site is affected by the signal coverage range of the single access site, and there is a problem of low prediction accuracy.

[0005] Summary of the Invention

[0006] The present application provides a crowd gathering prediction method, device and system, which can solve the problem of low accuracy in predicting future crowd gathering conditions based on traffic data of a single access site.

[0007] In a first aspect, a crowd gathering prediction method is provided. The crowd gathering prediction method includes: first, a computing device obtains historical user data and user switching data of multiple sites. The site is a base station or a base station cell. Then, based on the historical user data and user switching data, the computing device uses a prediction model to predict whether there will be an abnormal increase in the number of users at multiple sites in the future. The increase in the number of users at the first site among the multiple sites indicates that a crowd is gathering in the coverage area of ​​the first site. The historical user data includes: the number of users at any historical moment in multiple historical moments at any historical moment in multiple historical moments. The user switching data includes: the frequency of switching from one site to another site among the multiple sites at any historical moment in multiple historical moments.

[0008] Based on the crowd gathering prediction method described above, the computing device incorporates user handoff data between multiple base stations when predicting whether a site will experience abnormal crowd gatherings. This takes into account the mutual influence between base stations or base station cells, and incorporates historical user and handoff data from multiple sites with larger coverage areas. Based on the correlation between abnormal user growth at a site and crowd gatherings, a multi-base station joint prediction of abnormal crowd gatherings is achieved, improving the accuracy of crowd gathering predictions.

[0009] As one possible implementation, the computing device uses different models to predict different types of abnormal user growth on a site. Types of abnormal user growth can include sudden abnormal user growth and widespread abnormal user growth. For example, if the number of users on a site does not conform to the site's historical user growth patterns, the site can be considered to have experienced sudden abnormal user growth. For another example, if the number of users on a site changes abnormally due to user switching between multiple sites, this can be considered to be widespread abnormal user growth.

[0010] Optionally, the prediction model includes a first prediction model and a second prediction model. The computing device inputs historical user data and user switching data into the first prediction model to obtain reconstructed user data output by the first prediction model. Taking the first historical moment of the first site as an example, the corrected number of users of the first site at the first historical moment is the normal number of users of the first site at the first historical moment predicted by the first prediction model based on the number of users of the first site at multiple historical moments. The computing device calculates the difference between the reconstructed user data and the historical user data. Taking the first historical moment of the first site as an example, the greater the difference between the reconstructed user data and the historical user data corresponding to the first site at the first historical moment, the greater the possibility that the number of users of the first site at the first historical moment will increase abnormally. The computing device inputs the user switching data and the difference into the second prediction model, and determines whether the number of users of multiple sites will increase abnormally at future moments based on the user number status prediction results output by the second prediction model.

[0011] Based on the above implementation, the computing device uses the first prediction model to predict the reconstructed user data to calculate the difference between the reconstructed user data and the historical user data, and then uses the second prediction model to predict whether there will be an abnormal increase in the number of users at multiple sites in the future based on the user switching data and the difference. Since the reconstructed user data is mainly predicted by the first prediction model based on the historical change pattern of the number of users at the site, the first prediction model can be regarded as predicting the sudden abnormal increase in the number of abnormal users. Since the second prediction model outputs the prediction results based on the user switching data and the difference, and integrates the impact of user switching between multiple sites and the change in the number of users at each site on a single site, the second prediction model can be regarded as predicting the propagation of abnormal growth in the number of users. In this way, the computing device models the prediction tasks of sudden and propagation of abnormal growth in the number of users separately, thereby improving the prediction accuracy of crowd gathering caused by different types of abnormal growth in the number of users and the interpretability of the prediction results.

[0012] As a possible implementation manner, the first prediction model and the second prediction model are deep learning models based on graph neural network (GNN).

[0013] Optionally, the input of the first prediction model is first matrix data and graph structure data. The first matrix data is used to represent historical user data. The graph structure data is used to represent user switching data. The graph structure data includes multiple nodes and edges between multiple nodes. Multiple nodes are used to represent multiple sites, and the weight of the edge between two nodes is used to represent the frequency of switching between the two sites. In this way, since the graph structure data can more accurately and comprehensively characterize the user communication relationship between multiple sites, and the graph neural network is suitable for processing graph structure data, the first prediction model can perceive more user communication relationships between multiple sites, thereby improving the accuracy of the prediction, so that the prediction results can better reflect the actual situation.

[0014] Optionally, the input of the second prediction model is the second matrix data and the central graph of each site. The second matrix data is used to represent the difference. The multiple sites include the first site, and the central graph of the first site includes a central node, an associated node, and an edge for connecting the nodes. The central node is used to represent the first site, the associated node is used to represent the site where the user switches with the first site exist, and the edge of the central graph is used to represent the frequency of switching between the two sites. In this way, the second prediction model can perceive the impact of user switching between multiple sites on a single site based on the central graph, thereby predicting the abnormal growth of the number of users of the site, and at the same time, it can obtain the dynamic trend of user switching between multiple sites, so as to discover the abnormal growth of the number of users of potential transmission more timely.

[0015] The central graph may be an ego-network graph.

[0016] As a possible implementation, the multiple sites include a first site, and the output of the second prediction model is third matrix data. The third matrix data includes multiple vectors. The multiple vectors include a first vector, and the first vector is used to represent the number of users of the first site at a future time. When the first vector is greater than or equal to a first threshold, it indicates that the number of users of the first site will increase abnormally at the future time, and it is determined that a crowd will gather in the coverage area of ​​the first site at the future time.

[0017] As a possible implementation method, the first prediction model is a deep learning model based on the long short-term memory network (LSTM), and the second prediction model is a deep learning model based on the graph neural network.

[0018] Optionally, the historical user data includes the number of terminals connected to any of the multiple sites at any one of the multiple historical moments. The computing device inputs the historical user data and user switching data into a first prediction model, i.e., a deep learning model based on a long short-term memory network, to obtain reconstructed user data output by the first prediction model. The reconstructed user data includes the revised number of terminals connected to any of the multiple sites at any one of the multiple historical moments. Taking the first historical moment of a first site as an example, the revised number of users at the first historical moment is the normal number of users connected to the first site at the first historical moment, as predicted by the first prediction model based on the number of users connected to the first site at multiple historical moments. The computing device calculates the difference between the reconstructed user data and the historical user data. Taking the first historical moment of the first site as an example, the greater the difference between the reconstructed user data and the historical user data corresponding to the first site at the first historical moment, the greater the likelihood that the first site experienced abnormal user growth at the first historical moment. The computing device inputs the user switching data and the difference into a second prediction model and, based on the user number status prediction results output by the second prediction model, determines whether the multiple sites will experience abnormal user growth in the future.

[0019] In a second aspect, a crowd gathering prediction device is provided, which includes various modules for executing the crowd gathering prediction method in the first aspect or any possible implementation of the first aspect. The crowd gathering prediction device in the present application may include a data processing module and a prediction module. The data processing module is used to obtain historical user data and user switching data of multiple sites. The site is a base station or a base station cell. The historical user data includes the number of users at any one of the multiple sites at any historical moment in multiple historical moments. The user switching data includes the frequency of switching from one site to another site in the multiple sites at any one of the multiple historical moments. The prediction module is used to use a prediction model to predict whether there will be an abnormal increase in the number of users at multiple sites in the future based on the historical user data and user switching data.

[0020] As a possible implementation, the prediction model includes a first prediction model and a second prediction model. The prediction module is specifically configured to: input historical user data and user switching data into the first prediction model to obtain reconstructed user data output by the first prediction model, wherein the reconstructed user data includes the corrected number of users at any one of multiple sites at any one of multiple historical moments; calculate the difference between the reconstructed user data and the historical user data; input the user switching data and the difference into the second prediction model, and determine whether the multiple sites will experience abnormal user growth in the future based on the user number status prediction results output by the second prediction model.

[0021] Optionally, the first prediction model and the second prediction model are deep learning models based on graph neural networks.

[0022] Optionally, the input of the first prediction model is first matrix data and graph structure data, the first matrix data is used to represent historical user data, the graph structure data is used to represent user switching data, the graph structure data includes multiple nodes and edges between multiple nodes, the multiple nodes are used to represent multiple sites, and the edge between two nodes is used to represent the frequency of switching between the two sites.

[0023] Optionally, the input of the second prediction model is the second matrix data and the central graph of each site, the second matrix data is used to represent the difference, the multiple sites include the first site, the central graph of the first site includes a central node, associated nodes and edges for connecting the nodes, the central node is used to represent the first site, the associated nodes are used to represent the site where user switching occurs with the first site, and the edges of the central graph are used to represent the frequency of switching between the two sites.

[0024] As a possible implementation method, multiple sites include a first site, the output of the second prediction model is third matrix data, the third matrix data includes multiple vectors, the multiple vectors include a first vector, the first vector is used to represent the user number status of the first site at a future moment, when the first vector is greater than or equal to a first threshold, it indicates that there will be an abnormal increase in the number of users of the first site at a future moment, that is, it is determined that a crowd will gather in the coverage area of ​​the first site at a future moment.

[0025] As a possible implementation method, the first prediction model is a deep learning model based on the long short-term memory network, and the second prediction model is a deep learning model based on the graph neural network.

[0026] Optionally, the historical user data includes the number of terminals accessing any one of the multiple sites at any one of the multiple historical moments, and the reconstructed user data includes the corrected number of terminals accessing any one of the multiple sites at any one of the multiple historical moments.

[0027] In a third aspect, a computing device is provided, comprising a memory and a processor, wherein the memory is used to store a set of computer instructions, and when the processor executes the set of computer instructions, it is used to execute the operating steps of the crowd gathering prediction method in the first aspect or any possible implementation of the first aspect.

[0028] In a fourth aspect, a crowd gathering prediction system is provided, which includes multiple terminals, multiple base stations and the computing device described in the third aspect, multiple base stations are connected to the multiple terminals, and the computing device is used to collect historical user data and user switching data of multiple base stations, and execute the operating steps of the crowd gathering prediction method in the first aspect or any possible implementation method of the first aspect.

[0029] In addition, the technical effects of the second, third and fourth aspects can refer to the technical effects of the crowd gathering prediction method described in the first aspect and will not be repeated here.

[0030] In a fifth aspect, a readable storage medium is provided. The readable storage medium includes a computer program or instructions that, when executed on a computer, causes the computer to execute the crowd gathering prediction method described in any possible implementation of the first aspect.

[0031] In a sixth aspect, a computer program product is provided. The computer program product includes a computer program or instructions, which, when executed on a computer, causes the computer to execute the crowd gathering prediction method described in any possible implementation of the first aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] FIG1 is a schematic diagram of the architecture of a computer system provided by the present application;

[0033] FIG2 is a flow chart of a crowd gathering prediction method provided by this application;

[0034] FIG3 is a flow chart of a crowd gathering prediction step provided by this application;

[0035] FIG4 is a schematic diagram of a graph structure data provided by this application;

[0036] FIG5 is a schematic diagram of an egocentric network diagram provided by this application;

[0037] FIG6 is a flow chart of another crowd gathering prediction step provided by the present application;

[0038] FIG7 is a schematic diagram of the structure of a crowd gathering prediction device provided by the present application;

[0039] FIG8 is a schematic diagram of the structure of a computing device provided by the present application;

[0040] FIG9 is a schematic diagram of the structure of a crowd gathering prediction system provided in this application. DETAILED DESCRIPTION

[0041] The crowd gathering prediction method, device and system provided in the embodiments of the present application can be used in scenarios where crowd gathering prediction and early warning are provided to telecommunications operators, planning departments, event organizers, etc. The following is a brief introduction to the technologies that may be involved in this application.

[0042] (1) Neural Network

[0043] A neural network can be composed of neurons, which can be represented by x s The output of this operation unit satisfies the following formula (1).

[0044] Among them, h W,b is the output of the operation unit, x is the input of the operation unit, s=1, 2, ... n, n is a natural number greater than 1, W s is x sThe weight of the operation unit, b is the bias of the operation unit. f is the activation function of the neuron, which is used to introduce nonlinear characteristics into the neural network to convert the input signal in the neuron into the output signal. The output signal of the activation function can be used as the input of the next layer, and the activation function can be a sigmoid function. A neural network is a network formed by connecting multiple single neurons mentioned above, that is, the output of one neuron can be the input of another neuron. The input of each neuron can be connected to the local receptive field of the previous layer to extract the features of the local receptive field. The local receptive field can be an area composed of several neurons. The weight represents the strength of the connection between different neurons. The weight determines the influence of the input on the output. A weight close to 0 means that changing the input does not change the output. A negative weight means that increasing the input reduces the output.

[0045] In some feasible embodiments, the neural network's input signals can be various forms of signals, such as video signals, image signals, matrix data, and graph structure data. Neural network input signals also include various other computer-processable engineering signals, which are not listed here. Using neural networks for deep learning in predicting crowd gatherings at base stations, for example, can improve the accuracy of the neural network's predictions of abnormal crowd gatherings at base stations.

[0046] (2) Graph structure data

[0047] Graph-structured data is data stored in a graph data structure, also known as a graph. Graph-structured data consists of one or more nodes. A node can have zero, one, or more adjacent nodes. The connection between two nodes is called an edge. Nodes are also called vertices.

[0048] Graph data types include undirected, directed, and weighted graphs. Undirected graphs have undirected edges, meaning two connected nodes can reach each other. Directed graphs have directed edges, meaning two connected nodes can only reach each other from one node, depending on the direction of the edge. In weighted graphs, each edge is assigned a weight, which represents the distance or cost from one vertex to another.

[0049] Graph structured data can be represented in the form of an adjacency matrix or an incidence matrix.

[0050] (3) Graph Neural Networks

[0051] Graph neural networks are a general term for algorithms that use neural networks to learn from graph-structured data, extracting and discovering features and patterns within it to meet the demands of graph learning tasks such as clustering, classification, prediction, segmentation, and generation. Graph neural networks are frameworks that leverage deep learning to directly learn from graph-structured data. By developing strategies for nodes and edges within a graph, they transform graph data into standardized representations, which are then fed into a variety of neural networks for training. They achieve excellent results in tasks such as node classification, edge information propagation, and graph clustering.

[0052] The graph neural networks mentioned in this application are not limited to specific types, such as graph convolutional network (GCN), graph auto-encoder (gAE), graph generative network (GGN), graph recurrent network (GRN), graph attention network (GAT), etc.

[0053] (4) Long Short-Term Memory Network

[0054] Long Short-Term Memory (LSTM) networks are a type of recurrent neural network (RNN) that address the short-term memory limitations of RNNs. When a sequence is long enough, RNNs struggle to transfer information from earlier time steps to later ones. However, LSTMs can learn long-term dependencies and remember information from earlier time steps, thus maintaining context. Specifically, LSTMs have three gates: the forget gate, the input gate (also known as the update gate or candidate gate), and the output gate. These gates control which information is discarded and which is retained or left unchanged.

[0055] In the issue of public place safety management, how to accurately and timely predict possible crowd gatherings is a key issue.Current crowd gathering prediction solutions include those based on computer vision technology and those based on non-visual technology.

[0056] Solutions based on computer vision technology primarily rely on cameras deployed in various locations. By analyzing the images captured by the cameras, they can calculate and predict crowd size and density in real time. First, this approach requires a large number of cameras, resulting in high deployment and maintenance costs. Second, for large and densely populated areas, relying solely on visual solutions to process large amounts of image data creates high system complexity, making it difficult to ensure timely predictions, which in turn affects prediction accuracy.

[0057] Solutions based on non-visual technologies primarily process and analyze the number of users accessing wireless access points (such as Wi-Fi access points) and base stations to estimate the number of people within a specific area. However, current research on crowd estimation for wireless access points like Wi-Fi access points primarily focuses on how to accurately estimate the number of people within a specific area through signal processing, while relatively little research has been conducted on how to use this information to predict the risk of crowd gatherings. Furthermore, research on crowd estimation for base stations only predicts abnormal crowd gatherings at a single base station based on data from that single base station, without considering the mutual influence between base stations. In reality, the signal coverage range of a single base station is small, and the number of user accesses may be affected by the surrounding base stations. Therefore, solutions based on non-visual technologies, which rely solely on data such as the number of access terminals at a single base station, also face accuracy issues.

[0058] The present application provides a crowd gathering prediction method, specifically relates to a crowd gathering prediction method of "introducing user switching data between base stations to perform abnormal crowd gathering prediction of a prediction model". First, a computing device obtains historical user data and user switching data of multiple sites. Then, based on the historical user data and user switching data of multiple sites, the computing device uses a prediction model to predict whether there will be an abnormal increase in the number of users at multiple sites in the future. The increase in the number of users at the first site among the multiple sites indicates that a crowd has gathered in the coverage area of ​​the first site. The site is a base station or a base station cell. The historical user data includes: the number of users at any site among the multiple sites at any historical moment in multiple historical moments. The user switching data includes: the frequency of switching from one site to another site among the multiple sites at any historical moment in multiple historical moments.

[0059] Based on the crowd gathering prediction method described above, the computing device incorporates user handoff data between multiple base stations when predicting whether a site will experience abnormal crowd gatherings. This takes into account the mutual influence between base stations or base station cells, and incorporates historical user and handoff data from multiple sites with larger coverage areas. Based on the correlation between abnormal user growth at a site and crowd gatherings, a multi-base station joint prediction of abnormal crowd gatherings is achieved, improving the accuracy of crowd gathering predictions.

[0060] The implementation of the embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0061] FIG1 is a schematic diagram of the architecture of a computer system provided by the present application. As shown in FIG1 , the computer system 100 includes an execution device 110 , a training device 120 , a database 130 , a terminal device 140 , a data storage system 150 , and a data acquisition device 160 .

[0062] The execution device 110 may be a terminal, such as a computer, a mobile phone terminal, a tablet computer, a laptop computer, a virtual reality (VR) device, an augmented reality (AR) device, a mixed reality (MR) device, an extended reality (ER) device, a camera, or an in-vehicle computer, or an edge device (e.g., a box with a processing chip). In the present application, the execution device 110 may be a computing device connected to a base station, or a computing device deployed at a base station, such as a server or a cloud device.

[0063] The training device 120 can be a terminal, or other computing device that supports integer calculations, such as a server or a cloud device.

[0064] As a possible embodiment, the execution device 110 and the training device 120 are deployed on different physical devices (such as a server or a server in a cluster), or the execution device 110 and the training device 120 are different physical devices. Exemplarily, the execution device 110 and the training device 120 are processors deployed on different physical devices. For example, the execution device 110 can be a graphics processing unit (GPU), a central processing unit, other general-purpose processors, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc. The training device 120 can be a graphics processing unit (GPU), a neural network processing unit (NPU), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits for controlling the execution of the program of the present application.

[0065] In another possible embodiment, the execution device 110 and the training device 120 are deployed on the same physical device, or the execution device 110 and the training device 120 are the same physical device.

[0066] The data collection device 160 is used to collect training data and store the training data in the database 120. The data collection device 160 can be the same as or different from the execution device 110 and the training device 120. In the embodiment of the present application, the data collection device 160 can be a computing device or network device at an access site such as a base station or a wireless access point.

[0067] The training device 120 is used to train the neural network using training data until the loss function in the neural network converges and the loss function value is less than a specific threshold, indicating that the neural network training is complete, thereby achieving a certain accuracy. For example, the training device 120 uses historical user data and user switching data as training data input and reconstructed user data as output to train the neural network. Another example is that the training device 120 uses user switching data and difference values ​​as training data input and predicts abnormal user growth on a site as output to train the neural network. During the training process, backpropagation is performed based on the loss function, and the neural network parameters are updated using gradients until the loss function value is less than a specific threshold. Alternatively, the neural network training is completed when all training data in the database 130 is used for training, resulting in the trained neural network having the target functions of reconstructing user data and predicting abnormal user growth. The training device 120 then configures the trained neural network 101 to the execution device 110. The execution device 110 is used to implement the function of predicting abnormal user growth based on the trained neural network 101.

[0068] The model obtained by training the neural network with reconstructed user data as output can be called the first prediction model, and the model obtained by training the neural network with the prediction result of abnormal number of users at the site as output can be called the second prediction model.

[0069] Optionally, neural network 101 is used to predict the corrected number of users of a site's historical user data at multiple historical moments, or to predict whether the site will experience an abnormal increase in the number of users in the future. Neural network 101 can be a network type suitable for processing historical user data, user switching data, and differences in the form of sequence data or graph data. For example, neural network 101 can be a convolutional neural network (CNN), a recurrent neural network, or a graph neural network.

[0070] In some embodiments, the execution device 110 and the training device 120 are the same computing device, and the computing device can configure the trained neural network 101 to itself and use the trained neural network 101 to achieve the above-mentioned target function.

[0071] In other embodiments, the training device 120 may deploy the trained neural network 101 to multiple execution devices 110. Each execution device 110 utilizes the trained neural network 101 to implement the aforementioned target function.

[0072] In combination with the computer system 100, the crowd gathering prediction method provided in this embodiment can be applied to any scenario where it is necessary to predict abnormal states of data traffic, such as fixed networks, mobile networks, and wireless access networks. In fixed network scenarios such as large enterprise networks or school networks, the fixed network may also encounter a situation where the number of users suddenly increases, resulting in excessive network pressure. Therefore, the crowd gathering prediction method provided in this application can be used to predict crowd gatherings and provide abnormal warnings. In mobile network scenarios such as the fourth generation mobile communication technology (4G) network and the fifth generation mobile communication technology (4G) network, the crowd gathering prediction method provided in this application can be used to predict crowd gatherings and provide abnormal warnings based on the service data of the base station. In scenarios of wireless access networks such as Wi-Fi, crowd gathering predictions and abnormal warnings can be performed based on the service data of Wi-Fi to assist administrators in managing Wi-Fi network resources. In the above-mentioned fixed network, mobile network, and wireless access network scenarios, the training and prediction process of the neural network 101 will be adaptively adjusted according to the different input data and output data, which will not be described in detail here.

[0073] In some embodiments, the crowd gathering prediction method can be applied to various public places in a city, such as commercial areas, schools, parks, and stadiums. For example, during large-scale music festivals or sports competitions, the computer system 100 can use the crowd gathering prediction method to predict and issue early warnings of abnormal crowd gatherings, helping managers to effectively allocate resources, ensure the normal operation of the network, and conduct crowd management in advance to avoid possible safety risks.

[0074] In other embodiments, the computer system 100 can use crowd gathering prediction methods to assist network operators in optimizing base station energy usage, improving network service quality, and reducing operating costs. For example, by using crowd gathering prediction methods to monitor and predict network traffic in real time, operators can dynamically adjust network resources based on demand, achieving more refined operational management.

[0075] It should be noted that, in actual applications, the training data maintained in database 130 does not necessarily come from data acquisition device 160, but may also be received from other devices. In addition, training device 120 does not necessarily train the neural network entirely based on the training data maintained in database 130, but may also obtain training data from the cloud or other places to train the neural network. The above description should not be construed as limiting the embodiments of the present application.

[0076] Furthermore, according to the functions performed by the execution device 110, the execution device 110 can be further subdivided into the architecture shown in Figure 1. As shown in Figure 1, the execution device 110 is configured with a computing module 111, an I / O interface 112 and a preprocessing module 113.

[0077] Taking the execution device 110 as an example of a computing device connected to an access site such as a base station or a wireless access point, the computing module 111 can be a GPU, CPU, other general-purpose processors, DSP, ASIC, FPGA or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. on an on-board computer.

[0078] For example, the computing module 111 is used to run the neural network 101 to implement the function of the neural network 101 to reconstruct the historical user data to obtain the reconstructed user data corresponding to the historical user data.

[0079] For another example, the calculation module 111 is configured to calculate the difference between the reconstructed user data and the historical user data, and use the difference as an abnormality coefficient representing a sudden abnormal increase in the number of users.

[0080] For example, the computing module 111 is further configured to run the neural network 101 to implement the neural network 101's function of predicting an abnormal increase in the number of users.

[0081] I / O interface 112 is used for data exchange with external devices. A user can input data into I / O interface 112 via terminal device 140, for example, to instruct execution device 110 to initiate a function to reconstruct historical user data or to predict abnormal user growth in order to implement a crowd gathering prediction method. Alternatively, the input data can come from database 130.

[0082] The preprocessing module 113 is configured to preprocess the input data received by the I / O interface 112. In an embodiment of the present application, the preprocessing module 113 can be configured to generate training data, such as a training set, a validation set, and a test set, based on the input data received from the I / O interface 112. Optionally, the preprocessing module 113 can also perform preprocessing operations such as denoising on the input data, such as matrix data and graph structure data of a base station, to eliminate irrelevant information and restore useful real information.

[0083] When the execution device 110 preprocesses the input data, or when the computing module 111 of the execution device 110 performs calculations and other related processing, the execution device 110 can call the data, code, etc. in the data storage system 150 for corresponding processing, and can also store the data and instructions obtained from the corresponding processing in the data storage system 150.

[0084] Finally, the I / O interface 112 returns the processing result to the terminal device 140, thereby providing it to the user so that the user can view the processing result. In an embodiment of the present application, the processing result can be a prediction result of whether the number of users of the site will increase abnormally in the future.

[0085] The terminal device 140 can serve as a data acquisition terminal, collecting the input data input into the I / O interface 112 and the processing results output from the I / O interface 112 as shown in FIG1 as new sample data, and storing the data in the database 130. Of course, the sample data can also be collected without passing through the terminal device 140, and the I / O interface 112 can store the input data input into the I / O interface 112 and the processing results output from the I / O interface 112 as new sample data in the database 130.

[0086] Figure 1 is only a schematic diagram of a system architecture provided by an embodiment of the present application. The positional relationship between the devices, components, modules, etc. shown in Figure 1 does not constitute any limitation. For example, in Figure 1, the data storage system 150 is an external memory relative to the execution device 110. In other cases, the data storage system 150 can also be placed in the execution device 110.

[0087] Next, the crowd gathering prediction method provided by this embodiment is described in detail with reference to the accompanying drawings.

[0088] In the mobile network scenario, the execution device 110 in the computer system 100 in Figure 1 above can be a computing device capable of obtaining user data from a base station. Therefore, the crowd gathering prediction method will be described in detail below, using the execution device 110 as an example. As shown in Figure 2, the crowd gathering prediction method may include the following steps 210 through 220.

[0089] Step 210: The execution device 110 obtains historical user data and user switching data of multiple sites.

[0090] As a possible example, the execution device 110 obtains historical user data of multiple sites at multiple historical moments, and user switching data, ie, user switching data of multiple sites switching from one site to another at multiple historical moments.

[0091] As a possible implementation, the execution device 110 collects historical user data and user switching data from multiple sites, or reads historical user data and user switching data from multiple sites from the data collection device 160. The historical user data and user switching data in the data collection device 160 are collected and stored by the data collection device 160 from multiple sites.

[0092] Optionally, the site may be a base station or a base station cell.

[0093] Among them, historical user data can be used to describe the number of users at multiple sites at multiple historical moments. User switching data can be used to describe the frequency of each site switching from one site to another at each historical moment. For example, if the site is a base station, the number of times the terminal of base station A switches to base station B is 100 times within 5 hours, and the user switching data can be recorded as 20 times / hour. For another example, if the site is a base station cell, the number of times the terminal of base station cell A switches to base station cell B is 200 times within 4 hours, and the user switching data can be recorded as 50 times / hour. Base station cell A and base station cell B can be base station cells covered by the same or different base stations.

[0094] Step 220: The execution device 110 uses a prediction model to predict whether there will be an abnormal increase in the number of users at multiple sites in the future based on historical user data and user switching data. The increase in the number of users at the first site among the multiple sites indicates that a crowd is gathering in the coverage area of ​​the first site.

[0095] As one possible example, the prediction model includes a first prediction model and a second prediction model. The execution device 110 inputs historical user data and user switching data into the first prediction model to obtain reconstructed user data output by the first prediction model. The reconstructed user data includes the corrected number of users at any one of the multiple sites at any one of the multiple historical moments. The execution device 110 calculates the difference between the reconstructed user data and the historical user data, and then uses the second prediction model to predict whether there will be an abnormal increase in the number of users at the multiple sites in the future based on the user switching data and the difference, that is, whether any one of the multiple sites will experience an abnormal crowd gathering in the future.

[0096] As one possible implementation, execution device 110 uses the second prediction model to predict whether multiple sites will experience an abnormal increase in user numbers in the future based on the user switching data and the difference. This may involve obtaining a user quantity status prediction result output by the second prediction model, and then determining whether multiple sites will experience an abnormal increase in user numbers in the future based on the user quantity status prediction result. Taking the first site among the multiple sites as an example, if the first site experiences an abnormal increase in user numbers in the future, it is determined that a crowd will gather in the coverage area of ​​the first site at the future.

[0097] Optionally, considering that graph-structured data can more intuitively and accurately describe user communication relationships across multiple sites, using a deep learning model for training and prediction based on graph-structured data can improve the accuracy of predicting abnormal crowd gatherings. Before inputting the historical user data and user switching data, execution device 110 preprocesses the historical user data and user switching data to obtain matrix data or graph-structured data. The matrix data and graph-structured data serve as input data for the first and second prediction models.

[0098] Based on the crowd gathering prediction method provided in steps 210-220 above, execution device 110 incorporates user handoff data between multiple base stations when predicting whether abnormal crowd gatherings will occur at a site. This takes into account the mutual influence between base stations or base station cells, incorporates historical user data and user handoff data from multiple sites with larger coverage areas, and implements a multi-base station joint prediction of abnormal crowd gatherings based on the correlation between abnormal user growth at a site and crowd gatherings, thereby improving the accuracy of crowd gathering predictions.

[0099] For the specific steps of determining the above crowd gathering situation, please refer to step 224 shown in Figure 3, which will not be repeated here.

[0100] The above describes the overall process of the crowd gathering prediction method in conjunction with Figure 2. The following describes the process of the execution device 110 using the first prediction model and the second prediction model to predict abnormal crowd gatherings, taking steps 221 to 224 shown in Figure 3 as an example.

[0101] Step 221 : The execution device 110 pre-processes the historical user data and the user switching data.

[0102] As a possible implementation method, the execution device 110 preprocesses the historical user data and user switching data of multiple sites to obtain historical user data of multiple sites in the form of first matrix data, user switching data in the form of graph structure data, and user switching data in the form of a central graph.

[0103] Optionally, the first matrix data is represented as a matrix D, and the element D[i][j] in the matrix D represents the number of users at the i-th site at the j-th moment.

[0104] Optionally, for example, as shown in FIG4 , the graph structure data includes node A representing the first site, node B representing the second site, node C representing the third site, and node C representing the fourth site. The weight of the edge between node A and node B is 3, the weight of the edge between node A and node C is 5, the weight of the edge between node B and node C is 4, the weight of the edge between node C and node D is 3, and the weight of the edge between node A and node D is 6. The numerical value of the edge weight between two nodes represents the frequency of user switching between sites.

[0105] Optionally, the central graph in the present application may be an egocentric network graph, which may also be considered as a graph structure data. As shown in FIG5 , taking the first site as an example, the central graph of the first site may be an egocentric network graph. The central node n of the egocentric network graph of the first site represents the first site, and the associated nodes 1 to 13 represent sites directly or indirectly connected to the first site. The edge between two nodes represents a communication connection between the sites corresponding to the two nodes, and the weight of the edge (not shown in FIG5 ) represents the frequency of user switching between the sites corresponding to the two nodes. S t-1 、S t-2 、S t-3 The indicated ring range is used to indicate the closeness of the relationship between the site corresponding to the associated node and the first site, for example, S t-1 Indicates the site that interacts with the first site through zero network device jumps, S t-2 Indicates the site that interacts with the first site through one network device jump, S t-3 Indicates the site that interacts with the first site through two network device hops.

[0106] For example, edge weights, in matrix form, represent the frequency of user switching between two sites at multiple historical moments. A single egocentric network graph corresponding to a first site can represent the frequency of user switching between the first site and other sites at multiple historical moments. Thus, the central graph of the first site used as input for the second prediction model's prediction of the first site can be a single egocentric network graph.

[0107] For example, edge weights represent the frequency of user switching between two sites at a specific historical moment. A single egocentric network graph corresponding to a first site can only represent the frequency of user switching between the first site and other sites at a specific historical moment. Thus, the central graph of the first site used as input for the second prediction model's prediction of the first site can be multiple egocentric network graphs of the first site at multiple historical moments.

[0108] The above-mentioned egocentric network diagram can completely and accurately characterize the impact of traffic interactions between various sites on the number of users at each site. Using the egocentric network diagram as the input of the second prediction model can enable the second prediction model to dynamically perceive the abnormal growth of the number of users of the spread, so as to more promptly predict whether there will be abnormal crowd gatherings at the site in the future, thereby improving the accuracy and timeliness of the prediction.

[0109] Since preprocessing of historical user data and user switching data is a conventional data format conversion, this application does not elaborate on the specific steps of data format conversion.

[0110] Step 222: The execution device 110 inputs the pre-processed historical user data and user switching data of the multiple sites into the first prediction model to obtain reconstructed user data output by the first prediction model.

[0111] As a possible implementation manner, the execution device 110 inputs the historical user data in the first matrix data form and the user switching data in the graph structure data form into the first prediction model to obtain the reconstructed user data output by the first prediction model.

[0112] Optionally, the reconstructed user data is represented in the form of second matrix data. For example, the second matrix data includes a matrix R, where the element R[i][j] in the matrix R represents the corrected number of users at the i-th site at the j-th moment.

[0113] In a possible embodiment of the present application, the first prediction model may further output predicted user data, which is represented in the form of a prediction matrix. For example, the prediction matrix data includes a matrix P, where the element P[i][j] in the matrix P represents the number of users at the i-th site at the j-th time.

[0114] Step 223: The execution device 110 calculates the difference between the reconstructed user data and the historical user data.

[0115] As a possible implementation manner, the execution device 110 calculates the difference between the reconstructed user data and the historical user data of each site based on a matrix algorithm.

[0116] Optionally, the difference between the reconstructed user data and the historical user data can be anomaly matrix data formed by the difference between matrix R and matrix D, represented as matrix S. The element S[i][j] in matrix S represents the difference between the corrected number of users at site i and the uncorrected number of users at time j. Matrix S can also be called an anomaly coefficient.

[0117] Taking the first site as an example, the larger the anomaly coefficient of the first site at the first historical moment, the more likely it is that the first site experienced an abnormal increase in user numbers at that moment. Therefore, since the reconstructed user data is primarily derived from the first prediction model based on the site's historical user number changes, the prediction results of the first prediction model can be considered to indicate a sudden, abnormal increase in user numbers.

[0118] In a possible embodiment of the present application, when historical user data is missing, the difference between the reconstructed user data and the predicted user data may be used to replace the difference between the reconstructed user data and the historical user data.

[0119] For example, if the historical user data lacks the number of users at the i-th site at time j, the execution device 110 replaces the historical user data with the number of users at the i-th site at time m in the predicted user data, and performs calculations with the corrected number of users at the i-th site at time j in the reconstructed user data to obtain the element value of the anomaly coefficient. The m-th time belongs to the same time period as the j-th time, for example, the m-th time is 8:00 AM on Wednesday of this week, and the j-th time is 8:00 AM on Wednesday of last week.

[0120] In a possible implementation of the present application, the abnormal coefficient may also be obtained by clustering or statistics performed by the execution device 110 based on historical user data, predicted user data and / or reconstructed user data, which will not be described in detail here.

[0121] Step 224 : The execution device 110 inputs the user switching data and the difference into the second prediction model to obtain a user quantity status prediction result output by the second prediction model.

[0122] As a possible implementation manner, the execution device 110 inputs the user switching data in the form of a central graph and the difference in the form of abnormal matrix data into the second prediction model to obtain the user quantity status prediction result output by the second prediction model.

[0123] Optionally, the execution device 110 inputs the user switching data in the form of a central graph of the first site and the difference in the form of abnormal matrix data into the second prediction model, and obtains that the second prediction model outputs the user quantity status prediction result of the first site. For example, the user quantity status prediction result is the third matrix data. Taking the first site as an example, the third matrix data is a vector, which is used to indicate whether there will be an abnormal increase in the number of users of the first site at a future moment. If the vector output by the second prediction model is greater than or equal to the first threshold, it means that the first site will appear at a future moment in the future moment, that is, it indicates that the coverage area of ​​the first site will have a crowd gathering at a future moment. If the vector output by the second prediction model is less than the first threshold, it means that there will be no abnormal increase in the number of users of the first site at a future moment, that is, it indicates that there will be no crowd gathering in the coverage area of ​​the first site at a future moment.

[0124] Optionally, the input of the second prediction model can also be the abnormal matrix data and center map of multiple sites. The second prediction model performs pre-prediction according to the abnormal matrix data and center map of each site in turn, and the vectors output for each site are combined into a third matrix data, and the third matrix data includes multiple vectors. V in the third matrix data i The vector corresponding to the i-th site is used to indicate whether the number of users of the i-th site will increase abnormally in the future.

[0125] In the present application, in order to match the types of input data of the above-mentioned first prediction model and second prediction model to obtain better prediction accuracy, the execution device 110 uses a graph neural network that has a better processing effect on graph structure data to implement the functions of the first prediction model and the second prediction model, that is, the first prediction model and the second prediction model are deep learning models based on graph neural networks.

[0126] Based on the selected deep learning model based on graph neural network, the difference between the training process of neural network based on graph neural network and the conventional model training process lies only in the adaptive changes of input, output and loss function, which will not be elaborated here.

[0127] After the execution device 110 obtains the third matrix data output by the second prediction model, it can also predict abnormal growth in the number of users of multiple sites at future moments based on the third matrix data.

[0128] Taking the first site as an example, when V1 in the third matrix data of the execution device 110 is greater than or equal to the first threshold, it is determined that the number of users at the first site will increase abnormally in the future, and further, it is determined that a crowd will gather in the coverage area of ​​the first site in the future.

[0129] In this way, the crowd gathering prediction method of the present application can effectively predict the crowd gathering situation at multiple sites, and issue early warnings in a timely manner when possible abnormal gathering situations are detected, thereby helping operators and managers take appropriate measures to avoid possible problems and risks.

[0130] The above description, in conjunction with Figures 2-5, illustrates the crowd gathering prediction method provided by this application, taking the first prediction model and the second prediction model as deep learning models based on graph neural networks as an example. This application does not limit the model types of the first prediction model and the second prediction model. For example, the first prediction model and the second prediction model can both be models based on convolutional neural networks or recurrent neural networks, or one can be a model based on a convolutional neural network and the other can be a model based on a recurrent neural network.

[0131] The crowd gathering prediction method provided in this application is described below with reference to Figure 6 , using as an example a first prediction model that is a deep learning model based on a long short-term memory network and a second prediction model that is a deep learning model based on a graph neural network. As shown in Figure 6 , the crowd gathering prediction method may include the following steps 221 '-step 224 '.

[0132] Step 221 ′: Step 221 : The execution device 110 pre-processes the historical user data and the user switching data.

[0133] As a possible implementation method, the execution device 110 preprocesses the historical user data and user switching data of multiple sites to obtain historical user data of multiple sites in the form of first matrix data, user switching data in the form of graph structure data, and user switching data in the form of a central graph.

[0134] The historical user data in the first matrix data format, the user switching data in the graph structure data format, and the user switching data in the central graph format for the multiple sites are the same as the data in step 221, except that the number of users in the historical user data refers to the number of connected terminals. For example, the historical user data includes the number of connected terminals at any one of the multiple sites at any one of the multiple historical moments.

[0135] Optionally, the number of terminals connected to the multiple sites at multiple historical moments is obtained by the execution device 110 collecting statistics on trajectory data of the terminals moving between the multiple sites or user behavior data.

[0136] The collection, storage, use, processing, transmission, provision, and disclosure of user personal information in the technical solutions of this application are in compliance with relevant laws and regulations and do not violate public order and good morals. For example, in the technical solutions of this application, the processing of user personal information is performed with the user's authorization, which is explained here and will not be repeated below.

[0137] Step 222 ′: the execution device 110 inputs the pre-processed historical user data and user switching data of the multiple sites into the first prediction model to obtain reconstructed user data output by the first prediction model.

[0138] As a possible implementation method, the specific implementation method of step 222' is different from the above-mentioned step 222 in that the first prediction model is a deep learning model based on a long short-term memory network, and the historical user data input into the first prediction model includes the number of terminals accessing any site among multiple sites at any historical moment in multiple historical moments. The reconstructed user data output by the first prediction model is obtained after correcting the number of terminals accessing any site at any historical moment where there may be anomalies in the historical user data.

[0139] Step 223 ′: the execution device 110 calculates the difference between the reconstructed user data and the historical user data.

[0140] As a possible implementation method, the specific implementation method of step 223' differs from the above step 223 in that the difference is the difference between the number of terminals accessed by the site at any historical moment after correction and the number of terminals accessed by the site at the corresponding historical moment before correction.

[0141] Step 224 ′: the execution device 110 inputs the user switching data and the difference into the second prediction model to obtain a user quantity status prediction result output by the second prediction model.

[0142] As a possible implementation manner, for the specific implementation manner of step 224 ′, please refer to the relevant description of step 224 above, which will not be repeated here.

[0143] The above description describes in detail the crowd gathering prediction method of the present application in conjunction with Figures 2 to 6. The present application also provides a crowd gathering prediction device in conjunction with the above crowd gathering prediction method. Figure 7 is a schematic structural diagram of a possible crowd gathering prediction device provided by this embodiment. The crowd gathering prediction device can be used to implement the functions of the execution device 110 or the training device 120 in the above method embodiment, and therefore also has the beneficial effects possessed by the above method embodiment. In this embodiment, the crowd gathering prediction device can be the execution device 110, training device 120 or other device with computing functions as shown in Figure 2, or a module (such as a chip) applied to a server.

[0144] As shown in FIG. 7 , the crowd gathering prediction device 700 includes a data processing module 710 and a prediction module 720 .

[0145] The crowd gathering prediction device 700 can be used to implement the functions of the execution device 110 in the method embodiments shown in Figures 2 and / or 3 above, and the various modules included in the crowd gathering prediction device 700 are specifically used to implement the following functions.

[0146] Data processing module 710 is configured to obtain historical user data and user handoff data for multiple sites. The historical user data includes the number of users at any one of the multiple sites at any one of the multiple historical moments. The user handoff data includes the frequency of handoffs from one site to another of the multiple sites at any one of the multiple historical moments. A site may be a base station or a base station cell. For example, data processing module 710 is configured to execute step 210 as shown in FIG. 2 .

[0147] The prediction module 720 is used to predict whether there will be abnormal growth in the number of users at multiple sites in the future based on historical user data and user switching data using a prediction model. For example, the prediction module 720 is used to execute step 220 as shown in FIG2 .

[0148] As a possible implementation, the prediction model includes a first prediction model and a second prediction model. Prediction module 720 is specifically configured to: input historical user data and user switching data into the first prediction model to obtain reconstructed user data output by the first prediction model, where the reconstructed user data includes the corrected number of users at any one of the multiple sites at any one of the multiple historical moments; calculate the difference between the reconstructed user data and the historical user data; input the user switching data and the difference into the second prediction model, and determine whether the multiple sites will experience abnormal user growth in the future based on the user number status prediction results output by the second prediction model.

[0149] Optionally, the first prediction model and the second prediction model are deep learning models based on graph neural networks.

[0150] Optionally, the input of the first prediction model is first matrix data and graph structure data, the first matrix data is used to represent historical user data, the graph structure data is used to represent user switching data, the graph structure data includes multiple nodes and edges between multiple nodes, the multiple nodes are used to represent multiple sites, and the edge between two nodes is used to represent the frequency of switching between the two sites.

[0151] Optionally, the input of the second prediction model is the second matrix data and the central graph of each site, the second matrix data is used to represent the difference, the multiple sites include the first site, the central graph of the first site includes a central node, associated nodes and edges for connecting the nodes, the central node is used to represent the first site, the associated nodes are used to represent the site where user switching occurs with the first site, and the edges of the central graph are used to represent the frequency of switching between the two sites.

[0152] As a possible implementation method, multiple sites include a first site, the output of the second prediction model is third matrix data, the third matrix data includes multiple vectors, the multiple vectors include a first vector, the first vector is used to represent the user number status of the first site at a future moment, and when the first vector is greater than or equal to a first threshold, it indicates that the number of users of the first site will have an abnormal increase in the future moment.

[0153] As a possible implementation method, the first prediction model is a deep learning model based on the long short-term memory network, and the second prediction model is a deep learning model based on the graph neural network.

[0154] Optionally, the historical user data includes the number of terminals accessing any one of the multiple sites at any one of the multiple historical moments, and the reconstructed user data includes the corrected number of terminals accessing any one of the multiple sites at any one of the multiple historical moments.

[0155] It should be understood that the crowd gathering prediction device 700 of the embodiment of the present application may also include any module for implementing the functions of each step in the data processing method, and any module may be implemented by software or hardware. Exemplarily, the implementation of the data processing module 710 is described below using the data processing module 710 as an example. Similarly, the implementation of the prediction module 720 can refer to the implementation of the data processing module 710.

[0156] As an example of a software functional unit, the data processing module 710 may include code running on a computing instance. The computing instance may include at least one of a physical host (computing device), a virtual machine, and a container. Furthermore, the computing instance may be one or more. For example, the data processing module 710 may include code running on multiple hosts / virtual machines / containers. It should be noted that the multiple hosts / virtual machines / containers used to run the code may be distributed in the same region or in different regions. Furthermore, the multiple hosts / virtual machines / containers used to run the code may be distributed in the same availability zone (AZ) or in different AZs, each AZ including one data center or multiple geographically close data centers. Typically, a region may include multiple AZs.

[0157] Similarly, multiple hosts / virtual machines / containers running the code can be distributed within the same virtual private cloud (VPC) or across multiple VPCs. Typically, a VPC is set up within a region. Cross-region communication between two VPCs within the same region, or between VPCs in different regions, requires a communication gateway within each VPC to interconnect the VPCs.

[0158] In this way, the crowd gathering prediction device 700 can be software, or it can be a service provided in the form of SaaS (software as a service). Users can subscribe to the service according to actual needs to obtain the functions of the crowd gathering prediction device 700 provided in this application.

[0159] As an example of a hardware functional unit, the data processing module 710 may include at least one computing device, such as a server. Alternatively, the data processing module 710 may be implemented using an application-specific integrated circuit (ASIC) or a programmable logic device (PLD). The PLD may be a complex programmable logical device (CPLD), a field-programmable gate array (FPGA), a generic array logic (GAL), or any combination thereof.

[0160] The multiple computing devices included in data processing module 710 can be distributed in the same region or in different regions. The multiple computing devices included in data processing module 710 can be distributed in the same AZ or in different AZs. Similarly, the multiple computing devices included in data processing module 710 can be distributed in the same VPC or in multiple VPCs. The multiple computing devices can be any combination of servers, ASICs, PLDs, CPLDs, FPGAs, GALs, and other computing devices.

[0161] It should be noted that, in other embodiments, the data processing module 710 can be used to execute any step in the crowd gathering prediction method, and the prediction module 720 can be used to execute any step in the crowd gathering prediction method. The steps that the data processing module 710 and the prediction module 720 are responsible for implementing can be specified as needed. The full functions of the crowd gathering prediction device 700 are realized by respectively implementing different steps in the crowd gathering prediction method through the data processing module 710 and the prediction module 720.

[0162] For example, FIG8 is a schematic diagram of the structure of a computing device provided in an embodiment of the present application. The computing device can be a terminal device, a network device, a robot, an onboard computer, etc., or a chip (system) or other component or assembly that can be provided in a terminal device, a network device, a robot, or an onboard computer. For example, the computing device provided in an embodiment of the present application is the execution device 110 or the training device 120 in the computer system 100 shown in FIG1 .

[0163] As shown in Figure 8, computing device 800 may include a processor 801. Optionally, computing device 800 may further include a memory 802 and / or a transceiver 803. Processor 801 is coupled to memory 802 and transceiver 803, for example, via a communication bus.

[0164] The following is a detailed introduction to the various components of the computing device 800 with reference to FIG8 :

[0165] The processor 801 is the control center of the computing device 800 and can be a single processor or a collective term for multiple processing elements. For example, the processor 801 can be one or more central processing units (CPUs), a dedicated integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present application, such as one or more microprocessors (digital signal processors, DSPs) or one or more field programmable gate arrays.

[0166] Optionally, the processor 801 can execute various functions of the computing device 800 by running or executing software programs stored in the memory 802 and calling data stored in the memory 802, such as steps 210-220 shown in Figure 2 or the functions implemented by the data processing module 710, prediction module 720 and other modules in the crowd gathering prediction device 700.

[0167] In a specific implementation, as an embodiment, the processor 801 may include one or more CPUs, such as CPU0 and CPU1 shown in FIG8 .

[0168] In a specific implementation, as an embodiment, the computing device 800 may also include multiple processors, such as the processor 801 and the processor 804 shown in FIG8 . Each of these processors may be a single-core processor (single-CPU) or a multi-core processor (multi-CPU). The processor herein may refer to one or more devices, circuits, and / or processing cores for processing data (e.g., computer program instructions).

[0169] Among them, the memory 802 is used to store software programs that execute the solution of the present application, such as a software program that enables the processor 801 to execute steps 210-220 as shown in Figure 2 or the functions implemented by the data processing module 710, prediction module 720 and other modules in the crowd gathering prediction device 700, and is controlled by the processor 801. The specific implementation method can refer to the above-mentioned method embodiment and will not be repeated here.

[0170] Alternatively, the memory 802 may be a read-only memory (ROM) or other type of static storage device that can store static information and instructions, a random access memory (RAM) or other type of dynamic storage device that can store information and instructions, or an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disc storage, an optical disc storage (including a compact disc, laser disc, optical disc, digital versatile disc, Blu-ray disc, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto. The memory 802 may be integrated with the processor 801 or exist independently and be coupled to the processor 801 through an interface circuit (not shown in FIG8 ) of the computing device 800. This embodiment of the present application does not specifically limit this.

[0171] The transceiver 803 is used for communicating with other devices. For example, if the computing device 800 is a base station in a mobile network or a computing device connected to a base station, the transceiver 803 can be used to communicate with another base station or another computing device.

[0172] Optionally, the transceiver 803 may include a receiver and a transmitter (not shown separately in FIG8 ), wherein the receiver is used to implement a receiving function, and the transmitter is used to implement a sending function.

[0173] Optionally, the transceiver 803 can be integrated with the processor 801, or can exist independently and be coupled to the processor 801 through an interface circuit (not shown in FIG. 8 ) of the computing device 800. This embodiment of the present application does not specifically limit this.

[0174] It should be noted that the structure of the computing device 800 shown in FIG8 does not constitute a limitation on the device. The actual device may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently.

[0175] In addition, the technical effects of the computing device 800 can refer to the technical effects of the crowd gathering prediction method described in the above method embodiment, and will not be repeated here.

[0176] Since the computing device 800 of the present application is used to process data from access sites such as base stations and wireless access points to implement the crowd gathering prediction method described in the above method embodiment, the present application also provides a crowd gathering prediction system 900.

[0177] As shown in Figure 9, the crowd gathering prediction system 900 includes multiple terminals 910, multiple base stations 920, and a computing device 800. The multiple terminals 910 are communicatively connected to the multiple base stations 920, and the computing device 800 is connected to the multiple base stations 920 or deployed in one or more of the multiple base stations 920.

[0178] Multiple base stations 920 are used to collect user quantity data of multiple terminals 910 and send the user quantity data to the computing device 800, so that the computing device 800 pre-processes the user quantity data to obtain historical user data and user switching data, etc., and executes the steps of the crowd gathering prediction method provided in this application, such as steps 210-220 shown in Figure 2 or the functions implemented by the data processing module 710, prediction module 720 and other modules in the crowd gathering prediction device 700.

[0179] It should be understood that the processor in the embodiments of the present application may be a central processing unit, but may also be other general-purpose processors, digital signal processors, application-specific integrated circuits, field programmable gate arrays or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0180] It should also be understood that the memory in the embodiments of the present application may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. Among them, the non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic random access memory (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0181] The above embodiments can be implemented in whole or in part by software, hardware (such as circuits), firmware or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer program are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center via a wired method (such as infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a tape), an optical medium (for example, a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0182] 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 beyond the scope of this application.

[0183] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems and devices can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0184] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the modules is merely a logical function division. In actual implementation, there may be other division methods, such as multiple modules 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 modules, which can be electrical, mechanical or other forms.

[0185] The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical modules, that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules may be selected to achieve the purpose of the present embodiment according to actual needs.

[0186] In addition, each functional module in each embodiment of the present application may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0187] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, 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 enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory, a random access memory, a magnetic disk, or an optical disk.

[0188] In this application, "at least one" means one or more, and "more" means two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone, where A and B can be singular or plural. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship. "At least one of the following items" or similar expressions refers to any combination of these items, including any combination of single items or plural items. For example, at least one of a, b or c can mean: a, b, c, a and b, a and c, b and c, or a and b and c, where a, b and c can be single or multiple.

[0189] It should be noted that, in this application, words such as "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described in this application as "exemplary" or "for example" should not be construed as being preferred or advantageous over other embodiments or designs. Rather, the use of words such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0190] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. A crowd gathering prediction method, characterized in that: include: Acquire historical user data and user switching data of multiple sites, wherein the historical user data includes the number of users of any site among the multiple sites at any historical moment of the multiple historical moments, and the user switching data includes the frequency of switching from one site to another site at any historical moment of the multiple historical moments, wherein the site is a base station or a base station cell; Based on the historical user data and the user switching data, a prediction model is used to predict whether there will be an abnormal increase in the number of users at multiple sites in the future. The increase in the number of users at a first site among the multiple sites indicates that a crowd is gathering in the coverage area of ​​the first site.

2. The method according to claim 1, characterized in that The prediction model includes a first prediction model and a second prediction model. The prediction model is used to predict whether there is an abnormal growth in the number of users of multiple sites at a future moment based on the historical user data and the user switching data, including: Inputting the historical user data and the user switching data into the first prediction model to obtain reconstructed user data output by the first prediction model, wherein the reconstructed user data includes a corrected number of users at any one of the multiple sites at any one of the multiple historical moments; Calculating a difference between the reconstructed user data and the historical user data; The user switching data and the difference are input into the second prediction model, and it is determined whether there will be an abnormal increase in the number of users at multiple sites in the future according to the user quantity status prediction result output by the second prediction model.

3. The method according to claim 2, characterized in that The first prediction model and the second prediction model are deep learning models based on graph neural networks.

4. The method according to claim 3, characterized in that The input of the first prediction model is first matrix data and graph structure data, the first matrix data is used to represent the historical user data, the graph structure data is used to represent the user switching data, the graph structure data includes multiple nodes and edges between the multiple nodes, the multiple nodes are used to represent the multiple sites, and the edge between two nodes is used to represent the frequency of switching between the two sites.

5. The method according to claim 3 or 4, characterized in that: The input of the second prediction model is second matrix data and a central graph of each site, the second matrix data is used to represent the difference, the multiple sites include a first site, the central graph of the first site includes a central node, associated nodes and edges for connecting nodes, the central node is used to represent the first site, the associated nodes are used to represent sites with user switching with the first site, and the edges of the central graph are used to represent the frequency of switching between the two sites.

6. The method according to any one of claims 2 to 5, characterized in that: The multiple sites include a first site, the output of the second prediction model is third matrix data, the third matrix data includes multiple vectors, the multiple vectors include a first vector, the first vector is used to represent the user quantity status of the first site at the future moment, when the first vector is greater than or equal to a first threshold value, it indicates that the number of users of the first site has an abnormal growth at the future moment.

7. The method according to claim 2, characterized in that: The first prediction model is a deep learning model based on a long short-term memory network, and the second prediction model is a deep learning model based on a graph neural network.

8. The method according to claim 7, characterized in that The historical user data includes the number of terminals accessing any of the multiple sites at any historical moment in the multiple historical moments, and the reconstructed user data includes the revised number of terminals accessing any of the multiple sites at any historical moment in the multiple historical moments.

9. A crowd gathering prediction device, characterized in that: The invention comprises a module for executing the operation steps of the method described in any one of claims 1 to 8.

10. A computing device, characterized in that The computing device includes a memory and a processor, wherein the memory is used to store a set of computer instructions; when the processor executes the set of computer instructions, the operation steps of any one of the methods described in claims 1-8 are performed.

11. A crowd gathering prediction system, characterized in that: The system includes multiple terminals, multiple sites and the computing device described in claim 10, wherein the multiple sites are connected to the multiple terminals, and the computing device is used to collect historical user data and user switching data of the multiple sites and execute the operating steps of any method described in claims 1-8.

12. A readable storage medium, characterized in that: The readable storage medium includes a computer program or instructions. When the computer program or instructions are executed on a computer, the computer is caused to execute the operation steps of any one of the methods described in claims 1 to 8.

Images