Capacity prediction, model training method and device, equipment, computer program

Abstract

The application relates to the technical field of communication, and discloses a capacity prediction method and device, a model training method and device, equipment and a computer program. The application obtains second space-time capacity fusion data sets at different moments according to collected second user historical network data, and inputs the second space-time capacity fusion data sets into a target capacity prediction model to obtain capacity prediction results of each network node at a preset moment, wherein the target capacity prediction model comprises space features and time features among the network nodes; the application solves the problem of low network capacity prediction accuracy in the prior art, the obtained target capacity prediction model comprises space features and time features among the network nodes, therefore, the capacity prediction results of each network node at the preset moment obtained by using the target capacity prediction model also comprise space features and time features among the network nodes, and the accuracy of network capacity prediction is improved.

Description

Technical Field

[0001] This invention relates to the field of communication technology, and in particular to a capacity prediction method, a capacity prediction model training method, a capacity prediction device, a capacity prediction model training device, a capacity prediction equipment, a capacity prediction model training device, and a computer program. Background Technology

[0002] Currently, wireless network traffic demand is growing rapidly and network structures are becoming increasingly complex. Accurate prediction of network capacity demand in a region or coverage area can effectively allocate resources, improve customer experience, and avoid investment waste.

[0003] In existing technologies, capacity prediction accuracy is low due to various factors; therefore, improving the accuracy of network capacity prediction is an urgent problem to be solved. Summary of the Invention

[0004] The main objective of this invention is to provide a capacity prediction and model training method, apparatus, device, and computer program, which aim to improve the accuracy of network capacity prediction.

[0005] To achieve the above objectives, the present invention provides a capacity prediction method, which includes the following steps:

[0006] Based on the collected historical network usage data of the second user, a second spatiotemporal capacity fusion dataset at different times is obtained;

[0007] The second spatiotemporal capacity fusion dataset is input into the target capacity prediction model to obtain the capacity prediction results of each network node at a preset time; wherein, the target capacity prediction model includes the spatial and temporal characteristics between each network node.

[0008] Optionally, before the step of obtaining the second spatiotemporal capacity fusion dataset at different times based on the collected historical network usage data of the second user, the capacity prediction method further includes:

[0009] Based on the collected historical network usage data of the first user, we can obtain various types of node and edge relationships between network nodes at different times.

[0010] Based on the aforementioned relationships, a capacity prediction model is constructed; wherein, the capacity prediction model includes the spatial and temporal characteristics between each network node;

[0011] Based on the first user's historical network usage data, the capacity prediction model is trained to obtain the target capacity prediction model.

[0012] Optionally, the step of obtaining various types of node-edge associations between network nodes at different times based on the collected historical network usage data of the first user includes:

[0013] Based on the collected historical network usage data of the first user, the first spatiotemporal capacity fusion dataset at different times is obtained;

[0014] Based on the first spatiotemporal capacity fusion dataset, various types of node and edge associations between network nodes at different times are obtained.

[0015] Optionally, before the step of obtaining the first spatiotemporal capacity fusion dataset at different times based on the collected historical network usage data of the first user, the capacity prediction method further includes:

[0016] Collect the first user's historical network usage data; wherein, the first user's historical network usage data includes at least one of the following: minimized drive test data, software-collected data, user detection and response platform data, engineering parameter data, and geolocation data.

[0017] Optionally, the step of obtaining the first spatiotemporal capacity fusion dataset at different times based on the collected historical network usage data of the first user includes:

[0018] If the first user's historical network usage data includes at least two of the following: minimized drive test data, software-collected data, user detection and response platform data, engineering parameter data, and geographic data, then the abnormal data corresponding to each data will be cleaned to obtain the cleaned data.

[0019] Based on the key fields, the various data are fused to obtain the first spatiotemporal capacity fusion dataset.

[0020] Optionally, the step of obtaining various types of node-edge associations between network nodes at different times based on the first spatiotemporal capacity fusion dataset includes:

[0021] From the first spatiotemporal capacity fusion dataset, obtain various types of node and edge association data between network nodes at different times;

[0022] The associated data is converted into association relationships A at different times. k (i,j); where A k (i,j)≠0 indicates that there is an edge of type k between node i and node j;

[0023] According to the aforementioned relationship A k (i,j) obtains adjacency matrices of different types at different times. Wherein, the adjacency matrix is ​​obtained by integrating associations belonging to the same type, K represents the total number of types, k represents the type identifier, N represents the matrix order, and R represents the set of real numbers;

[0024] The step of constructing a capacity prediction model based on the correlation includes:

[0025] Based on the different types of adjacency matrices at different times Build a capacity prediction model.

[0026] Optionally, the types include geographical proximity, topological proximity, scene relevance, and route connectivity;

[0027] The process involves converting the associated data into association relationship A. k The steps for (i,j) include:

[0028] Geographically adjacent related data is converted into a relationship A1(i,j) = 1; where A1(i,j) = 1 indicates that node i and node j are geographically adjacent.

[0029] Convert the topologically adjacent related data into an association relationship A2(i,j)=1; where A2(i,j)=1 indicates that node i and node j are physically topologically adjacent;

[0030] Transform the scene-related associated data into the association relationship A3(i,j)=EJ(s) i ,s j )∈[0,1]; where, A3(i,j)=EJ(s i ,s j )∈[0,1] indicates scene correlation between node i and node j; EJ(.) represents the Tanimoto operator, s i With s j Let i and j represent the scene feature vectors of node i and node j, respectively.

[0031] Convert the associated data of the route connections into A4(i,j)=max{0,I(v i ,v j )-A1(i,j)}∈{0,1}; where, A4(i,j)=max{0,I(v i ,v j )-A1(i,j)}∈{0,1} indicates that nodes i and j are not geographically adjacent but have a connected path; I(.) represents the indicator operator, and v represents a node.

[0032] Optionally, the adjacency matrix of different types at different times... The steps to build a capacity prediction model include:

[0033] Adjacency matrices of different types at the same time The data is spliced ​​together to obtain the spatial characteristics between various network nodes at different times.

[0034] The spatial characteristics of each network node at different times are subjected to spatiotemporal weighted transformation to obtain the temporal characteristics of each network node at different times.

[0035] The temporal characteristics of each network node are fused to construct a capacity prediction model.

[0036] Optionally, the adjacency matrices of different types at the same time can be... The steps for stitching together the data to obtain the spatial characteristics of each network node at different times include:

[0037] According to the adjacency matrix Using Formula 1, obtain the adjacency matrix of the corresponding path;

[0038] Formula 1 is:

[0039]

[0040] Where l represents the maximum number of steps (l) required to reach node j from node i, c represents the number of path types of different lengths, and λ (l) For A k (k = 1, ..., K) represents the weight of each type of edge at layer l, and o represents the dot product operation. express The degree matrix;

[0041] Based on the adjacency matrix of the corresponding path Using Formula 2, the spatial characteristics between the network nodes are obtained;

[0042] Formula 2 is as follows:

[0043]

[0044] in, express The degree matrix, W represents the weight matrix, X represents the input node feature matrix, and σ(.) represents the sigmoid function; C Indicates the number of channels.

[0045] Optionally, the step of performing spatiotemporal weighted transformation on the spatial characteristics between network nodes at different times to obtain the temporal characteristics between network nodes at different times includes:

[0046] Based on the spatial characteristics between network nodes at different times, the pooling result is obtained using Formula 3;

[0047] Formula 3 is as follows:

[0048]

[0049] Where t represents the current time, t' represents different historical times, T represents the set time, v represents a node, V represents the set of nodes, and f pool Represents the pooling function;

[0050] According to the pooling result p (t ' ) Using Formula 4, we obtain the first weighted result;

[0051] Formula four is as follows:

[0052] q=σ(W2δ(W1p)),p (t ' ) ∈p

[0053] Where W1 and W2 represent weight matrices, δ(.) represents the ReLU function, and p represents the set of different time points after pooling all nodes;

[0054] Based on the first weighted result q, the second weighted result at different historical moments is obtained using Formula 5;

[0055] Formula five is:

[0056]

[0057] in, This represents the second weighted result at different historical moments.

[0058] To achieve the above objectives, the present invention provides a capacity prediction model training method, which includes the following steps:

[0059] Based on the collected historical network usage data of the first user, we can obtain various types of node and edge relationships between network nodes at different times.

[0060] Based on the aforementioned relationships, a capacity prediction model is constructed; wherein, the capacity prediction model includes the spatial and temporal characteristics between each network node;

[0061] Based on the first user's historical network usage data, the capacity prediction model is trained to obtain the target capacity prediction model.

[0062] Optionally, the step of obtaining various types of node-edge associations between network nodes at different times based on the collected historical network usage data of the first user includes:

[0063] Based on the collected historical network usage data of the first user, the first spatiotemporal capacity fusion dataset at different times is obtained;

[0064] Based on the first spatiotemporal capacity fusion dataset, various types of node and edge associations between network nodes at different times are obtained.

[0065] Optionally, before the step of obtaining the first spatiotemporal capacity fusion dataset at different times based on the collected historical network usage data of the first user, the capacity prediction method further includes:

[0066] Collect the first user's historical network usage data; wherein, the first user's historical network usage data includes at least one of the following: minimized drive test data, software-collected data, user detection and response platform data, engineering parameter data, and geolocation data.

[0067] Optionally, the step of obtaining the first spatiotemporal capacity fusion dataset at different times based on the collected historical network usage data of the first user includes:

[0068] If the first user's historical network usage data includes at least two of the following: minimized drive test data, software-collected data, user detection and response platform data, engineering parameter data, and geographic data, then the abnormal data corresponding to each data will be cleaned to obtain the cleaned data.

[0069] Based on the key fields, the various data are fused to obtain the first spatiotemporal capacity fusion dataset.

[0070] Optionally, the step of obtaining various types of node-edge associations between network nodes at different times based on the first spatiotemporal capacity fusion dataset includes:

[0071] From the first spatiotemporal capacity fusion dataset, obtain various types of node and edge association data between network nodes at different times;

[0072] The associated data is converted into association relationships A at different times. k (i,j); where A k (i,j)≠0 indicates that there is an edge of type k between node i and node j;

[0073] According to the aforementioned relationship A k (i,j) obtains adjacency matrices of different types at different times. Wherein, the adjacency matrix is ​​obtained by integrating associations belonging to the same type, K represents the total number of types, k represents the type identifier, N represents the matrix order, and R represents the set of real numbers;

[0074] The step of constructing a capacity prediction model based on the correlation includes:

[0075] Based on the different types of adjacency matrices at different times Build a capacity prediction model.

[0076] Optionally, the types include geographical proximity, topological proximity, scene relevance, and route connectivity;

[0077] The process involves converting the associated data into association relationship A. k The steps for (i,j) include:

[0078] Geographically adjacent related data is converted into a relationship A1(i,j) = 1; where A1(i,j) = 1 indicates that node i and node j are geographically adjacent.

[0079] Convert the topologically adjacent related data into an association relationship A2(i,j)=1; where A2(i,j)=1 indicates that node i and node j are physically topologically adjacent;

[0080] Transform the scene-related associated data into the association relationship A3(i,j)=EJ(s) i ,s j )∈[0,1]; where, A3(i,j)=EJ(s i ,s j )∈[0,1] indicates scene correlation between node i and node j; EJ(.) represents the Tanimoto operator, s i With s j Let i and j represent the scene feature vectors of node i and node j, respectively.

[0081] Convert the associated data of the route connections into A4(i,j)=max{0,I(v i ,v j )-A1(i,j)}∈{0,1}; where, A4(i,j)=max{0,I(v i ,v j )-A1(i,j)}∈{0,1} indicates that nodes i and j are not geographically adjacent but have a connected path; I(.) represents the indicator operator, and v represents a node.

[0082] Optionally, the adjacency matrix of different types at different times... The steps to build a capacity prediction model include:

[0083] Adjacency matrices of different types at the same time The data is spliced ​​together to obtain the spatial characteristics between various network nodes at different times.

[0084] The spatial characteristics of each network node at different times are subjected to spatiotemporal weighted transformation to obtain the temporal characteristics of each network node at different times.

[0085] The temporal characteristics of each network node are fused to construct a capacity prediction model.

[0086] Optionally, the adjacency matrices of different types at the same time can be... The steps for stitching together the data to obtain the spatial characteristics of each network node at different times include:

[0087] According to the adjacency matrix Using Formula 1, the adjacency matrix of the corresponding path can be obtained; Formula 1 is:

[0088]

[0089] Where l represents the maximum number of steps (l) required to reach node j from node i, c represents the number of path types of different lengths, and λ (l) For A k (k = 1, ..., K) represents the weight of each type of edge at layer l, and o represents the dot product operation. express The degree matrix;

[0090] Based on the adjacency matrix of the corresponding path Using Formula 2, the spatial characteristics between the network nodes are obtained;

[0091] Formula 2 is as follows:

[0092]

[0093] in, express The degree matrix, W represents the weight matrix, X represents the input node feature matrix, and σ(.) represents the sigmoid function; C Indicates the number of channels.

[0094] Optionally, the step of performing spatiotemporal weighted transformation on the spatial characteristics between network nodes at different times to obtain the temporal characteristics between network nodes at different times includes:

[0095] Based on the spatial characteristics between network nodes at different times, the pooling result is obtained using Formula 3;

[0096] Formula 3 is as follows:

[0097]

[0098] Where t represents the current time, t' represents different historical times, T represents the set time, v represents a node, V represents the set of nodes, and f pool Represents the pooling function;

[0099] According to the pooling result p (t ' ) Using Formula 4, we obtain the first weighted result;

[0100] Formula four is as follows:

[0101] q=σ(W2δ(W1p)),p (t ' ) ∈p

[0102] Where W1 and W2 represent weight matrices, δ(.) represents the ReLU function, and p represents the set of different time points after pooling all nodes;

[0103] Based on the first weighted result q, the second weighted result at different historical moments is obtained using Formula 5;

[0104] Formula five is:

[0105]

[0106] in, This represents the second weighted result at different historical moments.

[0107] Furthermore, to achieve the above objectives, the present invention also provides a capacity prediction device, the capacity prediction device comprising:

[0108] The second acquisition module is used to obtain the second spatiotemporal capacity fusion dataset at different times based on the collected historical network usage data of the second user.

[0109] The third acquisition module is used to input the second spatiotemporal capacity fusion dataset into the target capacity prediction model to obtain the capacity prediction results of each network node at a preset time; wherein, the target capacity prediction model includes the spatial and temporal characteristics between each network node.

[0110] Furthermore, to achieve the above objectives, the present invention also provides a capacity prediction model training device, the capacity prediction model training device comprising:

[0111] The first acquisition module is used to obtain various types of node and edge associations between network nodes at different times based on the collected historical network usage data of the first user.

[0112] A construction module is used to construct a capacity prediction model based on the aforementioned relationships; wherein the capacity prediction model includes spatial and temporal characteristics between various network nodes;

[0113] The training module is used to train the capacity prediction model based on the first user's historical network usage data to obtain the target capacity prediction model.

[0114] In addition, to achieve the above objectives, the present invention also provides a capacity prediction device, the capacity prediction device comprising: a memory, a processor, and a capacity prediction program stored in the memory and running on the processor, wherein the capacity prediction program, when executed by the processor, implements the steps of the capacity prediction method as described above.

[0115] In addition, to achieve the above objectives, the present invention also provides a capacity prediction model training device, the capacity prediction model training device comprising: a memory, a processor, and a capacity prediction model training program stored in the memory and running on the processor, wherein when the capacity prediction model training program is executed by the processor, it implements the steps of the capacity prediction model training method described above.

[0116] In addition, to achieve the above objectives, the present invention also provides a computer program that stores a capacity prediction program, which, when executed by a processor, implements the steps of the capacity prediction method described above.

[0117] Alternatively, the computer program may store a capacity prediction model training program, which, when executed by a processor, implements the steps of the capacity prediction model training method described above.

[0118] The technical solution provided by this invention obtains a second spatiotemporal capacity fusion dataset at different times based on the collected historical network usage data of the second user; the second spatiotemporal capacity fusion dataset is input into a target capacity prediction model to obtain the capacity prediction results of each network node at a preset time; wherein, the target capacity prediction model includes the spatial and temporal characteristics between each network node; thus solving the problem of low accuracy in network capacity prediction in the prior art.

[0119] In other words, the target capacity prediction model obtained by the technical solution provided by the present invention includes the spatial and temporal characteristics between each network node; therefore, the capacity prediction results of each network node at a preset time obtained by using the target capacity prediction model also include the spatial and temporal characteristics between each network node. This takes into account the spatial and temporal dependencies between each network node, thereby avoiding the different influences on capacity prediction caused by distant network nodes and historical observation results of each network node in the prior art, and greatly improving the accuracy of network capacity prediction.

[0120] The technical solution provided by this invention obtains various types of node and edge associations between network nodes at different times based on the collected historical network usage data of the first user; then, based on the associations, a capacity prediction model is constructed; wherein, the capacity prediction model includes the spatial and temporal characteristics between various network nodes; and then, based on the historical network usage data of the first user, the capacity prediction model is trained to obtain the target capacity prediction model; thus solving the problem of low training accuracy of network capacity prediction models in the prior art.

[0121] In other words, the technical solution provided by this invention obtains a target capacity prediction model that includes the spatial and temporal characteristics between various network nodes; that is, the spatial and temporal dependencies between various network nodes are considered during the training process of the capacity prediction model, which greatly improves the accuracy of the network capacity prediction model training. Attached Figure Description

[0122] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0123] Figure 1 This is a schematic diagram of the device structure of the hardware operating environment involved in the embodiments of the present invention;

[0124] Figure 2 This is a flowchart illustrating the first embodiment of the capacity prediction method of the present invention;

[0125] Figure 3 This is a flowchart illustrating the second embodiment of the capacity prediction method of the present invention;

[0126] Figure 4 A schematic diagram of the network structure for constructing the capacity prediction model in the first embodiment of the capacity prediction method of the present invention. Figure 1 ;

[0127] Figure 5 A schematic diagram of the network structure for constructing the capacity prediction model in the first embodiment of the capacity prediction method of the present invention. Figure 2 ;

[0128] Figure 6 This is a flowchart illustrating the first embodiment of the capacity prediction model training method of the present invention;

[0129] Figure 7 This is a structural block diagram of the first embodiment of the capacity prediction device of the present invention;

[0130] Figure 8 This is a structural block diagram of the first embodiment of the capacity prediction model training device of the present invention.

[0131] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0132] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0133] Please see Figure 1 As shown, Figure 1 This is a schematic diagram of the device structure of the hardware operating environment involved in the embodiments of the present invention.

[0134] The equipment can be a capacity prediction device or a training device for a capacity prediction model.

[0135] When the device is a capacity prediction device, the capacity prediction device includes: at least one processor 101, a memory 102, and a capacity prediction program stored in the memory and executable on the processor, the capacity prediction program being configured to implement the steps of the capacity prediction method of any of the following embodiments.

[0136] When the device is a training device for a capacity prediction model, the training device for the capacity prediction model includes: at least one processor 101, a memory 102, and a capacity prediction model training program stored in the memory and executable on the processor, the capacity prediction model training program being configured to implement the steps of the capacity prediction model training method of any of the following embodiments.

[0137] Processor 101 may include one or more processing cores, such as a quad-core processor, an octa-core processor, etc. Processor 101 may be implemented using at least one hardware form selected from DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). Processor 101 may also include a main processor and a coprocessor. The main processor, also known as a CPU (Central Processing Unit), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state. In some embodiments, processor 101 may integrate a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the screen. Processor 101 may also include an AI (Artificial Intelligence) processor, which handles operations related to the capacity prediction model training method, enabling the capacity prediction model to train and learn autonomously, improving efficiency and accuracy.

[0138] The memory 102 may include one or more computer programs, which may be non-transitory. The memory 102 may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices or flash memory devices. In some embodiments, the non-transitory computer program in the memory 102 is used to store at least one instruction, which is executed by the processor 101 to implement the capacity prediction model training method provided in the method embodiments of this application.

[0139] In some examples, the device may also optionally include a communication interface 103 and at least one peripheral device. The processor 101, memory 102, and communication interface 103 can be connected via a bus or signal lines. Each peripheral device can be connected to the communication interface 103 via a bus, signal lines, or a circuit board. Specifically, the peripheral device includes at least one of a radio frequency circuit 104, a display screen 105, and a power supply 106.

[0140] The communication interface 103 can be used to connect at least one I / O (Input / Output) related peripheral device to the processor 101 and the memory 102. In some embodiments, the processor 101, the memory 102, and the communication interface 103 are integrated on the same chip or circuit board; in some other embodiments, any one or two of the processor 101, the memory 102, and the communication interface 103 can be implemented on separate chips or circuit boards, which is not limited in this embodiment.

[0141] The radio frequency (RF) circuit 104 is used to receive and transmit RF (Radio Frequency) signals, also known as electromagnetic signals. The RF circuit 104 communicates with communication networks and other communication devices via electromagnetic signals. The RF circuit 104 converts electrical signals into electromagnetic signals for transmission, or converts received electromagnetic signals back into electrical signals. Optionally, the RF circuit 104 includes: an antenna system, an RF transceiver, one or more amplifiers, a tuner, an oscillator, a digital signal processor, a codec chipset, a user identity module card, etc. The RF circuit 104 can communicate with other terminals via at least one wireless communication protocol. This wireless communication protocol includes, but is not limited to: metropolitan area networks (MANs), various generations of mobile communication networks (2G, 3G, 4G, and 5G), wireless local area networks (WLANs), and / or WiFi (Wireless Fidelity) networks. In some embodiments, the RF circuit 104 may also include circuitry related to NFC (Near Field Communication), which is not limited in this application.

[0142] Display screen 105 is used to display a UI (User Interface). This UI may include graphics, text, icons, videos, and any combination thereof. When display screen 105 is a touch display screen, it also has the ability to collect touch signals on or above its surface. These touch signals can be input as control signals to processor 101 for processing. In this case, display screen 105 can also be used to provide virtual buttons and / or a virtual keyboard, also known as soft buttons and / or a soft keyboard. In some embodiments, display screen 105 can be a single screen, the front panel of the device; in other embodiments, display screen 105 can be at least two, respectively disposed on different surfaces of the device or in a folded design; in some embodiments, display screen 105 can be a flexible display screen, disposed on a curved or folded surface of the device. Furthermore, display screen 105 can be configured as a non-rectangular, irregular shape, i.e., a non-rectangular screen. Display screen 105 can be made of materials such as LCD (Liquid Crystal Display) or OLED (Organic Light-Emitting Diode).

[0143] Power source 106 is used to supply power to the various components in the device. Power source 106 can be AC ​​power, DC power, a disposable battery, or a rechargeable battery. When power source 106 includes a rechargeable battery, the rechargeable battery can support wired or wireless charging. The rechargeable battery can also be used to support fast charging technology.

[0144] Those skilled in the art will understand that Figure 1 The structure shown does not constitute a limitation on the device and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0145] Based on the above hardware structure, various embodiments of the present invention are proposed.

[0146] Please see Figure 2 As shown, Figure 2 This is a flowchart illustrating the first embodiment of the capacity prediction method of the present invention. The capacity prediction method includes the following steps:

[0147] Step S201: Based on the collected historical network usage data of the second user, obtain the second spatiotemporal capacity fusion dataset at different times.

[0148] It should be noted that this embodiment describes how to use a trained target capacity prediction model to predict the capacity of each network node at a preset time. Subsequent embodiments will describe how to construct and train the target capacity prediction model. Both the capacity prediction process and the training of the capacity prediction model involve user historical network usage data, spatiotemporal capacity fusion datasets, etc. To avoid repetitive explanations, please refer to subsequent embodiments.

[0149] It should be clarified that, in this embodiment, the data required for predicting the capacity of each network node at a preset time using the target capacity prediction model is the historical network usage data of the second user, resulting in a second spatiotemporal capacity fusion dataset; in subsequent embodiments, the data required for constructing and training the target capacity prediction model is the historical network usage data of the first user, resulting in a first spatiotemporal capacity fusion dataset. The historical network usage data of the second user and the historical network usage data of the first user can be the same, different, or partially the same; correspondingly, the second spatiotemporal capacity fusion dataset and the first spatiotemporal capacity fusion dataset can be the same, different, or partially the same; in practical applications, these can be flexibly adjusted according to specific application scenarios.

[0150] Step S202: Input the second spatiotemporal capacity fusion dataset into the target capacity prediction model to obtain the capacity prediction results of each network node at a preset time; wherein, the target capacity prediction model includes the spatial and temporal characteristics between each network node.

[0151] Accordingly, in this embodiment, after obtaining the second spatiotemporal capacity fusion dataset at different times based on the collected historical network usage data of the second user, it is necessary to input the second spatiotemporal capacity fusion dataset at different times into the target capacity prediction model to obtain the capacity prediction results of each network node at a preset time.

[0152] For example, the second user's historical network usage data is the user's historical network usage data during the period from May 1, 2020 to May 3, 2020; therefore, the second user's historical network usage data at different times is input into the target capacity prediction model to obtain the capacity prediction results of each network node during the preset period from May 1, 2021 to May 3, 2021.

[0153] In this embodiment, since the obtained target capacity prediction model includes the spatial and temporal characteristics between each network node, the capacity prediction results of each network node at a preset time obtained using this target capacity prediction model also include the spatial and temporal characteristics between each network node. This considers the spatial and temporal dependencies between network nodes, thus avoiding the different impacts on capacity prediction caused by distant network nodes and historical observations of each network node in existing technologies, greatly improving the accuracy of network capacity prediction. Furthermore, this embodiment can more accurately and efficiently locate capacity anomalies and high-load areas (sites), identify future capacity demands in advance, and help predict the flow and aggregation of people or vehicles, providing strong support for network optimization, planning and construction, network quality assurance, and prediction of social gathering events, which is conducive to establishing a differentiated brand advantage.

[0154] Please see Figure 3 As shown, Figure 3 This is a flowchart illustrating a second embodiment of the capacity prediction method of the present invention. In this embodiment, before step S201, which obtains the second spatiotemporal capacity fusion dataset at different times based on the collected historical network usage data of the second user, the capacity prediction method may further include the following steps:

[0155] Step S301: Based on the collected historical network usage data of the first user, obtain the various types of node and edge associations between network nodes at different times.

[0156] It should be noted that steps S301-S303 in this embodiment and step S201 in the first embodiment can be executed in parallel.

[0157] In this embodiment, the first user's historical network usage data refers to the user's network usage data over a past period of time.

[0158] The historical network usage data of the first user in this embodiment includes, but is not limited to, minimized drive test (MDT) data, software-collected data, user detection and response (XDR) platform data, engineering parameter data, and geolocation data. Specifically, minimized drive test (MDT) data optimizes the network based on measurement reports from commercial terminals; software-collected data is generated by analyzing and statistically processing signaling information using software at the Operation and Maintenance Center (OMC) to produce Measurement Report (MR) files; user detection and response (XDR) platform data refers to relevant data retrieved from network sensors, endpoint sensors, and cloud sensors; engineering parameter data is parameter data related to the construction process; and geolocation data is data related to geographical location. It is worth noting that the examples listed here are only a few types of historical network usage data for the first user; in practical applications, these can be flexibly adjusted according to specific application scenarios.

[0159] In this embodiment, step S301, based on the collected historical network usage data of the first user, obtains various types of node and edge associations between network nodes at different times, which may include the following steps:

[0160] First, based on the collected historical network usage data of the first user, the first spatiotemporal capacity fusion dataset at different times is obtained;

[0161] Then, based on the first spatiotemporal capacity fusion dataset, we obtain various types of node and edge associations between network nodes at different times.

[0162] In other words, this embodiment obtains various types of node-edge relationships between network nodes at different times based on the collected historical network usage data of the first user. Specifically, this can be achieved by obtaining a first spatiotemporal capacity fusion dataset at different times based on the collected historical network usage data of the first user, and then obtaining various types of node-edge relationships between network nodes at different times based on the first spatiotemporal capacity fusion dataset. By dividing the historical network usage data of the first user by time, obtaining the first spatiotemporal capacity fusion dataset for the corresponding time, and obtaining various types of node-edge relationships between network nodes at different times, the capacity prediction model constructed in the subsequent process includes temporal features.

[0163] In this embodiment, before obtaining the first spatiotemporal capacity fusion dataset at different times based on the collected historical network usage data of the first user, the capacity prediction method may further include:

[0164] Collect the first user's historical network usage data.

[0165] In other words, in this embodiment, historical network usage data of the first user is first collected, and then a first spatiotemporal capacity fusion dataset at different times is obtained based on the collected historical network usage data of the first user. The historical network usage data of the first user includes data from multiple platforms and involves different types; thus, the obtained first spatiotemporal capacity fusion dataset at different times is more comprehensive.

[0166] In this embodiment, when the first user's historical network usage data is MDT data, the main fields that need to be collected include, but are not limited to: timestamp, MME group ID, International Mobile Subscriber Identification Number (IMSI), International Mobile Equipment Identity (IMEI), User Experience (UE) unique identifier on the MME-side S1 interface (MME UE S1AP ID), longitude, latitude, uplink data traffic at the Packet Data Convergence Protocol (PDCP) layer, and downlink data traffic at the PDCP layer.

[0167] In this embodiment, when the first user's historical network usage data is collected using software-collected data, the main fields that need to be collected include, but are not limited to: MME UE S1AP ID, number of uplink service bytes, number of downlink service bytes, number of terabytes (TB) in uplink Quadrature Phase Shift Keying (QPSK) modulation, number of TB in downlink QPSK modulation, number of TB in uplink Quadrature Amplitude Modulation (QAM) modulation, number of TB in downlink 16QAM modulation, number of TB in uplink 64QAM modulation, number of TB in downlink 64QAM modulation, and number of concurrent online users.

[0168] In this embodiment, when the first user's historical network usage data collected is user XDR data, the main data stream collected is the S1-U / S11 / Gn interface data stream. The main fields to be collected include, but are not limited to: Owner Province, Owner City, IMSI, IMEI, Procedure Start Time, Procedure End Time, longitude, latitude, coordinate system, App Type, App Sub-type, App Content, Uplink Bytes (UL Data), Downlink Bytes (DL Data), etc.

[0169] In this embodiment, when the first user's historical network usage data is engineering parameter data, the main fields that need to be collected include, but are not limited to: province, city, region, scenario type, scenario name, base station name, cell name, E-UTRAN Cell Global Identifier (ECGI), site longitude, site latitude, equipment room location, cabinet number, frame number, slot number, etc.

[0170] In this embodiment, when the first user's historical network usage data is geographic data, the main fields to be collected include, but are not limited to: city, bus information (BusInfo), point of interest (POI) information, bus route information (BusLineResult), cycling route information (BikingRouteResult), driving route information (DrivingRouteResult), longitude, latitude, etc.

[0171] In this embodiment, if the first user's historical network usage data includes at least two of the following: minimized drive test data, soft-collected data, user XDR data, engineering parameter data, and geospatial data, obtaining the first spatiotemporal capacity fusion dataset at different times based on the collected first user's historical network usage data may include the following steps:

[0172] First, the abnormal data corresponding to each data point is cleaned to obtain the cleaned data.

[0173] Then, based on the key fields, the various data are fused to obtain the first spatiotemporal capacity fused dataset.

[0174] That is, in this embodiment, the first spatiotemporal capacity fusion dataset at different times is obtained based on the collected historical network usage data of the first user. Specifically, this can be achieved by cleaning the abnormal data corresponding to each data point to obtain the cleaned data, and then fusing the data based on key fields to obtain the first spatiotemporal capacity fusion dataset.

[0175] In this embodiment, abnormal data refers to abnormal data containing outliers, missing values, etc., which cannot be used as training data for the capacity prediction model.

[0176] In this embodiment, cleaning refers to filtering and removing abnormal data to obtain normal data. It is understood that since the first user's historical network usage data comes from different platforms, cleaning in this embodiment also refers to standardizing the naming format of fields with the same physical meaning across platforms, as well as standardizing the collection time granularity, thereby unifying the same fields and collection time granularity across different platforms to facilitate subsequent fusion.

[0177] In this embodiment, based on key fields, various data are fused to obtain a first spatiotemporal capacity fused dataset. Specifically, this can be achieved by associating the MDT dataset obtained through TimeStamp and MME UE S1AP ID with the software-collected dataset to form fused dataset 1; further, fused dataset 1 is associated with user XDR data through IMSI / IMEI to form fused dataset 2; and finally, fused dataset 2 is fused with engineering parameter data and geospatial data through latitude and longitude matching to obtain the final fused dataset. Since the final fused dataset is obtained from various first user historical network usage data, it includes timestamps, geographic information (latitude and longitude, point of interest information, scene tags, road information, etc.), wireless network capacity information (number of bytes of air interface service, number of users, application type, etc.), and coverage site information; therefore, in this embodiment, the final fused dataset is referred to as a spatiotemporal fused dataset.

[0178] In this embodiment, obtaining various types of node-edge associations between network nodes at different times based on the first spatiotemporal capacity fusion dataset may include the following steps:

[0179] First, from the first spatiotemporal capacity fusion dataset, we obtain various types of node and edge association data between network nodes at different times;

[0180] Then, the associated data is converted into association relationships A at different times. k (i,j); where A k (i,j)≠0 indicates that there is an edge of type k between node i and node j;

[0181] Furthermore, based on the relationship Ak (i,j) obtains adjacency matrices of different types at different times. The adjacency matrix is ​​obtained by integrating associations belonging to the same type. K represents the total number of types, k represents the type identifier, N represents the matrix order, and R represents the set of real numbers.

[0182] That is, in this embodiment, based on the first spatiotemporal capacity fusion dataset, the association relationships between various network nodes of different types at different times are obtained. Specifically, this can be achieved by obtaining the association data of various network nodes of different types at different times from the first spatiotemporal capacity fusion dataset, and then converting the association data into association relationships A at different times. k (i,j), and then based on the association relationship A k (i,j) obtains adjacency matrices of different types at different times. accomplish.

[0183] In this embodiment, the types include, but are not limited to, geographical proximity, topological proximity, scene correlation, and route connectivity.

[0184] The step of converting related data into relationships at different times can include the following scenarios:

[0185] Case 1: Convert geographically adjacent related data into a relationship A1(i,j) = 1; where A1(i,j) = 1 indicates that node i and node j are geographically adjacent.

[0186] Case 2: Convert the topologically adjacent related data into an association relationship A2(i,j) = 1; where A2(i,j) = 1 indicates that node i and node j are physically topologically adjacent;

[0187] Scenario 3: Convert the scene-related associated data into an association relationship A3(i,j) = EJ(s) i ,s j )∈[0,1]; where, A3(i,j)=EJ(s i ,s j )∈[0,1] indicates scene correlation between node i and node j; EJ(.) represents the Tanimoto operator, s i With s j Let i and j represent the scene feature vectors of node i and node j, respectively.

[0188] Case 4: Convert the associated data of the route connections into A4(i,j)=max{0,I(v i ,v j )-A1(i,j)}∈{0,1}; where, A4(i,j)=max{0,I(v i ,vj )-A1(i,j)}∈{0,1} indicates that nodes i and j are not geographically adjacent but have a connected path; I(.) represents the indicator operator, and v represents a node.

[0189] Among them, according to the association relationship A k (i,j) obtains adjacency matrices of different types at different times. The steps can be implemented by integrating the same type of relationships at the same time.

[0190] For example:

[0191] Please refer to Table 1 below, which shows the association data of various types of nodes and edges between network nodes at different times obtained from the first spatiotemporal capacity fusion dataset.

[0192] Table 1

[0193]

[0194] Please refer to Table 2 below, which shows how to convert related data into relationships A at different times. k (i,j).

[0195] Table 2

[0196]

[0197]

[0198] It is worth noting that only four types of relationships are listed in this embodiment. In practical applications, these can be flexibly adjusted according to specific application scenarios.

[0199] Please refer to Table 3 below for the four types of adjacency matrices between network nodes at different times.

[0200] Table 3

[0201]

[0202] It is understandable that in Table 3 above, for time 1, the adjacency matrix is ​​obtained by integrating the geographically adjacent A1 values. Integrating the topologically adjacent A2 values ​​yields the adjacency matrix. The adjacency matrix is ​​obtained by integrating the various A3s related to the scene. Integrate the A4s connected by the route to obtain the adjacency matrix. Similarly, the four types of adjacency matrices for time 2 will not be elaborated here.

[0203] Step S302: Construct a capacity prediction model based on the correlation; wherein, the capacity prediction model includes the spatial and temporal characteristics between each network node.

[0204] Accordingly, after obtaining the various types of node and edge associations between network nodes at different times based on the first spatiotemporal capacity fusion dataset, it is necessary to construct a capacity prediction model based on the associations.

[0205] In this embodiment, step S22, which constructs a capacity prediction model based on the correlation, may include:

[0206] Based on different types of adjacency matrices at different times Build a capacity prediction model.

[0207] In this embodiment, different types of adjacency matrices are used at different times. The steps to build a capacity prediction model may include the following:

[0208] First, consider the adjacency matrices of different types at the same time. The data is spliced ​​together to obtain the spatial characteristics between various network nodes at different times.

[0209] Then, the spatial characteristics between network nodes at different times are subjected to spatiotemporal weighted transformation to obtain the temporal characteristics between network nodes at different times.

[0210] Furthermore, the temporal characteristics of each network node are fused to construct a capacity prediction model.

[0211] That is, in this embodiment, different types of adjacency matrices are used at different times. Constructing a capacity prediction model can be achieved by using adjacency matrices of different types at the same time. The spatial features between network nodes at different times are spliced ​​together. Then, the spatial features between network nodes at different times are subjected to spatiotemporal weighted transformation to obtain the temporal features between network nodes at different times. Finally, the temporal features between network nodes are fused to construct a capacity prediction model.

[0212] For example, please refer to: Figure 4 As shown, this is an adjacency matrix based on different types of data at different times. A schematic diagram of the network structure for constructing the capacity prediction model is shown. The network structure includes a Spatial Transformer Networks (S-GTN) layer and a Temporal Weighted Correlation Modeling with Transformer Network (TWCTN) layer. Specifically, the S-GTN layer first obtains the spatial features between network nodes at different times. Then, the TWCTN layer obtains the temporal features between network nodes at different times. Finally, the temporal features between network nodes at different times are fused to obtain the prediction result, thus constructing the capacity prediction model.

[0213] In this embodiment, adjacency matrices of different types at the same time are... The steps of stitching together the data to obtain the spatial characteristics of network nodes at different times may include the following:

[0214] First, based on the adjacency matrix Using Formula 1, obtain the adjacency matrix of the corresponding path;

[0215] Formula 1 is:

[0216]

[0217] Where l represents the maximum number of steps (l) required to reach node j from node i, c represents the number of path types of different lengths, and λ (l) For A k (k = 1, ..., K) represents the weight of each type of edge at layer l, and o represents the dot product operation. express The degree matrix;

[0218] Then, based on the adjacency matrix of the corresponding path Using Formula 2, the spatial characteristics between each network node can be obtained;

[0219] Formula 2 is as follows:

[0220]

[0221] in, express The degree matrix, W represents the weight matrix, X represents the input node feature matrix, and σ(.) represents the sigmoid function; C Indicates the number of channels.

[0222] That is, in this embodiment, different types of adjacency matrices are used at the same time. The input is fed into the S-GTN layer to obtain the spatial features between network nodes at different times, that is, to complete the adjacency matrices of different types at the same time in the S-GTN layer. splicing.

[0223] For example, please refer to Table 4, which shows the concatenated matrix (i.e., spatial features) obtained by concatenating different types of adjacency matrices between network nodes at different times.

[0224] Table 4

[0225] time splicing matrix Time 1 H Time 2 H …… ……

[0226] In this embodiment, the step of performing spatiotemporal weighted transformation on the spatial characteristics of each network node at different times to obtain the temporal characteristics of each network node at different times may include the following steps:

[0227] First, based on the spatial characteristics between network nodes at different times, the pooling result is obtained using Formula 3;

[0228] Formula 3 is:

[0229]

[0230] Where t represents the current time, t' represents different historical times, T represents the set time, v represents a node, V represents the set of nodes, and f pool Represents the pooling function;

[0231] Then, based on the pooling result p (t ' ) Using Formula 4, we obtain the first weighted result;

[0232] Formula four is:

[0233] q=σ(W2δ(W1p)),p (t ' ) ∈p

[0234] Where W1 and W2 represent weight matrices, δ(.) represents the ReLU function, and p represents the set of different time points after pooling all nodes;

[0235] Then, based on the first weighted result q, the second weighted result at different historical moments is obtained using Formula 5;

[0236] Formula 5 is:

[0237]

[0238] in, This represents the second weighted result at different historical moments.

[0239] In other words, in this embodiment, the spatial features between each network node at different times are input to the TWCTN layer to obtain the temporal features between each network node at different times. That is, the spatiotemporal transformation of the spatial features between each network node at different times is completed in the S-GTN layer.

[0240] For example, please refer to Table 5, which shows the temporal characteristics between various network nodes at different times.

[0241] Table 5

[0242]

[0243] Furthermore, in this embodiment, the temporal characteristics of each network node are fused to construct a capacity prediction model; for example, please refer to [link to relevant documentation]. Figure 5 As shown, the temporal characteristics between various network nodes at different times are obtained. The graph is fused together using a Transformer layer into a single graph H. (t) Output H (t) This is the capacity prediction model that has been constructed.

[0244] Step S303: Train the capacity prediction model based on the first user's historical network usage data to obtain the target capacity prediction model.

[0245] Accordingly, in this embodiment, after constructing the capacity prediction model based on the correlation, it is necessary to train the capacity prediction model based on the historical network usage data of the first user; specifically, the capacity prediction model can be trained using the first spatiotemporal capacity fusion dataset at different times to obtain the target capacity prediction model.

[0246] In this embodiment, the obtained target capacity prediction model includes the spatial and temporal features between each network node; that is, the spatial and temporal dependencies between each network node are considered during the training of the capacity prediction model, which greatly improves the accuracy of the network capacity prediction model training.

[0247] Furthermore, based on the capacity prediction method described above, this embodiment also proposes a capacity prediction model training method, which can be found in [link to relevant documentation]. Figure 6 As shown, Figure 6 This is a flowchart illustrating the second embodiment of the capacity prediction model training method of the present invention. The capacity prediction model training method includes the following steps:

[0248] Step S601: Based on the collected historical network usage data of the first user, obtain the various types of node and edge associations between network nodes at different times;

[0249] Step S602: Construct a capacity prediction model based on the correlation; wherein, the capacity prediction model includes the spatial and temporal characteristics between each network node;

[0250] Step S603: Train the capacity prediction model based on the historical network usage data of the first user to obtain the target capacity prediction model.

[0251] The capacity prediction method of the present invention adopts all the technical solutions of the second embodiment described above, and therefore has at least all the beneficial effects brought about by the technical solutions of the second embodiment described above, which will not be repeated here.

[0252] In addition, please see Figure 7 As shown, based on the above-described capacity prediction method, this embodiment of the invention also proposes a capacity prediction device, which includes:

[0253] The second acquisition module 701 is used to obtain a second spatiotemporal capacity fusion dataset at different times based on the collected historical network usage data of the second user.

[0254] The third acquisition module 702 is used to input the second spatiotemporal capacity fusion dataset into the target capacity prediction model to obtain the capacity prediction results of each network node at a preset time; wherein, the target capacity prediction model includes the spatial and temporal characteristics between each network node.

[0255] The capacity prediction device of the present invention adopts all the technical solutions of the first / second embodiments of the above-described capacity prediction method, and therefore has at least all the beneficial effects brought about by the technical solutions of the first / second embodiments of the above-described capacity prediction method, which will not be described in detail here.

[0256] In addition, please see Figure 8 As shown, based on the above-described capacity prediction model training method, this embodiment of the invention also proposes a capacity prediction model training device, which includes:

[0257] The first acquisition module 801 is used to obtain various types of node and edge association relationships between network nodes at different times based on the collected historical network usage data of the first user.

[0258] Module 802 is used to construct a capacity prediction model based on the correlation relationship; wherein, the capacity prediction model includes the spatial and temporal characteristics between each network node;

[0259] Training module 803 is used to train the capacity prediction model based on the first user's historical network usage data to obtain the target capacity prediction model.

[0260] The capacity prediction model training device of the present invention adopts all the technical solutions of the first embodiment of the capacity prediction model training method described above, and therefore has at least all the beneficial effects brought about by the technical solutions of the first embodiment of the capacity prediction model training method described above, which will not be described in detail here.

[0261] In addition, this embodiment also proposes a computer program that stores a capacity prediction program, which, when executed by a processor, implements the capacity prediction method as described above; or a computer program that stores a capacity prediction model training program, which, when executed by a processor, implements the steps of the capacity prediction model training method as described above.

[0262] The computer program includes volatile or non-volatile, removable or non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, computer program modules, or other data). The computer program includes, but is not limited to, RAM (Random Access Memory), ROM (Read-Only Memory), EEPROM (Electrically Erasable Programmable Read-Only Memory), flash memory or other memory technologies, CD-ROM (Compact Disc Read-Only Memory), DVD or other optical disc storage, cartridges, magnetic tapes, disk storage, or any other medium that can be used to store desired information and is accessible to a computer.

[0263] Therefore, those skilled in the art should understand that all or some of the steps, systems, and functional modules / units in the methods disclosed above can be implemented as software, firmware, hardware, or suitable combinations thereof. In hardware implementations, the division between functional modules / units mentioned in the above description does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed collaboratively by several physical components. Some or all physical components may be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as integrated circuits, such as application-specific integrated circuits (ASICs).

[0264] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A capacity prediction method, characterized in that, The capacity prediction method includes the following steps: Based on the collected historical network usage data of the first user, the first spatiotemporal capacity fusion dataset at different times is obtained; Based on the first spatiotemporal capacity fusion dataset, various types of node-edge associations between network nodes at different times are obtained; the various types include geographical adjacency, topological adjacency, scene correlation, and route connectivity; Based on the aforementioned relationships, a capacity prediction model is constructed; wherein, the capacity prediction model includes the spatial and temporal characteristics between each network node; Based on the first user's historical network usage data, the capacity prediction model is trained to obtain the target capacity prediction model; Based on the collected historical network usage data of the second user, a second spatiotemporal capacity fusion dataset at different times is obtained; The second spatiotemporal capacity fusion dataset is input into the target capacity prediction model to obtain the capacity prediction results of each network node at a preset time; wherein, the target capacity prediction model includes the spatial and temporal characteristics between each network node; The step of obtaining various types of node-edge associations between network nodes at different times based on the first spatiotemporal capacity fusion dataset includes: From the first spatiotemporal capacity fusion dataset, obtain various types of node and edge association data between network nodes at different times; Convert the associated data into association relationships. ;in, This indicates that there is an edge of type k between node i and node j; Based on the aforementioned relationship Obtain adjacency matrices of different types at different times. Wherein, the adjacency matrix is ​​obtained by integrating associations belonging to the same type, K represents the total number of types, k represents the type identifier, N represents the matrix order, and R represents the set of real numbers; The step of constructing a capacity prediction model based on the correlation includes: Adjacency matrices of different types at the same time The data is spliced ​​together to obtain the spatial characteristics between various network nodes at different times. The spatial characteristics of each network node at different times are subjected to spatiotemporal weighted transformation to obtain the temporal characteristics of each network node at different times. The temporal characteristics of each network node are fused to construct a capacity prediction model.

2. The capacity prediction method as described in claim 1, characterized in that, Before the step of obtaining the first spatiotemporal capacity fusion dataset at different times based on the collected historical network usage data of the first user, the capacity prediction method further includes: Collect the first user's historical network usage data; wherein, the first user's historical network usage data includes at least one of the following: minimized drive test data, software-collected data, user detection and response platform data, engineering parameter data, and geolocation data.

3. The capacity prediction method as described in claim 2, characterized in that, The step of obtaining the first spatiotemporal capacity fusion dataset at different times based on the collected historical network usage data of the first user includes: If the first user's historical network usage data includes at least two of the following: minimized drive test data, software-collected data, user detection and response platform data, engineering parameter data, and geographic data, then the abnormal data corresponding to each data will be cleaned to obtain the cleaned data. Based on the key fields, the various data are fused to obtain the first spatiotemporal capacity fusion dataset.

4. The capacity prediction method as described in claim 1, characterized in that, The process of converting the associated data into association relationships. The steps include: Transform geographically adjacent related data into relationships. ;in, This indicates that node i and node j are geographically adjacent; Transform topologically adjacent related data into relationships. ;in, This indicates that node i and node j are physically topologically adjacent; Transform scene-related data into relationships. ;in, This indicates that there is a scene-related relationship between node i and node j; Represents the Tanimoto operator. and Let i and j represent the scene feature vectors of node i and node j, respectively. Convert the associated data of route connections into ;in, This indicates that node i and node j are not geographically adjacent but have a connected path; This indicates the indicator operator. Represents a node.

5. The capacity prediction method as described in claim 1, characterized in that, The adjacency matrix of different types at the same time The steps for stitching together the data to obtain the spatial characteristics of each network node at different times include: According to the adjacency matrix Using Formula 1, the adjacency matrix of the corresponding path can be obtained; Formula 1 is: in, Indicates the maximum number of times node i has been visited. Node j can be reached in one step. This indicates different path types based on length. for Representing the different types of edges in The weights corresponding to the layers, This represents the dot product operation. , express The degree matrix; Based on the adjacency matrix of the corresponding path Using Formula 2, the spatial characteristics between the network nodes are obtained; Formula 2 is as follows: in, , express The degree matrix, , Represents the weight matrix. This represents the input node feature matrix. Represents the sigmoid function; Indicates the number of channels.

6. The capacity prediction method as described in claim 1, characterized in that, The step of performing spatiotemporal weighted transformation on the spatial characteristics of each network node at different times to obtain the temporal characteristics of each network node at different times includes: Based on the spatial characteristics between network nodes at different times, the pooling result is obtained using Formula 3; Formula 3 is as follows: in, Indicates the current moment. Representing different moments in history, Indicates the set time. Represents a node. Represents a set of nodes. Represents the pooling function; Based on the pooling results Using Formula 4, we obtain the first weighted result; Formula four is as follows: in, and Represents the weight matrix. express function, This represents the set of different times after all nodes have been pooled. Based on the first weighted result Using Formula 5, we obtain the second weighted result for different historical moments; Formula five is: in, This represents the second weighted result at different historical moments.

7. A method for training a capacity prediction model, characterized in that, The capacity prediction model training method includes the following steps: Based on the collected historical network usage data of the first user, the first spatiotemporal capacity fusion dataset at different times is obtained; Based on the first spatiotemporal capacity fusion dataset, various types of node-edge associations between network nodes at different times are obtained; the various types include geographical adjacency, topological adjacency, scene correlation, and route connectivity; Based on the aforementioned relationships, a capacity prediction model is constructed; wherein, the capacity prediction model includes the spatial and temporal characteristics between each network node; Based on the first user's historical network usage data, the capacity prediction model is trained to obtain the target capacity prediction model; The step of obtaining various types of node-edge associations between network nodes at different times based on the first spatiotemporal capacity fusion dataset includes: From the first spatiotemporal capacity fusion dataset, obtain various types of node and edge association data between network nodes at different times; Convert the associated data into association relationships. ;in, This indicates that there is an edge of type k between node i and node j; Based on the aforementioned relationship Obtain adjacency matrices of different types at different times. Wherein, the adjacency matrix is ​​obtained by integrating associations belonging to the same type, K represents the total number of types, k represents the type identifier, N represents the matrix order, and R represents the set of real numbers; The step of constructing a capacity prediction model based on the correlation includes: Adjacency matrices of different types at the same time The data is spliced ​​together to obtain the spatial characteristics between various network nodes at different times. The spatial characteristics of each network node at different times are subjected to spatiotemporal weighted transformation to obtain the temporal characteristics of each network node at different times. The temporal characteristics of each network node are fused to construct a capacity prediction model.

8. The capacity prediction model training method as described in claim 7, characterized in that, Before the step of obtaining the first spatiotemporal capacity fusion dataset at different times based on the collected historical network usage data of the first user, the capacity prediction model training method further includes: Collect the first user's historical network usage data; wherein, the first user's historical network usage data includes at least one of the following: minimized drive test data, software-collected data, user detection and response platform data, engineering parameter data, and geolocation data.

9. The capacity prediction model training method as described in claim 8, characterized in that, The step of obtaining the first spatiotemporal capacity fusion dataset at different times based on the collected historical network usage data of the first user includes: If the first user's historical network usage data includes at least two of the following: minimized drive test data, software-collected data, user detection and response platform data, engineering parameter data, and geographic data, then the abnormal data corresponding to each data will be cleaned to obtain the cleaned data. Based on the key fields, the various data are fused to obtain the first spatiotemporal capacity fusion dataset.

10. The capacity prediction model training method as described in claim 7, characterized in that, The process of converting the associated data into association relationships. The steps include: Transform geographically adjacent related data into relationships. ;in, This indicates that node i and node j are geographically adjacent; Transform topologically adjacent related data into relationships. ;in, This indicates that node i and node j are physically topologically adjacent; Transform scene-related data into relationships. ;in, This indicates that there is a scene-related relationship between node i and node j; Represents the Tanimoto operator. and Let i and j represent the scene feature vectors of node i and node j, respectively. Convert the associated data of route connections into ;in, This indicates that node i and node j are not geographically adjacent but have a connected path; This indicates the indicator operator. Represents a node.

11. The capacity prediction model training method as described in claim 7, characterized in that, The adjacency matrix of different types at the same time The steps for stitching together the data to obtain the spatial characteristics of each network node at different times include: According to the adjacency matrix Using Formula 1, the adjacency matrix of the corresponding path can be obtained; Formula 1 is: in, Indicates the maximum number of times node i has been visited. Node j can be reached in one step. This indicates different path types based on length. for Representing the different types of edges in The weights corresponding to the layers, This represents the dot product operation. , express The degree matrix; Based on the adjacency matrix of the corresponding path Using Formula 2, the spatial characteristics between the network nodes are obtained; Formula 2 is as follows: in, , express The degree matrix, , Represents the weight matrix. This represents the input node feature matrix. Represents the sigmoid function; Indicates the number of channels.

12. The capacity prediction model training method as described in claim 7, characterized in that, The step of performing spatiotemporal weighted transformation on the spatial characteristics of each network node at different times to obtain the temporal characteristics of each network node at different times includes: Based on the spatial characteristics between network nodes at different times, the pooling result is obtained using Formula 3; Formula 3 is as follows: in, Indicates the current moment. Representing different moments in history, Indicates the set time. Represents a node. Represents a set of nodes. Represents the pooling function; Based on the pooling results Using Formula 4, we obtain the first weighted result; Formula four is as follows: in, and Represents the weight matrix. express function, This represents the set of different times after all nodes have been pooled. Based on the first weighted result Using Formula 5, we obtain the second weighted result for different historical moments; Formula five is: in, This represents the second weighted result at different historical moments.

13. A capacity prediction device, characterized in that, The capacity prediction device includes: The second acquisition module is used to obtain the second spatiotemporal capacity fusion dataset at different times based on the collected historical network usage data of the second user. The third acquisition module is used to input the second spatiotemporal capacity fusion dataset into the target capacity prediction model to obtain the capacity prediction results of each network node at a preset time; wherein, the target capacity prediction model includes the spatial and temporal characteristics between each network node; The capacity prediction device further includes: The first acquisition module is used to obtain a first spatiotemporal capacity fusion dataset at different times based on the collected historical network usage data of the first user; and to obtain various types of node-edge association relationships between network nodes at different times based on the first spatiotemporal capacity fusion dataset; the various types include geographical adjacency, topological adjacency, scene correlation, and route connectivity. A construction module is used to construct a capacity prediction model based on the aforementioned relationships; wherein the capacity prediction model includes spatial and temporal characteristics between various network nodes; The training module is used to train the capacity prediction model based on the first user's historical network usage data to obtain the target capacity prediction model. The step of obtaining various types of node-edge associations between network nodes at different times based on the first spatiotemporal capacity fusion dataset includes: From the first spatiotemporal capacity fusion dataset, obtain various types of node and edge association data between network nodes at different times; Convert the associated data into association relationships. ;in, This indicates that there is an edge of type k between node i and node j; Based on the aforementioned relationship Obtain adjacency matrices of different types at different times. Wherein, the adjacency matrix is ​​obtained by integrating associations belonging to the same type, K represents the total number of types, k represents the type identifier, N represents the matrix order, and R represents the set of real numbers; The step of constructing a capacity prediction model based on the correlation includes: Adjacency matrices of different types at the same time The data is spliced ​​together to obtain the spatial characteristics between various network nodes at different times. The spatial characteristics of each network node at different times are subjected to spatiotemporal weighted transformation to obtain the temporal characteristics of each network node at different times. The temporal characteristics of each network node are fused to construct a capacity prediction model.

14. A training device for a capacity prediction model, characterized in that, The training device for the capacity prediction model includes: The first acquisition module is used to obtain a first spatiotemporal capacity fusion dataset at different times based on the collected historical network usage data of the first user; and to obtain various types of node-edge association relationships between network nodes at different times based on the first spatiotemporal capacity fusion dataset; the various types include geographical adjacency, topological adjacency, scene correlation, and route connectivity. A construction module is used to construct a capacity prediction model based on the aforementioned relationships; wherein the capacity prediction model includes spatial and temporal characteristics between various network nodes; The training module is used to train the capacity prediction model based on the first user's historical network usage data to obtain the target capacity prediction model. The step of obtaining various types of node-edge associations between network nodes at different times based on the first spatiotemporal capacity fusion dataset includes: From the first spatiotemporal capacity fusion dataset, obtain various types of node and edge association data between network nodes at different times; Convert the associated data into association relationships. ;in, This indicates that there is an edge of type k between node i and node j; Based on the aforementioned relationship Obtain adjacency matrices of different types at different times. Wherein, the adjacency matrix is ​​obtained by integrating associations belonging to the same type, K represents the total number of types, k represents the type identifier, N represents the matrix order, and R represents the set of real numbers; The step of constructing a capacity prediction model based on the correlation includes: Adjacency matrices of different types at the same time The data is spliced ​​together to obtain the spatial characteristics between various network nodes at different times. The spatial characteristics of each network node at different times are subjected to spatiotemporal weighted transformation to obtain the temporal characteristics of each network node at different times. The temporal characteristics of each network node are fused to construct a capacity prediction model.

15. A capacity prediction device, characterized in that, The capacity prediction device includes: a memory, a processor, and a capacity prediction program stored in the memory and running on the processor, wherein the capacity prediction program, when executed by the processor, implements the steps of the capacity prediction method as described in any one of claims 1-6.

16. A training device for a capacity prediction model, characterized in that, The training device for the capacity prediction model includes: a memory, a processor, and a training program for the capacity prediction model stored in the memory and running on the processor. When the training program for the capacity prediction model is executed by the processor, it implements the steps of the capacity prediction model training method as described in any one of claims 7-12.

17. A computer program product, characterized in that, The computer program product stores a capacity prediction program, which, when executed by a processor, implements the steps of the capacity prediction method as described in any one of claims 1-6; or the computer program product stores a capacity prediction model training program, which, when executed by a processor, implements the steps of the capacity prediction model training method as described in any one of claims 7-12.

Images