Wireless cell multi-index space-time sequence prediction method and system

Abstract

The invention relates to a wireless cell multi-index space-time sequence prediction method and system, and the method comprises the steps: carrying out the preprocessing of the multi-index historical data of a wireless cell in a target region, and constructing a multi-index dynamic graph; constructing an adaptive mixed space-time graph neural network model according to the multi-index dynamic graph, wherein the adaptive mixed space-time graph neural network model comprises a plurality of adaptive mixed space-time learning blocks consisting of a time convolution module, an adaptive mixed graph learning module and a space-time adaptive module; dividing the preprocessed multi-index historical data into a training set, a verification set and a test set, and inputting an adaptive mixed space-time graphic neural network model for training after multi-cycle data division; and inputting wireless cell multi-index data collected in real time into the trained adaptive mixed space-time graphic neural network model to obtain a wireless cell multi-index predicted value. The prediction deviation caused by single index modeling or a traditional method can be reduced, and the accuracy of a prediction result is improved.

Description

Background Art

[0002] In a wireless communication network, spatio-temporal sequence prediction is a core technology for optimizing network resource scheduling and improving service quality. As a technology for processing time-varying data, spatio-temporal sequence prediction has been widely applied in multiple fields, such as traffic flow, weather forecasting, etc. The operation data of wireless cells, such as traffic volume, number of users, signal interference, etc., has significant spatio-temporal coupling characteristics: in the time dimension, the data shows periodicity and sudden fluctuations; in the space dimension, the indicators of adjacent cells have strong correlations due to user mobility, signal coverage overlap, etc.

[0003] In the prior art, wireless communication spatio-temporal sequence prediction methods usually only model a single indicator, such as only modeling traffic data, and use traditional time series models such as ARIMA, SARIMA or single-indicator deep learning models such as LSTM, Transformer in the modeling to achieve spatio-temporal sequence prediction; this kind of spatio-temporal sequence prediction method does not consider the dynamic coupling effect between multiple indicators, such as the increase in the number of users directly leads to an increase in traffic and an increase in interference, resulting in a large deviation between the prediction result and the actual network state; moreover, traditional time series models such as ARIMA, SARIMA rely on artificially designed spatio-temporal features, manually set time windows, and construct an adjacency matrix based on fixed rules, and then input into a machine learning model after artificially designed feature splicing. Since the interaction relationships such as user behavior, service type, signal quality, etc. of wireless data show high non-linearity, the simple features designed manually cannot cover complex patterns and rely on prior knowledge, resulting in the lack of high-dimensional non-linear features and difficulty in adapting to the network structure changes in different regions and different time periods; at the same time, traditional models assume that all cells have the same spatio-temporal feature distribution, adopt a unified model parameter and feature processing method, thereby introducing a large amount of spatial noise and reducing the prediction accuracy.

[0004] Therefore, it is necessary to improve one or more problems existing in the above related technical solutions.

[0005] It should be noted that the information disclosed in the above Background Art section is only used to enhance the understanding of the background of the present disclosure, and thus may include information that does not constitute the prior art known to those of ordinary skill in the art. Summary of the Invention

[0006] The purpose of the embodiments of the present disclosure is to provide a method and system for spatio-temporal sequence prediction of multiple indicators in a wireless cell, so as to at least to some extent overcome one or more problems caused by the limitations and defects of the related technologies.

[0007] In the first aspect, the present application provides a method for spatio-temporal sequence prediction of multiple indicators in a wireless cell, including:

[0008] Preprocess the multi - metric historical data of wireless cells in the target area and construct a multi - metric dynamic graph;

[0009] Construct an adaptive hybrid spatio - temporal graph neural network model according to the multi - metric dynamic graph. The adaptive hybrid spatio - temporal graph neural network model includes multiple adaptive hybrid spatio - temporal learning blocks composed of a temporal convolution module, an adaptive hybrid graph learning module, and a spatio - temporal adaptive module;

[0010] Divide the preprocessed multi - metric historical data into a training set, a validation set, and a test set. After multi - cycle data division, input it into the adaptive hybrid spatio - temporal graph neural network model for training;

[0011] Input the real - time collected multi - metric data of wireless cells into the trained adaptive hybrid spatio - temporal graph neural network model to obtain the multi - metric prediction values of wireless cells.

[0012] In a possible implementation manner, the step of preprocessing the multi - metric historical data of wireless cells in the target area and constructing a multi - metric dynamic graph includes:

[0013] Obtain the multi - metric historical data of wireless cells in the target area. The multi - metric historical data includes traffic data and user number data;

[0014] Clean and normalize the multi - metric historical data;

[0015] Take each wireless cell as a graph node and construct a multi - metric dynamic graph according to the geographical adjacency relationship or communication association relationship between graph nodes.

[0016] In a possible implementation manner, the multi - metric dynamic graph is \(G=(V, E, A, X)\), where \(V\) is the set of base station nodes, \(E\) is the set of physical connection edges, \(A\) is the index - aware adjacency tensor, and \(X\) is the multi - metric feature tensor.

[0017] In a possible implementation manner, the step of constructing an adaptive hybrid spatio - temporal graph neural network model according to the multi - metric dynamic graph, where the adaptive hybrid spatio - temporal graph neural network model includes multiple adaptive hybrid spatio - temporal learning blocks composed of a temporal convolution module, an adaptive hybrid graph learning module, and a spatio - temporal adaptive module, includes:

[0018] Construct a temporal convolution module. The temporal convolution module is used to receive multi - cycle time data, extract the time - dimension features of each type of index through a dual - path temporal convolution network, and splice all the time - dimension features of the indexes;

[0019] Construct an adaptive hybrid graph learning module, which is used to receive multi-metric dynamic graphs and obtain spatial dimension features for each type of metric through an independent graph attention network and cross-metric gating;

[0020] Construct a spatio-temporal adaptive module, which is used to receive the time dimension features and spatial dimension features input in parallel, superimpose the time dimension features and spatial dimension features, and generate spatio-temporal gating weights through a multi-layer perceptron to output the spatio-temporal joint features after adaptive fusion;

[0021] Connect the outputs of all adaptive hybrid spatio-temporal learning blocks to the output layer through skip connections, fuse the spatio-temporal joint features output by each adaptive hybrid spatio-temporal learning block, and obtain the multi-metric prediction result through the first formula.

[0022] In a possible implementation manner, the expression of the time convolution module is

[0023] where, H time is the time feature, is the time feature of m types of metrics, X (m) is the feature tensor of m types of metrics;

[0024] The expression of the adaptive hybrid graph learning module is

[0025] where, α m = Softmax(W m ·X (m) ), H space is the spatial dimension feature, α m is the learnable metric importance weight, is the spatial dimension feature of m types of metrics, A (m) is the m-type metric perception adjacency tensor, W m is the parameter matrix;

[0026] The expression of the spatio-temporal adaptive module is

[0027] where, is the spatio-temporal joint feature;

[0028] The first formula is

[0029] where, is the multi-metric prediction result, H out is the skip connection fusion result, W f1 、b f1 、W f2 、bf2 is a learnable parameter, K is the number of adaptive hybrid spatio-temporal learning blocks, is the spatio-temporal joint feature output by the i-th adaptive hybrid spatio-temporal learning block.

[0030] In a possible implementation, the step of dividing the preprocessed multi-index historical data into a training set, a validation set, and a test set, and after performing multi-period data division, inputting the data into the adaptive hybrid spatio-temporal graph neural network model for training includes:

[0031] Dividing the preprocessed multi-index historical data into a training set, a validation set, and a test set at a ratio of 7:1.5:1.5;

[0032] Dividing the divided multi-index historical data in chronological order to obtain multi-period data, where the multi-period data includes recent data, daily data, and weekly data;

[0033] Using the Adam optimizer and the mean absolute error loss function, training the adaptive hybrid spatio-temporal graph neural network model with multi-period data, and obtaining the trained adaptive hybrid spatio-temporal graph neural network model when the loss value on the validation set does not decrease within multiple rounds.

[0034] In a possible implementation, the mean absolute error loss function is

[0035] where L(θ) is the loss function, M is the number of metrics, Y i is the true value of the i-th sample, is the predicted value of the i-th sample.

[0036] In a possible implementation, the step of inputting the real-time collected multi-index data of the wireless cell into the trained adaptive hybrid spatio-temporal graph neural network model to obtain the predicted values of the multi-index of the wireless cell includes:

[0037] Preprocessing the real-time collected multi-index data of the wireless cell and performing multi-period data division to obtain real-time multi-period data;

[0038] Inputting the real-time multi-period data into the trained adaptive hybrid spatio-temporal graph neural network model to obtain the predicted values of the multi-index of the wireless cell.

[0039] In a possible implementation, the real-time multi-period data trains the model through the Adam optimizer, and the learning rate is 0.001.

[0040] Second aspect, the present application provides a multi - metric spatio - temporal sequence prediction system for a wireless cell. The system is used to execute the above - mentioned multi - metric spatio - temporal sequence prediction method for a wireless cell. The system includes:

[0041] A data processing module, configured to pre - process the multi - metric historical data of wireless cells in a target area and construct a multi - metric dynamic graph;

[0042] A model construction module, configured to construct an adaptive hybrid spatio - temporal graph neural network model according to the multi - metric dynamic graph. The adaptive hybrid spatio - temporal graph neural network model includes multiple adaptive hybrid spatio - temporal learning blocks composed of a temporal convolution module, an adaptive hybrid graph learning module, and a spatio - temporal adaptation module;

[0043] A model training module, configured to divide the pre - processed multi - metric historical data into a training set, a validation set, and a test set, and after performing multi - cycle data division, input the data into the adaptive hybrid spatio - temporal graph neural network model for training;

[0044] A model prediction module, configured to input the multi - metric data of the wireless cell collected in real - time into the trained adaptive hybrid spatio - temporal graph neural network model to obtain the multi - metric prediction values of the wireless cell.

[0045] The technical solution provided by the present application may include the following beneficial effects:

[0046] Through the multi - metric spatio - temporal sequence prediction method and system for a wireless cell of the present application, the dynamic coupling effect between metrics can be obtained through multi - metric joint modeling, and through multi - dimensional spatio - temporal feature fusion, the prediction deviation caused by single - metric modeling or traditional methods can be reduced, and the accuracy of the prediction result can be improved;

[0047] Moreover, by avoiding manual feature design, through dynamic hybrid graph learning, the spatial dependence differences of different metrics are distinguished, and at the same time, the spatial relationships of static adjacency and adaptive attention between cells are obtained. High - dimensional non - linear spatio - temporal features are automatically extracted by the neural network to adapt to the network structure changes in different regions and different time periods, and the model's representation ability for complex data patterns is enhanced;

[0048] At the same time, the spatio - temporal adaptation module eliminates the interference of heterogeneity such as device type and user density between cells on the prediction through a node - level gating mechanism, and reduces the spatial noise.

[0049] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] The accompanying drawings here are incorporated into the specification and form a part of this specification, showing embodiments consistent with the present disclosure, and are used together with the specification to explain the principles of the present disclosure. Obviously, the accompanying drawings in the following description are only some embodiments of the present disclosure, and for those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0051] Figure 1 A flowchart showing the method for predicting multi - metric spatio - temporal sequences of a wireless cell in an exemplary embodiment of the present disclosure;

[0052] Figure 2 A detailed flowchart showing step S100 of the method for predicting multi - metric spatio - temporal sequences of a wireless cell in an exemplary embodiment of the present disclosure;

[0053] Figure 3 A detailed flowchart showing step S200 of the method for predicting multi - metric spatio - temporal sequences of a wireless cell in an exemplary embodiment of the present disclosure;

[0054] Figure 4 A detailed flowchart showing step S300 of the method for predicting multi - metric spatio - temporal sequences of a wireless cell in an exemplary embodiment of the present disclosure;

[0055] Figure 5 A detailed flowchart showing step S400 of the method for predicting multi - metric spatio - temporal sequences of a wireless cell in an exemplary embodiment of the present disclosure;

[0056] Figure 6 A schematic diagram showing the node - limited transmission communication model of the method for predicting multi - metric spatio - temporal sequences of a wireless cell in an exemplary embodiment of the present disclosure;

[0057] Figure 7 A schematic diagram showing the structure of the system for predicting multi - metric spatio - temporal sequences of a wireless cell in an exemplary embodiment of the present disclosure. Detailed implementation manners

[0058] Now, example embodiments will be described more fully with reference to the accompanying drawings. However, the example embodiments can be implemented in various forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that this disclosure will be more thorough and complete, and will fully convey the concept of the example embodiments to those skilled in the art. The described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments.

[0059] In addition, the accompanying drawings are only schematic illustrations of the present disclosure and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and thus the repeated descriptions thereof will be omitted. Some of the block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities may be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.

[0060] In this exemplary embodiment, a method for predicting multi-metric spatio-temporal sequences of a wireless cell is first provided. This method can be applied to a terminal device, such as a mobile terminal like a mobile phone, a desktop computer, a personal digital assistant, a laptop computer, a tablet computer, a smart watch, etc. Referring to Figure 1 as shown, this method may include the following steps:

[0061] Step S100: Preprocess the multi-metric historical data of the wireless cell in the target area and construct a multi-metric dynamic graph.

[0062] Step S200: Construct an adaptive hybrid spatio-temporal graph neural network model according to the multi-metric dynamic graph. The adaptive hybrid spatio-temporal graph neural network model includes a plurality of adaptive hybrid spatio-temporal learning blocks composed of a temporal convolution module, an adaptive hybrid graph learning module, and a spatio-temporal adaptation module.

[0063] Step S300: Divide the preprocessed multi-metric historical data into a training set, a validation set, and a test set. After performing multi-period data division, input them into the adaptive hybrid spatio-temporal graph neural network model for training.

[0064] Step S400: Input the real-time collected multi-metric data of the wireless cell into the trained adaptive hybrid spatio-temporal graph neural network model to obtain the multi-metric prediction values of the wireless cell.

[0065] The above method can, by constructing an adaptive hybrid spatio-temporal graph neural network model, as Figure 6 shown, utilize the input of multi-period data of X R , X D , X W , combine the static and dynamic graph learning mechanisms, and introduce a spatio-temporal adaptation module to accurately predict key metrics such as the traffic and the number of users of the wireless cell, providing strong support for the resource optimization configuration and management decision-making of the wireless communication network.

[0066] Next, each step of the above method in this exemplary embodiment will be described in more detail with reference to Figures 1 to 6 the following.

[0067] In step S100, preprocess the multi - metric historical data of wireless cells in the target area and construct a multi - metric dynamic graph.

[0068] It can be understood that in constructing the multi - metric dynamic graph, the spatial relationship of wireless cells is transformed into a computable mathematical expression through a graph structure, providing a basic framework for subsequent spatio - temporal feature decoupling (separating time periods and spatial dependencies), multi - metric joint modeling (differentiated spatial weights), and heterogeneity adaptation (dynamic gating regulation). Its essence is to combine the physical characteristics of wireless communication networks with data - driven model learning, solve the problems of rough spatial dependence modeling and insufficient multi - metric coupling processing in traditional methods, and ultimately improve the accuracy and robustness of multi - metric prediction.

[0069] In one embodiment, as Figure 2 shown, step S100 may include the following sub - steps:

[0070] In step S110, obtain the multi - metric historical data of wireless cells in the target area, where the multi - metric historical data includes traffic data and user number data.

[0071] It should be noted that, optionally, the multi - metric historical data can be obtained by extracting the historical data of user number data and cell traffic data through a network management system, base station device logs, or a third - party data collection platform.

[0072] In step S120, clean and normalize the multi - metric historical data.

[0073] It should be noted that, optionally, the obtained raw data may have problems such as noise, missing values, and outliers. For missing values, methods such as mean filling and interpolation can be used for processing; for outliers, statistical methods such as the 3σ principle or machine - learning - based anomaly detection algorithms such as Isolation Forest can be used to identify and process them, either correcting the outliers or directly removing them.

[0074] In step S130, take each wireless cell as a graph node and construct a multi - metric dynamic graph according to the geographical adjacency relationship or communication association relationship between the graph nodes.

[0075] It should be noted that the multi - metric dynamic graph is \(G=(V, E, A, X)\), where \(V\) is the set of base - station nodes, \(E\) is the set of physical connection edges, \(A\) is the metric - aware adjacency tensor, and \(X\) is the multi - metric feature tensor, where \(A\in\mathbb{R}^{M\times N\times N}\), \(X\in\mathbb{R}^{M\times N\times P}\), \(M\) is the number of metric categories, \(N\) is the number of cells, and \(P\) is the number of historical time steps. Each sub - matrix \(A_{m}\) is generated through adaptive graph learning and represents the spatial correlation intensity of the \(m\) - th type of metric between base stations. N×N×M 、\(X\in\mathbb{R}^{M\times N\times P}\) N×M×P , \(M\) is the number of metric categories, \(N\) is the number of cells, \(P\) is the number of historical time steps. Each sub - matrix \(A_{m}\) (m) is generated through adaptive graph learning and represents the spatial correlation intensity of the \(m\) - th type of metric between base stations.

[0076] It is understandable that in determining the physical connection edges: according to the physical locations of the cell base stations, it is judged whether they are adjacent (for example, if the Euclidean distance is less than the threshold, it is considered geographically adjacent, and the threshold is 500 meters). If they are adjacent, an edge is connected to depict the impact of spatial proximity on the indicators (such as the traffic linkage caused by user movement); according to the actual connections of the wireless communication network (such as signal interference between cells, user handover records), the communication association strength is determined to depict the impact of communication logic on the indicators (such as the change in the user number distribution caused by interference).

[0077] The multi-index adjacency matrix is a matrix set that describes the spatial association strength between different cells for each index. In constructing the multi-index adjacency matrix: for each index (such as traffic, user number), an adjacency matrix (or adjacency tensor) is constructed respectively. The element value in the matrix represents the spatial association strength between the corresponding cells (such as 1 representing adjacent / associated, or a quantization value such as signal strength representing the weight). Finally, a multi-index adjacency matrix set is formed, and each matrix corresponds to the spatial dependence relationship of a class of indicators.

[0078] The preprocessed data (such as the historical index sequences of each cell) is used as the features of the nodes in the graph, and the adjacency matrix is used as the edge structure of the graph, jointly constituting the input of the model.

[0079] In step S200, an adaptive hybrid spatio-temporal graph neural network model is constructed according to the multi-index dynamic graph. The adaptive hybrid spatio-temporal graph neural network model includes multiple adaptive hybrid spatio-temporal learning blocks composed of a time convolution module, an adaptive hybrid graph learning module, and a spatio-temporal adaptation module.

[0080] It should be noted that optionally, a multi-index prediction objective function is established, and based on the historical n-day observation data, the multi-index sequence for the next T days is predicted:

[0081] Y t+1:t+T =[X t+1 ,…,X t+T =f θ (G;(X t-n ,…,X t-1 ,X t )),Y∈R N×M×T ;

[0082] where f θ is the mapping function of the adaptive hybrid spatio-temporal graph neural network model, which needs to satisfy cross-index spatial coupling modeling: generating different spatial attention weights for different indicators; asynchronous time series decomposition: separating the long-term trends, periodicity, and sudden fluctuation characteristics of each indicator; heterogeneous feature normalization: eliminating the impact of dimension differences on joint prediction through a dynamic gating mechanism.

[0083] In one embodiment, as Figure 3 shown, step S200 may include the following sub-steps:

[0084] In step S210, a temporal convolutional module is constructed. The temporal convolutional module is used to receive multi-period time data, extract the time-dimensional features of each type of metric through a dual-path temporal convolutional network, and concatenate all the time-dimensional features of the metrics.

[0085] It should be noted that, optionally, the expression of the temporal convolutional module is:

[0086] where H time is the time feature, is the time feature of m types of metrics, and X (m) is the metric feature tensor of m types.

[0087] It can be understood that the time features of all M metrics are concatenated according to the metric dimension to form a feature tensor containing all the time information of the metrics. At the same time, the dual-path temporal convolutional network (TCN) designs two branches of long period and short period for each type of metric. The long-period branch is used to capture the daily / weekly level trends, such as the weekly cycle pattern of the number of users; the short-period branch is used to extract the hourly-level fluctuations, such as the peak-hour features of the traffic. The time features of different periods are fused through feature concatenation.

[0088] In step S220, an adaptive hybrid graph learning module is constructed. The adaptive hybrid graph learning module is used to receive the multi-metric dynamic graph and obtain the spatial-dimensional features of each type of metric through an independent graph attention network and cross-metric gating.

[0089] It should be noted that, optionally, the expression of the adaptive hybrid graph learning module is:

[0090]

[0091] α m = Softmax(W m ·X (m) ), where H space is the spatial-dimensional feature, α m is the learnable metric importance weight, is the spatial-dimensional feature of m types of metrics, A (m) is the m-type metric perception adjacency tensor, and W m is the parameter matrix.

[0092] It can be understood that the wireless cell network is modeled as a dynamic graph G, where the node set is the base station nodes and the edge set is the physical connection edges. An index-aware adjacency tensor is constructed, and each sub-matrix is generated through adaptive graph learning, representing the spatial correlation strength of specific metrics between base stations. For each type of metric, an independent graph attention network (GAT) is used to generate spatial features, and then the spatial features are fused through a cross-metric gating mechanism. A learnable metric importance weight and parameter matrix are introduced to realize the differential generation of spatial attention weights for different metrics. The GAT uses a multi-head attention mechanism, such as a 3-head attention mechanism. The gated output is normalized by sigmoid to dynamically adjust the attention weights between nodes and capture the implicit spatial correlation. At the same time, it is normalized by softmax to realize the dynamic weighted fusion of spatial features of different metrics.

[0093] In step S230, a spatio-temporal adaptive module is constructed. The spatio-temporal adaptive module is used to receive the time-dimensional feature and the space-dimensional feature input in parallel, superimpose the time-dimensional feature and the space-dimensional feature, and generate a spatio-temporal gating weight through a multi-layer perceptron, and output the spatio-temporal joint feature after adaptive fusion.

[0094] It should be noted that, optionally, the expression of the spatio-temporal adaptive module is:

[0095] Where is the spatio-temporal joint feature.

[0096] It can be understood that for the cell heterogeneity problem, the spatio-temporal contribution weights of each metric are adjusted through a spatio-temporal gating mechanism. The spatio-temporal gating generates a spatio-temporal gating weight Gate ST , weights the spatio-temporal features, dynamically fuses the spatio-temporal features across metrics, and reduces the spatial noise interference.

[0097] In step S240, all the outputs of the adaptive hybrid spatio-temporal learning blocks are connected to the output layer through skip connections, the spatio-temporal joint features output by each adaptive hybrid spatio-temporal learning block are fused, and the multi-metric prediction result is obtained through the first formula.

[0098] It should be noted that, optionally, the first formula is

[0099] Where is the multi-metric prediction result, H out is the skip connection fusion result, W f1 , b f1 , W f2 , b f2 are learnable parameters, K is the number of adaptive hybrid spatio-temporal learning blocks, The spatio-temporal joint feature output for the i-th adaptive hybrid spatio-temporal learning block.

[0100] It can be understood that the outputs of all adaptive hybrid spatio-temporal learning blocks are connected to the output layer through skip connections. After fusing the spatio-temporal joint features output by each adaptive hybrid spatio-temporal learning block, two fully connected layers are passed through, and then the multi-index prediction results for multiple future steps are generated through the first formula. The role of the fully connected layer is to perform a linear transformation on the input features, change the dimension of the features, and at the same time add a bias to give the model more learning flexibility.

[0101] In step S300, the preprocessed multi-index historical data is divided into a training set, a validation set, and a test set. After multi-cycle data division, it is input into the adaptive hybrid spatio-temporal graph neural network model for training.

[0102] In one embodiment, as Figure 4 shown, step S300 may include the following sub-steps:

[0103] In step S310, the preprocessed multi-index historical data is divided into a training set, a validation set, and a test set in a ratio of 7:1.5:1.5.

[0104] It can be understood that after division, the data volume of the training set is 70% of the preprocessed multi-index historical data, the validation set is 15% of the preprocessed multi-index historical data, and the test set is 15% of the preprocessed multi-index historical data.

[0105] In step S320, the divided multi-index historical data is divided according to time order to obtain multi-cycle data, and the multi-cycle data includes recent data, daily data, and weekly data.

[0106] It should be noted that, optionally, as Figure 6 the recent data (X R ) is the data within the past 1-2 hours, used to capture short-term fluctuation features. The daily data (X D ) is the data at the same time period every day within the past week, used to extract daily cycle rules. The weekly data (X W ) is the data at the same time of the same day of the same week within the past four weeks, used to mine weekly cycle trends.

[0107] In step S330, the Adam optimizer and the mean absolute error loss function are used to train the adaptive hybrid spatio-temporal graph neural network model through multi-cycle data, and when the loss value on the validation set does not decrease within multiple rounds, the trained adaptive hybrid spatio-temporal graph neural network model is obtained.

[0108] It should be noted that, optionally, the mean absolute error loss function is:

[0109]

[0110] Among them, L(θ) is the loss function, M is the number of metrics, and Y i is the true value of the i-th sample, is the predicted value of the i-th sample.

[0111] It can be understood that the model is trained by backpropagation using the Adam optimizer through the mean absolute error loss function, and the loss of the validation set is monitored to avoid overfitting.

[0112] In step S400, the multi-metric data of the wireless cell collected in real time is input into the trained adaptive hybrid spatio-temporal graph neural network model to obtain the multi-metric prediction value of the wireless cell.

[0113] In one embodiment, as Figure 5 shown, step S400 may include the following sub-steps:

[0114] In step S410, the multi-metric data of the wireless cell collected in real time is preprocessed and divided into multi-period data to obtain real-time multi-period data.

[0115] It can be understood that the multi-metric data of the wireless cell collected in real time is processed according to the input format and preprocessing method set by the model to obtain recent, daily, and weekly data.

[0116] In step S420, the real-time multi-period data is input into the trained adaptive hybrid spatio-temporal graph neural network model to obtain the multi-metric prediction value of the wireless cell.

[0117] It should be noted that the real-time multi-period data is used to train the model through the Adam optimizer with a learning rate of 0.001.

[0118] Furthermore, in this exemplary embodiment, a multi-metric spatio-temporal sequence prediction system for a wireless cell is also provided. Referring to Figure 7 shown, the system may include:

[0119] A data processing module for preprocessing the multi-metric historical data of the wireless cells in the target area and constructing a multi-metric dynamic graph;

[0120] A model construction module for constructing an adaptive hybrid spatio-temporal graph neural network model according to the multi-metric dynamic graph, and the adaptive hybrid spatio-temporal graph neural network model includes a plurality of adaptive hybrid spatio-temporal learning blocks composed of a temporal convolution module, an adaptive hybrid graph learning module, and a spatio-temporal adaptive module;

[0121] A model training module, configured to divide the pre - processed multi - metric historical data into a training set, a validation set, and a test set, and after performing multi - cycle data division, input the data into the adaptive hybrid spatio - temporal graph neural network model for training;

[0122] A model prediction module, configured to input the multi - metric data of the wireless cell collected in real - time into the trained adaptive hybrid spatio - temporal graph neural network model to obtain the predicted values of the multi - metrics of the wireless cell.

[0123] Regarding the device in the above - mentioned embodiments, the specific manners in which each module performs operations have been described in detail in the embodiments related to the method, and will not be elaborated herein.

[0124] It should be noted that although several modules or units of the device for action execution are mentioned in the above - detailed description, this division is not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more of the above - described modules or units can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units. The components shown as modules or units may or may not be physical units, that is, they may be located in one place, or may be distributed over multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of the present disclosure. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0125] In an exemplary embodiment of the present disclosure, an electronic device is further provided. The electronic device may include a processor and a memory for storing executable instructions of the processor. Wherein, the processor is configured to execute the steps of the multi - metric spatio - temporal sequence prediction method of the wireless cell in any one of the above - mentioned embodiments by executing the executable instructions.

[0126] Those skilled in the art to which the present invention pertains can understand that various aspects of the present invention can be implemented as a system, a method, or a program product. Therefore, various aspects of the present invention can be specifically implemented in the following forms, namely: a complete hardware implementation, a complete software implementation (including firmware, micro - code, etc.), or an implementation combining hardware and software aspects, which can be collectively referred to as "circuit", "module", or "system" here.

[0127] Through the description of the above embodiments, those skilled in the art can easily understand that the exemplary embodiments described herein can be implemented by software or by a combination of software and necessary hardware. Therefore, the technical solutions according to the embodiments of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a portable hard drive, etc.) or on a network, including several instructions to enable a computing device (which can be a personal computer, a server, or a network device, etc.) to execute the above-described wireless cell multi-metric spatio-temporal sequence prediction method according to the embodiments of the present disclosure.

[0128] In an exemplary embodiment of the present disclosure, there is also provided a computer storage medium, on which a computer program is stored, and when the program is executed by, for example, a processor, the steps of the wireless cell multi-metric spatio-temporal sequence prediction method described in any one of the above embodiments can be implemented.

[0129] In some possible embodiments, various aspects of the present invention can also be implemented in the form of a computer program product, which includes computer programs or instructions. When the computer program product runs on a terminal device, the computer program code or instructions are used to cause the terminal device to execute the steps according to various exemplary embodiments of the present invention described in the above wireless cell multi-metric spatio-temporal sequence prediction method section of this specification.

[0130] The above program product can be written in any combination of one or more programming languages for programming code to perform the operations of the present invention. The programming languages include object-oriented programming languages - such as Java, C++, etc., and also include conventional procedural programming languages - such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computing device, partially on the user's device, executed as an independent software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In the case of a remote computing device, the remote computing device can be connected to the user's computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (for example, by using an Internet service provider to connect through the Internet).

[0131] The computer software product can be stored in a computer storage medium, which includes Read-Only Memory (ROM), Random Access Memory (RAM), Programmable Read-only Memory (PROM), Erasable Programmable Read Only Memory (EPROM), One-time Programmable Read-Only Memory (OTPROM), Electrically-Erasable Programmable Read-Only Memory (EEPROM), Compact Disc Read-Only Memory (CD-ROM) or other optical disc memories, magnetic disk memories, magnetic tape memories, or any other computer-readable medium capable of carrying or storing data.

[0132] Those skilled in the art will readily conceive of other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present disclosure, which follow the general principles of the present disclosure and include known common general knowledge or conventional technical means in the technical field not disclosed by the present disclosure. The specification and examples are only regarded as exemplary, and the true scope and spirit of the present disclosure are pointed out by the appended claims.

Claims

1. A method for predicting multi-index spatio-temporal sequences in a wireless cell, characterized in that, Including: Preprocessing the multi - metric historical data of wireless cells in the target area and constructing a multi - metric dynamic graph; Constructing an adaptive hybrid spatio - temporal graph neural network model according to the multi - metric dynamic graph, where the adaptive hybrid spatio - temporal graph neural network model includes multiple adaptive hybrid spatio - temporal learning blocks composed of a temporal convolution module, an adaptive hybrid graph learning module, and a spatio - temporal adaptive module; Dividing the preprocessed multi - metric historical data into a training set, a validation set, and a test set, and after performing multi - cycle data division, inputting it into the adaptive hybrid spatio - temporal graph neural network model for training; Inputting the real - time collected multi - metric data of wireless cells into the trained adaptive hybrid spatio - temporal graph neural network model to obtain the multi - metric prediction values of wireless cells.

2. The wireless cell multi-index spatio-temporal sequence prediction method according to claim 1, wherein The step of preprocessing the multi - metric historical data of wireless cells in the target area and constructing a multi - metric dynamic graph includes: Obtaining the multi - metric historical data of wireless cells in the target area, where the multi - metric historical data includes traffic data and user number data; Cleaning and normalizing the multi - metric historical data; Regarding each wireless cell as a graph node, and constructing a multi - metric dynamic graph according to the geographical adjacency relationship or communication association relationship between each graph node.

3. The method for predicting multi-index spatio-temporal sequences of a wireless cell according to claim 2, characterized in that The multi - metric dynamic graph is \(G=(V, E, A, X)\), where \(V\) is the set of base station nodes, \(E\) is the set of physical connection edges, \(A\) is the metric - aware adjacency tensor, and \(X\) is the multi - metric feature tensor.

4. The wireless cell multi-index spatio-temporal sequence prediction method according to claim 1, wherein The step of constructing an adaptive hybrid spatio - temporal graph neural network model according to the multi - metric dynamic graph, where the adaptive hybrid spatio - temporal graph neural network model includes multiple adaptive hybrid spatio - temporal learning blocks composed of a temporal convolution module, an adaptive hybrid graph learning module, and a spatio - temporal adaptive module, includes: Constructing a temporal convolution module, which is used to receive multi - cycle time data, extract the time - dimension features of each type of metric through a dual - path temporal convolution network, and splice all the time - dimension features of the metrics; Constructing an adaptive hybrid graph learning module, which is used to receive the multi - metric dynamic graph and obtain the spatial - dimension features of each type of metric through an independent graph attention network and cross - metric gating; Constructing a spatio - temporal adaptive module, which is used to receive the time - dimension features and spatial - dimension features input in parallel, stack the time - dimension features and spatial - dimension features, and generate spatio - temporal gating weights through a multi - layer perceptron to output the adaptively fused spatio - temporal joint features; Connecting the outputs of all adaptive hybrid spatio - temporal learning blocks to the output layer through skip connections, fusing the spatio - temporal joint features output by each adaptive hybrid spatio - temporal learning block, and obtaining the multi - metric prediction result through the first formula.

5. The method for predicting multi-index spatio-temporal sequences of a wireless cell according to claim 4, wherein The expression of the time convolution module is Among them, H time is a time feature, is the time feature of the m-type index, X (m) is the feature tensor of the m-type index; The expression of the adaptive hybrid graph learning module is Among them, α m = Softmax(W m ·X (m) ), H space is the spatial dimension feature, α m is the importance weight of learnable indicators, is the spatial dimension feature of m types of indicators, A (m) is the perception adjacency tensor of m types of indicators, W m is the parameter matrix; The expression of the spatio-temporal adaptive module is Among them, is the spatio-temporal joint feature; The first formula is Among them, is the multi-index prediction result, H out is the skip connection fusion result, W f1 , b f1 , W f2 , b f2 are learnable parameters, K is the number of adaptive hybrid spatio-temporal learning blocks, is the spatio-temporal joint feature output by the i-th adaptive hybrid spatio-temporal learning block.

6. The method for predicting multi-index spatio-temporal sequences of a wireless cell according to claim 1, wherein, The step of dividing the preprocessed multi - metric historical data into a training set, a validation set, and a test set, and after performing multi - cycle data division, inputting it into the adaptive hybrid spatio - temporal graph neural network model for training includes: Dividing the preprocessed multi - metric historical data into a training set, a validation set, and a test set at a ratio of 7:1.5:1.5; The divided multi-index historical data is partitioned in chronological order to obtain multi-period data, and the multi-period data includes recent data, daily data, and weekly data; The Adam optimizer and the mean absolute error loss function are used to train the adaptive hybrid spatio-temporal graph neural network model with the multi-period data, and when the loss value on the validation set does not decrease within multiple rounds, the trained adaptive hybrid spatio-temporal graph neural network model is obtained.

7. The method for predicting multi-index spatio-temporal sequences of a wireless cell according to claim 6, wherein The mean absolute error loss function is Among them, \(L(\theta)\) is the loss function, \(M\) is the number of metrics, and \(Y\) i is the true value of the \(i\)-th sample, and \(\hat{Y}\) is the predicted value of the \(i\)-th sample.

8. The wireless cell multi-index spatio-temporal sequence prediction method according to claim 1, wherein The step of inputting the wireless cell multi-index data collected in real time into the trained adaptive hybrid spatio-temporal graph neural network model to obtain the wireless cell multi-index prediction value includes: Preprocessing the wireless cell multi-index data collected in real time and performing multi-period data partitioning to obtain real-time multi-period data; Inputting the real-time multi-period data into the trained adaptive hybrid spatio-temporal graph neural network model to obtain the wireless cell multi-index prediction value.

9. The method for predicting the spatio-temporal sequence of multiple indicators of a wireless cell according to claim 8, wherein, The real-time multi-period data is used to train the model through the Adam optimizer with a learning rate of 0.

001.

10. A multi-index spatio-temporal sequence prediction system for a wireless cell, characterized in that, The system is used to execute the method described in any one of claims 1 to 9, and the system includes: A data processing module for preprocessing the multi-index historical data of wireless cells in the target area and constructing a multi-index dynamic graph; A model construction module for constructing an adaptive hybrid spatio-temporal graph neural network model according to the multi-index dynamic graph, and the adaptive hybrid spatio-temporal graph neural network model includes multiple adaptive hybrid spatio-temporal learning blocks composed of a temporal convolution module, an adaptive hybrid graph learning module, and a spatio-temporal adaptive module; A model training module for dividing the preprocessed multi-index historical data into a training set, a validation set, and a test set, and after performing multi-period data partitioning, inputting it into the adaptive hybrid spatio-temporal graph neural network model for training; A model prediction module for inputting the wireless cell multi-index data collected in real time into the trained adaptive hybrid spatio-temporal graph neural network model to obtain the wireless cell multi-index prediction value.

Images