Network capacity prediction method and device, electronic equipment and storage medium

Abstract

The application provides a network capacity prediction method and device, electronic equipment and storage medium. The method comprises the following steps: obtaining a fusion evaluation parameter of a to-be-predicted area; inputting the fusion evaluation parameter and the correlation weight between any network management side performance parameter in the fusion evaluation parameter and each fusion evaluation parameter except the any network management side performance parameter into a capacity prediction model to obtain a network capacity prediction value. According to the method provided by the application, the correlation weight between the fusion evaluation parameter and any network management side performance parameter and each fusion evaluation parameter except the any network management side performance parameter is input into the capacity prediction model, the correlation between the user track change and the space change reflected by the correlation weight is indicated, the capacity prediction model is instructed to capture the correlation between the user track change and the space change in the fusion evaluation parameter, and fast and accurate network capacity prediction is realized.

Description

Technical Field

[0001] This invention relates to the field of communication technology, and in particular to a network capacity prediction method, apparatus, electronic device, and storage medium. Background Technology

[0002] Large venues, compared to general areas, are characterized by high population density and higher concurrent service demands. Therefore, network capacity prediction in large venues has always been a hot topic for improving wireless network user performance. Traditional wireless network optimization cannot quickly adjust to rapid changes in service models, exhibiting a "lag." Consequently, current common network capacity prediction methods typically rely on Transformer and ARIMA models, along with feature decomposition of historical data, to achieve network capacity prediction.

[0003] However, current common network service forecasting methods are mainly based on single historical indicators, and some features are biased, so the accuracy of the network capacity forecasting results needs to be further improved. Summary of the Invention

[0004] This invention provides a network capacity prediction method, apparatus, electronic device, and storage medium to address the shortcomings of existing technologies in short-term network capacity prediction, where the accuracy of prediction results needs further improvement.

[0005] This invention provides a network capacity prediction method, comprising:

[0006] Obtain the fusion evaluation parameters for the region to be predicted;

[0007] The association weights between the fusion evaluation parameters and any network management side performance parameter among the fusion evaluation parameters and each fusion evaluation parameter other than the network management side performance parameter are input into the capacity prediction model to obtain the network capacity prediction value of the region to be predicted output by the capacity prediction model.

[0008] The capacity prediction model is obtained by training the initial capacity prediction model based on the sample fusion evaluation parameters.

[0009] According to the present invention, a network capacity prediction method is provided, wherein the capacity prediction model includes a graph convolutional unit and a gated recurrent unit;

[0010] The step of inputting the association weights between the fusion evaluation parameters and any network management-side performance parameter among the fusion evaluation parameters and each fusion evaluation parameter excluding the network management-side performance parameter into the capacity prediction model to obtain the network capacity prediction value of the region to be predicted output by the capacity prediction model includes:

[0011] The parameter features of the fusion evaluation parameters for the historical moments are extracted based on the graph convolutional units and the associated weights.

[0012] Based on the gated loop unit and the self-attention mechanism, the parameter features of the historical moments are processed to obtain the parameter features of the current moment;

[0013] Based on the parameter characteristics at the current moment, the predicted network capacity value is obtained.

[0014] According to a network capacity prediction method provided by the present invention, the gated loop unit includes an update gate, a reset gate, a candidate hidden gate, and an update hidden gate;

[0015] The step of performing feature processing on the parameter features of the historical time based on the gated recurrent unit and the self-attention mechanism to obtain the parameter features of the current time includes:

[0016] Based on the update gate and the first attention weight corresponding to the update gate, the parameter features of the historical moment are preserved to obtain the first preserved features;

[0017] Based on the reset gate and the second attention weight corresponding to the reset gate, the parameter features of the historical moment are discarded to obtain the second retained features;

[0018] Based on the candidate hidden gate and the third attention weight corresponding to the candidate hidden gate, feature processing is performed on the parameter features of the historical moment and the second preserved features to obtain the third preserved features;

[0019] Based on the updated hidden gate, feature processing is performed on the first retained feature, the second retained feature, and the third retained feature to obtain the current time parameter feature.

[0020] According to a network capacity prediction method provided by the present invention, obtaining the predicted network capacity value based on the parameter characteristics at the current time includes:

[0021] Based on the prediction function and the parameter characteristics at the current moment, the initial capacity prediction value is obtained;

[0022] Based on the network adjustment parameters of the region to be predicted, network capacity prediction is performed on the region to be predicted to obtain the capacity adjustment prediction value.

[0023] The initial capacity prediction value is corrected based on the capacity adjustment prediction value to obtain the network capacity prediction value.

[0024] According to a network capacity prediction method provided by the present invention, the network adjustment parameters include at least the number of inter-base station system handovers and cell individual offset parameters.

[0025] According to the network capacity prediction method provided by the present invention, the fusion evaluation parameters further include base station-side physical parameters and software configuration parameters;

[0026] The steps for obtaining the correlation weight between any network management side performance parameter and each other fusion evaluation parameter (excluding the network management side performance parameter) in the fusion evaluation parameters include:

[0027] Based on a graph convolutional network, the connection edges between any network management side performance parameter and each fusion evaluation parameter are calculated to obtain the association weights between any network management side performance parameter and each fusion evaluation parameter other than the any network management side performance parameter.

[0028] The present invention also provides a network capacity prediction device, comprising:

[0029] The acquisition unit acquires the fusion evaluation parameters for the region to be predicted.

[0030] The prediction unit inputs the association weights between the fusion evaluation parameters and any network management side performance parameter among the fusion evaluation parameters and each fusion evaluation parameter excluding the network management side performance parameter into the capacity prediction model, and obtains the network capacity prediction value of the region to be predicted output by the capacity prediction model.

[0031] The capacity prediction model is obtained by training the initial capacity prediction model based on the sample fusion evaluation parameters.

[0032] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the network capacity prediction method as described above.

[0033] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the network capacity prediction method as described above.

[0034] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the network capacity prediction method as described above.

[0035] The network capacity prediction method, apparatus, electronic device, and storage medium provided by this invention input the correlation weights between the fusion evaluation parameters and any network management side performance parameter and each fusion evaluation parameter excluding any network management side performance parameter into the capacity prediction model. Through the correlation weights reflecting the relationship between user trajectory changes and spatial changes, the capacity prediction model is instructed to capture the relationship between user trajectory changes and spatial changes in the fusion evaluation parameters, thereby achieving fast and accurate network capacity prediction. Attached Figure Description

[0036] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0037] Figure 1 This is one of the flowcharts illustrating the network capacity prediction method provided by the present invention;

[0038] Figure 2 This is the second flowchart of the network capacity prediction method provided by the present invention;

[0039] Figure 3 This is a comparative schematic diagram of feature reconstruction using the fusion evaluation parameters provided by the present invention;

[0040] Figure 4 This is a schematic diagram of the feature reconstruction relationship graph obtained based on the maximum number of users in RRC provided by the present invention;

[0041] Figure 5 This is the third flowchart of the network capacity prediction method provided by the present invention;

[0042] Figure 6 This is a schematic diagram illustrating the relationship between user movement trajectories and physical characteristics provided by the present invention;

[0043] Figure 7 This is a schematic diagram of the network capacity prediction device provided by the present invention;

[0044] Figure 8 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0045] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0046] Currently, short-term prediction methods for large venues mainly employ two models: Transformer and ARIMA. Among them, Transformer is a widely used and versatile prediction model. Its main approach involves optimizing the self-attention mechanism and performing feature decomposition on the input sequence. This design allows Transformer to capture richer temporal dependencies with lower complexity when processing longer historical data. However, although these optimizations reduce some computational burden, the large parameter size of the Transformer model results in a significantly higher computational cost than simple linear network models, and this increased computational burden has not significantly improved the model's prediction accuracy.

[0047] Furthermore, the basic idea of ​​the ARIMA model is to use historical information from the data itself to predict the future. The label value at a point in time is influenced by both label values ​​over a past period and by random events over that period. This means that the label values ​​in the ARIMA model fluctuate around a major trend over time, where the trend is influenced by historical labels and the fluctuations are influenced by random events over a period of time. Moreover, the major trend itself is not necessarily stable. This method is not effective in handling short-term predictions.

[0048] Therefore, current network capacity prediction schemes are not very accurate in short-term predictions for large venues. To address this problem, this invention provides a network capacity prediction method that overcomes the shortcomings of low accuracy in short-term network predictions for large venues, thereby achieving accurate and efficient short-term network capacity prediction. Figure 1 This is one of the flowcharts illustrating the network capacity prediction method provided by the present invention, such as... Figure 1 As shown, the method includes:

[0049] Step 110: Obtain the fusion evaluation parameters for the region to be predicted;

[0050] Step 120: Input the association weights between the fusion evaluation parameters and any network management side performance parameter among the fusion evaluation parameters and each fusion evaluation parameter other than the network management side performance parameter into the capacity prediction model to obtain the network capacity prediction value of the region to be predicted output by the capacity prediction model.

[0051] The capacity prediction model is obtained by training the initial capacity prediction model based on the sample fusion evaluation parameters.

[0052] Here, the area to be predicted refers to large venues or other areas with high population density and high business concurrency demand. The fusion evaluation parameters for the area to be predicted can include network management side performance parameters and other network parameters that do not change over time. The network management side performance parameters here can include the maximum number of RRC (Radio Resource Control) users, uplink and downlink PRB (Physical Resource Block) utilization, traffic, etc. It is understandable that the network management side performance parameters reflect parameters that change over time.

[0053] Furthermore, the network management-side performance parameters in the fusion evaluation parameters can reflect the operational status of the network in the area under test, and thus reflect the network's actual carrying capacity and performance. In other words, the network management-side performance parameters can reflect time characteristics. It is understandable that areas with higher traffic volume have stronger actual network carrying capacity; conversely, areas with lower traffic volume have relatively weaker actual network carrying capacity. In addition, other network parameters that do not change over time can reflect the physical parameter characteristics of the area under test, such as location, coverage area, and signal strength. Therefore, other fusion evaluation parameters can reflect spatial characteristics.

[0054] The capacity prediction model here can be built based on graph convolutional networks, and the initial capacity prediction model can be iteratively trained using pre-acquired sample fusion evaluation parameters to optimize the model parameters of the initial capacity prediction model until an accurate capacity prediction model is obtained.

[0055] Specifically, when performing short-term network capacity prediction for the region to be predicted, the first step is to obtain the fusion evaluation parameters of the region at historical times. For example, obtain the fusion evaluation parameters for each minute of the previous 10 minutes. Then, the fusion evaluation parameters and the correlation weights between any pre-obtained network management side performance parameter and each fusion evaluation parameter excluding the network management side performance parameter can be input into the capacity prediction model. The correlation weights reflect the relationship between user trajectories and spatial features, helping the capacity prediction model capture the features in the fusion evaluation parameters that are associated with the network capacity prediction value. The capacity prediction model then outputs the network capacity prediction value for the region to be predicted, achieving accurate network capacity prediction. Here, the network capacity prediction value can be a future network management side performance parameter, such as uplink PRB utilization, downlink PRB utilization, maximum number of RRC users, and traffic.

[0056] It should be noted that feature calculations are performed on the network management side performance parameters and other fusion evaluation parameters in the fusion evaluation parameters to reconstruct the correlation between temporal and spatial features, thereby capturing the correlation between the temporal and spatial features of the fusion evaluation parameters. That is, the correlation weight between any network management side performance parameter and each of the fusion evaluation parameters other than that parameter can be used to reflect the degree of correlation between user trajectories and physical parameters in the area to be predicted. It can be understood that by capturing the relationship between user trajectories and physical parameters in the fusion evaluation parameters, network usage demand at different times and locations can be predicted more accurately, thus facilitating better network performance optimization.

[0057] It should also be noted that the correlation weights between any network management side performance parameter and each fusion evaluation parameter other than the network management side performance parameter are pre-calculated. Therefore, when performing network capacity prediction through the capacity prediction model, the calculation process of correlation weights can be removed, which not only simplifies the model structure of the capacity prediction model, but also greatly improves the efficiency of network capacity prediction, thereby enabling short-term capacity prediction for large venue networks.

[0058] The method provided in this invention inputs the correlation weights between the fusion evaluation parameters and any network management side performance parameter and each fusion evaluation parameter excluding any network management side performance parameter into the capacity prediction model. By reflecting the relationship between user trajectory changes and spatial changes through the correlation weights, the capacity prediction model is instructed to capture the relationship between user trajectory changes and spatial changes in the fusion evaluation parameters, thereby achieving fast and accurate network capacity prediction.

[0059] Based on any of the above embodiments, the capacity prediction model includes a graph convolution unit and a gated recurrent unit;

[0060] Step 120 above includes:

[0061] The parameter features of the fusion evaluation parameters for the historical moments are extracted based on the graph convolutional units and the associated weights.

[0062] Based on the gated loop unit and the self-attention mechanism, the parameter features of the historical moments are processed to obtain the parameter features of the current moment;

[0063] Based on the parameter characteristics at the current moment, the predicted network capacity value is obtained.

[0064] Here, the capacity prediction model can include an input layer, graph convolutional units, gated recurrent units, and an output layer. Furthermore, the input layer, graph convolutional units, gated recurrent units, and output layer are sequentially connected. Additionally, the graph convolutional units can be constructed based on a multi-layer GCN (Graph Convolutional Network), for example, by setting the number of GCN layers to 3, to reduce short-term computational load, enabling real-time prediction and facilitating short-term capacity prediction for large venue networks.

[0065] Specifically, firstly, the fusion evaluation parameters and associated weights from historical moments are input into the input layer. Then, the input layer transmits these parameters and weights to a graph convolutional unit. Within the graph convolutional unit, the relationship graph corresponding to the fusion evaluation parameters is reconstructed using the associated weights. That is, using each fusion evaluation parameter as a node and the associated weights as connecting edges between corresponding nodes, feature reconstruction is performed to obtain the reconstructed relationship graph corresponding to the fusion evaluation parameters. Then, the graph convolutional unit performs convolution on the reconstructed relationship graph to extract the node features of each node, thus obtaining the parameter features. It can be understood that these parameter features reflect the relationship between user trajectories and spatial features among the fusion evaluation parameters.

[0066] Furthermore, the parameter features are input into a gated recurrent unit (GRU). Based on the GRU and the self-attention mechanism, feature processing is performed on the parameter features to enhance and increase the attention given to features related to network capacity prediction, thus obtaining the parameter features at the current time step. Finally, the parameter features at the current time step are input into the output layer. Based on the prediction function of the output layer and the parameter features at the current time step, the network capacity value is predicted to obtain the predicted network capacity value.

[0067] Based on any of the above embodiments, the gated loop unit includes an update gate, a reset gate, a candidate hidden gate, and an update hidden gate;

[0068] The step of performing feature processing on the parameter features of the historical time based on the gated recurrent unit and the self-attention mechanism to obtain the parameter features of the current time includes:

[0069] Based on the update gate and the first attention weight corresponding to the update gate, the parameter features of the historical moment are preserved to obtain the first preserved features;

[0070] Based on the reset gate and the second attention weight corresponding to the reset gate, the parameter features of the historical moment are discarded to obtain the second retained features;

[0071] Based on the candidate hidden gate and the third attention weight corresponding to the candidate hidden gate, feature processing is performed on the parameter features of the historical moment and the second preserved features to obtain the third preserved features;

[0072] Based on the updated hidden gate, feature processing is performed on the first retained feature, the second retained feature, and the third retained feature to obtain the current time parameter feature.

[0073] Specifically, the update gate is used to preserve features; that is, based on the first attention weight corresponding to the update gate, features that need to be retained in the next time step are selected from the parameter features of the previous time step, resulting in the first retained features. These first retained features can be calculated using the following formula, as shown below:

[0074]

[0075] In the formula, This represents the first preserved feature at time t; This represents the sigmoid activation function; Indicates the first attention weight; express The final computed value of a feature of a hidden layer at a given time; This represents the parameter characteristics at time t.

[0076] The reset gate is used to discard features. Specifically, based on the second attention weight corresponding to the reset gate, features that need to be discarded in the next time step are selected from the parameter features of the previous time step, resulting in the second retained features. These second retained features can be calculated using the following formula, as shown below:

[0077]

[0078] In the formula, This represents the second preserved feature at time t.

[0079] Candidate hidden gates are used to represent features with smaller weights. Based on the third attention weights corresponding to the candidate hidden gates, feature processing is performed on the parametric features and second preserved features at historical time points to obtain the third preserved feature. This third preserved feature can be calculated using the following formula, as shown below:

[0080]

[0081] In the formula, This represents the third preserved feature at time t; It is the hyperbolic tangent function.

[0082] The updated hidden gate is used to process the first, second, and third retained features to obtain the final features that need to be retained, i.e., the current-time parameter features. These current-time parameter features can be calculated using the following formula, as shown below:

[0083]

[0084] In the formula, This indicates the parameter characteristics at the current moment.

[0085] It should be noted that the gated recurrent unit (GRU) controls the flow and forgetting of information through update and reset gates, thereby achieving modeling of sequence data. The update gate determines how much of the information at the current time step should come from the hidden state at the previous time step, while the reset gate controls the influence of past information on the current time step. This gating mechanism enables the GRU to better capture important information in the sequence when processing feature sequence data, improving the accuracy of network capacity prediction. It also mitigates the gradient vanishing problem, making the model easier to train and optimize.

[0086] In practical applications, wireless network optimization is a dynamic process. That is, load balancing between base stations typically prioritizes adjusting CIO parameters, and the number of handovers between base stations also changes significantly with CIO adjustments. Therefore, to further improve the accuracy of network capacity prediction, based on any of the above embodiments, obtaining the predicted network capacity value based on the parameter characteristics at the current moment includes:

[0087] Based on the prediction function and the parameter characteristics at the current moment, the initial capacity prediction value is obtained;

[0088] Based on the network adjustment parameters of the region to be predicted, network capacity prediction is performed on the region to be predicted to obtain the capacity adjustment prediction value.

[0089] The initial capacity prediction value is corrected based on the capacity adjustment prediction value to obtain the network capacity prediction value.

[0090] Based on any of the above embodiments, the network adjustment parameters include at least the number of inter-base station system handovers and cell-specific offset parameters.

[0091] Specifically, the initial capacity prediction value can be calculated by inputting the current moment's parameter features into the prediction function. This enables short-term capacity prediction of large venues based on the prediction function generated by the GCN model, and allows prediction of future 5-minute performance changes based on historical 10-minute performance metrics, supporting 5-minute granular time-series feature prediction and user trajectory prediction.

[0092] It should be noted that time series forecasting essentially captures patterns that change over time in the input data, and fits these patterns to a mathematical model to minimize the error between the model's predictions and actual observations. The capacity forecasting model here uses a Graph Convolutional Network (GCN) to achieve automatic graph updates by learning graph structures through data-driven learning. By constructing an adjacency matrix, hidden spatial dependencies are preserved, and invisible graph structures can be discovered from the data without any prior knowledge. During the training of the capacity forecasting model, the core convolutional layer parameters and variables are derived from training the GCN model. Therefore, in practical applications of capacity forecasting, the prediction function can be represented by the following formula, as shown below:

[0093]

[0094] In the formula, Represents the prediction function; Indicates the number of layers in the GCN neural network; Represents the normal transformation matrix; Represents the inverse transformation matrix; , , They represent The weights of the layer features; X represents the input parameters at the current time step.

[0095] Furthermore, network capacity prediction for the region to be predicted can be performed using network adjustment parameters, yielding a capacity adjustment prediction value. These parameters can include the number of inter-base station system handovers and individual cell offset parameters in historical data for the region. The number of inter-base station system handovers can include intra-system handovers (e.g., handovers between 4G and 4G base stations, or between 5G and 5G base stations) and inter-system handovers (e.g., handovers between 4G and 5G base stations). Therefore, network capacity prediction can be performed for the region to be predicted based on the dynamically optimized network parameters from historical data, yielding a capacity adjustment prediction value. Subsequently, the initial capacity prediction value can be corrected based on the capacity adjustment prediction value. For example, the capacity adjustment prediction value can be added to the initial capacity prediction value to obtain the final network capacity prediction value, achieving a more accurate network capacity prediction.

[0096] In one embodiment, Figure 2 This is the second flowchart of the network capacity prediction method provided by the present invention, as shown below. Figure 2 As shown, the method includes:

[0097] First, network management side performance data, base station side physical information, software side configuration information, and corresponding weights are input into the GRU gating unit. The GRU gating unit then performs deep learning time-series prediction to obtain an initial capacity prediction value. Simultaneously, network adjustment parameters for the region to be predicted are input into the model bias correction module to obtain the capacity adjustment prediction value. Finally, the initial capacity prediction value is corrected based on the capacity adjustment prediction value to obtain the final network capacity prediction value.

[0098] The method provided in this embodiment of the invention takes into account that the optimization and adjustment of wireless networks is a dynamic process during short-term network capacity prediction. Therefore, after the initial network capacity prediction based on the fusion evaluation parameters, a model correction unit is added. This unit uses the network optimization and adjustment parameters from past times to predict the network capacity after optimization and adjustment, and corrects the initial capacity prediction value based on the capacity adjustment prediction value. This fully considers the impact of wireless network optimization and adjustment on the network capacity prediction results, and greatly improves the accuracy of short-term network capacity prediction.

[0099] Based on any of the above embodiments, the fusion evaluation parameters also include base station-side physical parameters and software configuration parameters;

[0100] The steps for obtaining the correlation weight between any network management side performance parameter and each other fusion evaluation parameter (excluding the network management side performance parameter) in the fusion evaluation parameters include:

[0101] Based on a graph convolutional network, the connection edges between any network management side performance parameter and each fusion evaluation parameter are calculated to obtain the association weights between any network management side performance parameter and each fusion evaluation parameter other than the any network management side performance parameter.

[0102] Here, base station-side physical parameters may include base station mounting height, antenna azimuth angle, antenna downtilt angle, inter-site spacing, and RRU (Remote Radio Unit) capabilities. It should be noted that base station-side physical parameters collectively influence the layout and performance of the wireless communication network. For example, base station mounting height and antenna azimuth angle determine the signal coverage direction and range, antenna downtilt angle is used to adjust the vertical coverage area of ​​the signal, and inter-site spacing affects the network density and capacity. It should also be noted that, compared to existing technologies, network capacity prediction no longer only considers dynamically changing performance indicators but also physical site information. By reconstructing features including both temporal and spatial domains, the shortcomings of traditional capacity prediction in multi-feature fusion are compensated for. The key to feature reconstruction lies in using the node relationships of a GCN (Graph Convolutional Network) to calculate weights, thereby improving the accuracy of network capacity prediction from the data source.

[0103] In addition, software-defined configuration parameters can include power configuration, frequency configuration, bandwidth configuration, handover parameters, load balancing parameters, etc. It should be noted that software-defined configuration parameters can be used to reflect network operating conditions and optimize performance.

[0104] Specifically, the original relationship graph of the rules can be constructed using each fusion evaluation parameter as a node. This original relationship graph is then input into the GCN network. Any parameter is selected from the network management side performance parameters, and connection edges are calculated between this parameter and the other fusion evaluation parameters to obtain the corresponding association weights. It should be noted that for a graph-structured dataset G, which has N nodes, each node has its own features. These features can be arranged into an N*D matrix x, where D represents the hidden state dimension of each node. The relationships between nodes can also be extracted as an N*N relation matrix x, also known as an adjacency matrix, where y represents the validation set. Using A to represent the edge information between nodes, the feature weight w is calculated based on the minimum value of the loss function minE((x,y),y′)∼p(ℓ(y′,y^(x,y))). Figure 3 This is a comparative schematic diagram of feature reconstruction using the fusion evaluation parameters provided by the present invention, as shown below. Figure 3 As shown, in the graph before feature reconstruction, the connecting edges between each fusion evaluation parameter (x1, x2, ...) are all the same, failing to convey the relationships between nodes, i.e., failing to express the intrinsic connection between user trajectories and spatial features. However, in the graph after feature reconstruction, the nodes are no longer Euclidean regular shapes, but rather represent more of the connections between nodes.

[0105] Figure 4 This is a schematic diagram of the feature reconstruction relationship graph obtained based on the maximum number of users in RRC provided by the present invention, as shown below. Figure 4 As shown, taking the maximum number of RRC users as an example, the correlation weights between the maximum number of RRC users and each other fusion evaluation parameter are calculated. After feature reconstruction, the weights between each feature can be obtained. A hierarchical structure is formed through weighted calculations. The first layer has the largest weight, such as the number of users and historical user count, traffic, and power configuration having the largest weights. The second layer has the next largest weight, such as the number of users and downlink PRB utilization, uplink PRB utilization, connection height, and handover parameters, and so on. Specifically, Table 1 shows the correlation weights between any network management side performance parameter and each fusion evaluation parameter other than the stated network management side performance parameter, as shown in Table 1 below:

[0106] Table 1

[0107]

[0108] It is understandable that by calculating each feature individually, relational maps of different features can be obtained separately.

[0109] In response to the insensitivity of traditional prediction algorithms to short-term business and their inability to provide timely early warnings, the method provided in this embodiment of the invention moves some computing power offline, greatly reducing the computing cost during the protection period and outputting the capacity prediction results in a timely manner.

[0110] Based on any of the above embodiments Figure 5 This is the third flowchart of the network capacity prediction method provided by the present invention, as shown below. Figure 5 As shown, the method includes:

[0111] First, input data, including network management side performance data, base station side physical information, and software configuration information, is used to construct the underlying feature dataset. Then, spatiotemporal feature reconstruction is performed on the underlying feature dataset. The spatiotemporal feature weights W are output based on the GCN feature reconstruction module.

[0112] Further, we move into the short-term business forecasting phase. Specifically, in the training phase of the GCN prediction model, a training set is first initialized, and then the GCN model is trained based on the training set to generate a prediction function. This prediction function is then used for time-series forecasting and user trajectory prediction. It should be noted that the Mean Absolute Error (MAE) is used to evaluate the model's prediction performance, as shown in the following formula:

[0113]

[0114] In the formula, Let i be the actual traffic value of the i-th test sample. Let be the predicted flow rate value for the i-th test sample; n is the total number of test samples. A reasonable range is MAE ≤ 5%. If MAE > 5%, the model bias correction module will be used for auxiliary correction. Additionally, the training data input x is the input data from the previous step, and the output y is the output data from the previous step. The training set:validation set ratio is set to 7:3.

[0115] In the application phase of the GCN prediction model, network capacity prediction values ​​can be obtained based on the input network management side performance data, base station side physical information, software configuration information, and spatiotemporal feature weights. It should be noted that the GCN prediction model has the following characteristics, including: (1) Temporal feature prediction: Based on the feature enhancement of the GCN proxy model, the mapping relationship between basic environmental variables and performance indicators is established, and the proxy model of PRB utilization and traffic, RRC maximum number of users, etc. is output. This can be achieved by calling the short-term service prediction model function. (2) User trajectory prediction: Based on the data of the first 10 nodes, the data of the last 5 nodes is predicted. The existing historical data has an accuracy of more than 90% in predicting the growth or decline trend of the next time node compared with the previous time for daily regional sites. The prediction function can be combined with the latitude and longitude information of the cell to present the geographic information. Figure 6 This is a schematic diagram illustrating the relationship between user movement trajectories and physical characteristics provided by the present invention, such as... Figure 6 As shown: the direction indicated by the arrow is the user's movement trajectory, corresponding to the user moving from base station A at time t1 to base station B at time t2, and finally arriving at base station C at time t3.

[0116] Based on any of the above embodiments Figure 7 This is a schematic diagram of the network capacity prediction device provided by the present invention, as shown below. Figure 7 As shown, the device includes:

[0117] Unit 710 is used to obtain the fusion evaluation parameters of the region to be predicted.

[0118] The prediction unit 720 inputs the fusion evaluation parameters and the correlation weights between any network management side performance parameter among the fusion evaluation parameters and each fusion evaluation parameter other than the network management side performance parameter into the capacity prediction model to obtain the network capacity prediction value of the region to be predicted output by the capacity prediction model.

[0119] The capacity prediction model is obtained by training the initial capacity prediction model based on the sample fusion evaluation parameters.

[0120] The apparatus provided in this invention inputs the correlation weights between the fusion evaluation parameters and any network management side performance parameter and each fusion evaluation parameter excluding any network management side performance parameter into the capacity prediction model. By reflecting the relationship between user trajectory changes and spatial changes through the correlation weights, the capacity prediction model is instructed to capture the relationship between user trajectory changes and spatial changes in the fusion evaluation parameters, thereby achieving fast and accurate network capacity prediction.

[0121] Figure 8 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 8As shown, the electronic device may include: a processor 810, a communications interface 820, a memory 830, and a communication bus 840, wherein the processor 810, the communications interface 820, and the memory 830 communicate with each other through the communication bus 840. The processor 810 can call logical instructions in the memory 830 to execute a network capacity prediction method, which includes: obtaining fusion evaluation parameters of the region to be predicted; inputting the fusion evaluation parameters and the correlation weights between any network management side performance parameter and each fusion evaluation parameter other than the network management side performance parameter into a capacity prediction model to obtain the network capacity prediction value of the region to be predicted output by the capacity prediction model; the capacity prediction model is obtained by training an initial capacity prediction model based on sample fusion evaluation parameters.

[0122] Furthermore, the logical instructions in the aforementioned memory 830 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0123] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the network capacity prediction method provided by the above methods. The method includes: obtaining fusion evaluation parameters of the region to be predicted; inputting the fusion evaluation parameters and the correlation weights between any network management side performance parameter and each fusion evaluation parameter other than the network management side performance parameter into a capacity prediction model to obtain the network capacity prediction value of the region to be predicted output by the capacity prediction model; the capacity prediction model is obtained by training an initial capacity prediction model based on sample fusion evaluation parameters.

[0124] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the network capacity prediction method provided by the above methods. The method includes: obtaining fusion evaluation parameters of the region to be predicted; inputting the fusion evaluation parameters and the correlation weights between any network management side performance parameter among the fusion evaluation parameters and each fusion evaluation parameter excluding the any network management side performance parameter into a capacity prediction model to obtain a network capacity prediction value of the region to be predicted output by the capacity prediction model; wherein the capacity prediction model is obtained by training an initial capacity prediction model based on sample fusion evaluation parameters.

[0125] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0126] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0127] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A network capacity prediction method, characterized in that, include: Obtain the fusion evaluation parameters for the region to be predicted; The association weights between the fusion evaluation parameters and any network management side performance parameter among the fusion evaluation parameters and each fusion evaluation parameter other than the network management side performance parameter are input into the capacity prediction model to obtain the network capacity prediction value of the region to be predicted output by the capacity prediction model. The capacity prediction model is obtained by training the initial capacity prediction model based on the sample fusion evaluation parameters; The capacity prediction model includes graph convolutional units and gated recurrent units; The step of inputting the association weights between the fusion evaluation parameters and any network management-side performance parameter among the fusion evaluation parameters and each fusion evaluation parameter excluding the network management-side performance parameter into the capacity prediction model to obtain the network capacity prediction value of the region to be predicted output by the capacity prediction model includes: The parameter features of the fusion evaluation parameters at historical moments are extracted based on the graph convolutional units and the associated weights. Based on the gated loop unit and the self-attention mechanism, the parameter features of the historical moments are processed to obtain the parameter features of the current moment; Based on the parameter characteristics at the current moment, the predicted network capacity value is obtained; The gated loop unit includes an update gate, a reset gate, a candidate hidden gate, and an update hidden gate; The step of performing feature processing on the parameter features of the historical time based on the gated recurrent unit and the self-attention mechanism to obtain the parameter features of the current time includes: Based on the update gate and the first attention weight corresponding to the update gate, the parameter features of the historical moment are preserved to obtain the first preserved features; Based on the reset gate and the second attention weight corresponding to the reset gate, the parameter features of the historical moment are discarded to obtain the second retained features; Based on the candidate hidden gate and the third attention weight corresponding to the candidate hidden gate, feature processing is performed on the parameter features of the historical moment and the second preserved features to obtain the third preserved features; Based on the updated hidden gate, feature processing is performed on the first retained feature, the second retained feature, and the third retained feature to obtain the current time parameter feature; The process of obtaining the network capacity prediction value based on the parameter characteristics at the current time includes: Based on the prediction function and the parameter characteristics at the current moment, the initial capacity prediction value is obtained; Based on the network adjustment parameters of the region to be predicted, network capacity prediction is performed on the region to be predicted to obtain the capacity adjustment prediction value. The initial capacity prediction value is corrected based on the capacity adjustment prediction value to obtain the network capacity prediction value; The network adjustment parameters include at least the number of system handovers between base stations and individual cell offset parameters.

2. The network capacity prediction method according to claim 1, characterized in that, The fusion evaluation parameters also include base station-side physical parameters and software configuration parameters; The steps for obtaining the correlation weight between any network management side performance parameter and each other fusion evaluation parameter (excluding the network management side performance parameter) in the fusion evaluation parameters include: Based on a graph convolutional network, the connection edges between any network management side performance parameter and each fusion evaluation parameter are calculated to obtain the association weights between any network management side performance parameter and each fusion evaluation parameter other than the any network management side performance parameter.

3. A network capacity prediction device, characterized in that, The apparatus performs the network capacity prediction method as described in any one of claims 1 to 2.

4. An electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the network capacity prediction method as described in any one of claims 1 to 2.

5. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the network capacity prediction method as described in any one of claims 1 to 2.

6. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the network capacity prediction method as described in any one of claims 1 to 2.

Images