Traffic suppression prediction method, electronic device, storage medium

Abstract

The application provides a traffic suppression prediction method, an electronic device and a storage medium, and the traffic suppression prediction method comprises the following steps: determining a suppression point traffic value according to a preset network traffic model, the network traffic model representing a mapping relationship between a value of a network parameter of a transmission network and a traffic value, the suppression point traffic value being a traffic threshold of the transmission network under a current operation strategy; determining a suppression reference value of a target network parameter corresponding to the suppression point traffic value; obtaining a parameter prediction value corresponding to the target network parameter, and determining a traffic suppression prediction result according to the parameter prediction value and the suppression reference value. According to the scheme provided in the embodiment of the application, the traffic suppression prediction result corresponding to the target network parameter can be predicted before traffic suppression occurs, data basis is provided for network optimization in advance, and user experience is effectively improved.

Description

Technical Field

[0001] This invention relates to, but is not limited to, the field of wireless communication technology, and particularly to a traffic suppression prediction method, an electronic device, and a storage medium. Background Technology

[0002] With the development of internet and communication technologies, wireless networks have become indispensable in daily life and work, bringing huge traffic demands to operators. However, the maximum traffic capacity of each cell is usually limited by the number of users, available resources, and hardware and software capabilities. When traffic reaches its maximum, increasing overhead to meet traffic demand not only fails to increase the cell's total traffic but may even lead to a decrease in traffic; this phenomenon is called traffic suppression. However, different cells have different operating strategies and working conditions, and currently there are no predictive methods for traffic suppression. Network diagnosis can only be performed after traffic suppression is detected, followed by adjustments to operating strategies to improve network conditions. This process has a certain lag and affects user experience. Summary of the Invention

[0003] The following is an overview of the subject matter described in detail herein. This overview is not intended to limit the scope of the claims.

[0004] This invention provides a traffic suppression prediction method, an electronic device, and a storage medium, which can predict traffic suppression, provide data for pre-optimization of the network, and thus improve user experience.

[0005] In a first aspect, embodiments of the present invention provide a traffic suppression prediction method, comprising:

[0006] The suppression point traffic value is determined based on a preset network traffic model, whereby the network traffic model represents the mapping relationship between the values ​​of network parameters of the transmission network and the traffic value, and the suppression point traffic value is the traffic threshold of the transmission network under the current operating strategy.

[0007] Determine the suppression reference value for the target network parameter corresponding to the suppression point flow value;

[0008] Obtain the parameter prediction values ​​corresponding to the target network parameters, and determine the traffic suppression prediction result based on the parameter prediction values ​​and the suppression reference values.

[0009] In a second aspect, embodiments of the present invention provide 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 computer program to implement the traffic suppression prediction method as described in the first aspect.

[0010] This invention includes: determining a suppression point traffic value based on a preset network traffic model, wherein the network traffic model characterizes the mapping relationship between the values ​​of network parameters of the transmission network and traffic values, and the suppression point traffic value is a traffic threshold of the transmission network under the current operating strategy; determining a suppression reference value for the target network parameter corresponding to the suppression point traffic value; obtaining a parameter prediction value for the target network parameter; and determining a traffic suppression prediction result based on the parameter prediction value and the suppression reference value. According to the solution provided by this invention, a suppression point traffic value can be determined based on a trained network traffic model, providing a data foundation for traffic suppression prediction. This allows for the prediction of a traffic suppression prediction result corresponding to the target network parameter before traffic suppression occurs, providing data for pre-optimization of the network and effectively improving user experience.

[0011] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the description, claims, and drawings. Attached Figure Description

[0012] The accompanying drawings are provided to further understand the technical solutions of the present invention and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the technical solutions of the present invention, and do not constitute a limitation on the technical solutions of the present invention.

[0013] Figure 1 This is a flowchart of a traffic suppression prediction method provided in one embodiment of the present invention;

[0014] Figure 2 This is a flowchart of training a network traffic model based on historical data, provided in another embodiment of the present invention;

[0015] Figure 3 This is a flowchart of training a network traffic model provided in another embodiment of the present invention;

[0016] Figure 4 This is a flowchart of a rasterized network parameter sample set provided in another embodiment of the present invention;

[0017] Figure 5 This is a flowchart for determining the flow rate value at the inhibition point provided in another embodiment of the present invention;

[0018] Figure 6 This is a flowchart of data preprocessing provided in another embodiment of the present invention;

[0019] Figure 7 This is a flowchart of obtaining parameter prediction values ​​provided in another embodiment of the present invention;

[0020] Figure 8 This is a flowchart of a training parameter prediction model provided in another embodiment of the present invention;

[0021] Figure 9 This is a flowchart for determining the degree of traffic suppression provided in another embodiment of the present invention;

[0022] Figure 10 This is a flowchart for determining root cause network parameters provided in another embodiment of the present invention;

[0023] Figure 11 This is a flowchart of the history acquisition provided in another embodiment of the present invention;

[0024] Figure 12 This is a structural diagram of a network traffic management system provided in another embodiment of the present invention;

[0025] Figure 13 This is a flowchart of Example 1 provided in another embodiment of the present invention;

[0026] Figure 14 This is a schematic diagram of a one-dimensional mapping relationship of a network traffic model provided in another embodiment of the present invention;

[0027] Figure 15 This is a flowchart of Example 2 provided in another embodiment of the present invention;

[0028] Figure 16 This is a schematic diagram of the two-dimensional mapping relationship of the network traffic model provided in another embodiment of the present invention;

[0029] Figure 17 This is a flowchart of Example 3 provided in another embodiment of the present invention;

[0030] Figure 18 This is a structural diagram of an electronic device provided in another embodiment of the present invention. Detailed Implementation

[0031] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0032] It should be noted that although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first," "second," etc., in the specification, claims, or the aforementioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0033] This invention provides a traffic suppression prediction method, an electronic device, and a storage medium. The traffic suppression prediction method includes: determining a suppression point traffic value based on a preset network traffic model, wherein the network traffic model characterizes the mapping relationship between the values ​​of network parameters of the transmission network and traffic values, and the suppression point traffic value is a traffic threshold of the transmission network under the current operating strategy; determining a suppression reference value of the target network parameter corresponding to the suppression point traffic value; obtaining a parameter prediction value of the target network parameter; and determining a traffic suppression prediction result based on the parameter prediction value and the suppression reference value. According to the solution provided by the embodiments of this invention, a traffic suppression prediction result corresponding to the target network parameter can be predicted before traffic suppression occurs, providing data basis for pre-optimization of the network and effectively improving user experience.

[0034] The embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0035] like Figure 1 As shown, Figure 1 This is a flowchart of a traffic suppression prediction method provided in an embodiment of the present invention. The traffic suppression method includes, but is not limited to, steps S110, S120 and S130.

[0036] Step S110: Determine the suppression point traffic value according to the preset network traffic model. The network traffic model represents the mapping relationship between the network parameters of the transmission network and the traffic value. The suppression point traffic value is the traffic threshold of the transmission network under the current operating strategy.

[0037] It is worth noting that the network traffic model can be a pre-trained offline model or an online model trained after data collection. It should be noted that when the network traffic model is trained online, a threshold for the number of training iterations can be preset. When the number of training failures reaches this threshold, subsequent steps are stopped. This prevents erroneous traffic suppression predictions due to an inaccurate network traffic model. Those skilled in the art are familiar with how to adjust the training threshold to ensure the network traffic model functions correctly, and will not be elaborated upon here.

[0038] It is understood that network parameters can be any parameters that can affect cell traffic, such as the number of connected users, the number of active users, traffic channel resource utilization, control channel resource utilization, spectrum efficiency (SE), cell traffic, average modulation and coding scheme (MCS), average channel quality indicator (CQI), average signal-to-interference plus noise ratio (SINR), etc. This embodiment does not impose any limitations on these parameters.

[0039] It should be noted that the network traffic model can be the law governing network traffic and network parameters. Since the suppression point represents the inflection point where traffic begins to decrease, the traffic threshold can be the average maximum traffic value that the transmission network can achieve. Therefore, one or more of the aforementioned network parameters can be selected for traffic suppression prediction. When selecting one network parameter for traffic prediction, such as the number of users, the correspondence between the number of active users and network traffic can be determined according to the network traffic model. The average maximum traffic value that the number of active users can achieve under different values ​​can be determined, and this average maximum traffic value is the suppression point traffic value. For example, when selecting multiple network parameters for traffic prediction, taking the number of users, service channel resource utilization, control channel resource utilization, and channel quality as examples, the average maximum traffic value that the cell can achieve under different values ​​of the above four network parameters can be obtained, and this average maximum traffic value is the suppression point traffic value. At this time, the suppression point traffic value is simultaneously associated with the four network parameters, thereby determining the maximum traffic that the cell can achieve under the influence of the above four network parameters.

[0040] It is understandable that, as the cell network adopts different strategies for network optimization based on the actual situation during operation, the suppression point traffic value of the cell network will be different under different operating strategies. That is, the suppression point will shift with different operating strategies. Therefore, in order to improve the accuracy of traffic suppression prediction, it is necessary to obtain the suppression point traffic value for the current operating strategy.

[0041] Step S120: Determine the suppression reference value of the target network parameter corresponding to the suppression point flow value.

[0042] It should be noted that, as described above, the number and types of target network parameters can be arbitrary. Therefore, in order to predict traffic suppression, it is necessary to determine the suppression reference values ​​for all target network parameters corresponding to the suppression point traffic value. For example, when the target network parameter is one type, taking the number of users as an example, it is necessary to determine the suppression reference value corresponding to the number of users. When the target network parameter includes the four types mentioned above, it is necessary to determine the suppression reference values ​​for the number of users, service channel resource utilization, control channel resource utilization, and channel quality respectively when the suppression point traffic value is reached, so as to provide an accurate data basis for traffic suppression prediction.

[0043] Step S130: Obtain the predicted values ​​of the target network parameters, and determine the traffic suppression prediction result based on the predicted values ​​and the suppression reference values.

[0044] It should be noted that the predicted parameter values ​​can be obtained through model prediction. For example, a parameter prediction model can be pre-set to obtain the predicted parameter values ​​that characterize the future traffic status of the cell. It is understood that when the predicted parameter values ​​are obtained through the parameter prediction model, the input data can be either a time series feature vector at a certain time granularity or a scalar representing the current state. For example, using a 15-minute granular data sequence from the past week as the feature vector as the input to the prediction model, the parameter values ​​of the cell at a 15-minute granularity for the next day can be predicted. Those skilled in the art are motivated to select a specific time dimension according to actual needs, and this embodiment does not impose any limitations on this.

[0045] Understandably, since the suppression point traffic value and suppression reference value have already been obtained through the network traffic model, after obtaining the parameter prediction value, it is possible to directly compare the parameter prediction value with the suppression reference value to obtain the traffic suppression prediction result. For example, if there is only one type of target network parameter, when the parameter prediction value is greater than the suppression reference value, it can be determined that as the value of the target network parameter increases, the network traffic is limited and less than the suppression point traffic value, resulting in traffic suppression. Alternatively, if the target network parameter is one of the four types mentioned above, the relationship model between the number of users and network traffic can be obtained through cross-sections using three parameters: service channel resource utilization, control channel resource utilization, and channel quality. Then, based on the current state of the network's service channel resource utilization, control channel resource utilization, and channel quality, the inflection point model currently in use is determined. A single-parameter method is used to determine whether suppression has occurred, and the corresponding suppression amount when suppression occurs is obtained for subsequent prediction.

[0046] It is worth noting that, in order to improve the accuracy of prediction, the parameter prediction model and the network traffic model can be trained using the same sample set. The corresponding training set and test set can be obtained from the sample set according to the specific model type selected. This embodiment does not impose any restrictions on this.

[0047] Additionally, refer to Figure 2 In one embodiment, during execution Figure 1 Before step S110 in the illustrated embodiment, the following steps may also be included, but are not limited to:

[0048] Step S210: Obtain a network parameter sample set from historical data of the transmission network. The network parameter sample set includes samples of network parameters.

[0049] Step S220: Train the network traffic model based on the network parameter sample set.

[0050] It is worth noting that since the network traffic model represents the relationship between network parameters and network traffic, obtaining a sample set of network parameters from historical data of the transmission network and training the network traffic model allows the model to reflect the historical operating state of the transmission network. When traffic suppression occurs during historical operation, the mathematical properties of the network traffic model can be used to quickly determine the suppression point traffic value. For example, the average maximum traffic value that occurred during historical operation can be used as the suppression point traffic value, improving prediction efficiency. Furthermore, assuming the operating strategy remains unchanged, suppression points in the transmission network are usually located approximately in the same area. Suppression points from historical operation can be used to determine the current network suppression points. Therefore, compared to training the network traffic model to predict suppression point traffic values, training the network traffic model using historical data and then using the model's mathematical properties to determine the suppression points effectively improves the accuracy of suppression points, thereby improving the accuracy of traffic prediction.

[0051] It should be noted that those skilled in the art are familiar with how to obtain historical data of transmission networks, such as by collecting historical data from devices such as the network management system, control network elements, or data service modules of the community network. This embodiment will not elaborate on this further.

[0052] Additionally, refer to Figure 3 In one embodiment, Figure 2 Step S220 in the illustrated embodiment also includes, but is not limited to, the following steps:

[0053] Step S310: Obtain the first training set and the first test set based on the network parameter sample set;

[0054] Step S320: Normalize the first training set and the first test set;

[0055] Step S330: Obtain the network traffic model, train the network traffic model based on the first training set, and verify the trained network traffic model based on the first test set.

[0056] It should be noted that before using the network parameter sample set, data cleaning and other preprocessing can be performed on the network parameter sample set to improve the integrity and usability of the data. This embodiment will not elaborate on this.

[0057] It should be noted that those skilled in the art are capable of using different strategies to divide the samples in the network parameter sample set into a first training set and a first test set according to actual needs, and this embodiment does not impose any limitations on this.

[0058] It is worth noting that the normalization of the first training set and the first test set can be performed using the min-max normalization method or other alternative normalization methods. Those skilled in the art are familiar with how to use the above methods to normalize the sample set, and this embodiment will not elaborate on this further.

[0059] It should be noted that the network traffic model can be a common machine learning model, such as a fully connected neural network model. Of course, machine learning models such as decision trees and regression models can also be used. Those skilled in the art are motivated to select the specific model type according to the actual situation. Taking a fully connected neural network model as an example, the network traffic model includes one input layer, one output layer, and multiple hidden layers. The input layer is used to input network parameters from the network parameter sample set, and the output layer is used to output the prediction result, such as SE or cell traffic. Furthermore, those skilled in the art are familiar with how to adjust the number of hidden layers and neurons according to the actual situation so that the trained model can meet the requirements. This embodiment does not impose any limitations on this.

[0060] It is understandable that since the first training set and the first test set have been normalized, after training and testing the network traffic model, it is necessary to perform inverse normalization on the output of the tested network traffic model to obtain the final output of the network traffic model. The specific inverse normalization method corresponds to the normalization method described above, and this embodiment will not elaborate on it further.

[0061] Additionally, refer to Figure 4 In one embodiment, Figure 3 Step S310 in the illustrated embodiment also includes, but is not limited to, the following steps:

[0062] Step S410: Determine the numerical range of the network parameter sample set;

[0063] Step S420: Divide the numerical range into several grid intervals according to the preset grid step size;

[0064] Step S430: Fill each sample of the network parameters into the raster interval according to the principle of numerical correspondence to obtain the raster dataset;

[0065] Step S440: Select the first training set and the first test set from the raster dataset according to the preset filtering strategy.

[0066] It should be noted that determining the numerical range of the network parameter sample set can be achieved by determining the maximum value of the network parameters, which will not be elaborated upon here. If multiple types of network parameters are involved, the maximum value of each type of network parameter can be determined separately, and the same value should be used for each type.

[0067] It is worth noting that the specific value of the grid step size can be determined according to actual needs. This embodiment does not impose any restrictions on this. For example, if the network parameter is the number of users, and its value range is [0, UserNumMax], where UserNumMax is the maximum value of the number of users, and the preset grid step size is UserNumMeshStep, then the value range can be divided into UserNumMeshNum grids, where UserNumMeshNum = (UserNumMax / UserNumMeshStep). If there are multiple types of network parameters, the corresponding grid ranges can be obtained by referring to the above method, which will not be elaborated here.

[0068] It should be noted that since the grid interval is actually a range of values, the sample values ​​of each network parameter can be used as keys to fill the corresponding grid interval with the sample values ​​of each network parameter.

[0069] It should be noted that those skilled in the art are motivated to select specific screening strategies based on actual circumstances. The following example illustrates a specific screening strategy: Set a grid threshold value MeshSampleMin. For each grid with more than 2×MeshSampleMin data, the average of half of the samples randomly selected from the grid data is used as the first training set for that grid, and the average of the remaining half of the samples is used as the first test set for that grid. For grids with less than MeshSampleMin data, the grid data is discarded. For grids with data between MeshSampleMin and 2*MeshSampleMin, the average data of that grid is used as the first test set.

[0070] Additionally, refer to Figure 5 In one embodiment, Figure 1 Step S120 in the illustrated embodiment also includes, but is not limited to, the following steps:

[0071] Step S510: Obtain a preset threshold value, determine the first difference of the network parameter samples, and determine the network parameter samples whose first difference is less than the threshold value as target samples.

[0072] Step S520: Determine the target flow value corresponding to the target sample based on the mapping relationship, and set the target flow value as the suppression point flow value.

[0073] Those skilled in the art will understand that the first-order difference is the difference between two consecutive adjacent terms in a discrete function, while the suppression point flow rate is the average maximum flow rate that the current transmission network can achieve. For the transmission network, the flow rate initially increases with the increase of network parameters, slows down before reaching the suppression point, stops increasing after reaching the suppression point, and then decreases with the increase of network parameters. Therefore, theoretically, when the first-order difference between two adjacent samples is zero, it can be determined that flow suppression has occurred, and the corresponding flow rate is the suppression point flow rate. However, in practical applications, only a limited number of samples can be collected, and it is difficult to ensure that at least two samples located in the flow suppression region are collected. Therefore, a small threshold value can be set. When the first-order difference between two samples is less than the threshold value, the sample can be considered to be close to the suppression point. The specific value of the threshold value can be adjusted according to the actual number of samples. This embodiment does not impose any limitations on this.

[0074] It should be noted that when multiple types of network parameters are involved, such as the number of users, service channel resource utilization, control channel resource utilization and channel quality in the example above, step S510 can be performed separately for each network parameter to determine the target sample, which will not be elaborated here.

[0075] It should be noted that since the network traffic model can characterize the mapping relationship between the values ​​of network parameters and network traffic, once the target sample is determined, the corresponding target traffic value can be determined through the value of the target sample, and then further determined as the suppression point traffic value. This will not be elaborated on here.

[0076] Additionally, refer to Figure 6 In one embodiment, Figure 2 Step S210 in the illustrated embodiment also includes, but is not limited to, the following steps:

[0077] Step S610: Obtain the preset acquisition granularity and acquisition period, wherein the acquisition granularity represents the frequency of acquiring network parameters within one acquisition period;

[0078] Step S620: Collect samples of network parameters from historical data according to the collection granularity and collection cycle;

[0079] Step S630: Perform data preprocessing on the collected network parameter samples to obtain a network parameter sample set.

[0080] It should be noted that the specific values ​​of the collection granularity and collection cycle can be determined according to actual needs. For example, if the collection granularity is one minute and the collection cycle is one day, then network parameters can be collected once per minute to obtain samples. Those skilled in the art will have the motivation to adjust according to actual needs, and no further restrictions will be imposed here.

[0081] It is worth noting that after the sample collection is completed, the dataset can be preprocessed, such as data timeline completion, data timeline deduplication, and data completion. Those skilled in the art are motivated to add or reduce preprocessing operations according to the actual situation.

[0082] Understandably, data timeline completion can resolve situations where data is missed at certain granularities during the acquisition process. If some data information on the timeline is missing, the corresponding data on the timeline can be filled in to avoid data acquisition failures from the dataset, which could affect subsequent operations.

[0083] It is understandable that deduplication through the data timeline can solve the problem of duplicate data collection at certain granularities during the data acquisition process. The deduplication rule can be to keep the first occurrence of the data in the dataset and delete the duplicate data thereafter. The specific rule can be selected according to the actual situation.

[0084] Understandably, data completion can resolve data gaps during the data collection process. Specifically, if the missing data is at the beginning of the dataset, empty data at the beginning can be filled with the first non-empty data point from the beginning. If the missing data is at the end of the dataset, empty data at the end can be filled with the first non-empty data point from the end. If the missing data is in the middle of the dataset, the first non-empty data point can be found both forward and backward, and linear interpolation can be used to fill the gap. Alternatively, mean-based filling can be used, depending on the specific needs.

[0085] Additionally, refer to Figure 7 In one embodiment, Figure 1 Step S130 in the illustrated embodiment also includes, but is not limited to, the following steps:

[0086] Step S710: Determine the acquisition time for each acquisition cycle based on the acquisition granularity;

[0087] Step S720: Obtain the prediction sample set from the network parameter sample set. The sample data of the prediction sample set includes the acquisition time and the parameter values ​​of the network parameters corresponding to the acquisition time.

[0088] Step S730: Train the parameter prediction model based on the prediction sample set.

[0089] Step S740: Obtain the predicted feature samples corresponding to the target network parameters, and input the predicted feature samples into the parameter prediction model to obtain the parameter prediction values.

[0090] It should be noted that the granularity of data collection usually represents the frequency of data collection. The collection time of the collection cycle is determined based on the collection frequency. For example, the collected prediction sample set can be referred to in the form of Table 1. Table 1 shows the sample data corresponding to one network parameter. For multiple network parameters, the data corresponding to the collection time can be added accordingly, which will not be repeated here. In Table 1, N is the data length of the network parameter sample set. Each row in the table is a sample of one network parameter, where data_i represents the parameter value of the i-th sample, and time_i is the collection time of data_i.

[0091] time time_0 time_1 time_2 time_i time_N-2 time_N-1 state_data data_0 data_1 data_2 ... data_i ... data_N-2 data_N-1

[0092] Table 1 Example of Prediction Sample Set Data

[0093] Understandably, the sample data in the prediction sample set can be K+L+M+1 dimensional, where K represents the parameter values ​​at the same acquisition time in the previous K periods, L represents the parameter values ​​at the previous L times, and M represents the time information of the current data, such as day of the week, hour, minute, and whether it is a holiday. Simultaneously, the first K+L+M dimensions of each sample represent the features of that sample, and the last dimension is the state value at the current time, used to represent the label of that sample, facilitating the differentiation between different samples. It is also understood that the first K+L dimensions of each sample can include all types of network parameters, which will not be elaborated upon here.

[0094] It should be noted that the predicted feature samples can be obtained by selecting the time dimension. For example, if it is necessary to predict the parameter prediction values ​​for the next K periods and L times, then K+L+M dimensional data can be obtained from the prediction sample set as the predicted feature samples. Those skilled in the art are motivated to adjust the dimensions of the predicted feature samples according to the actual prediction needs, and no further restrictions will be imposed here.

[0095] Additionally, refer to Figure 8 In one embodiment, Figure 7 Step S730 in the illustrated embodiment also includes, but is not limited to, the following steps:

[0096] Step S810: Normalize the data for the prediction sample set;

[0097] Step S820: Divide the normalized prediction sample set into a second training set and a second test set according to the preset segmentation ratio.

[0098] Step S830: Obtain the initial parameter prediction model and determine the training parameters of the initial parameter prediction model;

[0099] Step S840: Train the initial parameter prediction model based on the second training set, and verify the trained initial parameter prediction model based on the second test set.

[0100] Step S850: The initial parameter prediction model after verification is determined as the parameter prediction model.

[0101] It should be noted that after obtaining the prediction sample set, when multiple network parameters are involved, the data of each sample in the prediction sample set can be normalized. For example, the collection time, number of users, service channel resource utilization, control channel resource utilization and channel quality of a sample can be normalized. Then, the normalized prediction sample set can be divided by a pre-set segmentation ratio to obtain the second training set and the second test set.

[0102] It is worth noting that the initial parameter prediction model can be a Long Short-Term Memory (LSTM) neural network, or other alternative machine learning models such as decision trees and regression models. Those skilled in the art are motivated to select a specific model based on actual needs, which will not be elaborated here.

[0103] Understandably, when initializing the model, the neural network weights and biases can be randomly initialized, an activation function can be selected, and the maximum number of training iterations and model error of the state prediction model can be defined, where the model error is defined as the mean absolute error. The absolute error between the predicted parameter value `state_predict` and the true value `state_true` can be obtained using the following formula:

[0104] Where N1 is the output feature dimension and N2 is the number of samples.

[0105] It should be noted that after obtaining the predicted feature samples, the feature prediction samples can also be normalized. Furthermore, the normalization parameters are the same as the normalization parameters of the predicted sample set, and the normalization method used is also the same to ensure data consistency.

[0106] It is worth noting that once the specific model and parameters are determined, those skilled in the art are well aware of how to train and test the model using a second training set and a second test set, so we will not elaborate further here.

[0107] Additionally, refer to Figure 9 In one embodiment, Figure 2 Step S210 in the illustrated embodiment also includes, but is not limited to, the following steps:

[0108] Step S910: Determine the target historical data, which is the historical data obtained by the transmission network according to the latest operating strategy;

[0109] Step S920: Obtain network parameter samples from target historical data.

[0110] It should be noted that after traffic suppression occurs, the decision-making module typically determines the root cause of the suppression based on the characteristic data of the suppression point, thereby deriving a network optimization strategy to improve cell traffic. Therefore, when the operating strategy changes, the suppression point will also change accordingly. Network traffic models and parameter prediction models trained on historical data from previous operating strategies cannot accurately predict the suppression point under the current operating strategy. Therefore, when a change in the operating strategy is detected, the historical data of the latest operating strategy can be identified as the target historical data, and network parameter samples can be re-obtained from the target historical data.

[0111] It is understandable that when the network parameter samples change, the previously obtained network traffic model and parameter prediction model need to be retrained and tested to ensure the accuracy of the prediction. The training and testing methods of the model will not be repeated here.

[0112] It is worth noting that the network traffic model and parameter prediction model can be updated in response to changes in operating strategies, or by pre-setting timers and update cycles, thereby regularly updating the model with the latest data and effectively improving the accuracy of the network traffic model and parameter prediction model.

[0113] Additionally, refer to Figure 10 In one embodiment, after execution Figure 1 Following step S130 in the illustrated embodiment, the following steps may also be included, but are not limited to:

[0114] Step S1010: When the traffic suppression prediction result indicates that traffic suppression has occurred, the parameter prediction value is input into the network traffic model to obtain the traffic prediction value.

[0115] Step S1020: Determine the degree of flow suppression based on the predicted flow value and the suppression reference value.

[0116] It should be noted that, with the operating strategy unchanged, the relationship between the parameter prediction values ​​and the traffic values ​​also conforms to the network traffic model. Therefore, by using the mapping relationship of the network traffic model, the traffic prediction value corresponding to the parameter prediction value can be determined.

[0117] It should be noted that while analyzing the degree of traffic suppression, the traffic loss situation can also be determined according to actual needs, such as the absolute amount of traffic loss and the relative degree of traffic loss. The absolute amount of traffic loss is Max(CellBestThp-CellPredictThp, 0), where CellBestThp is the traffic value at the suppression point and CellPredictThp is the predicted traffic value; the relative degree of traffic loss is Max(CellBestThp-CellPredictThp, 0) / CellBestThp. Those skilled in the art are also motivated to select other parameters that can characterize traffic loss according to actual needs, which will not be limited here.

[0118] It should be noted that if the network parameter sample set involves only one type of network parameter, then the analysis can be performed only on that network parameter. If multiple network parameters are involved, then... Figure 1 The parametric cross-plane method described in the embodiment shown in step S130 is used to obtain the relationship between the number of users and network traffic.

[0119] In addition, in one embodiment, the network traffic model includes at least two different types of target network parameters, referring to... Figure 11 After execution Figure 10 Following step S1020 in the illustrated embodiment, the following steps may also be included, but are not limited to:

[0120] Step S1111: Determine the degree of traffic suppression corresponding to each target network parameter;

[0121] Step S1120: Determine the root cause network parameter that causes traffic suppression. The root cause network parameter is the target network parameter corresponding to the highest level of traffic suppression.

[0122] It should be noted that if multiple different target network parameters are involved, the degree of influence of each target network parameter on traffic is different. Therefore, in order to obtain a more accurate network optimization strategy, it is necessary to determine the target network parameter with the greatest impact on traffic suppression. The following uses four target network parameters as examples to illustrate the technical solution of this embodiment. The four target network parameters are the number of users, service channel resource utilization, control channel resource utilization, and channel quality.

[0123] The network traffic model is sliced ​​based on the suppression reference values ​​to obtain the first component slice of the suppression reference values ​​for each target network parameter. Since traffic values ​​typically rise to a maximum and then fall, the average of the first-order differences of the first component slice can be obtained and denoted as the first average. For example, the first half of the data in each first component slice can be obtained, and the specific data range can be determined based on the actual data distribution. The network traffic model is then sliced ​​based on the traffic prediction values ​​to obtain the second component slice of the traffic prediction values ​​for each target network parameter. The second average is obtained in the same way as the first component slice. The degree of suppression is calculated based on the average of the first-order differences. Taking the number of users as an example, the degree of suppression of the number of users satisfies the following relationship:

[0124] UserNum_inhibition_ratio=(diff_UserNum_best-diff_UserNum) / diff_UserNum_best,

[0125] Where UserNum_inhibition_ratio represents the degree of inhibition, diff_UserNum_best represents the first average value, and diff_UserNum represents the second average value.

[0126] It is worth noting that once the degree of traffic suppression corresponding to each target network parameter is determined, the greater the degree of traffic suppression, the higher the degree of suppression, meaning that traffic suppression is more likely to occur. Therefore, the root cause network parameters can be determined based on the value of the degree of traffic suppression, thereby identifying the main factors of traffic suppression. Optimizing the network based on these main factors can alleviate traffic suppression to a greater extent.

[0127] In another embodiment, the suppression point traffic value is the average maximum traffic value of the transmission network under the current operating strategy.

[0128] It should be noted that, according to the description of the suppression point in the above embodiments, the traffic threshold can be the average maximum traffic value of the transmission network under the current operating strategy. For example, it can be obtained by training the network traffic model to obtain multiple maximum traffic values, and then calculate the average maximum traffic value based on the multiple maximum traffic values, thereby improving the predictive reference value of the suppression point traffic value.

[0129] Additionally, refer to Figure 12 The present invention also provides a network traffic management system, including a data processing module 1210, a network traffic model 1220, a parameter prediction model 1230, and a traffic suppression analysis module 1240.

[0130] The data processing module 1210 is used to acquire data to obtain a network parameter sample set, and also to perform data preprocessing, including but not limited to data deduplication and data completion. The methods for data deduplication and data completion can be found in [reference needed]. Figure 6 The description of the illustrated embodiments will not be repeated here.

[0131] Network traffic model 1220 is used to divide the network parameter sample set into grids and obtain the first training set and the second test set according to the selection strategy. For specific principles and methods, please refer to [link / reference]. Figure 4 The illustrated embodiment describes how, after initializing the model, iterative training and model validation are performed to obtain a trained network traffic model. The specific principles and methods can be found in [reference needed]. Figure 3 The description of the illustrated embodiments will not be repeated here.

[0132] The parameter prediction model 1230 is used to segment the preprocessed data to obtain a second training set and a second test set. The initialization, training, and validation of the parameter prediction model are then performed using these two sets. For details, please refer to [reference needed]. Figure 8 The embodiment shown is described; furthermore, it is used to obtain future parameter prediction values ​​for a cell after inputting prediction feature samples, as detailed in the following reference. Figure 7 Description of the illustrated embodiment.

[0133] The flow suppression analysis module 1240 is used to determine the flow suppression results, and to determine the degree of suppression and identify the cause of suppression when flow suppression occurs. For details, please refer to [reference needed]. Figure 10 and Figure 11 The description of the illustrated embodiments will not be repeated here.

[0134] In addition, to better illustrate the technical solution of the present invention, three specific examples are presented below.

[0135] Example 1: In this example, a one-dimensional modeling approach is used, employing only one network parameter (e.g., the number of users). The network traffic model and parameter prediction model are trained online, with parameter predictions obtained through model prediction. Both models are updated periodically. Figure 13 This example includes, but is not limited to, the following steps:

[0136] Step S1310: Determine that both the network traffic model and the parameter prediction model are generated online. Check if the number of repeated executions due to training failure of the network traffic model has reached the first training threshold. If the threshold has not been reached, proceed to step S1320; otherwise, end the process.

[0137] Step S1321: Collect data according to the collection granularity and collection cycle, and preprocess the data.

[0138] It should be noted that the collected data is historical data from modules such as network management, network elements, or data services, including the number of RRC connected users or active users and cell traffic. Preprocessing includes data time extraction and completion, data time extraction deduplication, and data completion.

[0139] Online data timeline completion is performed. During the data collection process, there may be omissions in certain granularities. If some data information on the timeline is missing, the timeline will be filled in, and the corresponding data will be empty.

[0140] Online data timeline deduplication is performed. During the data collection process, duplicate data may be collected at certain granularities. It is necessary to deduplicatize the duplicate data. The deduplication rule is to keep the first occurrence of the data in the dataset and delete the duplicate data thereafter.

[0141] Online data completion is performed. If the missing data is at the beginning of the dataset, the empty data at the beginning is filled with the first non-empty data starting from the beginning. If the missing data is at the end of the dataset, the empty data at the end is filled with the first non-empty data starting from the end. If the missing data is in the middle of the dataset, the first non-empty data is searched forward and backward respectively, and linear interpolation is performed to fill the missing data. The filled dataset is the network parameter sample set.

[0142] Step S1322: Obtain the maximum number of users in the network parameter sample set, divide the user count grid into several grids with the grid step size, and deliver the samples in the network parameter sample set to the corresponding user count grids.

[0143] It should be noted that the maximum number of users is UserNumMax, the number of user grids is [0, UserNumMax], the grid step size is UserNumMeshStep, and the number of grids is UserNumMeshNum. Wherein, UserNumMeshNum = Ceil(UserNumMax / UserNumMeshStep), and then the data in the network parameter sample set is delivered to the corresponding grid according to the number of users.

[0144] Step S1323: Normalize the data in the user count grid and obtain the first training set and the first test set according to the filtering strategy.

[0145] It should be noted that a threshold value (MeshSampleMin) for the number of valid rasters can be preset. When the number of data points for a user count raster is greater than 2 × MeshSampleMin, the average of half of the samples randomly selected from that user count raster is used as the candidate training set for that raster, and the average of the remaining half is used as the candidate test set for that raster. When the number of data points for a user count raster is less than MeshSampleMin, that raster is discarded. When the number of data points for a user count raster is between MeshSampleMin and 2 × MeshSampleMin, the average of that raster's data is used as the candidate training set. Then, online normalization is performed on both the candidate training set and the candidate test set to obtain the first training set and the first test set.

[0146] Step S1324: Complete the training and testing of the network traffic model based on the first training set and the first test set.

[0147] It should be noted that the network traffic model is constructed as a neural network model and is randomly initialized. The network traffic model is trained using the first training set. After training to the maximum number of training iterations, the model is evaluated. If the training error meets the requirements, the model is tested using the first test set. If the test error meets the requirements, the network traffic model is output. Otherwise, the number of training failures of the network traffic model is incremented by 1 and step S1310 is executed.

[0148] Step S1330: When the update timer of the network traffic model reaches the update cycle of the network traffic model, execute step S1310; otherwise, execute step S1331.

[0149] Step S1331: Determine the suppression point traffic value and the number of suppressed users based on the network traffic model.

[0150] refer to Figure 14 , Figure 14 This is a schematic diagram of the mapping relationship of the network traffic model. The horizontal axis represents one of the network parameters, such as the number of users, service channel resource utilization, control channel resource utilization, or channel quality. The vertical axis represents the value of cell traffic or cell SE. It can be seen that as the horizontal axis increases from the origin, the cell traffic or cell SE gradually increases, and after reaching the suppression point, it begins to decrease, resulting in traffic suppression.

[0151] Step S1341: If the number of repetitions for training failure of the parameter prediction model has not reached the second training threshold, proceed to S1342; otherwise, obtain the real-time parameter value of the number of users, and use the filtered value of the real-time parameter value as the prediction parameter value to proceed to S1350.

[0152] Step S1342: Construct a prediction sample set on the network parameter sample set based on the time series, and obtain the second training set and the second test set based on the prediction sample set.

[0153] It should be noted that each initial data set includes the collection time and the corresponding number of users. Each row of the sample data set is a prediction sample. Each prediction sample has a dimension of K+L+M+1, where K is the user value at the same time in the previous K periods, L is the user value at the previous L times, M is the time information of the current data, including weekday, hour, minute, etc., and the last dimension is the user value at the current time, which is used to represent the label of the prediction sample.

[0154] It should be noted that after obtaining the prediction sample set, the prediction sample set can be normalized to each column of the prediction sample set by the normalization parameter. The normalized data is then divided into a second training set and a second test set according to the training set and test set splitting ratio parameter. The parameter prediction model is then trained and tested using the second training set and the second test set. If the training and testing errors meet the preset parameters, the parameter prediction model is output. Otherwise, the number of training failures of the parameter prediction model is incremented by 1 and step S1341 is executed.

[0155] Step S1350: Obtain the parameter prediction values ​​and perform traffic suppression analysis based on the future time state prediction values ​​output by the parameter prediction model.

[0156] It should be noted that before obtaining the parameter prediction values, it is also necessary to load the parameter prediction model and start the update timer of the parameter prediction model. When the update timer of the parameter prediction model reaches the update cycle, step S1310 is executed.

[0157] It should be noted that, in order to obtain the predicted parameter values, the feature data of the predicted feature samples can be normalized using normalization parameters. If there are data values ​​greater than 1 after normalization, they should be uniformly updated to 1. The predicted feature samples are then input into the parameter prediction model to obtain the intermediate predicted values ​​for future periods. The intermediate predicted values ​​are then subjected to an inverse normalization transformation to obtain the future predicted parameter values.

[0158] It should be noted that traffic suppression analysis can determine whether traffic will be suppressed and the degree of suppression. If the parameter prediction value is less than or equal to the suppression reference value, then the traffic will not be suppressed in the future for that period; otherwise, it will be suppressed in the future for that period. It can also determine the degree of traffic loss. For example, by inputting the parameter prediction value into the network traffic model, the predicted traffic value for the cell is CellPredictThp. Then, the absolute amount of traffic loss is Max(CellBestThp - CellPredictThp, 0), where CellBestThp is the traffic value at the suppression point, and the relative degree of traffic loss is Max(CellBestThp - CellPredictThp, 0) / CellBestThp.

[0159] Example 2: In this example, a multi-dimensional modeling approach is used, employing multiple types of network parameters, such as the number of users, service channel resource utilization, control channel resource utilization, and channel quality. Simultaneously, the network traffic model and parameter prediction model utilize offline models; the predicted parameter values ​​are obtained through model prediction. (Refer to...) Figure 15 This example includes, but is not limited to, the following steps:

[0160] Step S1510: Determine that both the network traffic model and the parameter prediction model are generated offline.

[0161] Step S1521: Collect data according to the collection granularity and collection cycle, and preprocess the data.

[0162] It should be noted that data preprocessing is performed offline, including data timeline completion. During the data acquisition process, there may be some omissions in certain granularities. If some data information on the timeline is missing, the timeline will be filled in, and the corresponding data will be empty.

[0163] It should be noted that data preprocessing also includes data completion. If the missing data is at the beginning of the dataset, the empty data at the beginning is uniformly filled with the first non-empty data starting from the beginning. If the missing data is at the end of the dataset, the empty data at the end is uniformly filled with the first non-empty data starting from the end. If the missing data is in the middle of the dataset, the first non-empty data is searched forward and backward respectively, and linear interpolation is performed to fill it. The data set after filling is the network parameter sample set.

[0164] Step S1522: Rasterize the network parameter sample set according to the maximum value of each type of network parameter, and deliver the samples in the network parameter sample set to the corresponding raster.

[0165] It should be noted that rasterization specifically includes the following steps: Obtain the maximum values ​​of the number of users, service channel resource utilization, control channel resource utilization, and channel quality, respectively, and denot them as UserNumMax, ShareChannelMax, ControlChannelMax, and ChannelQualityMax in the above order. Divide the user number interval [0, UserNumMax] into UserNumMeshNum = Ceil(UserNumMax / UserNumMeshStep) grids with a grid step size UserNumMeshStep; divide the service channel resource utilization interval [0, ShareChannelMax] into ShareChannelMeshNum = Ceil(ShareChannelMax / ShareChannelMeshStep) grids with a grid step size ShareChannelMeshStep; divide the control channel resource utilization interval [0, ControlChannelMax] into ControlChannelMeshNum = Ceil( The channel quality interval [0, ChannelQualityMax] is divided into ChannelQualityMeshNum = Ceil(ChannelQualityMax / ChannelQualityMeshStep) grids with a grid step size of ChannelQualityMeshStep, resulting in a total of UserNumMeshNum × ShareChannelMeshNum × ControlChannelMeshNum × ChannelQualityMeshNum grids. The data in the dataset is then delivered to the corresponding grids using the number of users, service channel resource utilization, control channel resource utilization, and channel quality as key values.

[0166] Step S1523: Normalize the rasterized data and obtain the first training set and the first test set according to the filtering strategy.

[0167] It should be noted that a threshold value (MeshSampleMin) for the number of valid rasters can be preset. When the number of data points in a raster is greater than 2 × MeshSampleMin, the average of half of the samples randomly selected from that raster's data is used as the candidate training set for that raster, and the average of the remaining half is used as the candidate test set for that raster. When the number of data points in a raster is less than MeshSampleMin, that raster's data is discarded. When the number of rasters is between MeshSampleMin and 2 × MeshSampleMin, the average of the data points in that raster is used as the candidate training set. The candidate training set and candidate test set are normalized offline to construct the first training set and the first test set for the network traffic model offline.

[0168] Step S1524: Complete the training and testing of the network traffic model based on the first training set and the first test set.

[0169] It should be noted that the training and testing of the network traffic model are both done offline.

[0170] Step S1530: Determine the suppression reference value for each type of network parameter based on the network traffic model.

[0171] It should be noted that the suppression point of the traffic suppression model is denoted as CellBestThp, and the suppression reference value corresponding to each network parameter is determined by the following formula: CellBest = [CellBestUserNum, CellBestShareChannel, CellBestControlChannel, CellBestChannelQuality], where CellBestUserNum is the suppression reference value for the number of users, CellBestShareChannel is the suppression reference value for the utilization of service channel resources, CellBestControlChannel is the suppression reference value for the utilization of control channel resources, and CellBestChannelQuality is the suppression reference value for channel quality.

[0172] refer to Figure 16 , Figure 16 This is a schematic diagram of the mapping relationship of the network traffic model under two-dimensional network parameters. The network parameters include the number of users and the utilization rate of service channel resources. The vertical axis represents the value of cell traffic or cell SE. It can be seen that as the number of users and the utilization rate of service channel resources increase from the origin, the cell traffic or cell SE gradually increases. After reaching the suppression point, it reaches the average maximum traffic value. Then, any increase in any network parameter will lead to a decrease in traffic, that is, traffic suppression occurs.

[0173] Step S1540: Construct a prediction sample set offline on the network parameter sample set based on the time series, and obtain the second training set and the second test set based on the prediction sample set.

[0174] It should be noted that the sample size of the prediction sample set is N_Sample. That is, the prediction sample set can be expressed as D = [data_0,...,data_i,...,data_N_Sample-1], where the i-th prediction sample is denoted as data_i. Each prediction sample has K+L+M+1 dimensions (where K is the parameter value at the same time point in the previous K periods, L is the parameter value at the previous L times, and M is the time information of the current data, including day of the week, hour, minute, etc.). The first K+L+M dimensions represent the features of each sample, and the last dimension is the parameter value at the current time, used to represent the label of the prediction sample. Max-min normalization is performed on each column of the sample dataset D, with the normalization parameter denoted as NormParameter, and the normalized dataset denoted as D_norm. The normalized data D_norm is then split into a second training set and a second test set according to the training-test set splitting ratio parameter alpha.

[0175] Step S1550: Train and test the parameter prediction model offline based on the second training set and the second test set.

[0176] Step S1560: Obtain the predicted parameter values ​​based on the parameter prediction model.

[0177] It should be noted that obtaining the parameter prediction values ​​includes the following steps: Prepare data according to the data processing method, constructing the feature data `data` of the prediction sample. The `data` dimension is K+L+M, where K represents the parameter values ​​at the same time point in the previous K periods, L represents the parameter values ​​at the previous L times, and M represents the time information of the current data, including day of the week, hour, minute, etc. Normalize the feature data of the prediction sample using the normalization parameter `NormParameter` to obtain the normalized feature data `data_norm`. Input `data_norm` into the model parameter prediction model to obtain the intermediate prediction values ​​for future periods. Perform an inverse normalization transformation on the intermediate prediction values ​​to obtain the parameter prediction values, denoted as `CellStatePredict`.

[0178] Step S1570: Load the offline network traffic model and the offline parameter prediction model to complete the traffic suppression analysis.

[0179] It should be noted that if the predicted value CellStatePredict = [UserNum, ShareChannel, Channel, ChannelQuality] for each type of network parameter is less than or equal to CellBest = [CellBestUserNum, CellBestShareChannel, CellBestControlChannel, CellBestChannelQuality], then the traffic for that period will not be suppressed in the future; otherwise, the traffic for that period will be suppressed in the future.

[0180] It should be noted that, in order to determine the specific traffic loss, the state prediction value CellStatePredict can be input into the network traffic model to obtain the cell traffic under this state as CellPredictThp. Then, the absolute amount of traffic loss is Max(CellBestThp-CellPredictThp, 0), and the relative degree of traffic loss is Max(CellBestThp-CellPredictThp, 0) / CellBestThp.

[0181] It should be noted that analyzing the reasons for traffic suppression can be accomplished through the following steps: Obtain slices of the network traffic model in CellBest corresponding to the components CellBestUserNum, CellBestShareChannel, CellBestControlChannel, and CellBestChannelQuality, and take the average of the first-order differences of the curves within a preset region, denoted as diff_UserNum_best, diff_ShareChannel_best, diff_ControlChannel_best, and diff_ChannelQuality_best, respectively. The preset region of the curve can be the first half near the origin. Then, obtain slices of the network traffic model in CellStatePredict corresponding to the components UserNum, ShareChannel, ControlChannel, and ChannelQuality, and take the first-order differences of one point before and after each slice, denoted as diff_UserNum, diff_ShareChannel, diff_ControlChannel, and diff_ChannelQuality, respectively.

[0182] After acquiring the data, the degree of inhibition is calculated. The degree of inhibition for the number of users is: UserNum_inhibition_ratio = (diff_UserNum_best - diff_UserNum) / diff_UserNum_best; the degree of inhibition for service channel resources is: ShareChannel_inhibition_ratio = (diff_ShareChannel_best - diff_ShareChannel) / diff_ShareChannel_best; the degree of inhibition for control channel resources is: ControlChannel_inhibition_ratio = (diff_ControlChannel_best - diff_ControlChannel) / diff_ControlChannel_best; and the degree of inhibition for channel quality is: ChannelQuality_inhibition_ratio = (diff_ChannelQuality_best - diff_ChannelQuality) / diff_ChannelQuality_best.

[0183] It should be noted that, in order to analyze the main factors causing traffic suppression, the maximum value among UserNum_inhibition_ratio, ShareChannel_inhibition_ratio, ControlChannel_inhibition_ratio, and ChannelQuality_inhibition_ratio can be obtained to determine the main cause of traffic suppression. If the maximum value is UserNum_inhibition_ratio, then the limited number of users is the main factor; if the maximum value is ShareChannel_inhibition_ratio, then the limited service channel resources are the main factor; if the maximum value is ControlChannel_inhibition_ratio, then the limited control channel resources are the main factor; and if the maximum value is ChannelQuality_inhibition_ratio, then the limited channel quality is the main factor.

[0184] Example 3: In this example, a two-dimensional modeling approach is used. Network parameters are exemplified by the utilization rates of service channel resources and control channel resources. Simultaneously, an online network traffic model is employed, and real-time parameters are used for parameter prediction. (Refer to...) Figure 17 This example includes, but is not limited to, the following steps:

[0185] Step S1710: Determine that both the network traffic model and the parameter prediction model are generated online.

[0186] Step S1721: Collect data according to the collection granularity and collection cycle, and preprocess the data.

[0187] It should be noted that the data is collected from historical data of the network management system, with a collection granularity of 15 minutes and a collection cycle of 28 days, including service channel resource utilization, control channel resource utilization, cell traffic, etc. Preprocessing includes data time extraction and completion, data time extraction deduplication, and data completion.

[0188] Online data timeline completion is performed. During the data collection process, there may be omissions in certain granularities. If some data information on the timeline is missing, the timeline will be filled in, and the corresponding data will be empty.

[0189] Online data timeline deduplication is performed. During the data collection process, duplicate data may be collected at certain granularities. It is necessary to deduplicatize the duplicate data. The deduplication rule is to keep the first occurrence of the data in the dataset and delete the duplicate data thereafter.

[0190] Online data completion is performed. If the missing data is at the beginning of the dataset, the empty data at the beginning is filled with the first non-empty data starting from the beginning. If the missing data is at the end of the dataset, the empty data at the end is filled with the first non-empty data starting from the end. If the missing data is in the middle of the dataset, the first non-empty data is searched forward and backward respectively, and linear interpolation is performed to fill the missing data. The filled dataset is the network parameter sample set.

[0191] Step S1722: Rasterize the network parameter sample set according to the maximum value of each type of network parameter, and deliver the samples in the network parameter sample set to the corresponding raster.

[0192] It should be noted that the rasterization process specifically includes the following steps: First, obtain the maximum values ​​of the service channel resource utilization and the control channel resource utilization, denoted as ShareChannelMax and ControlChannelMax, respectively. Second, divide the service channel resource utilization interval [0, ShareChannelMax] into ShareChannelMeshNum = Ceil(ShareChannelMax / ShareChannelMeshStep) grids using the grid step size ShareChannelMeshStep. Third, divide the control channel resource utilization interval [0, ControlChannelMax] into ControlChannelMeshNum = Ceil(ControlChannelMax / ControlChannelMeshStep) grids using the grid step size ControlChannelMeshStep. A total of ShareChannelMeshNum × ControlChannelMeshNum grids are obtained. Finally, the data in the dataset is delivered to the corresponding grids according to the service channel resource utilization and control channel resource utilization.

[0193] Step S1723: Normalize the rasterized data and obtain the first training set and the first test set according to the filtering strategy.

[0194] It should be noted that a threshold value (MeshSampleMin) for the number of valid graticles can be preset. When the number of data points in a grate is greater than 2 × MeshSampleMin, the average of half of the samples randomly selected from that grate is used as the candidate training set for that grate, and the average of the remaining half is used as the candidate test set. When the number of data points in a grate is less than MeshSampleMin, that grate is discarded. When the number of grates is between MeshSampleMin and 2 × MeshSampleMin, the average of the data points in that grate is used as the candidate training set. The candidate training set and candidate test set are then normalized to construct the first training set and the first test set for the network traffic model.

[0195] Step S1724: Construct a network traffic model as a neural network model and randomly initialize the network traffic model. Complete the training and testing of the network traffic model based on the first training set and the first test set.

[0196] Step S1730: Determine the suppression point traffic value and the suppression reference value for each type of network parameter based on the network traffic model.

[0197] It should be noted that the suppression point of the traffic suppression model is denoted as CellBestThp, and the suppression reference value corresponding to each network parameter is determined by the following formula: CellBest = [CellBestShareChannel, CellBestControlChannel], where CellBestShareChannel is the suppression reference value for service channel resource utilization, and CellBestControlChannel is the suppression reference value for control channel resource utilization. Simultaneously, the suppression point traffic value of the network traffic model is denoted as CellBestThp, and the corresponding feature is denoted as CellBest = [CellBestShareChannel, CellBestControlChannel].

[0198] Step S1740: Load the network traffic model and complete the traffic suppression analysis.

[0199] It should be noted that the real-time filtered value of the network parameters is used as the parameter prediction value, denoted as CellStatePredict. If the future state CellStatePredict is less than or equal to CellBest, then the traffic in that future time period will not be suppressed; otherwise, the traffic in that future time period will be suppressed.

[0200] It should be noted that, in order to determine the specific traffic loss, the state prediction value CellStatePredict can be input into the network traffic model to obtain the cell traffic under this state as CellPredictThp. Then, the absolute amount of traffic loss is Max(CellBestThp-CellPredictThp, 0), and the relative degree of traffic loss is Max(CellBestThp-CellPredictThp, 0) / CellBestThp.

[0201] It should be noted that analyzing the reasons for traffic suppression can be accomplished through the following steps: Obtain slices of the network traffic model at the CellBest corresponding components CellBestShareChannel and CellBestControlChannel, and take the average of the first-order differences of the curves within a preset region, denoted as diff_ShareChannel_best and diff_ControlChannel_best, respectively. The preset region of the curves can be the first half near the origin. Next, obtain slices of the network traffic model at the CellStatePredict corresponding components ShareChannel and ControlChannel, take the first-order differences of one point before and after the first-order differences, and take the average, denoted as diff_ShareChannel and diff_ControlChannel, respectively.

[0202] After acquiring the data, the degree of inhibition is calculated. The degree of inhibition for service channel resources is: ShareChannel_inhibition_ratio = (diff_ShareChannel_best - diff_ShareChannel) / diff_ShareChannel_best; the degree of inhibition for control channel resources is: ControlChannel_inhibition_ratio = (diff_ControlChannel_best - diff_ControlChannel) / diff_ControlChannel_best.

[0203] It should be noted that, in order to analyze the main factors causing traffic suppression, the maximum values ​​of UShareChannel_inhibition_ratio and ControlChannel_inhibition_ratio can be obtained to determine the main cause of traffic suppression. If the maximum value is ShareChannel_inhibition_ratio, the main factor is limited service channel resources; if the maximum value is ControlChannel_inhibition_ratio, the main factor is limited control channel resources.

[0204] Additionally, refer to Figure 18 An embodiment of the present invention also provides an electronic device 1800, which includes a memory 1810, a processor 1820, and a computer program stored in the memory 1810 and executable on the processor 1820.

[0205] The processor 1820 and memory 1810 can be connected via a bus or other means.

[0206] The non-transient software program and instructions required to implement the traffic suppression prediction method of the above embodiments are stored in the memory 1810. When executed by the processor 1820, the traffic suppression prediction method in the above embodiments is executed, for example, the method described above is executed. Figure 1 Method steps S110 to S130, Figure 2 Method steps S210 to S220, Figure 3 Method steps S310 to S330, Figure 4 Method steps S410 to S440 Figure 5 Method steps S510 to S520 Figure 6 Method steps S610 to S630, Figure 7 Method steps S710 to S740 Figure 8 Method steps S810 to S850 Figure 9 Method steps S910 to S920 Figure 10 Method steps S1010 to S1020 Figure 11 Method steps S1110 to S1120, Figure 13 Method steps S1310 to S1350, Figure 15 Method steps S1510 to S1570, Figure 17 Method steps S1710 to S1740.

[0207] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; 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.

[0208] Furthermore, one embodiment of the present invention provides a computer-readable storage medium storing computer-executable instructions that are executed by a processor or controller, for example, by a processor in the above-described electronic device embodiment, causing the processor to perform the traffic suppression prediction method in the above-described embodiment, for example, performing the above-described... Figure 1 Method steps S110 to S130, Figure 2 Method steps S210 to S220, Figure 3 Method steps S310 to S330, Figure 4 Method steps S410 to S440 Figure 5 Method steps S510 to S520 Figure 6 Method steps S610 to S630, Figure 7 Method steps S710 to S740 Figure 8 Method steps S810 to S850 Figure 9 Method steps S910 to S920 Figure 10 Method steps S1010 to S1020 Figure 11 Method steps S1110 to S1120, Figure 13 Method steps S1310 to S1350, Figure 15 Method steps S1510 to S1570, Figure 17 The method steps S1710 to S1740 are described above. Those skilled in the art will understand that all or some of the steps in the methods and systems disclosed above can be implemented as software, firmware, hardware, and suitable combinations thereof. Some or all physical components can be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit. Such software can be distributed on a computer-readable medium, which can include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible to a computer. Furthermore, as is known to those skilled in the art, communication media typically contain computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0209] The above is a detailed description of the preferred embodiments of the present invention. However, the present invention is not limited to the above embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention. All such equivalent modifications or substitutions are included within the scope defined by the claims of the present invention.

Claims

1. A method for predicting traffic suppression, comprising: The suppression point traffic value is determined based on a preset network traffic model, whereby the network traffic model represents the mapping relationship between the values ​​of network parameters of the transmission network and the traffic value, and the suppression point traffic value is the traffic threshold of the transmission network under the current operating strategy. Determine the suppression reference value for the target network parameter corresponding to the suppression point flow value; Obtain the parameter prediction values ​​corresponding to the target network parameters, and determine the traffic suppression prediction result based on the parameter prediction values ​​and the suppression reference values; Before determining the suppression point traffic value based on a preset network traffic model, the method further includes: A network parameter sample set is obtained from historical data of the transmission network, the network parameter sample set including samples of the network parameters; The network traffic model is trained based on the network parameter sample set.

2. The method according to claim 1, characterized in that, Training the network traffic model based on the network parameter sample set includes: The first training set and the first test set are obtained based on the network parameter sample set; Normalize the first training set and the first test set; Obtain a network traffic model, train the network traffic model based on the first training set, and verify the trained network traffic model based on the first test set.

3. The method according to claim 2, characterized in that, The process of obtaining the first training set and the first test set based on the network parameter sample set includes: Determine the numerical range of the network parameter sample set; The numerical range is divided into several grid intervals according to the preset grid step size; Each sample of the network parameters is filled into the grid interval according to the principle of numerical correspondence, to obtain the grid dataset; According to the preset filtering strategy, the first training set and the first test set are selected from the raster dataset.

4. The method according to claim 1, characterized in that, The step of determining the suppression point traffic value based on a preset network traffic model includes: Obtain a preset threshold value, determine the first-order difference of the network parameter samples, and determine the network parameter samples whose first-order difference is less than the threshold value as target samples; The target flow value corresponding to the target sample is determined based on the mapping relationship, and the target flow value is determined as the inhibition point flow value.

5. The method according to claim 1, characterized in that, The step of obtaining a network parameter sample set from historical data of the transmission network includes: Obtain a preset acquisition granularity and acquisition period, wherein the acquisition granularity represents the frequency of acquiring the network parameters within one acquisition period; Samples of network parameters are collected from the historical data according to the collection granularity and the collection period; The collected network parameter samples are preprocessed to obtain the network parameter sample set.

6. The method according to claim 5, characterized in that, The step of obtaining the parameter prediction values ​​corresponding to the target network parameters includes: The acquisition time for each acquisition cycle is determined based on the acquisition granularity. A prediction sample set is obtained from the network parameter sample set, wherein the sample data of the prediction sample set includes the acquisition time and the parameter value of the network parameter corresponding to the acquisition time; A parameter prediction model is trained based on the predicted sample set. Obtain the predicted feature samples corresponding to the target network parameters, and input the predicted feature samples into the parameter prediction model to obtain the predicted parameter values.

7. The method according to claim 6, characterized in that, The step of training a parameter prediction model based on the prediction sample set includes: The data in the predicted sample set is normalized; The normalized prediction sample set is divided into a second training set and a second test set according to a preset segmentation ratio. Obtain the initial parameter prediction model and determine the training parameters of the initial parameter prediction model; The initial parameter prediction model is trained according to the second training set, and the trained initial parameter prediction model is tested according to the second test set. The initial parameter prediction model after verification is determined as the parameter prediction model.

8. The method according to claim 1, characterized in that, The process of obtaining network parameter samples from historical data of the transmission network includes: Determine the target historical data, which is the historical data obtained by the transmission network according to the latest operating strategy; Obtain network parameter samples from the target's historical data.

9. The method according to claim 1, characterized in that, After determining the traffic suppression prediction result based on the predicted parameter value and the suppression reference value, the method further includes: When the traffic suppression prediction result indicates that traffic suppression has occurred, the predicted parameter value is input into the network traffic model to obtain the traffic prediction value; The degree of traffic suppression is determined based on the predicted traffic volume and the suppression reference value.

10. The method according to claim 9, characterized in that, The network traffic model includes at least two different types of target network parameters. After determining the degree of traffic suppression based on the traffic prediction value and the suppression reference value, the method further includes: Determine the degree of traffic suppression corresponding to each of the target network parameters; The root cause network parameter that causes traffic suppression is determined, wherein the root cause network parameter is the target network parameter corresponding to the highest value of the traffic suppression degree.

11. The method according to claim 1, characterized in that: The suppression point traffic value is the average maximum traffic value of the transmission network under the current operating strategy.

12. 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 traffic suppression prediction method as described in any one of claims 1 to 11.

13. A computer-readable storage medium storing computer-executable instructions for performing the traffic suppression prediction method as described in any one of claims 1 to 11.

Images