Communication network load prediction method, device, equipment and readable storage medium

Abstract

This application discloses a method, apparatus, device, and readable storage medium for predicting the load of a communication network. The method includes: determining whether a load prediction model for the target communication network is suitable for the current operating scenario of the target communication network based on the historical actual load and the historical predicted load of the historical actual load; if suitable, predicting the target predicted load of the target communication network at the next moment based on the recent load data of the target communication network and the load prediction model; if not suitable, updating the load prediction model based on the recent historical load data of the target communication network, and predicting the target predicted load based on the updated load prediction model and the recent load data, wherein the recent historical load data refers to the historical load data of the target communication network within a second preset time period prior to the current moment. Real-time model updates enable the prediction model to better adapt to the real-time operating scenario, thereby obtaining more accurate load prediction results.

Description

Technical Field

[0001] This application relates to the field of communication technology, and in particular to a method, apparatus, device and readable storage medium for predicting the load of a communication network. Background Technology

[0002] Mobile communication network load level is a key performance indicator affecting communication network operation and maintenance decisions (such as network resource allocation and topology management). As mobile networks continue to expand, the communication load levels of base station access nodes vary significantly across different regions and time periods, exhibiting dynamic changes that exacerbate the difficulty and cost of communication network management. Predicting future mobile communication network load can provide prior information for network operation and maintenance decisions, thereby enabling the early development of effective response plans. Therefore, a method for accurately predicting the future communication load of current mobile networks is urgently needed.

[0003] The above content is only used to help understand the technical solution of this application and does not represent an admission that the above content is prior art. Summary of the Invention

[0004] The main objective of this application is to provide a method, apparatus, device, and readable storage medium for predicting the load of a communication network, with the aim of improving the accuracy of future communication load prediction results for mobile communication networks.

[0005] To achieve the above objectives, this application provides a communication network load prediction method, which includes the following steps:

[0006] Based on the historical actual load of the target communication network and the historical predicted load of the historical actual load, it is determined whether the load prediction model of the target communication network is suitable for the current working scenario of the target communication network.

[0007] If compatible, the target predicted load of the target communication network at the next moment is predicted based on the recent load data of the target communication and the load prediction model, wherein the recent load data is the historical load data of the target communication network within a first preset time period before the current moment;

[0008] If not suitable, the load prediction model is updated based on the recent historical load data of the target communication network, and the target predicted load is predicted based on the updated load prediction model and the recent load data. The recent historical load data is the historical load data of the target communication network within a second preset time period before the current time, and the second preset time period is longer than the first preset time period.

[0009] Furthermore, the step of determining whether the load prediction model of the target communication network is suitable for the current working scenario of the target communication network based on the historical actual load of the target communication network and the historical predicted load of the historical actual load includes:

[0010] The first load vector feature is obtained by extracting vector features from the historical actual load.

[0011] The second load vector feature is obtained by extracting vector features from the historical predicted load;

[0012] Calculate the similarity between the first load vector feature and the second load vector feature;

[0013] If the similarity is greater than a preset similarity threshold, then the load prediction model is determined to be suitable for the current working scenario of the target communication network.

[0014] If the similarity is less than or equal to the preset similarity threshold, then the load prediction model is determined to be unsuitable for the current working scenario of the target communication network.

[0015] Furthermore, before the step of determining whether the load prediction model of the target communication network is suitable for the current working scenario of the target communication network based on the historical actual load of the target communication network and the historical predicted load of the historical actual load, the method includes:

[0016] Training samples are generated based on the historical load data of the target communication network.

[0017] The load data that are temporally adjacent in the training samples are integrated into sample features and input into the load prediction model to obtain the predicted load of the training samples.

[0018] The model prediction loss of the load prediction model is calculated based on the predicted load and the labels of the training samples;

[0019] The load prediction model is iteratively updated based on the predicted loss from the model.

[0020] Furthermore, the step of iteratively updating the load prediction model based on the model prediction loss includes:

[0021] Determine whether the model's prediction loss has converged;

[0022] If the model prediction loss is determined to converge, then the load prediction model training is complete.

[0023] If it is determined that the model prediction loss does not converge, the parameters in the load prediction model are updated based on the model prediction loss and the gradient descent method.

[0024] Based on the new training samples and the updated load prediction model, return to the step of integrating temporally adjacent load data in the training samples into sample features and inputting them into the load prediction model to obtain the predicted load of the training samples, and subsequent steps.

[0025] Furthermore, the target communication network includes multiple communication base stations, and the step of generating training samples based on the historical load data of the target communication network includes:

[0026] Historical load data of each of the aforementioned communication base stations are obtained as a training dataset;

[0027] The training samples are extracted from the training dataset based on a preset time window and the temporal order of the training dataset. Each training sample includes load data of different communication base stations at different times within the preset time window during the same period. The label of each training sample is the load data of different communication base stations at the next time in the corresponding period of the training sample.

[0028] Furthermore, the step of updating the load prediction model based on recent historical load data of the target communication network includes:

[0029] The recent historical load data of each of the target communication networks is obtained as the training dataset;

[0030] The steps of extracting the training samples from the training dataset based on the preset interval and the temporal order of the training dataset, and subsequent steps, are performed.

[0031] Furthermore, the communication network load prediction method is applied to the digital twin network of the target communication network. Before the step of determining whether the load prediction model of the target communication network is suitable for the current working scenario of the target communication network based on the historical actual load of the target communication network and the historical predicted load of the historical actual load, the method further includes:

[0032] Historical load data of the target communication network is obtained based on the digital twin network;

[0033] The recent historical load data is extracted in real time from the latest historical load data in the historical load data by using a preset time sliding window.

[0034] Furthermore, to achieve the above objectives, this application also provides a communication network load prediction device, which includes:

[0035] The judgment module is used to determine whether the load prediction model of the target communication network is suitable for the current working scenario of the target communication network based on the historical actual load of the target communication network and the historical predicted load of the historical actual load.

[0036] The first prediction module is used, if adapted, to predict the target predicted load of the target communication network at the next moment based on the recent load data of the target communication and the load prediction model, wherein the recent load data is the historical load data of the target communication network within a first preset time period before the current moment;

[0037] The second prediction module is used to update the load prediction model based on the recent historical load data of the target communication network if it is not suitable, and to predict the target predicted load based on the updated load prediction model and the recent load data. The recent historical load data is the historical load data of the target communication network within a second preset time period before the current time, and the second preset time period is longer than the first preset time period.

[0038] In addition, to achieve the above objectives, this application also provides a communication network load prediction device, which includes: a memory, a processor, and a communication network load prediction program stored in the memory and executable on the processor. When the communication network load prediction program is executed by the processor, it implements the steps of the communication network load prediction method described above.

[0039] In addition, to achieve the above objectives, this application also provides a readable storage medium storing a communication network load prediction program, which, when executed by a processor, implements the steps of the communication network load prediction method as described above.

[0040] This application proposes a method, apparatus, device, and medium for predicting the load of a communication network. This application determines whether a load prediction model for the target communication network is suitable for the current operating scenario of the target communication network based on the historical actual load and the historical predicted load of the historical actual load. If suitable, the target predicted load of the target communication network at the next moment is predicted based on the recent load data of the target communication network and the load prediction model, wherein the recent load data is the historical load data of the target communication network within a first preset time period before the current moment. If unsuitable, the load prediction model is updated based on the recent historical load data of the target communication network, and the target predicted load is predicted based on the updated load prediction model and the recent load data, wherein the recent historical load data is the historical load data of the target communication network within a second preset time period before the current moment, and the second preset time period is longer than the first preset time period. In other words, the load prediction model at this time is used to determine whether it is suitable for the current operating scenario by using the historical actual results and their corresponding historical predicted load results. If suitable, the load prediction model at this time is used for prediction. If the model is not suitable, recent historical data is used as training samples to update the load prediction model, thereby enabling real-time adjustment of the load prediction model. The prediction is then made based on the updated and adjusted model. It is understandable that real-time adjustment of model parameters allows the prediction model to better adapt to the real-time working scenario, resulting in more accurate load prediction results. Attached Figure Description

[0041] Figure 1 This is a schematic diagram of the device structure of the hardware operating environment involved in the embodiments of this application;

[0042] Figure 2 This is a flowchart illustrating the first embodiment of the communication network load prediction method of this application;

[0043] Figure 3 This is a flowchart illustrating the second embodiment of the communication network load prediction method of this application;

[0044] Figure 4 This is a flowchart illustrating the third embodiment of the communication network load prediction method in this application;

[0045] Figure 5 This is a flowchart illustrating the fourth embodiment of the communication network load prediction method in this application;

[0046] Figure 6 This is a schematic diagram of the network structure used to calculate similarity in the communication network load prediction method of this application.

[0047] Figure 7 This is a schematic diagram of the load prediction framework in the communication network load prediction method of this application;

[0048] Figure 8 This is a diagram showing the internal structure of the ConvLSTM unit in the communication network load prediction method of this application;

[0049] Figure 9 This is a schematic diagram of the overall logic flow of the communication network load prediction method of this application;

[0050] Figure 10 This is a schematic diagram of the digital twin system framework in the communication network load prediction method of this application.

[0051] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0052] It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit this application.

[0053] like Figure 1 As shown, Figure 1 This is a schematic diagram of the device structure of the hardware operating environment involved in the embodiments of this application.

[0054] The device in this application embodiment can be a server, or it can be an electronic terminal device such as a smartphone, PC, tablet computer, or portable computer.

[0055] like Figure 1 As shown, the device may include: a processor 1001, such as a CPU; a network interface 1004; a user interface 1003; a memory 1005; and a communication bus 1002. The communication bus 1002 is used to enable communication between these components. The user interface 1003 may include a display screen or an input unit such as a keyboard; optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface). The memory 1005 may be high-speed RAM or non-volatile memory, such as a disk drive. Optionally, the memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0056] Optionally, the device may also include a camera, RF (Radio Frequency) circuitry, sensors, audio circuitry, a WiFi module, and so on. Sensors may include light sensors, motion sensors, and other sensors. Specifically, light sensors may include ambient light sensors and proximity sensors. The ambient light sensor can adjust the display brightness according to the ambient light level, while the proximity sensor can turn off the display and / or backlight when the mobile terminal is moved to the ear. As a type of motion sensor, a gravity accelerometer can detect the magnitude of acceleration in various directions (generally three axes). When stationary, it can detect the magnitude and direction of gravity, and can be used for applications that identify the mobile terminal's posture (such as landscape / portrait switching, related games, magnetometer posture calibration), vibration recognition functions (such as pedometers, taps), etc. Of course, the mobile terminal may also be equipped with other sensors such as gyroscopes, barometers, hygrometers, thermometers, and infrared sensors, which will not be elaborated here.

[0057] Those skilled in the art will understand that Figure 1 The device structure shown does not constitute a limitation on the terminal and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0058] like Figure 1 As shown, the memory 1005, which serves as a computer storage medium, may include an operating system, a network communication module, a user interface module, and a communication network load prediction program.

[0059] exist Figure 1 In the device shown, network interface 1004 is mainly used to connect to the backend server and communicate with it; user interface 1003 is mainly used to connect to the client (user terminal) and communicate with it; while processor 1001 can be used to call the communication network load prediction program stored in memory 1005 and perform the following operations:

[0060] Based on the historical actual load of the target communication network and the historical predicted load of the historical actual load, it is determined whether the load prediction model of the target communication network is suitable for the current working scenario of the target communication network.

[0061] If compatible, the target predicted load of the target communication network at the next moment is predicted based on the recent load data of the target communication and the load prediction model, wherein the recent load data is the historical load data of the target communication network within a first preset time period before the current moment;

[0062] If not suitable, the load prediction model is updated based on the recent historical load data of the target communication network, and the target predicted load is predicted based on the updated load prediction model and the recent load data. The recent historical load data is the historical load data of the target communication network within a second preset time period before the current time, and the second preset time period is longer than the first preset time period.

[0063] Furthermore, the processor 1001 can call the communication network load prediction program stored in the memory 1005 and also perform the following operations:

[0064] The steps for determining whether the load prediction model of the target communication network is suitable for the current working scenario of the target communication network based on the historical actual load of the target communication network and the historical predicted load of the historical actual load include:

[0065] The first load vector feature is obtained by extracting vector features from the historical actual load.

[0066] The second load vector feature is obtained by extracting vector features from the historical predicted load;

[0067] Calculate the similarity between the first load vector feature and the second load vector feature;

[0068] If the similarity is greater than a preset similarity threshold, then the load prediction model is determined to be suitable for the current working scenario of the target communication network.

[0069] If the similarity is less than or equal to the preset similarity threshold, then the load prediction model is determined to be unsuitable for the current working scenario of the target communication network.

[0070] Furthermore, the processor 1001 can call the communication network load prediction program stored in the memory 1005 and also perform the following operations:

[0071] Before the step of determining whether the load prediction model of the target communication network is suitable for the current working scenario of the target communication network based on the historical actual load of the target communication network and the historical predicted load of the historical actual load, the method includes:

[0072] Training samples are generated based on the historical load data of the target communication network.

[0073] The load data that are temporally adjacent in the training samples are integrated into sample features and input into the load prediction model to obtain the predicted load of the training samples.

[0074] The model prediction loss of the load prediction model is calculated based on the predicted load and the labels of the training samples;

[0075] The load prediction model is iteratively updated based on the predicted loss from the model.

[0076] Furthermore, the processor 1001 can call the communication network load prediction program stored in the memory 1005 and also perform the following operations:

[0077] The step of iteratively updating the load prediction model based on the model prediction loss includes:

[0078] Determine whether the model's prediction loss has converged;

[0079] If the model prediction loss is determined to converge, then the load prediction model training is complete.

[0080] If it is determined that the model prediction loss does not converge, the parameters in the load prediction model are updated based on the model prediction loss and the gradient descent method.

[0081] Based on the new training samples and the updated load prediction model, the process returns to the step of integrating temporally adjacent load data from the training samples into sample features and inputting them into the load prediction model to obtain the predicted load of the training samples, and subsequent steps. Further, the processor 1001 can call the communication network load prediction program stored in the memory 1005 and also perform the following operations:

[0082] The target communication network includes multiple communication base stations, and the step of generating training samples based on the historical load data of the target communication network includes:

[0083] Historical load data of each of the aforementioned communication base stations are obtained as a training dataset;

[0084] The training samples are extracted from the training dataset based on a preset time window and the temporal order of the training dataset. Each training sample includes load data of different communication base stations at different times within the preset time window during the same period. The label of each training sample is the load data of different communication base stations at the next time in the corresponding period of the training sample.

[0085] Furthermore, the processor 1001 can call the communication network load prediction program stored in the memory 1005 and also perform the following operations:

[0086] The step of updating the load prediction model based on recent historical load data of the target communication network includes:

[0087] The recent historical load data of each of the target communication networks is obtained as the training dataset;

[0088] The steps of extracting the training samples from the training dataset based on the preset interval and the temporal order of the training dataset, and subsequent steps, are performed.

[0089] Furthermore, the processor 1001 can call the communication network load prediction program stored in the memory 1005 and also perform the following operations:

[0090] The communication network load prediction method is applied to the digital twin network of the target communication network. Before the step of determining whether the load prediction model of the target communication network is suitable for the current working scenario of the target communication network based on the historical actual load of the target communication network and the historical predicted load of the historical actual load, the method further includes:

[0091] Historical load data of the target communication network is obtained based on the digital twin network;

[0092] The recent historical load data is extracted in real time from the latest historical load data in the historical load data by using a preset time sliding window.

[0093] Reference Figure 2 The first embodiment of the communication network load prediction method of this application includes:

[0094] Step A10: Based on the historical actual load of the target communication network and the historical predicted load of the historical actual load, determine whether the load prediction model of the target communication network is suitable for the current working scenario of the target communication network.

[0095] It should be noted that, in this embodiment, the target communication network refers to the communication network whose load needs to be predicted over a future period. A communication network may include multiple base stations, and the load can refer to the frequency of users accessing each base station. In this embodiment, a trained load prediction model will be used to predict the future load of the target communication network. This load prediction model can be a neural network model such as LSTM (Long Short-Term Memory) and GRU (gated recurrent neural network). However, with social development and increased population mobility, the operating environment of the target communication network becomes more variable, and its load variation patterns also change accordingly. Therefore, the load prediction model may not be well adapted to the operating environment of the target communication network. To address this problem, this embodiment proposes a communication network load prediction method.

[0096] For example, to ensure that the load prediction model used to predict the load of a target communication network over a future period is well-suited to the current operating environment of the target communication network, a preliminary assessment of the load prediction model can be performed before prediction. This involves comparing the historical actual load of the target communication network with its historical predicted load, which corresponds to the same point in time (e.g., the current actual load and its predicted load). If the difference is small or negligible, the load prediction model is accurate and suited to the current operating environment of the target communication network. Conversely, a large difference indicates that the model's stored results are inaccurate and it is not suitable for the current operating environment of the target communication network. In cases of incompatibility, the structure of the load model needs to be adjusted.

[0097] In one possible implementation, step A10, which determines whether the load prediction model of the target communication network is suitable for the current operating scenario of the target communication network based on the historical actual load of the target communication network and the historical predicted load of the historical actual load, includes steps A110-A150:

[0098] Step A110: Extract vector features from the historical actual load to obtain the first load vector feature;

[0099] Step A120: Extract vector features from the historical predicted load to obtain the second load vector features;

[0100] Step A130: Calculate the similarity between the first load vector feature and the second load vector feature;

[0101] Step A140: If the similarity is greater than a preset similarity threshold, then it is determined that the load prediction model is adapted to the current working scenario of the target communication network.

[0102] Step A150: If the similarity is less than or equal to the preset similarity threshold, then it is determined that the load prediction model is not suitable for the current working scenario of the target communication network.

[0103] For example, vector features are extracted from historical actual load and historical predicted load to obtain first load vector features and second load vector features, respectively. Then, the similarity between the first load vector features and the second load vector features is used. If the similarity is greater than a preset similarity threshold, the load prediction model is determined to be suitable for the current operating scenario of the target communication network. Conversely, if the similarity is less than or equal to the similarity threshold, the load prediction model is determined to be unsuitable for the current operating scenario of the target communication network. (See reference...) Figure 6The diagram below illustrates the network structure for calculating similarity in this application. Suppose we need to predict the load at time T, then the actual base station load data a in the diagram... T-1 (i.e., historical actual load) and predicted base station load data (i.e., historical predicted load) all correspond to time T-1, and a T-1 and The inputs are fed into convolutional neural networks with the same structure. W(W1, W2, W3) / B(B1, B2, B3) in the convolutional neural network are sub-network parameters, which can be set according to the actual situation. T-1 The input is fed into a convolutional neural network for vector feature extraction to obtain O p (i.e., the first load vector feature), The input is fed into a convolutional neural network for vector feature extraction to obtain O r (i.e., the second load vector feature). The vector feature extraction and calculation method is as follows:

[0104]

[0105] o r =σ(w3*σ(w2*σ(w1*a) T-1 +b1)+b2)+b3)

[0106] In the formula, W1, W2, W3, B1, B2, and B3 represent the parameters of the neural network sub-network used for vector feature extraction, σ represents the sub-network activation function, and O p For the first load vector feature, O r This is the second load vector feature.

[0107] Calculate O again p and O r The similarity between them is calculated as follows:

[0108] E w (o p o r )=||o p -o r ||2

[0109] In the formula, O p For the first load vector feature, O r For the second load vector feature, Ew(O) p O r ) represents the degree of difference, 1 / Ew(O p O r The similarity score is denoted as . Similarly, the preset similarity threshold can be set by technical personnel according to their requirements for the model's prediction accuracy.

[0110] Step A20: If compatible, predict the target predicted load of the target communication network at the next moment based on the recent load data of the target communication and the load prediction model, wherein the recent load data is the historical load data of the target communication network within a first preset time period before the current moment;

[0111] For example, if the load prediction model is determined to be suitable for the current working scenario, the recent load data of the target communication network at the current moment is input into the load prediction model to predict the target predicted load of the target communication network at the next moment, that is, the load of the target communication network in the future. The recent load data refers to the historical load data of the target communication network within a first preset time period from the current moment. For example, at the current moment T, to predict the load of the target communication network at the next moment T+1, it is necessary to use the historical load data of the target communication network from moment TN to moment T. At moment T+1, to predict the load of the target communication network at moment T+2, it is necessary to use the historical data from moment T-N+1 to moment T+1, where N is equivalent to the length of the aforementioned first preset time period, and the value of N is set during the model training phase. The obtained target predicted load can then be used to guide network operation and maintenance decisions.

[0112] Step A30: If it does not fit, update the load prediction model based on the recent historical load data of the target communication network, and predict the target predicted load based on the updated load prediction model and the recent load data. The recent historical load data is the historical load data of the target communication network within a second preset time period before the current time, and the second preset time period is longer than the first preset time period.

[0113] For example, if it is determined that the load prediction model is not suitable for the current working scenario, the load prediction model is updated based on the recent historical load data of the target communication network. The recent historical load data refers to the historical load data of the target communication network within a second preset time period prior to the current moment. Typically, the second preset time period needs to be much longer than the first preset time period to ensure the model's prediction effectiveness. Furthermore, retraining the load prediction model using recent historical load data as training samples and updating the model parameters makes the updated load model more suitable for the current working scenario of the target communication network. The target predicted load is then predicted using the updated load model.

[0114] In addition, refer to Figure 9 This is a schematic diagram of the overall logic flow of the communication network load prediction method in this embodiment.

[0115] Step S200: Train the neural network model based on historical base station load data;

[0116] Step S201: Deploy the trained base station load neural network model online.

[0117] Step S202: Before predicting the base station load at the next time point T, determine whether the model parameters need to be updated.

[0118] Step S203: If it is determined that the model parameters do not need to be updated, the already trained model is used to predict the load based on historical load data before time T.

[0119] Step S204: Store and output the load of all base stations at time T;

[0120] Step S205: Continue to predict the load of all base stations at time T = T+1; and return to step S202.

[0121] Step S206: If it is determined that the model parameters need to be updated, the prediction model is retrained based on historical data before time T. Return to step S200.

[0122] It can be understood that the aforementioned historical base station load data is the historical load data of the target communication network, the neural network model is the load prediction model, and determining whether the model parameters need to be updated is to determine whether the load prediction model is adapted to the current working scenario. The historical data before time T is the recent historical load data of the target communication network.

[0123] In this embodiment, the load prediction model of the target communication network is determined to be suitable for the current working scenario of the target communication network based on the historical actual load and the historical predicted load of the historical actual load. If suitable, the target predicted load of the target communication network at the next moment is predicted based on the recent load data of the target communication network and the load prediction model, wherein the recent load data is the historical load data of the target communication network within a first preset time period before the current moment. If unsuitable, the load prediction model is updated based on the recent historical load data of the target communication network, and the target predicted load is predicted based on the updated load prediction model and the recent load data, wherein the recent historical load data is the historical load data of the target communication network within a second preset time period before the current moment, and the second preset time period is longer than the first preset time period. That is, in this application, the load prediction model at this time is used to determine whether it is suitable for the current working scenario by using the historical actual results and their corresponding historical predicted load results. If suitable, the load prediction model at this time is used for prediction. If the model is not suitable, recent historical data is used as training samples to update the load prediction model, thereby enabling real-time adjustment of the load prediction model. The prediction is then made based on the updated and adjusted model. It is understandable that real-time adjustment of model parameters allows the prediction model to better adapt to the real-time working scenario, resulting in more accurate load prediction results.

[0124] Reference Figure 3 Based on the first embodiment of the communication network load prediction method of this application, a second embodiment of the communication network load prediction method of this application is proposed. The parts identical to those in the above embodiment can be referred to the above content, and will not be repeated here. Before step A10, which determines whether the load prediction model of the target communication network is suitable for the current working scenario of the target communication network based on the historical actual load of the target communication network and the historical predicted load of the historical actual load, the method includes steps B10-B40:

[0125] Step B10: Generate training samples based on the historical load data of the target communication network;

[0126] In one feasible implementation, the target communication network includes multiple communication base stations, and the step of generating training samples based on historical load data of the target communication network includes:

[0127] Step B110: Obtain historical load data of each of the communication base stations as a training dataset;

[0128] Step B120: Extract the training samples from the training dataset based on a preset time window and the temporal order of the training dataset. Each training sample includes load data of different communication base stations at different times within the preset time window during the same period. The label of each training sample is the load data of different communication base stations at the next time in the corresponding period of the training sample.

[0129] For example, in this embodiment, training samples are generated based on historical load data of the target communication network. Historical load data refers to the load of different communication base stations under the target communication network at various past moments. This historical load data is used as the training dataset, meaning the training dataset includes the load data of each communication base station in the time domain. Training samples are extracted from the training dataset based on a preset time window and the temporal order of the training dataset. For example, training samples are extracted according to the order of the load data in the training dataset using a preset time window. Any training sample includes load data from different communication base stations at different moments within the same time period within a preset time window. The label of any training sample is the load data of different communication base stations at the next moment within the corresponding time period of the training sample. For example, assuming the interval between load data is Δt, training data samples are constructed based on a time window of size w, resulting in any training sample being {[a t+(N-w)Δt , ..., a t+(N-1)Δt ], a t+NΔt}, where a t Let a represent the load of each communication base station at time t. t+NΔt Let t+NΔt be the label, and t+NΔt is the next moment from the time interval t+(Nw)Δt to t+(N-1)Δt.

[0130] Step B20: Integrate the temporally adjacent load data in the training samples into sample features and input them into the load prediction model to obtain the predicted load of the training samples.

[0131] For example, the training samples mentioned above contain load data at different times within a time period, such as the load from time period t1 to t9. When training the model based on the training samples, load data that are temporally adjacent in the training samples are integrated into sample features and input into the load prediction model. For example, t1, t2, and t3 are integrated into one sample feature of the training sample. Correspondingly, the sample features may also include t4, t5, and t6, as well as t7, t8, and t9. All the sample features are input into the network nodes of each input layer of the load prediction model. In addition, taking the load from time period t1 to t9 as an example, the integration method can also be t1, t2, and t3; t2, t3, and t4; t3, t4, and t5; ..., t7, t8, and t9. The number of network nodes in each input layer of the load prediction model can be adjusted according to the sample feature integration method. The load prediction model then predicts the predicted load corresponding to the sample data based on the sample features.

[0132] Step B30: Calculate the model prediction loss of the load prediction model based on the predicted load and the labels of the training samples;

[0133] For example, the model prediction loss of the load prediction model is calculated based on the predicted load of the training samples and the labels of those training samples. This can be calculated using a loss function, as follows:

[0134]

[0135] In the formula, a t+ΔNt The labels for the training samples. F is the prediction load for the training samples, and F is a parameter. The loss function can also be set by technicians as needed.

[0136] Step B40: Iteratively update the load prediction model based on the model prediction loss.

[0137] In one feasible implementation, step B40, which iteratively updates the load prediction model based on the model prediction loss, includes steps B410-B440:

[0138] Step B410: Determine whether the model prediction loss has converged;

[0139] Step B420: If it is determined that the model prediction loss has converged, then the load prediction model training is complete.

[0140] Step B430: If it is determined that the model prediction loss does not converge, then the parameters in the load prediction model are updated based on the model prediction loss and the gradient descent method.

[0141] Step B440, based on the new training samples and the updated load prediction model, returns to the step of integrating temporally adjacent load data in the training samples into sample features and inputting them into the load prediction model to obtain the predicted load of the training samples, and subsequent steps.

[0142] For example, the load prediction model is iteratively updated based on the model prediction loss. First, it is determined whether the model prediction loss has converged, for example, whether the model prediction loss is less than a preset loss threshold. If it is less, the model prediction loss is considered converged, indicating that the load prediction model training is complete and the load prediction model has the ability to predict load. Conversely, the model prediction loss is considered not converged, indicating that the model's prediction ability is poor and training needs to continue. Then, the parameters in the load prediction model are updated based on the model prediction loss and gradient descent. Then, using new training samples and the updated load prediction model, the steps of integrating temporally adjacent load data in the training samples into sample features and inputting them into the load prediction model to obtain the predicted load of the training samples, and subsequent steps, are executed. This continues until the model prediction loss converges. For example, refer to... Figure 7 This is a schematic diagram of the load prediction framework for the communication network load prediction method in this application. The load prediction model in this embodiment can be a ConvLSTM neural network (convolutional neural network). Data grouping S100 in the diagram represents the process of integrating load data from any training sample to obtain sample features; that is, dividing the time-series load dataset in the training sample into multiple subsets, each subset being a sample feature. For example, dividing the time-series load dataset in a training sample into M subsets, [a t+(N-w)Δt , ..., a t+(N-1)Δt ] = [A1, ..., A M Therefore, the sample features include A1-Am, where A1 = [a t+(N-w)Δt , ..., a t+(N-w+w / M)Δt ]·, A M =[a t+(N-1-w / M)Δt , ..., a t+(N-1)Δt The sample features A1-Am of the training samples are then input into a ConvLSTM neural network for training. By grouping the input time-series load dataset, the convolutional capability of ConvLSTM can extract the load relationship between base stations at adjacent time points, thereby improving the accuracy of its prediction. The ConvLSTM neural network obtains the prediction result based on the input A1-Am. Again and predicting target a t+ΔNtThe labels of the training samples (i.e., the model's predicted loss) are calculated using the loss function S102. Then, based on this predicted loss, gradient descent is used to determine the network parameters, which are then used to update the model. For any ConvLSTM unit, such as ConvLSTM unit S101, refer to... Figure 8 Here is a diagram of the internal structure of a ConvLSTM cell. The internal state and output of the ConvLSTM at any time t are calculated as follows:

[0143]

[0144]

[0145]

[0146]

[0147]

[0148] In the formula, * denotes the convolution operation. Let W represent the Hadamard product, and σ represent the activation function. cf W ci W co W hf W Af b f W hi W Ai b i W hc W Ac b c W ho W Ac b c W ho W Ao b o These are the parameters within the ConvLSTM cell. C t-1 H t-1 A t H is the input parameter. t and C t For output parameters, f t i t and o t Tanh is the intermediate parameter, and the other internal structures of the ConvLSTM unit are the same as those of the existing convolutional neural network. In fact, the internal structure of the ConvLSTM unit can also refer to the structure of the existing convolutional neural network, which will not be elaborated here.

[0149] Reference Figure 4Based on the third embodiment of the communication network load prediction method of this application, a fourth embodiment of the communication network load prediction method of this application is proposed. The parts identical to those in the above embodiments can be referred to the above content and will not be repeated here. The communication network load prediction method is applied to the digital twin network of the target communication network. Before the step of determining whether the load prediction model of the target communication network is suitable for the current working scenario of the target communication network based on the historical actual load of the target communication network and the historical predicted load of the historical actual load, the method further includes:

[0150] Step C10: Obtain historical load data of the target communication network based on the digital twin network;

[0151] Step C20: Extract the recent historical load data in real time from the latest historical load data in the historical load data using a preset time sliding window.

[0152] It should be noted that in practical applications, a mobile communication network includes K ground base station nodes (i.e., base stations) and one central cloud server node. The base station nodes in the mobile communication network upload status data to the central cloud server in real time. The central cloud server node is responsible for generating a digital twin network of the mobile communication network. The base station twins in the digital twin network have a high degree of consistency with the actual base stations in the real environment. That is, after the status data uploaded by the base station nodes is input into the digital twin network, the digital twin network can simulate the real feedback of the communication network under the condition of uploaded status data, thereby helping network operation and maintenance decisions. The digital twin network also stores the status data (including load data) uploaded by each base station. Therefore, the above-mentioned communication network load method in this embodiment can be applied to the digital twin network of the target communication network. Wherein, any ground base station node k, in any digital twin network, is defined as: In the formula, l k This represents the actual location coordinates of base station node k. This represents the communication service range of base station k, which is generally determined according to the Euclidean distance principle and does not change over time. a(t) k This represents the communication load of base station k at time t, reflecting the frequency of users currently accessing the base station, and is a factor influencing base station k's decision d at time t. k Key performance indicators for (t) (such as resource allocation, base station hibernation, etc.), a(t+NΔt) k This represents the base station communication load at the next time interval NΔt, where N is a positive integer. k (t+NΔt) represents the decision action of base station k in the next time interval. The digital twin load of all base stations can be represented as a. t = [a1(t), a2(t), ..., aK (t)] T The entire digital twin system framework is as follows: Figure 10 As shown in the figure, there is a mapping relationship between the digital twin network and the base stations in the real target communication network. After configuring the above-mentioned communication network load prediction method in the digital twin network, the target network's future load data (i.e., the result of data inference) can be predicted using this method. The predicted load data is then input into the decision module, which generates a decision and sends it to the digital twin network. The digital twin network can simulate the state of the target communication network after executing the decision, thereby providing feedback to the decision. This feedback can then be used to guide the generation of further decisions.

[0153] For example, historical load data of the target communication network can be obtained through a digital twin network. Recent historical load data can be extracted in real time from the latest historical load data using a preset time sliding window. For instance, suppose there are historical load data: a1, a2, ..., an, an+1, an+2, ..., am. Here, a1 is the latest historical load data, and the preset time sliding window size is n. Then, the recent historical load data extracted from the latest historical load data based on this preset time sliding window is: a1, a2, ..., an. It is understood that recent historical load data has relatively strong real-time characteristics and can reflect the characteristic changes in the target communication model's load over a recent period. Therefore, updating the load prediction model using recent historical load data allows the updated load prediction model to adapt to the working scenario of the target communication network.

[0154] According to Figure 5 Based on the second embodiment of the communication network load prediction method of this application, a third embodiment of the communication network load prediction method of this application is proposed. The parts identical to those in the above embodiments can be referred to the above content, and will not be repeated here. The step A30, which updates the load prediction model based on the recent historical load data of the target communication network, includes steps A310-A320:

[0155] Step A310: Obtain recent historical load data of each of the target communication networks as the training dataset;

[0156] Step A320: Execute the step of extracting the training samples from the training dataset based on the preset interval and the temporal order of the training dataset, and subsequent steps.

[0157] For example, in this embodiment, when it is determined that the load prediction model is incompatible with the current operating scenario of the target communication network, recent historical load data of the communication base station is obtained and used as a training dataset. Based on this training dataset, the steps of extracting training samples from the training dataset based on a preset interval and the temporal order of the training dataset, and subsequent steps, are performed to complete the training and update of the load model. The specific process can be referred to the above content and will not be repeated here.

[0158] Furthermore, embodiments of this application also provide a communication network load prediction device, the communication network load prediction device comprising:

[0159] The judgment module is used to determine whether the load prediction model of the target communication network is suitable for the current working scenario of the target communication network based on the historical actual load of the target communication network and the historical predicted load of the historical actual load.

[0160] The first prediction module is used, if adapted, to predict the target predicted load of the target communication network at the next moment based on the recent load data of the target communication and the load prediction model, wherein the recent load data is the historical load data of the target communication network within a first preset time period before the current moment;

[0161] The second prediction module is used to update the load prediction model based on the recent historical load data of the target communication network if it is not suitable, and to predict the target predicted load based on the updated load prediction model and the recent load data. The recent historical load data is the historical load data of the target communication network within a second preset time period before the current time, and the second preset time period is longer than the first preset time period.

[0162] Optionally, the determination module is further configured to:

[0163] The first load vector feature is obtained by extracting vector features from the historical actual load.

[0164] The second load vector feature is obtained by extracting vector features from the historical predicted load;

[0165] Calculate the similarity between the first load vector feature and the second load vector feature;

[0166] If the similarity is greater than a preset similarity threshold, then the load prediction model is determined to be suitable for the current working scenario of the target communication network.

[0167] If the similarity is less than or equal to the preset similarity threshold, then the load prediction model is determined to be unsuitable for the current working scenario of the target communication network.

[0168] Optionally, the communication network load prediction device further includes a first training module, the training module being used for:

[0169] Training samples are generated based on the historical load data of the target communication network.

[0170] The load data that are temporally adjacent in the training samples are integrated into sample features and input into the load prediction model to obtain the predicted load of the training samples.

[0171] The model prediction loss of the load prediction model is calculated based on the predicted load and the labels of the training samples;

[0172] The load prediction model is iteratively updated based on the predicted loss from the model.

[0173] Optionally, the first training module is further configured to:

[0174] Determine whether the model's prediction loss has converged;

[0175] If the model prediction loss is determined to converge, then the load prediction model training is complete.

[0176] If it is determined that the model prediction loss does not converge, the parameters in the load prediction model are updated based on the model prediction loss and the gradient descent method.

[0177] Based on the new training samples and the updated load prediction model, return to the step of integrating temporally adjacent load data in the training samples into sample features and inputting them into the load prediction model to obtain the predicted load of the training samples, and subsequent steps.

[0178] Optionally, the target communication network includes multiple communication base stations, and the first training module is further configured to:

[0179] Historical load data of each of the aforementioned communication base stations are obtained as a training dataset;

[0180] The training samples are extracted from the training dataset based on a preset time window and the temporal order of the training dataset. Each training sample includes load data of different communication base stations at different times within the preset time window during the same period. The label of each training sample is the load data of different communication base stations at the next time in the corresponding period of the training sample.

[0181] Optionally, the communication network load prediction device further includes a second training module, the second training module being used for:

[0182] The recent historical load data of each of the target communication networks is obtained as the training dataset;

[0183] The steps of extracting the training samples from the training dataset based on the preset interval and the temporal order of the training dataset, and subsequent steps, are performed.

[0184] Optionally, the communication network load prediction device further includes an acquisition module, the acquisition module being used for:

[0185] Historical load data of the target communication network is obtained based on the digital twin network;

[0186] The recent historical load data is extracted in real time from the latest historical load data in the historical load data by using a preset time sliding window.

[0187] The communication network load prediction device provided in this application employs the communication network load prediction method in the above embodiments, aiming to improve the accuracy of future communication load prediction results for mobile communication networks. Compared with the prior art, the beneficial effects of the communication network load prediction device provided in this application are the same as those of the communication network load prediction method provided in the above embodiments, and other technical features in this communication network load prediction device are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0188] In addition, to achieve the above objectives, this application also provides a communication network load prediction device, which includes: a memory, a processor, and a communication network load prediction program stored in the memory and executable on the processor. When the communication network load prediction program is executed by the processor, it implements the steps of the communication network load prediction method as described above.

[0189] The specific implementation of the communication network load prediction device in this application is basically the same as the embodiments of the communication network load prediction method described above, and will not be repeated here.

[0190] In addition, to achieve the above objectives, this application also provides a computer medium storing a communication network load prediction program, which, when executed by a processor, implements the steps of the communication network load prediction method as described above.

[0191] The specific implementation of the computer medium in this application is basically the same as the embodiments of the above-described communication network load prediction method, and will not be repeated here.

[0192] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0193] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0194] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, 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 is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0195] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A method for predicting the load of a communication network, characterized in that, The communication network load prediction method includes the following steps: Based on the historical actual load of the target communication network and the historical predicted load of the historical actual load, it is determined whether the load prediction model of the target communication network is suitable for the current working scenario of the target communication network, wherein the load prediction model is a convolutional neural network; If compatible, the target predicted load of the target communication network at the next moment is predicted based on the recent load data of the target communication and the load prediction model, wherein the recent load data is the historical load data of the target communication network within a first preset time period before the current moment; If it is not suitable, the load prediction model is updated based on the recent historical load data of the target communication network, and the target predicted load is predicted based on the updated load prediction model and the recent load data. The recent historical load data is the historical load data of the target communication network within a second preset time period before the current time, and the second preset time period is longer than the first preset time period. The target communication network includes multiple communication base stations. Before the step of determining whether the load prediction model of the target communication network is suitable for the current working scenario of the target communication network based on the historical actual load of the target communication network and the historical predicted load of the historical actual load, the communication network load prediction method further includes: Historical load data of each of the aforementioned communication base stations are obtained as a training dataset; Training samples are extracted from the training dataset based on a preset time window and the temporal order of the training dataset. Each training sample includes load data of different communication base stations at different times within the preset time window during the same period. The label of each training sample is the load data of different communication base stations at the next time in the corresponding period of the training sample. The load data that are temporally adjacent in the training samples are integrated into sample features and input into the load prediction model to obtain the predicted load of the training samples; The model prediction loss of the load prediction model is calculated based on the predicted load and the labels of the training samples; The load prediction model is iteratively updated based on the model prediction loss. The communication network load prediction method is applied to the digital twin network of the target communication network. For any ground base station k in the target communication network, its digital twin in the digital twin network is: ,in, This represents the actual location coordinates of base station k. This indicates the communication service range of base station k. This represents the communication load of base station k at time t. This represents the decision action of base station k at time t. Indicates the time interval from the current time. The communication load of base station k at the next moment. Indicates the time interval from the current time. The decision action of base station k at the next moment; before the step of determining whether the load prediction model of the target communication network is suitable for the current working scenario of the target communication network based on the historical actual load of the target communication network and the historical predicted load of the historical actual load, the method further includes: Historical load data of the target communication network is obtained based on the digital twin network; The recent historical load data is extracted in real time from the latest historical load data in the historical load data by using a preset time sliding window.

2. The communication network load prediction method as described in claim 1, characterized in that, The steps for determining whether the load prediction model of the target communication network is suitable for the current working scenario of the target communication network based on the historical actual load of the target communication network and the historical predicted load of the historical actual load include: The first load vector feature is obtained by extracting vector features from the historical actual load. The second load vector feature is obtained by extracting vector features from the historical predicted load; Calculate the similarity between the first load vector feature and the second load vector feature; If the similarity is greater than a preset similarity threshold, then the load prediction model is determined to be suitable for the current working scenario of the target communication network. If the similarity is less than or equal to the preset similarity threshold, then the load prediction model is determined to be unsuitable for the current working scenario of the target communication network.

3. The communication network load prediction method as described in claim 1, characterized in that, The step of iteratively updating the load prediction model based on the model prediction loss includes: Determine whether the model's prediction loss has converged; If the model prediction loss is determined to converge, then the load prediction model training is complete. If it is determined that the model prediction loss does not converge, the parameters in the load prediction model are updated based on the model prediction loss and the gradient descent method. Based on the new training samples and the updated load prediction model, return to the step of integrating temporally adjacent load data in the training samples into sample features and inputting them into the load prediction model to obtain the predicted load of the training samples, and subsequent steps.

4. The communication network load prediction method as described in claim 3, characterized in that, The step of updating the load prediction model based on recent historical load data of the target communication network includes: The recent historical load data of each of the target communication networks is obtained as the training dataset; The steps of extracting training samples from the training dataset based on a preset time window and the time sequence of the training dataset, and subsequent steps, are performed.

5. A communication network load prediction device, characterized in that, The communication network load prediction device includes: The judgment module is used to determine whether the load prediction model of the target communication network is suitable for the current working scenario of the target communication network based on the historical actual load of the target communication network and the historical predicted load of the historical actual load. The target communication network includes multiple communication base stations, and the load prediction model is a convolutional neural network. The first prediction module is used, if adapted, to predict the target predicted load of the target communication network at the next moment based on the recent load data of the target communication and the load prediction model, wherein the recent load data is the historical load data of the target communication network within a first preset time period before the current moment; The second prediction module is used to update the load prediction model based on the recent historical load data of the target communication network if it is not suitable, and to predict the target predicted load based on the updated load prediction model and the recent load data. The recent historical load data is the historical load data of the target communication network within a second preset time period before the current time, and the second preset time period is longer than the first preset time period. The first training module is used to acquire historical load data of each communication base station as a training dataset; extract training samples from the training dataset based on a preset time window and the temporal order of the training dataset, wherein any training sample includes load data of different communication base stations at different times within the preset time window during the same period, and the label of any training sample is the load data of different communication base stations at the next time of the corresponding period of the training sample; integrate temporally adjacent load data in the training samples into sample features and input them into the load prediction model to obtain the predicted load of the training samples; calculate the model prediction loss of the load prediction model based on the predicted load and the label of the training samples; and iteratively update the load prediction model based on the model prediction loss. The acquisition module is used to acquire historical load data of the target communication network based on the digital twin network; and to extract recent historical load data in real time from the latest historical load data through a preset time sliding window, wherein any ground base station k in the target communication network has a digital twin in the digital twin network as follows: ,in, This represents the actual location coordinates of base station k. This indicates the communication service range of base station k. This represents the communication load of base station k at time t. This represents the decision action of base station k at time t. Indicates the time interval from the current time. The communication load of base station k at the next moment. Indicates the time interval from the current time. The decision action of base station k at the next moment.

6. A communication network load prediction device, characterized in that, The communication network load prediction device includes: a memory, a processor, and a communication network load prediction program stored in the memory and executable on the processor. When the communication network load prediction program is executed by the processor, it implements the steps of the communication network load prediction method as described in any one of claims 1 to 4.

7. A readable storage medium, characterized in that, The readable storage medium stores a communication network load prediction program, which, when executed by a processor, implements the steps of the communication network load prediction method as described in any one of claims 1 to 4.

Images