Bandwidth prediction method, device and electronic equipment

Abstract

The present application discloses a bandwidth prediction method, device and electronic device, the method comprising: predicting the bandwidth demand of the sub-service corresponding to the user in the first preset time period according to the user's network behavior, and predicting the real-time quality of the user network corresponding to the user in the second preset time period according to the physical layer signal characteristic data, and predicting the real-time bandwidth of the user network in the second preset time period; combining the bandwidth demand prediction result, the network real-time quality prediction result and the real-time bandwidth prediction result, predicting the bandwidth value that the user network can provide in the second preset time period. The technical solution provided by the embodiment of the present application takes into account the influence of the user's network behavior and the physical layer signal fluctuation on the bandwidth, so that the accuracy of bandwidth prediction can be significantly improved.

Description

Technical Field

[0001] The present application relates to the technical field of data transmission, and in particular to a bandwidth prediction method, device and electronic device. Background Art

[0002] In network data transmission, especially in real-time streaming data transmission, accurate bandwidth prediction is the core of algorithms such as congestion control and bitrate adaptation, and is crucial in encoding and transmission strategies. In addition, accurate bandwidth prediction can optimize the user experience quality in the network environment. Therefore, accurate bandwidth prediction is of great significance.

[0003] At present, the bandwidth prediction method usually adopts a machine learning prediction method, such as using a reinforcement learning model for bandwidth prediction, or combining a machine learning bandwidth prediction method with a rule-based bandwidth prediction method through decision-level or feature-level fusion to achieve bandwidth prediction.

[0004] However, the above methods do not take into account the impact of user personalized network behavior and physical layer signal fluctuations on bandwidth, resulting in inaccurate bandwidth prediction results (i.e., the bandwidth value that the network can provide in the future). Summary of the invention

[0005] The present application provides a bandwidth prediction method, device and electronic device to solve the problem that the existing technology ignores the influence of user network behavior and physical layer signal fluctuation on bandwidth, resulting in low bandwidth prediction accuracy. The specific implementation scheme is as follows:

[0006] In a first aspect, the present application provides a bandwidth prediction method, the method comprising:

[0007] Predicting bandwidth requirements of the sub-service corresponding to the user in a first preset time period according to the user's network behavior, and predicting the real-time quality of the user network corresponding to the user in a second preset time period according to the physical layer signal characteristic data, and predicting the real-time bandwidth of the user network in the second preset time period;

[0008] Combined with the bandwidth demand prediction result corresponding to the bandwidth demand of the sub-service, the network real-time quality prediction result corresponding to the real-time quality of the user network, and the real-time bandwidth prediction result corresponding to the real-time bandwidth of the user network, the bandwidth value that the user network can provide in the second preset time period is predicted.

[0009] Through the user's network behavior, the bandwidth demand of the user's corresponding sub-service in the first preset time period is predicted, and the user's long-term network behavior can be perceived. Combined with the real-time network quality prediction result of the second preset time period predicted by the physical layer signal characteristic data, and combined with the predicted real-time bandwidth prediction result of the second preset time period, a bandwidth prediction result that is closer to the user's bandwidth demand is provided (that is, the bandwidth value that the user network can provide in the second preset time period), thereby taking into account the user's network behavior and the impact of physical layer signal fluctuations on bandwidth, so that the accuracy of bandwidth prediction is significantly improved.

[0010] In a possible implementation, predicting the bandwidth demand of the sub-service corresponding to the user in the first preset time period according to the network behavior of the user includes:

[0011] Collecting the network service data of the user, and obtaining the statistical results of the network bandwidth demand of the sub-service corresponding to the user in different time periods according to the network service data; wherein the network service data represents the network behavior of the user;

[0012] Selecting a bandwidth demand prediction model according to the statistical results of the network bandwidth demand of the sub-service in different time periods, and predicting the bandwidth demand of the sub-service in the first preset time period by using the bandwidth demand prediction model to obtain a first bandwidth demand prediction result; and

[0013] Obtaining a second bandwidth demand prediction result of the sub-service in the first preset time period according to a fusion of the network bandwidth demand statistics results of the sub-service in different time periods;

[0014] The selected first bandwidth demand prediction result and the second bandwidth demand prediction result are integrated to obtain the bandwidth demand prediction result of the sub-service in the first preset time period.

[0015] Through the above application embodiments, long-term time-segment prediction of the statistical results of the user's sub-service network bandwidth demand is achieved, so that the user's long-term network behavior can be characterized, and more personalized network bandwidth prediction and allocation strategies can be provided for the subsequent period.

[0016] In a possible implementation manner, predicting the real-time quality of the user network corresponding to the user in the second preset time period according to the physical layer signal characteristic data includes:

[0017] Collecting the physical layer signal characteristic data and the network indicator data of the user network;

[0018] Processing the physical layer signal feature data and the network indicator data through an individual perception model based on incremental learning to obtain a first network real-time quality prediction result of the user network in the second preset time period;

[0019] The first network real-time quality prediction result is processed by a universal aggregated perception model to obtain a second network real-time quality prediction result of the user network in the second preset time period; wherein the universal aggregated perception model is a model obtained by processing the individual perception model based on incremental learning in a federated learning manner;

[0020] The first network real-time quality prediction result and the second network real-time quality prediction result are integrated to obtain the network real-time quality prediction result of the user network in the second preset time period.

[0021] Through the above application embodiments, combined with the physical layer signal characteristic data, the network quality of the user network (i.e., the real-time network quality prediction result) is perceived in real time, so that the impact of the physical layer signal characteristic data on the bandwidth can be considered, which helps to improve the accuracy of the bandwidth value that the user network can provide in the second preset time period. And through the prediction of the real-time network quality by the individual perception model and the prediction of the real-time network quality by the general aggregate perception model, the user's individual network quality perception ability and the network's general information perception ability are enhanced, and the actual network changes can be perceived more timely, further improving the accuracy of the real-time prediction of the user network's network quality.

[0022] In a possible implementation manner, the predicting the real-time bandwidth of the user network in the second preset time period includes:

[0023] Collecting network indicator data of the user network;

[0024] Based on the network indicator data, predicting the real-time bandwidth of the user network in the second preset time period by using a reinforcement learning model and a recursive least squares RLS model respectively, to obtain a first real-time bandwidth prediction result corresponding to the reinforcement learning model and a second real-time bandwidth prediction result corresponding to the RLS model;

[0025] The first real-time bandwidth prediction result and the second real-time bandwidth prediction result are integrated to obtain the real-time bandwidth prediction result of the user network in the second preset time period.

[0026] By predicting the real-time bandwidth of the user network in the second preset time period using a reinforcement learning model, the accuracy of the first real-time bandwidth prediction result is guaranteed. By predicting the real-time bandwidth of the user network in the second preset time period using an RLS model, the accuracy of the second real-time bandwidth prediction result is guaranteed. Based on the combination of the first real-time bandwidth prediction result and the second real-time bandwidth prediction result, the accuracy of the real-time bandwidth prediction result of the user network in the second preset time period is further improved.

[0027] In a possible implementation manner, before predicting the bandwidth requirement of the sub-service corresponding to the user in the first preset time period according to the network behavior of the user, predicting the real-time quality of the user network corresponding to the user in the second preset time period according to the physical layer signal characteristic data, and predicting the real-time bandwidth of the user network in the second preset time period, the method further includes:

[0028] updating the model parameters of the bandwidth demand prediction model by training the bandwidth demand prediction model at first preset intervals; and

[0029] updating the model parameters of the individual perception model based on incremental learning by training the individual perception model based on incremental learning at every second preset time, and updating the model parameters of the general aggregation model by training the general aggregation model at every third preset time; and

[0030] Every fourth preset time, the model parameters of the reinforcement learning model are updated through the training of the reinforcement learning model, and every fifth preset time, the model parameters of the RLS model are updated through the training of the RLS model.

[0031] Through the above-mentioned application embodiments, the ability to update model parameters is provided, and the adaptability of the model to a changing network environment is enhanced, thereby helping to further improve the prediction accuracy of real-time bandwidth.

[0032] In a possible implementation manner, the combining the bandwidth demand prediction result corresponding to the bandwidth demand of the sub-service, the network real-time quality prediction result corresponding to the real-time quality of the user network, and the real-time bandwidth prediction result corresponding to the real-time bandwidth of the user network to predict the bandwidth value that the user network can provide in the second preset time period includes:

[0033] Combining the bandwidth demand prediction result corresponding to the bandwidth demand of the sub-service with the network real-time quality prediction result corresponding to the real-time quality of the user network, obtaining a maximum bandwidth demand result for the second preset time period;

[0034] The maximum bandwidth requirement result is combined with the real-time bandwidth prediction result corresponding to the real-time bandwidth of the user network to predict the bandwidth value that the user network can provide in the second preset time period.

[0035] Through the above application embodiment, the bandwidth demand prediction results of the sub-services, the network real-time quality prediction results of the user network and the real-time bandwidth prediction results of the user network are combined to accurately predict the bandwidth value that the user network can provide. And under limited network resources, the user service demand can be prioritized to achieve accurate prediction of the bandwidth value.

[0036] In a possible implementation manner, after predicting the bandwidth value that can be provided by the user network in the second preset time period, the method further includes:

[0037] Selecting a target bandwidth allocation strategy based on the bandwidth demand prediction result of the sub-service in the first preset time period and the bandwidth value that can be provided by the user network in the second preset time period;

[0038] Bandwidth allocation is performed according to the target bandwidth allocation strategy.

[0039] Through the above application embodiment, reasonable allocation of bandwidth is achieved according to the bandwidth demand prediction result of the sub-service in the first preset time period and the bandwidth value that can be provided by the user network in the second preset time period.

[0040] In a second aspect, the present application further provides a bandwidth prediction device, the device comprising:

[0041] A prediction module, configured to predict the bandwidth demand of the sub-service corresponding to the user in a first preset time period according to the network behavior of the user, and to predict the real-time quality of the user network corresponding to the user in a second preset time period according to the physical layer signal characteristic data, and to predict the real-time bandwidth of the user network in the second preset time period;

[0042] The processing module is used to combine the bandwidth demand prediction result corresponding to the bandwidth demand of the sub-service, the network real-time quality prediction result corresponding to the real-time quality of the user network, and the real-time bandwidth prediction result corresponding to the real-time bandwidth of the user network, to predict the bandwidth value that the user network can provide in the second preset time period.

[0043] In a possible implementation, the prediction module is specifically used to collect the network service data of the user, and obtain the statistical results of the network bandwidth demand of the sub-service corresponding to the user in different time periods according to the network service data; wherein the network service data represents the network behavior of the user;

[0044] Selecting a bandwidth demand prediction model according to the statistical results of the network bandwidth demand of the sub-service in different time periods, and predicting the bandwidth demand of the sub-service in the first preset time period by using the bandwidth demand prediction model to obtain a first bandwidth demand prediction result; and

[0045] Obtaining a second bandwidth demand prediction result of the sub-service in the first preset time period according to a fusion of the network bandwidth demand statistics results of the sub-service in different time periods;

[0046] The selected first bandwidth demand prediction result and the second bandwidth demand prediction result are integrated to obtain the bandwidth demand prediction result of the sub-service in the first preset time period.

[0047] In a possible implementation manner, the prediction module is specifically used to collect the physical layer signal characteristic data and the network indicator data of the user network;

[0048] Processing the physical layer signal feature data and the network indicator data through an individual perception model based on incremental learning to obtain a first network real-time quality prediction result of the user network in the second preset time period;

[0049] The first network real-time quality prediction result is processed by a universal aggregated perception model to obtain a second network real-time quality prediction result of the user network in the second preset time period; wherein the universal aggregated perception model is a model obtained by processing the individual perception model based on incremental learning in a federated learning manner;

[0050] The first network real-time quality prediction result and the second network real-time quality prediction result are integrated to obtain the network real-time quality prediction result of the user network in the second preset time period.

[0051] In a possible implementation, the prediction module is specifically used to collect network indicator data of the user network;

[0052] Based on the network indicator data, predicting the real-time bandwidth of the user network in the second preset time period by using a reinforcement learning model and a recursive least squares RLS model respectively, to obtain a first real-time bandwidth prediction result corresponding to the reinforcement learning model and a second real-time bandwidth prediction result corresponding to the RLS model;

[0053] The first real-time bandwidth prediction result and the second real-time bandwidth prediction result are integrated to obtain the real-time bandwidth prediction result of the user network in the second preset time period.

[0054] In a possible implementation manner, the prediction module is specifically configured to update the model parameters of the bandwidth demand prediction model by training the bandwidth demand prediction model at every first preset time; and

[0055] updating the model parameters of the individual perception model based on incremental learning by training the individual perception model based on incremental learning at every second preset time, and updating the model parameters of the general aggregation model by training the general aggregation model at every third preset time; and

[0056] Every fourth preset time, the model parameters of the reinforcement learning model are updated through the training of the reinforcement learning model, and every fifth preset time, the model parameters of the RLS model are updated through the training of the RLS model.

[0057] In a possible implementation, the processing module is specifically used to combine the bandwidth demand prediction result corresponding to the bandwidth demand of the sub-service with the network real-time quality prediction result corresponding to the real-time quality of the user network to obtain the maximum bandwidth demand result of the second preset time period;

[0058] The maximum bandwidth requirement result is combined with the real-time bandwidth prediction result corresponding to the real-time bandwidth of the user network to predict the bandwidth value that the user network can provide in the second preset time period.

[0059] In a possible implementation manner, the processing module is specifically configured to select a target bandwidth allocation strategy by combining the bandwidth demand prediction result of the sub-service in the first preset time period with the bandwidth value that can be provided by the user network in the second preset time period;

[0060] Bandwidth allocation is performed according to the target bandwidth allocation strategy.

[0061] In a third aspect, the present application provides an electronic device, including:

[0062] Memory, used to store computer programs;

[0063] The processor is used to implement the above-mentioned bandwidth prediction method steps when executing the computer program stored in the memory.

[0064] In a fourth aspect, the present application provides a computer-readable storage medium, in which a computer program is stored. When the computer program is executed by a processor, the above-mentioned bandwidth prediction method steps are implemented.

[0065] For each aspect from the second to the fourth aspect and the technical effects that may be achieved by each aspect, please refer to the above description of the technical effects that can be achieved by the first aspect or various possible schemes in the first aspect, and no further details will be given here. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 A schematic diagram of a media data transmission process provided in an embodiment of the present application;

[0067] Figure 2 A schematic diagram of bandwidth occupancy statistics provided in an embodiment of the present application;

[0068] Figure 3 A flowchart of a bandwidth prediction method provided in an embodiment of the present application;

[0069] Figure 4 A flowchart of bandwidth demand prediction provided for an embodiment of the present application;

[0070] Figure 5 A flowchart of real-time network quality prediction provided by an embodiment of the present application;

[0071] Figure 6 A flowchart of real-time bandwidth prediction provided by an embodiment of the present application;

[0072] Figure 7 A schematic diagram of the processing process of the bandwidth prediction method provided in an embodiment of the present application;

[0073] Figure 8 A schematic diagram of a bandwidth prediction device provided in an embodiment of the present application;

[0074] Fig. 9 A schematic diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0075] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings. The specific operating methods in the method embodiments can also be applied to device embodiments or system embodiments. It should be noted that in the description of the present application, "multiple" is understood as "at least two". "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. A is connected to B, which can represent: A is directly connected to B and A is connected to B through C. In addition, in the description of the present application, words such as "first" and "second" are only used to distinguish the purpose of description, and cannot be understood as indicating or implying relative importance, nor can they be understood as indicating or implying order.

[0076] The embodiments of the present application are described in detail below in conjunction with the accompanying drawings.

[0077] With the development of the economy, a user may have multiple terminal devices, and because of the different needs of individual users, their usage behaviors on different terminal devices will also be quite different. With the advent of the self-media era, everyone is a producer and consumer of media data. The media services used by everyone include but are not limited to the following media services: video on demand, short video push, lossless music playback, high-definition video live broadcast, video conferencing, infant and elderly patient monitoring, and immersive virtual reality enhanced experience, etc.

[0078] Real-time media data transmission process such as Figure 1 As shown, media data is generated in real time by the sender, and after encoding, a flow control algorithm (i.e., media data flow control algorithm) generates a sending strategy, and then the media data is transmitted to the receiver via the Internet for real-time media data parsing and playback.

[0079] In order to ensure that users have a good experience on terminal devices, a key technical point is that the flow control algorithm can accurately predict the real-time bandwidth.

[0080] However, each user has different media services and usage habits. For example, user A prefers to use service A (such as a service for watching live broadcasts) from 21:00 to 22:00 every night, and service B (such as a service for listening to lossless music in real time) from 22:00 to 22:30; while user B is accustomed to using service C (such as a service for online on-demand video) from 21:30 to 22:30.

[0081] In a possible scenario, the bandwidth usage statistics of four services (service A, service B, service C, and service D) used by a user during the period from 18:30 to 22:00 in the evening are stacked as shown in Figure 2. Then, Figure 2 The data shown generates priorities for different services at different time periods.

[0082] Therefore, a solution is needed to perceive the user's service needs (such as the user's usage habits for the various media services they use) and predict the user's service demand bandwidth (that is, the bandwidth demand of the service used by the user). Then, based on the existing network conditions perceived by the device, a solution is needed to perform bandwidth prediction based on the user's service demand priority under limited network resources, and serve as the basis for subsequent bandwidth allocation.

[0083] However, existing bandwidth prediction methods do not take into account the user's service needs and the impact of network conditions (such as physical layer signal fluctuations) on bandwidth, resulting in inaccurate bandwidth prediction results.

[0084] Therefore, the present application proposes a bandwidth prediction method, which first predicts the bandwidth demand of the sub-service corresponding to the user in the first preset time period according to the user's network behavior, predicts the real-time quality of the user network in the second preset time period according to the physical layer signal characteristic data, and predicts the real-time bandwidth of the user network in the second preset time period, and then combines the prediction result of the bandwidth demand (i.e., the bandwidth demand prediction result), the prediction result of the real-time quality of the user network (i.e., the network real-time quality prediction result) and the prediction result of the real-time bandwidth of the user network (i.e., the real-time bandwidth prediction result) to predict the bandwidth value that the user network can provide in the second preset time period, thereby taking into account the user's service demand (i.e., the user's personalized network behavior) and the impact of network conditions (such as physical layer signal fluctuations) on bandwidth, and providing a bandwidth prediction result (i.e., the bandwidth value that can be provided) that is closer to the user's bandwidth demand, so that the prediction result of the bandwidth value is more accurate, which is conducive to improving the user experience quality in network data transmission.

[0085] Reference Figure 3 FIG. 1 is a flow chart of a bandwidth prediction method provided in an embodiment of the present application, the method comprising:

[0086] S301, predicting the bandwidth demand of the sub-service corresponding to the user in the first preset time period based on the user's network behavior, and predicting the real-time quality of the user network corresponding to the user in the second preset time period based on the physical layer signal characteristic data, and predicting the real-time bandwidth of the user network in the second preset time period.

[0087] The first preset time period and the second preset time period may be the same or different and may be adjusted according to specific application scenarios.

[0088] S302, predicting the bandwidth value that the user network can provide in the second preset time period by combining the bandwidth demand prediction result corresponding to the bandwidth demand of the sub-service, the network real-time quality prediction result corresponding to the real-time quality of the user network, and the real-time bandwidth prediction result corresponding to the real-time bandwidth of the user network.

[0089] In order to achieve accurate prediction of the bandwidth value that can be provided by the user network corresponding to the user in the second preset time period, the bandwidth demand of the sub-service corresponding to the user in the first preset time period is first predicted, so as to perceive the user's long-term network behavior and predict the real-time quality of the user network in the second preset time period and the real-time bandwidth of the user network in the second preset time period.

[0090] In an embodiment of the present application, a target model may be used to predict the bandwidth requirement of a sub-service in a first preset period, the real-time quality of a user network in a second preset period, and the real-time bandwidth of a user network in the second preset period.

[0091] The above-mentioned target model at least includes: a bandwidth demand prediction model for predicting the bandwidth demand of the sub-service in the first preset time period, an individual perception model and a general aggregate perception model based on incremental learning for predicting the real-time quality of the user network in the second preset time period, and a reinforcement learning model and a recursive least squares algorithm (English: Recursive Least Squares algorithm, abbreviated as RLS) model for predicting the real-time bandwidth of the user network in the second preset time period.

[0092] The general aggregated perception model may be obtained by processing a large number of individual perception models based on incremental learning by using a federated learning method to process each individual perception model based on incremental learning.

[0093] The reinforcement learning model and the RLS model for predicting the real-time bandwidth of the user network in the second preset time period may be used as a time series prediction model.

[0094] In order to further improve the prediction accuracy of bandwidth demand prediction model on bandwidth demand, the prediction accuracy of individual perception model and general aggregate perception model based on incremental learning on network real-time quality, and the prediction accuracy of reinforcement learning model and RLS model on network real-time bandwidth, the aforementioned models can be trained separately at preset time intervals, so that the model parameters are updated through model training, so that the model parameters of the model used to predict the bandwidth demand of the first preset time period, the network real-time quality and the network real-time bandwidth of the second preset time period are the latest version of the model parameters, so that the model used to predict the bandwidth demand of the first preset time period, the network real-time quality and the network real-time bandwidth of the second preset time period is the optimal model within the time period, so as to further improve the prediction accuracy of bandwidth demand, network real-time quality and network real-time bandwidth.

[0095] Furthermore, by updating the model parameters, the model parameters of the above model can change with the changes of the network, so that it can have excellent adaptability in a changeable network.

[0096] Therefore, in an embodiment of the present application, the ability to update model parameters (such as model parameters of a bandwidth demand prediction model, model parameters of an individual perception model based on incremental learning, model parameters of a general aggregate perception model, model parameters of a reinforcement learning model, and model parameters of an RLS model) is also provided, thereby enhancing the model's ability to adapt to a changing network environment, and helping to further improve the accuracy of predicting the bandwidth value that can be provided by the user network.

[0097] Specifically, the bandwidth demand prediction model is trained based on the first training data at every first preset time, so that the model parameters of the bandwidth demand prediction model are updated through the training of the bandwidth demand prediction model.

[0098] Every second preset time, the individual perception model based on incremental learning is trained based on the second training data, so that the model parameters of the individual perception model based on incremental learning are updated through the training of the individual perception model based on incremental learning.

[0099] The individual perception model based on incremental learning can be deployed on the user side. After updating the model parameters of the individual perception model based on incremental learning, the model parameters of the individual perception model based on incremental learning can also be uploaded to the cloud, so that the cloud can obtain the latest model parameters of the individual perception model based on incremental learning in a timely manner.

[0100] The universal aggregated perception model is trained based on the third training data at every third preset time, so that the model parameters of the universal aggregated perception model are updated through the training of the universal aggregated perception model.

[0101] The update of the model parameters of the universal perception aggregation model can be updated in the cloud.

[0102] After the model parameters are updated, they can also be pushed to individual users (i.e., user terminals) so that individual users can obtain the latest model parameters of the universal perception aggregation model in a timely manner, so that when predicting the real-time quality of the user network of individual users, the latest universal aggregation perception model can be used for prediction.

[0103] Every fourth preset time, the reinforcement learning model is trained based on the fourth training data, and after continuing incremental learning in the cloud, it is pushed to the user end to update the model parameters of the reinforcement learning model.

[0104] The RLS model is trained based on the fifth training data at every fifth preset time, so that the model parameters of the RLS model are updated through the training of the RLS model.

[0105] Since the computational overhead of the RLS model is much smaller than that of the machine learning method, the model parameters of the RLS model can be updated locally (such as on the user side) at a higher update frequency, so that the RLS model can better adapt to complex and changing networks.

[0106] In the embodiment of the present application, the first preset time period, the second preset time, the third preset time, the fourth preset time, and the fifth preset time may all be the same, or all be different, or may be partially the same and partially different.

[0107] The adjustment of the first preset time, the second preset time, the third preset time, the fourth preset time and the fifth preset time can be flexibly adjusted according to specific application scenarios.

[0108] The first training data may include historical network bandwidth demand statistics, which may include the bandwidth required by the corresponding service in the corresponding historical period.

[0109] For example, the statistical result of the network bandwidth demand of service A used by user a during the period from 19:00 to 19:30 on the i-th day is as follows: the bandwidth required by service A during the period from 19:00 to 19:30 on the i-th day is 9 megabits per second (Mbps).

[0110] The second training data may include historical physical layer signal characteristic data and historical network indicator data.

[0111] The historical physical layer signal characteristic data is the physical layer signal characteristic data in a historical period.

[0112] The physical layer signal characteristic data may include signal strength, reference signal receiving power, reference signal receiving quality, signal-to-noise ratio, block error rate, etc.

[0113] The historical network indicator data is network indicator data of a historical period. The network indicator data may be transport layer network indicator data.

[0114] The transport layer network indicator data may include round-trip time (RTT), packet loss rate, delay, delay gradient, throughput, etc.

[0115] The third training data may include output values ​​of the aforementioned training of the individual perception model based on incremental learning.

[0116] The fourth training data and the fifth training data may both include historical network indicator data.

[0117] Optionally, the specific prediction process of predicting the bandwidth demand of the sub-service in the first preset time period by using the bandwidth demand prediction model can be as follows: Figure 4 As shown in steps S401-S405 in. The steps S401-S405 are specifically as follows:

[0118] S401, collecting the user's network service data, and obtaining the statistical results of the network bandwidth demand of the user's corresponding sub-services in different time periods according to the network service data.

[0119] In the embodiment of the present application, the network service data of the user may include the network usage time period, usage duration, startup frequency, data volume change of the sub-service, etc. The sub-service may be a plurality of media services used by the user, such as a service for the user to watch live broadcasts, a service for listening to lossless music in real time, an online on-demand video service, etc., but is not limited thereto.

[0120] In addition, in the embodiment of the present application, the user's network service data can be collected at fixed time intervals, and the collection frequency of the network service data (such as fixed time intervals) can be flexibly adjusted according to the specific application scenario.

[0121] After executing step S401, execute step S402 and execute step S404. Step S402 and step S404 can be executed simultaneously, or step S404 can be executed after executing step S402, or step S402 can be executed after executing step S404.

[0122] S402: Select a bandwidth demand prediction model according to the statistical results of network bandwidth demand of different services in different time periods.

[0123] The bandwidth demand prediction models corresponding to the same service in different time periods may be the same or different; the bandwidth demand prediction models corresponding to different services in the same time period may be the same or different.

[0124] In the embodiment of the present application, the specific selection process of the above bandwidth demand prediction model can be as follows:

[0125] Specifically, if the bandwidth demand in the network bandwidth demand statistics of a service in a time period is greater than the first bandwidth demand threshold, that is, when the bandwidth demand is large, a complex model is selected as the bandwidth demand prediction model of the service in the time period, such as selecting a long short-term memory neural network (English: Long Short-Term Memory, abbreviated as LSTM) as the bandwidth demand prediction model.

[0126] If the bandwidth demand of a service in the network bandwidth demand statistics result in a time period is less than or equal to the first bandwidth demand threshold, that is, when the bandwidth demand is small, a simple model is selected as the bandwidth demand prediction model of the service in the time period, such as selecting the RLS model as the bandwidth demand prediction model.

[0127] Then, according to the time periods included in the first preset time period to be predicted and the statistical results of the network bandwidth requirements of the sub-services in different time periods, a bandwidth requirement prediction model for the sub-services is selected.

[0128] Exemplarily, the bandwidth requirement threshold is 50 Mpbs. The sub-services include A service and B service.

[0129] The statistical results of network bandwidth requirements for service A in different time periods are as follows: the bandwidth required for service A from 18:00 to 18:30 is 10 Mbps, the bandwidth required for service A from 18:30 to 19:00 is 80 Mbps, and the bandwidth required for service A from 19:00 to 19:30 is 60 Mbps.

[0130] The statistical results of network bandwidth requirements for service B in different time periods are as follows: the bandwidth required for service B from 18:00 to 18:30 is 100 Mbps, the bandwidth required for service B from 18:30 to 19:00 is 60 Mbps, and the bandwidth required for service B from 19:00 to 19:30 is 5 Mbps.

[0131] The first preset time period to be predicted is from 18:00 to 19:00, so service A selects the RLS model as the bandwidth demand prediction model in the time period from 18:00 to 18:30 in the first preset time period, and selects the LSTM model as the bandwidth demand prediction model in the time period from 18:30 to 19:00 in the first preset time period.

[0132] Service B selects the LSTM model as the bandwidth demand prediction model during the period from 18:00 to 18:30 in the first preset time period, and also selects the LSTM model as the bandwidth demand prediction model during the period from 18:30 to 19:00 in the first preset time period.

[0133] Through the above method, according to the comparison result of the bandwidth demand statistics of each service in each time period with the bandwidth demand threshold, it is determined whether the service selects a simple model or a complex model as the bandwidth demand prediction model in the time period, thereby realizing flexible selection of the bandwidth demand prediction model, so that the selected bandwidth demand prediction model is more in line with the specific demand situation.

[0134] At the same time, by selecting a simple model as the demand bandwidth prediction model when the bandwidth demand is small, the calculation overhead is reduced while ensuring the appropriate bandwidth demand prediction accuracy, thereby saving computing resources and accelerating the prediction efficiency of bandwidth demand.

[0135] Furthermore, by selecting a complex model as the bandwidth demand prediction model when the bandwidth demand is large, the bandwidth demand prediction model is prevented from failing to accurately predict the bandwidth demand, thereby making the bandwidth demand prediction result more accurate.

[0136] Optionally, the selection of the bandwidth demand prediction model may also be made according to a received user instruction, but is not limited thereto, and the selection method of the bandwidth demand prediction model may be adjusted according to an application scenario.

[0137] After executing step S402, execute step S403.

[0138] S403: predicting the bandwidth demand of the sub-service in the first preset time period by using the bandwidth demand prediction model to obtain a first bandwidth demand prediction result.

[0139] Specifically, the bandwidth demand of the sub-service in the first preset period is predicted by using the bandwidth demand prediction model of the sub-service in the first preset period, and a first bandwidth demand prediction result of the sub-service in the first preset period is obtained.

[0140] Exemplarily, the sub-services include service A and service B, the first preset time period to be predicted is from 18:00 to 18:40, the bandwidth demand prediction models for service A and service B in the time period from 18:00 to 18:30 are RLS model and LSTM model respectively, and the bandwidth demand prediction models in the time period from 18:30 to 18:40 are both LSTM models.

[0141] Therefore, service A uses the RLS model to predict bandwidth demand during the period from 18:00 to 18:30 in the first preset time period, and uses the LSTM model to predict bandwidth demand during the period from 18:30 to 18:40 in the first preset time period.

[0142] The bandwidth demand of service B is predicted by the LSTM model in the first preset time period from 18:00 to 18:30 and from 18:30 to 18:40.

[0143] The specific process of predicting the bandwidth demand of the sub-service in the first preset time period by using the bandwidth demand prediction model of the sub-service in the first preset time period can be as follows:

[0144] For each service, the network bandwidth demand statistics of the service in different time periods are respectively input into the corresponding bandwidth demand prediction model, and the network bandwidth demand statistics are processed by the bandwidth demand prediction model to obtain the network bandwidth demand statistics of the corresponding time period.

[0145] The output of the bandwidth demand prediction model is the first bandwidth demand prediction result of the service in the corresponding time period.

[0146] Exemplarily, the first preset time period to be predicted is 18:00 to 18:40. The statistical results of the network bandwidth demand of service A during the time period from 18:00 to 18:30 on the i-th day to the j-th day are input into the bandwidth demand prediction model of service A during the time period from 18:00 to 18:30, and the statistical results of the network bandwidth demand of service A during the time period from 18:30 to 18:40 on the i-th day to the j-th day are input into the bandwidth demand prediction model of service A during the time period from 18:30 to 18:40, so as to obtain the first bandwidth demand prediction result of service A during the time period from 18:00 to 18:40 by processing the statistical results of network bandwidth demand by the bandwidth demand prediction model.

[0147] After executing step S403, execute step S405.

[0148] S404, obtaining a second bandwidth demand prediction result of the sub-service in the first preset time period based on the integration of the network bandwidth demand statistics results of the sub-service in different time periods.

[0149] In the embodiment of the present application, a weighted average method may be used to merge the statistical results of network bandwidth requirements of sub-services in different time periods, but the method is not limited thereto.

[0150] In the embodiment of the present application, when the weighted average method is used to merge the statistical results of network bandwidth demand, the weights in the weighted average can be dynamically adjusted, or the weights and methods of weighting can be pre-set and adjusted according to user needs.

[0151] Exemplarily, the sub-services include service A and service B, and the first preset time period is 18:00 to 18:40. The statistical results of network bandwidth requirements of service A from 18:00 to 18:30 on the i-th day to the j-th day are weighted averaged to obtain the first bandwidth requirement forecast result of service A from 18:00 to 18:30, and the statistical results of network bandwidth requirements of service A from 18:30 to 18:40 on the i-th day to the j-th day are weighted averaged to obtain the first bandwidth requirement forecast result of service A from 18:00 to 18:40.

[0152] After executing step S404, execute step S405.

[0153] S405: The first bandwidth demand prediction result and the second bandwidth demand prediction result are integrated to obtain a bandwidth demand prediction result of the sub-service in the first preset time period.

[0154] In the embodiment of the present application, the fusion of the first bandwidth demand prediction result of the sub-service in the first preset time period and the second bandwidth demand prediction result of the sub-service in the first preset time period can be fused in a weighted average manner, but is not limited to this.

[0155] Through the above steps S401-S405, the bandwidth demand forecast result of the sub-service in the first preset time period is accurately predicted, thereby describing the user's long-term network behavior, which is helpful for predicting the accuracy of the bandwidth value that the user network can provide based on the user's network behavior, and is helpful for providing more personalized network bandwidth prediction and allocation strategies in the future.

[0156] Optionally, the specific prediction process of predicting the real-time quality of the user network in the second preset time period by the individual perception model and the general aggregate perception model based on incremental learning can be as follows: Figure 5 As shown in steps S501-S504 in the embodiment. The steps S501-S504 are specifically as follows:

[0157] S501, collecting physical layer signal characteristic data and network indicator data of user networks.

[0158] After executing step S501, execute step S502.

[0159] S502, processing the physical layer signal feature data and the network indicator data through an individual perception model based on incremental learning to obtain a first network real-time quality prediction result of the user network in a second preset time period.

[0160] Specifically, the physical layer signal feature data and the network indicator data are spliced ​​into an input vector of an individual perception model based on incremental learning, and then the input vector is input into the individual perception model based on incremental learning. Thus, by processing the input vector (i.e., the physical layer signal feature data and the network indicator data) through the individual perception model based on incremental learning, the real-time quality prediction result of the first network of the user network in the second preset time period is obtained, which in turn helps to improve the prediction accuracy of the real-time quality prediction result of the first network.

[0161] The output of the individual perception model based on incremental learning is the first network real-time quality prediction result of the user network in the second preset time period.

[0162] After executing step S502, execute step S503.

[0163] S503: Process the first network real-time quality prediction result by using a general aggregate perception model to obtain a second network real-time quality prediction result of the user network in a second preset time period.

[0164] Specifically, the first network real-time quality prediction result obtained in step S503 is input into the universal aggregate perception model, so that the first network real-time quality prediction result is processed by the universal aggregate perception model to obtain the second network real-time quality prediction result of the user network in the second preset time period, which helps to improve the prediction accuracy of the first network real-time quality prediction result.

[0165] The output of the general aggregate perception model is a second network real-time quality prediction result of the user network in the second preset time period.

[0166] After executing step S503, execute step S504.

[0167] S504: Fusing the first network real-time quality prediction result and the second network real-time quality prediction result to obtain a real-time network quality prediction result of the user network in a second preset time period.

[0168] In the embodiment of the present application, the first network real-time quality prediction result and the second network real-time quality prediction result can be fused through the attention module, but it is not limited to this.

[0169] Specifically, the first network real-time quality prediction result and the second network real-time quality prediction result are input into the attention module to obtain the network real-time quality prediction result of the user network in the second preset time period, so that the final network real-time quality prediction result takes into account both the user network information of the individual user and the general information of the user network, thereby improving the accuracy of the predicted network quality prediction result.

[0170] Through the above steps S501-S504, accurate prediction of the real-time network quality prediction result of the user network in the second preset time period can be achieved.

[0171] Optionally, the specific prediction process of predicting the real-time bandwidth of the user network in the second preset time period by using the reinforcement learning model and the RLS model can be as follows: Figure 6 As shown in steps S601-S603 in the embodiment. The steps S601-S603 are specifically as follows:

[0172] S601, collecting network indicator data of the user network.

[0173] In the embodiment of the present application, the network indicator data of the user network collected during the prediction process of the real-time bandwidth is consistent with the network indicator data of the user network collected in step S501, and will not be repeated here.

[0174] After executing step S601, execute step S602.

[0175] S602, based on the network indicator data, predict the real-time bandwidth of the user network in the second preset time period using a reinforcement learning model and an RLS model respectively, to obtain a first real-time bandwidth prediction result corresponding to the reinforcement learning model and a second real-time bandwidth prediction result corresponding to the RLS model.

[0176] In an embodiment of the present application, based on the network indicator data, the real-time bandwidth of the user network in the second preset time period is predicted by the reinforcement learning model, and the specific prediction process of obtaining the first real-time bandwidth prediction result corresponding to the reinforcement learning model can be as follows:

[0177] The network indicator data is input into the reinforcement learning model, and the reinforcement learning model processes the network indicator data to realize the prediction of the real-time bandwidth of the user network in the second preset time period by the reinforcement learning model, thereby obtaining the first real-time bandwidth prediction result of the user network in the second preset time period, which helps to improve the prediction accuracy of the first real-time bandwidth prediction result. The output of the reinforcement learning model is the first real-time bandwidth prediction result of the user network in the second preset time period.

[0178] In a possible implementation, based on the network indicator data, the real-time bandwidth of the user network in the second preset time period is predicted by the RLS model, and a specific prediction process of obtaining a second real-time bandwidth prediction result corresponding to the RLS model can be as follows:

[0179] First, based on the network indicator data, an initial real-time bandwidth prediction result is obtained. The initial real-time bandwidth prediction result is a rough estimate of the network bandwidth. Then, the initial real-time bandwidth prediction result is input into the RLS model, so that the RLS model can process the initial real-time bandwidth prediction result to realize the prediction of the user network in the second preset time period by the RLS model, thereby obtaining the second real-time bandwidth prediction result of the user network in the second preset time period, which helps to improve the prediction accuracy of the second real-time bandwidth prediction result. The output of the RLS model is the second real-time bandwidth prediction result of the user network in the second preset time period.

[0180] Optionally, obtaining the initial real-time bandwidth prediction result based on the network indicator data may be inputting the network indicator data into a first model, and obtaining the initial real-time bandwidth prediction result by processing the network indicator data through the first model.

[0181] The first model may be the reinforcement learning model or the LSTM model, but is not limited thereto. The first model may be selected according to a specific application scenario to make the second real-time bandwidth prediction result more accurate.

[0182] However, it should be noted that when the first model is the above-mentioned reinforcement learning model, the above-mentioned initial real-time bandwidth prediction result is the first real-time bandwidth prediction result.

[0183] In another possible implementation, based on the network indicator data, the real-time bandwidth of the user network in the second preset time period is predicted by the RLS model, and a specific prediction process of obtaining a second real-time bandwidth prediction result corresponding to the RLS model can be as follows:

[0184] The network indicator data is directly input into the RLS model, and the RLS model processes the network indicator data to realize the RLS model's prediction of the user network's real-time bandwidth in the second preset period, thereby helping to improve the accuracy of the second real-time bandwidth prediction result. The output of the RLS model is the second real-time bandwidth prediction result.

[0185] After executing step S602, execute step S603.

[0186] S603: The first real-time bandwidth prediction result and the second real-time bandwidth prediction result are integrated to obtain a real-time bandwidth prediction result of the user network in a second preset time period.

[0187] In an embodiment of the present application, a weighted average method can be used to merge the first real-time bandwidth prediction result and the second real-time bandwidth prediction result. The weight and method in the weighted average can be dynamically adjusted according to the specific application scenario to further improve the accuracy of the real-time bandwidth prediction result.

[0188] The first real-time bandwidth prediction result and the second real-time bandwidth prediction result may also be fused by a model fusion method. The fusion method of the first real-time bandwidth prediction result and the second real-time bandwidth prediction result may be adjusted according to a specific application scenario, so that the obtained real-time bandwidth prediction result of the user network in the second preset time period is more accurate.

[0189] Through the above steps S601-S603, accurate prediction of the real-time bandwidth prediction result of the user network in the second preset time period can be achieved.

[0190] Furthermore, in order to enable the bandwidth prediction result of the user network to take into account the impact of the user's personalized network behavior and physical layer signal fluctuations on the bandwidth, so as to further improve the accuracy of the bandwidth prediction result, after predicting the bandwidth demand prediction result of the sub-service in the first preset time period, the network real-time quality prediction result of the user network in the second preset time period, and the real-time bandwidth prediction result of the user network in the second preset time period in step S301, the bandwidth demand prediction result, the network real-time quality prediction result, and the real-time bandwidth prediction result are combined to obtain the bandwidth value that the user network can provide in the second preset time period.

[0191] Specifically, first, the maximum bandwidth requirement result of the second preset period is obtained by combining the bandwidth requirement prediction result of the sub-service in the first preset period with the network real-time quality prediction result of the user network in the second preset period. The maximum bandwidth requirement result can also be called a reference maximum bandwidth requirement result.

[0192] Combined with the maximum bandwidth demand result of the second preset time period and the real-time bandwidth prediction result of the second preset time period, the bandwidth value that the user network can provide in the second preset time period is predicted. Therefore, under limited network resources, the user service demand can be prioritized to accurately predict the bandwidth value that the user can provide in the second preset time period.

[0193] Optionally, the combination of the bandwidth demand prediction result of the sub-service in the first preset time period and the real-time network quality prediction result of the user network in the second preset time period can be combined in a weighted average manner.

[0194] For example, firstly, by combining the bandwidth demand prediction results of each service in the first preset period and the network real-time quality prediction results of the second preset period in a weighted average manner, the best network bandwidth that the user network can provide for each service in the second preset period can be obtained. Thus, based on each best network bandwidth, the maximum bandwidth demand result of the second preset period can be obtained.

[0195] The weights in the above weighted average can be flexibly adjusted according to specific application scenarios.

[0196] Optionally, the combination of the bandwidth demand prediction result of the sub-service in the first preset period and the network real-time quality prediction result of the user network in the second preset period can also be obtained through a second model, but is not limited to this. The combination method of the bandwidth demand prediction result and the network real-time quality prediction result can be selected according to the specific application scenario.

[0197] Among them, the second model can be selected according to the specific application scenario to make the maximum bandwidth demand prediction result more accurate.

[0198] In an embodiment of the present application, the combination of the maximum bandwidth demand result of the above-mentioned second preset time period and the real-time bandwidth prediction result of the second preset time period can be combined through decision-level fusion to avoid the real-time bandwidth prediction result being too small or exceeding the maximum bandwidth demand result. Therefore, by combining the real-time bandwidth prediction result with the maximum bandwidth demand result, the real-time bandwidth demand prediction result can be adjusted, so that the accuracy of the final bandwidth value is further improved.

[0199] The combination of the maximum bandwidth demand result of the second preset time period and the real-time bandwidth prediction result of the second preset time period can also be achieved through the results of the third model, but is not limited to this. The combination method of the maximum bandwidth demand prediction result and the real-time bandwidth prediction result can be selected according to the specific application scenario.

[0200] Furthermore, after obtaining the bandwidth value that the user network can provide in the second preset period, the target bandwidth allocation strategy is selected by combining the bandwidth value with the bandwidth demand prediction result of the sub-service in the first preset period. Then, bandwidth allocation for the sub-service is performed according to the target bandwidth allocation strategy.

[0201] Optionally, when the bandwidth demand prediction result is greater than the second bandwidth demand threshold (i.e., the bandwidth demand is large), and the bandwidth that the user network can ultimately provide (i.e., the bandwidth value that the user network can provide in the second preset time period) cannot meet the bandwidth demand prediction result, a rule-based conservative bandwidth allocation strategy can be selected as the target bandwidth allocation strategy, thereby allocating the bandwidth of the sub-service in the time period based on the rule-based conservative bandwidth allocation strategy.

[0202] The above rule-based conservative bandwidth allocation strategy can determine the demand weight of the sub-service according to the bandwidth demand prediction result of the sub-service, and allocate bandwidth to the sub-service without affecting the minimum demand index of the sub-service.

[0203] Exemplarily, the sub-services include service A and service B. The bandwidth demand prediction result of service A at time t1 is 200 Mbps, the bandwidth demand prediction result of service B at time t1 is 100 Mbps, and the bandwidth value that the user network can provide at time t1 is 30 Mbps.

[0204] It can be seen that the bandwidth available on the user network at time t1 cannot meet the needs of service A and service B. At this time, the bandwidth of service A and service B at time t1 is allocated based on the conservative bandwidth allocation strategy based on rules.

[0205] Specifically, first determine the demand weights of service A and service B. The demand weight of service A is the ratio of the bandwidth demand forecast result of service A at time t1 (i.e., 200Mbps) to the sum of the bandwidth demand forecast results of the sub-services at time t1 (i.e., the sum of 200Mbps and 100Mbps), so the demand weight of service A is 2 / 3. Similarly, the demand weight of service B is 1 / 3. Then, the product of the demand weight of service A and the bandwidth value that can be provided by the user network at time t1 is used as the bandwidth allocated to service A, that is, the bandwidth allocated to service A is 20Mbps. Similarly, the bandwidth allocated to service B is 10Mbps.

[0206] The above demand weights can also be flexibly adjusted according to specific application scenarios.

[0207] When the bandwidth value that the user network can provide in the second preset time period meets the corresponding bandwidth demand prediction result, the rule-based aggressive bandwidth allocation strategy can be selected as the target bandwidth allocation strategy, so as to allocate bandwidth based on the rule-based aggressive bandwidth allocation strategy.

[0208] In summary, the bandwidth prediction method proposed in this application, through the characterization of the user's long-term network behavior and the combination of physical layer signal feature data, takes into account the user's service needs (i.e., the user's personalized network behavior) and the impact of network conditions (such as physical layer signal fluctuations) on bandwidth, and provides a bandwidth prediction result (i.e., the bandwidth value that can be provided) that is closer to the user's bandwidth needs, making the prediction result of the bandwidth value more accurate, which is beneficial to improving the user experience quality in network data transmission.

[0209] In addition, bandwidth demand prediction models (such as the RLS model and the LSTM model) are used to make long-term predictions of users' network bandwidth needs by service and time period, so as to provide more personalized network bandwidth prediction and bandwidth allocation strategies.

[0210] At the same time, combined with the physical layer signal feature data, the real-time network quality of the user network is perceived in real time, forming a maximum bandwidth demand prediction result for reference. And through the general aggregation perception model of the individual perception model based on incremental learning, the network quality perception ability of individual users is enhanced, and the actual network changes can be perceived in time, so that the accuracy of the final predicted bandwidth value can be further improved.

[0211] In addition, by updating the model parameters, the adaptability of the models (such as the bandwidth demand prediction model, the individual perception model based on incremental learning, the general aggregate perception model, the reinforcement learning model, and the RLS model) to the changing network environment is enhanced, which helps to further improve the prediction accuracy of the bandwidth value that the user network can provide.

[0212] The technical solution of this application is further explained below in conjunction with a specific application process.

[0213] like Figure 7 The figure shows a schematic diagram of the processing process of the bandwidth prediction method. First, in the model update module, the model parameters of the bandwidth demand prediction model, the model parameters of the individual perception model based on incremental learning, the model parameters of the general aggregate perception model, the model parameters of the reinforcement learning model, and the model parameters of the RLS model are regularly updated to make these models the optimal models in the corresponding time period. The updated bandwidth demand prediction module is transmitted to the user network behavior perception module; the updated individual perception model based on incremental learning and the general aggregate perception model are transmitted to the real-time network quality perception module; the reinforcement learning model and the RLS model are transmitted to the real-time network bandwidth prediction module.

[0214] In the collection module, the user's network service data, physical layer signal characteristic data and network indicator data of the user's network are collected. The network service data is transmitted to the user network behavior perception module, so that the user network behavior perception module can predict the user's bandwidth demand based on the network service data. The physical layer signal characteristic data and the network indicator data are transmitted to the real-time network quality perception module, so that the real-time network quality perception module can predict the real-time quality of the user network based on the physical layer signal characteristic data and the network indicator data. The network indicator data is transmitted to the real-time network bandwidth prediction module, so that the real-time network bandwidth prediction module can predict the real-time bandwidth of the user network based on the network indicator data.

[0215] In the user network behavior perception module, based on the network service data received from the collection module, the statistical results of the network bandwidth demand of the sub-service corresponding to the user in different time periods are obtained. Then, based on the statistical results of the network bandwidth demand, a bandwidth demand prediction model is selected, so that the bandwidth demand of the sub-service in the first preset time period is predicted through the bandwidth demand prediction model to obtain the first bandwidth demand prediction result. And based on the fusion of the statistical results of the network bandwidth demand of the sub-service in different time periods, the second bandwidth demand prediction result of the sub-service in the first preset time period is obtained. Then, the first bandwidth demand prediction result and the second bandwidth demand prediction result are fused to obtain the bandwidth demand prediction result of the sub-service in the first preset time period. Then, the bandwidth demand prediction result is transmitted to the final prediction module, so that the final prediction module can obtain the bandwidth value that the user network can provide in the second preset time period based on the bandwidth demand prediction result of the first preset time period. And the bandwidth demand prediction result of the first preset time period is transmitted to the bandwidth allocation module, so that the bandwidth allocation module can allocate bandwidth based on the bandwidth demand prediction result.

[0216] In the network quality real-time perception module, the physical layer signal feature data and network indicator data received from the collection module are spliced ​​into an input vector of an individual perception model based on incremental learning. Then the input vector is input into the individual perception model based on incremental learning to obtain the first network real-time quality prediction result for the second preset period. Then the first network real-time quality prediction result is input into the general aggregation perception model to obtain the second network real-time quality prediction result for the second preset period. Then, the first network real-time quality prediction result and the second network real-time quality prediction result are input into the attention module to obtain the network real-time quality prediction result corresponding to the user network in the second preset period. And the network real-time quality prediction result is transmitted to the final prediction module, so that the final prediction module can obtain the bandwidth value that the user network can provide in the second preset period based on the network real-time quality prediction result.

[0217] In the real-time network bandwidth prediction module, based on the network indicator data received from the collection module, the real-time bandwidth of the user network is predicted by the reinforcement learning model and the RLS model respectively, and the first real-time bandwidth prediction result corresponding to the reinforcement learning model and the second real-time bandwidth prediction result corresponding to the RLS model are obtained. The first real-time bandwidth prediction result and the second real-time bandwidth prediction result are then integrated to obtain the real-time bandwidth prediction result of the user network in the second preset time period. The real-time bandwidth prediction result is transmitted to the final prediction module so that the final prediction module can obtain the bandwidth value that the user network can provide in the second preset time period based on the real-time bandwidth prediction result.

[0218] In the final prediction module, the bandwidth demand prediction result of the sub-service in the first preset period transmitted by the user network behavior perception module and the real-time network quality prediction result of the user network in the second preset period transmitted by the real-time network quality perception module are first combined to obtain the maximum bandwidth demand result of the second preset period. Then, the maximum bandwidth demand result and the real-time bandwidth prediction result of the second preset period transmitted by the real-time network bandwidth prediction module are combined to obtain the bandwidth value that the user can finally provide in the second preset period. The bandwidth value is transmitted to the bandwidth allocation module so that the bandwidth allocation module can allocate bandwidth based on the bandwidth value.

[0219] In the bandwidth allocation module, the bandwidth demand prediction result of the sub-service in the first preset period transmitted by the user network behavior perception module and the bandwidth value that the user network can provide in the second preset period transmitted by the final prediction module are received. Combined with the bandwidth demand prediction result of the sub-service in the first preset period and the bandwidth value that the user network can provide in the second preset period, a target bandwidth allocation strategy is selected. Then, bandwidth allocation is performed according to the target bandwidth allocation strategy.

[0220] Based on the same inventive concept, the present application also provides a bandwidth prediction device, such as Figure 8 The figure is a schematic diagram of the structure of a bandwidth prediction device provided by the present application, the device comprising:

[0221] Prediction module 801, used to predict the bandwidth demand of the sub-service corresponding to the user in the first preset time period according to the network behavior of the user, and predict the real-time quality of the user network corresponding to the user in the second preset time period according to the physical layer signal characteristic data, and predict the real-time bandwidth of the user network in the second preset time period;

[0222] The processing module 802 is used to combine the bandwidth demand prediction result corresponding to the bandwidth demand of the sub-service, the network real-time quality prediction result corresponding to the real-time quality of the user network, and the real-time bandwidth prediction result corresponding to the real-time bandwidth of the user network, to predict the bandwidth value that the user network can provide in the second preset time period.

[0223] In a possible implementation, the prediction module 801 is specifically used to collect the network service data of the user, and obtain the statistical results of the network bandwidth demand of the sub-service corresponding to the user in different time periods according to the network service data; wherein the network service data represents the network behavior of the user;

[0224] Selecting a bandwidth demand prediction model according to the statistical results of the network bandwidth demand of the sub-service in different time periods, and predicting the bandwidth demand of the sub-service in the first preset time period by using the bandwidth demand prediction model to obtain a first bandwidth demand prediction result; and

[0225] Obtaining a second bandwidth demand prediction result of the sub-service in the first preset time period according to a fusion of the network bandwidth demand statistics results of the sub-service in different time periods;

[0226] The selected first bandwidth demand prediction result and the second bandwidth demand prediction result are integrated to obtain the bandwidth demand prediction result of the sub-service in the first preset time period.

[0227] In a possible implementation, the prediction module 801 is specifically configured to collect the physical layer signal characteristic data and the network indicator data of the user network;

[0228] Processing the physical layer signal feature data and the network indicator data through an individual perception model based on incremental learning to obtain a first network real-time quality prediction result of the user network in the second preset time period;

[0229] The first network real-time quality prediction result is processed by a universal aggregated perception model to obtain a second network real-time quality prediction result of the user network in the second preset time period; wherein the universal aggregated perception model is a model obtained by processing the individual perception model based on incremental learning in a federated learning manner;

[0230] The first network real-time quality prediction result and the second network real-time quality prediction result are integrated to obtain the network real-time quality prediction result of the user network in the second preset time period.

[0231] In a possible implementation, the prediction module 801 is specifically configured to collect network indicator data of the user network;

[0232] Based on the network indicator data, predicting the real-time bandwidth of the user network in the second preset time period by using a reinforcement learning model and a recursive least squares RLS model respectively, to obtain a first real-time bandwidth prediction result corresponding to the reinforcement learning model and a second real-time bandwidth prediction result corresponding to the RLS model;

[0233] The first real-time bandwidth prediction result and the second real-time bandwidth prediction result are integrated to obtain the real-time bandwidth prediction result of the user network in the second preset time period.

[0234] In a possible implementation manner, the prediction module 801 is specifically configured to update the model parameters of the bandwidth demand prediction model by training the bandwidth demand prediction model at every first preset time; and

[0235] updating the model parameters of the individual perception model based on incremental learning by training the individual perception model based on incremental learning at every second preset time, and updating the model parameters of the general aggregation model by training the general aggregation model at every third preset time; and

[0236] Every fourth preset time, the model parameters of the reinforcement learning model are updated through the training of the reinforcement learning model, and every fifth preset time, the model parameters of the RLS model are updated through the training of the RLS model.

[0237] In a possible implementation, the processing module 802 is specifically configured to obtain a maximum bandwidth requirement result for the second preset time period by combining the bandwidth requirement prediction result corresponding to the bandwidth requirement of the sub-service with the network real-time quality prediction result corresponding to the real-time quality of the user network;

[0238] The maximum bandwidth requirement result is combined with the real-time bandwidth prediction result corresponding to the real-time bandwidth of the user network to predict the bandwidth value that the user network can provide in the second preset time period.

[0239] In a possible implementation, the processing module 802 is specifically configured to select a target bandwidth allocation strategy by combining the bandwidth demand prediction result of the sub-service in the first preset time period with the bandwidth value that can be provided by the user network in the second preset time period;

[0240] Bandwidth allocation is performed according to the target bandwidth allocation strategy.

[0241] Based on the same inventive concept, an electronic device is also provided in the embodiment of the present application. The electronic device can realize the function of the bandwidth prediction device. Fig. 9 , the electronic equipment includes:

[0242] At least one processor 901, and a memory 902 connected to the at least one processor 901. The specific connection medium between the processor 901 and the memory 902 is not limited in the embodiment of the present application. Fig. 9In the example, the processor 901 and the memory 902 are connected via a bus 900. Fig. 9 The bus 900 is represented by a bold line, and the connection between other components is only for schematic illustration and is not intended to be limiting. The bus 900 can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Fig. 9 Only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus. Alternatively, the processor 901 can also be called a controller, and there is no limitation on the name.

[0243] In the embodiment of the present application, the memory 902 stores instructions that can be executed by at least one processor 901. The at least one processor 901 can execute the bandwidth prediction method discussed above by executing the instructions stored in the memory 902. The processor 901 can implement Figure 8 The functions of each module in the device shown.

[0244] Among them, processor 901 is the control center of the device, and can use various interfaces and lines to connect various parts of the entire control device. By running or executing instructions stored in memory 902 and calling data stored in memory 902, various functions of the device and processing data, the device can be monitored as a whole.

[0245] In one possible design, the processor 901 may include one or more processing units, and the processor 901 may integrate an application processor and a modem processor, wherein the application processor mainly processes an operating system, a user interface, and application programs, and the modem processor mainly processes wireless communications. It is understandable that the modem processor may not be integrated into the processor 901. In some embodiments, the processor 901 and the memory 902 may be implemented on the same chip, and in some embodiments, they may also be implemented separately on separate chips.

[0246] Processor 901 can be a general-purpose processor, such as a central processing unit (CPU), a digital signal processor, an application-specific integrated circuit, a field programmable gate array or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component, and can implement or execute the methods, steps and logic block diagrams disclosed in the embodiments of the present application. A general-purpose processor can be a microprocessor or any conventional processor, etc. The steps of the bandwidth prediction method disclosed in the embodiments of the present application can be directly embodied as a hardware processor, or can be executed by a combination of hardware and software modules in the processor.

[0247] The memory 902 is a non-volatile computer-readable storage medium that can be used to store non-volatile software programs, non-volatile computer executable programs and modules. The memory 902 may include at least one type of storage medium, such as flash memory, hard disk, multimedia card, card-type memory, random access memory (English: Random Access Memory, abbreviated as RAM), static random access memory (English: Static Random Access Memory, abbreviated as SRAM), programmable read-only memory (English: Programmable Read Only Memory, abbreviated as PROM), read-only memory (English: Read Only Memory, abbreviated as ROM), electrically erasable programmable read-only memory (English: Electrically Erasable Programmable Read-Only Memory, abbreviated as EEPROM), magnetic memory, disk, optical disk, etc. The memory 902 is any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto. The memory 902 in the embodiment of the present application can also be a circuit or any other device that can realize a storage function, used to store program instructions and / or data.

[0248] By programming the processor 901, the code corresponding to the bandwidth prediction method described in the above embodiment can be fixed into the chip, so that the chip can execute the method when running. Figure 3 The steps of the bandwidth prediction method of the embodiment shown are as follows: How to design and program the processor 901 is a technology well known to those skilled in the art and will not be described in detail here.

[0249] Based on the same inventive concept, an embodiment of the present application further provides a storage medium, which stores computer instructions. When the computer instructions are executed on a computer, the computer executes the bandwidth prediction method discussed above.

[0250] In some possible implementations, various aspects of the bandwidth prediction method provided in the present application may also be implemented in the form of a program product, which includes a program code. When the program product is run on an apparatus, the program code is used to enable the control device to execute the steps of the bandwidth prediction method according to various exemplary implementations of the present application described above in this specification.

[0251] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0252] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0253] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0254] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0255] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A bandwidth prediction method, characterized in that: include: Collect the user's network service data, and obtain the network bandwidth demand statistics of the sub-service corresponding to the user in different time periods based on the network service data; wherein the network service data characterizes the user's network behavior; select a bandwidth demand prediction model based on the network bandwidth demand statistics of the sub-service in different time periods, and predict the bandwidth demand of the sub-service in a first preset time period through the bandwidth demand prediction model to obtain a first bandwidth demand prediction result; and obtain a second bandwidth demand prediction result of the sub-service in the first preset time period based on the fusion of the network bandwidth demand statistics of the sub-service in different time periods; obtain the bandwidth demand prediction result of the sub-service in the first preset time period by fusing the selected first bandwidth demand prediction result with the second bandwidth demand prediction result, and predict the real-time quality of the user network corresponding to the user in the second preset time period and predict the real-time bandwidth of the user network in the second preset time period based on the physical layer signal characteristic data; Combined with the bandwidth demand prediction result corresponding to the bandwidth demand of the sub-service, the network real-time quality prediction result corresponding to the real-time quality of the user network, and the real-time bandwidth prediction result corresponding to the real-time bandwidth of the user network, the bandwidth value that the user network can provide in the second preset time period is predicted.

2. The method according to claim 1, characterized in that The predicting, according to the physical layer signal characteristic data, the real-time quality of the user network corresponding to the user in the second preset time period includes: Collecting the physical layer signal characteristic data and the network indicator data of the user network; Processing the physical layer signal feature data and the network indicator data through an individual perception model based on incremental learning to obtain a first network real-time quality prediction result of the user network in the second preset time period; The first network real-time quality prediction result is processed by a universal aggregated perception model to obtain a second network real-time quality prediction result of the user network in the second preset time period; wherein the universal aggregated perception model is a model obtained by processing the individual perception model based on incremental learning in a federated learning manner; The first network real-time quality prediction result and the second network real-time quality prediction result are integrated to obtain the network real-time quality prediction result of the user network in the second preset time period.

3. The method according to claim 1, characterized in that The predicting the real-time bandwidth of the user network in the second preset time period includes: Collecting network indicator data of the user network; Based on the network indicator data, predicting the real-time bandwidth of the user network in the second preset time period by using a reinforcement learning model and a recursive least squares RLS model respectively, to obtain a first real-time bandwidth prediction result corresponding to the reinforcement learning model and a second real-time bandwidth prediction result corresponding to the RLS model; The first real-time bandwidth prediction result and the second real-time bandwidth prediction result are integrated to obtain the real-time bandwidth prediction result of the user network in the second preset time period.

4. The method according to any one of claims 1 to 3, characterized in that: Before predicting the bandwidth demand of the sub-service corresponding to the user in the first preset time period according to the network behavior of the user, predicting the real-time quality of the user network corresponding to the user in the second preset time period according to the physical layer signal characteristic data, and predicting the real-time bandwidth of the user network in the second preset time period, the method further includes: updating the model parameters of the bandwidth demand prediction model by training the bandwidth demand prediction model at first preset intervals; and updating the model parameters of the individual perception model based on incremental learning by training the individual perception model based on incremental learning at every second preset time, and updating the model parameters of the general aggregation model by training the general aggregation model at every third preset time; and Every fourth preset time, the model parameters of the reinforcement learning model are updated through the training of the reinforcement learning model, and every fifth preset time, the model parameters of the RLS model are updated through the training of the RLS model.

5. The method according to claim 1, characterized in that The predicting of a bandwidth value that can be provided by the user network in the second preset time period by combining the bandwidth demand prediction result corresponding to the bandwidth demand of the sub-service, the network real-time quality prediction result corresponding to the real-time quality of the user network, and the real-time bandwidth prediction result corresponding to the real-time bandwidth of the user network includes: Combining the bandwidth demand prediction result corresponding to the bandwidth demand of the sub-service with the network real-time quality prediction result corresponding to the real-time quality of the user network, obtaining a maximum bandwidth demand result for the second preset time period; The maximum bandwidth requirement result is combined with the real-time bandwidth prediction result corresponding to the real-time bandwidth of the user network to predict the bandwidth value that the user network can provide in the second preset time period.

6. The method according to claim 1 or 5, characterized in that After predicting the bandwidth value that can be provided by the user network in the second preset time period, the method further includes: Selecting a target bandwidth allocation strategy based on the bandwidth demand prediction result of the sub-service in the first preset time period and the bandwidth value that can be provided by the user network in the second preset time period; Bandwidth allocation is performed according to the target bandwidth allocation strategy.

7. A bandwidth prediction device, characterized in that: include: A prediction module, used to collect network service data of users, and obtain network bandwidth demand statistics of the sub-services corresponding to the users in different time periods based on the network service data; wherein the network service data characterizes the network behavior of the users; select a bandwidth demand prediction model based on the network bandwidth demand statistics of the sub-services in different time periods, and predict the bandwidth demand of the sub-services in a first preset time period through the bandwidth demand prediction model to obtain a first bandwidth demand prediction result; and obtain a second bandwidth demand prediction result of the sub-services in the first preset time period based on the fusion of the network bandwidth demand statistics of the sub-services in different time periods; obtain a bandwidth demand prediction result of the sub-services in the first preset time period by fusing the selected first bandwidth demand prediction result with the second bandwidth demand prediction result, and predict the real-time quality of the user network corresponding to the user in the second preset time period and predict the real-time bandwidth of the user network in the second preset time period based on the physical layer signal characteristic data; A processing module is used to combine the bandwidth demand prediction result corresponding to the bandwidth demand of the sub-service, the network real-time quality prediction result corresponding to the real-time quality of the user network, and the real-time bandwidth prediction result corresponding to the real-time bandwidth of the user network, to predict the bandwidth value that the user network can provide in the second preset time period.

8. An electronic device, characterized in that: include: Memory, used to store computer programs; A processor, configured to implement the method steps of any one of claims 1 to 6 when executing the computer program stored in the memory.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method steps described in any one of claims 1 to 6 are implemented.

Images