Flow prediction and broadband utilization rate improvement method and system based on VPN (Virtual Private Network) accelerator

Abstract

The invention provides a traffic prediction and broadband utilization rate improvement method and system based on a VPN accelerator, and relates to the technical field of machine learning. The method comprises the following steps: inputting traffic scene parameters and a traffic size sequence into a traffic prediction model, and determining a first suspected predicted traffic size under the condition that sudden traffic increase occurs in a target application program in the future, inputting a second historical actual traffic size and a corresponding time sequence dependency relationship into a pre-trained long-short-term memory network model under the condition that the sudden traffic increase of the target application program occurs in the future corresponding to the plurality of traffic scene parameters in the history, according to the method, the second suspected predicted traffic size of the target application program under the condition that sudden traffic increase occurs in the future is determined, the reliability of the obtained second suspected predicted traffic size is also high, so that the reliability of the finally determined actual predicted traffic size is also high, the phenomena of lagging, packet loss and the like when the service of the target service type is executed are avoided, and the service performance of the target application program is improved. And the bandwidth utilization rate is improved.

Description

Technical Field

[0001] This application relates to the field of machine learning technology, and in particular to methods and systems for traffic prediction and broadband utilization improvement based on VPN accelerators. Background Technology

[0002] Machine learning (IPC classification G06F16 at the patent level) is a technology that enables computers to automatically learn from data and make predictions or decisions through algorithms and models; it is a branch of artificial intelligence. Its core goal is to allow computers to analyze large amounts of data, identify patterns and rules, and thus build models that adapt to new data without explicit programming instructions. Machine learning includes different types such as supervised learning, unsupervised learning, and reinforcement learning, and is widely used in image recognition, natural language processing, and communications.

[0003] In the communications field, when performing network services based on VPN accelerators, pre-trained machine learning models can be used to predict future network traffic based on historical network traffic patterns, allowing for the allocation of appropriate bandwidth. However, sudden surges in network traffic can cause buffering, packet loss, and other issues during network service execution, resulting in a poor user experience and low bandwidth utilization. Summary of the Invention

[0004] This application provides a method and system for traffic prediction and bandwidth utilization improvement based on VPN accelerators, which is used to solve the problem in the prior art that when encountering sudden network traffic, it will cause lag, packet loss and other phenomena when performing network services, resulting in a poor user experience.

[0005] Firstly, this application provides a method for traffic prediction and bandwidth utilization improvement based on a VPN accelerator. The method is applied to a VPN accelerator, which communicates with a terminal device. The terminal device has a target application installed, and the target application executes a target service type during runtime. The method provided in this application includes: The system continuously receives user behavior feature sequences triggered by the user in the target application during runtime, the running status parameter sequence of the target application, and the traffic volume sequence generated by the target application during runtime from the terminal device. Both the user behavior feature sequence and the running status parameters are related to the traffic volume required in the future. Extract the temporal dependencies of user behavior feature sequences; Traffic scenario parameters are constructed based on user behavior feature sequences, the temporal dependencies of user behavior feature sequences, and the runtime state parameter sequences of the target application. By inputting traffic scenario parameters into a pre-trained burst traffic surge prediction model, it can be determined whether a burst traffic surge will occur in the target application in the future. If a sudden surge in traffic is determined to occur in the future, the traffic scenario parameters and traffic size sequence are input into the pre-trained traffic prediction model to determine the first suspected predicted traffic size in the event of a sudden surge in traffic for the target application. Obtain the second historical actual traffic size of the target application under the condition of a sudden surge in traffic in the future, corresponding to multiple traffic scenario parameters in history, and extract the temporal dependency of the second historical actual traffic size of the target application under the condition of a sudden surge in traffic in the future, corresponding to multiple traffic scenario parameters in history. The second historical actual traffic size of the target application under the condition of a sudden surge in traffic in the future, corresponding to multiple traffic scenario parameters in history, and the corresponding temporal dependencies are input into a pre-trained long short-term memory network model to determine the second suspected predicted traffic size under the condition of a sudden surge in traffic in the future. Based on the first and second suspected predicted traffic sizes, determine the actual predicted traffic size of the target application in the event of a sudden surge in traffic in the future. Based on the actual predicted traffic volume, increase the bandwidth allocated to the target service type and decrease the bandwidth allocated to other service types. The increase in bandwidth allocated to the target service type is positively correlated with the actual predicted traffic volume.

[0006] In some implementations, the target application is a live streaming application, and the user behavior feature sequence includes the user interaction activity sequence and the number of online users in the live streaming room. The running status parameters of the target application include the planned time of the hot content in the live streaming schedule, the packet loss rate of the live streaming application during data transmission, and the picture clarity of the live streaming room. Based on the temporal dependencies of extracted user behavior feature sequences, including: using a deep belief DBN network, extracting the temporal dependencies of the user interaction activity sequence and the online number sequence in the live stream; Traffic scenario parameters are input into a pre-trained burst traffic surge prediction model to determine whether a target application will experience a burst traffic surge in the future, including: Based on the pre-trained burst traffic surge prediction model, a burst traffic surge is determined to occur when the following conditions are identified: the average user interaction activity in the live stream is greater than a set activity threshold; the time-series dependency of the user interaction activity sequence indicates that the rate of increase of user interaction activity is greater than a set first rate threshold; the average number of online users in the live stream is greater than a set number threshold; the time-series dependency of the online user sequence indicates that the rate of increase of online users is greater than a set second rate threshold; the planned time of the appearance of hot content in the live stream schedule is less than a set duration threshold; the packet loss rate of the live stream application during data transmission is greater than a set packet loss rate threshold; and the image clarity of the live stream is reduced.

[0007] In some implementations, the target service type is live streaming, and the method further includes: If a sudden surge in traffic is anticipated in the future, preload the live streaming resources associated with the live streaming service at the edge nodes.

[0008] In some implementations, the target application is a cloud gaming application, and the user behavior feature sequence includes the user-input game command sequence and voice data packet traffic sequence. The target application's runtime status parameters include the game screen's image clarity and whether the boss battle trigger conditions are met. Extracting the temporal dependencies of user behavior feature sequences, including: Based on the deep belief DBN network and the sequence of game commands input by the user, the temporal dependency of the number of game commands input by the user per unit time and the temporal dependency of the voice data packet traffic sequence are extracted. Traffic scenario parameters are input into a pre-trained burst traffic surge prediction model to determine whether a target application will experience a burst traffic surge in the future, including: Based on the pre-trained burst traffic surge prediction model, if the average number of game commands input by users per unit time is greater than a set threshold, and the temporal dependency of the number of game commands input by users per unit time indicates that the rate of increase of the number of game commands input by users per unit time is greater than a set rate threshold, if the average traffic generated by voice data packets is greater than a set traffic threshold, and the temporal dependency of the voice data packet traffic sequence indicates that the rate of increase of the voice data packet traffic is greater than a set rate threshold, if the clarity of the live broadcast room is reduced, and if the boss battle trigger conditions are met, then a burst traffic surge is predicted to occur in the future.

[0009] In some implementations, the target service type is cloud gaming, and the method further includes: If a sudden surge in traffic is anticipated in the future, preloadable cloud gaming resources associated with the cloud gaming service will be preloaded at the edge nodes.

[0010] In some implementations, determining the actual predicted traffic volume of the target application in the event of a future sudden surge in traffic, based on a first suspected predicted traffic volume and a second suspected predicted traffic volume, includes: The first and second suspected predicted traffic sizes are weighted and averaged to determine the actual predicted traffic size of the target application in the event of a sudden surge in traffic.

[0011] In some implementations, after increasing the bandwidth allocated to the target service type and decreasing the bandwidth allocated to other service types based on the actual predicted traffic volume, the method further includes: Based on the increased bandwidth allocated to the target service type, execute the service of the target service type; Determine the actual traffic volume generated by executing the target business type after a preset time in the future; If the difference between the actual predicted traffic volume and the actual traffic volume is greater than a set difference threshold, the network parameters of the traffic prediction model and the long short-term memory network model are updated based on the difference between the actual predicted traffic volume and the actual traffic volume.

[0012] Secondly, this application also provides a traffic prediction and bandwidth utilization improvement system based on a VPN accelerator, applied to a VPN accelerator. The VPN accelerator is connected to a terminal device, which has a target application installed. The target application executes a target service type during runtime. The system includes: The data receiving unit is used to continuously receive user behavior feature sequences triggered by the user in the target application, the running status parameter sequence of the target application, and the traffic volume sequence generated by the target application during runtime, all sent by the terminal device. The user behavior feature sequences and running status parameters are both related to the future traffic volume demand. The dependency extraction unit is used to extract the temporal dependencies of user behavior feature sequences based on the deep belief DBN network. The scenario parameter construction unit is used to construct traffic scenario parameters based on user behavior feature sequences, the temporal dependencies of user behavior feature sequences, and the runtime status parameter sequences of the target application. The traffic surge determination unit is used to input traffic scenario parameters into a pre-trained burst traffic surge prediction model to determine whether a burst traffic surge will occur in the target application in the future. The first traffic prediction unit is used to input traffic scenario parameters and traffic size sequence into a pre-trained traffic prediction model if it is determined that a sudden traffic surge will occur in the future, and to determine the first suspected predicted traffic size in the event of a sudden traffic surge in the target application. The dependency extraction unit is used to obtain the second historical actual traffic size of the target application under the condition of a sudden traffic surge in the future, corresponding to multiple traffic scenario parameters in history, and extract the temporal dependency of the second historical actual traffic size of the target application under the condition of a sudden traffic surge in the future, corresponding to multiple historical traffic scenario parameters, based on the deep belief DBN network. The second traffic prediction unit is also used to input the second historical actual traffic size of the target application under the condition of a sudden traffic surge in the future, corresponding to multiple traffic scenario parameters in history, and the corresponding temporal dependencies, into the pre-trained long short-term memory network model, so as to determine the second suspected predicted traffic size under the condition of a sudden traffic surge in the future of the target application. The traffic prediction determination unit is used to determine the actual predicted traffic size of the target application in the event of a sudden surge in traffic, based on the first suspected predicted traffic size and the second suspected predicted traffic size. The bandwidth allocation unit is used to increase the bandwidth allocated to the target service type and decrease the bandwidth allocated to other service types based on the actual predicted traffic volume. The increase in bandwidth allocated to the target service type is positively correlated with the actual predicted traffic volume.

[0013] Thirdly, this application also provides a VPN accelerator, including a processor and a computer program stored in a memory and executable on the processor, wherein when the processor executes the computer program, the VPN accelerator performs the method provided in the first aspect of this application.

[0014] Fourthly, this application also provides a storage medium storing a computer program, which, when executed by a processor, causes the computer to perform the method provided in the first aspect of this application.

[0015] The method and system for traffic prediction and bandwidth utilization improvement based on VPN accelerators provided in this application can construct traffic scenario parameters based on user behavior feature sequences, the temporal dependencies of user behavior feature sequences, and the runtime state parameter sequences of the target application. These traffic scenario parameters are then input into a pre-trained burst traffic surge prediction model to determine whether the target application will experience a burst traffic surge in the future. Since each first training sample in training the burst traffic surge prediction model includes labeled historical traffic scenario parameters, with the labels indicating whether a burst traffic surge will or will not occur after the generation of these historical traffic scenario parameters, the prediction results are highly reliable.

[0016] If a sudden surge in traffic is predicted in the future, the traffic scenario parameters and traffic size sequence are input into a pre-trained traffic prediction model to determine the first suspected predicted traffic size under the condition of a future sudden surge in traffic for the target application. Furthermore, since each second training sample in training the traffic prediction model includes historical traffic scenario parameters, the corresponding historical traffic size sequence generated by the target application during runtime, and the corresponding first historical actual traffic size under the condition of a future sudden surge in traffic for the target application, the reliability of the determined first suspected predicted traffic size is high.

[0017] Furthermore, the second historical actual traffic size corresponding to multiple traffic scenario parameters in the future under the condition of a sudden surge in traffic of the target application, as well as the corresponding temporal dependencies, are input into a pre-trained long short-term memory network model to determine the second suspected predicted traffic size under the condition of a sudden surge in traffic of the target application in the future. The reliability of the obtained second suspected predicted traffic size is also high.

[0018] Furthermore, based on the first and second suspected predicted traffic volumes, the accuracy of determining the actual predicted traffic volume for the target application in the event of a sudden traffic surge is very high. Based on the actual predicted traffic volume, the bandwidth allocated to the target service type is increased, while the bandwidth allocated to other service types is decreased. The increase in bandwidth allocated to the target service type is positively correlated with the actual predicted traffic volume. Understandably, when a sudden traffic surge occurs, users are generally focused on executing the target service type, while other service types require very little bandwidth. Therefore, to improve bandwidth utilization, some of the bandwidth allocated to other service types can be reallocated to the target service type. This ensures that even if a sudden traffic surge occurs while executing the target service type, sufficient bandwidth for that service can be guaranteed. This avoids stuttering and packet loss when executing the target service type, improving user experience and bandwidth utilization. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 A flowchart illustrating the VPN accelerator-based traffic prediction and bandwidth utilization improvement method provided in this application embodiment; Figure 2 This is a functional block diagram of a VPN accelerator-based traffic prediction and broadband utilization improvement system provided in an embodiment of this application. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments made by those skilled in the art under the guidance of these embodiments are within the scope of protection of this application.

[0022] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0023] This application provides a method for traffic prediction and bandwidth utilization improvement based on a VPN accelerator. The method is applied to a VPN accelerator, which communicates with a terminal device. The terminal device has a target application installed, and the target application executes services of the target service type during runtime. For example... Figure 1 As shown, the method provided in this application embodiment includes: S101: Continuously receive user behavior feature sequences triggered by the user in the target application during runtime, the running status parameter sequence of the target application, and the traffic volume sequence generated by the target application during runtime from the terminal device, and the user behavior feature sequence and the running status parameters are all related to the traffic volume required in the future.

[0024] For example, the target application can be, but is not limited to, a live streaming application; the user behavior feature sequence includes the user interaction activity sequence in the live streaming room (such as the sequence of likes per unit time, the sequence of gift rewards per unit time, etc.) and the sequence of online users in the live streaming room; the running status parameters of the target application include the planned time of the hot content in the live streaming schedule (such as the time when celebrities appear to promote products), the packet loss rate of the live streaming application in data transmission, and the picture clarity of the live streaming room.

[0025] For example, if the target application is a cloud gaming application, the user behavior feature sequence includes the user's input game command sequence and voice data packet traffic sequence, and the target application's running status parameters include the clarity of the game screen and whether the boss battle trigger conditions are met.

[0026] S102: Extract the temporal dependencies of user behavior feature sequences based on the deep belief DBN network.

[0027] S103: Construct traffic scenario parameters based on user behavior feature sequences, the temporal dependencies of user behavior feature sequences, and the runtime state parameter sequences of the target application.

[0028] For example, traffic scenario parameters can be constructed by concatenating user behavior feature sequences, the temporal dependencies of user behavior feature sequences, and the runtime status parameter sequences of the target application.

[0029] S104: Input the traffic scenario parameters into the pre-trained burst traffic surge prediction model to determine whether the target application will experience a burst traffic surge in the future. If so, execute S105.

[0030] The sudden traffic surge prediction model is obtained by inputting multiple first training samples into the neural network for training. Each first training sample includes historical traffic scenario parameters labeled with identifiers. The identifiers are used to characterize whether a sudden traffic surge will occur or not after the generation of the historical traffic scenario parameters.

[0031] S105: Input the traffic scenario parameters and traffic size sequence into the pre-trained traffic prediction model to determine the first suspected predicted traffic size in the event of a sudden surge in traffic in the target application in the future. The traffic prediction model is trained based on multiple second training samples. Each second training sample includes historical traffic scenario parameters, the corresponding historical traffic size sequence generated by the target application during runtime, and the corresponding first historical actual traffic size in the event of a sudden surge in traffic in the target application in the future.

[0032] S106: Obtain the second historical actual traffic size of the target application under the condition of a sudden surge in traffic in the future, corresponding to multiple traffic scenario parameters in history, and extract the temporal dependency of the second historical actual traffic size of the target application under the condition of a sudden surge in traffic in the future, corresponding to multiple historical traffic scenario parameters, based on the deep belief DBN network.

[0033] S107: Input the second historical actual traffic size corresponding to multiple traffic scenario parameters of the target application in the event of a sudden traffic surge in the future, as well as the corresponding temporal dependencies, into a pre-trained long short-term memory network model to determine the second suspected predicted traffic size in the event of a sudden traffic surge in the target application in the future.

[0034] Understandably, when under the same traffic scenario parameters, the target application's future traffic characteristics will have certain commonalities. For example, if the traffic scenario parameters at 19:50 tonight (10 minutes before the celebrity's appearance to promote products) are the same as those at 19:50 last night and the night before (10 minutes before the celebrity's appearance to promote products), then the traffic characteristics of the three at the celebrity's official appearance time (e.g., 20:00) will show certain patterns (such as linear increase or linear decrease). Thus, the second suspected predicted traffic size can be obtained through S107 mentioned above.

[0035] The Long Short-Term Memory (LSTM) network model combines the second historical actual traffic size of the target application under the condition of a sudden surge in traffic in the future, corresponding to multiple traffic scenario parameters in the past, the corresponding temporal dependencies, and the second historical actual traffic size of the target application under the condition of a sudden surge in traffic in the future.

[0036] S108: Based on the first suspected predicted traffic size and the second suspected predicted traffic size, determine the actual predicted traffic size in the event of a sudden surge in traffic to the target application in the future.

[0037] For example, a weighted average of the first and second suspected predicted traffic sizes can be taken to determine the actual predicted traffic size in the event of a sudden surge in traffic to the target application in the future. This results in a more reliable actual predicted traffic size.

[0038] S109: Based on the actual predicted traffic volume, increase the bandwidth allocated to the target service type and decrease the bandwidth allocated to other service types. The increase in bandwidth allocated to the target service type is positively correlated with the actual predicted traffic volume.

[0039] In some implementations, after S109, the method provided in this application embodiment further includes: Based on the increased bandwidth allocated to the target service type, the target service type is executed; the actual traffic volume generated by executing the target service type after a preset time period is determined; if the difference between the actual predicted traffic volume and the actual traffic volume is greater than a set difference threshold, the network parameters of the Long Short-Term Memory (LSTM) network model are updated based on the difference. In this way, the traffic prediction model and the LSM network model can be continuously optimized, resulting in higher reliability of the first and second suspected predicted traffic volumes subsequently determined by the traffic prediction model and the LSM network model, respectively.

[0040] In some implementations, when the target application is a live streaming application, the user behavior feature sequence includes the user interaction activity sequence and the number of online users in the live streaming room. The target application's runtime parameters include the planned times when hot content appears in the live streaming schedule, the packet loss rate of the live streaming application during data transmission, and the image clarity of the live streaming room.

[0041] Furthermore, based on the deep belief DBN network, the temporal dependencies of the user interaction activity sequence and the online number sequence in the live stream can be extracted. Then, based on the pre-trained burst traffic surge prediction model, if the following conditions are identified: the average user interaction activity in the live stream is greater than a set activity threshold; the temporal dependency of the user interaction activity sequence indicates that the rate of increase of user interaction activity is greater than a set first rate threshold; the average online number sequence in the live stream is greater than a set number threshold; the temporal dependency of the online number sequence indicates that the rate of increase of online number is greater than a set second rate threshold; the planned time of the hot content in the live stream schedule is less than a set duration threshold; the packet loss rate of the live stream application during data transmission is greater than a set packet loss rate threshold; and the image clarity in the live stream is reduced (e.g., a celebrity is about to appear and promote products), a burst traffic surge is predicted (e.g., a burst traffic surge occurs 5 minutes after a celebrity appears and promotes products).

[0042] In addition, the method provided in this application embodiment also includes: If a sudden surge in traffic is anticipated in the future, preload the live streaming resources associated with the live streaming business (such as images of products being sold in the live stream and UI resources for the flash sale countdown) at the edge nodes.

[0043] In some implementations, the target application is a cloud gaming application, and the user behavior feature sequence includes the user-input game command sequence and voice data packet traffic sequence. The target application's running status parameters include the clarity of the game screen and whether the boss battle trigger conditions are met.

[0044] Furthermore, based on the deep belief DBN network and the sequence of game commands input by the user, the temporal dependency of the number of game commands input by the user per unit time, as well as the temporal dependency of the voice data packet traffic sequence, can be extracted. Thus, based on the pre-trained burst traffic surge prediction model, if the average number of game commands input by the user per unit time exceeds a set threshold, and the temporal dependency of the number of game commands input by the user per unit time indicates that the rate of increase of the number of game commands input by the user per unit time exceeds a set rate threshold; if the average traffic generated by the voice data packets exceeds a set traffic threshold, and the temporal dependency of the voice data packet traffic sequence indicates that the rate of increase of the voice data packet traffic exceeds a set rate threshold; if player game characters are all converging towards the geographical coordinates of the boss on the game map; if the clarity of the live stream image decreases; and if the boss battle trigger conditions are met (such as the boss's aggro list being updated, locking onto the primary target player; or the boss's health dropping to 50%, putting the boss into berserk mode), it indicates that a boss battle is likely to occur soon, thus confirming a future burst traffic surge.

[0045] In some implementations, the method provided in this application further includes: If a sudden surge in traffic is anticipated in the future, preload preloadable cloud gaming resources associated with the cloud gaming service (such as preloading boss models and game scene maps of the boss's location) at the edge nodes.

[0046] In addition, this application embodiment also provides a traffic prediction and bandwidth utilization improvement system based on a VPN accelerator, applied to a VPN accelerator. It should be noted that the basic principle and technical effects of the traffic prediction and bandwidth utilization improvement system based on a VPN accelerator provided in this embodiment are the same as those in the above embodiments. For the sake of brevity, any parts not mentioned in the embodiments of this invention can be referred to the corresponding content in the above embodiments. The VPN accelerator is communicatively connected to a terminal device, which has a target application installed. The target application executes the target service type during runtime. The system includes a data receiving unit, a dependency extraction unit, a scenario parameter construction unit, a traffic surge determination unit, a first traffic prediction unit, a dependency extraction unit, a second traffic prediction unit, a predicted traffic determination unit, and a bandwidth allocation unit, wherein... The data receiving unit is used to continuously receive user behavior feature sequences triggered by the user in the target application, the running status parameter sequence of the target application, and the traffic volume sequence generated by the target application during runtime, all sent by the terminal device. The user behavior feature sequences and running status parameters are both related to the future traffic volume demand. The dependency extraction unit is used to extract the temporal dependencies of user behavior feature sequences based on the deep belief DBN network. The scenario parameter construction unit is used to construct traffic scenario parameters based on user behavior feature sequences, the temporal dependencies of user behavior feature sequences, and the runtime status parameter sequences of the target application. The traffic surge determination unit is used to input traffic scenario parameters into a pre-trained sudden traffic surge prediction model to determine whether a sudden traffic surge will occur in the target application in the future. The sudden traffic surge prediction model is trained by inputting multiple first training samples into a neural network. Each first training sample includes historical traffic scenario parameters labeled with an identifier. The identifier is used to characterize whether a sudden traffic surge will occur or not after the historical traffic scenario parameters are generated. The first traffic prediction unit is used to input traffic scenario parameters and traffic size sequence into a pre-trained traffic prediction model if it is determined that a sudden traffic surge will occur in the future, and to determine the first suspected predicted traffic size of the target application in the event of a sudden traffic surge. The traffic prediction model is trained based on multiple second training samples. Each second training sample includes historical traffic scenario parameters, the corresponding historical traffic size sequence generated by the target application during runtime, and the corresponding first historical actual traffic size of the target application in the event of a sudden traffic surge. The dependency extraction unit is used to obtain the second historical actual traffic size of the target application under the condition of a sudden traffic surge in the future, corresponding to multiple traffic scenario parameters in history, and extract the temporal dependency of the second historical actual traffic size of the target application under the condition of a sudden traffic surge in the future, corresponding to multiple historical traffic scenario parameters, based on the deep belief DBN network. The second traffic prediction unit is also used to input the second historical actual traffic size corresponding to multiple historical traffic scenario parameters for the target application under the condition of a sudden traffic surge in the future, as well as the corresponding temporal dependencies, into a pre-trained long short-term memory network model to determine the second suspected predicted traffic size under the condition of a sudden traffic surge in the future; wherein, the long short-term memory network model is the second historical actual traffic size corresponding to multiple historical traffic scenario parameters for the target application under the condition of a sudden traffic surge in the future, the corresponding temporal dependencies, and the corresponding second historical actual traffic size under the condition of a sudden traffic surge in the future. The traffic prediction determination unit is used to determine the actual predicted traffic size of the target application in the event of a sudden surge in traffic, based on the first suspected predicted traffic size and the second suspected predicted traffic size. The bandwidth allocation unit is used to increase the bandwidth allocated to the target service type and decrease the bandwidth allocated to other service types based on the actual predicted traffic volume. The increase in bandwidth allocated to the target service type is positively correlated with the actual predicted traffic volume.

[0047] In addition, this application also provides a VPN accelerator, including a processor and a computer program stored in a memory and executable on the processor. When the processor executes the computer program, it causes the VPN accelerator to perform the method provided in the above embodiments of this application.

[0048] In addition, this application embodiment also provides a storage medium storing a computer program, which, when executed by a processor, causes the computer to perform the method provided in the above embodiments of this application.

[0049] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A method for traffic prediction and bandwidth utilization improvement based on VPN accelerator, characterized in that, The method, applied to a VPN accelerator, which is communicatively connected to a terminal device, wherein the terminal device has a target application installed, and the target application executes a target service type during runtime, includes: The terminal device continuously receives user behavior feature sequences triggered by the user in the target application during runtime, a sequence of running status parameters of the target application, and a sequence of traffic volume generated by the target application during runtime, wherein the user behavior feature sequences and the running status parameters are all associated with the future traffic volume demand. Extract the temporal dependencies of the user behavior feature sequences, and construct traffic scenario parameters based on the user behavior feature sequences, the temporal dependencies of the user behavior feature sequences, and the runtime state parameter sequence of the target application; The traffic scenario parameters are input into a pre-trained burst traffic surge prediction model to determine whether the target application will experience a burst traffic surge in the future. If it is determined that a sudden surge in traffic will occur in the future, the traffic scenario parameters and the traffic size sequence are input into a pre-trained traffic prediction model to determine the first suspected predicted traffic size in the event of a sudden surge in traffic for the target application in the future. Obtain the second historical actual traffic size of the target application under the condition of a sudden traffic surge in the future, corresponding to multiple traffic scenario parameters in history, and extract the temporal dependency of the second historical actual traffic size of the target application under the condition of a sudden traffic surge in the future, corresponding to multiple traffic scenario parameters in history. The second historical actual traffic size of the target application under the condition of a sudden surge in traffic in the future, corresponding to the traffic scenario parameters of the previous history, and the corresponding temporal dependencies are input into a pre-trained long short-term memory network model to determine the second suspected predicted traffic size under the condition of a sudden surge in traffic in the future of the target application. Based on the first suspected predicted traffic size and the second suspected predicted traffic size, determine the actual predicted traffic size of the target application in the event of a sudden surge in traffic in the future; Based on the actual predicted traffic volume, the bandwidth allocated to the target service type is increased, and the bandwidth allocated to other service types is decreased, wherein the increase in the bandwidth allocated to the target service type is positively correlated with the actual predicted traffic volume.

2. The method of claim 1, wherein, The target application is a live streaming application. The user behavior feature sequence includes the user interaction activity sequence and the number of online users in the live streaming room. The running status parameters of the target application include the planned time of the hot content in the live streaming schedule, the packet loss rate of the live streaming application during data transmission, and the picture clarity of the live streaming room. The step of extracting the temporal dependencies of the user behavior feature sequence includes: extracting the temporal dependencies of the user interaction activity sequence and the online number sequence of the live stream room respectively, based on the deep belief DBN network. The step of inputting the traffic scenario parameters into a pre-trained burst traffic surge prediction model to determine whether the target application will experience a burst traffic surge in the future includes: Based on the pre-trained burst traffic surge prediction model, a burst traffic surge is determined to occur in the future if the following conditions are identified: the average user interaction activity in the live stream is greater than a set activity threshold; the time-series dependency of the user interaction activity sequence indicates that the rate of increase of user interaction activity is greater than a set first rate threshold; the average number of online users in the live stream is greater than a set number threshold; the time-series dependency of the online user sequence indicates that the rate of increase of the online user is greater than a set second rate threshold; the planned time of the occurrence of hot content in the live stream schedule is less than a set duration threshold; the packet loss rate of the live stream application during data transmission is greater than a set packet loss rate threshold; and the image clarity of the live stream is reduced.

3. The method of claim 2, wherein, The target business type is a live streaming business, and the method further includes: If a sudden surge in traffic is anticipated in the future, preloadable live streaming resources associated with the live streaming service will be preloaded at the edge nodes.

4. The method according to claim 1, characterized in that, The target application is a cloud gaming application. The user behavior feature sequence includes the user-input game command sequence and voice data packet traffic sequence. The target application's running status parameters include the game screen clarity and whether the boss battle trigger conditions are met. The step of extracting the temporal dependencies of the user behavior feature sequence includes: Based on the deep belief DBN network and the sequence of game commands input by the user, the temporal dependency of the number of game commands input by the user per unit time and the temporal dependency of the voice data packet traffic sequence are extracted. The step of inputting the traffic scenario parameters into a pre-trained burst traffic surge prediction model to determine whether the target application will experience a burst traffic surge in the future includes: Based on the pre-trained burst traffic surge prediction model, if the average number of game commands input by users per unit time is greater than a set threshold, and the temporal dependency of the number of game commands input by users per unit time indicates that the rate of increase of the number of game commands input by users per unit time is greater than a set rate threshold; if the average traffic generated by voice data packets is greater than a set traffic threshold, and the temporal dependency of the voice data packet traffic sequence indicates that the rate of increase of the voice data packet traffic is greater than a set rate threshold; if the clarity of the live broadcast room decreases; and if the boss battle trigger conditions are met, then a burst traffic surge is determined to occur in the future.

5. The method according to claim 4, characterized in that, The target business type is cloud gaming, and the method further includes: If a sudden surge in traffic is anticipated in the future, preloadable cloud gaming resources associated with the cloud gaming service will be preloaded at the edge nodes.

6. The method according to claim 1, characterized in that, The step of determining the actual predicted traffic volume of the target application in the event of a sudden surge in traffic, based on the first suspected predicted traffic volume and the second suspected predicted traffic volume, includes: The first suspected predicted traffic size and the second suspected predicted traffic size are weighted and averaged to determine the actual predicted traffic size of the target application in the event of a sudden surge in traffic in the future.

7. The method according to any one of claims 1-6, characterized in that, After increasing the bandwidth allocated to the target service type and decreasing the bandwidth allocated to other service types based on the actual predicted traffic volume, the method further includes: Based on the increased bandwidth allocated to the target service type, execute the service of the target service type; Determine the actual traffic volume generated when the target service type is executed after a preset time period in the future; If the difference between the actual predicted traffic volume and the actual traffic volume is greater than a set difference threshold, the network parameters of the traffic prediction model and the long short-term memory network model are updated according to the difference between the actual predicted traffic volume and the actual traffic volume.

8. A traffic prediction and broadband utilization improvement system based on VPN accelerator, characterized in that, A system for use with a VPN accelerator, wherein the VPN accelerator is communicatively connected to a terminal device, the terminal device has a target application installed, and the target application executes a target service type during runtime; the system includes: The data receiving unit is used to continuously receive user behavior feature sequences triggered by the user in the target application during runtime, the running status parameter sequence of the target application, and the traffic volume sequence generated by the target application during runtime, all sent by the terminal device. The user behavior feature sequences and the running status parameters are both associated with the traffic volume required in the future. A dependency extraction unit is used to extract the temporal dependencies of the user behavior feature sequence; The scenario parameter construction unit is used to construct traffic scenario parameters based on the user behavior feature sequence, the temporal dependency of the user behavior feature sequence, and the running state parameter sequence of the target application. A traffic surge determination unit is used to input the traffic scenario parameters into a pre-trained sudden traffic surge prediction model to determine whether the target application will experience a sudden traffic surge in the future. The first traffic prediction unit is used to input the traffic scenario parameters and the traffic size sequence into a pre-trained traffic prediction model if it is determined that a sudden traffic surge will occur in the future, and to determine the first suspected predicted traffic size in the event of a sudden traffic surge in the target application. The dependency extraction unit is used to obtain the second historical actual traffic size of the target application under the condition of a sudden traffic surge in the future, corresponding to multiple traffic scenario parameters in history, and to extract the temporal dependency of the second historical actual traffic size of the target application under the condition of a sudden traffic surge in the future, corresponding to multiple traffic scenario parameters in history. The second traffic prediction unit is also used to input the second historical actual traffic size of the target application under the condition of a sudden traffic surge in the future, corresponding to the multiple traffic scenario parameters in history, and the corresponding temporal dependency relationship, into a pre-trained long short-term memory network model, so as to determine the second suspected predicted traffic size under the condition of a sudden traffic surge in the future of the target application. The traffic prediction determination unit is used to determine the actual predicted traffic size of the target application in the event of a sudden surge in traffic, based on the first suspected predicted traffic size and the second suspected predicted traffic size. A bandwidth allocation unit is used to increase the bandwidth allocated to the target service type and decrease the bandwidth allocated to other service types based on the actual predicted traffic volume, wherein the increase in bandwidth allocated to the target service type is positively correlated with the actual predicted traffic volume.

9. A VPN accelerator, comprising a processor and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it causes the VPN accelerator to perform the method as described in any one of claims 1 to 7.

10. A storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it causes the computer to perform the method as described in any one of claims 1 to 7.

Images