Streaming media adaptive transmission method and system for edge device
By building and fine-tuning lightweight pre-trained models in edge devices and dynamically adjusting streaming media transmission parameters, the problem of poor environmental adaptability of edge devices is solved, and the optimization of streaming video traffic and the reduction of transmission delay is achieved.
Patent Information
- Application Number
- CN202411019463.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-29
- Publication Date
- 2024-11-15
AI Technical Summary
In the prior art, the built communication model has poor adaptability to the edge device environment, resulting in poor traffic optimization effect and unable to effectively reduce streaming media transmission delay.
By obtaining the environment-aware data of edge devices, a lightweight pre-trained model is built, and a transfer learning method is used to fine-tune model parameters, dynamically adjust the streaming media transmission parameters to adapt to the edge device environment, and improve traffic optimization effect.
Effectively optimize streaming video traffic, reduce streaming media transmission delay, improve the adaptability of lightweight pre-trained models in edge devices, and improve traffic optimization effect.
Smart Images

Figure CN118972649A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of electrical communication technology, and in particular to an edge device streaming media adaptive transmission method and system. Background Art
[0002] In the current IoV architecture, edge computing servers are often used to store or pre-fetch content to assist content delivery. Compared with traditional networks, the design of high-speed cache algorithms in IoV is more challenging, mainly due to the high mobility of vehicles, frequently changing content requirements, and harsh communication environments. To this end, the commonly used technical means is to optimize traffic through a large communication model built with a large amount of data, reduce the impact of vehicles on content caching and delivery, and thus achieve high-quality video transmission.
[0003] In order to save data transmission time, the constructed communication model can be deployed in the edge device. However, the construction of the communication model is based on global considerations, and the special environment of different edge devices is not considered during training. Therefore, the constructed communication model has poor adaptability to the environment of edge devices, resulting in poor traffic optimization effect of the communication model after deployment in the edge device. Summary of the invention
[0004] In view of the shortcomings of the prior art, the present invention proposes an edge device streaming media adaptive transmission method and system, which can improve the flow optimization effect and reduce the streaming media transmission delay. The specific technical solution is as follows:
[0005] In a first aspect, a method for adaptively transmitting streaming media of an edge device is provided. In a first implementable manner of the first aspect, the method includes:
[0006] Obtain the environmental perception data corresponding to different users collected by each edge device, and build data based on all the acquired environmental perception data to train the lightweight pre-training model;
[0007] According to the environmental perception data collected by the edge device, a transfer learning method is used to fine-tune the model parameters of the lightweight pre-trained model to obtain an edge model that matches the edge device;
[0008] Acquire the real-time environmental perception data collected by the edge device, and predict the target parameters of the streaming media through the edge model;
[0009] The transmission parameters of the streaming media are dynamically adjusted in combination with the current parameters and target parameters of the streaming media.
[0010] In combination with the first implementable method of the first aspect, in the second implementable method of the first aspect, the environmental perception data includes: user behavior data, network bandwidth data, delay jitter data, packet loss rate, buffer data, video quality data, device performance data, user environment data, service usage data, user feedback data and / or content feature data.
[0011] In combination with the first implementable manner of the first aspect, in a third implementable manner of the first aspect, obtaining the environmental perception data includes:
[0012] Obtaining user behavior history data, and using a Gaussian mixture model to determine the satisfaction probability functions of different users under different user attributes based on the user behavior history data;
[0013] The corresponding satisfaction probability functions are used to determine the satisfaction probabilities at different time points under different user attributes;
[0014] The user satisfaction is determined by combining the satisfaction probabilities corresponding to different user attributes at different time points.
[0015] In combination with the first implementable method of the first aspect, in a fourth implementable method of the first aspect, a transfer learning method is used to fine-tune the model parameters of the lightweight pre-trained model, including: when the edge device collects new environmental perception data, an incremental learning method is used to update and optimize the parameters of the trained edge model.
[0016] In combination with the first implementable manner of the first aspect, in a fifth implementable manner of the first aspect, the model parameters of the edge model are regularly updated according to the model parameters of the lightweight pre-trained model obtained through training.
[0017] In combination with the first implementable manner of the first aspect, in a sixth implementable manner of the first aspect, the model parameters of the lightweight pre-trained model are regularly updated according to the model parameters of the edge model obtained through training.
[0018] In combination with the first implementable manner of the first aspect, in a seventh implementable manner of the first aspect, the target parameter is a buffer of the streaming media, and the transmission parameter is a bit rate of the streaming media.
[0019] In combination with the seventh implementable manner of the first aspect, in an eighth implementable manner of the first aspect, dynamically adjusting the transmission parameters of the streaming media in combination with the current parameters and the target parameters of the streaming media includes:
[0020] The transmission parameters of the streaming media are calculated according to the current parameters and the target parameters. The specific calculation formula is as follows:
[0021] R new =Rold +γ(B target -B buf );
[0022] Among them, R old is the current bit rate of the streaming media, γ is the set adjustment step size, B target is the target buffer size for streaming media, B buf The current buffer size of the streaming media.
[0023] In a second aspect, a streaming media adaptive transmission system for an edge device is provided. In a first implementable manner of the second aspect, the system includes:
[0024] A pre-training module is configured to obtain environmental perception data corresponding to different users collected by each edge device, and to build data based on all the acquired environmental perception data to train a lightweight pre-training model;
[0025] A transfer learning module is configured to fine-tune the model parameters of the lightweight pre-trained model using a transfer learning method according to the environmental perception data collected by the edge device to obtain an edge model matching the edge device;
[0026] A target prediction module, configured to obtain the real-time environmental perception data collected by the edge device and predict the target parameters of the streaming media through the edge model;
[0027] The transmission adjustment module is configured to dynamically adjust the transmission parameters of the streaming media in combination with the current parameters and target parameters of the streaming media.
[0028] In combination with the first implementable manner of the second aspect, in a second implementable manner of the second aspect, the pre-training module includes:
[0029] An estimation unit is configured to obtain user behavior history data, and use a Gaussian mixture model to estimate user satisfaction corresponding to different time points under different user attributes according to the user behavior history data;
[0030] An analysis unit configured to determine satisfaction levels corresponding to different user attributes at different time points through corresponding satisfaction probability functions;
[0031] A prediction unit is configured to predict the user satisfaction corresponding to each current user attribute according to the corresponding change trend;
[0032] The comprehensive unit is configured to determine the user feedback data of the user by comprehensively analyzing the user satisfaction corresponding to the user under all user attributes.
[0033] Beneficial effect: The edge device streaming media adaptive transmission method and system of the present invention can effectively optimize the streaming video traffic and reduce the streaming media transmission delay by training a lightweight pre-trained model using a data set composed of environmental perception data collected by a large number of different edge devices as samples. The environmental parameters of the edge device where the lightweight pre-trained model is deployed are fully considered, and the model parameters of the lightweight pre-trained model are adaptively adjusted through the transfer learning method, thereby improving the adaptability of the lightweight pre-trained model deployed in the edge device, improving the traffic optimization effect, and further reducing the streaming media transmission delay. BRIEF DESCRIPTION OF THE DRAWINGS
[0034] In order to more clearly illustrate the specific implementation of the present invention, the following will briefly introduce the drawings required for use in the specific implementation. In all the drawings, each element or part is not necessarily drawn according to the actual scale.
[0035] Figure 1 A flowchart of a method for adaptively transmitting streaming media to an edge device provided by an embodiment of the present invention;
[0036] Figure 2 A system block diagram of an edge device streaming media adaptive transmission system provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0037] The following embodiments of the technical solution of the present invention are described in detail in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and are therefore only used as examples, and cannot be used to limit the protection scope of the present invention.
[0038] like Figure 1 The flowchart of the edge device streaming media adaptive transmission method shown in the figure includes:
[0039] Step 1: Obtain the environmental perception data corresponding to different users collected by each edge device, and build data based on all the acquired environmental perception data to train the lightweight pre-training model;
[0040] Step 2: According to the environmental perception data collected by the edge device, a transfer learning method is used to fine-tune the model parameters of the lightweight pre-trained model to obtain an edge model that matches the edge device;
[0041] Step 3: Acquire the real-time environment perception data collected by the edge device, and predict the target parameters of the streaming media through the edge model;
[0042] Step 4: Dynamically adjust the transmission parameters of the streaming media based on the current parameters and target parameters of the streaming media.
[0043] Specifically, first, the cloud platform can use its powerful computing power to train the neural network model through a data set composed of environmental perception data collected by a large number of different edge devices as samples, thereby obtaining a lightweight pre-trained model that can effectively optimize streaming video traffic and reduce streaming transmission delay. Then, the cloud platform can deploy the trained lightweight pre-trained model in the edge device, and the edge device will fully consider its own environmental parameters and adaptively adjust the model parameters of the lightweight pre-trained model through the transfer learning method, thereby improving the adaptability of the lightweight pre-trained model deployed in the edge device, improving the traffic optimization effect, and further reducing the streaming transmission delay. After that, the edge device can obtain the user's real-time environmental perception data and obtain the corresponding target parameters through the adjusted edge prediction model. Finally, combined with the current parameters and target parameters of the streaming media, the transmission parameters of the streaming media can be dynamically adjusted to reduce the streaming media transmission delay.
[0044] In this embodiment, optionally, the environmental perception data includes: user behavior data, network bandwidth data, delay jitter data, packet loss rate, buffer data, video quality data, device performance data, user environment data, service usage data, user feedback data and / or content feature data.
[0045] Specifically, user behavior data includes viewing history data and interaction data. Viewing history data may include the type, duration, and frequency of videos that users have watched in the past. Interaction data may include user operations such as fast forward, rewind, pause, and play during streaming media playback.
[0046] Network bandwidth data includes real-time bandwidth data and bandwidth change trend data. Real-time bandwidth data is the network bandwidth currently available to users, and bandwidth change trend data is the change in the peak and valley values of network bandwidth over time. Delay jitter data includes network delay and delay jitter, and video quality data includes the resolution, bit rate, and frame rate of the played video. Device performance data includes the CPU and GPU performance of the user's device, as well as available memory and storage capacity. User environment data includes regional information of the user's location and the type of network accessed, such as Wi-Fi, 5G, etc. Service usage data includes the number of concurrent servers for video transmission and the storage load of the server. User feedback data includes direct feedback from users on the quality of video playback, such as ratings and comments. Content feature data includes the type, length, and popularity of the video. Different video content features require different transmission strategies.
[0047] In this embodiment, optionally, obtaining the environmental perception data includes:
[0048] Obtaining user behavior history data, and using a Gaussian mixture model to determine the satisfaction probability functions of different users under different user attributes based on the user behavior history data;
[0049] The corresponding satisfaction probability functions are used to determine the satisfaction probabilities at different time points under different user attributes;
[0050] The user satisfaction is determined by combining the satisfaction probabilities corresponding to different user attributes at different time points.
[0051] Specifically, first, the edge device collects historical data related to user satisfaction through user behavior analysis such as page dwell time, click-through rate, conversion rate, etc. Secondly, the cloud and edge devices use Gaussian mixture models to determine the satisfaction probability of different users under different user attributes based on the collected historical data. Finally, the streaming media transmission parameters are trained and predicted based on the user satisfaction corresponding to different time points to achieve adaptive streaming media transmission.
[0052] In this embodiment, optionally, a transfer learning method is used to fine-tune the model parameters of the lightweight pre-trained model, including: when the edge device collects new environmental perception data, an incremental learning method is used to update and optimize the parameters of the trained edge model.
[0053] Specifically, after the lightweight pre-trained model is deployed on the edge device, a small-scale data set can be built based on the environmental perception data collected by the edge device to fine-tune the lightweight pre-trained model, update the model parameters of the lightweight pre-trained model, and obtain the corresponding edge model. The specific algorithm for updating the model parameters is as follows:
[0054]
[0055] Among them, θ new is the model parameter of the marginal model, θ old is the model parameter of the lightweight pre-trained model, f(x; θ old ) is the update cost function, η is the weight factor of the edge device, and Dedge is the dataset constructed by the edge device.
[0056] After the knowledge in the lightweight pre-trained model is transferred to the edge model using the transfer learning method, when the edge device obtains new environmental perception data, the model parameters of the edge model can be updated using the incremental learning method so that the edge model can adapt to the new data stream and achieve continuous optimization of the edge model. The specific update algorithm is as follows:
[0057]
[0058] Among them, θ updatedis the model parameter of the updated edge model
[0059] After the model parameters of the edge model are optimized, the edge device can input the acquired real-time environment perception data into the edge model, and the edge model can output the corresponding target parameters according to the input real-time environment perception data, and dynamically adjust the transmission parameters of the streaming media according to the target parameters. In this embodiment, the target parameter is the buffer of the streaming media, and the transmission parameter is the bit rate of the streaming media. The specific adjustment method of the bit rate is as follows:
[0060] R new =R old +γ(B target -B buf );
[0061] Among them, R old is the current bit rate of the streaming media, γ is the set adjustment step size, B target is the target buffer size for streaming media, B buf The current buffer size of the streaming media.
[0062] In this embodiment, optionally, the model parameters of the edge model are regularly updated according to the model parameters of the lightweight pre-trained model obtained through training.
[0063] Specifically, during use, the cloud platform can also use an incremental learning method to update the model parameters of the lightweight pre-trained model according to the newly acquired environmental perception data, and synchronously update the model parameters of the edge model deployed on the edge device according to the updated model parameters, so that the edge model can be globally optimized. The specific update method of the model parameters is as follows:
[0064] θ edge =θ cloud +β(θ edge -θ cloud );
[0065] Among them, θ edge is the model parameter of the marginal model, θ cloud is the model parameter of the lightweight pre-training model, and β is the synchronization coefficient.
[0066] In this embodiment, optionally, the model parameters of the lightweight pre-trained model are regularly updated according to the model parameters of the edge model obtained through training.
[0067] Specifically, during use, not only can the model parameters of the edge model deployed on the edge device be regularly optimized according to the model parameters of the lightweight pre-trained model trained by the cloud platform, but the model parameters of the lightweight pre-trained model on the cloud platform can also be updated in real time according to the model parameters of the edge model adjusted by the edge device, and the updated model parameters of the lightweight pre-trained model can be synchronously updated to the edge models of other edge devices.
[0068] like Figure 2 The system block diagram of the edge device streaming media adaptive transmission system shown in the figure includes:
[0069] A pre-training module is configured to obtain environmental perception data corresponding to different users collected by each edge device, and to build data based on all the acquired environmental perception data to train a lightweight pre-training model;
[0070] A transfer learning module is configured to fine-tune the model parameters of the lightweight pre-trained model using a transfer learning method according to the environmental perception data collected by the edge device to obtain an edge model matching the edge device;
[0071] A target prediction module, configured to obtain the real-time environmental perception data collected by the edge device and predict the target parameters of the streaming media through the edge model;
[0072] The transmission adjustment module is configured to dynamically adjust the transmission parameters of the streaming media in combination with the current parameters and target parameters of the streaming media.
[0073] Specifically, the transmission system is composed of a pre-training module, a transfer learning module, a target prediction module and a transmission adjustment module. Among them, the pre-training module can use its own powerful computing power to train the neural network model through a data set composed of environmental perception data collected by a large number of different edge devices as samples, so as to obtain a lightweight pre-trained model that can effectively optimize the streaming video traffic and reduce the streaming transmission delay. The transfer learning module can fully consider the environmental parameters of the edge device, and adaptively adjust the model parameters of the lightweight pre-trained model through the transfer learning method, thereby improving the adaptability of the lightweight pre-trained model deployed in the edge device, improving the traffic optimization effect, and further reducing the streaming transmission delay. The target prediction module can obtain the user's real-time environmental perception data and obtain the corresponding target parameters through the adjusted edge prediction model. The transmission adjustment module can dynamically adjust the transmission parameters of the streaming media by combining the current parameters and target parameters of the streaming media, thereby reducing the streaming media transmission delay.
[0074] In this embodiment, optionally, the pre-training module includes:
[0075] An estimation unit is configured to obtain user behavior history data, and use a Gaussian mixture model to estimate user satisfaction corresponding to different time points under different user attributes according to the user behavior history data;
[0076] An analysis unit configured to determine satisfaction levels corresponding to different user attributes at different time points through corresponding satisfaction probability functions;
[0077] A prediction unit is configured to predict the user satisfaction corresponding to each current user attribute according to the corresponding change trend;
[0078] The comprehensive unit is configured to determine the user feedback data of the user by comprehensively analyzing the user satisfaction corresponding to the user under all user attributes.
[0079] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. These modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be included in the scope of the claims and specification of the present invention.
Claims
1. A method for adaptive transmission of streaming media on edge devices, It is characterized in that include: Obtain the environmental perception data corresponding to different users collected by each edge device, and build data based on all the acquired environmental perception data to train the lightweight pre-training model; According to the environmental perception data collected by the edge device, a transfer learning method is used to fine-tune the model parameters of the lightweight pre-trained model to obtain an edge model that matches the edge device; Acquire the real-time environmental perception data collected by the edge device, and predict the target parameters of the streaming media through the edge model; The transmission parameters of the streaming media are dynamically adjusted in combination with the current parameters and target parameters of the streaming media.
2. The method for adaptively transmitting streaming media to an edge device according to claim 1, It is characterized in that The environmental perception data includes: user behavior data, user satisfaction, network bandwidth data, delay jitter data, packet loss rate, buffer data, video quality data, device performance data, user environment data, service usage data, user feedback data and / or content feature data.
3. The method for adaptively transmitting streaming media on an edge device according to claim 1, It is characterized in that Acquiring the environmental perception data includes: Obtaining user behavior history data, and using a Gaussian mixture model to determine the satisfaction probability functions of different users under different user attributes based on the user behavior history data; The corresponding satisfaction probability functions are used to determine the satisfaction probabilities at different time points under different user attributes; The user satisfaction is determined by combining the satisfaction probabilities corresponding to different user attributes at different time points.
4. The method for adaptively transmitting streaming media on an edge device according to claim 1, It is characterized in that The model parameters of the lightweight pre-trained model are fine-tuned using a transfer learning method, including: when the edge device collects new environmental perception data, the parameters of the trained edge model are updated and optimized using an incremental learning method.
5. The method for adaptively transmitting streaming media on an edge device according to claim 1, It is characterized in that The model parameters of the edge model are regularly updated according to the model parameters of the lightweight pre-trained model obtained through training.
6. The method for adaptively transmitting streaming media on an edge device according to claim 1, It is characterized in that The model parameters of the lightweight pre-trained model are regularly updated according to the model parameters of the edge model obtained through training.
7. The method for adaptively transmitting streaming media on an edge device according to claim 1, It is characterized in that The target parameter is the buffer of the streaming media, and the transmission parameter is the bit rate of the streaming media.
8. The method for adaptively transmitting streaming media to an edge device according to claim 7, It is characterized in that The dynamically adjusting the transmission parameters of the streaming media in combination with the current parameters and the target parameters of the streaming media includes: The transmission parameters of the streaming media are calculated according to the current parameters and the target parameters. The specific calculation formula is as follows: R new =R old +γ(B target -B buf ); Among them, R old is the current bit rate of the streaming media, γ is the set adjustment step size, B target is the target buffer size for streaming media, B buf The current buffer size of the streaming media.
9. An edge device streaming media adaptive transmission system, It is characterized in that include: A pre-training module is configured to obtain environmental perception data corresponding to different users collected by each edge device, and to build data based on all the acquired environmental perception data to train a lightweight pre-training model; A transfer learning module is configured to fine-tune the model parameters of the lightweight pre-trained model using a transfer learning method according to the environmental perception data collected by the edge device to obtain an edge model that matches the edge device; A target prediction module, configured to obtain the real-time environmental perception data collected by the edge device and predict the target parameters of the streaming media through the edge model; The transmission adjustment module is configured to dynamically adjust the transmission parameters of the streaming media in combination with the current parameters and target parameters of the streaming media.
10. The edge device streaming media adaptive transmission system according to claim 9, It is characterized in that The pre-training module includes: An estimation unit is configured to obtain user behavior history data, and estimate user satisfaction corresponding to different time points under different user attributes using a Gaussian mixture model according to the user behavior history data; An analysis unit configured to determine satisfaction levels corresponding to different user attributes at different time points through corresponding satisfaction probability functions; A prediction unit is configured to predict the user satisfaction corresponding to each current user attribute according to the corresponding change trend; The comprehensive unit is configured to determine the user feedback data of the user by comprehensively analyzing the user satisfaction corresponding to the user under all user attributes.
