A method for allocating operator bandwidth based on network video traffic QoE identification

Abstract

The present invention discloses an operator bandwidth allocation method based on network video traffic QoE identification, which relates to the field of computer network technology. Based on users who simultaneously play set videos, the set videos are input into a trained bandwidth allocation model to output the user's bandwidth allocation strategy. The training process of the bandwidth allocation model is as follows: N set videos are pre-specified as optimization targets; the set videos and QoEs are obtained based on a webpage fine-grained recognition module and a user QoE recognition module. For a set video and a set of set bandwidths, the corresponding QoE value is unique, and a recurrent neural network is used to fit the QoE and bandwidth functions; with maximizing the sum of the user QoE values ​​of the N set videos as the final goal, the optimal bandwidth allocation strategy is obtained based on Bayesian optimization in combination with the set constraints and boundary conditions as well as the QoE and bandwidth functions. The operator bandwidth allocation method improves the accuracy and effectiveness of bandwidth allocation.

Description

Technical Field

[0001] The present invention relates to the technical field of computer networks, and in particular to an operator bandwidth allocation method based on network video traffic QoE identification. Background Art

[0002] In recent years, with the rapid development of communication technology and smart devices, the number of internet users has continued to grow, and network transmission speeds have continued to accelerate, leading to a significant increase in network traffic, most of which is generated by video services. Optimizing the quality of user experience for watching videos by allocating bandwidth is crucial for operators, as it directly impacts user satisfaction, network operational efficiency, and enterprise competitiveness.

[0003] IP bandwidth control is a rate control method implemented by operators on routers or network devices. It limits the network speed of specific devices by checking the IP addresses of devices connected to the network. Operators can pre-set the upload and download bandwidth size for each user or device based on a certain policy, according to the total network bandwidth and the number of users. Through software-defined networking (SDN), operators can flexibly and dynamically adjust bandwidth allocation strategies based on specific algorithms to optimize the utilization of network resources to cope with real-time traffic changes. Video streaming usually requires high bandwidth stability, especially during real-time transmission. If the bandwidth fluctuates, it may cause video freezes or quality degradation. The existing leaky bucket algorithm ensures that the transmission rate of video streaming remains stable by controlling the rate at which data flows in, preventing sudden traffic from affecting video playback.

[0004] During bandwidth allocation, QoE is a key metric for measuring the quality of the video streaming experience. It encompasses multiple dimensions, including visual quality, the frequency and duration of rebuffering (stuttering), and the magnitude of quality changes. These metrics often involve trade-offs. For example, improving visual quality may result in insufficient buffering, increasing the risk of stuttering.

[0005] Previous bandwidth allocation algorithms have mostly focused on Quality of Service (QoS) metrics such as latency, jitter, and packet loss. These technical indicators primarily measure the performance and stability of network transmission, ensuring that data can be efficiently and quickly transmitted from the source node to the destination node. However, while these QoS metrics are important in improving network efficiency and reliability, they do not fully consider the user's subjective experience (QoE), which refers to the user's actual experience and satisfaction when using the network. For applications such as video streaming, the user experience is not only determined by QoS metrics such as jitter and latency, but also by factors such as visual quality, rebuffering (stuttering), and quality fluctuations.

[0006] The main reason for this limitation is that operators cannot directly obtain users' actual viewing experience. For example, operators can monitor technical indicators such as bandwidth utilization, data transmission latency, and packet loss rate through network equipment, but these indicators cannot accurately reflect the visual quality or lag experienced by users when watching videos. Visual quality involves video resolution, while lag is related to the frequency and duration of buffering events, which are the experience factors that users care about most when watching videos. However, because many streaming media transmissions in the network are encrypted (for example, via HTTPS or TLS), operators find it difficult to obtain fine-grained client status information and cannot directly monitor the video encoding rate or the frequency of rebuffering events.

[0007] This information asymmetry prevents operators' bandwidth allocation algorithms from incorporating user-side QoE into their optimization models. Consequently, even when network transmission performance meets the expected QoS standards, the user's actual viewing experience (such as video clarity and smoothness) may still be subpar. For example, in some cases, even if network bandwidth is sufficient, if the adaptive video playback algorithm selects a lower-quality encoding stream, the user experience will still be subpar.

[0008] Before the widespread use of encrypted traffic, video service QoE evaluation relied on direct analysis of unencrypted traffic. Operators could easily access detailed data such as video encoding information, bitrate, and rebuffering (stuttering). However, with the adoption of encrypted protocols like HTTPS, direct traffic observation has become difficult, making it difficult for operators to assess the user experience. Encrypted traffic hides video content details, limiting the application of traditional QoE monitoring methods.

[0009] With the development of encrypted network traffic identification technology, operators can indirectly infer user QoE by deeply analyzing patterns and network behavior within encrypted traffic. For example, by analyzing packet size, transmission time, traffic patterns, round-trip time (RTT), and transport layer protocol behavior, operators can identify key events in video streaming, such as resolution changes, freezes, and video clarity. This technology uses machine learning and traffic feature analysis to effectively infer users' actual viewing experience, even when traffic is encrypted.

[0010] However, although operators can measure user QoE more accurately, it does not mean that they can directly use QoE indicators to replace the original QoS indicators for optimization. This is because:

[0011] 1. QoE evaluation typically involves complex algorithms and data processing, such as objective evaluation models (e.g., PSNR, SSIM) used in video quality measurement or prediction models based on deep learning. These algorithms typically require long computation times and struggle to provide real-time feedback at the millisecond level. QoS metrics such as latency and packet loss rate, on the other hand, can be quickly measured and adjusted on network devices, making them suitable for real-time optimization.

[0012] 2. Current network control mechanisms (such as traffic shaping, queuing policies, and congestion control) are designed based on QoS metrics. Operators' equipment and protocols are accustomed to using specific metrics such as latency, jitter, and packet loss rate for management and optimization. If QoE metrics were used directly, existing control mechanisms would likely require significant adjustments, which are often complex and could significantly increase the cost and complexity of network management.

[0013] Therefore, operators need to devise a new solution, such as storing actual network data (such as bandwidth) and user experience data (such as QoE metrics) for subsequent black-box model training. This will inform bandwidth allocation decisions and optimize future network management and service quality. Because different videos vary in size and resolution, decision-making algorithms also require identifying the specific videos users are watching to determine the mapping between bandwidth and QoE metrics for each video.

[0014] Existing network application identification technology can effectively address this issue by performing QoE and application identification based on encrypted network traffic identification technology. Operators can represent network traffic as a graph structure and leverage the powerful feature learning and pattern recognition capabilities of graph neural networks (GNNs) to identify the web pages users are viewing in a fine-grained manner. Specifically, each data packet in the network traffic can be viewed as a node in the graph, while the communication relationships or transmission directions between these nodes constitute the edges of the graph. In this way, encrypted traffic, which was originally presented as time series data, is converted into a graph model that can capture complex relationships and structures. Even if the data is encrypted and the specific content cannot be directly obtained, the characteristics of the traffic (such as packet size, time interval, transmission direction, etc.) can still provide sufficient discriminative information to help the model identify the differences between different applications, thereby achieving application classification.

[0015] In summary, optimizing network bandwidth resource allocation based on QoS is unreasonable. Since existing encrypted network traffic identification technology can identify QoE and applications, operators can consider optimizing bandwidth resource allocation by establishing mathematical models that correlate different video bandwidths with QoE. Summary of the Invention

[0016] Based on the technical problems existing in the background technology, the present invention proposes an operator bandwidth allocation method based on network video traffic QoE identification, which improves the accuracy and effectiveness of bandwidth allocation.

[0017] The present invention proposes an operator bandwidth allocation method based on network video traffic QoE identification. Based on the users who play set videos simultaneously, the set video traffic information and predicted bandwidth information are input into a trained bandwidth allocation model to output the user's bandwidth allocation strategy.

[0018] The training process of the bandwidth allocation model is as follows:

[0019] Step 1: Specify N videos as optimization targets in advance, and select the first k seconds of each video for optimization to align the video lengths.

[0020] Step 2: Deploy the webpage fine-grained recognition module and the user QoE recognition module at the network egress node. The webpage fine-grained recognition module filters the multi-stream network traffic to obtain a set video. The user QoE recognition module's recognition interval is t seconds. In a video length of k seconds, a total of k / t QoEs are identified.

[0021] Step 3: Monitor the user's bandwidth. Based on the recognition interval of the user's QoE recognition module and the time length of the video website switching resolution, the bandwidth of k / t monitoring times is obtained. For a set video and a set of set bandwidths, the corresponding QoE mn The value of is unique, so the recurrent neural network is used to fit the QoE and bandwidth function;

[0022] Step 4: Taking maximizing the sum of the user QoE values ​​of N set videos as the ultimate goal, combining the set constraints and boundary conditions as well as the QoE and bandwidth functions, the optimal bandwidth allocation strategy is obtained based on Bayesian optimization.

[0023] Furthermore, in step 4, the final goal is specifically:

[0024]

[0025] Among them, i is the index of the set video, l i is the number of users who watched the i-th set video, Z i The bandwidth required for setting the video for the i-th video, QoE i is the QoE of the i-th set video recognition.

[0026] Furthermore, in step 4, the constraints and boundary conditions are:

[0027] Constraint: In any time period, the sum of the bandwidth allocated to each set video is less than or equal to the total bandwidth;

[0028] Boundary condition: The bandwidth allocated to each set video is not negative and is less than or equal to the total bandwidth.

[0029] Furthermore, in step 4, the optimal bandwidth allocation strategy is obtained based on Bayesian optimization, specifically:

[0030] Assuming that a bandwidth allocation scheme Z is given, a prediction model of the objective function is constructed based on the Gaussian process. The prediction model takes the bandwidth allocation scheme Z as input and outputs the predicted mean μ(Z) and uncertainty σ(Z);

[0031] Based on the acquisition function that introduces constraints and boundary conditions, the next bandwidth allocation scheme to be evaluated is selected. The acquisition function is optimized using an optimization algorithm to maximize the improvement value of the current solution relative to the optimal solution. The bandwidth allocation scheme corresponding to the improvement value is used as the optimal bandwidth allocation strategy.

[0032] Furthermore, the acquisition function of constraints and boundary conditions is introduced, specifically:

[0033] α CEI (Z) = α EI (Z)·P(c1(Z)≤0)·P(c2(Z)≤0)·P(c3(Z)≤0);

[0034]

[0035] Among them, α CEI is the improvement value of the current solution corresponding to the acquisition function that introduces constraints and boundary conditions relative to the optimal solution, α EI is the improvement value of the current solution relative to the optimal solution corresponding to the acquisition function without introducing constraints and boundary conditions, Z * is the optimal bandwidth allocation scheme, Φ and φ are the cumulative distribution function and probability density function of the standard normal distribution, c1(Z) is the constraint condition that the sum of the bandwidth allocated to each set video is less than or equal to the total bandwidth, c2(Z) is the boundary condition that the bandwidth allocated to each set video is less than or equal to the total bandwidth, c3(Z) is the boundary condition that the bandwidth allocated to each set video is not negative, and P(·) represents the probability of the constraints and boundary conditions occurring.

[0036] A computer device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned operator bandwidth allocation method when executing the computer program.

[0037] A computer-readable storage medium stores a plurality of classification programs, wherein the plurality of classification programs are used to be called by a processor and execute the operator bandwidth allocation method as described above.

[0038] Signal connections include wired connections and wireless connections; wired connections include broadband connections and fiber optic connections; wireless connections include WIFI connections and radio connections.

[0039] Those skilled in the art will understand that all or part of the steps of implementing the above-mentioned method embodiment can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above-mentioned method embodiment; and the aforementioned storage medium includes: ROM, RAM, disk or optical disk, etc. Various media that can store program codes.

[0040] The advantages of the operator bandwidth allocation method based on network video traffic QoE identification provided by the present invention are: comprehensively considering the QoE characteristics and web page fingerprint characteristics that the operator can identify through encrypted traffic, allocating bandwidth with the goal of maximizing user QoE; first concentrating and then allocating the bandwidth of users watching set videos, improving the QoE of users watching set videos without affecting the experience of unrelated users; inputting encrypted traffic into a graph neural network and a convolutional neural network to construct a QoE-bandwidth dataset, and then using a recurrent neural network fitting function, so that the operator can estimate the user QoE experience under different bandwidth conditions, and at the same time use the Bayesian optimization algorithm to make reasonable bandwidth allocation decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 Figure 1. Training framework diagram for bandwidth allocation model.

[0042] Figure 2 The execution framework diagram after bandwidth allocation model training is completed. DETAILED DESCRIPTION

[0043] The technical solutions of the present invention are described in detail below through specific embodiments. Numerous specific details are set forth in the following description to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways than those described herein, and those skilled in the art may make similar modifications without departing from the scope of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0044] like Figure 1 and 2As shown, the present invention proposes an operator bandwidth allocation method based on network video traffic QoE identification. Based on users who play set videos at the same time, the set video traffic information and predicted bandwidth information are input into the trained bandwidth allocation model to output the user's bandwidth allocation strategy.

[0045] The bandwidth allocation model consists of a training phase and an application phase. During the training phase, operators deploy the QoE identification module and the fine-grained web page identification module at network egress nodes (such as campus network egress nodes). When the fine-grained web page identification module detects a designated video web page, it adds the corresponding web page information, QoE information, and bandwidth information to the dataset. When the dataset is large enough, a feedforward neural network (RNN) is used to fit the QoE-bandwidth curve for each designated video, and the resulting model is saved.

[0046] During the application phase, videos played within a b-second time interval are considered "simultaneous." For users playing the specified videos simultaneously, their bandwidth vectors for the next period are predicted and summed to form the total bandwidth. A Bayesian optimization algorithm is then used to calculate the appropriate bandwidth vector for each video from the total bandwidth and allocate it to the relevant users. New QoE-bandwidth data is also generated and added to the dataset. Once the dataset expands to a certain size, the QoE-bandwidth function is fitted to improve the accuracy of the fit.

[0047] Specifically, the training process of the bandwidth allocation model is as follows:

[0048] Step 1: Specify N set videos as optimization targets in advance, and select the first k seconds of each set video for optimization to align the video lengths;

[0049] Step 2: Place the web page fine-grained recognition module and the user QoE recognition module at the network egress node. Based on the web page fine-grained recognition module, filter the multi-stream network traffic to obtain the set video. The recognition interval of the user QoE recognition module is recorded as t seconds. In the k-second video length, a total of k / t QoEs are identified.

[0050] Web page fine-grained recognition module

[0051] To understand what videos users are watching, fine-grained web page recognition technology is used. Existing technologies, such as graph neural networks (GNNs), can extract effective information based on the characteristics of network data streams and the spatiotemporal correlations between them, enabling fine-grained classification of web pages. Multi-stream network traffic is fed into the fine-grained web page recognition module (FWI module). The pre-trained GNN model outputs the video of the web page being accessed by the operator's user. If the video is one of the N videos that require optimization, further operations such as bandwidth prediction can be performed.

[0052] User QoE identification module

[0053] In order to optimize user QoE, it is very necessary for operators to understand the QoE of users watching videos under different bandwidth conditions. Existing experiments have shown that the observation results of using data packet round-trip time (RTT) information to infer fine-grained QoE indicators are more effective and can better identify heavy buffering events and video resolutions. The user QoE identification module is deployed at the network exit node, combined with bandwidth monitoring and web page fine-grained identification modules to provide effective data for bandwidth-QoE fitting. Since resolution switching will affect the overall QoE, the QoE identification interval of the user QoE identification module needs to be consistent with the time length of the video website switching resolution, recorded as t seconds. In the video length of k seconds, a total of k / t QoEs are identified; that is, for the i-th set video, the QoE size of the p-th identification is QoE ip , p∈[1,k / t].

[0054] Step 3: Monitor the user's bandwidth. Based on the recognition interval of the user's QoE recognition module and the time length of the video website switching resolution, the bandwidth of k / t monitoring times is obtained. For a set video and a set of set bandwidths, the corresponding QoE mn The value of is unique, so the recurrent neural network is used to fit the QoE and bandwidth function;

[0055] Bandwidth prediction module

[0056] Operators can collect historical user bandwidth usage data from channels such as network logs and traffic monitoring systems. By analyzing historical user bandwidth usage data, operators can predict future bandwidth usage. Specifically, at a large scale, user bandwidth usage exhibits certain regularities, allowing the use of a moving average method to reveal potential long-term trends. At a small scale, however, there is some randomness, allowing the use of an exponential smoothing algorithm that assigns higher weights to recent data points. By combining these two prediction algorithms, bandwidth usage can be accurately predicted for the next k / t times.

[0057] Bandwidth-QoE relationship modeling module

[0058] The operator can monitor the user's bandwidth. For the i-th set video, the bandwidth monitored k / t times is listed as a vector B = [b1, b2, ..., b k / t ], and its corresponding QoE is ∑QoE ip For a given video n, and a specific set of bandwidths B m , which corresponds to QoE mnThe value of is unique, which depends on the size of the video itself and the download algorithm set by the content provider. Although the operator cannot directly know this information, the user QoE identification module and bandwidth monitoring module can provide the bandwidth vector and QoE corresponding data of each set video. When there is enough data, RNN can be used to fit the QoE. n The relationship between bandwidth B and QoE n Set the QoE of the nth video.

[0059] Step 4: Taking maximizing the sum of the user QoE values ​​of N set videos as the ultimate goal, combining the set constraints and boundary conditions as well as the QoE and bandwidth functions, the optimal bandwidth allocation strategy is obtained based on Bayesian optimization.

[0060] Bayesian Optimization Module

[0061] Bayesian optimization is a global optimization algorithm used to optimize black-box functions that are computationally expensive or lack explicit expressions. It constructs a surrogate model (usually a Gaussian process) to approximate the target function based on known data, giving the predicted value μ(Z) and uncertainty σ(Z) for each point.

[0062] Bayesian optimization then uses an acquisition function (e.g., expected improvement, EI) to select the next point in the input space to evaluate. The acquisition function balances exploring new areas with leveraging existing information, ensuring that the global optimal solution is found more quickly.

[0063] Each time a new point is selected, the objective function is evaluated, and the new data point is added to the existing data, the proxy model is continuously updated, and this process is repeated until the optimal solution is reached or the termination condition is met.

[0064] In this embodiment, it is assumed that N set videos are [l1,l2,..l N ] people watching and the total bandwidth is limited, l n For the number of users who watch the Nth set video, the amount that needs to be decided is how much bandwidth to allocate to these N set videos, such as the bandwidth Z required to be allocated for the i-th set video i =[z 1i ,z 2i ,...,z fi ], f = k / t, z fi The bandwidth corresponding to the k / tth QoE identification; bandwidth Z i Divide by the number of users l i , the result Z i / l i That is, the bandwidth allocated to each user. The ultimate goal is to maximize the sum of the QoE values ​​of all users watching these N set videos:

[0065]

[0066] Among them, i is the index of the set video, l i is the number of users who watched the i-th set video, Z i The bandwidth required for setting the video for the i-th video, QoE i is the QoE of the i-th set video recognition.

[0067] ∑l i The total bandwidth X of each user is predicted, and the constraints and boundary conditions are as follows: Constraint: In any time period, the sum of the bandwidth allocated to each set video must be less than or equal to the total bandwidth; Boundary condition: The bandwidth allocated to each set video is not negative and is less than or equal to the total bandwidth.

[0068]

[0069] Among them, c1(Z) is the constraint condition that the sum of the bandwidth allocated to each set video is less than or equal to the total bandwidth, c2(Z) is the boundary condition that the bandwidth allocated to each set video is less than or equal to the total bandwidth, c3(Z) is the boundary condition that the bandwidth allocated to each set video is not negative, and Z ij represents the bandwidth allocated to the i-th video in the j-th time period, X j Indicates the total bandwidth that the operator can allocate in the jth time period.

[0070] For these problems, Bayesian optimization is suitable for solving complex objective functions, lacking a clear QoE expression, and high computational costs. Instead of directly exploring the global objective function, Bayesian optimization evaluates a small number of sample points and gradually finds the optimal solution.

[0071] The steps of Bayesian optimization to solve the bandwidth allocation problem are as follows:

[0072] First, the Gaussian process is used to build a prediction model for the objective function f(Z). Assuming a given bandwidth allocation scheme Z, the prediction model will return a predicted value (mean) μ(Z) and uncertainty (standard deviation) σ(Z). The objective function f(Z) represents the QoE value to be maximized, and the prediction model of the Gaussian process is

[0073]

[0074] Next, the acquisition function is used to select the next bandwidth allocation solution to be evaluated. A commonly used acquisition function is the expected improvement (EI), which measures the potential improvement of the current solution relative to the optimal solution. The formula for EI is:

[0075]

[0076] Among them, α EI is the improvement value of the current solution relative to the optimal solution corresponding to the acquisition function without introducing constraints and boundary conditions, Z * For the optimal bandwidth allocation solution, Φ and φ are the cumulative distribution function and probability density function of the standard normal distribution, respectively. This acquisition function combines the mean and uncertainty to effectively explore the solution space.

[0077] The constraints and boundary conditions are then introduced into the acquisition function:

[0078] α CEI (Z) = α EI (Z)·P(c1(Z)≤0)·P(c2(Z)≤0)·P(c3(Z)≤0);

[0079] Among them, α CEI is the improvement value of the current solution corresponding to the acquisition function that introduces constraints and boundary conditions relative to the optimal solution. Φ and φ are the cumulative distribution function and probability density function of the standard normal distribution, respectively. c1(Z) is the constraint condition that the sum of the bandwidth allocated to each set video is less than or equal to the total bandwidth. c2(Z) is the boundary condition that the bandwidth allocated to each set video is less than or equal to the total bandwidth. c3(Z) is the boundary condition that the bandwidth allocated to each set video is not negative. P(·) represents the probability of the occurrence of the constraints and boundary conditions.

[0080] Finally, an optimization algorithm (such as L-BFGS) is used to optimize the acquisition function that introduces constraints and boundary conditions to find the one that maximizes α CEI (Z) means that the optimal bandwidth allocation strategy can be found.

[0081] Bandwidth allocation module

[0082] Based on the calculated Z, the operator can allocate bandwidth to users watching a specific video in k / t time periods over the next k seconds. It's worth noting that since operators can't precisely control bandwidth allocation to be exactly as expected, the actual bandwidth allocated will follow a normal distribution with the optimal bandwidth as the mean.

[0083] Through steps one to three, compared to traditional bandwidth allocation algorithms, this embodiment comprehensively considers the QoE characteristics and webpage fingerprint characteristics that operators can identify through encrypted traffic, and allocates bandwidth with the goal of maximizing user QoE. Compared to traditional methods, this embodiment first concentrates and then redistributes the bandwidth of users watching set videos, improving the QoE of users watching set videos without affecting the experience of unrelated users. This embodiment inputs encrypted traffic into a graph neural network and a convolutional neural network to construct a QoE-bandwidth dataset, and then uses a recurrent neural network fitting function to enable operators to estimate the user QoE experience under different bandwidth conditions, while using a Bayesian optimization algorithm to make reasonable bandwidth allocation decisions.

[0084] This embodiment introduces joint optimization based on QoE metrics. Operators can leverage large-scale user data and traffic characteristics to train more refined models, predict the impact of different bandwidth allocations on user experience, and then determine the bandwidth required for different video viewers. This not only maximizes video service quality, but also reduces unnecessary network congestion and optimizes bandwidth usage, thereby improving the accuracy and effectiveness of bandwidth allocation.

[0085] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A method for allocating operator bandwidth based on network video traffic QoE identification, characterized in that: Based on the users who are playing the specified videos at the same time, the specified video traffic information and the predicted bandwidth information are input into the trained bandwidth allocation model to output the user's bandwidth allocation strategy; The training process of the bandwidth allocation model is as follows: Step 1: Specify N videos as optimization targets in advance, and select the first k seconds of each video for optimization to align the video lengths. Step 2: Place the web page fine-grained recognition module and the user QoE recognition module at the network exit node. Based on the web page fine-grained recognition module, filter the multi-stream network traffic to obtain the set video. The recognition interval time of the user QoE recognition module is recorded as Seconds, in In the video length of seconds, a total of Secondary QoE; Step 3: Monitor the user's bandwidth. Based on the recognition interval of the user QoE recognition module and the time length of the video website switching resolution, we can get The bandwidth of the monitoring, for a set video and a set of set bandwidth, corresponds to The value of is unique, so the recurrent neural network is used to fit the QoE and bandwidth function; Step 4: Taking maximizing the sum of the user QoE values ​​of N set videos as the ultimate goal, the optimal bandwidth allocation strategy is obtained based on Bayesian optimization, combining the set constraints and boundary conditions as well as the QoE and bandwidth functions; In step 4, the final goal is to: in, To set the index of the video, To watch the The number of users who set the video, For the Set the bandwidth that the video needs to allocate. For the Set the QoE of video recognition; In step 4, the optimal bandwidth allocation strategy is obtained based on Bayesian optimization, specifically: Assuming a given bandwidth allocation scheme Z, the objective function is constructed based on the Gaussian process The prediction model takes the bandwidth allocation scheme Z as input and outputs the predicted mean and uncertainty , represents the QoE value to be maximized, and the prediction model of the Gaussian process is ; Based on the acquisition function that introduces constraints and boundary conditions, the next bandwidth allocation solution to be evaluated is selected. The acquisition function is optimized using an optimization algorithm to maximize the improvement value of the current solution relative to the optimal solution. The bandwidth allocation solution corresponding to this improvement value is used as the optimal bandwidth allocation strategy. The acquisition function that introduces constraints and boundary conditions is as follows: in, is the improvement value of the current solution corresponding to the acquisition function that introduces constraints and boundary conditions relative to the optimal solution, is the improvement value of the current solution relative to the optimal solution corresponding to the acquisition function without introducing constraints and boundary conditions, is the optimal bandwidth allocation scheme for the current round, and are the cumulative distribution function and probability density function of the standard normal distribution, The constraint condition is that the sum of bandwidth allocated to each set video is less than or equal to the total bandwidth. For each set video, the bandwidth allocated is less than or equal to the total bandwidth. For each set of video, the bandwidth allocated is not negative. Represents the probability of constraints and boundary conditions occurring.

2. The method for allocating operator bandwidth based on network video traffic QoE identification according to claim 1, characterized in that: In step 4, the constraints and boundary conditions are: Constraint: In any time period, the sum of the bandwidth allocated to each set video is less than or equal to the total bandwidth; Boundary condition: The bandwidth allocated to each set video is not negative and is less than or equal to the total bandwidth.

3. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the operator bandwidth allocation method according to any one of claims 1 to 2 is implemented.

4. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a plurality of classification programs, which are used to be called by a processor and execute the operator bandwidth allocation method according to any one of claims 1 to 2.

Images