A joint video download and dynamic bitrate adaptation method for short video streams

Through the joint video download and dynamic bit rate adaptive method, the problem of bandwidth wasted when watching short videos is solved. By dynamically adjusting the buffer and bit rate, it can maximize bandwidth saving and user experience without reducing the user experience quality promote.

CN115914680BActive Publication Date: 2024-11-08COMMUNICATION UNIVERSITY OF CHINA
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202211251336.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-10-13
Publication Date
2024-11-08
Estimated Expiration
2042-10-13

AI Technical Summary

Technical Problem

The prior art is difficult to save bandwidth resources during short video viewing without degrading the quality of user experience, especially the bandwidth waste caused by users frequently switching videos has not been effectively solved.

Method used

Using a joint video download and dynamic bit rate adaptive method for short video streams, the maximum buffer size and the bit rate of video blocks are dynamically adjusted through the preloading mechanism and the bit rate decision algorithm, combining the past real bandwidth and future video blocks Retention rate, optimize video download strategies and bit rate selection to ensure bandwidth saving without reducing user experience quality.

Benefits of technology

It achieves a reduction in bandwidth waste of up to 66.74% under various network conditions and improves the quality of user experience (QoE) by 42.5% and effectively utilizes bandwidth resources.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115914680B_ABST
    Figure CN115914680B_ABST
Patent Text Reader

Abstract

A joint video download and dynamic bitrate adaptation method for short video streams relates to the technical field of short video bitrate adaptation. The present invention considers how to combine the video download mechanism with the bitrate adaptation technology when the user watches the short video, and provides a new short video download mechanism and ABR algorithm to save bandwidth overhead without reducing the user experience quality (QoE). The present invention includes two specific implementation algorithms: a preloading mechanism and a hybrid bitrate adaptation method.
Need to check novelty before this filing date? Find Prior Art

Description

Technical field:

[0001] The present invention relates to the technical field of short video bitrate adaptation, which is a combined video download and dynamic bitrate adaptation method for short video streams. The main consideration is how to combine the video download mechanism with the bitrate adaptation technology when users watch short videos, so that the algorithm can both ensure the user experience quality and save bandwidth resources. Background technology:

[0002] Adaptive Bitrate Algorithm is referred to as ABR algorithm. Its main research purpose is to improve the quality of user experience, or QoE. ABR algorithm takes the future bandwidth and the buffer size of the player as input and the bitrate of the future video block as output. It is mainly divided into three categories, including rate-based ABR algorithm, buffer-based ABR algorithm and reinforcement learning-based ABR algorithm.

[0003] The ABR algorithm based on rate prediction is a very intuitive adaptive bitrate algorithm, and the representative algorithm is FESITIVE. The main idea of ​​this type of method is to predict the current or future network bandwidth based on the network bandwidth during the past video download, and make decisions on the adaptive bitrate of the video. When the network condition is good, the past network bandwidth is high, and the current download video block chooses a high bitrate; when the network condition is poor, the past network bandwidth is low, and the current download video block chooses a low bitrate to maximize the video quality without lag. The buffer-based ABR algorithm only predicts based on the current buffer time of the player, and the representative algorithms are BBA and BOLA. When the buffer time is long, it means that the network condition is good or the current download bitrate does not effectively utilize the bandwidth, and the bitrate of the next video block is increased; when the buffer time is short, it means that the network condition is poor or the current download bitrate is too high, and the bitrate of the next video block is reduced. The structures of these two algorithms are relatively simple, but the ABR algorithm must balance various QoE goals, including maximizing video quality, minimizing the occurrence of interruption events, and maintaining video quality smoothness. However, many of these goals are inherently conflicting, and it is difficult for rate-based and buffer-based ABR algorithms to meet these requirements at the same time. In order to learn a better ABR model, the latest ABR algorithm currently uses reinforcement learning to predict the bit rate, and the representative algorithm is Pensieve. The ABR algorithm based on reinforcement learning is a data-driven algorithm, so a large-scale data set is required to train a more accurate model. When the data set is small, the trained model has poor performance and cannot be generalized.

[0004] Currently, there is no ABR algorithm model specifically designed for short videos. The design of the ABR algorithm for short video streams needs to take into account the user's QoE and the utilization of bandwidth resources. Summary of the invention:

[0005] The invention is due to the fact that when watching short videos, if the user is not interested in the content, he will switch videos frequently. In this way, the buffered videos are not watched, which will cause a large amount of bandwidth resources to be wasted, and it will not help to improve the quality of user experience. Since saving bandwidth conflicts with the high QoE that users want to obtain, it is very challenging to set the download mechanism of short video streams. The purpose of the invention is to solve the problem of saving bandwidth overhead without reducing the QoE perceived by users.

[0006] The current mainstream short video sharing solutions are: Figure 1 As shown in the figure: Short videos are generated by content producers and published on the Internet. After being processed by the network server, users can watch these short videos on the client. Usually, the video recommendation system will recommend some videos to users, and these videos form a recommendation queue. In addition to watching the current video, users can also switch down and watch videos in the recommendation queue. In order to ensure the smoothness of users' viewing when switching short videos, the system needs to preload the current video as well as the videos in the recommendation queue. The current video can be divided into three parts: watched, preloaded but not watched, and not loaded. Figure 1 In the figure, they are represented by black, gray and white respectively.

[0007] Each short video is divided into several video blocks. In order to make a more fine-grained decision on the video download method, the length of each video block L is set to 1s. The bit rate R of the video block is divided into three levels R∈{0,1,2}, representing [750,1200,1850]kb / s respectively. The bit rate level needs to be determined according to the bit rate of the video being encoded. Every 5 videos form a queue. When the user finishes watching a video, a video is added to the queue.

[0008] The retention rate describes the probability of a user watching a certain video block. The user's retention rate is obtained by counting the user's behavior habits. The short video retention rate dataset used is a public dataset, and its data structure pattern is as follows: Figure 5 As shown, each video block corresponds to a block size and a user retention rate.

[0009] A buffer is a storage space of a specified size reserved in the memory for temporary storage of input / output data. This reserved memory space is called a buffer. The maximum buffer refers to the maximum range that the buffer can reach.

[0010] The overall structure of the present invention is as follows Figure 2As shown in the figure, it mainly includes two key points: first, determine the video block of the current video or the video in the recommended queue to be downloaded, implement the "preloading mechanism", first set the dynamic maximum buffer, and then determine the download mechanism of the video queue. Second, determine the bit rate size of the video block to be downloaded, implement the "bit rate decision algorithm", first predict the future bandwidth, and then make an adaptive bit rate selection.

[0011] The first is the preloading mechanism for short video streams, which inputs the real bandwidth of past video blocks and the retention rate of future video blocks, and outputs the sequence number of the video to be downloaded.

[0012] like Figure 3 The flowchart of the algorithm is shown below: before downloading each video block, the maximum buffer size of the currently downloaded video block needs to be calculated based on the past real bandwidth and the retention rate of future video blocks. The maximum buffer size changes dynamically according to different network conditions and video blocks. When the video being watched has not been downloaded and the current video buffer is smaller than the maximum buffer, the system will select the current video for download; if the current video has been downloaded to the maximum buffer, it is determined whether the video in the recommended queue meets the download conditions. It is determined in turn whether the videos in the recommended queue have not been downloaded to the maximum buffer and the retention rate of the next video block is greater than the threshold, and then the videos that meet the conditions are downloaded in turn. Until the current video and the videos in the queue are downloaded to the preload size that meets the above requirements, the player will temporarily stop downloading for a period of time. The stop time calculation method is fully described in the specific implementation method. After the time is up, the next video block will continue to be downloaded and the above judgment and decision will be repeated.

[0013] The second is the bitrate decision algorithm for the video block. The input is the actual bandwidth and the actual buffer size in the past, and the output is the bitrate size of the next video block to be downloaded.

[0014] like Figure 4 The following is the flowchart of the algorithm: After the preloading mechanism algorithm determines the serial number of the video to be downloaded, the bitrate decision algorithm determines the bitrate of the video block to be downloaded. First, based on the actual bandwidth of the past 5 seconds, the harmonic mean method is used to preliminarily predict the future bandwidth. Then, the preliminarily predicted bandwidth is smoothed and corrected with a penalty factor and a secondary correction is performed based on the buffer size. The two corrections are to reduce the impact of bandwidth extreme values ​​and to make full use of the cache area. Then, all possible bitrates for the next 5 video blocks are arranged and combined, and the performance benefits of all bitrate combinations of the past block and the next 5 blocks are calculated. Performance is a comprehensive indicator composed of four elements: bitrate, interruption time, smoothness, and bandwidth utilization, which represents the performance benefits brought by each bitrate combination. Then, an exhaustive method is used to find the bitrate combination with the highest performance, and the bitrate of the first future block in the combination is used as the decision value. Description of the drawings:

[0015] The accompanying drawing shows a schematic diagram of the invention principle of the joint video downloading and dynamic bit rate adaptation method for short video streams.

[0016] Figure 1 : Short video sharing solution.

[0017] Figure 2 : The overall architecture of the joint video downloading and dynamic bitrate adaptation method for short video streams.

[0018] Figure 3 : Flowchart of preloading mechanism for short video streams.

[0019] Figure 4 : Flowchart of bitrate decision for video blocks.

[0020] Figure 5 : Data structure diagram of short video dataset. Specific implementation plan:

[0021] 1. Preloading mechanism of short video stream:

[0022] The preloading mechanism is the first key point of the present invention. The real bandwidth of the past video blocks and the retention rate of the future video blocks are input, and the sequence number of the video to be downloaded is output, where -1 represents pausing the download, 0 represents the current video, and the sequence number greater than or equal to 1 represents the video in the queue. In order to save bandwidth waste, the present invention can dynamically adjust the maximum buffer according to the feedback of the last operation. The preloading mechanism first calculates the maximum buffer size, and then determines the download strategy according to the buffer size and the retention rate of the video block. The following are the steps for the preloading mechanism to download the video:

[0023] Step 101: i represents the i-th video in the recommendation queue, and i=0 represents the video currently being watched. The present invention sets the maximum buffer of the k-th block in the i-th video to The bit rate of the kth block in the i-th video is set to The kth block in the i-th video has a bit rate The size of the code is is the average download speed of the kth chunk in the ith video, Δt k is the time to pause downloading, L is the duration of each video block, and t k+1 is the download time of the k+1th block, t k is the download time of the kth block. Then the download time of the kth block in the ith video is:

[0024]

[0025] Step 102: When all videos are downloaded to the maximum allowed download size, the player stops downloading. L represents the duration of each video block. represents the buffer occupancy rate when the video player downloads the kth block in the i-th video. The download pause duration Δt k for:

[0026]

[0027] The value range of is:

[0028]

[0029] Step 103: Set is the retention rate of the kth block in the ith video. In order to reduce the waste of bandwidth resources, a dynamic maximum buffer is designed as shown below Model:

[0030]

[0031]

[0032]

[0033] X stands for The first possible value of, in the X expression, Indicates e rebuf is the time of freezing. If freezing occurs at the kth block in the ith video, it indicates that the network condition is poor and the maximum buffer size needs to be increased. If the download is paused after the kth block in the ith video, it indicates that the network condition is good and the maximum buffer size needs to be reduced. In order to prevent freezing due to a small buffer size when playing the k+1th block in the ith video, the Y expression is added. Y represents The second possible value of is to set the maximum buffer to the length L of the previous video block of video i plus the total operation time of a video block, where the total operation time includes the download time of the previous video block. Pause download duration Δt k As mentioned earlier, X is used to determine the maximum buffer size, and Y is used to prevent video freezes, so the dynamic maximum buffer It is the maximum value of X and Y. max(X,Y) means taking the maximum value of X and Y.

[0034] Step 104: The player first downloads the currently playing video until the buffer size The maximum buffer size reached at this moment

[0035] Step 105: After the current video is downloaded to the maximum buffer, the videos in the recommended queue are then downloaded. The maximum allowed preload size of the videos in the recommended queue is determined by and the retention rate threshold P of video i i To prevent buffering of the next video to be viewed When the user swipes away from the video, the length of the preloaded video fluctuates. The maximum buffer of the video in the recommendation queue is set to be consistent with the maximum buffer of the currently playing video. The video you are currently watching Be consistent.

[0036] Step 106: Set is the retention rate of the kth block in the ith video. The retention rate in the real dataset is obtained by counting the user's viewing behavior, and the retention rate in the simulated dataset is generated by software. This reflects the video's appeal to users. i The retention rate of each video block of downloaded video i is lower than that of the previous video block by a certain percentage, which indicates the degree and trend of the decrease in retention rate. i When it reaches the maximum, it means that the attractiveness of the current video block has been reduced to the minimum compared with the previous video block. As the retention rate threshold O i .O i The calculation formula is as follows, d i Indicates Dangdang i When the maximum O i The value of

[0037]

[0038]

[0039] Step 107: The videos in the queue need to be downloaded to the video blocks whose retention rate is greater than or equal to the threshold and whose buffer does not exceed the maximum buffer. When all the videos in the queue are downloaded to the blocks that meet the above constraints, the download is suspended for a pause time Δt k Calculated in step 102.

[0040] When the time is up, the next video block is downloaded and steps 101 to 107 are repeated.

[0041] 2. Video block bit rate decision algorithm:

[0042] The bitrate decision algorithm is the second key point of the present invention. The input is the past real bandwidth and the real buffer size, and the output is the bitrate size of the next video block to be downloaded. The bitrate decision algorithm first performs bandwidth prediction before selecting the bitrate, and uses the bandwidth prediction result combined with the buffer size as the main factor for bitrate decision. Since the serial number of the video to be downloaded has been determined in the preloading mechanism, the bitrate decision algorithm does not involve multiple videos, but involves multiple video blocks of the same video. Let k represent the k-th video block.

[0043] Step 201: When initially predicting the future bandwidth, use a sliding window of length W to observe the real bandwidth of the past W blocks. Currently, W is set to 5, and then use the harmonic mean method to initially predict the future bandwidth. Assume that the real bandwidth when downloading the k-th block in the past is b k , where the value range of k is 0 < k ≤ W, then the initially predicted future bandwidth b1 is

[0044]

[0045] Step 202: Since the harmonic mean is vulnerable to extreme values, when the real bandwidth fluctuates greatly, the bandwidth predicted in Step 201 may also have a large error. Use the penalty factor p to correct the bandwidth predicted in Step 201. The bandwidth correction formula is as follows, where p is the standard deviation of the past W blocks, and b2 is the smoothed and corrected bandwidth.

[0046]

[0047] b2 = b1 / (1 + p)

[0048] Step 203: Define the bandwidth below 2 Mbps as low bandwidth, and the bandwidth above or equal to 2 Mbps as high bandwidth. Mbps = Mbit / s, that is, megabits per second, which means the average number of bits of data transmitted per unit time. When the bandwidth is low bandwidth and the buffer has not reached the maximum buffer, reduce the bandwidth predicted by the bandwidth prediction module to make the bitrate selection conservative. Define the buffer correction factor m as follows, where b3 is the size of the future bandwidth corrected according to the buffer size:

[0049]

[0050] b3 = m * b2

[0051] where m1 and m2 are the upper and lower bounds of m respectively. B k (k) represents the current capacity of the buffer, and B max (k) represents the maximum capacity of the buffer. m reflects the scaling degree of the bandwidth predicted by the bandwidth prediction module according to the buffer state.

[0052] Step 204: Arrange and combine all possible bitrates for the next five video blocks, and then calculate the performance of all bitrate combinations for the past block and the next five blocks. Performance is a comprehensive indicator composed of four elements: bitrate, rebuf, smoothness, and bandwidth usage, bandwidth_usage, which represents the performance benefits brought by each bitrate combination. Bitrate is the sum of the encoding bitrates selected for each video block, rebuf is the sum of the duration of playback interruptions due to poor network conditions, smoothness is the sum of the bitrate differences between two video blocks, and bandwidth usage bandwidth_usage is the size of each video block. a, b, c, d are the coefficients of each element, representing the importance of each element to the final performance benefit. Then, an exhaustive method is used to find the bitrate combination with the greatest performance benefit, and the bitrate of the first future block in the combination is used as the decision value. The performance calculation formula is as follows:

[0053] performance

[0054] =a*bitrate+b*smooth-c*rebuf-d*bandwidth_usage

[0055] Step 205: In order to adapt to different network conditions, the factors considered in the bitrate decision under different bandwidths have different focuses. When the bandwidth is good, the bitrate decision algorithm should focus more on considering the use of a higher bitrate and making full use of the bandwidth. When the bandwidth is very poor, the bitrate decision algorithm should focus more on avoiding interruptions, reducing bitrate fluctuations, and maintaining smooth playback. Therefore, the coefficients a, b, c, and d of each element should take different values ​​under different bandwidths. The design idea of ​​each element coefficient is to observe the monotonic relationship between the coefficient value and the final actual performance, and then select a suitable function to fit the optimal parameter configuration offline based on the observation results. According to the observation of the monotonic relationship between the parameter value and the performance, when the network conditions gradually improve, that is, when the bandwidth gradually increases, if you want to maintain good performance, the coefficients a, b, c, and d of each element should be changed as follows: the bitrate coefficient a gradually increases, and the smoothness coefficient b, the interruption time (rebuf) coefficient c, and the bandwidth utilization (bandwidth_usage) coefficient d gradually decrease. That is, as the average bandwidth increases, the impact of bit rate on the final performance increases, while the impact of smoothness, interruption time, and bandwidth utilization on the final performance decreases. This can achieve the effect of using a higher bit rate under high bandwidth, fully utilizing bandwidth, maintaining smooth playback under low bandwidth, and reducing bandwidth waste. The specific settings of each element coefficient a, b, c, d are as follows, where level represents the average bandwidth of the past five video blocks, and level is used as an intermediate parameter to dynamically adjust each element coefficient a, b, c, d;

[0056]

[0057]

[0058]

[0059] c=exp(-level)

[0060] d=exp(-(level-1.3) 2 )

[0061] The coefficients of each element a, b, c, and d are obtained through the above-mentioned monotonic relationship and fitting it, so the functions and constants involved in the formula are obtained through parameter adjustment.

[0062] A joint video download and dynamic bitrate adaptation method for short video streams fully utilizes the past bandwidth and buffer size, combines the preloading mechanism with the bitrate decision algorithm, and ensures that the effect of saving bandwidth resources is achieved without reducing the user's QoE. Experiments on public data sets have shown that compared with existing download algorithms for short video streams, under various network conditions, the method of the present invention can reduce bandwidth waste by up to 66.74% and improve QoE by up to 42.5%.

Claims

1. A joint video download and dynamic bitrate adaptation method for short video streams, characterized in that Includes the following: Preloading mechanism for short video streams: Input the real bandwidth of past video blocks and the retention rate of future video blocks, and output the sequence number of the video to be downloaded, where -1 represents pausing the download, 0 represents the current video, and a sequence number greater than or equal to 1 represents a video in the queue; including the following steps: Step 101: i represents the i-th video in the recommendation queue, i=0 represents the video currently being watched; set the maximum buffer of the k-th block in the i-th video to The bit rate of the kth block in the ith video is set to The kth block in the i-th video has a bit rate The size of the code is is the average download speed of the kth chunk in the ith video, Δt k is the time to pause downloading, L is the duration of each video block, and t k+1 is the download time of the k+1th block, t k is the download time of the kth block; then the download time of the kth block in the ith video is: Step 102: When all videos are downloaded to the maximum allowed download size, the player stops downloading; L represents the duration of each video block, represents the buffer occupancy rate when the video player downloads the kth block in the i-th video; the download pause duration Δt k for: The value range of is: Step 103: Set is the retention rate of the kth block in the i-th video; in order to reduce the waste of bandwidth resources, a dynamic maximum buffer is designed as shown below Model: X stands for The first possible value of, in the X expression, Indicates e rebuf represents the time of freezing. If freezing occurs at the kth block in the i-th video, it indicates that the network condition is poor and the maximum buffer size needs to be increased. If the download is paused after the kth block in the i-th video, it indicates that the network condition is good and the maximum buffer size needs to be reduced. In order to prevent freezing due to a small buffer size when playing the k+1th block in the i-th video, the Y expression is added. Y represents The second possible value of is to set the maximum buffer to the length L of the previous video block of video i plus the total operation time of a video block, where the total operation time includes the download time of the previous video block. Pause download duration Δt k And the jam time rebuf; Use X to determine the maximum buffer size, and use Y to prevent video stuttering, so dynamic maximum buffer It is the maximum value of X and Y; max(X, Y) means taking the maximum value of X and Y; Step 104: The player first downloads the currently playing video until the buffer size The maximum buffer size reached at this moment Step 105: After the current video is downloaded to the maximum buffer, the videos in the recommended queue are then downloaded; the maximum allowed preload size of the videos in the recommended queue is determined by and the retention rate threshold P of video i i Joint decision; to prevent buffering of the next video to be viewed When the user swipes away from the video, the length of the preloaded video fluctuates. The maximum buffer of the video in the recommendation queue is set to be consistent with the maximum buffer of the currently playing video. The video you are currently watching Stay consistent; Step 106: Set is the retention rate of the kth block in the ith video. The retention rate in the real data set is obtained by counting the user's viewing behavior, and the retention rate in the simulated data set is generated by software; This reflects the video's appeal to users. i The retention rate of each video block of downloaded video i is lower than that of the previous video block by a certain percentage, which indicates the degree and trend of the decrease in retention rate. i When it reaches the maximum, it means that the attractiveness of the current video block has been reduced to the minimum compared with the previous video block; therefore, the retention rate of the previous video block is reduced to As the retention rate threshold P i ;P i The calculation formula is as follows, Indicates that when d i When the maximum P i The value of Step 107: The videos in the queue need to be downloaded to the video blocks whose retention rate is greater than or equal to the threshold and whose buffer does not exceed the maximum buffer; when all the videos in the queue are downloaded to the blocks that meet the above constraints, the download is suspended for a suspension time Δt k Calculated by step 102; When the time is up, continue downloading the next video block and repeat steps 101 to 107; Video block bit rate decision algorithm: The input is the past real bandwidth and the real buffer size, and the output is the bitrate size of the next video block to be downloaded; the bitrate decision algorithm first predicts the bandwidth before selecting the bitrate, and uses the bandwidth prediction result combined with the buffer size as the main factor for bitrate decision; since the sequence number of the video to be downloaded has been determined in the preloading mechanism, the bitrate decision algorithm does not involve multiple videos, but multiple video blocks of the same video, and k represents the kth video block; Step 201: When making a preliminary prediction of future bandwidth, a sliding window of length W is used to observe the actual bandwidth of the past W blocks, where W is currently 5, and then the harmonic mean method is used to make a preliminary prediction of future bandwidth; assuming that the actual bandwidth when downloading the kth block in the past is b k , the value range of k is 0<k≤W, then the preliminary predicted future bandwidth b1 is as follows; Step 202: Use the penalty factor p to correct the bandwidth predicted in step 201. The bandwidth correction formula is as follows: p is the standard deviation of the past W blocks, and b2 is the bandwidth after smoothing correction; b2=b1 / (1+p) Step 203: bandwidth lower than 2 Mbps is called low bandwidth, and bandwidth higher than or equal to 2 Mbps is called high bandwidth. Mbps = Mbit / s, i.e., megabits per second, which means the average number of bits of data transmitted within a time period. When the bandwidth is low bandwidth and the buffer does not reach the maximum buffer, the bandwidth predicted by the bandwidth prediction module is reduced to make the bit rate selection conservative. The buffer correction factor m is defined as follows, where b3 is the size of the future bandwidth after correction according to the buffer size: b3=m*b2 Among them, m1 and m2 are the upper and lower bounds of m respectively, and B k (k) represents the capacity of the current buffer, B max (k) represents the maximum capacity of the cache area; m reflects the degree of scaling of the bandwidth predicted by the bandwidth prediction module according to the cache status; Step 204: All possible bitrates of the next five video blocks are arranged and combined, and then the performance of all bitrate combinations of the past block and the next five blocks is calculated; performance is a comprehensive indicator composed of four elements: bitrate, rebuf, smoothness, and bandwidth usage, bandwidth_usage, which represents the performance benefit brought by each bitrate combination; bitrate is the sum of the encoding bitrates selected for each video block, rebuf is the sum of the duration of playback interruption due to poor network conditions, smoothness is the sum of the bitrate differences between two video blocks, and bandwidth usage bandwidth_usage is the size of each video block; a, b, C, d are coefficients of each element, representing the importance of each element to the final performance benefit; then an exhaustive method is used to find the bitrate combination with the maximum performance benefit performance, and the bitrate of the first future block in the combination is used as the decision value; the performance calculation formula is as follows: performance=a*bitrate+b*smooth-c*rebuF-d*bandwidth_usage Step 205: According to the observation of the monotonic relationship between parameter values ​​and performance, when the network conditions gradually improve, that is, the bandwidth gradually increases, in order to maintain good performance, the coefficients a, b, c, and d of each element should be changed as follows: the bit rate coefficient a gradually increases, and the smoothness coefficient b, interruption time coefficient c, and bandwidth utilization coefficient d gradually decrease; The specific settings of each element coefficient a, b, c, d are as follows, where level represents the average bandwidth of the past five video blocks, and level is used as an intermediate parameter to dynamically adjust each element coefficient a, b, c, d; c=exp(-level) d=exp(-(level-1.3) 2 ) The coefficients of each element a, b, c, and d are obtained through the above-mentioned monotonic relationship and fitting it, so the functions and constants involved in the formula are obtained through parameter adjustment.