Statistical multiplexing method and system, electronic equipment and storage medium

Abstract

The invention provides a statistical multiplexing method and system, electronic equipment and a storage medium, and the method comprises the steps: initializing model parameters, and determining a regression model according to the initialized model parameters; the encoder pre-encodes the frame to be encoded from the first moment to generate pre-encoding feature data, and sends the pre-encoding feature data to the statistical multiplexer; and the statistical multiplexer sequentially substitutes the pre-coding feature data within T time from the second moment into the regression model by taking a frame as a unit, and predicts a prediction coding rate within the T time through the quantization parameter and the regression model, and a quantization parameter and an iterative regression model parameter are adjusted to enable the predicted coding rate to meet the allowable range of the virtual cache region and the bandwidth, so that the statistical multiplexer multiplexes code streams of a plurality of encoders in one transmission channel. According to the method and the device, different characteristics of different encoders can be isolated, and deep binding of parameters of the encoders is not needed, so that the application range of a statistical multiplexing function is widened.

Description

Technical Field

[0001] The present application relates to the field of information communication, and in particular to a statistical multiplexing method, system, electronic device and storage medium. Background Art

[0002] Statistical multiplexing technology is one of the key technologies in modern communication systems. It is mainly used in wireless communication, network transmission and multimedia processing. The core idea of ​​this technology is to manage multiple signal sources as a whole through statistical methods to achieve efficient data transmission. Statistical multiplexing technology in the field of video coding and decoding is mainly used to optimize data transmission efficiency and improve system capacity.

[0003] In the current multimedia transmission scenario, it is necessary to cope with the pressure brought by the increasing demand for media quantity and quality with limited transmission bandwidth. In the same scenario, more media quantity and higher media quality will both increase the amount of information in the transmission process, but since the cost of upgrading the transmission link is very high and may lead to waste of resources to meet short-term peak demand, multimedia service providers, such as TV stations and radio stations, need to improve their compression efficiency and improve the effective utilization of bandwidth when generating multimedia data streams.

[0004] A higher compression efficiency metric can be defined as a smaller encoded file under the same quality, which is mainly related to the encoding protocol and encoder gear parameters. If the encoding quality remains unchanged in encoders with the same parameters, the encoding size will be positively correlated with the complexity of the video, that is, the more complex the video, the larger the encoding size. Since the complexity of each video fluctuates in the time domain, the corresponding encoding size also fluctuates. The response strategy for this fluctuation will directly affect the effective use of bandwidth.

[0005] In the application scenario of multi-channel video encoding, this fluctuation may be amplified. For example, when three programs need to be transmitted in the same fixed bandwidth channel, the complexity of program A is constantly high, the complexity of program B is constantly low, and the complexity of program C fluctuates. The codec rate control can be divided into three categories: CQ (constant quantization parameter), CBR (constant bitrate), and ABR (average bitrate). CQ will cause the encoding bitrate of program B to fluctuate greatly and is not recommended. CBR and ABR, due to the fixed target bitrate, will make the encoding quality of the highly complex channel A and some channels B lower than the average to a certain extent, resulting in unreasonable bandwidth allocation.

[0006] Statistical multiplexing technology analyzes the complexity distribution of video data of each program at different times, merges the contents of multiple programs together for bit rate control, and can dynamically allocate more bandwidth to complex programs, making the picture quality of each program closer, thereby improving bandwidth utilization and solving the problem of uneven bandwidth distribution.

[0007] Specifically, video coding statistical multiplexing technology analyzes the complexity of current and future encoding of all video program streams in a program group in frames from both time and space dimensions, and predicts the required encoding bit rate based on the analysis data, thereby dynamically adjusting the encoding parameters of each frame of video for all programs in the group, so that the overall bit rate of all programs in the group is controlled within the bandwidth limit.

[0008] Currently, other manufacturers have developed and produced technologies and products related to statistical multiplexing of video coding, but there are problems such as high binding of functions to encoders, inaccurate adjustment of encoding parameters, and large delay in bit rate control that need to be solved. These problems limit the use scenarios and effects of statistical multiplexing. Summary of the invention

[0009] In order to solve one of the above technical defects, a statistical multiplexing method, system, electronic device and storage medium are provided in the embodiments of the present application.

[0010] A first aspect of an embodiment of the present application provides a statistical multiplexing method, the method comprising: Initializing model parameters, and determining a regression model according to the initialized model parameters; The encoder pre-encodes the frame to be encoded from the first moment to generate pre-encoding characteristic data, and sends the pre-encoding characteristic data to the statistical multiplexer; The statistical multiplexer sequentially substitutes the pre-coding feature data within a time T from a second moment into the regression model in units of frames, estimates the predicted coding bit rate within a time T by using quantization parameters and the regression model, and adjusts the quantization parameters and iterative regression model parameters so that the predicted coding bit rate meets the allowable range of the virtual buffer area and the bandwidth, so that the statistical multiplexer multiplexes the bit streams of multiple encoders into a transmission channel, and the second moment is later than the first moment.

[0011] A second aspect of an embodiment of the present application provides a statistical multiplexing system, the system comprising: At least one encoder, configured to pre-encode a frame to be encoded starting from a first moment to generate pre-encoding characteristic data, and send the pre-encoding characteristic data to a statistical multiplexer; A statistical multiplexer, used to sequentially substitute pre-encoded feature data within a time T from a second moment into a regression model in units of frames, estimate a predicted coding rate within the time T by using a quantization parameter and the regression model, and adjust the quantization parameter and iterative regression model parameters so that the predicted coding rate meets an allowable range of a virtual buffer area and a bandwidth, so that the statistical multiplexer multiplexes code streams of multiple encoders into one transmission channel, and the second moment is later than the first moment; The online learning module is used to initialize the model parameters and determine the regression model according to the initialized model parameters.

[0012] A third aspect of the embodiments of the present application provides an electronic device, including: a processor and a memory; Among them, one or more computer programs are stored in the memory, and the one or more computer programs include instructions; when the instructions are executed by the processor, the electronic device executes the method described in the first aspect of the embodiment of the present application.

[0013] A fourth aspect of an embodiment of the present application provides a computer storage medium, including computer instructions. When the computer instructions are executed on an electronic device, the electronic device executes the method described in the first aspect of the embodiment of the present application.

[0014] By using the statistical multiplexing method provided in the embodiment of the present application, each encoder will generate a corresponding regression model for predicting the encoding bit rate of each program, so that the different characteristics between different encoders can be isolated, and there is no need to deeply bind the parameters of the encoders, thereby broadening the scope of application of the statistical multiplexing function. At the same time, the statistical multiplexing method of the present application can allocate encoding size and quantization parameters for each program in units of frames, which is more real-time. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings: Figure 1 This is a flow chart of a statistical multiplexing method described in Example 1 of the present application; Figure 2 This is a key node time diagram of the encoder and statistical multiplexer described in Example 1 of the present application; Figure 3 The remaining space of the virtual buffer area VBV rslt Greater than the maximum limit space VBV max Schematic diagram of Figure 4 The remaining space of the virtual buffer area VBV rsltLess than the minimum limit space VBV min Schematic diagram of Figure 5 The remaining space of the virtual buffer area VBV rslt In the maximum restricted space VBV max and minimum confined space VBV min Schematic diagram between; Figure 6 This is a flow chart of another statistical multiplexing method described in Example 1 of the present application; Figure 7 This is a schematic diagram of the principle of a statistical multiplexing system described in Example 2 of the present application. DETAILED DESCRIPTION

[0016] In order to make the technical solutions and advantages in the embodiments of the present application more clearly understood, the exemplary embodiments of the present application are further described in detail below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than an exhaustive list of all the embodiments. It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict. Example 1

[0017] like Figure 1 As shown, this embodiment provides a statistical multiplexing method, which includes: S101, initializing model parameters, and determining a regression model according to the initialized model parameters.

[0018] Specifically, the statistical multiplexing method proposed in this embodiment is implemented based on an online learning network. Before statistical multiplexing, it is necessary to determine the online learning model of the online learning network. This embodiment uses a regression model to predict the encoding bit rate.

[0019] First, the parameters of the regression model are initialized. The process of initializing the model parameters depends on the results of the historical data training of statistical multiplexing. Before training, it should be determined what kind of feature data can be provided by encoders of different protocols for predicting the coding size in the process of applying statistical multiplexing, and these data are used as feature vectors of the input model. In this embodiment, the feature data at least includes the complexity of a single frame, such as SAD (Sum of Absolute Difference), SATD (Sum of Absolute Transformed Difference after Hadamard Transformation), etc. In addition, variables such as coding protocol, coded frame type, prediction mode, etc. that are correlated with the coding compression result may also be required.

[0020] S102: The encoder pre-encodes the frame to be encoded starting from the first moment to generate pre-encoding characteristic data, and sends the pre-encoding characteristic data to a statistical multiplexer.

[0021] Specifically, the statistical multiplexing method of this embodiment can support encoding protocols including AVS series and H26X series. Figure 2 The diagram shows the key node timings of the encoder and statistical multiplexer. Figure 2 In the example, when the encoder sends C_lookahead pre-coding feature data, the time corresponding to the encoder pre-coding thread is T_cur. The encoder pre-coding thread pre-codes the frames to be encoded from the first time T_cur to generate pre-coding feature data, and sends it to the statistical multiplexer. The statistical multiplexer starts processing the frame at T_0. The time difference from T_cur to T_0 is C_lookahead / FPS, and FPS is the number of pre-coded frames per second of the encoder.

[0022] S103 , the statistical multiplexer sequentially substitutes the pre-coding feature data within the time T starting from the second moment into the regression model in units of frames, and estimates the predicted coding bit rate within the time T by using the quantization parameter and the regression model.

[0023] Specifically, after receiving the pre-coding feature data, the statistical multiplexer will substitute the pre-coding feature data within the T time from the second moment T_0 into the regression model determined in S101 in units of frames. In this regression model, the quantization parameter Q is also required to participate in learning and training. The quantization parameter Q can control the bit rate of each frame of each program. The quantization parameter Q is inversely proportional to the bit rate, that is, as the quantization parameter Q increases, the bit rate becomes lower, and correspondingly, as the quantization parameter Q decreases, the bit rate becomes higher. Each program corresponds to its own quantization parameter. In order to ensure the quality of each encoding, the quantization parameter Q needs to be limited to the minimum quantization parameter Q during the statistical multiplexing process. min and the maximum quantization parameter Q max The predicted coding rate S within T time can be estimated by using the quantization parameters and regression model. pred :

[0024] Wherein, Param_lookahead is precoding characteristic data, and N is a positive integer.

[0025] In some optional embodiments, after the statistical multiplexer receives the pre-encoding characteristic data sent by the encoder, the statistical multiplexer stores the received pre-encoding characteristic data in the message queue. During C_lookahead, the statistical multiplexer sequentially substitutes the pre-coded feature data within the time T from the second time T_0 into the regression model in units of frames.

[0026] S104: Adjust the quantization parameter and the iterative regression model parameter so that the predicted coding bit rate meets the allowable range of the virtual buffer area and the bandwidth.

[0027] Specifically, the relationship between the quantization parameter Q and the bit rate has been explained in S103. If statistical multiplexing of multiple encoders on a statistical multiplexer is to be implemented, the bit rate needs to be controlled within the allowable range of the virtual buffer area and the bandwidth. To this end, the quantization parameter needs to be adjusted so that the predictive coding bit rate meets the allowable range of the virtual buffer area VBV and the bandwidth BW.

[0028] When the statistical multiplexer performs statistical multiplexing of multiple encoders, the virtual buffer is used to record the remaining space VBV of the virtual buffer in the current state. rslt In this virtual buffer area, a maximum limit space VBV is set max and minimum confined space VBV min Ideally, the remaining space in the virtual buffer area is VBV. rslt Should be in the maximum restricted space VBV max and minimum confined space VBV min However, in actual applications, there are often three situations, namely, the remaining space of the virtual buffer area VBV rslt In the maximum restricted space VBV max and minimum confined space VBV min Between, the remaining space of the virtual buffer area VBV rslt Greater than the maximum limit space VBV max , Virtual buffer area remaining space VBV rslt Less than the minimum limit space VBV min .

[0029] In this embodiment, the calculation process of the remaining space of the virtual buffer area is as follows: First, the current space VBV of the virtual buffer area is obtained. cur , and then calculate the total bandwidth BW within time T T, and finally the current space of the virtual buffer area VBV cur and total bandwidth BW After summing T, it is combined with the predicted coding rate S pred After the difference calculation, the remaining space VBV of the virtual buffer area after T time can be obtained. rslt .

[0030] For the virtual buffer area remaining space VBV rslt Greater than the maximum limit space VBV maxThe situation can be expressed as:

[0031] like Figure 3 As shown, according to the current encoding conditions, the remaining space in the virtual buffer area (the rightmost cylinder) exceeds the maximum limit space VBV after T time max , the bandwidth will be wasted, so the output bit rate of the encoder needs to be increased, then correspondingly, the current quantization parameter Q needs to be reduced cur , then the updated quantization parameter Q new for:

[0032] Make the encoding size of each frame larger under the premise of keeping other parameters unchanged, factor represents the factor variable, and after multiple iterations, the remaining space VBV in the virtual buffer area after T time encoding rslt satisfy:

[0033] For the virtual buffer area remaining space VBV rslt Less than the minimum limit space VBV min The situation can be expressed as:

[0034] like Figure 4 As shown, this situation is similar to Figure 3 On the contrary, according to the current encoding conditions, the virtual buffer area will be over-consumed after T time. In extreme cases, the virtual buffer area space will be emptied, resulting in excessive transmission delay. The output bit rate of the encoder needs to be reduced. Accordingly, the current quantization parameter Q needs to be increased. cur , then the updated quantization parameter Q new for:

[0035] Make the virtual buffer area have free space VBV rslt satisfy: .

[0036] For the virtual buffer area remaining space VBV rslt In the maximum restricted space VBV max and minimum confined space VBV min The situation between can be expressed as:

[0037] like Figure 5 As shown, when the remaining space in the virtual buffer area VBV rslt satisfy When the current quantization parameter Qcur Have the encoder encode the frame.

[0038] In some optional embodiments, in order to reduce the amount of calculation and possible delay in the statistical multiplexing process, the bit rate control of each program in this embodiment is performed in units of mini_gop, that is, every time an I frame (intra frame) or P frame (predictive frame) appears, the remaining space VBV in the virtual buffer is adjusted. rslt Calculation.

[0039] After the above process, the ES stream bit rates output by multiple encoders in a statistical multiplexing group have been controlled within the overall bandwidth operating range, and the TS stream output bit streams of multiple encoders can be multiplexed and transmitted through a statistical multiplexer.

[0040] In this embodiment, each encoder will generate a corresponding regression model to predict the encoding bit rate of each program, so that different characteristics between different encoders can be isolated, and there is no need to deeply bind the parameters of the encoders, thereby broadening the scope of application of the statistical multiplexing function. At the same time, the statistical multiplexing method of the present application can allocate encoding size and quantization parameters for each program in units of frames, which is more real-time.

[0041] In some optional embodiments, Figure 6 As shown, the statistical multiplexing method also includes: S105. The statistical multiplexer sends the current quantization parameter to the encoder.

[0042] Specifically, after obtaining the required encoding bit rate, the statistical multiplexer feeds back the quantization parameter Q of the current frame at time T_0 to the encoder. Figure 2 Due to the uncertain transmission delay Δ, the encoder will always receive the feedback result of the frame at the time T_feedback-Δ, and encode the frame image when receiving the QP.

[0043] S106: The encoder encodes the frame to be encoded at the third moment according to the pre-encoding characteristic data and the current quantization parameter to obtain a real encoding code rate of the frame to be encoded at the third moment and sends the real encoding code rate to the statistical multiplexer.

[0044] Specifically, the encoder performs a quantization operation on the third moment T_enc (eg, Figure 2 The frame image is encoded and the actual encoding size S of the frame is enc Feedback to the statistical multiplexer.

[0045] S107. The statistical multiplexer obtains a loss function of a regression model according to the predicted encoding rate and the actual encoding rate.

[0046] Specifically, after the encoder sends the actual encoding rate to the statistical multiplexer, the statistical multiplexer calculates the predicted encoding rate S pred and the actual encoding rate S enc Calculate the loss function LOSS of the regression model:

[0047] S108. The statistical multiplexer updates the regression model parameters according to the loss function.

[0048] Specifically, when the loss function LOSS in S107 exceeds the threshold, it means that the current model parameters cannot accurately estimate the coding size, and the new coding data needs to be substituted into the regression model, so that the regression model adjusts the parameters to make the absolute value of the difference between the predicted coding rate and the actual coding rate as small as possible. In the selection of the model, a regression model is selected. If the parameters fed back by the encoder include various information such as the complexity information of the frame, the frame type, and the prediction mode, a polynomial regression equation can also be constructed, which has a better fitting effect under the premise that the exponent of the independent variable is not one. However, the relationship between the complexity of the frame, the quantization parameter, and the coding size of the frame is not an exponential relationship. Of course, in the process of regression model learning, the model parameters can be updated in units of one frame, or they can be updated and iterated through a set of data based on mini_gop or gop, which is similar to the difference between the Stochastic randomness mode and the mini-batch mode in the gradient descent process. Example 2

[0049] Corresponding to Example 1, Figure 7 As shown, this embodiment provides a statistical multiplexing system, which includes: At least one encoder, configured to pre-encode a frame to be encoded starting from a first moment to generate pre-encoding characteristic data, and send the pre-encoding characteristic data to a statistical multiplexer; A statistical multiplexer, used to sequentially substitute pre-encoded feature data within a time T from a second moment into a regression model in units of frames, estimate a predicted coding rate within the time T by using a quantization parameter and the regression model, and adjust the quantization parameter and iterative regression model parameters so that the predicted coding rate meets an allowable range of a virtual buffer area and a bandwidth, so that the statistical multiplexer multiplexes code streams of multiple encoders into one transmission channel, and the second moment is later than the first moment; The online learning module is used to initialize the model parameters and determine the regression model according to the initialized model parameters.

[0050] Specifically, this embodiment mainly involves three specific objects, a statistical multiplexer, an encoder, and an online learning module, wherein a regression model is embedded in the online learning module. In a statistical multiplexing group of multiple encoders, if there are n encoders, n regression models will be generated to predict the encoding bit rate of each program, thereby isolating the different features between different encoders. Therefore, even if encoders with different encoding protocols are used, such as H264, AVS series, and MPEG-2, bit rate statistics and bit stream multiplexing can be achieved in the same statistical multiplexer. Example 3

[0051] This embodiment provides an electronic device, including: a processor and a memory; Wherein, one or more computer programs are stored in the memory, and the one or more computer programs include instructions; when the instructions are executed by the processor, the electronic device performs the following method: Initializing model parameters, and determining a regression model according to the initialized model parameters; The encoder pre-encodes the frame to be encoded from the first moment to generate pre-encoding characteristic data, and sends the pre-encoding characteristic data to the statistical multiplexer; The statistical multiplexer sequentially substitutes the pre-coding feature data within a time T from a second moment into the regression model in units of frames, estimates the predicted coding bit rate within a time T by using quantization parameters and the regression model, and adjusts the quantization parameters and iterative regression model parameters so that the predicted coding bit rate meets the allowable range of the virtual buffer area and the bandwidth, so that the statistical multiplexer multiplexes the bit streams of multiple encoders into a transmission channel, and the second moment is later than the first moment.

[0052] The specific process of the method can refer to the contents described in Example 1, and will not be described in detail in this example.

[0053] In this embodiment, each encoder will generate a corresponding regression model to predict the encoding bit rate of each program, so that the different characteristics between different encoders can be isolated, and there is no need to deeply bind the parameters of the encoders, thereby broadening the scope of application of the statistical multiplexing function. At the same time, the statistical multiplexing method of the present application can allocate encoding size and quantization parameters for each program in units of frames, which is more real-time. Example 4

[0054] A computer storage medium includes computer instructions. When the computer instructions are executed on an electronic device, the electronic device executes the following method: Initializing model parameters, and determining a regression model according to the initialized model parameters; The encoder pre-encodes the frame to be encoded from the first moment to generate pre-encoding characteristic data, and sends the pre-encoding characteristic data to the statistical multiplexer; The statistical multiplexer sequentially substitutes the pre-coding feature data within a time T from a second moment into the regression model in units of frames, estimates the predicted coding bit rate within a time T by using quantization parameters and the regression model, and adjusts the quantization parameters and iterative regression model parameters so that the predicted coding bit rate meets the allowable range of the virtual buffer area and the bandwidth, so that the statistical multiplexer multiplexes the bit streams of multiple encoders into a transmission channel, and the second moment is later than the first moment.

[0055] The specific process of the method can refer to the contents described in Example 1, and will not be described in detail in this example.

[0056] In this embodiment, each encoder will generate a corresponding regression model to predict the encoding bit rate of each program, so that the different characteristics between different encoders can be isolated, and there is no need to deeply bind the parameters of the encoders, thereby broadening the scope of application of the statistical multiplexing function. At the same time, the statistical multiplexing method of the present application can allocate encoding size and quantization parameters for each program in units of frames, which is more real-time.

[0057] In this application, unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection, an electrical connection, or can communicate with each other; it can be a direct connection, or an indirect connection through an intermediate medium, it can be the internal connection of two elements or the interaction relationship between two elements. For ordinary technicians in this field, the specific meanings of the above terms in this application can be understood according to specific circumstances.

[0058] Although the preferred embodiments of the present application have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0059] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A statistical multiplexing method, characterized in that: The method comprises: Initializing model parameters, and determining a regression model according to the initialized model parameters; The encoder pre-encodes the frame to be encoded from the first moment to generate pre-encoding characteristic data, and sends the pre-encoding characteristic data to the statistical multiplexer; The statistical multiplexer sequentially substitutes the pre-coding feature data within a time T from the second moment into the regression model in units of frames, estimates the predicted coding rate within the time T through the quantization parameter and the regression model, and adjusts the quantization parameter and the iterative regression model parameter to make the predicted coding rate meet the allowable range of the virtual buffer area and the bandwidth, so that the statistical multiplexer multiplexes the code streams of multiple encoders into a transmission channel, and the second moment is later than the first moment.

2. The method according to claim 1, characterized in that The process of initializing the model parameters and determining the regression model according to the initialized model parameters includes: Determine the characteristic data used by encoders of different protocols to predict the encoding size; Training the statistically multiplexed historical data according to the feature data; Model parameters are initialized according to the historical data training results of statistical multiplexing, and a regression model is determined according to the initialized model parameters.

3. The method according to claim 1, characterized in that Before the statistical multiplexer sequentially substitutes the pre-coded feature data within the time T from the second moment into the regression model in units of frames, the method further includes: The statistical multiplexer stores the pre-coded characteristic data in a message queue; When the message queue length reaches a preset value, the statistical multiplexer sequentially substitutes the pre-coded feature data within a time period T from the second moment into the regression model in units of frames.

4. The method according to claim 1, characterized in that: The process of adjusting the quantization parameter and the iterative regression model parameter so that the predicted coding rate meets the allowable range of the virtual buffer area and the bandwidth includes: Calculate the remaining space in the virtual buffer after T time; When the remaining space in the virtual buffer exceeds the maximum limit after T time, the quantization parameter is reduced; When the remaining space in the virtual buffer is less than the minimum limit space after T time, the quantization parameter is increased; When the remaining space of the virtual buffer area is between the maximum limit space and the minimum limit space after T time, the current quantization parameter is maintained.

5. The method according to claim 4, characterized in that The process of calculating the remaining space of the virtual buffer area after T time includes: Get the current space of the virtual buffer; Calculate the total bandwidth within time T; After the maximum virtual buffer space and the total bandwidth are summed, a difference calculation is performed between the sum and the predicted coding bit rate to obtain the remaining virtual buffer space after the T time.

6. The method according to claim 4, characterized in that When an I frame or a P frame appears at any time, the remaining space in the virtual buffer after T time from the current time is calculated.

7. The method according to claim 1, characterized in that The method further comprises: The statistical multiplexer sends the current quantization parameter to the encoder; The encoder encodes the frame to be encoded at a third moment according to the pre-encoding characteristic data and the current quantization parameter to obtain a real encoding rate of the frame to be encoded at the third moment, and sends the real encoding rate to the statistical multiplexer, wherein the third moment is later than the second moment; The statistical multiplexer obtains a loss function of the regression model according to the predicted coding rate and the actual coding rate, and updates the regression model parameters according to the loss function.

8. A statistical multiplexing system, characterized in that: The system comprises: At least one encoder, configured to pre-encode a frame to be encoded starting from a first moment to generate pre-encoding characteristic data, and send the pre-encoding characteristic data to a statistical multiplexer; A statistical multiplexer, used to sequentially substitute pre-encoded feature data within a time T from a second moment into a regression model in units of frames, estimate a predicted coding rate within the time T by using a quantization parameter and the regression model, and adjust the quantization parameter and iterative regression model parameters so that the predicted coding rate meets an allowable range of a virtual buffer area and a bandwidth, so that the statistical multiplexer multiplexes code streams of multiple encoders into one transmission channel, and the second moment is later than the first moment; The online learning module is used to initialize the model parameters and determine the regression model according to the initialized model parameters.

9. An electronic device, characterized in that: include: Processor and memory; One or more computer programs are stored in the memory, and the one or more computer programs include instructions; when the instructions are executed by the processor, the electronic device executes the method as claimed in any one of claims 1 to 7.

10. A computer storage medium, characterized in that: The method comprises computer instructions. When the computer instructions are executed on an electronic device, the electronic device executes the method according to any one of claims 1 to 7.

Images