Network packet loss simulation method, related method, device, equipment and storage medium

Abstract

The invention discloses a network packet loss simulation method, a related method, a device, equipment and a storage medium, and the method comprises the steps: obtaining a packet loss state sequence and a target packet loss rate; wherein the packet loss state sequence comprises packet loss labels of the current time step and historical time steps before the current time step, and the packet loss labels represent whether packet loss is transmitted or not; in the process that a simulation platform used for simulating network transmission carries out transmission simulation according to the packet loss state sequence, a multi-dimensional network state obtained by monitoring the simulation platform is obtained; performing prediction based on the packet loss state sequence, the target packet loss rate and the multi-dimensional network state to obtain a packet loss label of the next time step of the current time step; and updating the packet loss state sequence based on the packet loss label of the next time step. According to the scheme, the cost and configuration complexity of packet loss simulation can be reduced on the premise of ensuring that the packet loss simulation can truly reproduce a complex phenomenon of a network environment as much as possible.

Description

Technical Field

[0001] This application relates to the field of communication network technology, and in particular to a network packet loss simulation method and related methods, devices, equipment and storage media. Background Technology

[0002] Currently, real-time communication technologies, primarily based on audio and video, are widely used in various scenarios, such as online office work, remote conferencing, telemedicine, online teaching, e-commerce live streaming, cloud gaming, and other digital applications.

[0003] In real-world network transmission, packet loss due to numerous possible factors remains a significant obstacle to improving service quality and user experience in digital scenarios. To combat packet loss in real-world network environments, many algorithmic models for network quality optimization have been developed. Therefore, before formal deployment, it is crucial to realistically simulate the network environment to optimize these algorithmic models. However, current packet loss simulation technologies are either implemented using software tools or hardware devices. The former typically uses random packet loss or rule-based loss strategies, failing to accurately reproduce the complexities of the network environment, while the latter suffers from high costs and complex configurations. Therefore, reducing the cost and configuration complexity of packet loss simulation while ensuring its accurate representation of the complexities of the network environment has become an urgent problem to solve. Summary of the Invention

[0004] The main technical problem addressed by this application is to provide a network packet loss simulation method and related methods, devices, equipment, and storage media, which can reduce the cost and configuration complexity of packet loss simulation while ensuring that the packet loss simulation can realistically reproduce the complex phenomena of the network environment as much as possible.

[0005] To address the aforementioned technical problems, the first aspect of this application provides a network packet loss simulation method, comprising: acquiring a packet loss state sequence and a target packet loss rate; wherein the packet loss state sequence includes packet loss labels for the current time step and all previous historical time steps, the packet loss labels indicating whether packet loss has occurred; during the transmission simulation process on a simulation platform used to simulate network transmission according to the packet loss state sequence, acquiring a multi-dimensional network state monitored by the simulation platform; predicting the packet loss label for the next time step based on the packet loss state sequence, the target packet loss rate, and the multi-dimensional network state; and updating the packet loss state sequence based on the packet loss label for the next time step.

[0006] To address the aforementioned technical problems, a second aspect of this application provides a method for tuning a network quality optimization model, comprising: acquiring a packet loss state sequence; wherein the packet loss state sequence is obtained through the network packet loss simulation method described in the first aspect; controlling a simulation platform for simulating network transmission to perform transmission simulation according to the packet loss state sequence; wherein the network quality optimization model is used for at least one of the sender and receiver of the simulation platform; and tuning the network quality optimization model based on the transmission records after the simulation platform completes the transmission simulation.

[0007] To address the aforementioned technical problems, a third aspect of this application provides a network packet loss simulation device, comprising: an acquisition module, a monitoring module, a prediction module, and an update module. The acquisition module is used to acquire a packet loss state sequence and a target packet loss rate; wherein, the packet loss state sequence includes packet loss labels for the current time step and each previous historical time step, and the packet loss labels characterize whether packet loss has occurred. The monitoring module is used to acquire a multi-dimensional network state monitored by the simulation platform during the transmission simulation process of the simulation platform used to simulate network transmission according to the packet loss state sequence. The prediction module is used to predict the packet loss label for the next time step after the current time step based on the packet loss state sequence, the target packet loss rate, and the multi-dimensional network state. The update module is used to update the packet loss state sequence based on the packet loss label of the next time step.

[0008] To address the aforementioned technical problems, a fourth aspect of this application provides a tuning device for a network quality optimization model, comprising: an acquisition module, a control module, and a tuning module. The acquisition module is used to acquire a packet loss state sequence, wherein the packet loss state sequence is obtained through the network packet loss simulation device described in the third aspect above. The control module is used to control a simulation platform for simulating network transmission to perform transmission simulation according to the packet loss state sequence, wherein the network quality optimization model is used for at least one of the sender and receiver of the simulation platform. The tuning module is used to tune the network quality optimization model based on the transmission records after the transmission simulation is completed by the simulation platform.

[0009] To address the aforementioned technical problems, the fifth aspect of this application provides an electronic device comprising at least a memory and a processor coupled to each other. The memory stores at least program instructions, and the processor executes the program instructions to implement the network packet loss simulation method in the first aspect or the network quality optimization model tuning method in the second aspect.

[0010] To address the aforementioned technical problems, the fourth aspect of this application provides a computer-readable storage medium storing program instructions executable by a processor. These program instructions are used to implement the network packet loss simulation method of the first aspect or the network quality optimization model tuning method of the second aspect.

[0011] The above scheme obtains a packet loss state sequence and a target packet loss rate. The packet loss state sequence includes packet loss labels for the current time step and all previous historical time steps. Each packet loss label indicates whether packet loss occurred. Then, during network transmission simulation on a simulation platform based on the packet loss state sequence, the multi-dimensional network state monitored by the simulation platform is acquired. Based on the packet loss state sequence, the target packet loss rate, and the multi-dimensional network state, predictions are made to obtain the packet loss label for the next time step. Finally, based on the packet loss label for the next time step, the packet loss state sequence is updated. This approach, on the one hand, ensures that the packet loss is measured according to the packet loss state sequence on the simulation platform. During packet state sequence transmission simulation, the simulation platform is monitored to obtain multi-dimensional network states. These multi-dimensional network states can reflect the network environment as realistically as possible. Furthermore, when predicting the packet loss label for the next time step, the packet loss state sequence, target packet loss rate, and multi-dimensional network states are combined for prediction. Therefore, this helps ensure that the packet loss simulation can realistically reproduce the complex phenomena of the network environment. On the other hand, since only the target packet loss rate and the packet loss labels from historical time steps are needed to start the packet loss simulation through the simulation platform, without complex hardware configuration, it helps reduce the cost and configuration complexity of packet loss simulation. Therefore, it is possible to reduce the cost and configuration complexity of packet loss simulation while ensuring that it can realistically reproduce the complex phenomena of the network environment as much as possible. Attached Figure Description

[0012] Figure 1 This is a flowchart illustrating an embodiment of the network packet loss simulation method of this application; Figure 2 This is a flowchart illustrating an embodiment of the tuning method for the network quality optimization model of this application; Figure 3 This is a schematic diagram of the framework of an embodiment of the network packet loss simulation device of this application; Figure 4 This is a schematic diagram of the framework of an embodiment of the tuning device for the network quality optimization model of this application; Figure 5 This is a schematic diagram of the framework of an embodiment of the electronic device of this application; Figure 6 This is a schematic diagram of a framework of an embodiment of the computer-readable storage medium of this application. Detailed Implementation

[0013] The embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0014] In the following description, specific details such as particular system architectures, interfaces, and technologies are presented for illustrative purposes rather than for limiting purposes, in order to provide a thorough understanding of this application.

[0015] In this paper, the terms "system" and "network" are often used interchangeably. The term "and / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. Additionally, the slash " / " generally indicates that the preceding and following related objects have an "or" relationship. Furthermore, "many" in this paper indicates two or more objects.

[0016] Please see Figure 1 , Figure 1 This is a flowchart illustrating an embodiment of the network packet loss simulation method of this application. Specifically, it may include the following steps: Step S11: Obtain the packet loss status sequence and target packet loss rate.

[0017] In this embodiment, the packet loss state sequence includes packet loss tags for the current time step and all previous historical time steps, with each tag indicating whether packet loss occurred. For example, if the current time step is denoted as t, the previous historical time steps can be denoted as 0, 1, 2, ..., t-1. Furthermore, packet loss tags can be represented by numbers, letters, or other characters. For example, when using numbers to represent packet loss tags, 0 can represent packet loss and 1 can represent no packet loss, or vice versa, 0 can represent no packet loss and 0 can represent packet loss; no limitation is made here. Of course, regardless of the character used to represent packet loss or no packet loss, it is preferable to keep the corresponding character representing packet loss or no packet loss unchanged during packet loss simulation.

[0018] In one implementation scenario, at the first time step of packet loss simulation, since there are no previous historical time steps, a packet loss label can be set for the first time step to successfully start the simulation. This allows the packet loss simulation to begin and the packet loss label for the next time step (i.e., the second time step) to be predicted. For example, the packet loss label for the first time step can be set by default to represent packet loss where no packets were transmitted. For instance, if the number 1 represents no packet loss, the packet loss label for the first time step can be set to 1 by default. Of course, the above example is only one possible way to set the packet loss label for the first time step; other possible settings are not limited here.

[0019] In one implementation scenario, the target packet loss rate can be characterized as the expected packet loss rate presented in the final packet loss state sequence generated by the packet loss simulation. For example, the target packet loss rate can be set according to application requirements, such as 5%, 10%, etc. The specific value of the target packet loss rate is not limited here, nor will it be listed in detail. For example, the target packet loss rate can be input by the user of the network packet loss simulation each time they need to start the simulation to generate the packet loss state sequence.

[0020] Step S12: During the transmission simulation process of the simulation platform used to simulate network transmission according to the packet loss state sequence, obtain the multi-dimensional network state monitored by the simulation platform.

[0021] In one implementation scenario, as a possible approach, the simulation platform can be built using software. For example, software can simulate the sender and receiver, and the transmission medium between them can be customized (e.g., wired media such as fiber optic cables or network cables, or wireless bands such as 2.4G or 5G, or a combination of both). Alternatively, as another possible implementation, the simulation platform can be built using physical devices. For instance, two terminal devices (e.g., smartphones, tablets, or other consumer electronics) can be selected, one as the sender and the other as the receiver, communicating via wired, wireless, or a combination of both. Of course, the above examples are merely a few possible implementation methods in practical applications, and the method of building the simulation platform is not limited here. For ease of simulation control, software-based construction is preferable for the simulation platform.

[0022] In one implementation scenario, after obtaining the packet loss state sequence through the aforementioned steps, the simulation platform can then perform transmission simulation according to this sequence. For example, still using the digit 1 to represent no packet loss and digit 0 to represent packet loss, the packet loss state sequence {1,1,1,0,1} represents controlling the simulation platform to transmit five data packets sequentially, with the fourth data packet resulting in packet loss, and the remaining data packets not experiencing packet loss. The simulation platform can then be controlled to perform transmission simulation according to this packet loss state sequence. Specifically, in the first time step, no packet loss occurred; in the second time step, no packet loss occurred; in the third time step, no packet loss occurred; in the fourth time step, no packet loss occurred; and in the fifth time step, no packet loss occurred. Of course, the above example is only one possible case in practical applications, and other possible scenarios will not be listed here.

[0023] In one implementation scenario, multidimensional network states can include any two or more of the following: network transmission delay, latency jitter, network congestion metrics, bandwidth utilization, RTT (Round-Trip Time) change rate, historical burst length, queue congestion level, and retransmission frequency. Of course, the above examples are merely a few possible illustrations of network states in practical applications; other possible network states will not be listed here. Furthermore, the specific meanings of each of the above-mentioned network states can be found in the technical details of communication networks, and will not be elaborated upon here.

[0024] In a specific implementation scenario, network transmission latency can be represented by a value between 0 and 1. For example, the network transmission latency that may occur in real-world applications, ranging from 0 to 1000 milliseconds, can be normalized. For ease of description, the network transmission latency can be denoted as f1.

[0025] In a specific implementation scenario, latency jitter can be represented by a value between 0 and 1. For example, the latency jitter of 0 to 100 milliseconds that may occur in a real-world application can be normalized. For ease of description, the latency jitter can be denoted as f2.

[0026] In a specific implementation scenario, network congestion metrics can be represented by values ​​between 0 and 1. For example, network congestion metrics can be determined based on a congestion window. For ease of description, the network congestion metric can be labeled f3.

[0027] In a specific implementation scenario, bandwidth utilization can be represented by a value between 0 and 1. For example, bandwidth utilization can be determined based on real-time bandwidth usage. For ease of description, bandwidth utilization can be denoted as f4.

[0028] In a specific implementation scenario, the RTT change rate can be represented by a value between 0 and 1. For example, the RTT change rate can be determined based on the degree of RTT fluctuation. For ease of description, the RTT change rate can be denoted as f5.

[0029] In a specific implementation scenario, the historical burst length can be represented by a value between 0 and 1. For example, the historical burst length can be determined based on the normalized burst index. For ease of description, the historical burst length can be denoted as f6.

[0030] In a specific implementation scenario, queue congestion can be represented by a value between 0 and 1. For example, queue congestion can be determined based on router queue occupancy. For ease of description, queue congestion can be denoted as f7.

[0031] In a specific implementation scenario, the retransmission frequency index can be represented by a value between 0 and 1. For example, the retransmission frequency index can be obtained by statistically analyzing retransmission requests. For ease of description, the retransmission frequency index can be denoted as f8.

[0032] Step S13: Based on the packet loss state sequence, target packet loss rate, and multidimensional network state, make a prediction to obtain the packet loss label for the next time step after the current time step.

[0033] In one implementation scenario, as a possible approach, a prompt instruction can be constructed based on the packet loss state sequence, the target packet loss rate, and the multidimensional network state. This prompt instruction instructs the large language model to predict the packet loss label for the next time step by combining the packet loss state sequence, the target packet loss rate, and the multidimensional network state. The output of the large language model in response to the prompt instruction can then be obtained as the packet loss label for the next time step. It should be noted that, to facilitate the large language model's understanding, the specific meanings of the packet loss state sequence, the target packet loss rate, and the multidimensional network state can be defined within the prompt instruction. Furthermore, the large language model can include, but is not limited to, open-source large models such as Llama and Bloom. Alternatively, the large language model can be obtained by fine-tuning an open-source large model based on a specific corpus (e.g., training data related to data transmission in the field of communication networks). Or, the large language model can be a custom large model; the specific source of the large language model is not limited here.

[0034] In another implementation scenario, as a possible approach, feature extraction can be performed based on the multidimensional network state to obtain network features. Temporal modeling is then performed based on the concatenated sequence of packet loss state sequences, target packet loss rates, and network features to obtain a first feature. This first feature can then be processed by multi-head attention, configured with different attention scopes for each attention head, to obtain a second feature. Finally, label mapping can be performed based on the fused features of the first and second features to obtain the packet loss label for the next time step. This method, through a series of operations including feature extraction, temporal modeling, multi-head attention, and label mapping, ultimately yields the packet loss label for the next time step. It fully utilizes the multidimensional network state and models temporal dependencies. Furthermore, by configuring multi-head attention with different attention scopes, it can fully consider changes in the network state, creating a synergistic effect with the multidimensional network state and contributing to improved accuracy in predicting the packet loss label for the next time step.

[0035] In a specific implementation scenario, to extract network features based on multidimensional network states, feature embedding can be performed separately for each dimension of the network state to obtain embedded features for each dimension. For ease of description, we will still use the aforementioned eight-dimensional network state as an example, which can be denoted as R. 8Then, it can obtain embedded features through several feature embeddings. For example, the eight-dimensional network state R can be first... 8 The first feature embedding is performed, and the above sampling yields the embedded features R. 16 Then embed the feature R 16 A second feature embedding is performed, and the embedded features R are obtained by sampling. 8 For ease of description, the specific process of embedding the above features can be represented as follows:

[0036] In the above formula, F(t) represents the embedding features of each dimension. Based on this, adaptive weight prediction can be performed based on the embedding features of each dimension to obtain the weight factors of each dimension. For example, the embedding features of each dimension can be predicted through fully connected layers or other network layers to obtain the weight factors of each dimension. For ease of description, the weight factors of each dimension can be represented as a(t). Finally, the network features can be obtained by weighting the weight factors and embedding features of each dimension. For ease of description, the network features F'(t) can be represented as:

[0037] In the above formula, This indicates that the matrix elements are multiplied accordingly. The above method embeds features into the network state of each dimension in the multidimensional network state, obtaining embedded features for each dimension. Adaptive weight prediction is then performed based on these embedded features to obtain weight factors for each dimension. Finally, the embedded features and weight factors for each dimension are weighted to obtain the network features. This method can express more complex feature information through embedded features, and the adaptive weighting can dynamically calculate the importance of each dimension, highlighting the relatively important dimension features in the packet loss simulation and suppressing the relatively minor dimension features.

[0038] In a specific implementation scenario, after obtaining the network features, temporal modeling can be performed by combining the packet loss state sequence and the target packet loss rate to obtain the first feature. Specifically, preprocessing can be performed based on the concatenated sequence of the packet loss state sequence, the target packet loss rate, and the network features to obtain sequence features. It should be noted that preprocessing can at least include feature mapping (e.g., upsampling, downsampling, etc.). On this basis, the sequence features can be processed based on the first GRU (Gated Recurrent Unit) layer to capture feature representations of short-term dependencies, then based on the second GRU layer to process the feature representations of short-term dependencies to model feature representations of medium-term patterns, and finally based on the third GRU layer to process the feature representations of medium-term patterns to extract feature representations of long-term features as the first feature. The above method, through feature processing using three GRU layers, can capture multi-scale temporal dependencies.

[0039] In a specific implementation scenario, after obtaining the first feature, multi-head attention can be used to process it. Specifically, the first feature can be processed by a first attention head whose scope of interest is the target time step interval, resulting in the first attention feature. For example, the target time step interval can be set according to the local range that needs attention in the actual application, such as 2 to 3 time steps before and after. This allows for the capture of short-term network state changes, such as the relationship between a packet loss and the previous 1 to 2 packets. Furthermore, the first feature can be processed by a second attention head whose scope of interest is the target time step interval, resulting in the second attention feature. For example, the target time step interval can be set according to the attention period needed in the actual application, such as every 10 to 20 time steps. This allows for the discovery of periodic congestion patterns in the network, such as periodic bursts every 20 milliseconds. Additionally, the first feature can be processed by a third attention head whose scope of interest is the target packet loss sequence interval, resulting in the third attention feature. It should be noted that the target packet loss sequence interval is a local interval in the packet loss state sequence where the packet loss rate is higher than a packet loss rate threshold. This helps in identifying the start and end of sudden packet loss events. For example, boundary detection of consecutive packet loss intervals. Furthermore, the first feature can be processed based on a fourth attention head whose scope is the entire input feature sequence, resulting in a fourth attention feature. For example, the entire input feature sequence is the entire sequence of the first feature. This helps in modeling long-term network state trends. For example, the long-term evolution of the network from good to congested. Based on this, the first, second, third, and fourth attention features can be fused (e.g., spliced) to obtain a second feature. It should be noted that the above description only focuses on illustrating the differences between multi-head attention and conventional multi-head attention in this embodiment of the disclosure. For other parts, please refer to the technical details of multi-head attention, which will not be repeated here. The above method processes the first feature based on the first attention head, which focuses on the target time step interval, to obtain the first attention feature. It then processes the first feature based on the second attention head, which focuses on the target time step interval, to obtain the second attention feature. Finally, it processes the first feature based on the third attention head, which focuses on the target packet loss sequence interval, to obtain the third attention feature. Finally, it processes the first feature based on the fourth attention head, which focuses on the entire input feature sequence, to obtain the fourth attention feature. This method can focus on local correlations through the first attention head, periodic patterns through the second attention head, burst detection through the third attention head, and global dependencies through the fourth attention head. The second feature is obtained by fusing the first, second, third, and fourth attention features. This method can overcome the forgetting problem that may exist in the temporal modeling process as much as possible, establish long-distance temporal relationships, and help to better identify and predict the start, duration, and end of burst packet loss.

[0040] In a specific implementation scenario, after obtaining the second feature, a label mapping can be performed based on the fused feature of the first and second features to obtain the packet loss label for the next time step. For example, the first and second features can be fused using methods such as addition or weighting to obtain a fused feature. Based on this, the fused feature can continue to pass through network layers such as fully connected layers and softmax layers to obtain the packet loss probability for the next time step. Then, the packet loss label for the next time step can be determined based on the packet loss probability. For example, threshold judgment or random sampling can be used to decide whether packet loss has occurred, thus obtaining the packet loss label. Taking threshold judgment as an example, it can be determined whether the packet loss probability is higher (or lower) than a probability threshold. If so, the packet loss label for the next time step can be determined to represent packet loss during transmission (e.g., the packet loss label is represented by the number 0); otherwise, the packet loss label for the next time step can be determined to represent no packet loss during transmission (e.g., the packet loss label is represented by the number 1).

[0041] In one implementation scenario, the packet loss label for the next time step is predicted by a packet loss simulation model. To improve the performance of the packet loss simulation model, it can be trained before prediction. Specifically, monitoring can be performed during real network transmission to obtain the actual packet loss state sequence and the actual multidimensional network state, where the actual packet loss state sequence includes the packet loss labels at each time step in the real network transmission. Based on this, the packet loss rate of the actual packet loss state sequence can be obtained as the target packet loss rate of the sample. The packet loss label of the last time step in the actual packet loss state sequence can be extracted as the sample packet loss label. The sample packet loss label is then removed from the actual packet loss state sequence to obtain the sample packet loss state sequence. Thus, predictions can be made based on the sample packet loss state sequence, the target packet loss rate, and the actual multidimensional network state to obtain the predicted packet loss probability of the last time step. Furthermore, the network parameters of the packet loss simulation model can be adjusted based on the difference between the sample packet loss label and the predicted packet loss probability. In this way, the model can be trained using relevant data from the actual network transmission process, which helps to improve the packet loss simulation model to reproduce the complex phenomena of the network environment as realistically as possible during packet loss simulation.

[0042] In a specific implementation scenario, as mentioned earlier, the actual packet loss state sequence can include packet loss labels at each time step in the actual network transmission. For example, the actual packet loss state sequence {1,0,1,0,1,1} indicates that packet loss occurred at the second and fourth time steps, and no packet loss occurred at the remaining four time steps. Furthermore, the specific meaning of the actual multidimensional network state can be found in the aforementioned description of multidimensional network states, and will not be repeated here.

[0043] In a specific implementation scenario, for ease of understanding, let's take the aforementioned real packet loss state sequence {1,0,1,0,1,1} as an example. We can obtain its packet loss rate, 33%, as the target packet loss rate for the sample. Furthermore, we can extract the packet loss label at the last time step of the real packet loss state sequence {1,0,1,0,1,1} as the sample packet loss label (i.e., "1"). Further, we can remove the aforementioned sample packet loss label from the real packet loss state sequence {1,0,1,0,1,1}, resulting in {1,0,1,0,1,}, as the sample packet loss state sequence. Of course, the above example is merely one possible instance in practical applications of obtaining the target packet loss rate, sample packet loss label, and sample packet loss state sequence from the real packet loss state sequence; other possible scenarios will not be illustrated here.

[0044] In a specific implementation scenario, after obtaining the sample packet loss state sequence, the target packet loss rate, and the actual multidimensional network state, the above data can be used as input data for the packet loss simulation model. The packet loss simulation model then makes predictions to obtain the predicted packet loss probability at the last time step. For example, the packet loss simulation model may include a feature extraction network (e.g., embedding layer, adaptive weight layer, etc.) for extracting network features; a temporal modeling network (e.g., GRU, etc.) for modeling the first feature; a multi-head attention network for acquiring the second feature; and a label mapping network (e.g., fully connected layer, softmax layer, etc.) for predicting the packet loss probability. The processing procedure of the packet loss simulation model can be found in the aforementioned descriptions and will not be repeated here.

[0045] In a specific implementation scenario, after obtaining the predicted packet loss probability at the last time step, the training loss can be calculated based on the cross-entropy loss function using the aforementioned sample packet loss labels, and the network parameters of the packet loss simulation model can be adjusted based on the training loss. For example, the training loss L can be expressed as:

[0046] In the above formula, N represents the batch sample size. This represents the packet loss label for the i-th sample. This represents the predicted packet loss probability for the i-th sample. It is a small positive number used to prevent the input to the log function from being zero or negative, helping to avoid numerical instability. Thus, by minimizing the training loss L, the packet loss simulation model is forced to better fit the data, making the output predicted probability approach the true packet loss label as quickly as possible.

[0047] Step S14: Update the packet loss state sequence based on the packet loss label of the next time step.

[0048] In one implementation scenario, after obtaining the packet loss label for the next time step, it can be appended to the end of the packet loss state sequence to update the sequence. Taking the aforementioned packet loss state sequence {1,1,1,0,1} as an example, if the predicted packet loss label for the next time step is 1, the packet loss state sequence can be updated to {1,1,1,0,1,1}. Of course, the above example is merely one possible illustration of updating the packet loss state sequence in practical applications; other possible scenarios will not be elaborated upon here.

[0049] In one implementation scenario, after updating the packet loss state sequence, the steps described above, which involve obtaining the multi-dimensional network state monitored by the simulation platform during the transmission simulation process according to the packet loss state sequence, can be returned to continuously update the packet loss state sequence until the latest packet loss state sequence meets the simulation requirements. In this way, a packet loss state sequence that meets the simulation requirements can be generated through continuous iteration.

[0050] In a specific implementation scenario, as a possible example, the simulation requirement can be set to include: the difference between the packet loss rate of the latest packet loss state sequence and the target packet loss rate is less than the allowable error. Taking the aforementioned packet loss state sequence {1,1,1,0,1} as an example, if the target packet loss rate is 15% and the allowable error is ±2%, then since the current packet loss rate of the packet loss state sequence is 20%, the difference between it and the target packet loss rate exceeds the allowable error. Therefore, further packet loss simulation is performed to update the packet loss state sequence to the aforementioned example {1,1,1,0,1,1}. Since the current packet loss rate of the packet loss state sequence is 16.7%, the difference between it and the target packet loss rate does not exceed the allowable error. Therefore, the latest packet loss state sequence {1,1,1,0,1,1} can be considered to meet the simulation requirement. Of course, the above example is only one possible example in practical applications; other possible scenarios will not be listed here.

[0051] In a specific implementation scenario, as another possible example, the simulation requirements can be set as follows: the sequence length of the latest packet loss state sequence reaches the target length, and the difference between the packet loss rate of the latest packet loss state sequence and the target packet loss rate is less than the allowable error. Taking the aforementioned packet loss state sequence {1,1,1,0,1} as an example, if the target packet loss rate is 15%, the allowable error is ±2%, and the target length is 10, then since the packet loss rate of the current packet loss state sequence is 20%, the difference between it and the target packet loss rate exceeds the allowable error. Therefore, further packet loss simulation is performed to update the packet loss state sequence to the aforementioned example {1,1,1,0,1,1}. Since the packet loss rate of this current packet loss state sequence is 16.7%, the difference between it and the target packet loss rate does not exceed the allowable error, but the sequence length is only 6, which has not yet reached the target length of 10. Therefore, it can be considered that the latest packet loss state sequence {1,1,1,0,1,1} still does not meet the simulation requirements, and further packet loss simulation is needed.

[0052] The above scheme obtains a packet loss state sequence and a target packet loss rate. The packet loss state sequence includes packet loss labels for the current time step and all previous historical time steps. Each packet loss label indicates whether packet loss occurred. Then, during network transmission simulation on a simulation platform based on the packet loss state sequence, the multi-dimensional network state monitored by the simulation platform is acquired. Based on the packet loss state sequence, the target packet loss rate, and the multi-dimensional network state, predictions are made to obtain the packet loss label for the next time step. Finally, based on the packet loss label for the next time step, the packet loss state sequence is updated. This approach, on the one hand, ensures that the packet loss is measured according to the packet loss state sequence on the simulation platform. During packet state sequence transmission simulation, the simulation platform is monitored to obtain multi-dimensional network states. These multi-dimensional network states can reflect the network environment as realistically as possible. Furthermore, when predicting the packet loss label for the next time step, the packet loss state sequence, target packet loss rate, and multi-dimensional network states are combined for prediction. Therefore, this helps ensure that the packet loss simulation can realistically reproduce the complex phenomena of the network environment. On the other hand, since only the target packet loss rate and the packet loss labels from historical time steps are needed to start the packet loss simulation through the simulation platform, without complex hardware configuration, it helps reduce the cost and configuration complexity of packet loss simulation. Therefore, it is possible to reduce the cost and configuration complexity of packet loss simulation while ensuring that it can realistically reproduce the complex phenomena of the network environment as much as possible.

[0053] Please see Figure 2 , Figure 2 This is a flowchart illustrating an embodiment of the network quality optimization model tuning method of this application. Specifically, it may include the following steps: Step S21: Obtain the packet loss status sequence.

[0054] In this embodiment, the packet loss state sequence is obtained by performing packet loss simulation through the process steps described in the above-described network packet loss simulation method embodiment. For details, please refer to the above-described packet loss simulation method embodiment, which will not be repeated here.

[0055] Step S22: Control the simulation platform used to simulate network transmission to perform transmission simulation according to the packet loss state sequence.

[0056] In this embodiment of the disclosure, the network quality optimization model is used for at least one of the sender and receiver in the simulation platform. Exemplarily, the network quality optimization model may include, but is not limited to, algorithms such as ARQ (Automatic Repeat-reQuest), FEC (Forward Error Correction), JitterBuffer (video network jitter buffer), NetEQ (Network Equalizer), PacedSender (smooth transmission controller), and GCC (Google Congestion Control), etc., which will not be listed here individually.

[0057] Step S23: Based on the transmission records after the transmission simulation on the simulation platform, optimize the network quality optimization model.

[0058] In one implementation scenario, transmission records may include, but are not limited to: whether the receiver recovers the lost data packets based on the network quality optimization model in the event of packet loss, and whether the sender retransmits the lost data packets in a timely manner based on the network quality optimization model, etc. The specific content of the transmission records is not limited here.

[0059] In one implementation scenario, as a possible example, after obtaining the transmission records, prompt instructions can be constructed based on the transmission records and the network quality optimization model. The prompt instructions are used to instruct the large language model to determine the items to be optimized and the recommended optimization methods for the items to be optimized in the network quality optimization model through the transmission records. Then, the network quality optimization model can be tuned based on the items to be optimized and their recommended optimization methods output by the large language model in response to the prompt instructions.

[0060] In another implementation scenario, as another possible example, after obtaining the transmission records, different network quality optimization models can be distinguished, and the transmission records can be used to optimize the network quality optimization models in different ways. For example, taking ARQ as the network quality optimization model, core monitoring indicators can be obtained from the transmission records, such as the raw packet loss rate (the proportion of packets lost when retransmission is not triggered), the retransmission success rate (the proportion of data packets successfully received after retransmission), the retransmission delay (the time from packet loss detection to the reception of retransmitted packets), and the retransmission number distribution (the proportion of a single data packet being retransmitted 1, 2, or more times). Based on this, the network quality optimization model can be tuned in terms of "dynamically adjusting the retransmission trigger threshold." For example, if the transmission record shows that "in low packet loss scenarios (e.g., packet loss rate <5%), the retransmission success rate is high, but the retransmission delay accounts for more than 20% of the total delay," the retransmission trigger threshold can be increased (e.g., only triggering retransmission for packets with two or more consecutive packet losses) to reduce unnecessary retransmissions and lower latency. Conversely, if "in high packet loss scenarios (e.g., packet loss rate >15%), the retransmission success rate is less than 60%," the retransmission trigger threshold can be lowered (e.g., triggering immediately upon detecting packet loss) and the maximum number of retransmissions can be increased (e.g., from 2 to 3) to prioritize recovery rate. Alternatively, the network quality optimization model can also be tuned in terms of "optimizing the retransmission strategy type." For example, if the record shows "rollback," the network quality optimization model can be tuned accordingly. "N-frame ARQ results in a large amount of non-lost data being retransmitted (the proportion of valid new data in retransmitted packets is <30%)": Switch to selective retransmission ARQ (only retransmit lost packets) to reduce bandwidth waste. Conversely, if "stop-and-wait ARQ results in low throughput (sending window utilization <0%)": Use continuous ARQ (such as Go-Back-N) to allow subsequent packets to be sent even without acknowledgment, improving transmission efficiency. Alternatively, the network quality optimization model can be tuned by "dynamically adjusting the timeout retransmission time (RTO) in conjunction with network latency." For example, if the log shows "RTO set too long (timeout waiting time exceeds twice the round-trip time (RTT), resulting in slow retransmission response": Dynamically compress RTO based on historical RTT fluctuations (such as the average of the last 100 RTTs + twice the standard deviation) to shorten the retransmission wait time. Of course, the above examples are only a few possible tuning methods when the network quality optimization model is ARQ. Other possible situations are not limited here, nor will they be listed one by one.

[0061] The above scheme obtains a packet loss state sequence, which is obtained through the process steps in the above network packet loss simulation method embodiment. The simulation platform used to simulate network transmission is then controlled to perform transmission simulation according to the packet loss state sequence. The network quality optimization model is used by at least one of the sender and receiver on the simulation platform. Based on the transmission records obtained after the transmission simulation, the network quality optimization model is tuned. Since the packet loss state sequence is obtained through the process steps in the above network packet loss simulation method embodiment, the packet loss state sequence can realistically reproduce the complex phenomena of the network environment as much as possible. Therefore, by tuning the network quality optimization model based on the transmission records obtained from the transmission simulation, the adaptability of the network quality optimization model to the network environment can be improved as much as possible, which helps to improve network transmission quality after formal deployment.

[0062] Please see Figure 3 , Figure 3 This is a schematic diagram of the framework of an embodiment of the network packet loss simulation device of this application. The network packet loss simulation device 30 includes: an acquisition module 31, a monitoring module 32, a prediction module 33, and an update module 34. The acquisition module 31 is used to acquire a packet loss state sequence and a target packet loss rate; wherein, the packet loss state sequence includes packet loss labels for the current time step and each previous historical time step, and the packet loss label indicates whether packet loss has occurred; the monitoring module 32 is used to acquire the multi-dimensional network state monitored by the simulation platform during the transmission simulation process of the simulation platform used to simulate network transmission according to the packet loss state sequence; the prediction module 33 is used to predict the packet loss label for the next time step of the current time step based on the packet loss state sequence, the target packet loss rate, and the multi-dimensional network state; the update module 34 is used to update the packet loss state sequence based on the packet loss label of the next time step.

[0063] In the above scheme, the network packet loss simulation device 30 acquires a packet loss state sequence and a target packet loss rate. The packet loss state sequence includes packet loss labels for the current time step and all previous historical time steps. The packet loss label indicates whether packet loss has occurred. Then, during the transmission simulation process on the simulation platform used to simulate network transmission according to the packet loss state sequence, the multi-dimensional network state monitored by the simulation platform is acquired. Based on the packet loss state sequence, the target packet loss rate, and the multi-dimensional network state, a prediction is made to obtain the packet loss label for the next time step after the current time step. Then, based on the packet loss label of the next time step, the packet loss state sequence is updated. On the one hand, because in the simulation... During the transmission simulation based on the packet loss state sequence, the platform monitors the simulation platform to obtain a multi-dimensional network state. This multi-dimensional network state can reflect the network environment as realistically as possible. Furthermore, when predicting the packet loss label for the next time step, the platform combines the packet loss state sequence, the target packet loss rate, and the multi-dimensional network state for prediction. Therefore, it helps to ensure that the packet loss simulation can realistically reproduce the complex phenomena of the network environment. On the other hand, since only the target packet loss rate and the packet loss labels of historical time steps are needed to start the packet loss simulation through the platform, without complex hardware configuration, it helps to reduce the cost and configuration complexity of packet loss simulation. Therefore, it can reduce the cost and configuration complexity of packet loss simulation while ensuring that it can realistically reproduce the complex phenomena of the network environment as much as possible.

[0064] In some disclosed embodiments, the prediction module 33 includes a feature extraction submodule for extracting features based on the multidimensional network state to obtain network features; the prediction module 33 includes a temporal modeling submodule for performing temporal modeling based on the packet loss state sequence, the target packet loss rate, and the concatenated sequence of network features to obtain a first feature; the prediction module 33 includes a multi-head attention submodule for processing the first feature based on the multi-head attention with each attention head configured to different attention ranges to obtain a second feature; and the prediction module 33 includes a label mapping submodule for performing label mapping based on the fused features of the first and second features to obtain the packet loss label for the next time step.

[0065] In some disclosed embodiments, the feature extraction submodule includes a feature embedding unit, which is used to embed features based on the network state of each dimension in the multidimensional network state to obtain embedded features of each dimension; the feature extraction submodule includes a weight prediction unit, which is used to perform adaptive weight prediction based on the embedded features of each dimension to obtain weight factors of each dimension; the feature extraction submodule includes a feature weighting unit, which is used to weight based on the embedded features and weight factors of each dimension to obtain network features.

[0066] In some disclosed embodiments, the time-series modeling submodule includes a preprocessing unit for preprocessing the spliced ​​sequence to obtain sequence features; wherein the preprocessing includes at least feature mapping; the time-series modeling submodule includes a first GRU unit for processing the sequence features based on a first GRU layer to capture feature representations representing short-term dependencies; the time-series modeling submodule includes a second GRU unit for processing feature representations representing short-term dependencies based on a second GRU layer to model feature representations representing intermediate-term patterns; the time-series modeling submodule includes a third GRU unit for processing feature representations representing intermediate-term patterns based on a third GRU layer to extract feature representations representing long-term features as first features.

[0067] In some disclosed embodiments, the multi-head attention submodule includes a first attention unit, used to process a first feature based on a first attention head whose attention range is a target time step interval, to obtain a first attention feature; the multi-head attention submodule includes a second attention unit, used to process the first feature based on a second attention head whose attention range is a target time step interval, to obtain a second attention feature; the multi-head attention submodule includes a third attention unit, used to process the first feature based on a third attention head whose attention range is a target packet loss sequence interval, to obtain a third attention feature; wherein, the target packet loss sequence interval is a local interval in the packet loss state sequence where the packet loss rate is higher than a packet loss rate threshold; the multi-head attention submodule includes a fourth attention unit, used to process the first feature based on a fourth attention head whose attention range is the entire input feature sequence, to obtain a fourth attention feature; the multi-head attention submodule includes a feature fusion unit, used to fuse the first attention feature, the second attention feature, the third attention feature, and the fourth attention feature to obtain a second feature.

[0068] In some disclosed embodiments, the network packet loss simulation device 30 includes a loop module for returning to the steps of acquiring the multi-dimensional network state monitored by the simulation platform during the transmission simulation of the simulation platform for simulating network transmission according to the packet loss state sequence, so as to continuously update the packet loss state sequence until the latest packet loss state sequence meets the simulation requirements.

[0069] In some disclosed embodiments, the simulation requirements include: the difference between the packet loss rate of the latest packet loss state sequence and the target packet loss rate is less than the allowable error; or, the simulation requirements include: the sequence length of the latest packet loss state sequence reaches the target length, and the difference between the packet loss rate of the latest packet loss state sequence and the target packet loss rate is less than the allowable error.

[0070] In some disclosed embodiments, the packet loss label for the next time step is predicted by the packet loss simulation model. The monitoring module 32 is also used to monitor during real network transmission to obtain the real packet loss state sequence and the real multidimensional network state. The real packet loss state sequence includes the packet loss labels at each time step in the real network transmission. The network packet loss simulation device 30 includes a preparation module for obtaining the packet loss rate of the real packet loss state sequence as the sample target packet loss rate, extracting the packet loss label of the last time step in the real packet loss state sequence as the sample packet loss label, and removing the sample packet loss labels from the real packet loss state sequence to obtain the sample packet loss state sequence. The prediction module 33 is also used to make predictions based on the sample packet loss state sequence, the sample target packet loss rate, and the real multidimensional network state to obtain the predicted packet loss probability of the last time step. The network packet loss simulation device 30 includes a parameter tuning module for adjusting the network parameters of the packet loss simulation model based on the difference between the sample packet loss label and the predicted packet loss probability.

[0071] In some publicly disclosed embodiments, the multidimensional network state includes any two or more of the following: network transmission delay, delay jitter, network congestion index, bandwidth utilization, RTT change rate, historical burst length, queue congestion level, and retransmission frequency index.

[0072] Please see Figure 4 , Figure 4 This is a schematic diagram of the framework of an embodiment of the network quality optimization model tuning device of this application. The network quality optimization model tuning device 40 includes: an acquisition module 41, a control module 42, and a tuning module 43. The acquisition module 41 is used to acquire a packet loss state sequence; wherein, the packet loss state sequence is obtained through the network packet loss simulation device in the above-described network packet loss simulation device embodiment; the control module 42 is used to control the simulation platform used to simulate network transmission to perform transmission simulation according to the packet loss state sequence; wherein, the network quality optimization model is used for at least one of the sender and receiver of the simulation platform; the tuning module 43 is used to tune the network quality optimization model based on the transmission records after the simulation platform completes the transmission simulation.

[0073] In the above scheme, the network quality optimization model tuning device 40 obtains the packet loss state sequence, which is obtained through the network packet loss simulation device in the above-described network packet loss simulation device embodiment. It then controls the simulation platform used to simulate network transmission to perform transmission simulation according to the packet loss state sequence. The network quality optimization model is used by at least one of the sender and receiver on the simulation platform. Based on the transmission records obtained after the transmission simulation by the simulation platform, the network quality optimization model is tuned. Since the packet loss state sequence is obtained through the network packet loss simulation device in the above-described network packet loss simulation device embodiment, the packet loss state sequence can realistically reproduce the complex phenomena of the network environment as much as possible. Therefore, by tuning the network quality optimization model using the transmission records obtained from the transmission simulation, the adaptability of the network quality optimization model to the network environment can be improved as much as possible, which helps to improve network transmission quality after formal deployment.

[0074] Please see Figure 5 , Figure 5 This is a schematic diagram of a framework of an embodiment of the electronic device of this application. The electronic device 50 includes at least a memory 51 and a processor 52 coupled to each other. The memory 51 stores at least program instructions, and the processor 52 is used to execute the program instructions to implement the steps in any of the above-described network packet loss simulation method embodiments, or to implement the steps in any of the above-described network quality optimization model tuning method embodiments. For details, please refer to the foregoing disclosed embodiments, which will not be repeated here. The electronic device 50 may include, but is not limited to, industrial control computers, servers, etc., and the specific type of the electronic device 50 is not limited here.

[0075] Specifically, processor 52 controls itself and memory 51 to implement the steps in any of the above-described network packet loss simulation method embodiments, or to implement the steps in any of the above-described network quality optimization model tuning method embodiments. Processor 52 can also be called a CPU (Central Processing Unit). Processor 52 may be an integrated circuit chip with signal processing capabilities. Processor 52 can also be a general-purpose processor, digital signal processor (DSP), application-specific integrated circuit (ASIC), field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. A general-purpose processor can be a microprocessor or any conventional processor. Furthermore, processor 52 can be implemented using integrated circuit chips.

[0076] In the above scheme, electronic device 50 acquires a packet loss state sequence and a target packet loss rate. The packet loss state sequence includes packet loss labels for the current time step and all previous historical time steps. The packet loss label indicates whether packet loss has occurred. Then, during the transmission simulation process on the simulation platform used to simulate network transmission according to the packet loss state sequence, the multi-dimensional network state monitored by the simulation platform is acquired. Based on the packet loss state sequence, the target packet loss rate, and the multi-dimensional network state, prediction is made to obtain the packet loss label for the next time step after the current time step. Then, based on the packet loss label of the next time step, the packet loss state sequence is updated. On the one hand, because in the simulation platform... During the transmission simulation based on the packet loss state sequence, the simulation platform is monitored to obtain a multi-dimensional network state. This multi-dimensional network state can reflect the network environment as realistically as possible. Furthermore, when predicting the packet loss label for the next time step, the packet loss state sequence, target packet loss rate, and multi-dimensional network state are combined for prediction. Therefore, this helps to ensure that the packet loss simulation can realistically reproduce the complex phenomena of the network environment. On the other hand, since only the target packet loss rate and the packet loss labels of historical time steps are needed to start the packet loss simulation through the simulation platform, without complex hardware configuration, it helps to reduce the cost and configuration complexity of packet loss simulation. Therefore, it is possible to reduce the cost and configuration complexity of packet loss simulation while ensuring that it can realistically reproduce the complex phenomena of the network environment as much as possible. Furthermore, a packet loss state sequence is obtained through the process steps described in the above-described network packet loss simulation method embodiment. The simulation platform used to simulate network transmission is then controlled to perform transmission simulation according to the packet loss state sequence. The network quality optimization model is used for at least one of the sender and receiver on the simulation platform. Based on the transmission records obtained after the transmission simulation is completed by the simulation platform, the network quality optimization model is tuned. Since the packet loss state sequence is obtained through the process steps described in the above-described network packet loss simulation method embodiment, the packet loss state sequence can ensure the realistic reproduction of complex phenomena in the network environment as much as possible. Therefore, by tuning the network quality optimization model based on the transmission records obtained from the transmission simulation, the adaptability of the network quality optimization model to the network environment can be improved as much as possible, which helps to improve network transmission quality after formal deployment.

[0077] Please see Figure 6 , Figure 6 This is a schematic diagram of a framework of an embodiment of the computer-readable storage medium of this application. The computer-readable storage medium 60 stores program instructions 61 that can be executed by a processor. The program instructions 61 are used to implement the steps in any of the above-described network packet loss simulation method embodiments, or to implement the steps in any of the above-described network quality optimization model tuning method embodiments.

[0078] In the above scheme, the computer-readable storage medium 60 acquires the packet loss state sequence and the target packet loss rate. The packet loss state sequence includes packet loss labels for the current time step and all previous historical time steps. The packet loss label indicates whether packet loss has occurred. Then, during the transmission simulation process on the simulation platform used to simulate network transmission according to the packet loss state sequence, the multi-dimensional network state monitored by the simulation platform is acquired. Based on the packet loss state sequence, the target packet loss rate, and the multi-dimensional network state, a prediction is made to obtain the packet loss label for the next time step after the current time step. Then, based on the packet loss label of the next time step, the packet loss state sequence is updated. On the one hand, because in the simulation... During the transmission simulation based on the packet loss state sequence, the platform monitors the simulation platform to obtain a multi-dimensional network state. This multi-dimensional network state can reflect the network environment as realistically as possible. Furthermore, when predicting the packet loss label for the next time step, the platform combines the packet loss state sequence, the target packet loss rate, and the multi-dimensional network state for prediction. Therefore, it helps to ensure that the packet loss simulation can realistically reproduce the complex phenomena of the network environment. On the other hand, since only the target packet loss rate and the packet loss labels of historical time steps are needed to start the packet loss simulation through the platform, without complex hardware configuration, it helps to reduce the cost and configuration complexity of packet loss simulation. Therefore, it can reduce the cost and configuration complexity of packet loss simulation while ensuring that it can realistically reproduce the complex phenomena of the network environment as much as possible. Furthermore, a packet loss state sequence is obtained through the process steps described in the above-described network packet loss simulation method embodiment. The simulation platform used to simulate network transmission is then controlled to perform transmission simulation according to the packet loss state sequence. The network quality optimization model is used for at least one of the sender and receiver on the simulation platform. Based on the transmission records obtained after the transmission simulation is completed by the simulation platform, the network quality optimization model is tuned. Since the packet loss state sequence is obtained through the process steps described in the above-described network packet loss simulation method embodiment, the packet loss state sequence can ensure the realistic reproduction of complex phenomena in the network environment as much as possible. Therefore, by tuning the network quality optimization model based on the transmission records obtained from the transmission simulation, the adaptability of the network quality optimization model to the network environment can be improved as much as possible, which helps to improve network transmission quality after formal deployment.

[0079] In some embodiments, the functions or modules of the apparatus provided in this disclosure can be used to perform the methods described in the above method embodiments. The specific implementation can be referred to the description of the above method embodiments, and for the sake of brevity, it will not be repeated here.

[0080] The description of the various embodiments above tends to emphasize the differences between the various embodiments. The similarities or similarities between them can be referred to, and for the sake of brevity, they will not be repeated here.

[0081] In the several embodiments provided in this application, it should be understood that the disclosed methods and apparatus can be implemented in other ways. For example, the apparatus implementations described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0082] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0083] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0084] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods of various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0085] If the technical solution of this application involves personal information, the product using this technical solution has clearly informed the user of the personal information processing rules and obtained the user's voluntary consent before processing the personal information. If the technical solution of this application involves sensitive personal information, the product using this technical solution has obtained the user's separate consent before processing the sensitive personal information, and also meets the requirement of "express consent". For example, at personal information collection devices such as cameras, clear and prominent signs are set up to inform users that they have entered the scope of personal information collection and that personal information will be collected. If an individual voluntarily enters the collection scope, it is deemed that they have agreed to the collection of their personal information; or on the personal information processing device, with clear signs / information informing users of the personal information processing rules, authorization is obtained from the individual through pop-up information or by asking the individual to upload their personal information; wherein, the personal information processing rules may include information such as the personal information processor, the purpose of personal information processing, the processing method, and the types of personal information processed.

Claims

1. A network packet loss emulation method, characterized by, The method comprises: obtaining a packet loss state sequence and a target packet loss rate; wherein the packet loss state sequence comprises packet loss labels of a current time step and each historical time step before the current time step, and the packet loss label represents whether to transmit a packet loss; obtaining a multi-dimensional network state monitored by a simulation platform for simulating network transmission during a transmission simulation process of the simulation platform according to the packet loss state sequence; predicting based on the packet loss state sequence, the target packet loss rate and the multi-dimensional network state to obtain a packet loss label of a next time step of the current time step; updating the packet loss state sequence based on the packet loss label of the next time step.

2. The method of claim 1, wherein, The predicting based on the packet loss state sequence, the target packet loss rate and the multi-dimensional network state to obtain a packet loss label of a next time step of the current time step comprises: extracting features based on the multi-dimensional network state to obtain network features; performing time series modeling based on a splicing sequence of the packet loss state sequence, the target packet loss rate and the network features to obtain first features; processing the first features based on multi-head attention in which each attention head is configured to have a different attention range to obtain second features; performing label mapping based on a fusion feature of the first features and the second features to obtain the packet loss label of the next time step.

3. The method of claim 2, wherein, The extracting features based on the multi-dimensional network state to obtain network features comprises: performing feature embedding based on network states of each dimension in the multi-dimensional network state to obtain embedded features of the each dimension; performing adaptive weight prediction based on the embedded features of the each dimension to obtain weight factors of the each dimension; performing weighting based on the embedded features and the weight factors of the each dimension to obtain the network features.

4. The method of claim 2, wherein, The performing time series modeling based on a splicing sequence of the packet loss state sequence, the target packet loss rate and the network features to obtain first features comprises: performing preprocessing based on the splicing sequence to obtain sequence features; wherein the preprocessing at least comprises feature mapping; processing the sequence features based on a first GRU layer to capture feature representations representing short-term dependencies; processing the feature representations representing short-term dependencies based on a second GRU layer to model feature representations representing medium-term patterns; processing the feature representations representing medium-term patterns based on a third GRU layer to extract feature representations representing long-term features as the first features.

5. The method of claim 2, wherein, The processing the first features based on multi-head attention in which each attention head is configured to have a different attention range to obtain second features comprises: processing the first features based on a first attention head having an attention range of a target time step interval to obtain first attention features; processing the first features based on a second attention head having an attention range of a target time step interval to obtain second attention features; processing the first features based on a third attention head having an attention range of a target packet loss sequence interval to obtain third attention features; wherein the target packet loss sequence interval is a local interval in the packet loss state sequence in which a packet loss rate is higher than a packet loss rate threshold. obtaining a fourth attention feature by processing the first feature based on the fourth attention head with a focus range of the entire input feature sequence; and performing fusion based on the first attention feature, the second attention feature, the third attention feature, and the fourth attention feature to obtain the second feature.

6. The method of claim 1, wherein, After the updating of the packet loss state sequence based on the packet loss label of the next time step, the method further comprises: In the process of the simulation platform for simulating network transmission performing transmission simulation according to the packet loss state sequence, the step of obtaining the multi-dimensional network state monitored by the simulation platform is returned to update the packet loss state sequence until the latest packet loss state sequence meets the simulation requirements.

7. The method of claim 6, wherein, The simulation requirements include that the difference between the packet loss rate of the latest packet loss state sequence and the target packet loss rate is less than the allowable error. Alternatively, the simulation requirements include that the sequence length of the latest packet loss state sequence reaches the target length, and the difference between the packet loss rate of the latest packet loss state sequence and the target packet loss rate is less than the allowable error.

8. The method of claim 1, wherein, The packet loss label of the next time step is obtained by a packet loss simulation model, and the training steps of the packet loss simulation model include: Monitoring during real network transmission to obtain a real packet loss state sequence and a real multi-dimensional network state; wherein the real packet loss state sequence includes packet loss labels of each time step in the real network transmission; Obtaining the packet loss rate of the real packet loss state sequence as a sample target packet loss rate, and obtaining the packet loss label of the last time step in the real packet loss state sequence as a sample packet loss label, and removing the sample packet loss label from the real packet loss state sequence as a sample packet loss state sequence; Based on the sample packet loss state sequence, the sample target packet loss rate, and the real multi-dimensional network state, a predicted packet loss probability of the last time step is obtained; Based on the difference between the sample packet loss label and the predicted packet loss probability, the network parameters of the packet loss simulation model are adjusted.

9. The method according to any one of claims 1 to 8, characterized in that, The multi-dimensional network state includes any two or more of network transmission delay, delay jitter degree, network congestion indicator, bandwidth utilization rate, RTT change rate, historical burst length, queue congestion degree, and retransmission frequency indicator.

10. A method for tuning a network quality optimization model, the method comprising: receiving a plurality of network quality optimization models; and tuning the plurality of network quality optimization models. It comprises: obtaining a packet loss state sequence; wherein the packet loss state sequence is obtained by the network packet loss simulation method of any one of claims 1 to 9; controlling a simulation platform for simulating network transmission to perform transmission simulation according to the packet loss state sequence; wherein the network quality optimization model is used for at least one of the sending side and the receiving side of the simulation platform; Based on the transmission record of the simulation platform after completing the transmission simulation, the network quality optimization model is optimized.

11. A network packet loss emulation apparatus, characterized by, It comprises: an obtaining module, configured to obtain a packet loss state sequence and a target packet loss rate; wherein the packet loss state sequence includes packet loss labels of the current time step and each historical time step before the current time step, and the packet loss label represents whether to transmit packet loss or not. a monitoring module, configured to acquire a multi-dimensional network state monitored by a simulation platform for simulating network transmission according to the packet loss state sequence during transmission simulation by the simulation platform; a prediction module, configured to predict, based on the packet loss state sequence, the target packet loss rate and the multi-dimensional network state, a packet loss label of a next time step of the current time step; an updating module, configured to update the packet loss state sequence based on the packet loss label of the next time step.

12. A tuning apparatus of a network quality optimization model, characterized by, comprise: an acquisition module, configured to acquire a packet loss state sequence; wherein the packet loss state sequence is acquired by the network packet loss simulation method in claim 11; a control module, configured to control a simulation platform for simulating network transmission to perform transmission simulation according to the packet loss state sequence; wherein the network quality optimization model is used for at least one of a sender and a receiver of the simulation platform; an optimization module, configured to optimize the network quality optimization model based on transmission records after the simulation platform completes the transmission simulation.

13. An electronic device, comprising: at least comprise a memory and a processor coupled with each other, the memory at least stores program instructions, and the processor is configured to execute the program instructions to implement the network packet loss simulation method in any one of claims 1 to 9 or the optimization method of the network quality optimization model in claim 10.

14. A computer-readable storage medium, characterized in that, store program instructions capable of being run by a processor, and the program instructions are used to implement the network packet loss simulation method in any one of claims 1 to 9 or the optimization method of the network quality optimization model in claim 10.

Images