Satellite data traffic measurement method and system

Abstract

The invention discloses a satellite data traffic measurement method and system, and belongs to the technical field of satellite networks. In order to solve the problems of poor satellite network flow measurement precision and high complexity in the prior art, the invention provides a satellite data flow measurement method, which comprises the following steps: determining a satellite node for data flow measurement, and initializing a configuration probability data structure algorithm and an imitation learning algorithm; sampling network flow in a period of time, judging whether a data flow jitter value in the period of time exceeds a threshold value or not, and if the data flow jitter value exceeds the threshold value, generating a flow value according to a label value and an expert strategy by utilizing an imitation learning algorithm; if not, using a probability data structure algorithm to measure and record a label value, and generating an expert strategy according to the label value; and completing measurement. According to the method, resource utilization is reasonably optimized while the measurement accuracy is improved, high-precision and high-timeliness satellite flow measurement is achieved through mutual complementation of the probability data structure algorithm and the imitation learning algorithm, and the method is low in hardware dependency, high in precision, high in intelligence and high in adaptability.

Description

Technical Field

[0001] The present invention belongs to the technical field of satellite network flow measurement, and more specifically, relates to a satellite data flow measurement method and system. Background Art

[0002] Satellite network is an extension of traditional terrestrial network, and is widely used in long-distance content distribution, emergency communication and multi-target positioning. Data flow measurement refers to the number of bytes transmitted per unit time in the network. Data flow measurement is used in network fault diagnosis, performance analysis and evaluation, billing, etc., and is an important network technology. Due to the large scale of satellite network, the transmission includes multiple protocols and is dynamic, so it is necessary to design data flow measurement methods according to its characteristics. Data flow measurement of terrestrial network cannot be well applied to satellite network.

[0003] Due to the characteristics of satellite network topology, large scale, high latency and high packet loss rate, its data flow measurement technology is highly dependent on hardware, such as the traffic monitoring equipment developed by Packteer in the United States, but it has low intelligence and cannot be well applied to satellite networks with dynamic network changes. Some scholars have also proposed artificial intelligence algorithms to measure network traffic based on historical experience. Its advantage is that it can adapt to dynamic topology, but it has disadvantages such as strong dependence on hardware (medium), high computational complexity and low precision. Ideal satellite data flow measurement should have the following characteristics: high accuracy, low complexity, high intelligence and high dynamics of adaptive satellite topology. The current status of satellite data flow measurement can only meet a few of them. Based on this, a satellite data flow measurement method with high intelligence, high accuracy, low complexity and low hardware dependence is needed.

[0004] Corresponding improvements have also been made to the above problems, such as China's patent application number CN202310468314.3, published on July 11, 2023. The patent discloses a high-density low-orbit satellite network traffic measurement method based on matrix filling, the method includes: determining the initial measurement satellite node based on a sampling model of discreteness and coverage; using a random walk model to evaluate the probability of satellite link failure, using a bipartite graph network structure recommendation algorithm to analyze the selection of sampling points at historical moments, and determining the measurement satellite node sequence at the new moment; comparing the error rate and standard error of the overlapping parts in the recovery matrix of adjacent windows to achieve dynamic adaptive adjustment of the number of measurement satellite nodes. The disadvantage of this patent is that although it reduces satellite resource consumption, the operation complexity is high.

[0005] Another example is Chinese patent application number CN202311600533.9, published on April 12, 2024. The patent discloses a satellite access traffic prediction method based on LSTM and access strategy, which belongs to the field of communication technology. The present invention regards the satellite access traffic in a certain ground coverage area as time series data, and uses a long short-term memory network model to predict the traffic at the next moment. In addition, by analyzing the satellite access strategy of the ground terminal, that is, analyzing which satellite the ground traffic accesses, the ground traffic is allocated to the corresponding satellite, thereby effectively solving the problem of satellite traffic prediction. The shortcoming of this patent is that although the operation complexity is low, the accuracy cannot be effectively guaranteed. Summary of the invention

[0006] 1. Problems to be solved

[0007] In view of the problem that the existing satellite network traffic measurement has poor accuracy and high complexity, the present invention provides a satellite data traffic measurement method and system. The measurement method of the present invention improves the measurement accuracy while rationally optimizing resource utilization, and realizes high-precision and timely satellite traffic measurement through the mutual complementation of probabilistic data structure algorithm and imitation learning algorithm, with low hardware dependence, high accuracy, high intelligence and strong adaptability. The system structure of the present invention is simple, the operation complexity is low and the intelligence level is high.

[0008] 2. Technical solution

[0009] To solve the above problems, the present invention adopts the following technical solutions.

[0010] A satellite data flow measurement method comprises the following steps:

[0011] S1: Determine the satellite node for data flow measurement, and initialize and configure the probabilistic data structure algorithm and the imitation learning algorithm at the satellite node;

[0012] S2: Sampling network traffic within a period of time Δt∈[t1,t2], and calculating the data flow jitter value of the period of time Δt. If the data flow jitter value is lower than or equal to the threshold, the probabilistic data structure algorithm is used to measure and record the label value, and an expert strategy is generated according to the label value, and the traffic value is generated according to the label value and the expert strategy; if the data flow jitter value exceeds the threshold, the imitation learning algorithm is used to generate the traffic value according to the label value and the expert strategy;

[0013] S3: Return the data value of the period of time Δt to the user, completing the data flow measurement.

[0014] Furthermore, the calculation of the data stream jitter value M in step S2 is as follows:

[0015]

[0016] Where i represents the total number of data packets; RTT i is the RTT of the ith data packet, RTT is the Round-Trip Time; n is the total number of sampled data packets; B is the total bandwidth; B′ is the total remaining bandwidth; It is the average RTT value of the data packets sampled in the previous stage.

[0017] Furthermore, the network traffic sampling performed in step S2 includes the start time, link number, remaining bandwidth value, packet loss rate, Round-Trip Time and the number of bytes transmitted.

[0018] Furthermore, the expert strategy L in step S2, is a set, number is the expert strategy number, In , l is the flow value, o is the standard operation, and x is the flow change feature; and the expert trajectory is applied to form the expert strategy, and the weights and relationships of different expert trajectories are defined as follows:

[0019]

[0020] Among them, w a is the weight of number a; b is the manual adjustment factor, which is used to dynamically adjust the value of a certain item.

[0021] Furthermore, in step S1, initializing the configuration probability data structure algorithm and the imitation learning algorithm includes:

[0022] Select the sketch algorithm as the probabilistic data structure algorithm and set the sketch algorithm to an m×n multidimensional matrix;

[0023] The reinforcement learning algorithm based on generative adversarial network, namely GAIL, is selected as the imitation learning algorithm, and the generator and discriminator in the reinforcement learning algorithm based on generative adversarial network are initialized and defined.

[0024] Furthermore, the method further comprises step S4: evaluating the performance of the entire measurement method according to the data value returned to the user, and updating the probabilistic data structure algorithm and the imitation learning algorithm according to the evaluation performance result.

[0025] Furthermore, a performance evaluation function is used to evaluate the performance of the measurement method. The performance evaluation function is:

[0026]

[0027] in, is the difference between the true value and the calculated value; N is the number of labels.

[0028] A system using a satellite data flow measurement method as described in any one of the above items, comprising:

[0029] Satellite node to be tested;

[0030] Calculation unit: used for sampling network traffic within a period of time Δt∈[t1, t2], and calculating the data flow jitter value of the period of time Δt;

[0031] Traffic generation unit: used to generate traffic.

[0032] 3. Beneficial effects

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] (1) In the whole process of satellite data flow measurement, the present invention selects appropriate measurement means and processing according to the size of the data flow jitter value within the adopted time. When the data flow jitter value is small, it indicates that the transmission data flow is relatively stable. The probability data structure algorithm is selected for measurement, which can record the frequency of the element to be measured or estimate its cardinality with a small memory space, and has low hardware dependence and little additional network overhead. The advantages are low measurement cost, low complexity, strong timeliness and high accuracy. The disadvantage is that it cannot handle complex data flows, burst data flows, etc. well. When the data flow jitter value is large, it indicates that the transmission data flow has large fluctuations and uncertainties, which can form or cause complex data flows, burst data flows, etc. According to the data flow, the imitation learning algorithm is used to measure according to previous historical experience to improve the accuracy and robustness of measuring complex data flows; since the label values ​​generated by the probabilistic data structure algorithm provide basic data samples and expert strategies for the simulation learning algorithm, the simulation learning can more accurately learn and predict the changes in data flow, which is conducive to optimizing its measurement strategy and action; the whole method improves the measurement efficiency and accuracy while reasonably optimizing resource utilization. The joint data flow measurement method composed of the probabilistic data structure algorithm and the imitation learning algorithm complements each other to achieve high-precision and timely satellite flow measurement, which has the characteristics of low hardware dependence, high precision, high intelligence and strong adaptability;

[0035] (2) When calculating the jitter value of a data stream, the present invention introduces the concepts of RTT, total bandwidth, and the sum of remaining bandwidth. RTT measures the delay of network communication and can more comprehensively evaluate the stability of the network; the sum of total bandwidth and remaining bandwidth provides information on network capacity, which helps to understand the performance of the network under high load; the calculation of the jitter value no longer depends on the variation of the traditional data packet arrival time, so that it can more comprehensively reflect the actual situation of the network data stream, thereby improving the accuracy of the jitter value calculation and providing guarantee for subsequent measurements;

[0036] (3) The present invention designs the expert strategy into a set form, which includes the characteristics of the data flow of the satellite network under certain conditions. Since the data flow in the satellite network is complex and diverse, the characteristics of the data flow under different conditions may be different. The set form can make the expert strategy accommodate multiple data flow characteristics, thereby ensuring the comprehensiveness of the expert strategy formation; and the expert trajectory is used to form the expert strategy, and the weights and relationships of different expert trajectories are defined, which can better ensure the accuracy of the expert strategy formation, and can also be flexibly adjusted according to changes in the network environment, data flow characteristics and user needs, thereby improving the flexibility and applicability of the entire expert strategy;

[0037] (4) The present invention uses the sketch algorithm as the probabilistic data structure algorithm, which is a hash-based multidimensional data structure that can store the frequency of a certain type of elements in multiple sets. It has been widely used in data measurement in traditional networks and has the characteristics of high precision, small memory usage and low complexity. GAIL is selected as the imitation learning algorithm, which only requires a small number of samples and expert strategies to make artificial intelligence decisions and has low operation complexity. After returning the data value to the user, the performance of the entire method will be evaluated to further ensure the accuracy of the entire measurement method, so that the true value of the satellite data flow under various conditions can be obtained. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is a schematic diagram of the overall process of the present invention;

[0039] Figure 2 It is a schematic diagram of the principle of the present invention. DETAILED DESCRIPTION

[0040] The present invention is further described below in conjunction with specific embodiments and drawings.

[0041] like Figure 1 and Figure 2 As shown, a satellite data flow measurement method comprises the following steps:

[0042] S1: Determine a satellite node for data flow measurement, and initialize and configure a probabilistic data structure algorithm and an imitation learning algorithm at the satellite node; wherein the probabilistic data structure algorithm may include a sketch algorithm, a hash function, a counter, and a multidimensional table, etc.;

[0043] S2: Sampling the network traffic within a period of time Δt∈[t1,t2], and calculating the data flow jitter value of the period of time Δt. If the data flow jitter value is lower than or equal to the threshold (the value of the threshold depends on the counter of the probabilistic data structure in the probabilistic data structure algorithm), the probabilistic data structure algorithm is used to measure and record the label value, and an expert strategy is generated according to the label value, and a traffic value is generated according to the label value and the expert strategy; if the data flow jitter value exceeds the threshold, the imitation learning algorithm is used to generate the traffic value according to the generated label value and the expert strategy;

[0044] Specifically, in this step, the scheme of the probabilistic data structure algorithm to measure and record the tag value is:

[0045] First, construct the experience replay pool (Experience Replay), which is a six-tuple: (s, s′, r, a, slot, score), where slot is the time slot number, used to distinguish different five-tuple values. Score is the score measured by the probability data structure algorithm this time. The higher the score, the higher the measurement efficiency.

[0046] Then, save the six-tuple with high scores, and the calculation formula is:

[0047]

[0048] data is the amount of successfully measured data in bytes, and Δt is the measurement time.

[0049] The sextuple with the highest score is the label value, which will be used to imitate the gradient learning of the neural network to generate an expert strategy for it, such as: from state s to state s′ to the need to apply action a, etc.

[0050] S3: Return the data value of the period of time Δt to the user, completing the data flow measurement.

[0051] At present, the mainstream usually uses a single algorithm for satellite data traffic measurement. Then the single algorithm measurement will result in a lack of flexibility to adapt to different network environments and traffic patterns, so that in some cases the measurement is inaccurate or cannot meet the measurement requirements; or the single algorithm meets the accuracy requirements, but the algorithm structure is complex. In a resource-constrained satellite network environment, this may lead to increased calculation delays and fail to meet real-time requirements. Therefore, the present application creatively uses two algorithms in a complementary manner to measure satellite network data traffic, thereby achieving high-precision and timely satellite traffic measurement. And the label values ​​generated by the probabilistic data structure algorithm can be used for imitation learning to more quickly capture the changing laws of traffic, thereby helping to improve the efficiency and accuracy of the imitation learning algorithm; and through the label values ​​generated by the sketch algorithm, imitation learning can be exposed to more diverse network traffic scenarios and conditions, thereby increasing the generalization ability of the imitation learning algorithm, enabling it to more flexibly respond to various complex network environments and generate more accurate traffic prediction results.

[0052] In the entire satellite data flow measurement process, the present application selects appropriate measurement means and processing according to the size of the data flow jitter value within the adopted time. When the data flow jitter value is small, the data flow is relatively stable. The use of the probabilistic data structure algorithm for measurement can record the frequency of the element to be measured or estimate its cardinality with a smaller memory space, and will not cause additional overhead to the network, thereby reducing the measurement cost; when the data flow jitter value is large, the data flow has large fluctuations and uncertainties, and the use of the simulation learning algorithm can more flexibly respond to changes in data flow, thereby improving the accuracy and robustness of the measurement; and the label value generated by the probabilistic data structure algorithm provides basic data for the simulation learning algorithm, so that the simulation learning can more accurately learn and predict changes in data flow, thereby optimizing its strategy and behavior; the entire method reasonably optimizes resource utilization while improving measurement efficiency and accuracy, and realizes high-precision and timely satellite flow measurement through the mutual complementation of the probabilistic data structure algorithm and the imitation learning algorithm, and has the characteristics of low hardware dependence, high precision, high intelligence and strong adaptability.

[0053] In a specific implementation, the data stream jitter value M in step S2 is calculated as follows:

[0054]

[0055] Where i represents the total number of data packets; RTT i is the RTT of the ith data packet, RTT is the Round-Trip Time; n is the total number of sampled data packets; B is the total bandwidth; B′ is the total remaining bandwidth; It is the average RTT value of the data packets sampled in the previous stage.

[0056] When calculating the jitter value of data flow, the concepts of RTT, total bandwidth and the sum of remaining bandwidth are introduced. RTT measures the delay of network communication and can more comprehensively evaluate the stability of the network; the sum of total bandwidth and remaining bandwidth provides information on network capacity, which helps to understand the performance of the network under high load; the calculation of jitter value no longer depends on the change of traditional data packet arrival time, so that it can more comprehensively reflect the actual situation of network data flow, thereby improving the accuracy of jitter value calculation and providing guarantee for subsequent measurements.

[0057] In a specific implementation, the network traffic sampling in step S2 includes the start time, link number, remaining bandwidth value, packet loss rate, Round-Trip Time, and the number of bytes transmitted. The start time marks the start time of traffic sampling, which helps to locate and sort the traffic data in time; the link number is used to uniquely identify a specific communication link in the satellite network; the remaining bandwidth value can evaluate the load of the link, predict possible network congestion, and provide decision support for traffic control and resource allocation; the packet loss rate is one of the important indicators for measuring network quality; the Round-Trip Time can evaluate the response speed of the network; the number of bytes transmitted records the total amount of data passing through the link in a specific time period; by collecting the above parameters, a comprehensive description of the satellite network traffic status can be obtained, and the sampling is comprehensive to ensure subsequent accuracy.

[0058] In a specific embodiment, the expert strategy L in step S2, is a set, number is the expert strategy number, which is used for imitation learning calculation and is unique. In , l is the flow value, o is the standard operation, and x is the flow change feature; and the expert trajectory is applied to form the expert strategy, and the weights and relationships of different expert trajectories are defined as follows:

[0059]

[0060] Among them, w a is the weight of number a; b is the manual adjustment factor, which is used to dynamically adjust the value of a certain item. It is explained here that the expert strategy L refers to the actions generated by imitating the learning agent to obtain valuable label values: including satellite nodes, state and action pairs, such as:

[0061]

[0062] W t is the number of the satellite in time slot t; It is the measurement status combination in time slot t, such as RTT, remaining bandwidth, etc.; a t yes The action corresponding to the state, such as Knowing that the burst flow is not high-speed at time t, the "20% sampling measurement" action in imitation learning is performed to combine multiple expert strategies Combined together to form the expert strategy trajectory, namely:

[0063]

[0064] The expert strategy is designed in the form of a set, which includes the characteristics of the data flow of the satellite network under certain conditions. Since the data flow in the satellite network is complex and diverse, the characteristics of the data flow may be different under different conditions. The set form allows the expert strategy to accommodate a variety of data flow characteristics, thereby ensuring the comprehensiveness of the expert strategy formation; and applying expert trajectories to form expert strategies and defining the weights and relationships of different expert trajectories can better ensure the accuracy of the expert strategy formation, and can also be flexibly adjusted according to changes in the network environment, data flow characteristics and user needs, thereby improving the flexibility and applicability of the entire expert strategy.

[0065] In a specific implementation, in step S1, initializing the configuration probability data structure algorithm and the imitation learning algorithm includes:

[0066] The sketch algorithm is selected as the probabilistic data structure algorithm, and the sketch algorithm is set to an m×n multidimensional matrix; the sketch algorithm can record the frequency of the elements to be measured or estimate their cardinality in a smaller memory space and will not cause additional overhead to the network. It has higher computing efficiency and lower resource consumption; it has been widely used in data measurement in traditional networks.

[0067] The reinforcement learning algorithm based on generative adversarial network, namely GAIL, is selected as the imitation learning algorithm, and the generator and discriminator in the reinforcement learning algorithm based on generative adversarial network are initialized and defined. GAIL enables the intelligent agent to efficiently imitate the expert strategy and adapt to high-dimensional complex environments through the framework of generative adversarial network. The two algorithms are selected to combine and complement each other to measure data with high accuracy and strong timeliness.

[0068] It is worth noting that a single sampling measurement refers to a method of measuring part of the flow and calculating all the flow values, assuming that the data flow is stable and constant within a certain period of time. Among them, the measurement process under different states can be regarded as a Markov decision process, so the entire measurement problem is modeled as an optimization problem solved by imitation learning, that is:

[0069]

[0070] V(l) is a certain reward (a value) under the measurement agent strategy π. The meaning of the above formula is to minimize the value between the agent and the expert agent. When the values ​​of the two are the smallest (the closer they are), the value generated is the largest (because the expert strategy is the best), and its action a is called the optimal measurement action. Among them, V(l,s) is the specific calculation formula, which is the improved Bellman equation (BellmanEquation), that is: In the above formula, is an expectation of applying strategy π represents the reward value (return) of action a in state s. γ∈[0,1] is the discount factor, which represents the discount of the Markov chain in the transfer. P(s′|s,a) represents the probability of the next state s′. P(s′|s,a)×V π (l,s) represents the sum of the rewards of the next state s' using the strategy π. Use this equation to find the value with the maximum value.

[0071] It should be noted here that the sketch algorithm can efficiently extract the key features of network traffic and convert them into label values. These label values, as the input of imitation learning, can greatly simplify the feature extraction steps in the learning process. In addition, the label values ​​contain the core information of network traffic, so that imitation learning can capture the changing patterns of traffic more quickly, thereby improving learning efficiency and accuracy.

[0072] In a specific implementation, step S4 is also included: the performance of the entire measurement method is evaluated according to the data value returned to the user, and the probabilistic data structure algorithm and the imitation learning algorithm are updated according to the evaluation performance results. In this embodiment, the significance of performance evaluation is to compare the output of the measurement method with the data value actually expected by the user, which can verify the effectiveness and accuracy of the measurement method; and the probabilistic data structure algorithm and the imitation learning algorithm are updated according to the results of the performance evaluation, which helps to further improve the accuracy of the algorithm and reduce the performance degradation caused by algorithm defects or data errors; thereby improving the user experience.

[0073] Preferably, in this embodiment, a performance evaluation function is used to evaluate the performance of the measurement method, and the performance evaluation function is:

[0074]

[0075] in, is the difference between the true value and the calculated value; n is the number of labels. This performance evaluation function is simple in structure and does not involve too many complex parameters. It can quickly implement performance evaluation while ensuring accuracy.

[0076] A system using the satellite data flow measurement method as described above comprises:

[0077] Satellite node to be tested;

[0078] Calculation unit: used for sampling network traffic within a period of time Δt∈[t1, t2], and calculating the data flow jitter value of the period of time Δt;

[0079] Traffic generation unit: used to generate traffic.

[0080] The system of the present invention can reasonably allocate measurement means according to the jitter value of the data flow within the sampling time of the satellite network. When the jitter value of the data flow is small, the sketch algorithm is directly used to measure and record the label value and generate the expert strategy. Conversely, the imitation learning generates the flow value according to the label value and the expert strategy. The entire system realizes efficient allocation of resources, which not only avoids the problems of high computing power operation pressure and serious resource consumption caused by the full use of imitation learning; it also avoids the problems of large errors and low precision caused by the full use of the sketch algorithm; at the same time, it improves the flexibility of the entire system, making it adaptable to various complex network environments and application scenarios; the structure is simple and the operation complexity is low.

[0081] Example 1

[0082] In order to further understand the technical solution of the present application, the following examples are given:

[0083] A satellite data flow measurement method specifically comprises the following steps:

[0084] S1: Determine the satellite node for data flow measurement, and initialize the configuration of the probabilistic data structure algorithm and the imitation learning algorithm at the satellite node; that is, install the sketch and the imitation learning class at the determined satellite node to be measured; and initialize the sketch and the imitation learning class as follows: initialize the sketch algorithm and the imitation learning class, the imitation learning class includes a variety of imitation learning algorithms, wherein the sketch is set to an m×n multidimensional matrix, and the generative adversarial imitation learning (GAIL) is selected in the imitation learning class, and the generator and the discriminator are defined;

[0085] S2: The start and end time of satellite node collection are set to t1 and t2, the number of measurement samples is i, and the measurement task is completed j times. Then the entire traffic matrix is:

[0086]

[0087] The network traffic within a period of time Δt∈[t1,t2] is sampled, including the start time, link number, remaining bandwidth value, packet loss rate, RTT (Round-Trip Time), number of bytes of transmitted data, etc. The jitter value of Δt is calculated, and according to the value of M within Δt, the decision is divided into the following two types:

[0088] When M in the satellite network is low, that is, lower than or equal to the threshold, the sketch algorithm is directly applied to measure and record the label value (for artificial intelligence imitation learning reasoning and decision-making) and generate an expert strategy; the expert strategy L refers to a set in imitation learning, which contains the characteristics of the data flow under a certain condition in the satellite network. For example, the link numbered lt011 has a transmission data volume of 153125 bytes when the remaining bandwidth value is 73Mb and the packet loss rate is 3%. Expert strategies can be applied to imitation learning;

[0089] On the contrary, when M in the satellite network exceeds the threshold, imitation learning generates traffic values ​​based on the label value and expert strategy; the generator and discriminator use the default ones. Finally, all the calculation results are summarized to obtain the final traffic value:

[0090] Where v is the number of measurements

[0091] According to the data value Traffic Δt Evaluate the performance of traffic measurement methods, update sketch and imitation learning expert strategies, hyperparameters, experience replay pools, etc.

[0092] It is explained here that in step S2, in order to facilitate artificial intelligence imitation learning to count the value of the data flow, it is necessary to record the sample label value in the previous measurement, and the sample label includes the label value and the expert strategy formed by the expert trajectory; the label value refers to certain flow values ​​(referring to constant flow, excluding sudden flow, extremely unbalanced flow, etc.) that are spaced (discontinuous), non-empty, and continuously increase or decrease in time; the expert strategy L, is a set, number is the expert strategy number, which is used for imitation learning calculation and is unique. In the figure, l is the flow value, o is the standard operation, also called the demonstration operation, and x is the flow change characteristic.

[0093] S3: Return the data value Traffic within time Δt to the user Δt , complete the data flow measurement.

[0094] Through a large number of experiments and simulations, the inventors were able to obtain the actual value of satellite data traffic under various conditions, proving that the solution described in this application is feasible and effective.

[0095] The examples described in the present invention are merely descriptions of the preferred implementation modes of the present invention, and are not intended to limit the concept and scope of the present invention. Without departing from the design concept of the present invention, various modifications and improvements made to the technical solutions of the present invention by engineers and technicians in this field should all fall within the protection scope of the present invention.

Claims

1. A satellite data flow measurement method, characterized in that: The steps include: S1: Determine the satellite node for data flow measurement, and initialize and configure the probabilistic data structure algorithm and the imitation learning algorithm at the satellite node; S2: Sampling network traffic within a period of time Δt∈[t1,t2], and calculating the data flow jitter value of the period of time Δt. If the data flow jitter value is lower than or equal to the threshold, the probabilistic data structure algorithm is used to measure and record the label value, and an expert strategy is generated according to the label value, and the traffic value is generated according to the label value and the expert strategy; if the data flow jitter value exceeds the threshold, the imitation learning algorithm is used to generate the traffic value according to the label value and the expert strategy; S3: Return the data value of the period of time Δt to the user, completing the data flow measurement.

2. A satellite data flow measurement method according to claim 1, characterized in that: The calculation of the data stream jitter value M in step S2 is as follows: Where i represents the total number of data packets; RTT i is the RTT of the ith data packet, RTT is the Round-Trip Time; n is the total number of sampled data packets; B is the total bandwidth; B′ is the total remaining bandwidth; It is the average RTT value of the data packets sampled in the previous stage.

3. A satellite data flow measurement method according to claim 2, characterized in that: The network traffic sampling in step S2 includes the start time, link number, remaining bandwidth value, packet loss rate, Round-Trip Time and the number of bytes transmitted.

4. A satellite data flow measurement method according to claim 1 or 2, characterized in that: The expert strategy L in step S2, is a set, number is the expert strategy number, In , l is the flow value, o is the standard operation, and x is the flow change feature; and the expert trajectory is applied to form the expert strategy, and the weights and relationships of different expert trajectories are defined as follows: Among them, w a is the weight of number a; b is the manual adjustment factor, which is used to dynamically adjust the value of a certain item.

5. The satellite data flow measurement method according to claim 1, characterized in that: In step S1, initializing the configuration probability data structure algorithm and the imitation learning algorithm includes: Select the sketch algorithm as the probabilistic data structure algorithm and set the sketch algorithm to an m×n multidimensional matrix; The reinforcement learning algorithm based on generative adversarial network, namely GAIL, is selected as the imitation learning algorithm, and the generator and discriminator in the reinforcement learning algorithm based on generative adversarial network are initialized and defined.

6. A satellite data flow measurement method according to claim 1, characterized in that: The method further includes step S4: evaluating the performance of the entire measurement method according to the data value returned to the user, and updating the probabilistic data structure algorithm and the imitation learning algorithm according to the evaluation performance results.

7. A satellite data flow measurement method according to claim 6, characterized in that: The performance evaluation function is used to evaluate the performance of the measurement method. The performance evaluation function is: in, is the difference between the true value and the calculated value; N is the number of labels.

8. A system using a satellite data flow measurement method according to any one of claims 1 to 7, characterized in that: include: Satellite node to be tested; Calculation unit: used for sampling network traffic within a period of time Δt∈[t1, t2], and calculating the data flow jitter value of the period of time Δt; Traffic generation unit: used to generate traffic.

Images