Low earth orbit satellite network scheduling method suitable for unbalanced data flow

Abstract

The invention discloses a low-earth-orbit satellite network scheduling method suitable for unbalanced data streams, relates to the field of low-earth-orbit satellite network data transmission, and is used for solving the problem that due to high-speed movement of satellites, links are periodically connected and disconnected, and finally transmission fails. The controller substitutes into a logistic regression formula to calculate a scheduling probability according to the flow distribution index, the link historical utilization rate, the delay jitter and the reachable efficiency of a prediction node, generates a scheduling value with the maximum reward by using deep reinforcement learning, perfects a low-delay scheduling strategy, and performs scheduling according to a high-bandwidth scheduling strategy. A group of fuzzy rules is formulated according to the historical scheduling bandwidth maximum probability and the link bandwidth stability index to perform fuzzy reasoning, and a high-bandwidth scheduling sequence scheme is determined, so that the reliability and stability of high-bandwidth transmission are improved, the bandwidth resource utilization efficiency is improved, the bandwidth waste is reduced, and the probability of network congestion is reduced.

Description

Technical Field

[0001] The present invention relates to the field of low-orbit satellite network data transmission, and more specifically, to a low-orbit satellite network scheduling method suitable for unbalanced data flow. Background Art

[0002] With its many advantages such as low cost, wide coverage, and low latency, low-orbit satellites have become an important part of the integrated space-ground network and are widely used in long-distance communication, massive data distribution, multi-target tracking and positioning, etc. Specifically, the current research on traffic scheduling in low-orbit satellite networks is still mainly based on ground networks. In ground networks, the traffic prediction model of the autoregressive sliding model is defined to obtain the global state and classify the data of the real-time allocation network nodes of the link, thereby improving the throughput;

[0003] The prior art has the following deficiencies:

[0004] At present, the scheduling methods of low-orbit satellite networks are relatively simple and low in intelligence. They are still mainly based on the scheduling methods in ground networks. Due to the high-speed movement of satellites, the links have the characteristics of periodic on-off, and the links between nodes will be connected and disconnected (on and off). This increases the probability of burst flow and unbalanced flow in satellite networks (compared with ground networks), affecting data transmission and even ultimately failing. Therefore, a low-orbit satellite network scheduling method suitable for unbalanced data flow is proposed.

[0005] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not constitute the prior art that is already known to one of ordinary skill in the art. Summary of the invention

[0006] The embodiments of the present invention provide a low-orbit satellite network scheduling method applicable to unbalanced data flows. In view of the problems existing in satellite network scheduling, a low-orbit satellite network scheduling applicable to burst flows and unbalanced flows is provided.

[0007] To achieve the above object, the present invention provides the following technical solution: a low-orbit satellite network scheduling method suitable for unbalanced data flow, comprising S1: a client sends a request to a server, and the server responds to the client for resources;

[0008] S2: The controller samples the traffic distribution index and link historical usage rate during the data transmission phase, initializes the multi-agent deep reinforcement learning model, and the SDN controller collects delay jitter and predicts the reachable efficiency of the node;

[0009] S3: The controller uses the traffic distribution index, link historical utilization rate, delay jitter, and predicted node reachability efficiency to calculate the scheduling probability through the logistic regression formula, and compares the scheduling probability with the preset scheduling threshold to determine the low-latency scheduling strategy and the high-bandwidth scheduling strategy.

[0010] S4: Obtain high bandwidth scheduling strategy, collect historical scheduling bandwidth maximum probability and link bandwidth stability index;

[0011] S5: Obtain the maximum probability of historical scheduling bandwidth and the link bandwidth stability index, and substitute them into fuzzy logic to determine the high-bandwidth scheduling order solution.

[0012] In a preferred embodiment, the client and the server establish a multi-path connection based on MPQUIC, the client sends an initial connection invitation, and when the client receives a handshake confirmation message, the handshake is completed and the data transmission phase is entered.

[0013] In a preferred embodiment, the traffic distribution index is obtained by weighted calculation based on the ratio of the average value to the maximum value of the remaining bandwidth, the ratio of the average value to the maximum value of the traffic, and the ratio of the average value to the maximum value of the data packet return delay;

[0014] The ratio of the real-time bandwidth usage of the link to the total bandwidth is collected within the specified time window, and the weighted average of the ratio is calculated to obtain the historical usage rate of the link;

[0015] Based on the round-trip delay of each data packet, the delay difference between each pair of adjacent data packets is calculated; the average value of the delay differences is taken to obtain the delay jitter;

[0016] By collecting network topology, link status and traffic data, substituting them into the machine learning model, obtaining the predicted nodes, calculating the similarity between the predicted nodes and the real nodes, and determining the reachable efficiency of the predicted nodes.

[0017] In a preferred embodiment, the traffic distribution index, the historical link usage rate, the delay jitter, and the reachable efficiency of the predicted node are substituted into the logistic regression formula to calculate the scheduling probability. The specific formula is expressed as follows:

[0018]

[0019] In the formula, L is the result of logistic regression calculation, that is, the scheduling probability, e is the natural base, and y is the linear combination term of the logistic regression model. Specifically, y can be set as:

[0020]

[0021] Where β0 is the bias term, β1, β2, β3 and β4 are the regression coefficients of traffic distribution index, link historical utilization rate, delay jitter and predicted node reachability efficiency, respectively.

[0022] In a preferred embodiment, the scheduling probability is compared with a preset scheduling threshold. If the scheduling probability is greater than or equal to the scheduling threshold, it means that the current path node is highly schedulable and is suitable for a low-latency scheduling strategy. If the scheduling probability is less than the scheduling threshold, it means that the path node requires a higher bandwidth to process traffic, and a high-bandwidth scheduling strategy is selected.

[0023] When the low-latency scheduling strategy is executed, a path with the least number of hops is selected;

[0024] When the high-bandwidth scheduling policy is executed, a path with sufficient bandwidth is selected.

[0025] In a preferred embodiment, the low-latency scheduling strategy is based on network status information, and the deep reinforcement learning model continuously adjusts the strategy to maximize the cumulative reward through interaction with the environment, and the scheduling value with the maximum reward is used as the optimal scheduling value under the current state.

[0026] In a preferred embodiment, the probability of the maximum value of the historical scheduling bandwidth is obtained by calculating the ratio of the number of times the link is selected as the maximum bandwidth transmission path to the total scheduling number;

[0027] Based on historical bandwidth data, the ratio of the bandwidth average value to the bandwidth fluctuation range is calculated to obtain the link bandwidth stability index.

[0028] In a preferred embodiment, the maximum probability of historical scheduling bandwidth and the link bandwidth stability index are defined as input variables and divided into different fuzzy sets respectively;

[0029] The high bandwidth scheduling order result is defined as the output variable and divided into fuzzy sets;

[0030] Formulate fuzzy rules to describe the impact of the maximum probability of historical scheduling bandwidth and link bandwidth stability index on the high-bandwidth scheduling order results;

[0031] Fuzzy reasoning is performed based on fuzzy rules to determine the high-bandwidth scheduling order plan.

[0032] Technical effects and advantages of the present invention:

[0033] 1. The present invention improves the impact of complex data flows (such as burst flows, unbalanced flows, etc.) caused by network dynamics on scheduling in current low-orbit satellite networks, and proposes a low-orbit satellite network scheduling method suitable for unbalanced data flows. The scheduling probability is calculated based on the traffic distribution index, link historical utilization rate, delay jitter, and predicted node reachability efficiency, and deep reinforcement learning is used to generate the scheduling value with the maximum reward, thereby reducing the probability of network congestion.

[0034] 2. The present invention obtains a high-bandwidth scheduling strategy, formulates a set of fuzzy rules for fuzzy reasoning based on the probability of the maximum historical scheduling bandwidth and the link bandwidth stability index, determines the high-bandwidth scheduling order plan, improves the utilization efficiency of bandwidth resources, reduces bandwidth waste, and improves the reliability and stability of high-bandwidth transmission. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 The present invention is a method flow chart of a low-orbit satellite network scheduling method applicable to unbalanced data streams. DETAILED DESCRIPTION

[0036] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0037] The present invention includes the following entities: one or more servers (with MPQUIC server source code installed), one or more clients (with MPQUIC client source code installed), a programmable switch (with sketch flow measurement module installed), an SDN controller (with deep reinforcement learning scheduling module installed, etc.);

[0038] The above entities consist of an integrated network of ground and low-orbit satellites, with the goal of achieving efficient data transmission between the client and the server;

[0039] Server (MPQUIC server): The server contains various resources (such as images, videos, etc.) and can provide the required resources to the client. The MPQUIC server source code is installed. The server responds to the client's request and provides data to the client according to the scheduling strategy;

[0040] Client (MPQUIC client): The user initiates a request to the server through the client to obtain the server's resources. The MPQUIC client source code is installed. The client requests and receives data through the MPQUIC protocol;

[0041] Programmable switch: It has the functions of an SDN switch and has an SDN flow table inside. It communicates with the SDN controller through the OpenFlow protocol and is responsible for traffic forwarding and data packet routing. It is installed with the Sketch module, which is used to measure network traffic in real time and collect information such as RTT and traffic size of data flows. Based on traffic measurement, the Sketch module calculates the real-time traffic distribution index and measures the link delay jitter (RTT fluctuation) and bandwidth usage through the Sketch algorithm.

[0042] SDN controller: responsible for obtaining the status information of the entire network, collecting and calculating the traffic distribution index, link historical utilization rate, delay jitter, and the reachable efficiency of the predicted node, and generating the optimal scheduling decision through the deep reinforcement learning scheduling module, and sending the decision to the programmable switch for execution;

[0043] The present invention is divided into the following stages: client request and server response, network traffic data collection and model initialization, scheduling strategy calculation, network scheduling based on low-latency scheduling strategy and high-bandwidth scheduling strategy;

[0044] Example 1

[0045] See also Figure 1 , a low-orbit satellite network scheduling method suitable for unbalanced data flow, the specific operation process is as follows:

[0046] S1: The client sends a request to the server, and the server responds with resources to the client;

[0047] First, the client and the server establish a multipath connection based on MPQUIC;

[0048] Specifically, the client sends an initial connection invitation (Client Hello, CHLO), which contains the basic information required for the connection;

[0049] Secondly, after receiving the request, the server verifies the connection and returns a handshake confirmation message (Server Hello, SHLO), and also returns the paths and resource capabilities supported by the server;

[0050] Finally, when the client receives the handshake confirmation message, the handshake is completed and the data transmission phase begins;

[0051] It should be noted that the handshake process refers to the process of establishing a reliable communication channel between entities such as the client, server, programmable switch and SDN controller based on their functional characteristics and data interaction requirements. The specific data interaction capacity and process are not limited, but are set by the experimenters based on the above entities, so they will not be elaborated here;

[0052] The client generates a resource request message (Resource Request) based on user needs, which contains the following key information: uniquely identifies the required resource, sets the priority according to the real-time requirements of the resource, and the path information currently available to the client, including the RTT and bandwidth of the low-orbit satellite path and the ground path;

[0053] The request message is packaged into a multi-path data packet through the MPQUIC protocol and sent to the server simultaneously using the low-orbit satellite and the ground path;

[0054] S2: The controller samples the traffic distribution index and link historical usage rate during the data transmission phase, initializes the multi-agent deep reinforcement learning model, and the SDN controller collects delay jitter and predicts the reachable efficiency of the node;

[0055] The traffic distribution index is calculated based on network status parameters, quantifies traffic characteristics in the network, dynamically responds to changes in network conditions, and optimizes traffic scheduling decisions. Its acquisition logic is obtained through weighted calculation of the ratio of the average value to the maximum value of the remaining bandwidth, the ratio of the average value to the maximum value of the traffic, and the ratio of the average value to the maximum value of the packet return delay;

[0056] Specifically, the weighted formula of the traffic distribution index is expressed as:

[0057]

[0058] In the formula, α, β, γ∈[0,1] and are all weight factors corresponding to the ratio, BAN avg With BAN max They are the average and maximum values ​​of the remaining bandwidth, respectively, obtained from the SDN API; TRA avg With TRA max They are the average and maximum values ​​of the traffic, obtained by reading the multi-programmable switch flow table by Sketch; RTT avg With RTT max They are the average and maximum packet return delays, obtained from the QUIC server;

[0059] It should be noted that α, β, and γ are determined by the network equipment hardware during actual transmission and are used to manually adjust the parameters, which will not be described in detail here;

[0060] Among them, SDN API, sketch and QUIC are all key technical components for realizing traffic distribution index calculation and network status collection. Specifically, SDN API provides an interactive interface between the software-defined network (SDN) controller and the switch, sketch is used to implement lightweight and fast traffic measurement and sampling in programmable switches, and QUIC provides an efficient data transmission protocol between the client and the server to obtain information related to the round-trip delay of data packets;

[0061] The link historical utilization rate refers to the average ratio of bandwidth resource usage of a network link within a certain historical period of time. Its acquisition logic is to collect the ratio of the link's real-time bandwidth usage to the total bandwidth within a specified time window, and perform weighted average calculation on the ratio to obtain the link historical utilization rate;

[0062] Specifically, the specific formula for calculating the link historical usage rate is:

[0063]

[0064] Where U is the historical link usage rate, T is the time window size, and B uesd (t) is the actual bandwidth usage of the link at time t, B total is the total bandwidth capacity of the link; T = 1, 2, 3, ... t, where t is the index of the current time window;

[0065] Specifically, the actual bandwidth usage is obtained by obtaining the actual bandwidth usage of the link in each time window through the SDN controller or programmable switch, and the static parameters of the link are queried through the switch or SDN API to obtain the total bandwidth capacity of the link;

[0066] It should be noted that the time window interval setting is set by the experimenters based on the low-orbit satellite network scheduling frequency and the number of network fluctuations, which will not be elaborated here;

[0067] Delay jitter refers to the degree of fluctuation in the delay of data packet transmission in the network. Its acquisition logic is based on the round-trip delay of each data packet, calculating the delay difference of each pair of adjacent data packets; taking the average of the delay differences to obtain the delay jitter;

[0068] Specifically, the calculation formula for delay jitter is:

[0069]

[0070] Where J is the delay jitter, N is the total number of data packets, i is the i-th data packet, and RTT i is the round trip delay of the i-th data packet;

[0071] Specifically, the SDN controller predicts the reachability efficiency of nodes through network topology information, link status information, and traffic statistics;

[0072] The reachability efficiency of a predicted node refers to the reachability or arrival efficiency of a node in the network, that is, the transmission efficiency of data packets from the source node to the target node. The acquisition logic is to collect network topology, link status and traffic data, substitute them into the machine learning model, obtain the predicted node, calculate the similarity between the predicted node and the real node, and determine the reachability efficiency P of the predicted node;

[0073] Specifically, the machine learning model can predict the node of the next time window based on a regression model, a decision tree, a neural network, etc. There is no limitation on the selection of the machine learning model, but it is set by the experimenter according to the data characteristics, the experimental objectives, the model complexity and the amount of training data, and is not limited here;

[0074] The similarity between the node of the next window and the real node is calculated. The similarity calculation can be based on cosine similarity. Specifically, the node of the next window and the real node are vectorized and substituted into the cosine similarity calculation to obtain the reachable efficiency of the predicted node;

[0075] Among them, the real node refers to the real value of the node of the next window determined by the SDN controller based on the machine learning model;

[0076] S3: The controller uses the traffic distribution index, link historical utilization rate, delay jitter, and predicted node reachability efficiency to calculate the scheduling probability through the logistic regression formula, and compares the scheduling probability with the preset scheduling threshold to determine the low-latency scheduling strategy and the high-bandwidth scheduling strategy.

[0077] Substitute the traffic distribution index, link historical utilization rate, delay jitter, and predicted node reachability efficiency into the logistic regression formula to calculate the scheduling probability. The specific formula is as follows:

[0078]

[0079] In the formula, L is the result of logistic regression calculation, that is, the scheduling probability, e is the natural base, and y is the linear combination term of the logistic regression model. Specifically, y can be set as:

[0080]

[0081] Where β0 is the bias term, β1, β2, β3 and β4 are the regression coefficients of traffic distribution index, link historical utilization rate, delay jitter and predicted node reachability efficiency respectively;

[0082] Compare the scheduling probability with the preset scheduling threshold. If the scheduling probability is greater than or equal to the scheduling threshold, it means that the current path node is highly schedulable and is suitable for a low-latency scheduling strategy. If the scheduling probability is less than the scheduling threshold, it means that the path node requires a higher bandwidth to process traffic and a high-bandwidth scheduling strategy is selected.

[0083] Specifically, when the low-latency scheduling strategy is executed, a path with the least number of hops is selected, which helps to reduce the transmission delay;

[0084] Specifically, when the high-bandwidth scheduling strategy is executed, a path with sufficient bandwidth is selected to ensure that the high-bandwidth transmission requirements are met;

[0085] Furthermore, the low-latency scheduling strategy generates the optimal scheduling value through deep reinforcement learning based on the network status information, and generates the scheduling value with the maximum reward, which is the optimal scheduling value under the current state;

[0086] Among them, the network status information (including RTT, bandwidth value, link utilization) will be used as input value and recorded as state (state refers to the current state of the network) and passed to the deep reinforcement learning model;

[0087] Specifically, in the reinforcement learning framework, an action is a decision that can be taken by the experimenter under the network state. In the low-latency scheduling strategy, the action includes selecting different routing paths and selecting different flow control mechanisms;

[0088] After each action is trained, the system returns a reward value, which reflects the quality of the current action. For low-latency scheduling, the reward function is usually designed based on latency. The larger the reward, the smaller the latency and the higher the transmission efficiency.

[0089] Finally, the deep reinforcement learning model continuously adjusts its strategy to maximize the cumulative reward through interaction with the environment, and takes the scheduling value with the maximum reward as the optimal scheduling value in the current state;

[0090] The present invention improves the impact of complex data flows (such as burst flows, unbalanced flows, etc.) caused by network dynamics on scheduling in current low-orbit satellite networks, and proposes a low-orbit satellite network scheduling method suitable for unbalanced data flows. The scheduling probability is calculated by substituting the traffic distribution index, link historical utilization rate, delay jitter and predicted node reachability into the logistic regression formula, and deep reinforcement learning is used to generate the scheduling value with the maximum reward, thereby reducing the probability of network congestion.

[0091] Example 2

[0092] In Example 1 of the present invention, examples are given to illustrate that the scheduling probability is calculated by substituting the traffic distribution index, the historical link usage rate, the delay jitter and the reachable efficiency of the predicted node into the logistic regression formula, and the operation strategy of the scheduling with the maximum reward is generated by deep reinforcement learning; however, in Example 1, the delay problem is only solved from the perspective of the low-latency scheduling strategy, and how to select a relatively high-bandwidth path node in the unbalanced data traffic has not been discussed, resulting in a lack of accurate scheduling when high-bandwidth transmission scheduling is required, thereby reducing network transmission efficiency and resource utilization; in view of the above problems, Example 2 of the present invention is further refined;

[0093] S4: Obtain high bandwidth scheduling strategy, collect historical scheduling bandwidth maximum probability and link bandwidth stability index;

[0094] The probability of the maximum value of historical scheduling bandwidth refers to the probability that, in the scheduling history in the specified time window, after filtering out the links selected by the low-latency scheduling strategy, the remaining links are selected as high-bandwidth transmission paths, and the links successfully carry large traffic transmission. The acquisition logic is to calculate the ratio of the number of times the link is selected as the maximum bandwidth transmission path to the total number of scheduling times to obtain the probability of the maximum value of historical scheduling bandwidth;

[0095] It should be noted that the link selected as the high-bandwidth transmission path is determined by the experimenter based on the specific network status and current network bandwidth, and will not be elaborated here;

[0096] The link bandwidth stability index refers to the link bandwidth stability index, which is used to evaluate the bandwidth fluctuation of the link within a specified time window. Its acquisition logic is to calculate the ratio of the bandwidth average value to the bandwidth fluctuation range through historical bandwidth data to obtain the link bandwidth stability index;

[0097] It should be noted that the bandwidth fluctuation amplitude refers to the standard deviation of the link bandwidth, which is calculated by the experimenters based on the historical sampling data of the link bandwidth and the bandwidth values ​​collected, and will not be elaborated here;

[0098] S5: Obtain the maximum probability of historical scheduling bandwidth and link bandwidth stability index, and substitute them into fuzzy logic to determine the high bandwidth scheduling order plan;

[0099] For example, "High", "Low", and "Medium" are used to judge the probability of the maximum value of the historical scheduling bandwidth, and "Stability", "Instability", and "Moderate" are used to judge the stability index of the link bandwidth.

[0100] Formulate a set of fuzzy rules to describe the impact of different input variables on output variables. The definition of rules can be based on professional knowledge or obtained through data analysis and experiments. For example:

[0101] The maximum probability of historical scheduling bandwidth is marked as X, the link bandwidth stability index is marked as U, and the high bandwidth scheduling order result is marked as C_results;

[0102] Then we can define:

[0103] Rule 1:IF(X is High)AND(Uis Stability)THEN(C_results is High)

[0104] Rule 2:IF(U is Low)AND(Uis Instability)THEN(C_results is Low) ...

[0106] Perform fuzzy reasoning based on fuzzy rules to determine the high bandwidth scheduling order plan;

[0107] It should be noted that the division of fuzzy sets can be adjusted according to actual conditions. For example, although three fuzzy sets are used as examples in this embodiment, they can actually be divided into more than three sets to facilitate better precise adjustment according to images at different angles.

[0108] Furthermore, for the judgment of the maximum probability of the historical scheduling bandwidth and the high, medium and low of the link bandwidth stability index, a threshold can be set according to the actual situation for judgment. For example, when the maximum probability of the historical scheduling bandwidth exceeds 72%, it is marked as "High", and when the link bandwidth stability index is higher than 68%, it is marked as "Stability", etc., which will not be elaborated here;

[0109] The present invention obtains a high-bandwidth scheduling strategy, formulates a set of fuzzy rules for fuzzy reasoning according to the probability of the maximum value of the historical scheduling bandwidth and the link bandwidth stability index, determines the high-bandwidth scheduling order plan, improves the utilization efficiency of bandwidth resources, reduces bandwidth waste, and improves the reliability and stability of high-bandwidth transmission.

[0110] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0111] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented by software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website site, computer, server or data center to another website site, computer, server or data center by wired or wireless (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.

[0112] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0113] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0114] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0115] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0116] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0117] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0118] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage media include: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROM), random access memories (RAM), magnetic disks or optical disks.

[0119] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A low-orbit satellite network scheduling method suitable for unbalanced data flow, characterized in that: include: S1: The client sends a request to the server, and the server responds with resources to the client; S2: The controller samples the traffic distribution index and link historical usage rate during the data transmission phase, initializes the multi-agent deep reinforcement learning model, and the SDN controller collects delay jitter and predicts the reachable efficiency of the node; S3: The controller uses the traffic distribution index, link historical utilization rate, delay jitter, and predicted node reachability efficiency to calculate the scheduling probability through the logistic regression formula, and compares the scheduling probability with the preset scheduling threshold to determine the low-latency scheduling strategy and the high-bandwidth scheduling strategy. S4: Obtain high bandwidth scheduling strategy, collect historical scheduling bandwidth maximum probability and link bandwidth stability index; S5: Obtain the maximum probability of historical scheduling bandwidth and the link bandwidth stability index, and substitute them into fuzzy logic to determine the high-bandwidth scheduling order solution.

2. The low-orbit satellite network scheduling method applicable to unbalanced data flow according to claim 1, characterized in that: The client and the server establish a multipath connection based on MPQUIC. The client sends an initial connection invitation. When the client receives the handshake confirmation message, the handshake is completed and the data transmission phase begins.

3. The low-orbit satellite network scheduling method applicable to unbalanced data flow according to claim 2, characterized in that: The traffic distribution index is obtained by weighted calculation based on the ratio of the average value to the maximum value of the remaining bandwidth, the ratio of the average value to the maximum value of the traffic, and the ratio of the average value to the maximum value of the packet return delay. The ratio of the real-time bandwidth usage of the link to the total bandwidth is collected within the specified time window, and the weighted average of the ratio is calculated to obtain the historical usage rate of the link; Based on the round-trip delay of each data packet, the delay difference between each pair of adjacent data packets is calculated; the average value of the delay differences is taken to obtain the delay jitter; By collecting network topology, link status and traffic data, substituting them into the machine learning model, obtaining the predicted nodes, calculating the similarity between the predicted nodes and the real nodes, and determining the reachable efficiency of the predicted nodes.

4. The low-orbit satellite network scheduling method applicable to unbalanced data flow according to claim 3, characterized in that: Substitute the traffic distribution index, link historical utilization rate, delay jitter, and predicted node reachability efficiency into the logistic regression formula to calculate the scheduling probability. The specific formula is as follows: In the formula, L is the result of logistic regression calculation, that is, the scheduling probability, e is the natural base, and y is the linear combination term of the logistic regression model. Specifically, y can be set as: Where β0 is the bias term, β1, β2, β3 and β4 are the regression coefficients of traffic distribution index, link historical utilization rate, delay jitter and predicted node reachability efficiency, respectively.

5. The low-orbit satellite network scheduling method applicable to unbalanced data flow according to claim 4, characterized in that: Compare the scheduling probability with the preset scheduling threshold. If the scheduling probability is greater than or equal to the scheduling threshold, it means that the current path node is highly schedulable and is suitable for a low-latency scheduling strategy. If the scheduling probability is less than the scheduling threshold, it means that the path node requires a higher bandwidth to process traffic and a high-bandwidth scheduling strategy is selected. When the low-latency scheduling strategy is executed, a path with the least number of hops is selected; When the high-bandwidth scheduling policy is executed, a path with sufficient bandwidth is selected.

6. The low-orbit satellite network scheduling method applicable to unbalanced data flow according to claim 5, characterized in that: The low-latency scheduling strategy is based on network status information. The deep reinforcement learning model continuously adjusts the strategy through interaction with the environment to maximize the cumulative reward, and the scheduling value with the maximum reward is used as the optimal scheduling value under the current state.

7. The low-orbit satellite network scheduling method applicable to unbalanced data flow according to claim 6, characterized in that: The probability of the maximum historical scheduling bandwidth value is obtained by calculating the ratio of the number of times the link is selected as the maximum bandwidth transmission path to the total scheduling number; Based on historical bandwidth data, the ratio of the bandwidth average value to the bandwidth fluctuation range is calculated to obtain the link bandwidth stability index.

8. The low-orbit satellite network scheduling method applicable to unbalanced data flow according to claim 7, characterized in that: The maximum probability of historical scheduling bandwidth and link bandwidth stability index are defined as input variables and divided into different fuzzy sets respectively; The high bandwidth scheduling order result is defined as the output variable and divided into fuzzy sets; Formulate fuzzy rules to describe the impact of the maximum probability of historical scheduling bandwidth and link bandwidth stability index on the high-bandwidth scheduling order results; Fuzzy reasoning is performed based on fuzzy rules to determine the high-bandwidth scheduling order plan.

Images