Deep space network optimal transmission path calculation method

Abstract

The invention discloses a deep space network optimal transmission path calculation method, relates to the technical field of deep space network data transmission path selection, and is used for solving the problem of transmission path selection in a deep space network. The method comprises the following steps; data communication is realized, link delay, packet loss rate and environmental factors are monitored, and an optimal transmission path is selected based on a link quality score; an integrated learning and multi-agent reinforcement learning model is constructed, and the path optimization effect is enhanced; designing a completely decentralized independent learning algorithm to realize autonomous path selection; a multi-objective optimization model is adopted to dynamically balance energy consumption and performance, and an optimal path is selected; collaborative optimization among nodes is performed based on the graph neural network, and the resource utilization rate is improved. According to the method, the characteristics of the deep space environment are creatively combined, link monitoring is fused, efficient and reliable deep space network transmission path planning is realized, and the stability and the resource utilization rate of deep space communication are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of deep space network data transmission path selection, and more specifically, to a method for calculating an optimal transmission path for a deep space network. Background Art

[0002] The deep space network refers to the network between the earth's ground station and the moon and other celestial bodies. Due to its huge development potential, it has been valued by many countries and regions, and the competition among major powers is very fierce, such as the exploration of celestial bodies proposed by NASA. Affected by the harsh network environment of the deep space network (long delay, high packet loss rate, etc.), when the ground station and the spacecraft in the deep space network communicate data, they will face the disconnection of the transmission link and cause transmission failure, so it is necessary to calculate and generate the optimal data transmission path to ensure the stability of data transmission. Therefore, in order to improve the success rate of data transmission in the deep space network, it is necessary to invent a method for calculating the optimal data transmission path suitable for it according to the characteristics of the deep space network.

[0003] In view of the above problems, the present invention proposes a solution. Summary of the invention

[0004] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a method for calculating an optimal transmission path of a deep space network to solve the problems raised in the above-mentioned background technology.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] In a preferred embodiment,

[0007] Step 1: Realize data transmission communication;

[0008] Step 2: Construct ensemble learning to combine multi-agent reinforcement learning;

[0009] Step 3: Design a completely decentralized independent learning algorithm to solve the problem. Each agent performs independent proximal strategy optimization in its own environment and samples.

[0010] Step 4: Introduce the multi-objective trade-off between node energy consumption and performance in path decision;

[0011] Step 5: Realize collaborative optimization between nodes based on graph neural network GNN.

[0012] In a preferred embodiment, an LLDP confirmation frame is sent to a spacecraft in deep space, and the spacecraft returns an LLDP frame after receiving it, and the frame format of the two exchanges;

[0013] On the one hand, the LLDP frame can return the transmission link status, link QoS array, and reachability of the transmitted data, and on the other hand, execute the transmission path generated according to the quality of service;

[0014] Design a mechanism to dynamically optimize paths based on link quality, combine link monitoring and environmental factors with path planning, and optimize dynamic paths;

[0015] The deep space link data is acquired through regular and real-time monitoring using the detection frame loopback method and multi-point monitoring algorithm, including:

[0016] Link delay;

[0017] Jitter;

[0018] Environmental factors data are obtained through regular and real-time monitoring using proton and electron detectors and radiation detectors, including:

[0019] Solar wind intensity: particle flow rate, density;

[0020] Space radiation: radiation levels;

[0021] The deep space link data and environmental factor data obtained are used to calculate the scores of each factor, according to the formula: data score = 100-(current data value / maximum acceptable data value) × 100;

[0022] The link quality score is calculated according to the scores of each factor, specifically according to the formula: LQS = Qy × yd + Qd × dd + Qt × td + Qk × kd, where LQS represents the link quality score, yd represents the link delay score, dd represents the jitter score, td represents the solar wind intensity score, kd represents the space radiation score, Qy represents the weight of the link delay, Qd represents the weight of the jitter, Qt represents the weight of the solar wind intensity, and Qk represents the weight of the space radiation;

[0023] A link score threshold YL is set. When the link quality score LQS is greater than or equal to YL, this link is preferred as the main path. When the link quality score LQS is less than YL, this link is used as an alternative path.

[0024] In a preferred embodiment, the transmission data is sampled and a training set is formed.

[0025] D1={(x1,y1,z1),(x2,y2,z2),…,(x n ,y n ,z n )}

[0026] Among them, x is a certain type of feature vector formed by the transmitted data; y is its label value; z is the decision conclusion; the entire deep space network environment is defined as MRJ = {O, S, A, R, P, J, L}, where J is the joint relationship with ensemble learning; O is the total number of agents; S is the state set of all decision-making agents S = s1×s2×…×s O ; A is the set of actions of all decision-making agents, A = a1×a2×…×a O ; R is the reward function set of all decision-making agents, R = r1×r2×…×r O ; P is the state transition probability of the environment; L is the final decision; the initial model is F(x,MRJ);

[0027] The mean square error is defined as the loss function; the smaller its value, the more accurate the joint model is, and vice versa;

[0028]

[0029] Where n is the total number of samples; w i and is the true value and predicted value of the i-th sample; is the prediction error of the ith sample.

[0030] In a preferred embodiment, for any agent o∈O, the strategy π of each environment is initialized o With the value function V o ;

[0031] Within the iteration number ξ∈[0,1,2,…], execute S4-3 to S4-5;

[0032] For any agent o∈O, obtain its trajectory data;

[0033] For any agent o∈O, calculate its value function V o ;

[0034] For any agent o∈O, apply the goal of maximizing PPO to update the corresponding strategy π o ;

[0035] For any agent o∈O, apply the loss function MSE(y,F(x)) to optimize the improved value function V o ;

[0036] ξ is the number of iterations;

[0037] End the iteration and summarize the decisions.

[0038] In a preferred embodiment, an energy-aware path planning mechanism is designed: the energy consumption of each node, including: communication power, processing power, node residual energy, is used as an optimization variable combined with the performance of each node, including: link delay, packet loss rate optimization decision;

[0039] Get node energy consumption and performance data

[0040] Node energy consumption data includes communication power, processing power, and node remaining energy. Specifically, the round-trip time between nodes is measured by sending a detection frame through RTT, and the communication distance is calculated according to the formula: Ptxgl = Pj + k × d^2, where Pj represents the basic power consumption, d represents the communication distance between nodes, and k is the signal attenuation factor. The number and size of data packets received and sent by the node are monitored in real time to calculate the processing volume per unit time. The processing power factor c of the node data processing is determined using a hardware current sensor. Specifically, the formula: Pclgl = c × Dp, where Dp is the data processing volume of the current node. The remaining power is read through the battery management system, and the remaining energy of the node is reported using a low-power wireless signal.

[0041] Node performance data includes link delay and packet loss rate. Use the loopback frame method to send detection frames with timestamps and calculate the time difference of frame returns. Use the data packet log system at both ends of the link to record the number of sent and received packets. The specific formula is: DBL = number of lost packets / number of sent packets, where DBL represents the packet loss rate.

[0042] The total energy consumption of the node is calculated based on the obtained node energy consumption data. The specific formula is: E = Et + Ec + jsy, where E represents the total energy consumption of the node, Et is the calculation formula for communication power consumption: communication power × transmission time on the path, Ec is the calculation formula for processing power consumption: processing power × data processing time, and jsy is the remaining energy of the node.

[0043] The performance score of the node is calculated based on the obtained node performance data. The specific formula is: P = (SYS + DBLS) / 2, where P represents the performance score of the node. SYS represents the link delay score calculation formula: 100-((current link delay / maximum link delay) × 100), and DBLS represents the packet loss rate score calculation formula: 100-((current packet loss rate / maximum packet loss rate) × 100)

[0044] The objective function is used to combine the node energy consumption and performance data to form a multi-objective optimization problem, and a threshold Yh is set. When the path objective function value is less than Yh, the path is marked as a better path. When the path objective function value is greater than or equal to Yh, the path is marked as an alternative path. The specific form of the objective function is: F = Q1 × E + Q2 × (100-P), where F represents the objective function, Q1 represents the weight of the total energy consumption of the node, and Q2 represents the weight of the node performance.

[0045] In a preferred embodiment, NetworkX is used to implement graph modeling, and the deep space network is modeled as an undirected weighted graph G = (JD, BE), where JD represents a node set: represents a spacecraft node, and the characteristics of each node include residual energy, communication power, and processing power; BE represents an edge set: represents a communication link between nodes, and the characteristics include delay and packet loss rate. The connection relationship between the nodes is represented by the adjacency matrix A, and the node feature matrix X and the edge feature matrix EJ are used to represent the specific state;

[0046] The GNN model is constructed using the PyTorch Geometric framework. The node feature matrix X and the adjacency matrix A are input, and the edge feature matrix EJ is used to assist in building a more accurate transmission relationship. In each layer, the node's own features and the features of the neighboring nodes are aggregated through the edge features, and the state of each node is updated according to the formula: H(i+1)=a×(A×H(i)×W(i)), where H(i) represents the node features of the i-th layer, W(i) represents the weight matrix, and a represents the activation function. After multi-layer convolution aggregation, the final state of each node is output for path selection and collaborative optimization.

[0047] The simulated deep space network state includes node features and link features. Through supervised learning, the model is trained with the known optimal path as the label. The trained GNN model is used for reasoning, and the node embedding vector ZX is output. The data transmission path is determined according to its value.

[0048] The present invention discloses a method for calculating the optimal transmission path of a deep space network, which relates to the technical field of deep space network data transmission path selection, and is used to solve the transmission path selection problem in a deep space network; the method includes the following steps: realizing data communication, monitoring link delay, packet loss rate and environmental factors, and selecting the optimal transmission path based on link quality score; constructing an integrated learning and multi-agent reinforcement learning model to enhance the path optimization effect; designing a completely decentralized independent learning algorithm to realize autonomous path selection; using a multi-objective optimization model to dynamically balance between energy consumption and performance and select the optimal path; and performing collaborative optimization between nodes based on a graph neural network to improve resource utilization. The present invention innovatively combines the characteristics of the deep space environment, integrates link monitoring, realizes efficient and reliable deep space network transmission path planning, and improves the stability and resource utilization of deep space communications. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 This is a scene diagram of a method for calculating an optimal transmission path for a deep space network according to the present invention.

[0050] Figure 2 This is an integrated learning and multi-agent reinforcement learning solution method proposed by the present invention for the deep space network environment.

[0051] Figure 3 The present invention is a schematic diagram of the operation of a method for calculating an optimal transmission path for a deep space network. DETAILED DESCRIPTION

[0052] 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.

[0053] Example

[0054] The present invention discloses a method for calculating an optimal transmission path of a deep space network, and the specific steps are as follows:

[0055] Step 1: Implement communication in data transmission.

[0056] Specifically, an LLDP confirmation frame is sent to a spacecraft in deep space, and the spacecraft returns an LLDP frame after receiving it. The frame format of the two exchanges;

[0057] On the one hand, the LLDP frame can return the transmission link status, link QoS array, and reachability of the transmitted data, and on the other hand, execute the transmission path generated according to the quality of service. Data transmission communication is realized. For example: a communication request is initiated through the ground station, and the ground station sends the request signal to the deep space satellite network. Satellite number S-001 in the satellite network receives the signal through its radio frequency, identifies the target as the spacecraft M-Explorer, and forwards the signal to the next-level satellite S-005. Finally, S-005 passes the signal to the spacecraft M-Explorer, confirms its online status, and sends an LLDP confirmation frame to the spacecraft. The frame format includes: source address, destination address (spacecraft), and link status monitoring field. After receiving the confirmation frame, the spacecraft M-Explorer returns a feedback frame containing the following information:

[0058] Current link delay: 350ms.

[0059] Data packet loss rate: 2%.

[0060] Available bandwidth: 50Mbps.

[0061] It should be noted that in deep space scenarios, the long delay, high jitter and uncertain environmental factors of the link, including space radiation and solar wind, make traditional protocols difficult to apply. In this embodiment, a mechanism for dynamically optimizing the path based on link quality is designed to combine link monitoring and environmental factors with path planning. Through real-time link quality monitoring, the dynamic path is optimized based on the influence of environmental factors, overcoming the technical difficulties of link fluctuations and instability in deep space environments, and significantly improving the reliability of link transmission.

[0062] The deep space link data is acquired through regular and real-time monitoring using the detection frame loopback method and multipoint monitoring algorithm, including:

[0063] Link delay (Delay): round trip time (unit: ms);

[0064] Jitter: The variation in the time interval between data packets (unit: ms);

[0065] Environmental factor data are obtained through regular and real-time monitoring using proton and electron detectors (Solar Wind Analyzer, SWA) and radiation detectors (Radiation Assessment Detector, RAD), including:

[0066] Solar wind intensity: particle velocity (unit: km / s), density (unit: particles / cm3);

[0067] Space radiation: radiation level (unit: Gy / h);

[0068] The link quality score is calculated by acquiring the deep space link data and environmental factor data. First, the score of each data is calculated according to the formula: data score = 100-(current data value / maximum acceptable data value)×100. For example, the score of the link delay data is calculated according to the formula, link delay score = 100-(current link delay value / maximum acceptable link delay value)×100.

[0069] The link quality score is calculated based on the scores of each data, specifically according to the formula: LQS = Qy × yd + Qd × dd + Qt × td + Qk × kd, where LQS represents the link quality score, yd represents the link delay score, dd represents the jitter score, td represents the solar wind intensity score, kd represents the space radiation score, Qy represents the weight of the link delay, Qd represents the weight of the jitter, Qt represents the weight of the solar wind intensity, and Qk represents the weight of the space radiation. For example: Qy is calculated to be 0.4, Qd is 0.3, Qt is 0.2, and Qk is 0.1. The obtained data is as follows:

[0070] Current delay value = 350ms; maximum acceptable delay = 1000ms.

[0071] Current jitter value = 5ms; maximum acceptable jitter = 50ms.

[0072] Current solar wind intensity = 600km / s; maximum solar wind threshold = 1000km / s.

[0073] Current radiation intensity = 0.8 Gy / h; maximum radiation threshold = 5 Gy / h.

[0074] Substituting the example data into the formula, the link quality score LQS is calculated to be 69.4. In this embodiment, the link score threshold YL is set, and the link selection is divided into preferred links and available links. When the link quality score LQS is greater than or equal to YL, this link is preferred as the main path, and when the link quality score LQS is less than YL, this link is used as an alternative path; for example: setting the threshold YL to 80, when the link quality score LQS is greater than or equal to 80, this link is preferred as the main path, and when the link quality score LQS is less than 80, this link is used as an alternative path; through the example link quality score is 69.4 less than 80, this link is used as an alternative path.

[0075] Step 2: Construct ensemble learning and joint multi-agent reinforcement learning to enhance the reinforcement learning decision conclusions by combining weak prediction models through multiple iterations.

[0076] Sample the transmission data and form a training set

[0077] D1={(x1,y1,z1),(x2,y2,z2),…,(x n ,y n ,z n )}

[0078] Among them, x is a certain type of feature vector formed by the transmitted data. y is its label value. z is the decision conclusion. The entire deep space network environment is defined as: MRJ = {O, S, A, R, P, J, L}, where J is the joint relationship with ensemble learning. O is the total number of agents. S is the state set of all decision-making agents S = s1×s2×…×s O A is the set of actions of all decision-making agents, and A=a1×a2×…×a O R is the reward function set of all decision-making agents, R = r1×r2×…×r O . P is the state transition probability of the environment. L is the final decision. The initial model is F(x,MRJ).

[0079] Mean Squared Error (MSE) is defined as a loss function. The smaller its value is, the more accurate the joint model of the present invention is, and vice versa, the greater the error is.

[0080]

[0081] Where n is the total number of samples. i and is the true value and predicted value of the i-th sample. is the prediction error of the ith sample.

[0082] All participating agents jointly integrate the learning algorithm through a certain strategy π o To maximize the overall reward and aggregate the respective conclusions to form the final conclusion.

[0083] Step 3: Design a completely decentralized independent learning algorithm to solve the problem. Each agent performs separate proximal strategy optimization in its own environment and samples.

[0084] Here are the steps:

[0085] S3-1 For any agent o∈O, initialize the strategy π of each environment o With the value function V o .

[0086] S3-2 executes S4-3 to S4-5 within the number of iterations ξ∈[0,1,2,…].

[0087] S3-3 For any agent o∈O, obtain its trajectory data.

[0088] S3-4 For any agent o∈O, calculate its value function V i .

[0089] S3-5 For any agent o∈O, update the corresponding strategy π by maximizing the PPO goal o .

[0090] S3-5 For any agent o∈O, apply the loss function MSE(y,F(x)) to optimize the improved value function V o .

[0091] S3-6 ξ is the number of iterations.

[0092] S3-7 ends the iteration and summarizes the decisions.

[0093] Step 4: Considering the energy constraints of spacecraft and the problem of node collaboration in deep space networks, a multi-objective trade-off between energy consumption and performance is introduced in path decision making, and the optimal transmission path is selected by optimizing the multi-objective function.

[0094] In this embodiment, an energy-aware path planning mechanism is designed: the energy consumption of each node, including communication power, processing power, and node residual energy, is used as an optimization variable combined with the performance of each node, including link delay and packet loss rate optimization decision;

[0095] The specific steps are as follows:

[0096] Step 4-1: Obtain node energy consumption and performance data

[0097] Node energy consumption data includes communication power, processing power, and node remaining energy; specifically, the round-trip time between nodes is measured by sending a detection frame through RTT (RoundTrip Time), so as to calculate the communication distance, according to the formula: Ptxgl = Pj + k × d^2, where Pj represents the basic power consumption, reflecting the minimum energy consumption of the node to maintain communication, d represents the communication distance between nodes, and k is the signal attenuation factor, which is calibrated according to the node communication power test data; a real-time traffic monitoring module is embedded in the node to record the processing rate of the transmitted data, and a hardware current sensor is used to determine the processing power factor c during node data processing, according to the formula: Pclgl = c × Dp, where Dp is the data processing capacity of the current node; the remaining power is read through the battery management system (BMS), and the remaining energy of the node is reported using a low-power wireless signal. For example: the following instance data is obtained:

[0098] Node 1:

[0099] Remaining energy: 80%

[0100] Data processing power consumption: 2W

[0101] Communication power consumption: 5W

[0102] Node 2:

[0103] Remaining energy: 75%

[0104] Data processing power consumption: 1.5W

[0105] Communication power consumption: 4W

[0106] Node 3:

[0107] Remaining energy: 60%

[0108] Data processing power consumption: 1.2W

[0109] Communication power consumption: 6W

[0110] Node 4:

[0111] Remaining energy: 50%

[0112] Data processing power consumption: 1.5W

[0113] Communication power consumption: 5W

[0114] Node 5:

[0115] Remaining energy: 85%

[0116] Data processing power consumption: 3W

[0117] Communication power consumption: 7W

[0118] Node performance data includes link delay and packet loss rate. Specifically, use the Loopback Frame Method to send a detection frame with a timestamp and calculate the time difference of the frame return. Use the data packet log system at both ends of the link to record the number of sent and received packets. Synchronize log data in real time and calculate the packet loss rate. Specifically based on the formula: DBL = number of lost packets / number of sent packets, where DBL represents the packet loss rate. For example, the following instance data is obtained:

[0119] Node 1 to Node 2: delay 100ms, packet loss rate 1%,

[0120] Node 1 to Node 3: delay 200ms, packet loss rate 2%,

[0121] Node 2 to Node 4: delay 300ms, packet loss rate 3%,

[0122] Node 3 to Node 5: delay 250ms, packet loss rate 1.5%,

[0123] Node 4 to Node 5: delay 400ms, packet loss rate 4%,

[0124] It should be noted that the acquisition of the above-mentioned node energy consumption and performance data belongs to conventional technical means and will not be elaborated here.

[0125] Step 4-2: Data normalization

[0126] Linear normalization is used to map the energy consumption data and performance data of each node to the interval [0,1], the formula is: Where Normalized_Score is the normalized value.

[0127] Step 4-3: Establish a multi-objective optimization model

[0128] The total energy consumption of the node is calculated according to the obtained node energy consumption data, specifically based on the formula: E=Et+Ec+jsy, where E represents the total energy consumption of the node, Et is the calculation formula for communication power consumption: communication power × transmission time on the path, Ec is the calculation formula for processing power consumption: processing power × data processing time, jsy is the remaining energy of the node, for example: according to the obtained data, node 1 to node 2: the communication power consumption is 5W, the processing power consumption is 2W, and the remaining energy of the node is 75%. By calculation, the transmission time on the path is 0.1h, and the processing time is 0.05h. Substituting the example data into the formula, the total node power consumption E is calculated to be 0.6Wh.

[0129] The performance score of the node is calculated according to the obtained node performance data, specifically according to the formula: P = (SYS + DBLS) / 2, where P represents the performance score of the node, wherein SYS represents the link delay score calculation formula: 100-((current link delay / maximum link delay) × 100), and DBLS represents the packet loss rate score calculation formula: 100-((current packet loss rate / maximum packet loss rate) × 100). For example: according to the obtained data: the current link delay is 100ms, the maximum delay is 500ms, the current packet loss rate is 2%, and the maximum packet loss rate is 10%. Substituting the example data into the formula, the node performance score P is calculated to be 80.

[0130] Use the objective function to integrate the node energy consumption and performance data to form a multi-objective optimization problem, and set the threshold Yh. When the path objective function value is less than Yh, mark this path as a better path. When the path objective function value is greater than or equal to Yh, mark this path as an alternative path. The form of the objective function is specifically based on: F = Q1 × E + Q2 × (100-P), where F represents the objective function, Q1 represents the weight of the total energy consumption of the node, and Q2 represents the weight of the node performance. Use the weighted sum method to transform the multi-objective optimization into a single-objective optimization problem, dynamically adjust the values ​​of Q1 and Q2, and adjust the weight distribution according to the real-time energy consumption status of the node. For example: The sample data obtained is as follows:

[0131] Path 1:

[0132] Total node energy consumption: E = 3.97Wh;

[0133] Node performance score: P = 85;

[0134] Path 2:

[0135] Total node energy consumption: E = 2.9Wh;

[0136] Node performance score: P = 75;

[0137] By calculation, Q1 is 0.6, Q2 is 0.4, and the example data is substituted into the formula to calculate the objective function F1=8.38 of path 1 and the objective function F2=11.74 of path 2. In this embodiment, the threshold Yh is set to 10. When the path objective function value is less than 10, the path is marked as a better path, and when the path objective function is greater than or equal to 10, the path is marked as an alternative path; according to the example data, the objective function F1=8.38 of path 1 is less than 10, and path 1 is marked as a better path. The objective function F2=11.74 of path 2 is greater than 10, and the path is marked as an alternative path.

[0138] Step 5: Based on the Graph Neural Network (GNN), realize the collaborative optimization between nodes, improve the resource utilization of nodes in the deep space network, and maintain the global stability and task completion rate of the network under the conditions of link limitation or node limitation;

[0139] Step 5-1: Use NetworkX to implement graph modeling and model the deep space network as an undirected weighted graph G = (JD, BE), where JD represents the node set: represents the spacecraft node, and the characteristics of each node include residual energy, communication power, and processing power. BE represents the edge set: represents the communication link between nodes, and the characteristics include latency and packet loss rate. Use the adjacency matrix A to represent the connection relationship between nodes, and use the node feature matrix X and the edge feature matrix EJ to represent the specific status. For example: the following sample data:

[0140] The network consists of 4 spacecraft, and the node characteristics are as follows:

[0141] A: Remaining energy 60%, communication power 4W, processing power 3W.

[0142] B: Remaining energy 50%, communication power 3W, processing power 2W.

[0143] Edge features:

[0144] (A→B): delay 20ms, packet loss rate 0.1.

[0145] Step 5-2: Use the PyTorch Geometric framework to build a GNN model, input the node feature matrix X and the adjacency matrix A, represent the initial state of the network, and use the edge feature EJ to assist in building a more accurate transmission relationship. In each layer, the node's own features and the features of neighboring nodes are aggregated through edge features, and the state of each node is updated according to the formula: H(i+1)=a×(A×H(i)×W(i)), where H(i) represents the node features of the i-th layer, W(i) represents the weight matrix, and a represents the activation function. After multi-layer convolution aggregation, the final state Z of each node is output for path selection and collaborative optimization.

[0146] Step 5-3: Simulate the generated deep space network state (including node features and link features). Train the model using the known optimal path as a label through supervised learning.

[0147] Step 5-4: Use the trained GNN model for inference, output the node embedding vector ZX, and determine the data transmission path based on its value.

[0148] 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.

[0149] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented by software, the above embodiments may be implemented in whole or in part in the form of a computer program product.

[0150] Those of ordinary skill in the art will appreciate that the modules 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 of the technical solution and the invention constraints. 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.

[0151] In addition, each functional module in each embodiment of the present application may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0152] 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.

[0153] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for calculating the optimal transmission path of deep space network, It is characterized by; including: Step 1: Realize data transmission communication; Step 2: Construct ensemble learning to combine multi-agent reinforcement learning; Step 3: Design a completely decentralized independent learning algorithm to solve the problem. Each agent performs independent proximal strategy optimization in its own environment and samples. Step 4: Introduce the multi-objective trade-off between node energy consumption and performance in path decision to optimize the multi-objective function; Step 5: Realize collaborative optimization between nodes based on graph neural network GNN.

2. The method for calculating the optimal transmission path of a deep space network according to claim 1, characterized in that: In step 1, an LLDP confirmation frame is sent to a spacecraft in deep space. After receiving the frame, the spacecraft returns an LLDP frame. The frame format of the two exchanges is as follows; On one hand, the LLDP frame can return the transmission link status, link QoS array, and reachability of the transmitted data, and on the other hand, it can execute the transmission path generated according to the quality of service; Design a mechanism to dynamically optimize paths based on link quality, combine link monitoring and environmental factors with path planning, and optimize dynamic paths; The deep space link data is acquired through regular and real-time monitoring using the detection frame loopback method and multi-point monitoring algorithm, including: Link delay; Jitter; Environmental factors data are obtained through regular and real-time monitoring using proton and electron detectors and radiation detectors, including: Solar wind intensity: particle flow rate, density; Space radiation: radiation levels; The deep space link data and environmental factor data obtained are used to calculate the scores of each factor, according to the formula: data score = 100-(current data value / maximum acceptable data value) × 100; The link quality score is calculated according to the scores of each factor, specifically according to the formula: LQS = Qy × yd + Qd × dd + Qt × td + Qk × kd, where LQS represents the link quality score, yd represents the link delay score, dd represents the jitter score, td represents the solar wind intensity score, kd represents the space radiation score, Qy represents the weight of the link delay, Qd represents the weight of the jitter, Qt represents the weight of the solar wind intensity, and Qk represents the weight of the space radiation; A link score threshold YL is set. When the link quality score LQS is greater than or equal to YL, this link is selected as the main path. When the link quality score LQS is less than YL, this link is selected as an alternative path to achieve data transmission communication.

3. The method for calculating the optimal transmission path of a deep space network according to claim 2, characterized in that: In step 2, the transmission data is sampled and formed into a training set D1={(x1,y1,z1),(x2,y2,z2),…,(x n ,y n ,z n )} Among them, x is a certain type of feature vector formed by the transmitted data; y is its label value; z is the decision conclusion; the entire deep space network environment is defined as MRJ = {O, S, A, R, P, J, L}, where J is the joint relationship with ensemble learning; O is the total number of agents; S is the state set of all decision-making agents S = s1×s2×…×s O ; A is the set of actions of all decision-making agents, A = a1×a2×…×a O ; R is the reward function set of all decision-making agents, R = r1×r2×…×r O ; P is the state transition probability of the environment; L is the final decision; the initial model is F(x,MRJ); The mean square error is defined as the loss function; the smaller its value, the more accurate the joint model is, and vice versa; Where n is the total number of samples; w i and is the true value and predicted value of the i-th sample; is the prediction error of the ith sample.

4. A method for calculating an optimal transmission path for a deep space network according to claim 3, characterized in that; In step 3, for any agent o∈O, initialize the strategy π in each environment o With the value function V o ; Within the iteration number ξ∈[0,1,2,…], execute S4-3 to S4-5; For any agent o∈O, obtain its trajectory data; For any agent o∈O, calculate its value function V o ; For any agent o∈O, apply the goal of maximizing PPO to update the corresponding strategy π o ; For any agent o∈O, apply the loss function MSE(y,F(x)) to optimize the improved value function V o ; Compute pseudo residual in the direction of negative gradient ξ is the number of iterations; End the iteration and summarize the decisions.

5. A method for calculating an optimal transmission path for a deep space network according to claim 4, characterized in that: In step 4, an energy-aware path planning mechanism is designed: the energy consumption of each node, including communication power, processing power, and node residual energy, is used as an optimization variable combined with the performance of each node, including link delay and packet loss rate optimization decision; Get node energy consumption and performance data Node energy consumption data includes communication power, processing power, and node remaining energy. Specifically, the round-trip time between nodes is measured by sending a detection frame through RTT, and the communication distance is calculated according to the formula: Ptxgl = Pj + k × d^2, where Pj represents the basic power consumption, d represents the communication distance between nodes, and k is the signal attenuation factor. The number and size of data packets received and sent by the node are monitored in real time to calculate the processing volume per unit time. The processing power factor c of the node data processing is determined using a hardware current sensor. Specifically, the formula: Pclgl = c × Dp, where Dp is the data processing volume of the current node. The remaining power is read through the battery management system, and the remaining energy of the node is reported using a low-power wireless signal. Node performance data includes link delay and packet loss rate. Use the loopback frame method to send detection frames with timestamps and calculate the time difference of frame returns. Use the data packet log system at both ends of the link to record the number of sent and received packets. The specific formula is: DBL = number of lost packets / number of sent packets, where DBL represents the packet loss rate. The total energy consumption of the node is calculated based on the obtained node energy consumption data. The specific formula is: E = Et + Ec + jsy, where E represents the total energy consumption of the node, Et is the calculation formula for communication power consumption: communication power × transmission time on the path, Ec is the calculation formula for processing power consumption: processing power × data processing time, and jsy is the remaining energy of the node. The performance score of the node is calculated based on the obtained node performance data. The specific formula is: P = (SYS + DBLS) / 2, where P represents the performance score of the node. SYS represents the link delay score calculation formula: 100-((current link delay / maximum link delay) × 100), and DBLS represents the packet loss rate score calculation formula: 100-((current packet loss rate / maximum packet loss rate) × 100) The objective function is used to combine the node energy consumption and performance data to form a multi-objective optimization problem, and a threshold Yh is set. When the path objective function value is less than Yh, the path is marked as a better path. When the path objective function value is greater than or equal to Yh, the path is marked as an alternative path. The specific form of the objective function is: F = Q1 × E + Q2 × (100-P), where F represents the objective function, Q1 represents the weight of the total energy consumption of the node, and Q2 represents the weight of the node performance.

6. A method for calculating an optimal transmission path for a deep space network according to claim 5, characterized in that: In step 5, NetworkX is used to implement graph modeling, and the deep space network is modeled as an undirected weighted graph G = (JD, BE), where JD represents the node set: represents the spacecraft node, and the characteristics of each node include residual energy, communication power, and processing power. BE represents the edge set: represents the communication link between nodes, and the characteristics include delay and packet loss rate. The adjacency matrix A is used to represent the connection relationship between nodes, and the node feature matrix X and the edge feature matrix EJ are used to represent the specific status; The GNN model is constructed using the PyTorch Geometric framework. The node feature matrix X and the adjacency matrix A are input, and the edge feature matrix EJ is used to assist in building a more accurate transmission relationship. In each layer, the node's own features and the features of the neighboring nodes are aggregated through the edge features, and the state of each node is updated according to the formula: H(i+1)=a×(A×H(i)×W(i)), where H(i) represents the node features of the i-th layer, W(i) represents the weight matrix, and a represents the activation function. After multi-layer convolution aggregation, the final state of each node is output for path selection and collaborative optimization. The simulated deep space network state includes node features and link features. Through supervised learning, the model is trained with the known optimal path as the label. The trained GNN model is used for reasoning and the node embedding vector ZX is output. The data transmission path is determined according to its value.

Images