An automatic satellite remote control command generation system based on artificial intelligence

Abstract

This invention discloses an artificial intelligence-based automatic satellite remote control command generation system, belonging to the field of satellite remote control command generation technology. The dynamic value assessment module of this invention integrates multi-source heterogeneous data, breaking through the rigidity of traditional fixed priority strategies and realizing real-time dynamic trade-off of mission value; the conflict resolution module transforms inter-satellite resource competition into a graph structure optimization problem, effectively solving the implicit command mutual interference phenomenon in multi-satellite collaborative observation; the knowledge graph library constructs cross-model influence relationship chains, eliminating the implicit conflict risk of heterogeneous satellite command templates; the physical rule engine embeds dynamic constraints into the decision-making process to prevent the generation of dangerous commands that violate the physical laws of spacecraft.

Description

Technical Field

[0001] This invention relates to the field of satellite remote control command generation technology, and in particular to an automatic satellite remote control command generation system based on artificial intelligence. Background Technology

[0002] With the rapid expansion of low-Earth orbit satellite constellations, multi-satellite collaborative observation has become a core mission mode for scenarios such as ocean monitoring and disaster early warning. Current mainstream systems adopt a distributed architecture, with each satellite independently generating command sequences and achieving basic collaboration through inter-satellite links. New-generation remote sensing constellations generally introduce dynamic priority scheduling algorithms to allocate observation tasks based on the real-time position and energy status of satellites. However, such systems are difficult to cope with global conflicts that emerge in complex scenarios at the constellation level.

[0003] To mitigate multi-satellite command conflicts, recent research has focused on strategies combining online optimization and rule engines. Some systems employ graph neural networks to model inter-satellite influences and generate conflict prediction models through offline training. Other approaches introduce a lightweight game theory framework to pre-compute resource occupancy matrices during command generation. While these methods can effectively identify explicit conflicts, they are not effective in identifying implicit problems such as command cannibalization. When multiple satellites respond to the same emergency event, the system may simultaneously generate a large number of homogeneous commands, leading to wasted observation resources and energy overload.

[0004] Some systems have introduced digital twin technology for instruction rehearsal, but the simulation granularity is mostly limited to the physical device layer and lacks in-depth modeling of the task logic layer. Existing conflict resolution algorithms mostly rely on manually defined rule bases, and when faced with sudden multi-satellite coordination needs, ground stations still need to manually select strategies and adjust parameters. Summary of the Invention

[0005] In view of the aforementioned existing problems, the present invention is proposed.

[0006] This invention provides an artificial intelligence-based automatic satellite remote control command generation system to solve the problems of existing systems relying on static rules and manual intervention, which make it difficult to cope with dynamic command conflicts and heterogeneous equipment adaptation issues in multi-satellite collaborative observation, thus restricting the expansion of constellation scale.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0008] This invention provides an artificial intelligence-based automatic satellite remote control command generation system, which includes:

[0009] The dynamic value assessment module is used to receive real-time satellite status data and mission requirement parameters, and generate a multi-satellite mission priority sequence.

[0010] The conflict resolution module is connected to the dynamic value assessment module and eliminates command conflicts based on the inter-satellite resource occupancy relationship graph.

[0011] The spatiotemporal prediction module analyzes the spatial environment monitoring data stream in real time and predicts abnormal events;

[0012] A knowledge graph database stores the influence relationship chains between heterogeneous satellite subsystems;

[0013] The physics rules engine embeds satellite dynamics constraints;

[0014] Federated learning architecture enables secure updates of multi-node model parameters.

[0015] As a preferred embodiment of the artificial intelligence-based automatic satellite remote control command generation system described in this invention, the dynamic value assessment module includes:

[0016] Satellite remaining power sensor interface, receives battery voltage and temperature data;

[0017] The task timeliness calculation unit calculates the observation urgency coefficient based on the spatiotemporal coordinates of the target event;

[0018] The multi-head attention computing layer performs feature fusion on the energy status, orbital phase, and historical imaging quality of at least three satellites, and outputs dynamic priority weights.

[0019] As a preferred embodiment of the artificial intelligence-based automatic satellite remote control command generation system described in this invention, the operation of the conflict resolution module includes:

[0020] Construct an inter-satellite resource occupancy graph, where nodes represent satellite observation missions and edge weights characterize the overlap of communication frequency bands and the conflict value of attitude adjustment range;

[0021] A graph attention network is used to iteratively update the node feature vectors and generate the optimal task allocation scheme.

[0022] When a risk of energy overload is detected, a power management sub-instruction sequence is automatically inserted.

[0023] As a preferred embodiment of the artificial intelligence-based automatic satellite remote control command generation system of the present invention, the step of constructing the inter-satellite resource occupancy relationship graph in the conflict resolution module includes:

[0024] Each satellite observation mission is mapped to a set of nodes N, and an undirected weighted graph is constructed:

[0025] G = (N, E, W)

[0026] Where N represents the set of all observation task nodes, i,j∈N represent the i-th and j-th tasks respectively, E represents the set of edges between nodes, (i,j)∈E when there is a conflict between tasks i and j, and W is the edge weight function, defined as W(i,j)=C ij , where C ij This represents the combined conflict value between the i-th and j-th tasks;

[0027] The conflict values ​​are linearly weighted using the following formula:

[0028] C ij =αB ij +βA ij ,

[0029] Among them, C ij This represents the overall conflict value, where α and β are normalized weighting coefficients, satisfying α + β = 1, B ij Indicates the overlap of communication frequency bands, A ij Indicates the conflict value within the attitude adjustment range;

[0030] Communication band overlap is quantified as follows:

[0031] in, These are the lower and upper limits of the communication frequency band for the i-th satellite, respectively. These are the lower and upper limits of the communication frequency band for the j-th satellite, respectively;

[0032] The attitude adjustment range conflict value is quantified as follows:

[0033]

[0034] Among them, c i ,c j , i and j respectively, are the attitude adjustment angles required for the i-th and j-th satellites to perform the observation mission, and Z is the maximum attitude adjustment angle range preset by the system, used for normalization processing.

[0035] As a preferred embodiment of the artificial intelligence-based automatic satellite remote control command generation system described in this invention, the spatiotemporal prediction module includes:

[0036] Causal convolutional neural network layer, inputting spatial particle flux and geomagnetic index time series data streams;

[0037] A sliding time window processor outputs the probability of abnormal events occurring within the next 5-10 minutes.

[0038] The pre-compiled instruction cache dynamically loads emergency instruction templates based on predicted probabilities.

[0039] In a preferred embodiment of the artificial intelligence-based automatic satellite remote control command generation system described in this invention, the method for constructing the knowledge graph database includes:

[0040] Parse the command-response pairs in the satellite interface control document;

[0041] A random walk algorithm was used to uncover hidden impact paths across satellite models;

[0042] The confidence level and minimum safe interval parameters for storing instruction conflict chains.

[0043] As a preferred embodiment of the artificial intelligence-based automatic satellite remote control command generation system described in this invention, the knowledge graph database represents heterogeneous satellite subsystems and their influence chains as a weighted graph G. K =(V,E) K W K ), where W K Let w be the edge weight function, and w ij =W K (i,j), and based on the Random Walk Restart Algorithm (RWR), the implicit influence paths between satellite models are mined. The specific process includes: constructing a node transition probability matrix P on a weighted graph, where each element is obtained by normalizing the edge weights. Let P be the node transition probability matrix on the graph, and its elements are P ij :

[0044]

[0045] Among them, P ij w represents the probability of transitioning from node i to node j. ij This represents the edge weight between nodes i and j. This represents the set of all nodes adjacent to node i, where k is the index of the adjacent node;

[0046] Initiate a random walk from the source node s, and update the node influence probability vector in each iteration as follows:

[0047] p (t+1) =(1-γ)P T p (t) +γe s ,

[0048] Where, p (t) Let P represent the node influence probability distribution vector in the t-th iteration, where γ represents the restart probability, satisfying 0 < γ < 1. T Denotes the transpose of the transition matrix, e s Let t represent the standard basis vector whose corresponding component is 1 and all other components are 0 at the source node s; t represents the number of iterations and s represents the starting node of the walk.

[0049] When the probability vector converges, the following conditions are met:

[0050] |p (t+1) -p (t) |1<∈,

[0051] Where |·|1 represents the L1 norm, and ∈ represents the preset convergence threshold;

[0052] After convergence, nodes with probabilities exceeding the threshold τ are selected based on the steady-state distribution to obtain the set of key influencing nodes:

[0053]

[0054] in, p represents the set of key nodes selected. i τ represents the component of the i-th node in the steady-state probability vector, and τ represents the preset probability threshold.

[0055] As a preferred embodiment of the artificial intelligence-based automatic satellite remote control command generation system described in this invention, the physical rule engine includes:

[0056] The orbital mechanics constraint unit verifies whether the command parameters conform to the current orbital phase.

[0057] Thermal control boundary check unit compares command-triggered actions with equipment temperature thresholds;

[0058] Differentiable verification layers transform satellite dynamic equations into neural network regularization terms.

[0059] As a preferred embodiment of the artificial intelligence-based automatic satellite remote control command generation system described in this invention, the operation of the federated learning architecture includes:

[0060] During local model training, a differential privacy algorithm is used to add Gaussian noise;

[0061] During the parameter aggregation phase, the model hash value is verified through a blockchain smart contract.

[0062] Before deploying the model, perform adversarial example stress tests in the digital twin environment.

[0063] As a preferred embodiment of the artificial intelligence-based automatic satellite remote control command generation system described in this invention, the federated learning architecture incorporates a Gaussian mechanism to update the gradient vector g of each client's local model. i The steps for trimming and injecting noise include:

[0064] Norm clipping is applied to the local gradient to limit sensitivity, using the following formula:

[0065]

[0066] Among them, g i This represents the local model update gradient vector for the i-th client, where C represents the gradient clipping threshold norm, used to limit gradient sensitivity, and |·|2 represents the L2 norm operation of the vector.

[0067] Adding Gaussian noise to the clipped gradient yields a privacy-preserving update:

[0068]

[0069] in, Let (σC) represent the gradient vector after adding noise, where σ represents the noise scaling factor, and I represents the identity matrix with the same dimension as the gradient vector. 2 For noise variance;

[0070] According to Gaussian mechanism theory, the noise scaling factor satisfies:

[0071]

[0072] Where ξ represents the privacy budget, which measures the strength of privacy protection, δ represents the upper limit of the failure probability allowed by differential privacy, and ln(·) represents the natural logarithm operation.

[0073] The beneficial effects of this invention are as follows: This invention significantly enhances the autonomous management and control capabilities of satellite constellations through multi-dimensional technological innovation; the dynamic value assessment module integrates multi-source heterogeneous data, breaking through the rigidity of traditional fixed priority strategies and achieving real-time dynamic balancing of mission value; the conflict resolution module transforms inter-satellite resource contention into a graph structure optimization problem, effectively solving the implicit command mutual interference phenomenon in multi-satellite collaborative observation; the knowledge graph library constructs cross-model influence relationship chains, eliminating the implicit conflict risk of heterogeneous satellite command templates; the physical rule engine embeds dynamic constraints into the decision-making process, preventing the generation of dangerous commands that violate the physical laws of spacecraft; and the federated learning architecture, while ensuring data privacy, achieves the collaborative evolution of multi-node models, enabling the system to continuously adapt to the addition of new satellites to the network. Attached Figure Description

[0074] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0075] Figure 1 This is a schematic diagram of the framework of the AI-based satellite remote control command automatic generation system in Example 1. Detailed Implementation

[0076] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0077] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0078] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0079] Example 1, referring to Figure 1 This embodiment provides an artificial intelligence-based automatic satellite remote control command generation system, including:

[0080] The dynamic value assessment module is used to receive real-time satellite status data and mission requirement parameters, and generate a multi-satellite mission priority sequence.

[0081] The dynamic value assessment module includes:

[0082] Satellite remaining power sensor interface, receives battery voltage and temperature data;

[0083] The task timeliness calculation unit calculates the observation urgency coefficient based on the spatiotemporal coordinates of the target event;

[0084] The multi-head attention computing layer performs feature fusion on the energy status, orbital phase, and historical imaging quality of at least three satellites, and outputs dynamic priority weights.

[0085] The conflict resolution module is connected to the dynamic value assessment module and eliminates command conflicts based on the inter-satellite resource occupancy relationship graph.

[0086] The operations of the conflict resolution module include:

[0087] Construct an inter-satellite resource occupancy graph, where nodes represent satellite observation missions and edge weights characterize the overlap of communication frequency bands and the conflict value of attitude adjustment range;

[0088] A graph attention network is used to iteratively update the node feature vectors and generate the optimal task allocation scheme.

[0089] When a risk of energy overload is detected, a power management sub-instruction sequence is automatically inserted;

[0090] The steps for constructing the inter-satellite resource occupancy graph in the conflict resolution module include:

[0091] Each satellite observation mission is mapped to a set of nodes N, and an undirected weighted graph is constructed:

[0092] G = (N, E, W)

[0093] Where N represents the set of all observation task nodes, i,j∈N represent the i-th and j-th tasks respectively, E represents the set of edges between nodes, (i,j)∈E when there is a conflict between tasks i and j, and W is the edge weight function, defined as W(i,j)=C ij , where C ij This represents the combined conflict value between the i-th and j-th tasks;

[0094] The conflict values ​​are linearly weighted using the following formula:

[0095] C ij =αB ij +βA ij ,

[0096] Among them, C ij This represents the overall conflict value, where α and β are normalized weighting coefficients, satisfying α + β = 1, B ij Indicates the overlap of communication frequency bands, A ij Indicates the conflict value within the attitude adjustment range;

[0097] Communication band overlap is quantified as follows:

[0098] in, These are the lower and upper limits of the communication frequency band for the i-th satellite, respectively. These are the lower and upper limits of the communication frequency band for the j-th satellite, respectively;

[0099] The attitude adjustment range conflict value is quantified as follows:

[0100]

[0101] Among them, c i ,c j These are the attitude adjustment angles required for the i-th and j-th satellites to perform their observation tasks, respectively, and Z is the maximum attitude adjustment angle range preset by the system, used for normalization processing;

[0102] Specifically, communication frequency band overlap and attitude differences are transformed into comparable conflict indicators to achieve a unified measurement of resource conflicts. Frequency band overlap is based on the ratio of spectrum intersection to minimum bandwidth, which can reflect the intensity of communication interference between satellites. Attitude conflict values ​​are normalized by angle differences to characterize the risk of attitude adjustment. Linear weighting introduces weight coefficients so that the system can flexibly adjust the contribution ratio of the two parts according to mission requirements.

[0103] The spatiotemporal prediction module analyzes the spatial environment monitoring data stream in real time and predicts abnormal events;

[0104] The spatiotemporal prediction module includes:

[0105] Causal convolutional neural network layer, inputting spatial particle flux and geomagnetic index time series data streams;

[0106] A sliding time window processor outputs the probability of abnormal events occurring within the next 5-10 minutes.

[0107] A pre-compiled instruction cache dynamically loads emergency instruction templates based on predicted probabilities.

[0108] A knowledge graph database stores the influence relationship chains between heterogeneous satellite subsystems;

[0109] Methods for constructing knowledge graph libraries include:

[0110] Parse command-response pairs in the satellite interface control document;

[0111] A random walk algorithm was used to uncover hidden impact paths across satellite models;

[0112] Confidence level and minimum safe interval parameters for storing instruction conflict chains;

[0113] In the knowledge graph database, heterogeneous satellite subsystems and their influence chains are represented as a weighted graph G. K =(V,E) K ,W K ), where W K Let w be the edge weight function, and w ij =W K (i,j), and based on the Random Walk Restart Algorithm (RWR), the implicit influence paths between satellite models are mined. The specific process includes: constructing a node transition probability matrix P on a weighted graph, where each element is obtained by normalizing the edge weights. Let P be the node transition probability matrix on the graph, and its elements are P ij :

[0114]

[0115] Among them, P ij w represents the probability of transitioning from node i to node j. ijThis represents the edge weight between nodes i and j. This represents the set of all nodes adjacent to node i, where k is the index of the adjacent node;

[0116] Initiate a random walk from the source node s, and update the node influence probability vector in each iteration as follows:

[0117] p (t+1) =(1-γ)P T p (t) +γe s ,

[0118] Where, p (t) Let P represent the node influence probability distribution vector in the t-th iteration, where γ represents the restart probability, satisfying 0 < γ < 1. T Denotes the transpose of the transition matrix, e s Let t represent the standard basis vector whose corresponding component is 1 and all other components are 0 at the source node s; t represents the number of iterations and s represents the starting node of the walk.

[0119] When the probability vector converges, the following conditions are met:

[0120] |p (t+1) -p (t) |1<ε,

[0121] Where |·|1 represents the L1 norm, and ε represents the preset convergence threshold;

[0122] After convergence, nodes with probabilities exceeding the threshold τ are selected based on the steady-state distribution to obtain the set of key influencing nodes:

[0123]

[0124] in, p represents the set of key nodes selected. i τ represents the component of the i-th node in the steady-state probability vector, and τ represents the preset probability threshold.

[0125] Specifically, the random walk restart algorithm can measure the influence intensity of each node on the source node on a global scale. By taking into account both local and global information through restart probability, it can discover non-obvious cross-model influence links. The transition probability matrix normalizes the original edge weights to ensure the comparability of influence weights between different models. The convergence criterion and threshold screening provide stability and controllability for path extraction. It does not rely on preset path length and can automatically discover influence paths of varying lengths.

[0126] The physics rules engine embeds satellite dynamics constraints;

[0127] The physics rule engine includes:

[0128] The orbital mechanics constraint unit verifies whether the command parameters conform to the current orbital phase.

[0129] Thermal control boundary check unit compares command-triggered actions with equipment temperature thresholds;

[0130] Differentiable verification layer transforms satellite dynamics equations into neural network regularization terms;

[0131] Federated learning architecture enables secure updates of multi-node model parameters;

[0132] The operation of the federated learning architecture includes:

[0133] During local model training, a differential privacy algorithm is used to add Gaussian noise;

[0134] During the parameter aggregation phase, the model hash value is verified through a blockchain smart contract.

[0135] Before deploying the model, perform adversarial example stress tests in the digital twin environment;

[0136] In the federated learning architecture, a Gaussian mechanism is introduced to update the gradient vector g of each client's local model. i The steps for trimming and injecting noise include:

[0137] Norm clipping is applied to the local gradient to limit sensitivity, using the following formula:

[0138]

[0139] Among them, g i This represents the local model update gradient vector for the i-th client, where C represents the gradient clipping threshold norm, used to limit gradient sensitivity, and |·|2 represents the L2 norm operation of the vector.

[0140] Adding Gaussian noise to the clipped gradient yields a privacy-preserving update:

[0141]

[0142] in, Let (σC) represent the gradient vector after adding noise, where σ represents the noise scaling factor, and I represents the identity matrix with the same dimension as the gradient vector. 2 For noise variance;

[0143] According to Gaussian mechanism theory, to satisfy (∈,δ) differential privacy, the noise scaling factor satisfies:

[0144]

[0145] Where ξ represents the privacy budget, which measures the strength of privacy protection, δ represents the upper limit of the failure probability allowed by differential privacy, and ln(·) represents the natural logarithm operation;

[0146] Specifically, this Gaussian mechanism first limits the sensitivity of each client's updates to a constant C range through pruning, ensuring that different clients contribute consistently to privacy leaks, and then injects a variance of (σC). 2 The Gaussian noise ensures that each update satisfies the (∈,δ) differential privacy guarantee. The relationship between the noise factor σ and the privacy budget ∈ and the failure probability δ is the theoretical minimum lower bound, which can minimize the noise intensity under given privacy requirements, thereby taking into account the model performance.

[0147] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. An artificial intelligence-based automatic satellite remote control command generation system, characterized in that, include, The dynamic value assessment module is used to receive real-time satellite status data and mission requirement parameters, and generate a multi-satellite mission priority sequence. The conflict resolution module is connected to the dynamic value assessment module and eliminates command conflicts based on the inter-satellite resource occupancy relationship graph. The spatiotemporal prediction module analyzes the spatial environment monitoring data stream in real time and predicts abnormal events; A knowledge graph database stores the influence relationship chains between heterogeneous satellite subsystems; The physics rules engine embeds satellite dynamics constraints; Federated learning architecture enables secure updates of multi-node model parameters; The operation of the conflict resolution module includes: Construct an inter-satellite resource occupancy graph, where nodes represent satellite observation missions and edge weights characterize the overlap of communication frequency bands and the conflict value of attitude adjustment range; A graph attention network is used to iteratively update the node feature vectors and generate the optimal task allocation scheme. When a risk of energy overload is detected, a power management sub-instruction sequence is automatically inserted; The steps for constructing the inter-satellite resource occupancy graph in the conflict resolution module include: Each satellite observation mission is mapped to a set of nodes N, and an undirected weighted graph is constructed: G = (N, E, W) Where N represents the set of all observation task nodes, i,j∈N represent the i-th and j-th tasks respectively, E represents the set of edges between nodes, (i,j)∈E when there is a conflict between tasks i and j, and W is the edge weight function, defined as W(i,j)=C ij , where C ij This represents the combined conflict value between the i-th and j-th tasks; The conflict values ​​are linearly weighted using the following formula: C ij =αB ij +βA ij , Among them, C ij This represents the overall conflict value, where α and β are normalized weighting coefficients, satisfying α + β = 1, B ij Indicates the overlap of communication frequency bands, A ij Indicates the conflict value within the attitude adjustment range; Communication band overlap is quantified as follows: in, These are the lower and upper limits of the communication frequency band for the i-th satellite, respectively. These are the lower and upper limits of the communication frequency band for the j-th satellite, respectively; The attitude adjustment range conflict value is quantified as follows: Among them, c i ,c j , i and j respectively, are the attitude adjustment angles required for the i-th and j-th satellites to perform the observation mission, and Z is the maximum attitude adjustment angle range preset by the system, used for normalization processing.

2. The artificial intelligence-based automatic satellite remote control command generation system as described in claim 1, characterized in that, The dynamic value assessment module includes: Satellite remaining power sensor interface, receives battery voltage and temperature data; The task timeliness calculation unit calculates the observation urgency coefficient based on the spatiotemporal coordinates of the target event; The multi-head attention computing layer performs feature fusion on the energy status, orbital phase, and historical imaging quality of at least three satellites, and outputs dynamic priority weights.

3. The artificial intelligence-based automatic satellite remote control command generation system as described in claim 1, characterized in that, The spatiotemporal prediction module includes: Causal convolutional neural network layer, inputting spatial particle flux and geomagnetic index time series data streams; A sliding time window processor outputs the probability of abnormal events occurring within the next 5-10 minutes. The pre-compiled instruction cache dynamically loads emergency instruction templates based on predicted probabilities.

4. The artificial intelligence-based automatic satellite remote control command generation system as described in claim 1, characterized in that, The method for constructing the knowledge graph database includes: Parse command-response pairs in the satellite interface control document; A random walk algorithm was used to uncover hidden impact paths across satellite models; The confidence level and minimum safe interval parameters for storing instruction conflict chains.

5. The artificial intelligence-based automatic satellite remote control command generation system as described in claim 4, characterized in that, In the knowledge graph database, heterogeneous satellite subsystems and their influence chains are represented as a weighted graph G. K =(V,E) K W K ), where W K Let w be the edge weight function, and w ij =W K (i,j), and based on the Random Walk Restart Algorithm (RWR), the implicit influence paths between satellite models are mined. The specific process includes: constructing a node transition probability matrix P on a weighted graph, where each element is obtained by normalizing the edge weights. Let P be the node transition probability matrix on the graph, and its elements are P ij : Among them, P ij w represents the probability of transitioning from node i to node j. ij This represents the edge weight between nodes i and j. This represents the set of all nodes adjacent to node i, where k is the index of the adjacent node; Initiate a random walk from the source node s, and update the node influence probability vector in each iteration as follows: p (t+1) =(1-γ)P T p (t) +ge s , Where, p (t) Let P represent the node influence probability distribution vector in the t-th iteration, where γ represents the restart probability, satisfying 0 < γ < 1. T Denotes the transpose of the transition matrix, e s Let t represent the standard basis vector whose corresponding component is 1 and all other components are 0 at the source node s; t represents the number of iterations and s represents the starting node of the walk. When the probability vector converges, the following conditions are met: |p (t+1) -p (t) |1<∈, Where |·|1 represents the L1 norm, and ∈ represents the preset convergence threshold; After convergence, nodes with probabilities exceeding the threshold τ are selected based on the steady-state distribution to obtain the set of key influencing nodes: in, p represents the set of key nodes selected. i τ represents the component of the i-th node in the steady-state probability vector, and τ represents the preset probability threshold.

6. The artificial intelligence-based automatic satellite remote control command generation system as described in claim 1, characterized in that, The physical rule engine includes: The orbital mechanics constraint unit verifies whether the command parameters conform to the current orbital phase. Thermal control boundary check unit compares command-triggered actions with equipment temperature thresholds; Differentiable verification layers transform satellite dynamic equations into neural network regularization terms.

7. The artificial intelligence-based automatic satellite remote control command generation system as described in claim 1, characterized in that, The operations of the federated learning architecture include: During local model training, a differential privacy algorithm is used to add Gaussian noise; During the parameter aggregation phase, the model hash value is verified through a blockchain smart contract. Before deploying the model, perform adversarial example stress tests in the digital twin environment.

8. The artificial intelligence-based automatic satellite remote control command generation system as described in claim 7, characterized in that, In the federated learning architecture, a Gaussian mechanism is introduced to update the gradient vector g of each client's local model. i The steps for trimming and injecting noise include: Norm clipping is applied to the local gradient to limit sensitivity, using the following formula: Among them, g i This represents the local model update gradient vector for the i-th client, where C represents the gradient clipping threshold norm, used to limit gradient sensitivity, and |·|2 represents the L2 norm operation of the vector. Adding Gaussian noise to the clipped gradient yields a privacy-preserving update: in, Let (σC) represent the gradient vector after adding noise, where σ represents the noise scaling factor, and I represents the identity matrix with the same dimension as the gradient vector. 2 For noise variance; According to Gaussian mechanism theory, the noise scaling factor satisfies: Where ξ represents the privacy budget, which measures the strength of privacy protection, δ represents the upper limit of the failure probability allowed by differential privacy, and ln(·) represents the natural logarithm operation.

Images