A load balancing routing method and system for satellite network

Abstract

The present invention provides a load balancing routing method and system for a satellite network. The method comprises the following steps: collecting satellite operation status of satellite nodes through a satellite sensor network to obtain satellite original status data; performing network status perception on the satellite original status data to obtain a satellite network status data set; performing task demand mining on communication participating nodes to obtain a task demand trend data set; performing task demand prediction modeling and task demand prediction according to a preset deep learning model and a task demand trend data set to obtain a task demand prediction data set. The present invention improves the energy efficiency of a multi-satellite system, realizes task load balancing and coordination in a cloud-edge collaborative environment, and ensures privacy security in an air-ground integrated network through technical methods such as satellite status perception, task demand prediction, distributed resource optimization, reinforcement learning resource scheduling, and privacy protection mechanisms.

Description

Technical Field

[0001] The present invention relates to the field of satellite communication technology, and in particular to a load balancing routing method and system for a satellite network. Background Art

[0002] With the rapid development of satellite communication technology and the continuous expansion of satellite network scale, satellite networks are facing increasingly complex challenges in task processing and resource allocation. Traditional ground computing models are difficult to meet the needs of real-time processing of massive data and dynamic service quality assurance. Therefore, aerial computing technology came into being. By integrating computing and storage resources on satellite nodes, data can be processed nearby, transmission delay can be reduced, and service quality can be improved.

[0003] However, the existing satellite network load balancing routing methods based on air computing still have some limitations:

[0004] Insufficient adaptability to dynamic network environments: The satellite network topology changes dynamically, and the channel conditions are complex and changeable. Existing routing methods are difficult to perceive network status changes in real time and make dynamic adjustments, resulting in inefficient routing decisions.

[0005] Insufficient data security and privacy protection: Task offloading involves the transmission and processing of user sensitive data. Existing methods lack effective security and privacy protection mechanisms, which can easily lead to data leakage and privacy infringement.

[0006] Inefficient task offloading and resource allocation: Existing methods usually adopt a centralized scheduling method, lacking global optimization and dynamic scheduling of satellite network resources, which easily causes unbalanced node load and affects the efficiency of task completion. Summary of the invention

[0007] In view of the above problems, the present invention proposes a load balancing routing method and system for satellite networks, which realizes load balancing routing of satellite networks through traffic analysis and routing optimization, thereby improving data transmission efficiency and network stability.

[0008] To achieve the above object, a load balancing routing method for a satellite network comprises the following steps:

[0009] Step S1: collect satellite operation status of satellite nodes through the satellite sensor network to obtain satellite original status data; perform network status perception on the satellite original status data to obtain a satellite network status data set;

[0010] Step S2: mining the task requirements of the communication participating nodes to obtain a task requirement trend data set; performing task requirement prediction modeling and task requirement prediction based on a preset deep learning model and the task requirement trend data set to obtain a task requirement prediction data set;

[0011] Step S3: Initialize and distribute the model according to the satellite network status data set and the mission demand prediction data set to obtain distributed model parameters; perform local model training on the satellite nodes according to the distributed model parameters to obtain local encrypted model parameters; perform global model aggregation according to the local encrypted model parameters to obtain a global training model; perform model evaluation on the global training model to obtain a distributed resource optimization model;

[0012] Step S4: encrypting the task demand prediction data set to obtain encrypted task demand information data; performing secure multi-party computing processing and privacy protection evaluation based on the encrypted task demand information data and the distributed resource optimization model to obtain a task allocation plan;

[0013] Step S5: According to the satellite network status data set, the task demand prediction data set and the task allocation plan, the satellite network environment status is sensed to obtain the environment status data; the reinforcement learning technology and the environment status data are used to perform resource scheduling learning to obtain the resource scheduling strategy; according to the resource scheduling strategy, the satellite nodes are scheduled for air computing resources to obtain the air computing resource scheduling plan;

[0014] Step S6: Perform routing policy traffic analysis based on the satellite network status data set, task allocation plan and air computing resource scheduling plan to obtain routing policy data; perform load balancing adjustment based on the routing policy data to obtain a load balancing routing plan to implement load balancing routing operations in the satellite network.

[0015] The present invention acquires the running status of satellite nodes in real time, builds an accurate satellite network status model, and provides basic data for subsequent resource scheduling, task allocation and routing strategy formulation. By analyzing historical task data and user behavior, future task requirements are predicted, providing key information for resource scheduling and load balancing. By analyzing historical data, the trend change of task requirements can be identified, and deep learning models can be used for prediction, thereby improving the accuracy and efficiency of prediction. Distributed training and encryption technology are used to improve the training efficiency of resource optimization models and protect data privacy. By distributing model training to various nodes, computing resources can be fully utilized to improve training efficiency. Encryption technology can protect sensitive information and avoid data leakage. Through secure multi-party computing technology, resource allocation schemes are jointly calculated without leaking any party's sensitive information, thereby ensuring data security and user privacy. Secure multi-party computing technology enables multiple participants to jointly calculate a function without sharing their private data, thereby ensuring data privacy. Reinforcement learning technology is used to dynamically adjust according to the real-time environmental status, thereby improving the flexibility and efficiency of resource allocation. The reinforcement learning model can continuously learn and optimize the scheduling strategy according to the environmental state and reward function, and finally find a scheduling strategy that can maximize the system benefit. Through traffic analysis and routing optimization, load balancing routing of satellite networks is realized, and data transmission efficiency and network stability are improved. By analyzing network traffic and topology, the best data transmission path can be planned, and dynamic adjustment can be made according to the real-time network status to achieve load balancing, improve data transmission efficiency and network stability. Therefore, the present invention provides a load balancing routing method and system for satellite networks, which improves the energy efficiency of multi-satellite systems, realizes task load balancing and coordination in a cloud-edge collaborative environment, and ensures privacy security in an integrated air-ground network through satellite state perception, task demand prediction, distributed resource optimization, reinforcement learning resource scheduling, privacy protection mechanism and other methods.

[0016] Preferably, step S1 further comprises:

[0017] Step S11: collecting satellite operation status of satellite nodes through the satellite sensor network to obtain satellite original status data;

[0018] Step S12: performing edge preprocessing on the satellite original state data to obtain satellite preprocessing state data;

[0019] Step S13: extracting state features from the satellite preprocessed state data to obtain satellite state feature data;

[0020] Step S14: performing status feature aggregation on the satellite status feature data to obtain a satellite network status data set.

[0021] The present invention obtains the real-time operating status of satellite nodes through a satellite sensor network, and provides basic data for subsequent network status perception and resource scheduling. It can comprehensively and timely collect various operating parameters of satellite nodes. By performing operations such as denoising, compression, and format conversion on the original data, data redundancy can be reduced, data transmission efficiency can be improved, and more concise and effective pre-processed data can be provided for subsequent feature extraction and aggregation. The advantage of edge preprocessing is that it can complete part of the data processing work near the satellite node, reduce the amount of data transmitted to the ground center, and improve system efficiency. Through the feature extraction algorithm, more effective feature information for satellite network status perception and resource scheduling is extracted from the pre-processed data. More effective information is provided for subsequent network status analysis and resource optimization. Feature extraction can convert raw data into features that are easier to understand and analyze, helping the system to better understand the operating status of the satellite network. By aggregating the feature data of multiple satellite nodes, a complete satellite network status data set is constructed. The data set can reflect the operating status of the entire satellite network and provide global information support for subsequent resource scheduling, task allocation, routing strategy formulation and other decisions. At the same time, through the aggregation operation, the data dimension can be reduced, the model training and analysis process can be simplified, and the system efficiency can be improved. State feature aggregation can integrate the information of each satellite node to form an overall network status view, providing more comprehensive information for system decision-making.

[0022] Preferably, step S2 further comprises:

[0023] Step S21: Collecting task history data from multiple sources on the communication participating nodes to obtain a task history data set;

[0024] Step S22: Perform user behavior analysis based on the task history data set to obtain a user behavior analysis data set;

[0025] Step S23: performing demand trend mining on the task history data set and the user behavior analysis data set to obtain a task demand trend data set;

[0026] Step S24: using a preset deep learning model and a task demand trend data set as training data to train a demand forecasting model and obtain a task demand forecasting model;

[0027] Step S25: Acquire real-time satellite network status data, use the mission demand prediction model and the real-time satellite network status data to perform mission demand prediction, and obtain a mission demand prediction data set.

[0028] The present invention provides basic materials for subsequent task demand prediction by accumulating task history data of communication participating nodes. By collecting task history data from different sources, such as task records submitted by users, historical task execution data, task types and parameters, etc., a comprehensive and rich task history data set can be constructed to lay the foundation for in-depth analysis of user behavior and prediction of future task requirements. By analyzing task history data, user behavior habits and task demand change trends can be identified, providing more accurate guidance for subsequent task demand prediction. By analyzing task history data and user behavior patterns, seasonal changes, periodic changes, sudden changes and other trends in task requirements can be identified, providing training data for subsequent task demand prediction models. The deep learning model can learn complex relationships and nonlinear patterns in the data, providing powerful capabilities for more accurately predicting future task requirements. By combining real-time satellite network status data, such as available resources of satellite nodes, communication link status, etc., future task requirements can be predicted more accurately and the reliability of the prediction results can be improved.

[0029] Preferably, step S3 further comprises:

[0030] Step S31: using a preset resource optimization model structure to initialize model parameters according to a satellite network status data set and a mission demand prediction data set to obtain an initial resource optimization model;

[0031] Step S32: Perform satellite node performance analysis based on the satellite original status data to obtain node computing capacity data and node data storage capacity data; distribute the initial resource optimization model to each satellite node, divide the model parameters based on the node computing capacity data and the node data storage capacity data to obtain distributed model parameters;

[0032] Step S33: Performing local model training on the satellite nodes according to the distributed model parameters to obtain local encrypted model parameters; performing global model aggregation according to the local encrypted model parameters to obtain a global training model;

[0033] Step S34: perform a global model evaluation on the global training model to obtain global model evaluation data; use the global model evaluation data to adjust model parameters of the global training model to obtain a distributed resource optimization model.

[0034] The present invention can initialize the model parameters according to the satellite network status data set and the task demand prediction data set, so that it can better adapt to the specific environment and task requirements of the satellite network, and improve the efficiency and accuracy of model training. By analyzing the performance of the satellite node, the information such as the computing power and data storage capacity of the node can be obtained, and the initial resource optimization model is distributed to each node, and the model parameters are divided according to the node's ability, so that each node can perform model training according to its own ability, thereby improving the overall training efficiency. Each satellite node performs local training according to its own model parameters, encrypts the training results, and then uploads the encrypted model parameters to the central server for global model aggregation, and finally obtains a global training model. This distributed training method can make full use of the computing resources of each node and improve the model training efficiency, while encryption technology can ensure data privacy and avoid sensitive information leakage. By evaluating the performance of the model, the advantages and disadvantages of the model can be analyzed, and the model can be adjusted according to the evaluation results, such as adjusting the model parameters, optimizing the model structure, etc., and finally a model that can effectively optimize resource allocation is obtained, thereby improving resource utilization and system efficiency.

[0035] Preferably, step S33 further comprises:

[0036] Step S331: performing local data preprocessing on the satellite node to obtain local data preprocessing data; performing local model training according to the local data preprocessing data and the distributed model parameters to obtain a local initial model;

[0037] Step S332: performing local model compression on the local initial model to obtain a local compressed model; performing local model encryption on the local compressed model to obtain local encrypted model parameters;

[0038] Step S333: transmitting the local encrypted model parameters to the central server, and using the preset model key to perform local model decryption on the local encrypted model parameters to obtain local model decryption data;

[0039] Step S334: perform federated learning model aggregation on the local model decrypted data to obtain model aggregation data; use the preset model aggregation strategy and model aggregation data to update the global model parameters to obtain global model aggregation data; perform global model encryption on the global model aggregation data to obtain a global training model.

[0040] The present invention performs local preprocessing on each satellite node according to the local data received by itself, and then uses the preprocessed data and the model parameters distributed to itself to perform local model training to obtain a preliminary local initial model. This distributed training method can make full use of the computing resources of each node and improve the training speed. Compressing the local initial model can reduce the amount of data transmission and improve the transmission efficiency. The compressed model is encrypted to prevent the model from being stolen or tampered with. The central server has a preset model key and can decrypt the encrypted model parameters to obtain the original model parameters, providing the necessary data for subsequent model aggregation. By aggregating the model parameters of each node, a more powerful global model can be obtained, while also protecting data privacy.

[0041] Preferably, step S4 further comprises:

[0042] Step S41: performing sensitive information identification on the task demand prediction data set to obtain task sensitive information data;

[0043] Step S42: using the preset privacy budget data to add noise to the task sensitive information data and perform noise control to obtain the task required noise data;

[0044] Step S43: encrypting the task requirement noise data to obtain encrypted task requirement information data;

[0045] Step S44: transmitting the encrypted task requirement information data and the distributed resource optimization model to the communication participating nodes, performing protocol initialization, and obtaining protocol initialization data;

[0046] Step S45: Perform secure multi-party encryption joint calculation according to the protocol initialization data, the encryption task requirement information data and the distributed resource optimization model to obtain a resource allocation encryption scheme;

[0047] Step S46: decrypt the encrypted resource allocation scheme and perform a privacy protection evaluation to obtain a task allocation scheme.

[0048] The present invention can take targeted privacy protection measures to prevent sensitive information leakage by identifying sensitive information, such as user's personal information, task content, task priority, etc. By adding noise, sensitive information can be hidden in the noise data, making it difficult for attackers to directly obtain sensitive information. Privacy budget data can control the degree of noise addition and balance the relationship between privacy protection and data availability. Encryption technology can convert data into unreadable ciphertext, and only people with decryption keys can decrypt the data, thereby ensuring data security. Transmitting encrypted data to communication participating nodes can prevent sensitive information from being leaked during transmission. Protocol initialization can ensure that a consistent computing protocol is reached between participating nodes, providing protection for subsequent secure multi-party computing. Secure multi-party computing can enable multiple participants to jointly calculate a function without sharing their private data, thereby ensuring data privacy. The decryption process needs to use a preset key to ensure that only authorized parties can decrypt the solution. Privacy protection evaluation can analyze the security of the solution, evaluate the risk of leakage of sensitive information, and ensure the security of the solution.

[0049] Preferably, step S5 further comprises:

[0050] Step S51: performing satellite network environment state perception according to the satellite network state data set, the task demand prediction data set and the task allocation plan to obtain environment state data;

[0051] Step S52: using the reinforcement learning technology and the environmental state data to define the state space, and obtain the state space data; using the reinforcement learning technology and the environmental state data to define the action space, and obtain the action space data; using the reinforcement learning technology and the environmental state data to design the reward function, and obtain the reward function data;

[0052] Step S53: Obtain historical scheduling strategy data; perform node load calculation based on the environment status data and the historical scheduling strategy data to obtain task load data;

[0053] Step S54: using an adaptive weighted scheduling optimization algorithm to perform adaptive resource scheduling weight calculation according to the task load data and the task demand prediction data set to obtain adaptive weighted scheduling data;

[0054] Step S55: performing scheduling strategy learning according to the adaptive weight scheduling data, the state space data, the action space data and the reward function data to obtain a resource scheduling strategy;

[0055] Step S56: Allocate resources to satellite nodes according to the resource scheduling strategy to obtain a resource allocation plan; execute tasks according to the resource allocation plan and monitor the execution process to obtain execution process monitoring data;

[0056] Step S57: Optimize the resource allocation plan according to the execution process monitoring data to obtain the air computing resource scheduling plan.

[0057] By integrating multiple data sets, the present invention can more comprehensively understand the current state of the satellite network, such as the available resources of satellite nodes, the status of communication links, etc., and provide support for formulating more effective resource scheduling strategies. Through the state space definition, the environmental state data can be converted into state information that can be recognized by the reinforcement learning model. Through the action space definition, the resource scheduling actions that can be taken can be converted into actions that can be selected by the model. Through the reward function design, different scheduling results can be converted into reward values ​​that can be understood by the model, guiding the model to learn the best scheduling strategy. By analyzing the historical scheduling strategy data, the advantages and disadvantages of different scheduling strategies can be understood, providing a reference for the formulation of the current scheduling strategy. According to the environmental state data and the historical scheduling strategy data, the load of each node can be calculated, providing more accurate information for adaptive weight scheduling. The constructed adaptive weight scheduling algorithm can dynamically adjust the scheduling weight of each node according to the changes in the load of the node and the task requirements, and assign tasks to nodes with lighter loads and sufficient resources, thereby improving resource utilization and system efficiency. Using reinforcement learning technology, a model that can perform optimal resource scheduling according to the environmental state is trained according to the adaptive weight scheduling data, state space data, action space data and reward function data. According to the trained scheduling strategy, tasks can be assigned to appropriate nodes to improve task execution efficiency. Execution process monitoring can collect data generated during task execution, such as task execution time, resource usage, etc., to provide reference for subsequent scheduling optimization. By analyzing the execution process monitoring data, the shortcomings of the scheduling strategy can be identified and targeted adjustments can be made, such as adjusting node weights, optimizing task allocation strategies, etc., to ultimately obtain a more effective in-flight computing resource scheduling solution.

[0058] Preferably, the adaptive weight scheduling optimization algorithm in step S54 is as follows:

[0059] ;

[0060] In the formula, Indicates The adaptive weights of satellite nodes, Indicates The mission load of satellite nodes, represents the summation variable, Indicates the total number of satellite nodes, Indicates the satellite node index value, Indicates The task priority of each satellite node, Indicates The mission demand forecast value of satellite nodes, represents the angle parameter of time variation, Indicates The resource utilization of satellite nodes, represents the reward function, represents the scheduling weight adjustment factor, Represents the reward function parameters.

[0061] The present invention constructs an adaptive weight scheduling optimization algorithm, where The core goal of this formula is to calculate the weight of each satellite node, which is used to decide which node to assign the task to during resource scheduling. The higher the weight, the more likely the node is to be assigned a task. It reflects the current load of the node. The lower the load, the higher the weight. When a node has a lower load, it has more resources to process new tasks, so the weight assigned to it should be higher. , which reflects the priority of the task. When the task priority is higher, the value of this part is larger and the weight is higher, which means that high-priority tasks should be assigned to nodes with sufficient resources. It reflects the node's prediction of future task demand. When the node estimates that the future demand is low, the larger the value of this part is, the higher the weight is, which means that the node with low estimated demand has more resources to process new tasks. The impact of time changes is taken into account. For example, in certain time periods, the task demand will increase. At this time, the node weight can be appropriately increased to cope with the higher demand. It reflects the resource utilization of the node. When the resource utilization of the node is high, the larger the value of this part is, the higher the weight is. This means that nodes with high resource utilization should be assigned tasks to make full use of resources. It reflects the influence of the reward function. The reward function can be designed according to different scheduling objectives. For example, it can reward tasks to nodes with low load or reward tasks to nodes with high resource utilization. and is an adjustment factor that can be adjusted according to specific circumstances to balance the weights between different objectives. This formula takes into account factors such as task load, task priority, task demand forecast, time change, resource utilization, and reward function. By calculating the weight of each node, adaptive resource scheduling is achieved, and resource allocation strategies can be dynamically adjusted according to different situations to improve resource utilization and system efficiency.

[0062] Preferably, step S6 further comprises:

[0063] Step S61: performing satellite network topology perception according to the satellite network status data set and the air computing resource scheduling scheme to obtain satellite network topology data;

[0064] Step S62: using the task allocation scheme and the air computing resource scheduling scheme to perform network traffic prediction and obtain traffic prediction data;

[0065] Step S63: performing routing strategy learning according to the satellite network topology data and the traffic prediction data to obtain routing strategy learning data; performing routing evaluation index calculation on the routing strategy learning data to obtain routing strategy data;

[0066] Step S64: performing data packet path planning according to the routing strategy data to obtain optimal routing path data; performing path adjustment according to the optimal routing path data and the satellite network status data set to obtain a load balancing routing initial solution;

[0067] Step S65: Perform data transmission and network status monitoring according to the load balancing routing initial plan to obtain network status monitoring data; use the network status monitoring data to optimize the routing of the load balancing routing initial plan to obtain a load balancing routing plan to implement load balancing routing operations in the satellite network.

[0068] The present invention can obtain information such as connection relationship, link bandwidth, node computing power, etc. between satellite nodes by integrating satellite network status data set and air computing resource scheduling scheme, so as to construct the topological structure of satellite network and provide necessary information for subsequent routing strategy formulation. According to the task allocation scheme and air computing resource scheduling scheme, the flow changes between different time periods and different nodes can be predicted, providing reference for subsequent routing strategy formulation. By combining satellite network topology data and flow prediction data, a model that can dynamically adjust routing strategy according to network status and flow changes can be trained. Routing evaluation index can evaluate the performance of different routing strategies, such as delay, packet loss rate, throughput, etc., to help select the optimal routing strategy. According to routing strategy data, the best transmission path of data packets from source node to target node can be planned, and the path can be dynamically adjusted according to real-time network status, such as link bandwidth, node load, etc., to achieve load balancing. Through actual data transmission and network status monitoring, more accurate network status information, such as actual link bandwidth, node load, etc., can be collected, and the routing scheme can be continuously optimized according to this information, and finally the load balancing routing of satellite network is realized, and the data transmission efficiency and network stability are improved.

[0069] The present invention also provides a load balancing routing system for a satellite network, the system comprising:

[0070] The satellite network status perception module is used to collect satellite operation status of satellite nodes through the satellite sensor network to obtain satellite original status data; perform network status perception on the satellite original status data to obtain a satellite network status data set;

[0071] The task demand prediction module is used to mine the task demand of the communication participating nodes to obtain a task demand trend data set; perform task demand prediction modeling and task demand prediction based on the preset deep learning model and the task demand trend data set to obtain a task demand prediction data set;

[0072] The federated learning collaborative optimization module is used to initialize and distribute the model according to the satellite network status data set and the mission demand prediction data set to obtain the distributed model parameters; perform local model training on the satellite nodes according to the distributed model parameters to obtain the local encrypted model parameters; perform global model aggregation according to the local encrypted model parameters to obtain the global training model; perform model evaluation on the global training model to obtain the distributed resource optimization model;

[0073] The privacy-preserving task offloading module is used to encrypt the task demand prediction data set to obtain encrypted task demand information data; perform secure multi-party computing processing and privacy protection evaluation based on the encrypted task demand information data and the distributed resource optimization model to obtain a task allocation plan;

[0074] The air computing resource scheduling module is used to perceive the satellite network environment status according to the satellite network status data set, the task demand prediction data set and the task allocation plan to obtain the environment status data; use the reinforcement learning technology and the environment status data to perform resource scheduling learning to obtain the resource scheduling strategy; perform air computing resource scheduling on the satellite nodes according to the resource scheduling strategy to obtain the air computing resource scheduling plan;

[0075] The load balancing routing module is used to perform routing strategy traffic analysis based on the satellite network status data set, task allocation plan and air computing resource scheduling plan to obtain routing strategy data; perform load balancing adjustment based on the routing strategy data to obtain a load balancing routing plan to achieve load balancing routing operations in the satellite network.

[0076] The present invention acquires the running status of satellite nodes in real time, builds an accurate satellite network status model, and provides basic data for subsequent resource scheduling, task allocation and routing strategy formulation. By analyzing historical task data and user behavior, future task requirements are predicted, providing key information for resource scheduling and load balancing. By analyzing historical data, the trend change of task requirements can be identified, and deep learning models can be used for prediction, thereby improving the accuracy and efficiency of prediction. Distributed training and encryption technology are used to improve the training efficiency of resource optimization models and protect data privacy. By distributing model training to various nodes, computing resources can be fully utilized to improve training efficiency. Encryption technology can protect sensitive information and avoid data leakage. Through secure multi-party computing technology, resource allocation schemes are jointly calculated without leaking any party's sensitive information, thereby ensuring data security and user privacy. Secure multi-party computing technology enables multiple participants to jointly calculate a function without sharing their private data, thereby ensuring data privacy. Reinforcement learning technology is used to dynamically adjust according to the real-time environmental status, thereby improving the flexibility and efficiency of resource allocation. The reinforcement learning model can continuously learn and optimize the scheduling strategy according to the environmental state and reward function, and finally find a scheduling strategy that can maximize the system benefit. Through traffic analysis and routing optimization, load balancing routing of satellite networks is realized, and data transmission efficiency and network stability are improved. By analyzing network traffic and topology, the best data transmission path can be planned, and dynamic adjustment can be made according to the real-time network status to achieve load balancing, improve data transmission efficiency and network stability. Therefore, the load balancing routing method and system of satellite networks provided by the present invention improve the energy efficiency of multi-satellite systems through technologies such as satellite state perception, task demand prediction, distributed resource optimization, reinforcement learning resource scheduling, and privacy protection mechanism, realize task load balancing and coordination in a cloud-edge collaborative environment, and ensure privacy security in an integrated space-ground network. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0078] Figure 1 It is a flow chart of a load balancing routing method for a satellite network provided by an embodiment of the present invention;

[0079] Figure 2 It is a schematic diagram of a process for obtaining a distributed resource optimization model provided by an embodiment of the present invention;

[0080] Figure 3 It is a flowchart of obtaining an air computing resource scheduling solution provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0081] 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.

[0082] The present invention provides a load balancing routing method for a satellite network, such as Figure 1 As shown, the following steps are included:

[0083] Step S1: collect satellite operation status of satellite nodes through the satellite sensor network to obtain satellite original status data; perform network status perception on the satellite original status data to obtain a satellite network status data set;

[0084] The sensors on each satellite node regularly collect the operating status data of the satellite node, which is stored in the local memory of the satellite node in the form of raw data. Then, the raw data is processed, key features are extracted, and aggregated into a complete satellite network status data set.

[0085] Step S2: mining the task requirements of the communication participating nodes to obtain a task requirement trend data set; performing task requirement prediction modeling and task requirement prediction based on a preset deep learning model and the task requirement trend data set to obtain a task requirement prediction data set;

[0086] Collect mission history data from each communication participating node and build a unified mission history data set. Use data mining technology to analyze the mission history data set, identify user behavior patterns, and build a user behavior analysis data set. Combine the mission history data set and the user behavior analysis data set to analyze the trend of mission requirements and build a mission requirement trend data set. Use the preset deep learning model and the mission requirement trend data set as training data to train the demand prediction model and obtain a model that can predict future mission requirements. Finally, use the trained model and real-time satellite network status data to predict future mission requirements and obtain a mission requirement prediction data set.

[0087] Step S3: Initialize and distribute the model according to the satellite network status data set and the mission demand prediction data set to obtain distributed model parameters; perform local model training on the satellite nodes according to the distributed model parameters to obtain local encrypted model parameters; perform global model aggregation according to the local encrypted model parameters to obtain a global training model; perform model evaluation on the global training model to obtain a distributed resource optimization model;

[0088] Using the preset resource optimization model structure, the model parameters are initialized according to the satellite network status data set and the mission demand prediction data set to obtain an initial resource optimization model. The initial resource optimization model is distributed to each satellite node, and the model parameters are divided according to the performance indicators of the node to obtain distributed model parameters. Each satellite node performs model training based on its own assigned model parameters and local data to obtain a local encrypted model parameter. Then, the encrypted model parameters are uploaded to the central server for global model aggregation to finally obtain a global training model, evaluate the global training model, and adjust the parameters of the global training model according to the evaluation results, and finally obtain a distributed resource optimization model that can effectively optimize resource allocation.

[0089] Step S4: encrypting the task demand prediction data set to obtain encrypted task demand information data; performing secure multi-party computing processing and privacy protection evaluation based on the encrypted task demand information data and the distributed resource optimization model to obtain a task allocation plan;

[0090] The task demand prediction data set is encrypted to obtain encrypted task demand information data. The encrypted task demand information data and the distributed resource optimization model are used to perform secure multi-party computing, and the resource allocation encryption scheme is jointly calculated without leaking any party's sensitive information. Finally, the resource allocation encryption scheme is decrypted to obtain the final task allocation scheme, and the privacy protection measures are evaluated.

[0091] Step S5: According to the satellite network status data set, the task demand prediction data set and the task allocation plan, the satellite network environment status is sensed to obtain the environment status data; the reinforcement learning technology and the environment status data are used to perform resource scheduling learning to obtain the resource scheduling strategy; according to the resource scheduling strategy, the satellite nodes are scheduled for air computing resources to obtain the air computing resource scheduling plan;

[0092] In an embodiment of the present invention, a satellite network status data set, a task demand prediction data set, and a task allocation scheme are integrated to construct a satellite network environment status data containing satellite node status, task demand, and task allocation information. Reinforcement learning technology is used to convert the environment status data into a language that can be understood by the reinforcement learning model, and a reward function is designed according to different scheduling results. Historical scheduling strategy data is obtained, and the load of each node is calculated based on the environment status data and the historical scheduling strategy data. An adaptive weight scheduling optimization algorithm is used to dynamically adjust the scheduling weight of each node based on the task load data and the task demand prediction data set. Reinforcement learning technology is used to train a model that can perform optimal resource scheduling according to the environment state based on the adaptive weight scheduling data, state space data, action space data, and reward function data.

[0093] Step S6: Perform routing policy traffic analysis based on the satellite network status data set, task allocation plan and air computing resource scheduling plan to obtain routing policy data; perform load balancing adjustment based on the routing policy data to obtain a load balancing routing plan to implement load balancing routing operations in the satellite network.

[0094] In an embodiment of the present invention, a topological structure of a satellite network is constructed according to a satellite network status data set and an air computing resource scheduling scheme. According to a task allocation scheme and an air computing resource scheduling scheme, traffic changes between different time periods and different nodes are predicted. According to the satellite network topology data and traffic prediction data, a model capable of dynamically adjusting a routing strategy according to network status and traffic changes is trained, and the learned routing strategy is evaluated using an evaluation index. According to the routing strategy data, an optimal transmission path for a data packet from a source node to a target node is planned. At the same time, according to the real-time network status, the path is dynamically adjusted to achieve load balancing and form an initial load balancing routing scheme. Data is transmitted according to the initial load balancing routing scheme, and the network status is monitored in real time. The initial load balancing routing scheme is optimized using the network status monitoring data, and finally a load balancing routing scheme that can effectively balance the network load is obtained.

[0095] Preferably, step S1 further comprises:

[0096] Step S11: collecting satellite operation status of satellite nodes through the satellite sensor network to obtain satellite original status data;

[0097] Step S12: performing edge preprocessing on the satellite original state data to obtain satellite preprocessing state data;

[0098] Step S13: extracting state features from the satellite preprocessed state data to obtain satellite state feature data;

[0099] Step S14: performing status feature aggregation on the satellite status feature data to obtain a satellite network status data set.

[0100] In an embodiment of the present invention, the satellite sensor network regularly collects the operating status data of the satellite nodes, such as attitude information, orbit information, energy consumption, communication link status, computing resource usage, etc., through sensors carried on each satellite node. These data are stored in the local memory of the satellite node in the form of raw data. In order to reduce the amount of data transmission and the complexity of calculation, the satellite node performs edge preprocessing on the raw data. For example, the data is subjected to operations such as denoising, compression, format conversion, etc., to remove redundant information, and the data is converted into a more concise and efficient format. The satellite node uses a feature extraction algorithm to extract feature information that is more effective for network status perception and resource scheduling from the preprocessed data. For example, key features such as the load status, communication link quality, and computing power of the satellite node can be extracted and stored as feature data. The feature data of all satellite nodes are aggregated to form a complete satellite network status data set. This data set contains the key feature information of all satellite nodes and can reflect the operating status of the entire satellite network.

[0101] Preferably, step S2 further comprises:

[0102] Step S21: Collecting task history data from multiple sources on the communication participating nodes to obtain a task history data set;

[0103] Step S22: Perform user behavior analysis based on the task history data set to obtain a user behavior analysis data set;

[0104] Step S23: performing demand trend mining on the task history data set and the user behavior analysis data set to obtain a task demand trend data set;

[0105] Step S24: using a preset deep learning model and a task demand trend data set as training data to train a demand forecasting model and obtain a task demand forecasting model;

[0106] Step S25: Acquire real-time satellite network status data; use the mission demand prediction model and the real-time satellite network status data to perform mission demand prediction to obtain a mission demand prediction data set.

[0107] In an embodiment of the present invention, task history data is collected from each communication participating node, including task records submitted by users, historical task execution data, task types and parameters, etc., and stored in a unified task history data set. The task history data set is analyzed using data mining technology to identify user behavior patterns, such as task submission time, task type, task scale, task priority, etc., and a user behavior analysis data set is constructed. The task history data set and the user behavior analysis data set are combined to analyze the trend change law of task requirements, such as identifying seasonal changes, periodic changes, sudden changes, etc. in task requirements, and a task requirement trend data set is constructed. Using a preset deep learning model, such as a recurrent neural network (RNN) or a long short-term memory network (LSTM), the task requirement trend data set is used as training data to train a demand prediction model to obtain a model that can predict future task requirements. Real-time satellite network status data, such as available resources of satellite nodes, communication link status, etc., is obtained from a satellite sensor network. Task requirements are predicted using the trained task requirement prediction model and real-time satellite network status data to obtain a prediction data set containing future task requirement prediction information.

[0108] Preferably, if Figure 2 As shown, step S3 further includes the following:

[0109] Step S31: using a preset resource optimization model structure to initialize model parameters according to a satellite network status data set and a mission demand prediction data set to obtain an initial resource optimization model;

[0110] Using a preset resource optimization model structure, such as a deep learning-based resource allocation model, the model parameters are initialized according to the satellite network status dataset and the mission demand prediction dataset to obtain an initial resource optimization model.

[0111] Step S32: Perform satellite node performance analysis based on the satellite original status data to obtain node computing capacity data and node data storage capacity data; distribute the initial resource optimization model to each satellite node, divide the model parameters based on the node computing capacity data and the node data storage capacity data to obtain distributed model parameters;

[0112] The original state data of the satellite is analyzed to obtain performance indicators such as the computing power and data storage capacity of each node. The initial resource optimization model is distributed to each satellite node, and the model parameters are divided according to the performance indicators of the node to obtain the distributed model parameters.

[0113] Step S33: Performing local model training on the satellite nodes according to the distributed model parameters to obtain local encrypted model parameters; performing global model aggregation according to the local encrypted model parameters to obtain a global training model;

[0114] Each satellite node is trained based on the received distributed model parameters and locally stored data to obtain the locally trained model parameters. Each satellite node encrypts the trained model parameters locally to obtain a locally encrypted model parameter. The encryption method can use privacy protection technologies such as homomorphic encryption and differential privacy to ensure the transmission and aggregation of model parameters without leaking the original data. Then, the encrypted model parameters are uploaded to the central server for global model aggregation, and finally a global training model is obtained.

[0115] Step S34: performing a global model evaluation on the global training model to obtain global model evaluation data; adjusting model parameters of the global training model using the global model evaluation data to obtain a distributed resource optimization model;

[0116] Evaluate the global training model, for example, using a test data set to evaluate the model's performance. Adjust the parameters of the global training model based on the evaluation results, such as adjusting the model structure, optimizing hyperparameters, etc., and ultimately obtain a distributed resource optimization model that can effectively optimize resource allocation.

[0117] Preferably, step S33 further comprises:

[0118] Step S331: performing local data preprocessing on the satellite node to obtain local data preprocessing data; performing local model training according to the local data preprocessing data and the distributed model parameters to obtain a local initial model;

[0119] Step S332: performing local model compression on the local initial model to obtain a local compressed model; performing local model encryption on the local compressed model to obtain local encrypted model parameters;

[0120] Step S333: transmitting the local encrypted model parameters to the central server, and using the preset model key to perform local model decryption on the local encrypted model parameters to obtain local model decryption data;

[0121] Step S334: perform federated learning model aggregation on the local model decrypted data to obtain model aggregation data; use the preset model aggregation strategy and model aggregation data to update the global model parameters to obtain global model aggregation data; perform global model encryption on the global model aggregation data to obtain a global training model.

[0122] In an embodiment of the present invention, each satellite node performs local data preprocessing based on the local data it receives, and the preprocessing includes data cleaning, data conversion, etc. to obtain local data preprocessing data. Local data refers to data collected and stored by each satellite node itself, including: satellite operation status data, such as the satellite's position, speed, attitude, remaining energy, computing resource utilization, storage space occupancy, etc., which can be collected through the satellite's own sensor network; communication link status data, such as signal strength, signal-to-noise ratio, transmission delay, etc. with other satellite nodes, ground base stations, and user terminals, which can reflect the current network connection status; historical task data, such as the type of tasks that have been processed, data volume, computing requirements, completion time, etc., which can be used to analyze task demand trends.

[0123] Then, the preprocessed data and the model parameters distributed to itself are used for local model training to obtain a preliminary local initial model. The local initial model is compressed, for example, by using model pruning or quantization techniques to reduce the size of model parameters and reduce the amount of data transmission. Next, the compressed model parameters are encrypted, for example, by using symmetric encryption or asymmetric encryption techniques to prevent the model from being stolen or tampered with. The encrypted model parameters are transmitted to the central server, and the central server decrypts the encrypted model parameters using the preset model key to obtain the original model parameters, providing necessary data for subsequent model aggregation. The central server uses the federated learning model aggregation strategy, such as weighted average or federated average, to aggregate the decrypted model parameters of each node to obtain model aggregation data. Then, the global model parameters are updated using the preset model aggregation strategy and aggregation data, for example, by using the gradient descent algorithm to update the parameters. Finally, the updated global model parameters are encrypted to obtain a global training model.

[0124] Preferably, step S4 further comprises:

[0125] Step S41: performing sensitive information identification on the task demand prediction data set to obtain task sensitive information data;

[0126] Step S42: using the preset privacy budget data to add noise to the task sensitive information data and perform noise control to obtain the task required noise data;

[0127] Step S43: encrypting the task requirement noise data to obtain encrypted task requirement information data;

[0128] Step S44: transmitting the encrypted task requirement information data and the distributed resource optimization model to the communication participating nodes, performing protocol initialization, and obtaining protocol initialization data;

[0129] Step S45: Perform secure multi-party encryption joint calculation according to the protocol initialization data, the encryption task requirement information data and the distributed resource optimization model to obtain a resource allocation encryption scheme;

[0130] Step S46: decrypt the encrypted resource allocation scheme and perform a privacy protection evaluation to obtain a task allocation scheme.

[0131] In an embodiment of the present invention, sensitive information is identified for a task demand prediction data set, for example, data items containing sensitive information such as user personal information, task content, and task priority are identified, and these data are marked as sensitive information data. Using preset privacy budget data, noise is added to sensitive information data, for example, noise is added using methods such as Laplace mechanism or Gaussian mechanism, and the degree of noise addition is controlled according to the privacy budget to obtain task demand noise data. The task demand noise data is encrypted, for example, using symmetric encryption or asymmetric encryption to obtain encrypted task demand information data. The encrypted task demand information data and the distributed resource optimization model are transmitted to the communication participating nodes, and the protocol is initialized, for example, using a secure multi-party computing protocol to initialize, so as to ensure that a consistent computing protocol is reached between the participating nodes. According to the protocol initialization data, the encrypted task demand information data and the distributed resource optimization model, secure multi-party encrypted joint calculation is performed, for example, using secret sharing, homomorphic encryption and other technologies, without leaking any sensitive information of any party, and jointly calculating the resource allocation encryption scheme. The resource allocation encryption scheme is decrypted, for example, using a preset key to decrypt, to obtain the final task allocation scheme. At the same time, privacy protection measures are evaluated, such as using differential privacy analysis and other methods to assess the risk of leakage of sensitive information to ensure the security of the solution.

[0132] Preferably, if Figure 3 As shown, step S5 further includes:

[0133] Step S51: performing satellite network environment state perception according to the satellite network state data set, the task demand prediction data set and the task allocation plan to obtain environment state data;

[0134] In an embodiment of the present invention, a satellite network status data set, a task demand prediction data set and a task allocation scheme are integrated to construct a satellite network environment status data including satellite node status, task demand and task allocation information.

[0135] Step S52: using the reinforcement learning technology and the environmental state data to define the state space, and obtain the state space data; using the reinforcement learning technology and the environmental state data to define the action space, and obtain the action space data; using the reinforcement learning technology and the environmental state data to design the reward function, and obtain the reward function data;

[0136] In the embodiment of the present invention, the reinforcement learning technology is used to convert the environmental state data into a language that can be understood by the reinforcement learning model. For example, the available resources, communication link status, task type and priority of the satellite node are defined as the state space, and the resource scheduling actions that can be taken, such as which node to assign the task to, how many resources to assign, etc., are defined as the action space, and the reward function is designed according to different scheduling results, such as rewarding the assignment of tasks to nodes with low load, or rewarding the assignment of tasks to nodes with high resource utilization.

[0137] Step S53: Obtain historical scheduling strategy data; perform node load calculation based on the environment status data and the historical scheduling strategy data to obtain task load data;

[0138] In an embodiment of the present invention, historical scheduling strategy data is obtained, such as recording scheduling decisions previously made for different environmental states. Based on the environmental state data and historical scheduling strategy data, the load of each node is calculated, such as calculating the CPU usage rate, memory usage rate, etc. of each node.

[0139] Step S54: using an adaptive weighted scheduling optimization algorithm to perform adaptive resource scheduling weight calculation according to the task load data and the task demand prediction data set to obtain adaptive weighted scheduling data;

[0140] In the embodiment of the present invention, the constructed adaptive weight scheduling optimization algorithm is used to dynamically adjust the scheduling weight of each node according to the task load data and the task demand prediction data set.

[0141] Step S55: performing scheduling strategy learning according to the adaptive weight scheduling data, the state space data, the action space data and the reward function data to obtain a resource scheduling strategy;

[0142] In an embodiment of the present invention, reinforcement learning technology is used to train a model that can perform optimal resource scheduling according to environmental conditions based on adaptive weight scheduling data, state space data, action space data, and reward function data.

[0143] Step S56: Allocate resources to satellite nodes according to the resource scheduling strategy to obtain a resource allocation plan; execute tasks according to the resource allocation plan and monitor the execution process to obtain execution process monitoring data;

[0144] In the embodiment of the present invention, according to the trained resource scheduling strategy, resources are allocated to satellite nodes, for example, tasks are allocated to appropriate nodes, and corresponding computing resources are allocated. At the same time, the task execution process is monitored, for example, information such as task execution time and resource usage is collected.

[0145] Step S57: optimizing the resource allocation plan according to the execution process monitoring data to obtain the air computing resource scheduling plan;

[0146] In the embodiment of the present invention, the resource allocation scheme is optimized according to the execution process monitoring data, such as adjusting the node weights, optimizing the task allocation strategy, etc., to obtain a more effective in-flight computing resource scheduling scheme.

[0147] Preferably, the adaptive weight scheduling optimization algorithm in step S54 is as follows:

[0148] ;

[0149] In the formula, Indicates The adaptive weights of satellite nodes, Indicates The mission load of satellite nodes, represents the summation variable, Indicates the total number of satellite nodes, Indicates the satellite node index value, Indicates The task priority of each satellite node, Indicates The mission demand forecast value of satellite nodes, represents the angle parameter of time variation, Indicates The resource utilization of satellite nodes, represents the reward function, represents the scheduling weight adjustment factor, Represents the reward function parameters.

[0150] The present invention constructs an adaptive weight scheduling optimization algorithm, where The core goal of this formula is to calculate the weight of each satellite node, which is used to decide which node to assign tasks to during resource scheduling. The higher the weight, the more likely the node is to be assigned tasks. ,This part reflects the current load of the node. The lower the load, the higher the weight. When a node has a lower load, it has more resources to process new tasks, so the weight assigned to it should be higher. , which reflects the priority of the task. When the task priority is higher, the value of this part is larger and the weight is higher. This means that high-priority tasks should be assigned to nodes with sufficient resources. , which reflects the node's prediction of future task demand. When the node estimates that the future demand is low, the larger the value of this part, the higher the weight. This means that nodes with low estimated demand have more resources to process new tasks. ,This part takes into account the impact of time changes.,For example, in certain time periods, the task demand will increase, and at this time, the node weight can be appropriately increased to cope with the higher demand. , which reflects the resource utilization of the node. When the resource utilization of the node is high, the larger the value of this part, the higher the weight. This means that nodes with high resource utilization should be assigned to tasks to make full use of resources. ,This part reflects the influence of the reward function. The reward function can be designed according to different scheduling objectives. For example, it can reward tasks to be assigned to nodes with low load, or reward tasks to be assigned to nodes with high resource utilization. and is an adjustment factor that can be adjusted according to specific circumstances to balance the weights between different objectives. This formula takes into account factors such as task load, task priority, task demand forecast, time change, resource utilization, and reward function. By calculating the weight of each node, adaptive resource scheduling is achieved, and resource allocation strategies can be dynamically adjusted according to different situations to improve resource utilization and system efficiency.

[0151] Preferably, step S6 further comprises:

[0152] Step S61: performing satellite network topology perception according to the satellite network status data set and the air computing resource scheduling scheme to obtain satellite network topology data;

[0153] Step S62: using the task allocation scheme and the air computing resource scheduling scheme to perform network traffic prediction and obtain traffic prediction data;

[0154] Step S63: performing routing strategy learning according to the satellite network topology data and the traffic prediction data to obtain routing strategy learning data; performing routing evaluation index calculation on the routing strategy learning data to obtain routing strategy data;

[0155] Step S64: performing data packet path planning according to the routing strategy data to obtain optimal routing path data; performing path adjustment according to the optimal routing path data and the satellite network status data set to obtain a load balancing routing initial solution;

[0156] Step S65: Perform data transmission and network status monitoring according to the load balancing routing initial plan to obtain network status monitoring data; use the network status monitoring data to optimize the routing of the load balancing routing initial plan to obtain a load balancing routing plan to implement load balancing routing operations in the satellite network.

[0157] In an embodiment of the present invention, a satellite network topology is constructed according to a satellite network state data set and an air computing resource scheduling scheme, for example, information such as connection relationships between satellite nodes, link bandwidth, node computing capabilities, etc. is recorded to form satellite network topology data. According to a task allocation scheme and an air computing resource scheduling scheme, traffic changes between different time periods and different nodes are predicted, for example, data transmission volume is predicted according to factors such as task size, task type, and node processing capabilities to form traffic prediction data. According to the satellite network topology data and traffic prediction data, a model capable of dynamically adjusting routing strategies according to network status and traffic changes is trained, for example, using a deep learning model or a reinforcement learning model. Then, evaluation indicators such as delay, packet loss rate, and throughput are used to evaluate the learned routing strategies to obtain the best routing strategy data. According to the routing strategy data, the best transmission path of the data packet from the source node to the target node is planned, for example, the path planning is performed using methods such as the shortest path algorithm or the minimum delay algorithm. At the same time, according to the real-time network status, such as link bandwidth, node load, etc., the path is dynamically adjusted to achieve load balancing and form a load balancing routing initial scheme. Data is transmitted according to the initial load balancing routing scheme, and the network status is monitored in real time, such as recording link bandwidth, node load, etc. The initial load balancing routing scheme is optimized using network status monitoring data, such as adjusting the path, allocating traffic, etc., and finally a load balancing routing scheme that can effectively balance the network load is obtained.

[0158] The present invention also provides a load balancing routing system for a satellite network, the system comprising:

[0159] The satellite network status perception module is used to collect satellite operation status of satellite nodes through the satellite sensor network to obtain satellite original status data; perform network status perception on the satellite original status data to obtain a satellite network status data set;

[0160] The task demand prediction module is used to mine the task demand of the communication participating nodes to obtain a task demand trend data set; perform task demand prediction modeling and task demand prediction based on the preset deep learning model and the task demand trend data set to obtain a task demand prediction data set;

[0161] The federated learning collaborative optimization module is used to initialize and distribute the model according to the satellite network status data set and the mission demand prediction data set to obtain the distributed model parameters; perform local model training on the satellite nodes according to the distributed model parameters to obtain the local encrypted model parameters; perform global model aggregation according to the local encrypted model parameters to obtain the global training model; perform model evaluation on the global training model to obtain the distributed resource optimization model;

[0162] The privacy-preserving task offloading module is used to encrypt the task demand prediction data set to obtain encrypted task demand information data; perform secure multi-party computing processing and privacy protection evaluation based on the encrypted task demand information data and the distributed resource optimization model to obtain a task allocation plan;

[0163] The air computing resource scheduling module is used to perceive the satellite network environment status according to the satellite network status data set, the task demand prediction data set and the task allocation plan to obtain the environment status data; use the reinforcement learning technology and the environment status data to perform resource scheduling learning to obtain the resource scheduling strategy; perform air computing resource scheduling on the satellite nodes according to the resource scheduling strategy to obtain the air computing resource scheduling plan;

[0164] The load balancing routing module is used to perform routing strategy traffic analysis based on the satellite network status data set, task allocation plan and air computing resource scheduling plan to obtain routing strategy data; perform load balancing adjustment based on the routing strategy data to obtain a load balancing routing plan to achieve load balancing routing operations in the satellite network.

[0165] Therefore, the embodiments should be regarded as illustrative and non-restrictive from all points, and the scope of the present invention is limited by the appended claims rather than the above description, and it is therefore intended that all changes falling within the meaning and range of equivalent elements of the application documents are included in the present invention.

[0166] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A load balancing routing method for a satellite network, characterized in that: The method comprises: Step S1: collect satellite operation status of satellite nodes through the satellite sensor network to obtain satellite original status data; perform network status perception on the satellite original status data to obtain a satellite network status data set; Step S2: mining the task requirements of the communication participating nodes to obtain a task requirement trend data set; performing task requirement prediction modeling and task requirement prediction based on a preset deep learning model and the task requirement trend data set to obtain a task requirement prediction data set; Step S3: Initialize and distribute the model according to the satellite network status data set and the mission demand prediction data set to obtain distributed model parameters; perform local model training on the satellite nodes according to the distributed model parameters to obtain local encrypted model parameters; perform global model aggregation according to the local encrypted model parameters to obtain a global training model; perform model evaluation on the global training model to obtain a distributed resource optimization model; Step S4: encrypting the task demand prediction data set to obtain encrypted task demand information data; performing secure multi-party computing processing and privacy protection evaluation based on the encrypted task demand information data and the distributed resource optimization model to obtain a task allocation plan; Step S5: According to the satellite network status data set, the task demand prediction data set and the task allocation plan, the satellite network environment status is sensed to obtain the environment status data; the reinforcement learning technology and the environment status data are used to perform resource scheduling learning to obtain the resource scheduling strategy; according to the resource scheduling strategy, the satellite nodes are scheduled for air computing resources to obtain the air computing resource scheduling plan; Step S6: Perform routing policy traffic analysis based on the satellite network status data set, task allocation plan and air computing resource scheduling plan to obtain routing policy data; perform load balancing adjustment based on the routing policy data to obtain a load balancing routing plan to implement load balancing routing operations in the satellite network.

2. The method according to claim 1, characterized in that The step S1 further comprises: Step S11: collecting satellite operation status of satellite nodes through the satellite sensor network to obtain satellite original status data; Step S12: performing edge preprocessing on the satellite original state data to obtain satellite preprocessing state data; Step S13: extracting state features from the satellite preprocessed state data to obtain satellite state feature data; Step S14: performing status feature aggregation on the satellite status feature data to obtain a satellite network status data set.

3. The method according to claim 1, characterized in that The step S2 further comprises: Step S21: Collecting task history data from multiple sources on the communication participating nodes to obtain a task history data set; Step S22: Perform user behavior analysis based on the task history data set to obtain a user behavior analysis data set; Step S23: performing demand trend mining on the task history data set and the user behavior analysis data set to obtain a task demand trend data set; Step S24: using a preset deep learning model and a task demand trend data set as training data to train a demand forecasting model and obtain a task demand forecasting model; Step S25: Acquire real-time satellite network status data; use the mission demand prediction model and the real-time satellite network status data to perform mission demand prediction to obtain a mission demand prediction data set.

4. The method according to claim 1, characterized in that: The step S3 further comprises: Step S31: using a preset resource optimization model structure to initialize model parameters according to a satellite network status data set and a mission demand prediction data set to obtain an initial resource optimization model; Step S32: Perform satellite node performance analysis based on the satellite original status data to obtain node computing capacity data and node data storage capacity data; distribute the initial resource optimization model to each satellite node, divide the model parameters based on the node computing capacity data and the node data storage capacity data to obtain distributed model parameters; Step S33: Performing local model training on the satellite nodes according to the distributed model parameters to obtain local encrypted model parameters; performing global model aggregation according to the local encrypted model parameters to obtain a global training model; Step S34: perform a global model evaluation on the global training model to obtain global model evaluation data; use the global model evaluation data to adjust model parameters of the global training model to obtain a distributed resource optimization model.

5. The method according to claim 4, characterized in that The step S33 further comprises: Step S331: performing local data preprocessing on the satellite node to obtain local data preprocessing data; performing local model training according to the local data preprocessing data and the distributed model parameters to obtain a local initial model; Step S332: performing local model compression on the local initial model to obtain a local compressed model; performing local model encryption on the local compressed model to obtain local encrypted model parameters; Step S333: transmitting the local encrypted model parameters to the central server, and using the preset model key to perform local model decryption on the local encrypted model parameters to obtain local model decryption data; Step S334: perform federated learning model aggregation on the local model decrypted data to obtain model aggregation data; use the preset model aggregation strategy and model aggregation data to update the global model parameters to obtain global model aggregation data; perform global model encryption on the global model aggregation data to obtain a global training model.

6. The method according to claim 1, characterized in that The step S4 further comprises: Step S41: performing sensitive information identification on the task demand prediction data set to obtain task sensitive information data; Step S42: using the preset privacy budget data to add noise to the task sensitive information data and perform noise control to obtain the task required noise data; Step S43: encrypting the task requirement noise data to obtain encrypted task requirement information data; Step S44: transmitting the encrypted task requirement information data and the distributed resource optimization model to the communication participating nodes, performing protocol initialization, and obtaining protocol initialization data; Step S45: Perform secure multi-party encryption joint calculation according to the protocol initialization data, the encryption task requirement information data and the distributed resource optimization model to obtain a resource allocation encryption scheme; Step S46: decrypt the encrypted resource allocation scheme and perform a privacy protection evaluation to obtain a task allocation scheme.

7. The method according to claim 1, characterized in that The step S5 further comprises: Step S51: performing satellite network environment state perception according to the satellite network state data set, the task demand prediction data set and the task allocation plan to obtain environment state data; Step S52: using the reinforcement learning technology and the environmental state data to define the state space, and obtain the state space data; using the reinforcement learning technology and the environmental state data to define the action space, and obtain the action space data; using the reinforcement learning technology and the environmental state data to design the reward function, and obtain the reward function data; Step S53: Obtain historical scheduling strategy data; perform node load calculation based on the environment status data and the historical scheduling strategy data to obtain task load data; Step S54: using an adaptive weighted scheduling optimization algorithm to perform adaptive resource scheduling weight calculation according to the task load data and the task demand prediction data set to obtain adaptive weighted scheduling data; Step S55: performing scheduling strategy learning according to the adaptive weight scheduling data, the state space data, the action space data and the reward function data to obtain a resource scheduling strategy; Step S56: Allocate resources to satellite nodes according to the resource scheduling strategy to obtain a resource allocation plan; execute tasks according to the resource allocation plan and monitor the execution process to obtain execution process monitoring data; Step S57: Optimize the resource allocation plan according to the execution process monitoring data to obtain the air computing resource scheduling plan.

8. The method according to claim 7, characterized in that The adaptive weight scheduling optimization algorithm in step S54 is: ； in, Indicates The adaptive weights of satellite nodes, Indicates The mission load of satellite nodes, represents the summation variable, Indicates the total number of satellite nodes, Indicates the satellite node index value, Indicates The task priority of each satellite node, Indicates The mission demand forecast value of satellite nodes, represents the angle parameter of time variation, Indicates The resource utilization of satellite nodes, represents the reward function, represents the scheduling weight adjustment factor, Represents the reward function parameters.

9. The method according to claim 1, characterized in that: The step S6 further comprises: Step S61: performing satellite network topology perception according to the satellite network status data set and the air computing resource scheduling scheme to obtain satellite network topology data; Step S62: using the task allocation scheme and the air computing resource scheduling scheme to perform network traffic prediction and obtain traffic prediction data; Step S63: performing routing strategy learning according to the satellite network topology data and the traffic prediction data to obtain routing strategy learning data; performing routing evaluation index calculation on the routing strategy learning data to obtain routing strategy data; Step S64: performing data packet path planning according to the routing strategy data to obtain optimal routing path data; performing path adjustment according to the optimal routing path data and the satellite network status data set to obtain a load balancing routing initial solution; Step S65: Perform data transmission and network status monitoring according to the load balancing routing initial plan to obtain network status monitoring data; use the network status monitoring data to optimize the routing of the load balancing routing initial plan to obtain a load balancing routing plan to implement load balancing routing operations in the satellite network.

10. A load balancing routing system for a satellite network, characterized in that: The system includes a satellite network state perception module, a task demand prediction module, a federated learning collaborative optimization module, a privacy protection task offloading module, an air computing resource scheduling module and a load balancing routing module, which are used to execute the load balancing routing method of the satellite network as claimed in claim 1, specifically: The satellite network status perception module is used to collect satellite operation status of satellite nodes through the satellite sensor network to obtain satellite original status data; perform network status perception on the satellite original status data to obtain a satellite network status data set; The task demand prediction module is used to mine the task demand of the communication participating nodes to obtain a task demand trend data set; perform task demand prediction modeling and task demand prediction based on the preset deep learning model and the task demand trend data set to obtain a task demand prediction data set; The federated learning collaborative optimization module is used to initialize and distribute the model based on the satellite network status data set and the mission demand prediction data set to obtain distributed model parameters; perform local model training on satellite nodes based on the distributed model parameters to obtain local encrypted model parameters; perform global model aggregation based on the local encrypted model parameters to obtain a global training model; Perform model evaluation on the global training model to obtain a distributed resource optimization model; The privacy-preserving task offloading module is used to encrypt the task demand prediction data set to obtain encrypted task demand information data; perform secure multi-party computing processing and privacy protection evaluation based on the encrypted task demand information data and the distributed resource optimization model to obtain a task allocation plan; The air computing resource scheduling module is used to perceive the satellite network environment status according to the satellite network status data set, the task demand prediction data set and the task allocation plan to obtain the environment status data; Use reinforcement learning technology and environmental status data to learn resource scheduling and obtain resource scheduling strategies; schedule satellite nodes for air computing resources according to the resource scheduling strategies and obtain air computing resource scheduling solutions; The load balancing routing module is used to perform routing strategy traffic analysis based on the satellite network status data set, task allocation plan, and air computing resource scheduling plan to obtain routing strategy data; Load balancing adjustment is performed according to the routing strategy data to obtain a load balancing routing solution to achieve load balancing routing operations in the satellite network.

Images