Satellite resource dynamic allocation method and system under cloud-native hybrid cloud architecture

Abstract

The application provides a satellite resource dynamic allocation method under a cloud native hybrid cloud architecture, encapsulates satellite payload resources and cloud resources through a containerization technology to form a unified virtual resource pool, constructs a deep learning model, predicts resource demand of each beam coverage area in a future period based on satellite task historical data and real-time demand, generates a resource matching scheme, dynamically adjusts a beam radius, channel allocation and time slot scheduling by using a reinforcement learning algorithm to match real-time load changes of the cloud resources, constructs a satellite and cloud digital twin model, simulates and predicts a scheduling strategy based on the resource matching scheme through the satellite and cloud digital twin model, mirrors a resource state in real time, and updates a strategy library according to simulation feedback data of the satellite and cloud digital twin model to realize automatic reorganization of multiple satellite tasks. The application also provides a satellite resource dynamic allocation system under the cloud native hybrid cloud architecture and a computer storage medium. Accordingly, the application realizes three-dimensional cooperation of dynamic resource pooling, intelligent prediction and elastic scheduling, and can effectively solve the problem that satellite task response time, resource utilization and safety cannot be achieved simultaneously.

Description

Technical Field

[0001] This invention relates to the field of satellite communication technology, and in particular to a method, system, and computer storage medium for dynamic allocation of satellite resources under a cloud-native hybrid cloud architecture. Background Technology

[0002] In the current satellite resource management environment, satellite resources (including transponders, orbits, power, etc.) are dispersed across heterogeneous cloud environments. This decentralized resource distribution leads to numerous problems. First, cross-platform scheduling is inefficient; resource coordination and scheduling between different cloud platforms consume significant time and effort, making rapid and efficient resource allocation difficult. Second, response latency is a serious issue. This latency not only affects the timely execution of satellite missions but may also lead to delays in critical tasks, thereby impacting the operational efficiency and reliability of the entire satellite system.

[0003] Traditional satellite resource allocation strategies primarily employ pre-allocation. This approach proves particularly inadequate when facing sudden mission demands. Because it cannot dynamically adjust resource allocation based on real-time mission requirements, it often leads to insufficient satellite resource utilization. Although existing technologies can improve resource utilization by optimizing traditional pre-allocation strategies, the improvement is limited. This inefficient resource utilization not only wastes resources but also restricts the overall performance and economic efficiency of the satellite system.

[0004] In data processing, the handling of sensitive data faces a dilemma. On the one hand, while private clouds can provide a relatively secure computing environment, their limited computing power makes it difficult to meet the needs of large-scale, complex data processing. On the other hand, while public clouds offer powerful computing capabilities, they pose a risk of data leakage and cannot guarantee the security of sensitive data. This contradiction between security and performance makes it difficult for satellite systems to find a balance in data processing, thus affecting the overall operational effectiveness of the satellite system and the extraction of data value.

[0005] Therefore, there is an urgent need for a method that can effectively resolve the contradiction between satellite mission response time, resource utilization, and security. Summary of the Invention

[0006] To address the aforementioned shortcomings, the present invention aims to provide a method, system, and computer storage medium for dynamic allocation of satellite resources under a cloud-native hybrid cloud architecture. By constructing a three-dimensional collaborative architecture of dynamic resource pooling, intelligent prediction, and elastic scheduling, the invention effectively resolves the contradiction between satellite mission response time, resource utilization, and security.

[0007] To achieve the aforementioned technical effects, this invention provides a method for dynamic allocation of satellite resources under a cloud-native hybrid cloud architecture, comprising the following steps:

[0008] Satellite payload resources and cloud resources are encapsulated using containerization technology to form a unified virtual resource pool; wherein, the cloud resources include computing nodes, storage capacity and network bandwidth of public cloud and private cloud;

[0009] A deep learning model is constructed, and the resource demand of each beam coverage area in the future period is predicted based on historical satellite mission data and real-time demand, and a resource matching scheme is generated.

[0010] Reinforcement learning algorithms are used to dynamically adjust beam radius, channel allocation, and time slot scheduling to match the real-time load changes of the cloud resources;

[0011] A satellite-cloud digital twin model is constructed. Based on the resource matching scheme, the satellite-cloud digital twin model is used to simulate and predict scheduling strategies and mirror resource status in real time. The strategy library is updated according to the simulation feedback data of the satellite-cloud digital twin model to achieve automatic reorganization of multi-satellite missions.

[0012] Optionally, constructing the deep learning model includes:

[0013] Construct a generator G, the input of which is a random noise sequence Z = {z1, z2, z1, ..., z2}. T The output is time series data. Where, θ G z is a parameter of the generator G. t It is random noise at time point t, where T is the length of the sequence;

[0014] Construct a discriminator D, the input of which is real time series data X = {x1, x2, x1, ..., x2}. T} or the time series data The output is the probability of the input data being true; where x t These are the observations at time point t, where T is the length of the sequence;

[0015] By alternately optimizing the discriminant loss function L D and generator loss function L G Train the model, where:

[0016]

[0017] P data For the distribution of real data, P Z The distribution of random noise;

[0018] After the model training is completed, the generator G predicts the resource demand of each beam coverage area in the future period based on the historical data of satellite missions and real-time requirements, and generates a resource matching scheme.

[0019] Optionally, the step of dynamically adjusting the beam radius, channel allocation, and time slot scheduling using reinforcement learning algorithms includes:

[0020] Initialize the policy network parameters θ and the value network parameters φ;

[0021] Predefined reinforcement learning elements, including state s t Action a t and the reward function r t , where state s t Includes satellite beam coverage information, channel allocation status, time slot allocation status, and user service requirements; Action a t This represents a dynamic adjustment command for beam radius, channel allocation, or time slot scheduling; the reward function r t Calculated based on a comprehensive assessment of task completion rate, resource utilization rate, and data transmission quality;

[0022] The objective function is constructed as follows:

[0023]

[0024] in, For the dominant function, It is a state-action value function. Value network estimation in state s t Expected return under; π θ (a t |s t ) indicates that in state s t Take action a t The probability, Let L be the probability of the old strategy, ∈ be the parameter of the clipping range, and L be the probability of the old strategy. VF (φ) is the loss function of the value function, L ENT (θ) is the entropy regularization term, and c1 and c2 are weight coefficients. It is the expected value;

[0025] The policy network parameters θ are updated by maximizing the objective function L(θ);

[0026] By minimizing the loss function L of the value function VF (φ) Update the value network parameter φ.

[0027] On the other hand, based on the same inventive structure, this invention also provides a satellite resource dynamic allocation system under a cloud-native hybrid cloud architecture, including a resource virtualization module, a resource prediction and scheduling module, a cloud-native management module, and a hybrid cloud interface module; wherein:

[0028] The resource virtualization module is configured to encapsulate satellite payload resources and cloud resources using containerization technology to form a unified virtual resource pool; wherein, the cloud resources include computing nodes, storage capacity and network bandwidth of public cloud and private cloud;

[0029] The resource prediction and scheduling module is configured to predict the resource demand of each beam coverage area in the future period based on historical satellite mission data and real-time demand using a deep learning model, and generate a resource matching scheme; and to dynamically adjust the beam radius, channel allocation and time slot scheduling using a reinforcement learning algorithm to match the real-time load changes of the cloud resources.

[0030] The cloud-native management module is configured to simulate and predict scheduling strategies based on the resource matching scheme through a satellite and cloud digital twin model and to mirror resource status in real time, and to update the strategy library according to the simulation feedback data of the satellite and cloud digital twin model, so as to realize automatic reorganization of multi-satellite missions.

[0031] The hybrid cloud interface module is configured to enable seamless integration of satellite resources with multiple cloud platforms.

[0032] Optionally, the resource prediction and scheduling module specifically includes:

[0033] The spatiotemporal prediction submodule is configured to predict user needs based on historical task data and / or historical user needs, using optimization algorithms or deep learning models.

[0034] The scheduling submodule is configured to dynamically adjust the resource allocation scheme based on the prediction results of the spatiotemporal prediction submodule, as well as user needs, satellite resource status, and other information.

[0035] Optionally, the spatiotemporal prediction submodule is specifically used for:

[0036] Construct a generator G, the input of which is a random noise sequence Z = {z1, z2, z1, ..., z2}. T The output is time series data. Where, θ G z is a parameter of the generator G. t It is random noise at time point t, where T is the length of the sequence;

[0037] Construct a discriminator D, the input of which is real time series data X = {x1, x2, x1, ..., x2}. T} or the time series data The output is the probability of the input data being true; where x t These are the observations at time point t, where T is the length of the sequence;

[0038] By alternately optimizing the discriminant loss function LD and generator loss function L G Train the model, where:

[0039]

[0040] P data For the distribution of real data, P Z The distribution of random noise;

[0041] After the model training is completed, the generator G predicts the resource demand of each beam coverage area in the future period based on the historical data of satellite missions and real-time requirements, and generates a resource matching scheme.

[0042] Optionally, the scheduling submodule is specifically used for:

[0043] Initialize the policy network parameters θ and the value network parameters φ;

[0044] Predefined reinforcement learning elements, including state s t Action a t and the reward function r t , where state s t Includes satellite beam coverage information, channel allocation status, time slot allocation status, and user service requirements; Action a t This represents a dynamic adjustment command for beam radius, channel allocation, or time slot scheduling; the reward function r t Calculated based on a comprehensive assessment of task completion rate, resource utilization rate, and data transmission quality;

[0045] The objective function is constructed as follows:

[0046]

[0047] in, For the dominant function, It is a state-action value function. Value network estimation in state s t Expected return under; π θ (a t |s t ) indicates that in state s t Take action a t The probability, Let L be the probability of the old strategy, ∈ be the parameter of the clipping range, and L be the probability of the old strategy. VF (φ) is the loss function of the value function, L ENT (θ) is the entropy regularization term, and c1 and c2 are weight coefficients. It is the expected value;

[0048] The policy network parameters θ are updated by maximizing the objective function L(θ);

[0049] By minimizing the loss function L of the value function VF (φ) Update the value network parameter φ.

[0050] Optionally, the cloud-native management module specifically includes:

[0051] The data acquisition submodule is configured to connect to sensors and devices in the satellite constellation and hybrid cloud via IoT technology, collect data on the satellite's position, status, mission execution, and cloud resource usage in real time, calculate the constellation topology, obtain the global satellite status and global resource usage, and send them to the multi-source data fusion submodule.

[0052] The communication monitoring submodule is configured to collect the bandwidth and delay of inter-satellite access throughout the entire constellation topology, generate a delay matrix between satellites, and send it to the multi-source data fusion submodule.

[0053] The multi-source data fusion submodule is configured to collect information from the data acquisition submodule and the communication monitoring submodule, and generate a dynamic panoramic resource view of the satellite constellation and hybrid cloud based on the global resource usage, the global satellite status, and the delay matrix.

[0054] The simulation submodule is configured to simulate the operation of satellite constellations and hybrid clouds in different scenarios using simulation technology.

[0055] The real-time feedback submodule is configured to compare the simulation results of the simulation submodule with the actual operating data and adjust the satellite mission and cloud resource configuration in real time.

[0056] The strategy pre-play submodule is configured to allow users to test and verify different scheduling strategies in a virtual environment.

[0057] Optionally, the hybrid cloud interface module specifically includes:

[0058] The cloud platform interface submodule is configured to interface with various mainstream cloud platforms to achieve resource integration and management.

[0059] The elastic scaling submodule is configured to dynamically adjust the scale of cloud computing resources according to resource demand in order to meet the real-time requirements of tasks.

[0060] The security isolation submodule is configured to ensure the security of resources and data between the satellite system and the cloud platform, preventing data leakage and malicious attacks.

[0061] The present invention also provides a computer storage medium storing a computer program thereon, wherein the computer program, when executed by a processor, implements the dynamic allocation method for satellite resources under a cloud-native hybrid cloud architecture as described in any of the preceding claims.

[0062] Compared with existing technologies, this invention offers at least the following advantages: It constructs a unified resource view through comprehensive analysis and virtualization of satellite resources. Employing a machine learning-based predictive model, combined with historical satellite mission data and current mission requirements, it estimates the resource demand trends for different missions and rationally schedules the resources of the satellite communication system. By monitoring resource utilization efficiency in real time and adjusting allocation strategies accordingly, it achieves dynamic optimization of satellite resource allocation within a cloud-native hybrid cloud architecture, improving resource utilization and mission adaptability, and ensuring stable operation of satellite services. Attached Figure Description

[0063] Figure 1 This is a flowchart illustrating the steps of the dynamic satellite resource allocation method under the cloud-native hybrid cloud architecture provided in the first embodiment of the present invention.

[0064] Figure 2 This is a structural block diagram of the satellite resource dynamic allocation system under the cloud-native hybrid cloud architecture provided in the second embodiment of the present invention;

[0065] Figure 3 This is a structural block diagram of the resource prediction and scheduling module of the satellite resource dynamic allocation system under the cloud-native hybrid cloud architecture provided in the second embodiment of the present invention;

[0066] Figure 4 This is a structural block diagram of the cloud-native management module of the satellite resource dynamic allocation system under the cloud-native hybrid cloud architecture provided in the second embodiment of the present invention;

[0067] Figure 5 The structural block diagram of the hybrid cloud interface module of the satellite resource dynamic allocation system under the cloud-native hybrid cloud architecture provided in the second embodiment of the present invention. Detailed Implementation

[0068] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0069] It should be noted that references to "an embodiment," "embodiment," "example embodiment," etc., in this specification refer to the described embodiment including specific features, structures, or characteristics, but not every embodiment must include these specific features, structures, or characteristics. Furthermore, such expressions do not refer to the same embodiment. Moreover, when describing specific features, structures, or characteristics in conjunction with embodiments, whether or not explicitly described, it is indicated that incorporating such features, structures, or characteristics into other embodiments is within the knowledge of those skilled in the art.

[0070] Furthermore, certain terms are used in the specification and subsequent claims to refer to specific components or parts. Those skilled in the art will understand that manufacturers may use different names or terms to refer to the same component or part. This specification and subsequent claims do not distinguish components or parts by differences in name, but rather by differences in function. The terms "comprising" and "including" used throughout the specification and subsequent claims are open-ended and should be interpreted as "including but not limited to." Additionally, the term "connection" here includes any direct and indirect electrical connection means. Indirect electrical connection means include connections made through other means.

[0071] Before describing the embodiments of this application in detail, the technical concept of this application will be briefly described first:

[0072] This invention achieves dynamic three-dimensional collaborative optimization of satellite resources through a cloud-native hybrid cloud architecture, specifically including: 1. Resource virtualization: Abstracting satellite payloads (computing / storage / communication) and cloud resources (public / private cloud) into a unified resource pool using containerization technology, breaking down the boundaries between satellite and ground resources. 2. Intelligent prediction: Utilizing generative adversarial networks (GANs) to analyze historical mission data, predicting future resource demands in beam areas, and generating allocation plans. 3. Elastic scheduling: Employing near-end strategy optimization algorithms to adjust beam radius, channel allocation, and time slot scheduling in real time to match changes in cloud load. 4. Digital twin closed-loop construction of a satellite-cloud digital twin model, using communication delay matrix and panoramic resource view simulation strategies, combined with real-time feedback to dynamically optimize task reorganization. This invention's three-dimensional collaborative mechanism (virtualization + prediction + scheduling) effectively solves the contradictions of resource dispersion, response latency, and low utilization; utilizing integrated satellite-ground resource scheduling, sensitive data is stored in the private cloud, while computationally intensive tasks are transferred to the public cloud, balancing security and performance.

[0073] The specific principles of the satellite resource dynamic allocation method under the cloud-native hybrid cloud architecture of this application will be described below with reference to specific embodiments.

[0074] Example 1

[0075] Figure 1This invention illustrates a method for dynamic allocation of satellite resources under a cloud-native hybrid cloud architecture, comprising the following steps:

[0076] S101: Satellite payload resources and cloud resources are encapsulated using containerization technology to form a unified virtual resource pool; wherein, the cloud resources include computing nodes, storage capacity and network bandwidth of public cloud and private cloud.

[0077] The satellite payload resources include hardware resources such as computing, storage, and communication.

[0078] Specifically, containerization technology abstracts the hardware and cloud resources of satellite payloads into flexibly allocable virtual resources; at the same time, it incorporates computing nodes, storage capacity, network bandwidth, etc. in public or private clouds into unified management; this virtualization and integration approach breaks down the boundaries between satellite resources and cloud computing resources, forming a unified resource pool.

[0079] S102: Construct a deep learning model to predict the resource demand of each beam coverage area in the future based on historical satellite mission data and real-time requirements, and generate a resource matching scheme. In other words, by constructing a deep learning model, the resource demand of each beam coverage area in the future is predicted, and the optimal matching scheme is generated.

[0080] In an optional implementation, constructing the deep learning model includes:

[0081] Construct a generator G, the input of which is a random noise sequence Z = {z1, z2, z1, ..., z2}. T The output is time series data. Where, θ G z is a parameter of the generator G. t It is random noise at time point t, where T is the length of the sequence;

[0082] Construct a discriminator D, the input of which is real time series data X = {x1, x2, x1, ..., x2}. T} or the time series data The output is the probability of the input data being true; where x t These are the observations at time point t, where T is the length of the sequence;

[0083] By alternately optimizing the discriminant loss function L D and generator loss function L G Train the model, where:

[0084]

[0085] P dataFor the distribution of real data, P Z The distribution of random noise;

[0086] After the model training is completed, the generator G predicts the resource demand of each beam coverage area in the future period based on the historical data of satellite missions and real-time requirements, and generates a resource matching scheme.

[0087] Generator G generates time series data, taking random noise Z as input and outputting a generated time series.

[0088] Discriminator D is used to distinguish between real data X and generated data. It takes time series data as input and outputs a probability value representing the probability that the input data is the true data. Mathematically, the discriminator D can be represented as: p = D(X; θ) D )or Where, θ D These are the parameters of the discriminator D.

[0089] The loss function L of discriminator D D The aim is to maximize the ability to distinguish between real and generated data; the loss function L of the generator G. G The aim is to minimize the discriminator D's ability to recognize generated data, that is, to make the generated data as real data as possible.

[0090] Alternate optimization of discriminator loss function L D and generator loss function L G This involves alternating training of the discriminator D and the generator G, specifically including:

[0091] Fixed θ G θ is updated using gradient descent. D To minimize L D ;

[0092] Fixed θ D θ is updated using gradient descent. G To minimize L G .

[0093] After training, a generator G is used to predict future time series data. Given a random noise sequence Z, the generator G will output a predicted time series. Therefore, it is possible to effectively predict and generate time-series data on user demand, so as to achieve forward-looking and scientific resource allocation.

[0094] S103: Employ reinforcement learning algorithms to dynamically adjust beam radius, channel allocation, and time slot scheduling to match the real-time load changes of the cloud resources.

[0095] In an optional implementation, step S103 includes:

[0096] Initialize the policy network parameters θ and the value network parameters φ;

[0097] Predefined reinforcement learning elements, including state s t Action a t and the reward function r t , where state s t Includes satellite beam coverage information, channel allocation status, time slot allocation status, and user service requirements; Action a t This represents a dynamic adjustment command for beam radius, channel allocation, or time slot scheduling; the reward function r t Calculated based on a comprehensive assessment of task completion rate, resource utilization rate, and data transmission quality;

[0098] Under the current strategy, interaction with the environment yields a series of states, actions, and rewards {s}. t a t r t}

[0099] Policy function π θ (a t |s t ) indicates that in state s t Take action a t The probability is output by the policy network in the algorithm.

[0100] Based on the sampled data, calculate the advantage function A(s) for each state-action pair. t a t ).

[0101] The objective function is constructed as follows:

[0102]

[0103] in, For the dominant function, It is a state-action value function. Value network estimation in state s t Expected return under; π θ (a t |s t ) indicates that in state s t Take action a t The probability, Let L be the probability of the old strategy, ∈ be the parameter of the clipping range, and L be the probability of the old strategy. VF (φ) is the loss function of the value function, L ENT (θ) is the entropy regularization term, and c1 and c2 are weight coefficients. It is the expected value;

[0104] The objective function combines policy optimization, value function update, and entropy regularization.

[0105] The policy network parameters θ are updated by maximizing the objective function L(θ);

[0106] By minimizing the loss function L of the value function VF (φ) Update the value network parameter φ.

[0107] Furthermore, repeated sampling, calculation of the advantage function, policy update, and value function update are performed until the policy converges. Corresponding actions are taken based on the current state of the satellite communication system to achieve reasonable scheduling of the satellite communication system's resources.

[0108] S104: Construct a satellite-cloud digital twin model. Based on the resource matching scheme, simulate and predict scheduling strategies using this model and mirror resource status in real time. Update the strategy library based on simulation feedback data from the satellite-cloud digital twin model to achieve automatic reconfiguration of multiple satellite missions. In other words, by constructing a satellite-cloud digital twin model, pre-simulating scheduling strategies and mirroring resource status in real time, and combining this with online feedback to update the strategy library, automatic reconfiguration of multiple satellite missions can be achieved.

[0109] Example 2

[0110] Figure 2 This invention illustrates a satellite resource dynamic allocation system 100 based on a cloud-native hybrid cloud architecture, comprising a resource virtualization module 10, a resource prediction and scheduling module 20, a cloud-native management module 30, and a hybrid cloud interface module 40; wherein:

[0111] The resource virtualization module 10 is configured to encapsulate satellite payload resources and cloud resources using containerization technology to form a unified virtual resource pool; wherein, the cloud resources include computing nodes, storage capacity, and network bandwidth of public and private clouds; the resource prediction and scheduling module 20 is configured to predict the resource demand of each beam coverage area in the future period based on historical satellite mission data and real-time demand using a deep learning model, and generate a resource matching scheme; and to dynamically adjust the beam radius, channel allocation, and time slot scheduling using a reinforcement learning algorithm to match the real-time load changes of the cloud resources; the cloud-native management module 30 is configured to simulate and predict the scheduling strategy based on the resource matching scheme using a satellite and cloud digital twin model and to mirror the resource status in real time, and to update the strategy library according to the simulation feedback data of the satellite and cloud digital twin model to achieve automatic reorganization of multiple satellite missions; the hybrid cloud interface module 40 is configured to achieve seamless integration between satellite resources and multiple cloud platforms.

[0112] Specifically, the resource virtualization module 10 uses containerization technology to abstract the hardware resources of the satellite payload (such as sensors, communication equipment, and processing units on the satellite) into flexibly allocable virtual resources. Simultaneously, it is tightly connected to the hybrid cloud interface, bringing resources in the cloud platform (such as computing nodes, storage capacity, and network bandwidth in public or private clouds) under unified management. This virtualization and integration approach breaks down the boundaries between satellite resources and cloud computing resources, forming a unified resource pool. The resource prediction and scheduling module 20 uses optimization algorithms or machine learning models to predict user needs and generate resource allocation schemes based on historical satellite mission data, satellite resource status, and user requirements. The cloud-native management module 30 synchronizes the satellite constellation and hybrid cloud status in real time, and is responsible for the overall management of the entire constellation's resources. The hybrid cloud interface module 40 is configured to achieve seamless integration between satellite resources and multiple cloud platforms, while ensuring efficient resource utilization, security, and scalability.

[0113] The advantages of the resource virtualization module 10 in this embodiment are its flexibility and integration. By using containerization technology, satellite hardware resources are abstracted into dynamically allocable virtual resources, avoiding resource waste. At the same time, combined with the hybrid cloud interface, it breaks down the boundaries between satellite and cloud platform resources, forming a unified resource pool, improving resource allocation capabilities and adaptability, providing a foundation for system optimization, reducing management complexity, and enhancing system reliability and stability.

[0114] In one application example, a constellation of low-Earth orbit (LEO) satellites provides broadband communication services to users worldwide. These satellites are equipped with various payloads, including communication antennas, processing units, and storage devices. Simultaneously, the system accesses public and private cloud resources to process satellite data and perform computationally intensive tasks. Users are distributed across different geographical areas, resulting in significant spatiotemporal differences in communication needs. Therefore, dynamic resource allocation is necessary to meet user demands and improve resource utilization.

[0115] Optionally, this embodiment is developed based on the resource management framework Kubernetes. Kubernetes is an open-source container orchestration platform used for automating the deployment, scaling, and management of containerized applications. kubectl is the official command-line tool for Kubernetes, providing a simple yet powerful interface for communicating with, managing, and monitoring Kubernetes clusters.

[0116] This embodiment preferably uses Kubernetes containerization technology to encapsulate the satellite payload's hardware resources (such as sensors, communication equipment, and processing units on the satellite) and cloud platform resources (such as computing nodes, storage capacity, and network bandwidth in public or private clouds). First, a container template is defined to abstract the satellite payload's hardware resources into dynamically allocable virtual resource units. For example, the satellite's communication antenna bandwidth is divided into multiple virtual channels, each with a fixed bandwidth capacity; the processing unit's computing power is divided into multiple virtual computing nodes, each with a specific processor and memory configuration; and storage devices are divided into multiple virtual storage volumes, each with a specific storage capacity. Then, the Kubernetes container orchestration tool is used to manage the encapsulated resources, forming a unified resource pool. Through Kubernetes' API interface, resource creation, deletion, updating, and query operations are implemented, ensuring flexible allocation and efficient utilization of resources.

[0117] like Figure 3 As shown, in an optional implementation, the resource prediction and scheduling module 20 specifically includes a spatiotemporal prediction submodule 201 and a scheduling submodule 202, wherein:

[0118] The spatiotemporal prediction submodule 201 is configured to predict user needs based on historical mission data and / or historical user needs, using optimization algorithms or deep learning models; the scheduling submodule 202 is configured to dynamically adjust the resource allocation scheme based on the prediction results of the spatiotemporal prediction submodule, user needs, satellite resource status, and other information, so as to maximize resource utilization and improve the efficiency of mission execution.

[0119] Therefore, the resource prediction and scheduling module 20 uses optimization algorithms or machine learning models to accurately predict user needs and generate resource allocation plans based on satellite mission historical data, resource status and user needs, thereby improving the scientific nature and foresight of resource allocation.

[0120] In one example, the spatiotemporal prediction submodule 201 uses a deep learning module to predict the resource requirements of each beam coverage area over a future period. Specifically, the spatiotemporal prediction submodule 201 is used for:

[0121] Construct a generator G, the input of which is a random noise sequence Z = {z1, z2, z1, ..., z2}. T The output is time series data. Where, θ G z is a parameter of the generator G. t It is random noise at time point t, where T is the length of the sequence;

[0122] Construct a discriminator D, the input of which is real time series data X = {x1, x2, x1, ..., x2}. T} or the time series data The output is the probability of the input data being true; where x t These are the observations at time point t, where T is the length of the sequence;

[0123] By alternately optimizing the discriminant loss function L D and generator loss function L G Train the model, where:

[0124]

[0125] P data For the distribution of real data, P Z The distribution of random noise;

[0126] After the model training is completed, the generator G predicts the resource demand of each beam coverage area in the future period based on the historical data of satellite missions and real-time requirements, and generates a resource matching scheme.

[0127] Before model training, historical data of satellite missions are collected, including information such as mission type, mission time, mission location, and resource usage. The data is then preprocessed, such as data cleaning and normalization.

[0128] In practice, X can be the resource demand data of each beam coverage area over a period of time.

[0129] A random noise sequence Z is a randomly generated numerical sequence used as input to a generator G, enabling it to generate diverse data.

[0130] Optionally, the generator G uses an LSTM (Long Short-Term Memory) network to generate time series data. It takes random noise Z as input and outputs a generated time series. In mathematics, a generator G can be represented as The goal of generator G is to generate resource demand forecast data for each beam coverage area over a future period based on input random noise and historical data.

[0131] LSTM (Long Short-Term Memory) is a special type of recurrent neural network architecture primarily used for processing sequential data. By introducing mechanisms such as input gates, forget gates, and output gates, it effectively solves the gradient vanishing or exploding problems that traditional recurrent neural networks encounter when processing long sequences. This allows it to better capture long-term dependencies in sequences, resulting in outstanding performance in fields such as natural language processing and time series prediction. For example, it is used for tasks such as text generation, machine translation, and stock price prediction, providing a powerful tool for sequence modeling.

[0132] Discriminator D is used to distinguish between real data X and generated data. It takes time series data as input and outputs a probability value representing the probability that the input data is real data. The discriminator D's task is to distinguish whether the input data is real data or synthetic data generated by the generator G.

[0133] Mathematically, the discriminator D can be represented as: p = D(X; θ) D )or Where, θ D These are the parameters of the discriminator D.

[0134] Discriminator loss function L D The first term in the formula This represents the recognition loss of the discriminator D on the real data, and we hope that the output probability D(X) of the discriminator D on the real data is close to 1; the second term This represents the loss of the discriminator D in recognizing the generated data pairs. We hope that the output probability D(G(Z)) of the discriminator D for the generated data is close to 0.

[0135] The loss function L of generator G G To minimize the ability of the discriminator D to recognize the generated data, that is, to make the generated data as real data as possible:

[0136] Alternately train the discriminator D and the generator G: fix θ G θ is updated using gradient descent. D To minimize L D ; Fix θ D θ is updated using gradient descent. G To minimize L G This alternating training method allows the generator and discriminator to compete against each other during training. The generator continuously generates more realistic data, while the discriminator D continuously improves its ability to distinguish.

[0137] After training, the generator G is used to predict future time series data. That is, given a random noise sequence Z, the generator G will output a predicted time series. Therefore, it is possible to effectively predict and generate time-series data on user demand, thereby achieving forward-looking and scientific resource allocation.

[0138] In one example, the scheduling submodule 202 uses a reinforcement learning algorithm to dynamically adjust the beam radius, channel allocation, and time slot scheduling to match the real-time load of cloud resources. Specifically, the scheduling submodule 202 is used for:

[0139] Initialize the policy network parameters θ and the value network parameters φ;

[0140] Predefined reinforcement learning elements, including state s t Action a t and the reward function r t , where state s t Includes satellite beam coverage information, channel allocation status, time slot allocation status, and user service requirements; Action a t This represents a dynamic adjustment command for beam radius, channel allocation, or time slot scheduling; the reward function r t Calculated based on a comprehensive assessment of task completion rate, resource utilization rate, and data transmission quality;

[0141] Under the current strategy, interaction with the environment yields a series of states, actions, and rewards {s}. t a t r t}

[0142] In the specific implementation, state s t This can be represented as a vector containing satellite beam coverage information, frequency allocation, time slot allocation, and user service requirements, such as action a. t This represents adjustments to beam resources, frequency resources, and time slot resources. The reward function r... t The design can be based on the system's performance metrics, such as throughput, user satisfaction, and interference level.

[0143] Specifically, a reward is a scalar value used to evaluate the merits of an action in a given state, thereby influencing future decisions. In the satellite resource scheduling scenario of this embodiment, the reward can be a comprehensive evaluation of various factors, such as task completion, resource utilization, data transmission quality, task priority satisfaction, energy consumption, and system stability. Specifically, if a satellite can efficiently complete its predetermined observation or data transmission tasks, while making reasonable use of resources, ensuring high data transmission quality, meeting the needs of high-priority tasks, reducing energy consumption, and maintaining stable system operation, then these factors combined can form a positive reward signal, incentivizing the agent to make better scheduling decisions in the future.

[0144] Policy function π θ (at |s t ) indicates that in state s t Take action a t The probability is output by the policy network in the algorithm.

[0145] Optionally, the policy network is implemented using an Actor-Critic network. The Actor-Critic network is a widely used reinforcement learning architecture that combines the advantages of policy gradients (Actors) and value function estimation (Critics). Actors are responsible for selecting actions based on the current policy, obtaining rewards through interaction with the environment, and updating the policy to improve the expected return of the actions. Critics, on the other hand, evaluate the value of the actions chosen by the Actors, providing feedback on the quality of the policy and updating the value function to more accurately estimate future returns. Working together, Actors utilize the Critic's evaluation to optimize the policy, while the Critic leverages the Actor's experience to improve value estimation, thus achieving a more efficient learning and decision-making process.

[0146] Based on the sampled data, calculate the advantage function A(s) for each state-action pair. t a t The advantage function measures the performance in state s. t Take action a t Advantages and disadvantages relative to the average case (value function).

[0147] The state-action value function can be estimated using a deep Q-network, Q(s) t a t ;θ)=f(s t a t ;θ), where θ are the parameters of the neural network. f(s) t a t ;θ) is the output of the neural network, representing the output in state s t Take action a t The expected cumulative reward.

[0148] The objective function is constructed as follows:

[0149]

[0150] in, For the dominant function, It is a state-action value function. Value network estimation in state s t Expected return under; π θ (a t |s t ) indicates that in state s t Take action a t The probability, Let L be the probability of the old strategy, ∈ be the parameter of the clipping range, and L be the probability of the old strategy. VF (φ) is the loss function of the value function, L ENT (θ) is the entropy regularization term, and c1 and c2 are weight coefficients. It is the expected value;

[0151] The objective function described above combines policy optimization, value function update, and entropy regularization.

[0152] Loss function of value function Entropy regularization term L ENT (θ)=H(π θ (s t )). Where H(π θ (s t ))=-∑ a π θ (a|s t logπ θ (a|s t ) represents the entropy of the policy, and represents the uncertainty of the policy distribution.

[0153] The objective function works by limiting the range of policy updates. If the policy update rate exceeds a preset range (i.e., greater than 1+∈ or less than 1-∈), the update will be pruned to prevent drastic policy changes. This is combined with mean squared error loss. To optimize the value function; by adding an entropy regularization term L. ENT (θ) encourages strategy exploration and prevents premature convergence to a local optimum.

[0154] Increasing entropy can prevent the strategy from converging to a local optimum too early.

[0155] The policy network parameters θ are updated by maximizing the objective function L(θ);

[0156] By minimizing the loss function L of the value function VF (φ) Update the value network parameter φ.

[0157] Furthermore, repeated sampling, calculation of the advantage function, policy update, and value function update are performed until the policy converges. Corresponding actions are taken in the current state of the satellite communication system to achieve reasonable scheduling of the satellite communication system's resources.

[0158] Through multiple iterative training iterations, the policy network converges, yielding the optimal resource scheduling strategy. In actual operation, resource allocation is dynamically adjusted based on real-time task requirements and resource status, improving resource utilization and task execution efficiency.

[0159] See Figure 4Optionally, the cloud-native management module 30 specifically includes a data acquisition submodule 301, a communication monitoring submodule 302, a multi-source data fusion submodule 303, a simulation submodule 304, a real-time feedback submodule 305, and a strategy pre-playback submodule 306, wherein:

[0160] The data acquisition submodule 301 is configured to connect sensors and devices in the satellite constellation and hybrid cloud via IoT technology, collect real-time data on satellite position, status, mission execution, and cloud resource usage, calculate the constellation topology, obtain global satellite status and global resource usage, and send them to the multi-source data fusion submodule 303; the communication monitoring submodule 302 is configured to collect bandwidth and latency data for inter-constellation topology communication, generate a latency matrix between satellites, and send it to the multi-source data fusion submodule 303; the multi-source data fusion submodule 303 is configured to collect data from the data acquisition submodule 301. Based on the information received from the communication monitoring submodule 302, and according to the global resource usage, global satellite status, and delay matrix, a dynamic panoramic resource view of the satellite constellation and hybrid cloud is generated. The simulation submodule 304 is configured to simulate the operation of the satellite constellation and hybrid cloud in different scenarios using simulation technology. The real-time feedback submodule 305 is configured to compare the simulation results of the simulation submodule 304 with the actual operation data and adjust the satellite mission and cloud resource configuration in real time. The strategy pre-drill submodule 306 is configured to allow users to test and verify different scheduling strategies in a virtual environment to identify potential problems and optimize strategies in advance.

[0161] In an optional implementation, the data acquisition submodule 301 is used to connect sensors and devices in the satellite constellation with resources in the hybrid cloud via Internet of Things (IoT) technology. For example, it uses a satellite communication link to transmit satellite telemetry data to a ground station and then uploads it to the cloud platform. It collects data such as satellite position information, mission execution status, and resource usage in real time, and also collects resource usage data in public and private clouds, such as CPU and memory utilization of computing nodes. The collected data is then cleaned, filtered, and formatted to remove invalid and noisy data, thereby improving the accuracy and consistency of the data.

[0162] In an optional implementation, the communication monitoring submodule 302 is used to monitor the communication bandwidth and latency within the satellite constellation in real time through the communication links between satellites, for example, by periodically sending test signals and measuring round-trip time using a response mechanism; monitor the communication links between satellites and ground stations and the network communication between ground stations and cloud platforms, measure communication bandwidth and latency using network monitoring tools, and generate communication performance data; generate a latency matrix based on the monitored communication latency data. This matrix is ​​a two-dimensional array representing the communication latency between nodes. For example, for a constellation of n satellites, the latency matrix size is n×n; since the communication latency changes in real time, the latency matrix needs to be updated periodically, for example, every 5 minutes.

[0163] In an optional implementation, the multi-source data fusion submodule 303 is used to fuse data from different sources, eliminate redundancy and contradictions, and extract valuable information. For example, it combines satellite position information with communication delay information to analyze communication performance, and combines cloud resource usage with satellite mission execution to assess resource demand and utilization. Based on the fused data, a resource view is generated, including information such as the distribution, usage, and communication performance of satellite and cloud resources. For example, the satellite constellation topology and cloud platform resource pool are displayed through a graphical interface. Since the resource status changes in real time, the resource view needs to be updated in real time, for example, once every minute.

[0164] In an optional implementation, the simulation submodule 304 is used to construct a virtual simulation environment based on the satellite constellation topology, satellite performance parameters, cloud platform resource configuration, and mission requirements, simulating processes such as satellite motion, mission execution, resource allocation, and communication interaction. The strategy pre-simulation submodule 306 is used to define different simulation scenarios, such as normal operation scenarios, emergency mission scenarios, and resource failure scenarios, to observe the system's dynamic adjustment capabilities under different scenarios; to run different scheduling strategies in the simulation environment, recording data such as resource usage, mission execution efficiency, and communication performance; to analyze the simulation results, evaluate the performance of the scheduling strategies, identify potential problems, and optimize the strategies, such as adjusting strategy parameters or introducing new mechanisms; and to allow users to pre-simulate different scheduling strategies in the virtual environment, identifying and optimizing problems in advance.

[0165] In an optional implementation, the real-time feedback submodule 305 is used to compare the simulation results with the actual operating data, analyze the differences, such as comparing resource utilization and task response time, and evaluate the accuracy of the simulation results and the rationality of the scheduling strategy; generate feedback information based on the comparison results and feed it back to the scheduling submodule 202, such as prompting to adjust the resource allocation strategy or optimize communication performance; the scheduling submodule 202 adjusts the satellite mission and cloud resource configuration in real time based on the feedback information, such as adjusting the mission priority or communication link.

[0166] The hybrid cloud interface module 40 is responsible for interfacing with various mainstream cloud platforms (including public and private clouds) to achieve resource integration and management. See also... Figure 5 Optionally, the hybrid cloud interface module 40 specifically includes a cloud platform interface submodule 401, an elastic expansion submodule 402, and a security isolation submodule 403, wherein:

[0167] The cloud platform interface submodule 401 is configured to interface with various mainstream cloud platforms to achieve resource integration and management; the elastic scaling submodule 402 is configured to dynamically adjust the scale of cloud computing resources according to resource needs to meet the real-time requirements of the task; and the security isolation submodule 403 is configured to ensure the security of resources and data between the satellite system and the cloud platform to prevent data leakage and malicious attacks.

[0168] In an optional implementation, the cloud platform interface submodule 401 is configured as follows: Select a supported cloud platform, such as public clouds like AWS, Azure, and Google Cloud, and private clouds like OpenStack and VMware vSphere, ensuring good interface support and security; configure the interface to manage cloud resources programmatically; register satellite resources and cloud platform resources to form a unified resource pool, and periodically synchronize resource status information to ensure the system can monitor resource usage in real time. For example, obtain information such as cloud platform resource utilization and virtual machine running status through the interface, and integrate it with the satellite resource status information.

[0169] In an optional implementation, the elastic expansion submodule 402 is specifically used to combine the prediction results of the resource prediction and scheduling module 20 to analyze the gap in resource demand. When the predicted resource demand exceeds the existing capacity, the elastic expansion mechanism is automatically triggered to dynamically increase resources such as computing nodes, storage capacity or network bandwidth, and allocate the new resources to the corresponding tasks to optimize the resource allocation scheme. At the same time, the resource allocation is dynamically adjusted according to the task execution status and resource utilization rate to reclaim excess resources in a timely manner.

[0170] In an optional implementation, the security isolation submodule 403 is specifically used to formulate strict security policies, covering access control, data encryption, network security, etc. For example, setting role-based access control policies to restrict users' access to resources, and encrypting the storage and transmission of sensitive data; achieving network isolation between the satellite system and the cloud platform through virtual private networks (VPNs) and firewall technology, configuring firewall rules to restrict unauthorized access and data leakage; encrypting data in the satellite and cloud platform to ensure data security during transmission and storage, and restricting access to the data through access control mechanisms; conducting regular security audits to check the implementation of security policies, and monitoring the system security status in real time through security monitoring tools to promptly detect and handle security threats.

[0171] The present invention also provides a computer storage medium storing a computer program thereon, wherein the computer program, when executed by a processor, implements the dynamic allocation method for satellite resources under the cloud-native hybrid cloud architecture as described in the above embodiments.

[0172] The advantages of the computer storage medium over existing technologies and the satellite resource dynamic allocation system are the same, and will not be elaborated further here.

[0173] In summary, the satellite resource dynamic allocation method and system under a cloud-native hybrid cloud architecture described in this invention achieves three-dimensional synergy of dynamic resource pooling, intelligent prediction, and elastic scheduling by adopting a cloud-native hybrid cloud architecture. Dynamic resource pooling integrates dispersed satellite resources into a unified resource pool, improving resource allocation flexibility and utilization. Intelligent prediction can forecast resource needs in advance based on historical data and real-time mission requirements, providing a basis for resource scheduling. Elastic scheduling dynamically adjusts resource allocation based on prediction results and real-time mission requirements, ensuring timely response and efficient execution of satellite missions. Through this innovative three-dimensional synergistic architecture, this invention provides a novel approach and solution for satellite resource management, possessing significant innovative value and practical application significance.

[0174] It should be noted that the present invention can be implemented in software and / or a combination of software and hardware, for example, using an application-specific integrated circuit (ASIC), a general-purpose computer, or any other similar hardware device. In one embodiment, the software program of the present invention can be executed by a processor to implement the steps or functions described above. Similarly, the software program of the present invention (including associated data structures) can be stored in a computer-readable recording medium, such as RAM memory, a magnetic or optical drive, a floppy disk, or similar devices. Furthermore, some steps or functions of the present invention can be implemented in hardware, for example, as circuitry that works with a processor to perform the various steps or functions.

[0175] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of the present invention is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0176] Of course, the present invention may have other various embodiments. Without departing from the spirit and essence of the present invention, those skilled in the art can make various corresponding changes and modifications according to the present invention, but these corresponding changes and modifications should all fall within the protection scope of the appended claims.

Claims

1. A method for dynamic allocation of satellite resources under a cloud-native hybrid cloud architecture, characterized in that, Including the following steps: Satellite payload resources and cloud resources are encapsulated using containerization technology to form a unified virtual resource pool; wherein, the cloud resources include computing nodes, storage capacity and network bandwidth of public cloud and private cloud; A deep learning model is constructed, and the resource demand of each beam coverage area in the future period is predicted based on historical satellite mission data and real-time demand, and a resource matching scheme is generated. Reinforcement learning algorithms are used to dynamically adjust beam radius, channel allocation, and time slot scheduling to match the real-time load changes of the cloud resources; The method of dynamically adjusting beam radius, channel allocation, and time slot scheduling using reinforcement learning algorithms includes: Initialize policy network parameters and value network parameters ; Predefined reinforcement learning elements, including states ,action and reward function , where state Includes satellite beam coverage information, channel allocation status, time slot allocation status, and user service requirements; actions This represents a dynamic adjustment command for beam radius, channel allocation, or time slot scheduling; reward function. Calculated based on a comprehensive assessment of task completion rate, resource utilization rate, and data transmission quality; The objective function is constructed as follows: ； in, For the dominant function, It is a state-action value function. Value network estimation in state The expected return; Indicates the state Take action below The probability, The probability of the old strategy. The parameters for the editing range. It is the loss function of the value function. It is the entropy regularization term. and These are weighting coefficients. It is the expected value; By maximizing the objective function Update the policy network parameters ; By minimizing the loss function of the value function Update the value network parameters ; A satellite-cloud digital twin model is constructed. Based on the resource matching scheme, the satellite-cloud digital twin model is used to simulate and predict scheduling strategies and mirror resource status in real time. The strategy library is updated according to the simulation feedback data of the satellite-cloud digital twin model to achieve automatic reorganization of multi-satellite missions.

2. The method for dynamic allocation of satellite resources under a cloud-native hybrid cloud architecture according to claim 1, characterized in that, The construction of the deep learning model includes: Construct a generator G, the input of which is a random noise sequence. The output is time series data. ;in, For the generator G The parameters, It is random noise at time point t. T The length of the sequence; Construct a discriminator D, whose input is real time series data. or the time series data The output is the probability of the input data being true; where, These are the observations at time point t. T The length of the sequence; By alternately optimizing the discriminator loss function and generator loss function Train the model, where: ； ； For the distribution of real data, The distribution of random noise; After the model training is completed, the generator G predicts the resource demand of each beam coverage area in the future period based on the historical data of satellite missions and real-time requirements, and generates a resource matching scheme.

3. A dynamic satellite resource allocation system under a cloud-native hybrid cloud architecture, characterized in that, It includes a resource virtualization module, a resource prediction and scheduling module, a cloud-native management module, and a hybrid cloud interface module; among which: The resource virtualization module is configured to encapsulate satellite payload resources and cloud resources using containerization technology to form a unified virtual resource pool; wherein, the cloud resources include computing nodes, storage capacity and network bandwidth of public cloud and private cloud; The resource prediction and scheduling module is configured to predict the resource demand of each beam coverage area in the future period based on historical satellite mission data and real-time demand using a deep learning model, and generate a resource matching scheme; and to dynamically adjust the beam radius, channel allocation and time slot scheduling using a reinforcement learning algorithm to match the real-time load changes of the cloud resources. The method of dynamically adjusting beam radius, channel allocation, and time slot scheduling using reinforcement learning algorithms includes: Initialize policy network parameters and value network parameters ; Predefined reinforcement learning elements, including states ,action and reward function , where state Includes satellite beam coverage information, channel allocation status, time slot allocation status, and user service requirements; actions This represents a dynamic adjustment command for beam radius, channel allocation, or time slot scheduling; reward function. Calculated based on a comprehensive assessment of task completion rate, resource utilization rate, and data transmission quality; The objective function is constructed as follows: ； in, For the dominant function, It is a state-action value function. Value network estimation in state The expected return; Indicates the state Take action below The probability, The probability of the old strategy. The parameters for the editing range. It is the loss function of the value function. It is the entropy regularization term. and These are weighting coefficients. It is the expected value; By maximizing the objective function Update the policy network parameters ; By minimizing the loss function of the value function Update the value network parameters ; The cloud-native management module is configured to simulate and predict scheduling strategies based on the resource matching scheme through a satellite and cloud digital twin model and to mirror resource status in real time, and to update the strategy library according to the simulation feedback data of the satellite and cloud digital twin model, so as to realize automatic reorganization of multi-satellite missions. The hybrid cloud interface module is configured to enable seamless integration of satellite resources with multiple cloud platforms.

4. The satellite resource dynamic allocation system under a cloud-native hybrid cloud architecture according to claim 3, characterized in that, The resource prediction and scheduling module specifically includes: The spatiotemporal prediction submodule is configured to predict user needs based on historical task data and / or historical user needs, using optimization algorithms or deep learning models. The scheduling submodule is configured to dynamically adjust the resource allocation scheme based on the prediction results of the spatiotemporal prediction submodule, user needs, and satellite resource status.

5. The satellite resource dynamic allocation system under a cloud-native hybrid cloud architecture according to claim 4, characterized in that, The spatiotemporal prediction submodule is specifically used for: Construct a generator G, the input of which is a random noise sequence. The output is time series data. ;in, For the generator G The parameters, It is random noise at time point t. T The length of the sequence; Construct a discriminator D, whose input is real time series data. or the time series data The output is the probability of the input data being true; where, These are the observations at time point t. T The length of the sequence; By alternately optimizing the discriminator loss function and generator loss function Train the model, where: ； ； For the distribution of real data, The distribution of random noise; After the model training is completed, the generator G predicts the resource demand of each beam coverage area in the future period based on the historical data of satellite missions and real-time requirements, and generates a resource matching scheme.

6. The satellite resource dynamic allocation system under a cloud-native hybrid cloud architecture according to claim 4, characterized in that, The scheduling submodule is specifically used for: Initialize policy network parameters and value network parameters ; Predefined reinforcement learning elements, including states ,action and reward function , where state Includes satellite beam coverage information, channel allocation status, time slot allocation status, and user service requirements; actions This represents a dynamic adjustment command for beam radius, channel allocation, or time slot scheduling; reward function. Calculated based on a comprehensive assessment of task completion rate, resource utilization rate, and data transmission quality; The objective function is constructed as follows: ； in, For the dominant function, It is a state-action value function. Value network estimation in state The expected return; Indicates the state Take action below The probability, The probability of the old strategy. The parameters for the editing range. It is the loss function of the value function. It is the entropy regularization term. and These are weighting coefficients. It is the expected value; By maximizing the objective function Update the policy network parameters ; By minimizing the loss function of the value function Update the value network parameters .

7. The satellite resource dynamic allocation system under a cloud-native hybrid cloud architecture according to claim 3, characterized in that, The cloud-native management module specifically includes: The data acquisition submodule is configured to connect to sensors and devices in the satellite constellation and hybrid cloud via IoT technology, collect data on the satellite's position, status, mission execution, and cloud resource usage in real time, calculate the constellation topology, obtain the global satellite status and global resource usage, and send them to the multi-source data fusion submodule. The communication monitoring submodule is configured to collect the bandwidth and delay of inter-satellite access throughout the entire constellation topology, generate a delay matrix between satellites, and send it to the multi-source data fusion submodule. The multi-source data fusion submodule is configured to collect information from the data acquisition submodule and the communication monitoring submodule, and generate a dynamic panoramic resource view of the satellite constellation and hybrid cloud based on the global resource usage, the global satellite status, and the delay matrix. The simulation submodule is configured to simulate the operation of satellite constellations and hybrid clouds in different scenarios using simulation technology. The real-time feedback submodule is configured to compare the simulation results of the simulation submodule with the actual operating data and adjust the satellite mission and cloud resource configuration in real time. The strategy pre-play submodule is configured to allow users to test and verify different scheduling strategies in a virtual environment.

8. The satellite resource dynamic allocation system under a cloud-native hybrid cloud architecture according to claim 3, characterized in that, The hybrid cloud interface module specifically includes: The cloud platform interface submodule is configured to interface with various mainstream cloud platforms to achieve resource integration and management. The elastic scaling submodule is configured to dynamically adjust the scale of cloud computing resources according to resource demand in order to meet the real-time requirements of tasks. The security isolation submodule is configured to ensure the security of resources and data between the satellite system and the cloud platform, preventing data leakage and malicious attacks.

9. A computer storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the dynamic satellite resource allocation method under the cloud-native hybrid cloud architecture as described in any one of claims 1 to 2.

Images