Big model-based space computing power network micro-service deployment optimization method and system

Abstract

The application discloses a space computing power network micro-service deployment optimization method and system based on a large model, and has the characteristics that the method comprises the following steps: constructing a processing, transmission and propagation delay model, and generating a structured deployment task description; performing variable coverage, constraint integrity, delay correspondence and resource dimension consistency checking; distinguishing execution errors and modeling errors and feeding back after correction, and outputting a micro-service deployment matrix to a satellite node. The system comprises a data acquisition module, a graph construction and delay modeling module, a task description module, a model generation module, a consistency checking module, a code generation and verification module, a feedback correction module and a deployment output module. Compared with the prior art, the application has the advantages that the checkability of model generation and deployment decision in a dynamic space network is improved, and the problems of high code writing cost of a solver, easy loss of model components, easy disconnection of delay modeling and link data, and insufficient reliability of deployment results of a space computing power network in a dynamic low-orbit satellite network are solved.

Description

Technical Field

[0001] This invention relates to the field of space computing network resource scheduling technology, and in particular to a large-model-based method and system for optimizing the deployment of microservices in space computing networks, including structured modeling, solver code generation, verification and error correction, and deployment decision output for remote sensing application microservices in low-Earth orbit satellite networks. Background Technology

[0002] Space-based computing networks provide computing power and coverage for on-orbit processing of remote sensing data by introducing onboard computing resources into satellite networks. Low-Earth orbit (LEO) satellite nodes, compared to terrestrial data centers, have limited computing, storage, and communication resources. Furthermore, the high-speed motion of LEO satellites causes the distance, propagation delay, and transmission rate of inter-satellite links to vary over time, necessitating the optimization of remote sensing microservice deployment to simultaneously handle heterogeneous resources, dynamic links, and service dependencies. Remote sensing applications typically include functions such as image acquisition, preprocessing, feature extraction, target recognition, image segmentation, and change detection. By breaking these applications down into microservices, different applications can share functionally identical microservices, and different microservices can be deployed on different LEO satellite nodes, thereby improving onboard resource utilization and system scalability. Microservice decomposition also transforms the deployment problem from placing a single application to optimizing the combination of multiple dependent nodes.

[0003] Existing heuristic deployment methods rely on fixed rules, making it difficult to stably adapt to different task inputs when inter-satellite links change dynamically. Reinforcement learning deployment methods typically require pre-constructing states, actions, and rewards and then training them, resulting in high adjustment costs when facing changes in satellite resource status or service quality goals. Manually building optimization models and writing solver code relies on domain experience, making it difficult to efficiently complete modeling, coding, and verification when space computing network tasks change rapidly. While large language models have the ability to extract modeling elements from task descriptions and generate code, directly using large language models to output deployment suggestions can easily lead to problems such as omitted variables, missing constraints, mismatched latency terms, mismatched resource dimensions, code execution failures, or model results that do not meet deployment constraints.

[0004] In summary, existing space computing power networks suffer from several drawbacks in dynamic low-Earth orbit satellite networks, including reliance on manual modeling, high solver code writing costs, susceptibility to missing model components, potential disconnection between latency modeling and link data, and insufficient reliability of deployment results. Therefore, there is an urgent need for a structured modeling, dedicated consistency verification, testing, and feedback correction method for deploying remote sensing microservices in space computing power networks. This method would ensure that the output of large language models is constrained to produce verifiable, solvable, and deployable results. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for optimizing the deployment of microservices in a space computing network based on a large model, addressing the shortcomings of existing technologies. It employs a structured deployment task description method, constructing a processing, transmission, and propagation latency model by incorporating a remote sensing microservice DAG, low-Earth orbit satellite resources, deployable satellite sets, and the time-varying states of inter-satellite links. The method constrains the output parameters, variables, constraints, objective functions, model expressions, and solver code snippets of the large language model through preset prompt templates. It then performs checks on variable coverage, constraint integrity, latency correspondence, and resource dimension consistency. After testing cases distinguish between execution errors and modeling errors and provide feedback for correction, it outputs a deployment matrix of microservices to satellite nodes. This method limits the role of the large language model to a modeling component generator constrained by structured input and prompt templates. Through variable coverage, constraint integrity, latency term correspondence, and resource dimension consistency verification, it transforms the output of the large language model into an inspectable optimized model component. By using pre-set test cases and dual-channel feedback for execution errors / modeling errors, it can separate and handle code-level errors and model-level errors, improving the feasibility and reliability of the deployment and optimization model generation process. It is particularly suitable for structured modeling, solver code generation, verification and error correction, and deployment decision output for remote sensing application microservices in low-Earth orbit satellite networks, greatly improving the verifiability of model generation and deployment decisions in dynamic space networks.

[0006] The specific technical solution for achieving the objective of this invention is: a microservice deployment optimization method for space computing power networks based on a large model, characterized by the following steps:

[0007] Step S101: Obtain the description of the remote sensing application microservice, the data dependencies between microservices, the resource requirements of each microservice, the resource status of low-orbit satellite nodes, the set of deployable satellites corresponding to each microservice, and the time-varying status of inter-satellite links within the target deployment time window;

[0008] Step S102: Construct a directed acyclic graph of microservices based on the data dependencies between microservices, and construct a satellite network graph based on the resource status of low-orbit satellite nodes and the time-varying status of inter-satellite links;

[0009] Step S103: Construct latency calculation relationships based on the microservice directed acyclic graph and satellite network graph;

[0010] Step S104: Generate a structured deployment task description based on the microservice directed acyclic graph, satellite network graph, resource requirements, deployable satellite set, and latency calculation relationship;

[0011] Step S105: Input the structured deployment task description into the large language model that runs according to the preset prompt template;

[0012] Step S106: Based on the input structured deployment task description, the large language model outputs the parameter set P, variable set V, constraint set C, objective function O, model expression, and solver code snippet;

[0013] Step S107: Perform deployment variable coverage verification, constraint integrity verification, latency item correspondence verification, and resource dimension verification on the parameter set P, variable set V, constraint set C, and objective function O;

[0014] Step S108: Generate a microservice deployment optimization model from the verified parameter set P, variable set V, constraint set C, objective function O, and model expression, and generate solver executable code based on solver code snippets;

[0015] Step S109: Run preset test cases to verify the solver executable code. The preset test cases in step S109 include: feasible deployment test cases, resource overrun test cases, microservice dependency test cases, and binary deployment test cases.

[0016] Step S110: Divide the errors generated during the verification process into execution errors and modeling errors, and feed back the error messages and code snippets corresponding to the execution errors, as well as the model expressions, test inputs and verification results corresponding to the modeling errors, to the large language model for correction;

[0017] Step S111: After the solver executable code passes the preset test cases, run the solver executable code and output a deployment matrix representing the correspondence between microservices and low-Earth orbit satellite nodes, thereby optimizing the deployment of microservices in the space computing network.

[0018] The delay calculation relationship in step S103 includes: processing delay. Transmission delay Propagation delay And the latest time the data of multiple predecessor microservices arrives at the startup time. .in, For microservices The computational load, For satellite nodes computing power For microservices Microservices The amount of data, For satellite nodes To satellite node The link transmission rate at time t, For satellite nodes To satellite node At a distance of time t, For signal propagation speed, For precursor microservices The completion time, For microservices The set of precursor microservices, To support the precursor microservice Satellite nodes.

[0019] The structured deployment task description in step S104 includes: scenario field, microservice field, dependency field, satellite resource field, link field, latency field, constraint field, target field, and output field; the microservice field records the microservice identifier, computational load, input data volume, output data volume, and resource requirement vector; the dependency field records the predecessor and successor microservices in the directed acyclic graph of microservices; the satellite resource field records the computational, storage, and memory resource status of low-Earth orbit satellite nodes; the link field records time slices, link reachability, link transmission rate, and distance between satellite nodes; the output field limits the output to a deployment matrix.

[0020] The preset prompt template in step S105 includes: task role field, scenario input field, symbol extraction field, modeling requirement field, code generation field, verification requirement field, and output format field. The output format field limits the output of the large language model in the order of parameter set P, variable set V, constraint set C, objective function O, model expression, and solver code fragment. The variable set V includes a binary deployment variable used to indicate whether the microservice is deployed to the corresponding low-Earth orbit satellite node.

[0021] The deployment variable coverage verification in step S107 is used to confirm that each microservice and its corresponding set of deployable satellites are included in the range of binary deployment variable indexes; the constraint integrity verification is used to confirm that the unique deployment constraint, deployable set constraint, resource capacity constraint, microservice dependency constraint, latency constraint, and binary value constraint are all written into the constraint set C; the latency term correspondence verification is used to confirm that the latency terms in the objective function O and the dependency constraints correspond to processing latency, transmission latency, and propagation latency, respectively; the resource dimension verification is used to confirm that the dimension of the resource requirement vector is consistent with the dimension of the satellite node resource capacity vector.

[0022] The microservice deployment optimization model in step S108 is a mixed integer programming model; the mixed integer programming model uses a deployment matrix. Let T be the decision variable, and let minimizing the total completion time T be the objective function. , , , as well as As a constraint relationship. Among them, For microservices Deployable satellite collection For microservices The resource demand vector, For satellite nodes The resource capacity vector, For microservices The completion time, For precursor microservices The completion time.

[0023] The execution errors in step S110 include: syntax errors, undefined variables, mismatched index dimensions, and runtime errors; the modeling errors include: unbounded results, infeasible results, results that violate unique deployment constraints, results that violate resource capacity constraints, results that violate microservice dependency constraints, and results that fail to meet latency calculation verification.

[0024] The rows of the deployment matrix in step S111 correspond to the microservices in the directed acyclic graph of microservices; the columns of the deployment matrix correspond to the low-Earth orbit satellite nodes; the values ​​in the deployment matrix are used to indicate whether the corresponding microservice is deployed to the corresponding low-Earth orbit satellite node, and the management satellite node issues microservice image acquisition, runtime parameter configuration and microservice startup instructions to the target low-Earth orbit satellite node according to the deployment matrix.

[0025] A space computing network microservice deployment optimization system based on a large model is characterized by comprising: a data acquisition module, a graph construction and latency modeling module, a task description module, a model generation module, a consistency verification module, a code generation and verification module, a feedback correction module, and a deployment output module. The data acquisition module acquires descriptions of remote sensing application microservices, data dependencies between microservices, resource requirements of each microservice, resource status of low-Earth orbit satellite nodes, the set of deployable satellites corresponding to each microservice, and the time-varying status of inter-satellite links within the target deployment time window. The graph construction and latency modeling module constructs a directed acyclic graph of microservices and a satellite network graph, and determines processing latency, transmission latency, propagation latency, and the latest arrival time of data from multiple predecessor microservices. The task description module generates structured... The deployment task description is input into the large language model, which runs according to a preset prompt template. The output parameters include P, V, C, O, the objective function, the model expression, and the solver code snippet. The consistency verification module performs deployment variable coverage verification, constraint integrity verification, latency term correspondence verification, and resource dimension verification. The code generation and verification module generates a microservice deployment optimization model and solver executable code, and runs preset test cases. The feedback correction module feeds back the error messages and code snippets corresponding to execution errors, as well as the model expressions, test inputs, and verification results corresponding to modeling errors, to the large language model. The deployment output module outputs the deployment matrix after the solver executable code passes verification and sends microservice deployment instructions.

[0026] Compared with the prior art, the present invention has the following beneficial technical effects and significant technical progress:

[0027] 1) This invention limits the role of the large language model to a modeling component generator constrained by structured input and prompt templates, and transforms the output of the large language model into an inspectable optimized model component through variable coverage, constraint integrity, time delay term correspondence and resource dimension consistency verification.

[0028] 2) By using pre-set test cases and dual-channel feedback for execution errors / modeling errors, code-level errors and model-level errors can be handled separately, which greatly improves the feasibility and reliability of the deployment and optimization model generation process.

[0029] 3) It has good application prospects and commercial development value, and is especially suitable for structured modeling, solver code generation, verification and error correction and deployment decision output of remote sensing application microservices in low orbit satellite networks. Attached Figure Description

[0030] Figure 1 This is a flowchart of the method of the present invention;

[0031] Figure 2 This is a schematic diagram of the system structure of the present invention;

[0032] Figure 3 This is a schematic diagram of the deployment scenario of remote sensing microservices in a space computing network, as shown in Example 1.

[0033] Figure 4 This is a schematic diagram of the optimized model generation, code generation, and self-correction verification process in Example 1. Detailed Implementation

[0034] The present invention provides a method for optimizing the deployment of microservices in a space computing network based on a large model, comprising the following steps: 1) obtaining descriptions of remote sensing application microservices, microservice dependencies, resource requirements, satellite resources, deployable satellite sets, and time-varying states of inter-satellite links within the target deployment time window; 2) constructing a directed acyclic graph of microservices and a satellite network graph; 3) establishing relationships between processing latency, transmission latency, propagation latency, and the latest arrival of multiple predecessor data; generating a structured deployment task description; 4) calling a large language model running according to a preset prompt template to output P / V / C / O, model expressions, and solver code snippets; 5) performing consistency checks; 6) generating an optimized model and executable code; 7) running preset test cases; 8) differentiating between execution errors and modeling errors and providing feedback corrections; and 9) outputting a deployment matrix after successful verification, thereby optimizing the deployment of microservices in the space computing network.

[0035] The embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the following embodiments are used to illustrate the technical solutions of the present invention, and not to limit the scope of protection of the present invention.

[0036] Example 1

[0037] See Figure 1 A microservice deployment optimization method for space computing power networks based on a large model, the specific steps of which are as follows:

[0038] Step S101: Obtain deployment input data

[0039] The system acquires the descriptions of remote sensing application microservices, data dependencies between microservices, resource requirements of each microservice, resource status of low-Earth orbit satellite nodes, the set of deployable satellites corresponding to each microservice, and the time-varying status of inter-satellite links within the target deployment time window. The microservice descriptions include: microservice identifier, computational load, input data volume, output data volume, and resource requirement vector. The low-Earth orbit satellite node resource status includes: computing power, storage capacity, memory capacity, and communication status. The time-varying status of inter-satellite links includes: time slice, link reachability, link transmission rate, and distance between satellite nodes.

[0040] The microservice set M in this invention is represented by the following formula:

[0041] .

[0042] Where M is the set of microservices, For the first microservice, Let n be the nth microservice, where n is the total number of microservices.

[0043] The set of low-Earth orbit satellite nodes N is represented by the following formula:

[0044]

[0045] Where N is the set of low-Earth orbit satellite nodes, As the first low-Earth orbit satellite node, Let q be the q-th low-Earth orbit satellite node, where q is the total number of low-Earth orbit satellite nodes.

[0046] The deployment matrix X to be solved is expressed by the following formula:

[0047]

[0048] Where X is the deployment matrix to be solved. =1 indicates microservices Deployed to satellite nodes , =0 indicates microservices Not deployed to satellite nodes , where n×q means that the matrix has n microservice rows and q satellite node columns.

[0049] The target deployment time window is denoted as [t_0, t_e]. The start time of the deployment window for the target. The end of the deployment window for the target.

[0050] Step S102: Construct the microservice DAG and satellite network graph

[0051] This step separates the application dependency structure and spatial network topology, facilitating the subsequent mapping of microservice dependency constraints to link latency constraints. Specifically, it includes:

[0052] 1) Construct a directed acyclic graph of microservices as shown in the following formula based on the data dependencies between microservices. :

[0053] .

[0054] in, Let M be a directed acyclic graph of microservices. It is a set of data dependency edges between microservices. For precursor microservices, For subsequent microservices.

[0055] 2) Construct the satellite network diagram shown in the following formula based on the resource status of low-Earth orbit satellite nodes and the time-varying status of inter-satellite links. :

[0056] .

[0057] in, Let N be the satellite network diagram at time t, where N is the set of low-Earth orbit satellite nodes and L(t) is the set of inter-satellite links reachable at time t.

[0058] Step S103: Constructing the delay calculation relationship

[0059] The system considers processing latency, transmission latency, propagation latency, and the latest arrival time of data from multiple predecessor microservices. The processing latency is determined by the microservice's computational load and the satellite's computing power. The transmission latency is determined by the amount of data between microservices and the inter-satellite link transmission rate within the target time window. The propagation latency is determined by the distance between satellite nodes and the signal propagation speed. The latest arrival time of data is defined when a microservice has multiple predecessor microservices, with the latest arrival time among the data arrival times of these predecessor microservices serving as the startup constraint for that microservice. The latest arrival time of data is expressed by the following formula:

[0060] ;

[0061] ;

[0062] .

[0063] in, For microservices At satellite nodes Processing latency on For precursor microservices To subsequent microservices Data transmission latency, For satellite nodes To satellite node The propagation delay For microservices The data arrives at the startup time. For microservices The computational load, For satellite nodes computing power For microservices To microservices The amount of data, For satellite nodes at time t To satellite node The link transmission rate, For satellite nodes at time t To satellite node distance, For signal propagation speed, For precursor microservices The completion time, For microservices A collection of precursor microservices.

[0064] Step S104: Generate a structured deployment task description

[0065] The aforementioned microservice DAG, satellite network graph, resource requirements, deployable sets, and latency calculation relationships are organized into a structured deployment task description. This structured deployment task description adopts a field-based format, including scenario fields, microservice fields, dependency fields, satellite resource fields, link fields, latency fields, constraint fields, target fields, and output fields. Through field-based input, the large language model receives enumerable, parsable, and verifiable modeling materials, rather than loose natural language descriptions.

[0066] Step S105: Invoke the preset prompt template and large language model

[0067] The structured deployment task description is input into the large language model that runs according to the preset prompt template. The preset prompt template includes: task role field, scenario input field, symbol extraction field, modeling requirement field, code generation field, verification requirement field, and output format field. The prompt template constrains the large language model as an optimization modeling assistant, requiring it to extract sets, parameters, variables, constraints, and objective functions from the input fields and output solver code snippets.

[0068] In one implementation, the preset prompt template includes the following: 1) The task role field requires the large language model to establish a mixed integer programming model for the deployment of microservices in the space computing network; 2) The scenario input field requires the microservice DAG, satellite network diagram, resource requirements, deployable set, and latency relationship; 3) The symbol extraction field requires the generation of parameter set P and variable set V; 4) The modeling requirements field requires the generation of constraint set C and objective function O; 5) The code generation field requires the generation of solver code snippets; 6) The verification requirements field requires the explanation of variable coverage, constraint integrity, latency term correspondence, and resource dimension consistency; 7) The output format field requires the output to be in the order of P, V, C, O, model expression, and solver code snippets.

[0069] Step S106: Output P / V / C / O and model expression

[0070] The large language model outputs a parameter set P, a variable set V, a constraint set C, an objective function O, a model expression, and a solver code snippet based on the input structured deployment task description. The parameter set P includes: microservice computing load, data volume, resource requirements, satellite resource capacity, link transmission rate, satellite distance, and deployable set; the variable set V includes: binary deployment variables. Variables at start time Completion time variable The total application completion time variable T; the constraint set C includes: unique deployment, deployable set, resource capacity, service dependency, link reachability, latency calculation and binary value constraints; the objective function O is to minimize the total application completion time or the completion time index within the target deployment time window.

[0071] The parameter set P is described by the following formula:

[0072] .

[0073] The set of variables V is described by the following formula:

[0074] .

[0075] The constraint set C is described by the following formula:

[0076] .

[0077] Where P is the parameter set, V is the variable set, C is the constraint set, and O is the objective function; For microservices The computational load, For microservices To microservices The amount of data, For microservices The resource demand vector, For satellite nodes The resource capacity vector, For satellite nodes at time t To satellite node The link transmission rate, For satellite nodes at time t To satellite node distance, For microservices Deployable satellite collection For microservices At the beginning of For microservices The completion time is T, where T is the total application completion time.

[0078] Step S107: Perform consistency check

[0079] 1) Perform consistency checks on the output of the large language model;

[0080] 2) Deployment variable coverage verification: Traverse each microservice and its set of deployable satellites, and check whether a binary deployment variable index exists in each one;

[0081] 3) Constraint integrity check: Check whether the unique deployment constraint, deployable set constraint, resource capacity constraint, microservice dependency constraint, latency constraint, and binary value constraint have all been written into the constraint set;

[0082] 4) Check whether the delay terms in the objective function and dependency constraints correspond to the processing delay, transmission delay, and propagation delay, respectively;

[0083] 5) Check whether the resource demand vector and the satellite resource capacity vector are consistent in terms of resource dimension verification.

[0084] The deployment variable constraints are described by the following formula:

[0085]

[0086] in, As the only deployment constraint, For deployable set constraints, Due to resource capacity constraints, For microservice dependency constraints, Binary value constraints; For microservices Deployable satellite collection Deploy variables for binary values. For microservices The resource demand vector, For satellite nodes The resource capacity vector.

[0087] Step S108: Generate optimization model and solver code

[0088] A microservice deployment optimization model is generated based on the parameter set P, variable set V, constraint set C, objective function O, and model expression. Solver executable code is generated based on solver code snippets. The microservice deployment optimization model is a mixed integer programming model with deployment matrix X as the decision variable, minimizing the total application completion time T as the objective, and low-Earth orbit satellite node resource status, microservice deployable satellite set, microservice dependencies, inter-satellite link reachability, and binary value conditions as constraints.

[0089] The objective function and the temporal dependency constraints are described by the following equation:

[0090] ;

[0091] .

[0092] Where O represents the optimization objective, and T represents the total application completion time. For microservices The completion time, For microservices At the beginning of Deploy variables for binary values. To handle latency, For transmission delay, To delay the transmission time, This is a set of data dependency edges between microservices.

[0093] Step S109: Run the preset test cases

[0094] Run preset test cases to verify the solver's executable code. The preset test cases include: feasible deployment test cases, resource overrun test cases, microservice dependency test cases, and binary deployment test cases. The feasible deployment test cases are used to verify that the deployment matrix can meet resource capacity and dependency timing constraints. The resource overrun test cases are used to verify that when resource requirements exceed the satellite node capacity, it can return an infeasible result or trigger capacity constraints. The microservice dependency test cases are used to verify that the start time of subsequent microservices is not earlier than the data arrival time of the predecessor microservices. The binary deployment test cases are used to verify that the deployment variable takes the value of 0 or 1 and that each microservice corresponds to only one selected deployment node.

[0095] Step S110: Feedback and correction based on error type

[0096] Errors generated during the verification process are categorized into execution errors and modeling errors. The error messages and code snippets corresponding to execution errors, as well as the model expressions, test inputs, and verification results corresponding to modeling errors, are fed back to the large language model for correction.

[0097] The execution errors include: syntax errors, undefined variable errors, index dimension mismatch errors, and runtime errors. Their feedback includes: error messages, line numbers, relevant code snippets, and runtime environment information, enabling the large language model to correct code-level issues. The modeling errors include: unbounded results, infeasible results, violations of unique deployment constraints, violations of resource capacity constraints, violations of microservice dependency constraints, and results that fail to meet latency calculation verification. Their feedback includes: model expressions, test inputs, the type of constraint violated, and verification results, enabling the large language model to correct model-level issues.

[0098] Step S111: Output the deployment matrix and execute the deployment.

[0099] 1) After the solver executable code passes the preset test cases, run the solver executable code and output the deployment matrix X;

[0100] 2) The rows of the deployment matrix correspond to microservices, and the columns correspond to low-Earth orbit satellite nodes. The values ​​in the matrix indicate whether the corresponding microservice is deployed to the corresponding low-Earth orbit satellite node.

[0101] 3) The management satellite node sends microservice image acquisition, operation parameter configuration and microservice start instructions to the target low-Earth orbit satellite node according to the deployment matrix, thereby completing the deployment of remote sensing microservices within the target deployment time window.

[0102] Example 2

[0103] See Figure 2A space computing network microservice deployment optimization system based on a large model includes: a data acquisition module, a graph construction and latency modeling module, a task description module, a model generation module, a consistency verification module, a code generation and verification module, a feedback correction module, and a deployment output module. Each module processes deployment input data, structured task description, model component generation, model consistency verification, solver code generation, test verification, error feedback, and deployment matrix output in sequence.

[0104] The data acquisition module acquires the description of remote sensing application microservices, data dependencies between microservices, resource requirements of each microservice, resource status of low-Earth orbit satellite nodes, the set of deployable satellites corresponding to each microservice, and the time-varying status of inter-satellite links within the target deployment time window. The graph construction and latency modeling module constructs a directed acyclic graph of microservices and a satellite network graph, and determines the processing latency, transmission latency, propagation latency, and the latest arrival time of data from multiple predecessor microservices. The task description module generates a structured deployment task description. The model generation module inputs the structured deployment task description into a large language model that runs according to a preset prompt template, and its output parameter set P The system comprises a variable set V, a constraint set C, an objective function O, a model expression, and a solver code snippet. The consistency verification module performs deployment variable coverage verification, constraint integrity verification, latency term correspondence verification, and resource dimension verification. The code generation and verification module generates a microservice deployment optimization model and solver executable code, and runs preset test cases. The feedback correction module feeds back the error messages and code snippets corresponding to execution errors, as well as the model expressions, test inputs, and verification results corresponding to modeling errors, to the large language model. The deployment output module outputs a deployment matrix after the solver executable code passes verification and sends microservice deployment instructions.

[0105] Example 3

[0106] See Figure 3 In the specific application scenario of deploying remote sensing microservices in a space computing network, remote sensing tasks are generated or received by satellite nodes and broken down into multiple microservices with dependencies. Different low-Earth orbit (LEO) satellite nodes have different computing resources, storage resources, and communication connection statuses. The management satellite nodes deploy the corresponding microservices to the target LEO satellite nodes according to the deployment matrix, enabling each microservice to complete remote sensing application processing while meeting resource capacity, link reachability, and dependency timing conditions.

[0107] Example 4

[0108] See Figure 4 Optimizing the model generation, code generation, and self-correction verification process includes:

[0109] 1) Input the structured deployment task description into the large language model to obtain the parameter set, variable set, constraint set, objective function, model expression and solver code snippet;

[0110] 2) Perform a consistency check on the output results;

[0111] 3) After successful verification, generate the mixed-integer programming model and solver executable code;

[0112] 4) Run the preset test cases. When an execution error occurs, provide an error message and code snippet. When a modeling error occurs, provide the model expression, test input, and verification result.

[0113] 5) Output the deployment matrix when all preset test cases pass.

[0114] The above embodiments are merely illustrative of the present invention and are not intended to limit the scope of the patent. All equivalent substitutions, modifications, or variations made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for optimizing the deployment of microservices in a space computing network based on a large model, characterized in that, The method includes the following steps: Step S101: Obtain the description of the remote sensing application microservice, the data dependencies between microservices, the resource requirements of each microservice, the resource status of low-orbit satellite nodes, the set of deployable satellites corresponding to each microservice, and the time-varying status of inter-satellite links within the target deployment time window; Step S102: Construct a directed acyclic graph of microservices based on the data dependencies between microservices, and construct a satellite network graph based on the resource status of low-orbit satellite nodes and the time-varying status of inter-satellite links; Step S103: Construct latency calculation relationships based on the microservice directed acyclic graph and satellite network graph; Step S104: Generate a structured deployment task description based on the microservice directed acyclic graph, satellite network graph, resource requirements, deployable satellite set, and latency calculation relationship; Step S105: Input the structured deployment task description into the large language model that runs according to the preset prompt template; Step S106: Based on the input structured deployment task description, the large language model outputs the parameter set P, variable set V, constraint set C, objective function O, model expression, and solver code snippet; Step S107: Perform deployment variable coverage verification, constraint integrity verification, latency item correspondence verification, and resource dimension verification on the parameter set P, variable set V, constraint set C, and objective function O; Step S108: Generate a microservice deployment optimization model from the verified parameter set P, variable set V, constraint set C, objective function O, and model expression, and generate solver executable code based on solver code snippets; Step S109: Run preset test cases to verify the solver executable code. The preset test cases in step S109 include: feasible deployment test cases, resource overrun test cases, microservice dependency test cases, and binary deployment test cases. Step S110: Divide the errors generated during the verification process into execution errors and modeling errors, and feed back the error messages and code snippets corresponding to the execution errors, as well as the model expressions, test inputs and verification results corresponding to the modeling errors, to the large language model for correction; Step S111: After the solver executable code passes the preset test cases, run the solver executable code and output a deployment matrix representing the correspondence between microservices and low-Earth orbit satellite nodes, thereby optimizing the deployment of microservices in the space computing network.

2. The method according to claim 1, characterized in that, The delay calculation relationship in step S103 includes: processing delay. Transmission delay Propagation delay And the latest time the data of multiple predecessor microservices arrives at the startup time. ;in, For microservices The computational load, For satellite nodes computing power For microservices Microservices The amount of data, For satellite nodes To satellite node The link transmission rate at time t, For satellite nodes To satellite node At a distance of time t, For signal propagation speed, For precursor microservices The completion time, For microservices The set of precursor microservices, To support the precursor microservice Satellite nodes.

3. The method for optimizing the deployment of microservices in a space computing network based on a large model as described in claim 1, characterized in that, The structured deployment task description in step S104 includes: scenario field, microservice field, dependency field, satellite resource field, link field, latency field, constraint field, target field, and output field; the microservice field records the microservice identifier, computational load, input data volume, output data volume, and resource requirement vector; the dependency field records the predecessor and successor microservices in the directed acyclic graph of microservices; the satellite resource field records the computational, storage, and memory resource status of low-Earth orbit satellite nodes; the link field records time slices, link reachability, link transmission rate, and distance between satellite nodes; the output field limits the output to a deployment matrix.

4. The method for optimizing the deployment of microservices in a space computing network based on a large model as described in claim 1, characterized in that, The preset prompt template in step S105 includes: task role field, scenario input field, symbol extraction field, modeling requirement field, code generation field, verification requirement field, and output format field. The output format field limits the output of the large language model in the order of parameter set P, variable set V, constraint set C, objective function O, model expression, and solver code fragment. The variable set V includes a binary deployment variable used to indicate whether the microservice is deployed to the corresponding low-Earth orbit satellite node.

5. The method for optimizing the deployment of microservices in a space computing network based on a large model according to claim 1, characterized in that, The deployment variable coverage verification in step S107 is used to confirm that each microservice and its corresponding set of deployable satellites are included in the range of binary deployment variable indexes; the constraint integrity verification is used to confirm that the unique deployment constraint, deployable set constraint, resource capacity constraint, microservice dependency constraint, latency constraint, and binary value constraint are all written into the constraint set C; the latency term correspondence verification is used to confirm that the latency terms in the objective function O and the dependency constraints correspond to processing latency, transmission latency, and propagation latency, respectively; the resource dimension verification is used to confirm that the dimension of the resource requirement vector is consistent with the dimension of the satellite node resource capacity vector.

6. The method for optimizing the deployment of microservices in a space computing network based on a large model according to claim 1, characterized in that, The microservice deployment optimization model in step S108 is a mixed integer programming model; the mixed integer programming model uses a deployment matrix. Let T be the decision variable, and let minimizing the total completion time T be the objective function. , , , ,as well as As a constraint relationship; where, For microservices Deployable satellite collection For microservices The resource demand vector, For satellite nodes The resource capacity vector, For microservices The completion time, For precursor microservices The completion time.

7. The method for optimizing the deployment of microservices in a space computing network based on a large model as described in claim 1, characterized in that, The execution errors in step S110 include: syntax errors, undefined variables, mismatched index dimensions, and runtime errors; the modeling errors include: unbounded results, infeasible results, results that violate unique deployment constraints, results that violate resource capacity constraints, results that violate microservice dependency constraints, and results that fail to meet latency calculation verification.

8. The method for optimizing the deployment of microservices in a space computing network based on a large model according to claim 1, characterized in that, The rows of the deployment matrix in step S111 correspond to the microservices in the directed acyclic graph of microservices. The columns of the deployment matrix correspond to low-Earth orbit satellite nodes; the values ​​in the deployment matrix are used to indicate whether the corresponding microservice is deployed to the corresponding low-Earth orbit satellite node, and the management satellite node sends microservice image acquisition, runtime parameter configuration and microservice startup instructions to the target low-Earth orbit satellite node according to the deployment matrix.

9. A system constructed using the space computing power network microservice deployment optimization method based on a large model as described in claim 1, characterized in that, The system includes: a data acquisition module, a graph construction and latency modeling module, a task description module, a model generation module, a consistency verification module, a code generation and verification module, a feedback correction module, and a deployment output module. The data acquisition module acquires descriptions of remote sensing application microservices, data dependencies between microservices, resource requirements of each microservice, resource status of low-Earth orbit satellite nodes, the set of deployable satellites corresponding to each microservice, and the time-varying status of inter-satellite links within the target deployment time window. The graph construction and latency modeling module constructs a directed acyclic graph of microservices and a satellite network graph, and determines processing latency, transmission latency, propagation latency, and the latest arrival time of data from multiple predecessor microservices. The task description module generates a structured deployment task description. The model generation module... The structured deployment task description is input into the large language model, which runs according to a preset prompt template. Its output includes a parameter set P, a variable set V, a constraint set C, an objective function O, a model expression, and a solver code snippet. The consistency verification module performs deployment variable coverage verification, constraint integrity verification, latency term correspondence verification, and resource dimension verification. The code generation and verification module generates a microservice deployment optimization model and solver executable code, and runs preset test cases. The feedback correction module feeds back the error messages and code snippets corresponding to execution errors, as well as the model expressions, test inputs, and verification results corresponding to modeling errors, to the large language model. The deployment output module outputs a deployment matrix after the solver executable code passes verification and sends microservice deployment instructions.

Images