A method for dynamically mapping a satellite-ground cooperative task and an edge intelligent terminal

Abstract

This invention relates to a dynamic mapping method for satellite-ground collaborative tasks and an edge intelligent terminal. The method includes: responding to the triggering of a task to be processed; acquiring task type, input data source, and service level information; and acquiring a dynamic constraint set in real time; generating a structured task parameter set corresponding to the task to be processed based on the task type, input data source, and service level information; selecting a task dynamic mapping strategy from a preset candidate strategy set based on the structured task parameter set and the dynamic constraint set, under the premise of meeting the hard constraint of the satellite overpass time window; and, according to the task dynamic mapping strategy, jointly scheduling the terminal's communication resources, heterogeneous computing resources, and multi-layer storage cache to perform task data processing and satellite-ground collaborative transmission, thereby improving the success rate and reliability of satellite-ground collaborative tasks.

Description

Technical Field

[0001] This invention relates to the field of sixth-generation mobile communication network technology, and in particular to a dynamic mapping method for satellite-ground collaborative tasks and an edge intelligent terminal. Background Technology

[0002] Low Earth Orbit (LEO) satellites and ground terminals working together is one of the core scenarios of 6G integrated space-ground networks. However, the high-speed movement of LEO satellites results in a strict overpass time window for their communication with ground terminals. A single window typically lasts only a few hundred seconds, and the window interval can be as long as tens of minutes.

[0003] Existing task offloading algorithms on the terminal side (such as latency / energy consumption optimization algorithms in mobile edge computing) are mainly developed based on terrestrial network scenarios, implicitly assuming the continuity of network connections. When these algorithms are directly applied to satellite-ground collaborative scenarios, they can lead to systemic failures due to neglecting the physical constraint of the "overhead time window": the terminal may make an offloading decision based on a momentarily good link state at the end of the window, but because it does not consider whether the remaining window duration is sufficient to cover data backhaul and round-trip latency, the task may be offloaded but cannot be completed before the window ends, resulting in a significant increase in task failure rate. To address the above problems, some existing patents focus on transmission optimization on the network side, but do not solve the problem of autonomous decision-making on the terminal side under strong spatiotemporal constraints. Summary of the Invention

[0004] The main objective of this invention is to provide a dynamic mapping method for satellite-ground collaborative missions and an edge intelligent terminal. By using the satellite overpass time window as a hard constraint for terminal decision-making, this invention fundamentally solves the problem of decision failure caused by ignoring window constraints in traditional offloading algorithms under dynamic satellite-ground environments, thereby improving the success rate and reliability of satellite-ground collaborative missions.

[0005] To achieve the above objectives, this invention provides a dynamic mapping method for space-ground collaborative missions, comprising the following steps: In response to the triggering of a pending task, the system acquires the task type, input data source, and service level information, as well as a set of dynamic constraints in real time. The set of dynamic constraints includes at least the satellite overpass time window, the current estimated uplink available bandwidth, and the current estimated round-trip time; wherein the satellite overpass time window is a hard constraint. Based on the task type, input data source, and business service level information, a structured task parameter set corresponding to the task to be processed is generated. The structured task parameter set includes at least the task data volume parameter, the task computation volume parameter, the maximum allowable delay of the task, the minimum precision requirement of the task, and the task divisibility flag. Under the premise of meeting the hard constraint of satellite overpass time window, based on the structured task parameter set and the dynamic constraint set, a task dynamic mapping strategy is selected from the preset candidate strategy set according to the current task to be processed. The candidate strategy set includes at least local execution strategy, split execution strategy, offload execution strategy and deferred execution strategy. Based on the aforementioned task dynamic mapping strategy, the communication resources, heterogeneous computing resources, and multi-layer storage cache of the jointly scheduled terminal are used to perform task data processing and satellite-ground collaborative transmission.

[0006] Furthermore, the steps of obtaining task type, input data source and business service level information in response to the triggering of a task to be processed, and obtaining a dynamic constraint set in real time, include: Receive pending task requests triggered by upper-layer applications, and parse the pending task requests to obtain task type, input data source and business service level information; The satellite overhead time window is obtained by calculating the satellite's overhead time using the terminal's global satellite navigation and positioning module in conjunction with real-time ephemeris data, or by calibrating the available window information broadcast by the network side or satellite side. The estimated uplink available bandwidth and estimated round-trip delay under the current link mode are obtained through link measurement and updated when the link fluctuates or the mode switches. The satellite overhead time window, the current uplink available bandwidth estimate, and the current round-trip delay estimate are integrated into a dynamic constraint set.

[0007] Further, the step of generating a structured task parameter set corresponding to the task to be processed based on the task type, input data source, and business service level information includes: The service level information is mapped to the maximum allowable latency of the task and the minimum accuracy requirement of the task. Based on the joint estimation of historical task profile information and real-time sampling information corresponding to the task type, the task data volume parameter and the task computation volume parameter are obtained. Based on the task dependency graph corresponding to the task type and the input data source, the task divisibility flag is determined. The task data volume parameter, the task computation volume parameter, the task maximum allowable delay, the task minimum accuracy requirement, and the task divisibility flag are combined into a structured task parameter set.

[0008] Furthermore, the step of determining the candidate strategy set includes: Configure a local execution strategy for completing tasks locally on the terminal; Configure an offloading execution strategy to offload the entire task to an on-board collaborative node for execution; Define a split execution strategy for the terminal to first execute a set of local preprocessing operations, and then unload the preprocessed task data or intermediate results to the on-board collaborative node to execute the remaining tasks; Configure a deferred execution strategy to postpone the execution of a task until the next available window or when the link conditions are met; The local execution strategy, the split execution strategy, the unload execution strategy, and the deferred execution strategy are determined as a candidate strategy set; When determining the candidate strategy set, the configurations of the local execution strategy, the split execution strategy, the unload execution strategy, and the deferred execution strategy must meet the following conditions: Mission-related data processing and satellite-to-ground data transmission must be completed before the satellite overpass time window ends; The amount of data required for uplink backhaul and the transmission schedule must be constrained by the estimated current uplink available bandwidth. The end-to-end completion delay of the task does not exceed the maximum allowable delay of the task and meets the minimum accuracy requirements of the task; The configuration of each strategy is optimized with the goal of minimizing the overall cost, which includes at least task completion latency, terminal energy consumption, and resource costs.

[0009] Furthermore, the step of selecting a task dynamic mapping strategy from a preset candidate strategy set based on the structured task parameter set and the dynamic constraint set, according to the current task to be processed, under the premise of satisfying the hard constraint of the satellite overpass time window, includes: Add the pending task to the task queue; Based on the task data volume parameters, task computation volume parameters, and the dynamic constraint set in the structured task parameter set, calculate the remaining window duration, the estimated local preprocessing latency, and the estimated uplink transmission latency. Based on the remaining window duration, the estimated local preprocessing delay, the estimated uplink transmission delay, and the current round-trip delay estimate, it is verified whether data upload and result return can be completed before the end of the satellite overpass time window if the unload execution strategy or the segmented execution strategy is used to execute the task to be processed; wherein, the current round-trip delay estimate is used to characterize the time required for result return. Based on the verification results, the task dynamic mapping strategy is selected from the candidate strategy set.

[0010] Furthermore, the step of selecting a task dynamic mapping strategy from a preset set of candidate strategies further includes: When the remaining time of the satellite overpass time window is less than the first threshold, or the estimated uplink available bandwidth is lower than the second threshold, the offload execution strategy and the segmented execution strategy are prohibited, and the local execution strategy or the deferred execution strategy is selected.

[0011] Furthermore, the step of jointly scheduling the communication resources, heterogeneous computing resources, and multi-layer storage cache of the terminal according to the task dynamic mapping strategy to perform data processing and satellite-ground cooperative transmission for the task includes: When the determined task dynamic mapping strategy is the segmentation execution strategy, the heterogeneous computing unit of the terminal is invoked to execute a set of local preprocessing operations to generate preprocessed task data or intermediate results; the set of local preprocessing operations includes at least one or more of the following: target detection, trajectory or pose coarse extraction, sparse feature encoding, keyframe selection or semantic summary compression. The task data or intermediate results are offloaded to the on-board collaborative node via a multi-mode communication interface; The remaining tasks are executed by the on-board collaborative node, and the task execution results are returned to the terminal.

[0012] Furthermore, the step of jointly scheduling the communication resources, heterogeneous computing resources, and multi-layer storage cache of the terminal according to the task dynamic mapping strategy to perform data processing and satellite-ground cooperative transmission for the task also includes: Prioritize the task queues, giving priority to task queues that meet the maximum allowable delay of the task. In the multi-layer storage cache, a cache reservation quota is set for the key summary data or high-confidence perception results generated by the local preprocessing operation set; When the satellite overpass time window enters the preset tail segment or the occupancy of the multi-layer storage cache reaches a preset threshold, the transmission priority of the key summary data or high confidence perception results is increased.

[0013] Further, the step of setting a cache reservation quota for the key summary data or high-confidence-aware results generated by the local preprocessing operation set in the multi-layer storage cache includes: Based on the priority level in the service level information, different cache reservation quotas are configured for key summary data or high confidence perception results of different levels. The cache reservation quota is dynamically adjusted based on the remaining duration of the satellite overpass time window, and the cache reservation quota is increased when the remaining duration is less than a preset threshold. Configure the cache reserved quota as a dedicated storage space independent of the ordinary data queue, and prohibit non-critical data from occupying it.

[0014] The present invention also provides an edge intelligent terminal, comprising: The global satellite navigation and positioning module is used to provide terminal location and timing information, and calculate or calibrate the satellite overpass time window by combining real-time ephemeris data or available window information broadcast by the network side or satellite side. A multi-mode communication interface is used to provide a satellite-to-ground backhaul link and obtain estimates of uplink available bandwidth and round-trip delay. Heterogeneous computing units include a CPU and an NPU, wherein the NPU is used to perform local preprocessing operations; Multi-layered storage cache is used to store raw sensing data, preprocessed results, and data packets to be uploaded in layers. The task formal modeling module is used to formalize the task to be processed into a structured set of task parameters; The dynamic constraint integration module is used to obtain the dynamic constraint set in real time. The online optimization decision module is used to jointly solve the structured task parameter set and the dynamic constraint set under the condition that the satellite overpass time window must be a hard constraint, and output the task dynamic mapping strategy. The strategy execution module is used to jointly schedule multi-mode communication interfaces, heterogeneous computing units and multi-layer storage caches according to the task dynamic mapping strategy, and to perform data processing and satellite-ground collaborative transmission of the tasks to be processed.

[0015] The satellite-ground collaborative mission dynamic mapping method and edge intelligent terminal provided by this invention have the following beneficial effects: This invention introduces the satellite overpass time window as a hard constraint into the terminal decision-making paradigm, fundamentally eliminating the systematic misjudgment of traditional algorithms at the end of the window; through the local preprocessing operation set executed by the NPU, the amount of data to be uploaded is significantly compressed, making collaborative transmission within a limited window possible; when the remaining window duration or uplink bandwidth reaches a threshold, unloading and segmentation are forcibly prohibited, reserving a safety margin for actual execution; through a three-layer mechanism of task queue priority control, cache reservation quota, and window end priority promotion, reliable transmission of key data is ensured; a complete closed loop is formed from task triggering to joint scheduling and execution, the key parameter acquisition path is clear, and it has good feasibility. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating a dynamic mapping method for space-ground collaborative missions in one embodiment of the present invention; Figure 2 This is a diagram showing the integrated relationship between task modeling and dynamic constraints in one embodiment of the present invention; Figure 3 This is a flowchart of the execution of a segmentation execution strategy according to an embodiment of the present invention.

[0017] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0018] 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.

[0019] Reference Figure 1 The diagram below illustrates a dynamic mapping method for space-ground collaborative missions proposed in this invention, comprising the following steps: S1, in response to the triggering of the task to be processed, obtain the task type, input data source and service level information, and obtain the dynamic constraint set in real time. The dynamic constraint set includes at least the satellite overpass time window, the current uplink available bandwidth estimate and the current round-trip delay estimate; wherein the satellite overpass time window is a hard constraint. S2, Based on the task type, input data source and business service level information, generate a structured task parameter set corresponding to the task to be processed. The structured task parameter set includes at least the task data volume parameter, the task computation volume parameter, the maximum allowable delay of the task, the minimum precision requirement of the task, and the task divisibility flag. S3, under the premise of meeting the hard constraint of the satellite overpass time window, based on the structured task parameter set and the dynamic constraint set, select a task dynamic mapping strategy from the preset candidate strategy set according to the current task to be processed. The candidate strategy set includes at least the local execution strategy, the split execution strategy, the offload execution strategy and the postponed execution strategy. S4. Based on the task dynamic mapping strategy, the communication resources, heterogeneous computing resources and multi-layer storage cache of the scheduling terminal are jointly used to perform task data processing and satellite-ground collaborative transmission.

[0020] In one embodiment, for step S1, The steps of responding to the triggering of the task to be processed, obtaining the task type, input data source and business service level information, and obtaining the dynamic constraint set in real time include: Receive pending task requests triggered by upper-layer applications, and parse the pending task requests to obtain task type, input data source and business service level information; The satellite overhead time window is obtained by calculating the satellite's overhead time using the terminal's global satellite navigation and positioning module in conjunction with real-time ephemeris data, or by calibrating the available window information broadcast by the network side or satellite side. The estimated uplink available bandwidth and estimated round-trip delay under the current link mode are obtained through link measurement and updated when the link fluctuates or the mode switches. The satellite overhead time window, the current uplink available bandwidth estimate, and the current round-trip delay estimate are integrated into a dynamic constraint set.

[0021] In practical implementation, when the upper-layer application of the terminal detects a specific event or triggers a pending task according to a preset strategy, it generates a pending task request. This request is sent to the terminal's task access and service constraint parsing module, from which the task type, input data source, and service level information are obtained. In the satellite-ground collaborative scenario, the continuous movement of satellites causes changes in the communication window, fluctuations in link quality, and jitter in transmission delay, making the terminal environment significantly dynamic and time-varying. This step, through the collaborative work of the terminal's hardware enhancement layer and link measurement module, obtains three types of core dynamic constraints (satellite overpass time window, current uplink available bandwidth estimate, and current round-trip delay estimate) and integrates them into a unified set of dynamic constraints: Acquisition and calibration of satellite overhead time window: The satellite overhead time window W is defined as the time interval during which the low-Earth orbit satellite is visible to the terminal, denoted as a closed interval W = [t start , t end As a hard constraint of this invention, it means that any mission mapping strategy involving satellite-to-ground transmission must ensure that all data processing and transmission actions occur within t. end The satellite overpass time window must be completed beforehand; otherwise, the strategy is considered infeasible. The satellite overpass time window is obtained through two methods and can be cross-calibrated to improve accuracy. Method 1: Autonomous Calculation: The terminal's global navigation and positioning module provides real-time location and precise timing information. Combined with real-time ephemeris data acquired by the terminal, a window calculation algorithm predicts the visible period of the satellite relative to the terminal, outputting the satellite overpass time window W. Method 2: Broadcast Calibration: The terminal directly receives available window information broadcast from the network or satellite side, obtaining a satellite overpass time window verified by the network side. When both methods are available simultaneously, the terminal performs cross-validation and fusion calibration on the results. Once the satellite overpass time window W is determined, it is marked as an inviolable hard constraint. All subsequent task mapping strategies are formulated to ensure that all data processing and transmission actions occur within the specified time frame. end This is contingent upon prior completion. Real-time measurement of uplink available bandwidth and round-trip delay estimates: The link measurement component built into the multi-mode communication interface continuously monitors the currently active link mode; uplink available bandwidth estimate B... up The values ​​are obtained through link-layer throughput statistics and end-to-end rate probing, and smoothed to suppress transient fluctuations. The estimated round-trip time (RTT) is obtained through transport-layer measurements and probe packet responses, and jitter is suppressed through sliding window statistics.

[0022] When a link mode switch or a significant change in bandwidth / latency is detected, a new round of measurement and reassessment is immediately triggered to ensure that the constraint information used for decision-making always reflects the current link state. The above three types of parameters are encapsulated into a unified set of dynamic constraints, represented in vector form: R(t)=[W, B up [(t), RTT(t)]. This set fully characterizes the satellite-to-ground communication environment faced by the terminal at the current moment: when communication is possible (W), and at what communication rate (B). up ), and the cost of communication delay (RTT).

[0023] In one embodiment, for step S2, The step of generating a structured task parameter set corresponding to the task to be processed based on the task type, input data source, and business service level information includes: The service level information is mapped to the maximum allowable latency of the task and the minimum accuracy requirement of the task. Based on the joint estimation of historical task profile information and real-time sampling information corresponding to the task type, the task data volume parameter and the task computation volume parameter are obtained. Based on the task dependency graph corresponding to the task type and the input data source, the task divisibility flag is determined. The task data volume parameter, the task computation volume parameter, the task maximum allowable delay, the task minimum accuracy requirement, and the task divisibility flag are combined into a structured task parameter set.

[0024] In practice, the task formal modeling module transforms the above information into a structured task parameter set: Service level information is converted into the maximum allowable latency L of the task through predefined mapping rules. max And the minimum accuracy requirement A for the task minThis ensures that business requirements are explicitly constrained in subsequent decision-making. The task data volume parameter D and the task computation volume parameter C are obtained using a joint estimation strategy: based on the historical data scale statistics of similar tasks in historical task profile information, combined with real-time sampling information, lightweight sampling of the input data source is performed at the moment the task is triggered, and an accurate estimate of the task data volume is obtained through weighted fusion or adaptive calibration. The task computation volume parameter is estimated based on the task type and preset model performance analysis information, and corrected based on the current load status of heterogeneous computing units. The task divisibility flag F is determined based on the task dependency graph. If the task dependency graph shows that the task can be decomposed into several subtasks that can be parallelized or piped, and the data coupling between subtasks is low, then F=1 is determined, indicating that the task has the conditions for split execution; otherwise, F=0 is determined, indicating that it can only be executed as a whole. The above five core parameters are combined and encapsulated to generate a structured task parameter set T = {D, C, L}. max A min , F} serves as the unified input interface for subsequent online decision-making.

[0025] In one embodiment, the step of determining the candidate strategy set includes: Configure a local execution strategy for completing tasks locally on the terminal; Configure an offloading execution strategy to offload the entire task to an on-board collaborative node for execution; Define a split execution strategy for the terminal to first execute a set of local preprocessing operations, and then unload the preprocessed task data or intermediate results to the on-board collaborative node to execute the remaining tasks; Configure a deferred execution strategy to postpone the execution of a task until the next available window or when the link conditions are met; The local execution strategy, the split execution strategy, the unload execution strategy, and the deferred execution strategy are determined as a candidate strategy set; When determining the candidate strategy set, the configurations of the local execution strategy, the split execution strategy, the unload execution strategy, and the deferred execution strategy must meet the following conditions: Mission-related data processing and satellite-to-ground data transmission must be completed before the satellite overpass time window ends; The amount of data required for uplink backhaul and the transmission schedule must be constrained by the estimated current uplink available bandwidth. The end-to-end completion delay of the task does not exceed the maximum allowable delay of the task and meets the minimum accuracy requirements of the task; The configuration of each strategy aims to minimize the overall cost, which includes at least task completion latency, terminal energy consumption, and resource costs.

[0026] In practical implementation, the present invention predefines four typical task mapping modes (local execution strategy, split execution strategy, offload execution strategy and deferred execution strategy), covering all possibilities from completely local processing to completely offload, as well as split execution and delayed execution due to insufficient conditions in between, ensuring that the terminal has at least one feasible strategy option under any foreseeable window constraints and link state.

[0027] The Local Execution (LOCAL) strategy means that the entire processing flow of the task to be processed is completed locally on the terminal, without involving any satellite-to-ground data transmission. It is suitable for scenarios where the satellite overpass time window is too short, uplink bandwidth is severely insufficient, or the task has extremely high real-time requirements and offloading would increase latency. The Offload Execution (OFFLOAD) strategy means that the task to be processed is packaged as a whole and completely offloaded to the on-board collaborative node for execution via a multi-mode communication interface. The terminal is only responsible for receiving the results. It is suitable for scenarios where the satellite overpass time window is ample, uplink bandwidth is sufficient, and the on-board computing power is superior to the terminal's computing power. The Split Execution (SPLIT) strategy means that the terminal first executes a set of local preprocessing operations P to generate intermediate results with significantly reduced data volume, and then offloads the intermediate results to the on-board collaborative node to execute the remaining high-complexity computation tasks. Its essence is "trading local computation for transmission time," using the NPU for rapid preprocessing to compress the amount of data to be uploaded from D to D', thereby achieving feasible collaboration within a limited window. The DEFER (Delayed Execution) strategy refers to delaying task execution until the next available window or when link conditions improve, when the remaining window duration is insufficient, link quality is poor, or terminal load is too high. This serves as a safety fallback mechanism to prevent task failure due to forced execution under unfavorable conditions. The above four strategies are collectively defined as a candidate strategy set, denoted as π∈{LOCAL, SPLIT, OFFLOAD, DEFER}, serving as the output space for subsequent online decision-making. The reason traditional methods fail in satellite-ground collaborative scenarios is fundamentally due to the absence of a hard constraint—the satellite overpass time window—in the domain of their decision paradigm. This invention introduces W into the constraint set and defines it as a hard constraint, transforming the mathematical structure of the decision problem from resource optimization within an infinite time frame to feasibility assurance within a finite time window, thus fundamentally eliminating systematic misjudgments at the end of the window.

[0028] Reference Figure 2 This is a diagram illustrating the integration of task modeling and dynamic constraints in one embodiment of the present invention. A policy configuration is only allowed to enter the candidate set if it meets the following core constraints: Constraint 1, Window Completion Constraint. Task-related data processing and satellite-to-ground backhaul actions must be completed before the satellite overpass time window ends. For the OFFLOAD strategy, the sum of the data upload latency and the result backhaul latency must be less than the remaining window duration; for the SPLIT strategy, the sum of the local preprocessing latency and the compressed data upload latency must be less than the remaining window duration. Any strategy configuration that fails to meet this constraint will be removed from the candidate set. Constraint 2, Bandwidth Matching Constraint. The amount of data required for uplink backhaul and the transmission schedule must be constrained by the current estimated uplink available bandwidth. For OFFLOAD and SPLIT strategies involving data transmission, the amount of data to be transmitted must match B_up to ensure that transmission can be completed within the expected time under available bandwidth, preventing transmission timeouts due to overestimation of bandwidth. Constraint 3, Service Quality Constraint. The end-to-end completion latency of the task must not exceed the maximum allowable latency L of the task. max And meet the minimum accuracy requirement A for the task. min For the LOCAL strategy, it is necessary to verify whether the local computation latency is less than L. max For the SPLIT strategy, it is necessary to verify whether the sum of local preprocessing delay, transmission delay, and on-board computation delay is less than L. max Meanwhile, does the accuracy loss introduced by preprocessing still exist in A? min Within the permissible range.

[0029] Under the premise of satisfying the above constraints, the strategy selection is guided by minimizing the overall cost. The overall cost function is defined as J = f(L, E, Cost), where: the task completion delay L includes the sum of local processing delay, transmission delay, on-board processing delay, and result return delay. The calculation method varies for different strategies, but all must satisfy L ≤ L max Terminal energy consumption E includes local computing energy consumption and communication transmission energy consumption; resource cost Cost includes link occupancy cost, computing resource quota consumption, etc., and can be flexibly configured according to the operation strategy. The comprehensive cost function J is implemented through weighted summation or hierarchical optimization. The specific weights can be dynamically adjusted according to the business strategy, but cost optimization must be carried out under the premise of meeting the aforementioned three hard constraints.

[0030] In one embodiment, for step S3, The step of selecting a task dynamic mapping strategy from a preset candidate strategy set based on the structured task parameter set and the dynamic constraint set, according to the current task to be processed, under the premise of meeting the hard constraint of the satellite overpass time window, includes: Add the task to be processed to the task queue; Based on the task data volume parameters, task computation volume parameters, and the dynamic constraint set in the structured task parameter set, calculate the remaining window duration, the estimated local preprocessing latency, and the estimated uplink transmission latency. Based on the remaining window duration, the estimated local preprocessing delay, the estimated uplink transmission delay, and the estimated round-trip delay, it is verified whether the task to be processed can be completed before the end of the satellite overpass time window if the offload execution strategy or the segmented execution strategy is adopted. Based on the verification results, the task dynamic mapping strategy is selected from the candidate strategy set.

[0031] In practical implementation, after completing task modeling and dynamic constraint acquisition, the terminal enters the online decision-making phase: at the current moment, for the current task to be processed, an optimal and feasible strategy is selected from a pre-set set of candidate strategies, enabling the task to be completed with minimal overall cost while satisfying the hard constraint of the satellite overpass time window. After the task to be processed is modeled, it is added to the task queue. The task queue is a priority queue maintained by the terminal, used to manage multiple concurrent or waiting tasks. Each task in the queue carries its structured task parameter set T={D,C,L}. max A min The action of adding a task to the queue triggers an online decision-making process for that task, and the decision-making module processes them sequentially according to the queue order or priority order.

[0032] Based on the structured task parameter set and dynamic constraint set, the quantitative parameters required for decision-making are calculated: Remaining window duration T rem According to the satellite overhead time window W=[t start , t end [and current time t] now Calculate T rem = t end - t now This represents the remaining time budget before the window closes.

[0033] Local preprocessing estimated latency T pre Based on the task computational load parameter C and the processing capabilities of the heterogeneous computing units on the terminal, estimate the time required to execute the local preprocessing operation set P. For indivisible tasks (F=0), T pre Meaningless; for divisible tasks, T pre It is a key parameter that determines the feasibility of the segmentation strategy.

[0034] Uplink transmission estimated delay T tx Based on the amount of data to be transmitted and the estimated uplink bandwidth B. up Estimate the transmission time. For the offload execution strategy, the amount of data to be transmitted is the total task data size D; for the split execution strategy, the amount of data to be transmitted is the compressed intermediate result data size D'. tx = Amount of data to be transmitted / Bup Protocol overhead and retransmission margin adjustments can be added.

[0035] Feasibility assessment for two strategies that may involve data transmission: The determination condition for the uninstallation execution strategy is: T pre (If the task requires local preprocessing before it can be unloaded, this will be included) +T tx (Complete data D) + RTT ≤ T rem The RTT (Return Time) is recorded to ensure that the task execution result can be sent back to the terminal before the window closes; otherwise, even if the data upload is completed, the task is considered to have failed if the result cannot be retrieved.

[0036] For the split execution strategy, the decision condition is: T pre +T tx (Compressed data D') + RTT ≤ T rem Because local preprocessing and uplink transmission are performed sequentially, latency accumulates.

[0037] The local execution strategy does not involve satellite-to-ground transmission, and the window constraint is naturally satisfied (but L must be satisfied). max (Constraints), no in-window completion verification is required. The postponement execution strategy delays the decision to the future, and no verification is required at the current moment. Based on the verification results, a subset of feasible strategies is obtained: if both unloading and splitting satisfy the in-window completion condition, both are feasible; if only splitting is feasible, splitting is the only transport-related option; if neither is feasible, transport-related strategies are prohibited, and only local execution or postponement execution can be chosen. Based on the subset of feasible strategies, optimization is performed with the goal of minimizing the overall cost. For each feasible strategy, the overall cost J = f(L, E, Cost) is calculated: L (task completion delay): local execution L = T local ; Segment execution L = T pre +T tx + T cloud + RTT; Unload execution L =T tx + T cloud + RTT; E (Terminal Power Consumption) is estimated based on local computing power consumption and communication transmission power consumption; Cost (Resource Cost) is estimated based on link occupancy, quota consumption, etc. The feasible strategy that minimizes the overall cost J is selected as the final dynamic task mapping strategy.

[0038] In one embodiment, the step of selecting a task dynamic mapping strategy from a preset set of candidate strategies further includes: When the remaining duration of the satellite overpass time window is less than the first threshold, or the estimated value of the available uplink bandwidth is lower than the second threshold, the offloading execution strategy and the splitting execution strategy are prohibited from being selected, and the local execution strategy or the suspension execution strategy is selected.

[0039] In a specific implementation, even if it is confirmed through the "can be completed within the window" determination that a certain strategy meets the time feasibility condition at the current moment, it may still fail due to unexpected situations during actual execution. This embodiment introduces a threshold-triggered security protection mechanism. By setting two key thresholds, high-risk strategy selections are杜绝 at the decision-making stage. The first threshold is the remaining window duration threshold: when the remaining window duration T rem is less than the first threshold W0, the decision module executes forced prohibition and strategy fallback: forced prohibition, regardless of the result of the "can be completed within the window" determination, prohibits the selection of the offloading execution strategy and the splitting execution strategy. At this time, the remaining time is already too tight, and any minor perturbation may cause timeout. Strategy fallback, automatically switches to evaluating local execution and suspension execution. If local execution can be completed within L max and meets the accuracy requirements, local execution is selected; otherwise, suspension execution is selected, and the task is delayed until the next available window. The second threshold is the available uplink bandwidth threshold: when the available uplink bandwidth B up is less than the second threshold B0, the decision module executes forced prohibition and strategy fallback: forced prohibition, prohibits the selection of the offloading execution strategy and the splitting execution strategy. At this bandwidth, even the compressed intermediate results may have too long transmission delays and a very high failure probability. Strategy fallback, the same processing logic as when triggered by the first threshold, selects local execution or suspension execution.

[0040] The first threshold W0 and the second threshold B0 can be triggered independently (T rem <W0 or B up <B0), or they can take effect in combination. The threshold parameters can be dynamically determined according to the service level of the service, historical transmission statistics, and safety margin configuration, and are stored in the terminal as system parameters for the decision module to call in real time through offline calibration, operation and maintenance configuration, or business policy distribution. For example, high-priority tasks can use more conservative thresholds (larger W0), and low-priority tasks can use more aggressive thresholds; the thresholds can also be dynamically adjusted according to historical transmission success rates.

[0041] In one embodiment, for step S4, The step of jointly scheduling the communication resources, heterogeneous computing resources, and multi-layer storage caches of the terminal according to the task dynamic mapping strategy and performing data processing and space-ground collaborative transmission of the task includes: When the determined task dynamic mapping strategy is the segmentation execution strategy, the heterogeneous computing unit of the terminal is invoked to execute a set of local preprocessing operations to generate preprocessed task data or intermediate results; the set of local preprocessing operations includes at least one or more of the following: target detection, trajectory or pose coarse extraction, sparse feature encoding, keyframe selection or semantic summary compression. The task data or intermediate results are offloaded to the on-board collaborative node via a multi-mode communication interface; The remaining tasks are executed by the on-board collaborative node, and the task execution results are returned to the terminal.

[0042] Reference Figure 3 The following is a flowchart of the segmented execution strategy according to an embodiment of the present invention. In specific implementation, the essence of the segmented execution strategy is "trading local computing for transmission time", that is, using the fast processing capability of the local heterogeneous computing unit of the terminal to perform lightweight preprocessing before the data is transmitted uplink, compressing the original sensing data into an intermediate result with a significantly reduced data volume, thereby achieving feasible satellite-ground coordination within a limited satellite overpass time window.

[0043] When the task dynamic mapping strategy output by the online decision module is a segmented execution strategy, the strategy execution module issues local preprocessing operation instructions to the heterogeneous computing units. The local preprocessing operation set P is pre-configured or dynamically selected according to the task type and business requirements. Its core objective is to extract core information from the raw sensing data in the shortest possible time while significantly compressing the amount of data to be transmitted.

[0044] The set of local preprocessing operations includes at least one or more of the following operations: Target detection: Run a lightweight target detection model on the input video stream, image or radar point cloud data to identify targets of interest in the scene, output target bounding boxes, category confidence and other information, and transform the raw data into a structured list of targets.

[0045] Coarse trajectory or attitude extraction: For tasks that require continuous tracking, the target tracking algorithm is run to extract the target's motion trajectory or coarse attitude information, and the continuous data frames are converted into a sequence of trajectory points.

[0046] Sparse feature encoding: Deep features of the original data are extracted using a deep learning model and compressed into a fixed-dimensional feature vector through sparse encoding or quantization techniques, which is then directly used as the input to the on-board recognition model.

[0047] Keyframe selection: For video streaming tasks, frames containing key information are selected from consecutive video frames through scene change detection or motion analysis, and only the selected keyframes are uploaded instead of the entire video stream.

[0048] Semantic summarization compression: Run semantic segmentation or scene understanding models to generate a semantic-level description of the scene, converting the original image into structured text or structured data summaries.

[0049] All of the above operations are performed on the terminal's NPU (Neural Processing Unit), with an estimated local preprocessing latency T. pre The process can be kept at a low level to ensure sufficient window space remains for uplink transmission. In this invention, the CPU is responsible for policy interpretation, task orchestration, and lightweight control logic, while the NPU is dedicated to performing local preprocessing operations. This heterogeneous division of labor, with "CPU decision-making + NPU execution," enables the terminal to complete data compression within milliseconds, securing valuable window time for subsequent uplink transmission.

[0050] After local preprocessing is completed, significantly compressed task data or intermediate results are generated. Examples of outputs for different operations: target detection outputs a list of targets (including location, category, and confidence); trajectory extraction outputs a sequence of trajectory points; feature encoding outputs feature vectors; keyframe selection outputs selected compressed keyframes; semantic summarization outputs a structured description. The data volume D' of the intermediate results is typically only 5%-30% of the original data volume D, with the specific compression ratio depending on the type of preprocessing operation and parameter configuration. The generated intermediate results are written to a key summary queue in a multi-layered storage cache, and a cache reservation quota is set to ensure that they are not overtaken by subsequent low-priority data. After local preprocessing is completed, the policy execution module, in conjunction with the multi-mode communication interface, initiates uplink transmission: determining the transmission rate and sending strategy based on the current link status and remaining window duration; and sending the intermediate results to the designated on-board collaborative node via the multi-mode communication interface. Upon receiving the intermediate results, the on-board collaborative node executes the remaining high-complexity computational tasks, such as large-scale deep learning inference, multi-target association analysis, and complex scene understanding. After completing processing, the on-board node generates the final task execution result and returns it to the terminal via the downlink. The terminal receives the mission execution results returned from the satellite and delivers them to the upper-layer application that initiated the request, completing the collaborative mission loop of "local preprocessing - on-board unloading - result return". The entire process is completed within the satellite's overhead time window W.

[0051] In one embodiment, the step of jointly scheduling the communication resources, heterogeneous computing resources, and multi-layer storage cache of the terminal according to the task dynamic mapping strategy to perform data processing and satellite-ground cooperative transmission for the task further includes: Prioritize the task queues, giving priority to task queues that meet the maximum allowable delay of the task. In the multi-layer storage cache, a cache reservation quota is set for the key summary data or high-confidence perception results generated by the local preprocessing operation set; When the satellite overpass time window enters the preset tail segment or the occupancy of the multi-layer storage cache reaches a preset threshold, the transmission priority of the key summary data or high confidence perception results is increased.

[0052] In practical implementation, to ensure that critical tasks and critical data are given priority in complex scenarios such as multi-task concurrency, resource contention, and window tail segments, and to avoid transmission failures due to resource contention, this embodiment introduces three collaborative mechanisms, including task queue priority control, cache reservation quota mechanism, and window tail segment priority promotion mechanism. Specifically: Task queue priority control: based on service level information or maximum allowable latency L for the task. max The task queue is divided into multiple priority levels, and each task is assigned a priority level based on its L. max Tasks are assigned to the corresponding priority queues. The policy execution module prioritizes retrieving tasks from the high-priority queues for local computation or uplink transmission, ensuring that latency-sensitive tasks receive the highest priority resource allocation. When a high-priority task arrives, if a low-priority task is currently being processed, the system supports task-level preemption, pausing the low-priority task to release resources for the high-priority task. Cache reservation quota mechanism: In the multi-layered storage cache, a dedicated cache area is allocated for critical summary data or high-confidence sensing results, configured with a fixed reservation quota, independent of ordinary data queues, prohibiting non-critical data from occupying it. The reservation quota can be dynamically configured based on business service level information. During cache writing, the data pipeline and multi-layered cache management module check the priority markers of data packets. For data packets carrying "critical summary" or "high confidence" markers, they are forcibly written to the reserved quota area. Window tail priority promotion mechanism: The policy execution module continuously monitors the remaining duration T of the satellite overpass time window. rem When T rem When the data volume is less than the preset tail-end threshold, the system determines that the data has entered the tail-end of the window. The transmission priority of the queue containing critical summary data or high-confidence perception results is automatically increased, allowing it to surpass ordinary data queues and gain priority scheduling rights. This is achieved by setting the scheduling weight of the critical summary queue to the highest level, pausing or limiting the transmission of ordinary data queues, or setting a high-priority marker for critical data packets at the link layer. When the total occupancy of multi-layer storage cache reaches the preset threshold, the priority increase mechanism is also triggered, proactively increasing the transmission priority of critical summary data to reduce cache pressure. These three mechanisms work together to ensure that critical tasks gain an advantageous position during resource allocation through task queue priority control; cache reservation quotas ensure that critical data is not overwhelmed during storage; and window tail-end priority increase ensures that critical data receives last-minute protection during transmission, forming a full-link priority guarantee from "task enqueuing - data storage - window transmission." This fundamentally avoids the window tail-end transmission failure problem caused by resource contention or improper scheduling in traditional solutions.

[0053] In one embodiment, the step of setting a cache reservation quota for key summary data or high-confidence-aware results generated by the local preprocessing operation set in the multi-layer storage cache includes: Based on the priority level in the service level information, different cache reservation quotas are configured for key summary data or high confidence perception results of different levels. The cache reservation quota is dynamically adjusted based on the remaining duration of the satellite overpass time window, and the cache reservation quota is increased when the remaining duration is less than a preset threshold. Configure the cache reserved quota as a dedicated storage space independent of the ordinary data queue, and prohibit non-critical data from occupying it.

[0054] In practical implementation, within the limited multi-tiered storage cache resources, dedicated protection zones are designated for critical data to ensure they are not overtaken by ordinary data. This embodiment further refines the quota setting mechanism: Tiered quota configuration based on service level: Differentiated configuration is applied based on the priority level in the service level information: high priority levels (such as real-time security and emergency response) are configured with larger cache reservation quotas; medium priority levels (such as routine monitoring and periodic reporting) are configured with medium-sized cache reservation quotas; low priority levels (such as background data collection and offline archiving) are configured with smaller cache reservation quotas or only dynamic quotas. Specific quota values ​​can be calibrated offline or dynamically adjusted online based on the total terminal cache capacity, historical peak load statistics, and operational strategies.

[0055] Dynamic quota adjustment based on remaining window duration: When the remaining window duration T... rem When T exceeds the preset adjustment threshold, maintain the normal tiered quota configuration; when T... rem When the value is less than the preset threshold, it is determined to enter the end of the window. The cache quota is automatically increased for key summary data or high confidence perception results to provide sufficient buffer for the end of the transmission.

[0056] Independent configuration and isolation protection of dedicated storage space: In a multi-tiered storage cache, a separate logical queue is established for critical digest data, mapping the configured reserved quota to the capacity limit of this independent queue. Data packets carrying "critical digest" or "high confidence" tags are written to the critical digest queue, and their capacity consumption is included in the reserved quota. Ordinary data packets are written to the ordinary data queue and must not occupy space in the critical digest queue. When the critical digest queue is close to the reserved quota limit, a write protection mechanism is triggered, preventing ordinary data from being written to the critical digest queue even if there is still free space in the ordinary data queue. If the critical digest queue is full, an in-queue eviction policy is initiated, but space is never preempted from the ordinary data queue. Conversely, when the ordinary data queue is full, if the critical digest queue still has free quota, ordinary data cannot borrow that space, ensuring the exclusivity of the reserved quota. Through this isolation design, critical digest data obtains a storage environment isolated from ordinary data. No matter how rapidly ordinary data grows, it will not impact the storage stability of critical data.

[0057] One embodiment of the present invention also provides an edge intelligent terminal, comprising: The global satellite navigation and positioning module is used to provide terminal location and timing information, and calculate or calibrate the satellite overpass time window by combining real-time ephemeris data or available window information broadcast by the network side or satellite side. A multi-mode communication interface is used to provide a satellite-to-ground backhaul link and obtain estimates of uplink available bandwidth and round-trip delay. Heterogeneous computing units include a CPU and an NPU, wherein the NPU is used to perform local preprocessing operations; Multi-layered storage cache is used to store raw sensing data, preprocessed results, and data packets to be uploaded in layers. The task formal modeling module is used to formalize the task to be processed into a structured set of task parameters; The dynamic constraint integration module is used to obtain the dynamic constraint set in real time. The online optimization decision module is used to jointly solve the structured task parameter set and the dynamic constraint set under the condition that the satellite overpass time window must be a hard constraint, and output the task dynamic mapping strategy. The strategy execution module is used to jointly schedule multi-mode communication interfaces, heterogeneous computing units and multi-layer storage caches according to the task dynamic mapping strategy, and to perform data processing and satellite-ground collaborative transmission of the tasks to be processed.

[0058] Specifically, the collaborative working process of the above modules is as follows: the task formal modeling module receives the task request triggered by the upper-layer application and generates a structured task parameter set T; the dynamic constraint integration module obtains W and B from the GNSS module and the multi-mode communication interface in real time. upThe system uses RTT to form a dynamic constraint set R(t); the online optimization decision module uses W as a hard constraint and solves for the policy π based on T and R(t); the policy execution module calls the heterogeneous computing unit to perform local preprocessing based on π, writes the intermediate results into a multi-layer storage cache, and offloads them to the satellite node through a multi-mode communication interface, and finally receives the results and returns them to the upper layer application, forming a complete closed loop.

[0059] In summary, this invention, in response to the triggering of a pending task, acquires task type, input data source, and service level information, and formalizes the pending task into a structured task parameter set; acquires a dynamic constraint set; under the premise of satisfying the hard constraint of the satellite overpass time window, based on the structured task parameter set and the dynamic constraint set, selects a task dynamic mapping strategy from a preset candidate strategy set according to the current pending task; and according to the task dynamic mapping strategy, jointly schedules the communication resources, heterogeneous computing resources, and multi-layer storage cache of the terminal to perform task data processing and satellite-ground collaborative transmission, thereby improving the success rate and reliability of satellite-ground collaborative tasks.

[0060] The above description is only a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A dynamic mapping method for space-ground collaborative missions, characterized in that, Includes the following steps: In response to the triggering of a pending task, the system obtains the task type, input data source and service level information, and acquires a set of dynamic constraints in real time. The set of dynamic constraints includes at least the satellite overhead time window, the current uplink available bandwidth estimate and the current round-trip delay estimate. The satellite overhead time window is a hard constraint. Based on the task type, input data source, and business service level information, a structured task parameter set corresponding to the task to be processed is generated. The structured task parameter set includes at least the task data volume parameter, the task computation volume parameter, the maximum allowable delay of the task, the minimum precision requirement of the task, and the task divisibility flag. Under the premise of meeting the hard constraint of satellite overpass time window, based on the structured task parameter set and the dynamic constraint set, a task dynamic mapping strategy is selected from the preset candidate strategy set according to the current task to be processed. The candidate strategy set includes at least local execution strategy, split execution strategy, offload execution strategy and deferred execution strategy. Based on the aforementioned task dynamic mapping strategy, the communication resources, heterogeneous computing resources, and multi-layer storage cache of the jointly scheduled terminal are used to perform task data processing and satellite-ground collaborative transmission.

2. The dynamic mapping method for space-ground collaborative missions according to claim 1, characterized in that, The steps of responding to the triggering of a pending task, obtaining task type, input data source and business service level information, and obtaining a dynamic constraint set in real time include: Receive pending task requests triggered by upper-layer applications, and parse the pending task requests to obtain task type, input data source and business service level information; The satellite overhead time window is obtained by calculating the satellite's overhead time using the terminal's global satellite navigation and positioning module in conjunction with real-time ephemeris data, or by calibrating the available window information broadcast by the network side or satellite side. The estimated uplink available bandwidth and estimated round-trip delay under the current link mode are obtained through link measurement and updated when the link fluctuates or the mode switches. The satellite overhead time window, the current uplink available bandwidth estimate, and the current round-trip delay estimate are integrated into a dynamic constraint set.

3. The dynamic mapping method for space-ground collaborative missions according to claim 1, characterized in that, The step of generating a structured task parameter set corresponding to the task to be processed based on the task type, input data source, and business service level information includes: The service level information is mapped to the maximum allowable latency of the task and the minimum accuracy requirement of the task. Based on the joint estimation of historical task profile information and real-time sampling information corresponding to the task type, the task data volume parameter and the task computation volume parameter are obtained. Based on the task dependency graph corresponding to the task type and the input data source, the task divisibility flag is determined. The task data volume parameter, the task computation volume parameter, the task maximum allowable delay, the task minimum accuracy requirement, and the task divisibility flag are combined into a structured task parameter set.

4. The dynamic mapping method for space-ground collaborative missions according to claim 1, characterized in that, The steps for determining the candidate strategy set include: Configure a local execution strategy for completing tasks locally on the terminal; Configure an offloading execution strategy to offload the entire task to an on-board collaborative node for execution; Define a split execution strategy for the terminal to first execute a set of local preprocessing operations, and then unload the preprocessed task data or intermediate results to the on-board collaborative node to execute the remaining tasks; Configure a deferred execution strategy to postpone the execution of a task until the next available window or when the link conditions are met; The local execution strategy, the split execution strategy, the unload execution strategy, and the deferred execution strategy are determined as a candidate strategy set; When determining the candidate strategy set, the configurations of the local execution strategy, the split execution strategy, the unload execution strategy, and the deferred execution strategy must meet the following conditions: Mission-related data processing and satellite-to-ground data transmission must be completed before the satellite overpass time window ends; The amount of data required for uplink backhaul and the transmission schedule must be constrained by the estimated current uplink available bandwidth. The end-to-end completion delay of the task does not exceed the maximum allowable delay of the task and meets the minimum accuracy requirements of the task; The configuration of each strategy aims to minimize the overall cost, which includes at least task completion latency, terminal energy consumption, and resource costs.

5. The dynamic mapping method for space-ground collaborative missions according to claim 1, characterized in that, The step of selecting a task dynamic mapping strategy from a preset candidate strategy set based on the structured task parameter set and the dynamic constraint set, according to the current task to be processed, under the premise of meeting the hard constraint of the satellite overpass time window, includes: Add the task to be processed to the task queue; Based on the task data volume parameters, task computation volume parameters, and the dynamic constraint set in the structured task parameter set, calculate the remaining window duration, the estimated local preprocessing latency, and the estimated uplink transmission latency. Based on the remaining window duration, the estimated local preprocessing delay, the estimated uplink transmission delay, and the current round-trip delay estimate, it is verified whether data upload and result return can be completed before the end of the satellite overpass time window if the unload execution strategy or the segmented execution strategy is used to execute the task to be processed; wherein, the current round-trip delay estimate is used to characterize the time required for result return. Based on the verification results, the task dynamic mapping strategy is selected from the candidate strategy set.

6. The dynamic mapping method for space-ground collaborative missions according to claim 5, characterized in that, The step of selecting a task dynamic mapping strategy from a preset set of candidate strategies further includes: When the remaining time of the satellite overpass time window is less than the first threshold, or the estimated uplink available bandwidth is lower than the second threshold, the offload execution strategy and the segmented execution strategy are prohibited, and the local execution strategy or the deferred execution strategy is selected.

7. The dynamic mapping method for space-ground collaborative missions according to claim 1, characterized in that, The steps of jointly scheduling the communication resources, heterogeneous computing resources, and multi-layer storage cache of the terminal according to the task dynamic mapping strategy, and performing data processing and satellite-ground cooperative transmission for the task, include: When the determined task dynamic mapping strategy is the segmentation execution strategy, the heterogeneous computing unit of the terminal is invoked to execute a set of local preprocessing operations to generate preprocessed task data or intermediate results; the set of local preprocessing operations includes at least one or more of the following: target detection, trajectory or pose coarse extraction, sparse feature encoding, keyframe selection or semantic summary compression. The task data or intermediate results are offloaded to the on-board collaborative node via a multi-mode communication interface; The remaining tasks are executed by the on-board collaborative node, and the task execution results are returned to the terminal.

8. The dynamic mapping method for space-ground collaborative missions according to claim 7, characterized in that, The step of jointly scheduling the communication resources, heterogeneous computing resources, and multi-layer storage cache of the terminal according to the task dynamic mapping strategy, and performing data processing and satellite-ground collaborative transmission for the task, further includes: Prioritize the task queues, giving priority to task queues that meet the maximum allowable delay of the task. In the multi-layer storage cache, a cache reservation quota is set for the key summary data or high-confidence perception results generated by the local preprocessing operation set; When the satellite overpass time window enters the preset tail segment or the occupancy of the multi-layer storage cache reaches a preset threshold, the transmission priority of the key summary data or high confidence perception results is increased.

9. The dynamic mapping method for space-ground collaborative missions according to claim 8, characterized in that, The step of setting a cache reservation quota for the key summary data or high-confidence-aware results generated by the local preprocessing operation set in the multi-layer storage cache includes: Based on the priority level in the service level information, different cache reservation quotas are configured for key summary data or high confidence perception results of different levels. The cache reservation quota is dynamically adjusted based on the remaining duration of the satellite overpass time window, and the cache reservation quota is increased when the remaining duration is less than a preset threshold. Configure the cache reserved quota as a dedicated storage space independent of the ordinary data queue, and prohibit non-critical data from occupying it.

10. An edge intelligent terminal, characterized in that, include: The global satellite navigation and positioning module is used to provide terminal location and timing information, and calculate or calibrate the satellite overpass time window by combining real-time ephemeris data or available window information broadcast by the network side or satellite side. A multi-mode communication interface is used to provide a satellite-to-ground backhaul link and obtain estimates of uplink available bandwidth and round-trip delay. Heterogeneous computing units include a CPU and an NPU, wherein the NPU is used to perform local preprocessing operations; Multi-layered storage cache is used to store raw sensing data, preprocessed results, and data packets to be uploaded in layers. The task formal modeling module is used to formalize the task to be processed into a structured set of task parameters; The dynamic constraint integration module is used to obtain the dynamic constraint set in real time. The online optimization decision module is used to jointly solve the structured task parameter set and the dynamic constraint set under the condition that the satellite overpass time window must be a hard constraint, and output the task dynamic mapping strategy. The strategy execution module is used to jointly schedule multi-mode communication interfaces, heterogeneous computing units and multi-layer storage caches according to the task dynamic mapping strategy, and to perform data processing and satellite-ground collaborative transmission of the tasks to be processed.

Images