Satellite and unmanned aerial vehicle cooperative task planning method and system for earth observation

Abstract

The application provides a satellite and unmanned aerial vehicle cooperative task planning method and system for earth observation, and relates to the technical field of earth observation. By generating satellite priority initial individuals and unmanned aerial vehicle priority initial individuals, the characteristics of heterogeneous resources can be fully considered, and the initial population generation strategy in stages can effectively solve the problems of low initial population quality and insufficient search diversity in the prior art. The global evolutionary search and local optimization are combined to improve the accuracy and convergence speed of the solution, and the defects of insufficient effectiveness of global search and insufficient pertinence of local improvement in the prior art are compensated. Through the initial population generation mechanism and the global-local cooperative optimization strategy, the complex challenges of satellite and unmanned aerial vehicle heterogeneous resource cooperative planning in the earth observation task can be more effectively coped with, so that a better task planning scheme is obtained, and the efficiency and success rate of task execution are improved.

Description

Technical Field

[0001] This invention relates to the field of Earth observation technology, and more specifically, to a satellite and UAV collaborative mission planning method and system for Earth observation. Background Technology

[0002] In recent years, Earth observation missions have been increasingly developing towards greater complexity, diversity, and timeliness. Satellite platforms offer a significant advantage in terms of wide coverage, but their flexibility is relatively limited; unmanned aerial vehicle (UAV) platforms, on the other hand, possess extremely high mobility but are constrained by insufficient endurance. Based on the complementarity of these two types of platforms, collaborative observation using multiple space-air platforms has gradually become an important development direction for addressing complex observation needs and compensating for the shortcomings of single platforms.

[0003] Currently, for mission planning problems involving heterogeneous observation resources such as satellites and drones, existing technologies mostly employ a hierarchical iterative framework. While such methods can achieve cross-platform collaborative scheduling to some extent, they suffer from significant shortcomings under complex constraints, both in terms of the effectiveness of global search and the targeted nature of local improvements, making it difficult to fully meet the actual needs of collaborative observation across heterogeneous platforms.

[0004] In the global search phase, existing methods have significant shortcomings in initial population construction and search perturbation design: they often rely on random generation or general mutation methods, failing to fully combine the resource characteristics of satellites and UAVs to construct high-quality initial solutions. This is not only detrimental to maintaining search diversity under complex heterogeneous constraints, but also restricts the improvement of global search efficiency to some extent. Summary of the Invention

[0005] The problem that this invention aims to solve is that existing global search methods rely on random generation or general mutation methods, which are not conducive to maintaining search diversity under complex heterogeneous constraints.

[0006] To address the aforementioned problems, in a first aspect, this invention provides a satellite and UAV collaborative mission planning method for Earth observation, comprising: Based on observation mission information and observation resource information, a mission set and a resource center set are constructed. The resource center set includes satellite resource centers and UAV resource centers. The initial allocation is carried out according to the allocation rule selected from the preset allocation rule set. Then, after satellite scheduling verification, reallocation based on the proximity principle and random reset, the first preset number of satellite priority initial individuals are obtained. Each allocation rule is selected less than or equal to 1 time. The proximity principle is the principle of allocating the task to the UAV resource center with the smallest distance from the task. K-means clustering is performed based on the K value selected from the preset set of cluster categories. All tasks in each cluster are initially assigned according to the proximity principle. After drone scheduling verification, reassignment according to the allocation rules, and random reset, a second preset number of drone priority initial individuals are obtained. Each K value is selected less than or equal to 1 time. The initial population is formed by using satellite-priority initial individuals, UAV-priority initial individuals, and random individuals generated according to the random generation principle. Each individual in the initial population is a task collaborative planning scheme. The initial population is iteratively optimized using a genetic algorithm and an improved local optimization algorithm to obtain the optimal solution. The tasks are then allocated to the satellite resource center and the UAV resource center according to the optimal solution.

[0007] Secondly, the present invention also provides a satellite and UAV collaborative mission planning system for Earth observation, comprising: The collection construction module is used to construct task collections and resource center collections based on observation task information and observation resource information. The resource center collections include satellite resource centers and UAV resource centers. The satellite priority individual generation module is used to perform initial allocation according to the allocation rules selected from the preset allocation rule set, and then perform satellite scheduling verification, reallocation based on the proximity principle, and random reset to obtain the first preset number of satellite priority initial individuals. Each allocation rule is selected less than or equal to 1 time. The proximity principle is the principle of allocating the task to the UAV resource center with the smallest distance from the task. The drone priority individual generation module is used to perform K-means clustering based on the K value selected from the preset set of cluster categories, and to initially allocate all tasks in each cluster according to the proximity principle. After drone scheduling verification, reallocation according to the allocation rules and random reset, a second preset number of drone priority initial individuals are obtained. Each K value is selected less than or equal to 1 time. The population completion module is used to form an initial population by using the obtained satellite-priority initial individuals, UAV-priority initial individuals, and random individuals generated according to the random generation principle. Each individual in the initial population is a task collaborative planning scheme. The optimal solution generation module is used to iteratively optimize the initial population using genetic algorithms and improved local optimization algorithms to obtain the optimal solution, and then allocate the task to the satellite resource center and the UAV resource center according to the optimal solution.

[0008] This invention provides a method and system for collaborative mission planning between satellites and unmanned aerial vehicles (UAVs) for Earth observation. Compared with existing technologies, it has the following advantages: By generating satellite-priority and UAV-priority initial individuals, the characteristics of heterogeneous resources can be fully considered, providing a high-quality starting point for subsequent evolution and avoiding the inefficiency of purely random generation. Introducing a phased initial population generation strategy effectively solves the problems of low initial population quality and insufficient search diversity in existing technologies. Furthermore, combining global evolutionary search with local optimization improves the accuracy and convergence speed of the solution, overcoming the shortcomings of insufficient global search effectiveness and local improvement targeting in existing technologies. Through the initial population generation mechanism and the global-local collaborative optimization strategy, the complex challenges of collaborative planning of heterogeneous resources between satellites and UAVs in Earth observation missions can be addressed more effectively, resulting in a better mission planning scheme and improving mission execution efficiency and success rate. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a flowchart illustrating a satellite and UAV collaborative mission planning method for Earth observation, provided as an embodiment of the present invention.

[0011] Figure 2 This is a schematic diagram of the multi-point mutation operation process provided in an embodiment of the present invention.

[0012] Figure 3 This is a schematic diagram illustrating the interaction process between the reinforcement learning module and the local optimization environment provided in an embodiment of the present invention.

[0013] Figure 4 This is a schematic diagram of a task exchange process provided in an embodiment of the present invention.

[0014] Figure 5 This is a schematic diagram of a satellite and UAV collaborative mission planning system for Earth observation, provided as an embodiment of the present invention. Detailed Implementation

[0015] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions in the embodiments of this application are described clearly and completely. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0016] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0017] like Figure 1 As shown in the embodiment of this application, a satellite and UAV collaborative mission planning method for Earth observation is provided, including: S1: Based on the observation mission information and observation resource information, construct the mission set and the resource center set. The resource center set includes satellite resource centers and UAV resource centers.

[0018] Specifically, observation task information may include the task's geographical location, observation time window, execution priority, and required payload type. Observation resource information may include satellite orbital parameters, payload capacity, and visible time window, as well as the UAV's range, payload capacity, and takeoff and landing locations. Based on constraints such as the task's visible time window, resolution requirements, and resource accessibility, a set of optional resource centers is selected for each task, and a one-dimensional integer chromosome encoding is used to represent the allocation relationship between tasks and resource centers. Let the task set be... The resource center is a collection of Then chromosomes can be represented as .in, Indicates task The execution center number is used. By restricting gene values ​​to the resource center set, the genetic operations are matched with heterogeneous resource constraints. This method is designed for static point target Earth observation scenarios, and performs collaborative allocation and scheduling optimization of observation tasks under constraints such as task time window, observation duration, satellite energy and storage downlink, and UAV range and return. In addition, this collaborative planning method is suitable for situations where resource centers are limited and the workload is large, making it difficult for resource centers to execute all tasks simultaneously.

[0019] During the population initialization phase, considering the different characteristics of satellites having wide coverage and UAVs having high mobility, a two-way priority population initialization strategy is designed. Heuristic individuals are constructed that are preferentially allocated to satellites and preferentially allocated to UAVs, respectively, and these individuals, together with random individuals, form the initial population. The specific steps are as follows: S2-S4.

[0020] S2: Initial allocation is performed according to the allocation rules selected from the preset allocation rule set. Then, after satellite scheduling verification, reallocation based on the proximity principle, and random reset, a first preset number of satellite priority initial individuals are obtained. Each allocation rule is selected less than or equal to 1 time. The proximity principle is the principle of allocating the task to the UAV resource center with the smallest distance from the task.

[0021] Specifically, based on a preset ratio and a preset total number of individuals within the population, a first preset quantity and a second preset quantity are determined. This preset ratio can be set according to the preference for satellite and drone resources in the actual application scenario. If there is a greater preference for utilizing satellite resources, a higher satellite priority ratio can be set. For example, when initializing the population, a preset ratio... , The remaining proportion is randomly generated, with a preset total number of individuals. N Calculate the number of individuals corresponding to the two preferred initial values, the first preset number. Second preset quantity And prioritize the initial number of times individual satellites are generated. The initial number of times the drone prioritizes initial individual generation The pre-defined allocation rule set includes various allocation rules such as priority based on the number of visible windows, priority based on time window margin, and priority based on low-attitude maneuver cost. For task scheduling within homogeneous resource centers, existing mature algorithms are employed, such as the improved tabu search algorithm and the adaptive large neighborhood search algorithm, to solve the multi-satellite task scheduling and multi-UAV task scheduling problems, respectively. Random reset refers to the process where, if a task cannot be successfully scheduled after initial allocation, satellite scheduling verification (e.g., the improved tabu search algorithm), and reassignment based on proximity, the task is randomly assigned to any available resource center to ensure that all tasks are attempted for allocation.

[0022] For this step, when generating satellite-priority initial individuals, an allocation rule can be selected from a set of preset allocation rules, such as "prioritizing allocation to satellites with the earliest visible time window." Tasks in the task set will be allocated to satellite resource centers according to this rule. After the initial pre-allocation is completed, satellite scheduling checks are performed on the generated individuals to verify whether each task can be successfully executed on the assigned satellite. For tasks that fail to be scheduled, a proximity principle can be used to reassign them to the nearest UAV resource center, and UAV scheduling checks can be performed again (e.g., using an adaptive large neighborhood search algorithm). If task scheduling still fails, these tasks are randomly reset and randomly allocated to any available resource center. The above process will be repeated until the required first preset number of satellite-priority initial individuals are generated.

[0023] S3: Perform K-means clustering based on the K value selected from the preset set of cluster categories, and initially allocate all tasks in each cluster according to the proximity principle. After drone scheduling verification, reallocation according to the allocation rules and random reset, a second preset number of drone priority initial individuals are obtained, wherein each K value is selected less than or equal to 1 time.

[0024] Specifically, when generating initial priority drone individuals, a K value can be selected from a preset set of clustering categories, for example, K=3, to perform K-means clustering on the task set, dividing the tasks into 3 clusters. Then, all tasks in each cluster are assigned to the drone resource center closest to the cluster center according to the proximity principle. After the initial pre-assignment is completed, drone scheduling is checked on the generated individuals. For tasks that fail to be scheduled, an allocation rule can be selected from a preset set of allocation rules, for example, the rule of "prioritizing allocation to the satellite with the best payload capacity," and the task is reassigned to a satellite resource center, and satellite scheduling is checked again. If there are still tasks that fail to be scheduled, these tasks are randomly reset and randomly assigned to any available resource center. The above process is repeated until the required second preset number of initial priority drone individuals are generated.

[0025] S4: The initial population is formed by using the satellite-priority initial individuals, the UAV-priority initial individuals, and random individuals generated according to the random generation principle. Each individual in the initial population is a task collaborative planning scheme.

[0026] Specifically, the remaining Individuals are generated randomly, meaning that tasks are randomly assigned to UAV resource centers and satellite resource centers. These two priority groups of individuals, along with the randomly generated individuals, form the initial population.

[0027] S5: Using a genetic algorithm and an improved local optimization algorithm, the initial population is iteratively optimized to obtain the optimal solution. The task is then allocated to the satellite resource center and the UAV resource center according to the optimal solution.

[0028] Specifically, a binary tournament selection strategy is used to select parent individuals from the current population. Crossover is performed using a single-point crossover method to recombine parent individuals. In single-point crossover, two parent individuals are randomly selected and swapped at a certain position in their task allocation sequence, generating two new offspring individuals. In the mutation phase, in addition to conventional single-point mutation, a multi-point mutation mechanism is further designed. After generating candidate task allocation solutions through global search, a target individual is selected to enter the local optimization phase. Local optimization can employ various strategies, such as neighborhood search, simulated annealing, or reinforcement learning. In neighborhood search, minor modifications can be made to the target individual's task allocation scheme (e.g., swapping resources between two tasks, moving one task to another resource), and the fitness of the new scheme is evaluated. If the new scheme is better, the modification is accepted. This process continues until a preset termination condition is reached, such as reaching the maximum number of local optimization steps or the fitness no longer improving. After multiple rounds of evolution and local optimization, the optimized population contains a large number of improved solutions. The best task planning scheme that the current algorithm can find is selected from these. This optimal solution will then be used to guide the execution of actual satellite and drone observation missions.

[0029] In this optional embodiment, by introducing a phased initial population generation strategy, the problems of low initial population quality and insufficient search diversity in existing technologies are effectively solved. Specifically, by generating satellite-priority and drone-priority initial individuals, the characteristics of heterogeneous resources can be fully considered, providing a high-quality starting point for subsequent evolution and avoiding the inefficiency of purely random generation. For example, in the disaster observation example, by pre-generating an initial scheme that favors satellites or drones, it can be ensured that the advantages of both platforms are initially utilized in the early stages of the algorithm, rather than relying entirely on random exploration.

[0030] Furthermore, combining global evolutionary search with local optimization improves the accuracy and convergence speed of the solution. This combination overcomes the shortcomings of existing technologies, such as insufficient effectiveness of global search and inadequate targeting of local improvements.

[0031] By employing an initial population generation mechanism and a global-local collaborative optimization strategy, we can more effectively address the complex challenges of collaborative planning of heterogeneous resources between satellites and UAVs in Earth observation missions, thereby obtaining a better mission planning scheme and improving the efficiency and success rate of mission execution.

[0032] The following is a detailed description of each step.

[0033] S1: Based on the observation mission information and observation resource information, construct the mission set and the resource center set. The resource center set includes satellite resource centers and UAV resource centers.

[0034] S2: Initial allocation is performed according to the allocation rules selected from the preset allocation rule set. Then, after satellite scheduling verification, reallocation based on the nearest principle, and random reset, a first preset number of satellite priority initial individuals are obtained. Each allocation rule is selected less than or equal to 1 time. This step specifically includes the following:

[0035] S210: Randomly select an unused allocation rule from the allocation rule set, and allocate the tasks in the task set to the satellite resource center to obtain the satellite priority initial individual.

[0036] Specifically, a mission coordination planning scheme based on the principle of prioritizing mission allocation to the satellite resource center is considered as a satellite-priority initial individual. Allocation rules can include, but are not limited to, allocation logic based on various factors such as mission execution priority, mission observation duration, mission geographical location, and resource availability. Randomly selecting a rule from the available allocation rule set that has not yet been used in the current batch of individual generation aims to increase the diversity of the initial population, prevent all initial individuals from being based on the same allocation logic, thereby broadening the search space and increasing the probability of subsequent evolutionary algorithms finding the global optimum.

[0037] S220: Perform satellite scheduling verification on the initial satellite priority individuals, and reassign the tasks that failed to be scheduled in the initial satellite priority individuals to the UAV resource center according to the principle of proximity, and perform UAV scheduling verification.

[0038] Specifically, scheduling verification refers to performing a feasibility check on assigned tasks to ensure that the tasks can be successfully executed by the designated resources (such as satellites or UAVs) while meeting all constraints (such as time windows, payload capacity, energy limits, etc.). For example, for satellites, satellite scheduling verification checks whether the task falls within the satellite's visible time window and whether the satellite's sensors meet the task's observation requirements. When a task fails to be scheduled at a satellite resource center, it will be attempted to be reassigned to the geographically nearest UAV resource center. UAV resource centers typically possess characteristics such as flexibility, low altitude, and high resolution, and can serve as an effective supplement to tasks that fail to be scheduled by satellites. This principle aims to leverage the local advantages of UAVs to improve the overall scheduling success rate of tasks.

[0039] S230: Randomly reset tasks that still fail to be scheduled, and finally construct a satellite-priority initial individual.

[0040] Specifically, for tasks that fail to be scheduled after being checked by satellite and UAV resource centers, a randomized approach is adopted, assigning these tasks to any resource center. Random reset helps break local deadlocks, introduces new possibilities, and ensures that each initial individual is a complete task allocation scheme, even if some tasks may not be optimally scheduled temporarily.

[0041] S240: Reconstruct the satellite-priority initial individuals until the first preset number of satellite-priority initial individuals are obtained. For example, let... ,like < If the condition is met, return to S210; otherwise, proceed to S3.

[0042] By randomly selecting an unused allocation rule from the set of allocation rules, each generated initial individual is ensured to be based on a different allocation strategy, thus introducing population diversity from the initial stage. Subsequently, tasks in the task set are preferentially allocated to the satellite resource center according to the selected rule, reflecting the priority utilization of the satellite's wide-area observation capabilities. Next, the initially formed satellite-priority initial individuals are tested by satellite scheduling to verify the feasibility of task allocation. For tasks that fail to be scheduled at the satellite resource center, the scheme does not simply abandon them, but introduces the principle of proximity, reassigning them to the UAV resource center and testing them again by UAV scheduling, which fully utilizes the flexibility and supplementary capabilities of UAVs in local areas. If the task still fails to be scheduled, it is processed by random reset to avoid task omission and introduce further exploration. The above process is repeated until a first preset number of satellite-priority initial individuals are generated. This multi-stage, multi-strategy construction approach not only ensures the feasibility of each initial individual, but also effectively avoids the homogenization of the initial population through the random selection of rules and multi-level resource allocation mechanisms. This provides rich and potential starting points for subsequent crossover, mutation, and local optimization, thereby significantly improving the convergence speed and optimization capability of the entire task planning algorithm.

[0043] S3: Perform K-means clustering based on the K value selected from the preset set of cluster categories, and initially allocate all tasks in each cluster according to the proximity principle. Then, after drone scheduling verification, reallocation according to the allocation rules, and random reset, a second preset number of drone priority initial individuals are obtained, and each K value is selected less than or equal to 1 time. This step specifically includes the following:

[0044] S310: Based on the spatial location features of the task, select an unused K value from the preset set of clustering categories to perform K-means clustering and obtain multiple task clusters.

[0045] Specifically, the K-means clustering algorithm is used to group tasks based on their geographical coordinates (such as longitude, latitude, and altitude). By grouping spatially adjacent tasks into a cluster, a more concentrated set of tasks can be provided for resources like UAVs that operate in specific regions. For example, a range of K values ​​can be pre-defined, such as [5, 15]. When generating initial priority individuals for UAVs each time, the system randomly selects an unused K value from this range for clustering to ensure the diversity of the initial population.

[0046] S320: Assign all tasks in each task cluster to the drone resource center that is closest to the task cluster center to obtain the initial drone individuals.

[0047] Specifically, after task clustering, the distance between the geometric center of each task cluster and all UAV resource centers is calculated. Then, all tasks in the entire task cluster are treated as a whole and assigned to the UAV resource center closest to its center. This initial allocation strategy based on the "proximity principle" aims to maximize the local operational efficiency of UAV resources and reduce unnecessary long-distance flights, thus forming a preliminary UAV-led task allocation scheme. For example, Euclidean distance or spherical distance can be calculated to determine the nearest UAV resource center. The distance calculation formula can be as follows: in, Indicates the distance from the cluster center to the th h The distance of the drone R Represents the Earth's radius. and These represent the latitude and longitude of the cluster center, respectively. and They represent the first h The latitude and longitude of the drone's location.

[0048] S330: Perform drone scheduling verification on the initial priority drone individuals, select allocation rules from the allocation rule set, allocate the tasks that failed to be scheduled in the initial priority drone individuals to the satellite resource center, and perform satellite scheduling verification.

[0049] Specifically, scheduling verification simulates the actual scheduling process, checking whether each UAV resource center can successfully execute all tasks assigned to it within its capabilities (e.g., considering the UAV's range, payload, mission time window, flight speed, etc.). Through UAV scheduling verification, tasks that cannot be effectively executed by the current UAV due to resource constraints or time conflicts can be identified. For example, a time window-based scheduling algorithm can be used to detect time conflicts and assess resource occupancy for the task sequences assigned to each UAV.

[0050] For tasks that fail to be scheduled by UAVs, a satellite resource center is introduced for secondary allocation. Since satellite resources typically have wider coverage and more flexible scheduling capabilities, transferring tasks that UAVs cannot handle to satellites can improve the overall task completion rate. A rule is selected from a pre-defined set of allocation rules (e.g., based on task execution priority, satellite availability window, satellite payload capacity, etc.) to attempt to assign the failed tasks to a suitable satellite resource center, and a second satellite scheduling check is performed to confirm whether the satellite can successfully execute the task.

[0051] S340: Randomly reset tasks that still fail to be scheduled, and finally build a drone-priority initial individual.

[0052] Specifically, after two rounds of scheduling checks by drones and satellites, some tasks still failed to be scheduled. These tasks need to be "reset," meaning they are removed from the current allocation scheme, and new, random allocation opportunities are sought for them. This random reset mechanism helps avoid local optima, increases the diversity of initial individuals, and provides a broader search space for subsequent evolutionary algorithms. For example, these failed tasks can be randomly assigned to any available resource center (whether satellite or drone) without immediate scheduling checks, relying instead on subsequent overall scheduling checks and optimization processes to correct them.

[0053] S350: Rebuild the initial drone priority individuals until a second preset number of initial drone priority individuals are obtained. For example, let... ,like < If the initial population is successful, proceed to S310; otherwise, proceed to S4. To ensure the diversity and quantity of the initial population, the process of constructing the initial individuals of the UAVs is repeated. Through multiple independent constructions, a sufficient number of initial individuals with different task allocation and scheduling schemes can be generated, providing rich starting points for subsequent evolutionary algorithms.

[0054] When generating priority initial individuals for drones, this approach fully utilizes the spatial location features of tasks. K-means clustering effectively groups geographically close tasks, and based on proximity, assigns task clusters to the most suitable drone resource centers. This significantly improves the rationality and efficiency of initial allocation, reducing unnecessary drone flight costs. Furthermore, for tasks where drone scheduling fails, satellite resources are introduced for secondary allocation, effectively compensating for the limitations of local drone capabilities and improving the overall task completion rate. For tasks still unschedulable, a random reset mechanism increases the diversity of the initial population, avoiding local optima and providing a broader search space and higher-quality initial solutions for subsequent evolutionary algorithms. This multi-stage, collaborative initial individual generation strategy ensures that the initial population not only has high quality but also good diversity, laying a solid foundation for subsequent optimization processes and improving the convergence speed and final solution quality of the entire collaborative task planning method.

[0055] S4: The initial population is formed by using the satellite-priority initial individuals, the UAV-priority initial individuals, and random individuals generated according to the random generation principle. Each individual in the initial population is a task collaborative planning scheme.

[0056] S5: Using a genetic algorithm and an improved local optimization algorithm, the initial population is iteratively optimized to obtain the optimal solution. Tasks are then allocated to the satellite resource center and the UAV resource center according to the optimal solution. This step specifically includes the following:

[0057] S510: The task coordination planning scheme of individuals in the initial population is evolved by using crossover and mutation operations to obtain the evolved population.

[0058] S511: Use a binary tournament selection strategy to select parent individuals from the current population.

[0059] Specifically, the binary tournament selection strategy selects individuals with higher fitness through competition, thereby passing on the genetic information of these superior individuals to the next generation and guiding the population towards a better outcome. For example, two individuals can be randomly selected from the current population, their fitness values ​​can be compared, and the individual with the better fitness value can be selected as the parent. Alternatively, a tournament size can be set, and a number of individuals of that size can be randomly selected from the population, from which the individual with the best fitness can be selected as the parent.

[0060] S512: Using a single-point crossover method, two task allocation resource centers are exchanged among the parent individuals to form a new allocation scheme, resulting in a recombined parent individual.

[0061] Specifically, the genetic information of two parent individuals is exchanged to generate new offspring individuals, i.e., recombinant parent individuals. This method, by combining gene segments from different parents, helps increase population diversity and explore new combination possibilities in the solution space. For example, a crossover point can be randomly selected at any position in the individual's encoding, and gene segments from two parent individuals can be exchanged at that crossover point to generate two new recombinant individuals.

[0062] S513: If the current iteration number is less than or equal to half of the preset iteration number, perform random single-point mutation on the recombined parent individuals, change the resource center of a task allocation, complete the evolution, and obtain the evolved population.

[0063] S514: If the current iteration count is greater than half of the preset iteration count, perform random multi-point mutation on the recombined parent individuals, randomly change the resource centers of multiple task allocations, complete the evolution, and obtain the evolved population.

[0064] Specifically, such as Figure 2 As shown, the parent individuals are decoded and scheduled; multiple gene loci corresponding to failed scheduling tasks are mutated, and execution centers (i.e., resource centers) are reallocated. This approach can introduce more significant gene changes, further enhancing the population's exploration capabilities, especially in the later stages of the algorithm, helping to escape local optima and perform deeper searches. For example, multiple gene loci in the recombinant parent individuals can be randomly selected, and the values ​​of these loci can be randomly changed respectively.

[0065] By adaptively adjusting the mutation strategy, random single-point mutation is used in the early stages of evolution, effectively promoting the preservation of population diversity and broad exploration of the solution space, preventing the algorithm from getting trapped in local optima too early. In the later stages of evolution, random multi-point mutation is adopted, introducing stronger perturbations that help the algorithm escape potential local optima and perform a deeper, more refined search for existing better solutions, thereby improving the algorithm's global search capability and convergence accuracy. This dynamic balance between exploration and development results in a satellite-UAV collaborative mission planning scheme with higher adaptability and robustness, enabling it to more effectively complete Earth observation tasks.

[0066] S520: Select the task collaboration planning scheme of the target individual within the evolutionary population and perform local optimization until the termination condition is met, thus completing one local optimization.

[0067] In the local improvement stage, existing technologies generally employ fixed neighborhood structures or empirical rules to drive optimization, failing to adequately consider the resource characteristics differences between satellite and UAV platforms. This makes it difficult to achieve accurate optimization of local solutions, further impacting the overall mission planning effectiveness. Local optimization uses simulated annealing as its basic framework, focusing on task exchange, rescheduling, neighborhood evaluation, and operator selection between resource centers to further explore the local potential of the current solution. During local optimization, if an improved solution is obtained, it replaces the original individual in subsequent population updates. In the local optimization process, the reinforcement learning module does not directly generate task allocation schemes but rather adaptively selects operator combinations in the local search. The task exchange and rescheduling verification process generates candidate neighborhood solutions, and the differentiated neighborhood evaluation and update acceptance mechanism measures the quality of candidate solutions and completes local updates. This step, based on the current search state, selects actions for the current local optimization from a preset operator combination and constructs rewards and updates the Q-table based on the evaluation results, thus achieving dynamic decision-making in the local search process. The interaction between the reinforcement learning module and the local optimization environment is illustrated in the following examples. Figure 3 As shown, this step specifically includes the following:

[0068] S521: Select an elite individual within the evolutionary population or randomly select an individual according to a preset probability as the target individual, wherein the elite individual is the individual with the highest fitness within the evolutionary population.

[0069] S522: Using the target individual as the current solution, extract the current state vector of the current solution. .

[0070] Specifically, to ensure that the operator selection aligns with the phased search characteristics of simulated annealing, the state vector is defined as: 1) Temperature stage characteristics Let the current temperature be... The initial temperature is According to the temperature ratio Search phase The discrete values ​​are divided into four levels, as shown in the following formula. 2) Characteristics of the degree of stagnation :set up For stagnation counts that have not produced any positive improvement, then See the formula below.

[0071] 3) Characteristics of the proportion of failed task pool Let the number of tasks in the failed task pool be... The total number of tasks is Define pool percentage , press The discrete approach is divided into three levels. The higher the pool percentage, the more tasks are not effectively placed and the stronger the perturbation and redistribution capabilities are required.

[0072] 4) Characteristics of recent improvement magnitude Let the fitness function be... The previous iteration was at the current solution The candidate solutions generated above are Define the positive improvement amount in the previous round as To avoid the impact of different scales of different examples on state discrimination, a relative improvement margin is introduced. The calculation method is shown in the following formula, where To predetermine the first minimum positive number, we need to avoid a denominator of zero. Finally, we will... Discretize into three levels according to the following formula, where To improve the threshold, this paper takes .

[0073] in, This represents the candidate solution generated in the previous iteration. Corresponding fitness Indicates the current solution The corresponding fitness level.

[0074] S523: Input the current state vector into the reinforcement learning module, and the reinforcement learning module selects the current operator combination from the action set.

[0075] Specifically, the reinforcement learning module selects a current operator combination from a predefined action set based on the current state vector. The action set includes various task swapping and rescheduling verification operations, which are specific means of local optimization. Through learning, the reinforcement learning module can adaptively select the most suitable operator combination for the current state to maximize the expected long-term reward. The action set is defined as a combination of task move-out and move-in operators and task insertion operators, consisting of six actions: ; in, This represents the move-in / move-out rules for random operator tasks. This represents the move-out and move-in rules for the greedy operator tasks. Indicates the insertion rules for random operator tasks. Represents the insertion rules for the greedy operator task. This represents the insertion rule for the operator task based on probability selection.

[0076] Action selection adopts -greedy strategy, based on probability Randomly select actions to explore, based on probability. Choose the action with the highest Q value (i.e., the operator combination): in, Indicates random selection of actions; Represents the current state vector In this case, choose an action v The Q value obtained after execution; Represents the current state vector In this case, the action that maximizes the Q value is chosen. This formula balances the exploratory ability in the early stages of the search and the utilization ability in the later stages.

[0077] For the task movement operator, two types of moveout and movein operators are designed in the task moveout and movein phases. Task moveout is used to release tasks from the failed task pool in the current resource center, and task movein is used to replenish tasks from the failed task pool to the center.

[0078] 1) Random operator task move-out and move-in rules Randomly select tasks to move out or in.

[0079] 2) The greedy operator task shifting rules There are two strategies based on the type of resource center; (1) For satellite resource centers, the mission benefit density is defined as: in, For the task The execution priority level is a numerical value. For example, there are five levels, with numerical values ​​of 1, 2, 3, 4, and 5 for each priority level. Each task is assigned an execution priority level when it is generated or issued. This execution priority level indicates the task's urgency or importance. For the task Observation duration; tasks with low benefit density are moved out first, and tasks with high benefit density are moved in first.

[0080] (2) For the UAV resource center, the definition is based on The local density of nearest neighbor distance is: in, Indicates task Local density, For the current set of tasks assigned to the drone resource center, Represents a set The number of tasks in the middle, These are distance normalization parameters; Indicates task Observation location and mission The distance between observation positions; tasks with low local density are moved out first, and tasks with high local density are moved in first.

[0081] For the satellite resource center, the assessment of mission benefit density ensures that limited satellite observation resources are prioritized for high-value, high-efficiency missions, thereby improving the overall efficiency of mission completion. For the UAV resource center, the assessment of local density based on k-nearest neighbor distance promotes the rational spatial clustering of missions, effectively optimizes UAV flight paths, reduces operating costs, and improves mission execution efficiency. This intelligent mission selection mechanism enables the local optimization process to explore and utilize the solution space more efficiently, avoiding blind operations, and thus significantly improving the optimization efficiency and mission completion quality of satellite-UAV collaborative mission planning schemes.

[0082] For the task insertion operator, during the task insertion phase, it is used to extract from... Select the task and insert it into The three insertion operators are described below.

[0083] 1) Random operator task insertion rules : Randomly select tasks for insertion.

[0084] 2) Greedy operator task insertion rules Based on the type of resource center, there are four types of task insertion scenarios, and different task insertion strategies should be selected for different task insertion scenarios.

[0085] (1) For the scenario of inserting a task from one satellite resource center into another, a time window conflict is used to measure the conflict between the task to be inserted and the specified task set. The degree of conflict between internal tasks, and the time window conflict are as follows: in, This indicates a specified set of tasks. For the inserted task set of the satellite resource center, the satellite resource center For the task The set of visible time windows is Similarly, the Satellite Resource Center For the task The set of visible time windows is , Set of visible time windows The Middle y A time window, Set of visible time windows The Middle y A time window, S Indicates a collection of satellite resource centers, located within the same satellite resource center. The two times for the two different tasks are respectively and , Indicates time window At the beginning of Indicates time window At the end of the day, Indicates time window At the beginning of Indicates time window The end time, combined with the start time Overall End Time ,like If there is overlap, then the task With the task Time window conflict ;otherwise ; For the task The execution priority is determined by the time window conflict. Tasks with fewer time window conflicts are inserted first.

[0086] (2) For the scenario of task insertion from one UAV resource center to another, the average flight insertion estimate is used to estimate the impact of task insertion on the UAV flight path. It should be noted that the candidate tasks are... Insert each candidate path segment into the current closed path of the UAV. The insertion of a candidate task only affects the inserted path and has no effect on other paths. Calculate the resulting path increment and take the average value. The average flight insertion estimate is then: in, This refers to a collection of drones within the drone resource center. For drones Set of scheduled tasks For drones The number of tasks in the already scheduled task set. Indicates task Observation location and mission The distance between observation locations For drones Total flight path distance after initial scheduling For the task The distance between the observation location and the UAV base. Missions with smaller average flight insertion estimates are prioritized for insertion.

[0087] (3) For the scenario of inserting a task from the satellite resource center into a task from the UAV resource center, define the first comprehensive index. It is composed of the time window conflict degree and the average flight insertion estimate. The smaller the value, the greater the conflict between tasks at the satellite resource center. It is a calculation Tasks in With The degree of time window conflict between all other tasks within the system. Simultaneously, to minimize the average flight insertion estimate at the drone resource center, in cross-resource center transfer scenarios from the satellite center to the drone center, priority should be given to... Smaller tasks are inserted into the corresponding drone center. The first comprehensive indicator is: ; in, This represents the time window conflict normalization value. This represents the first preset positive value, at which point the calculation... within At that time, specify the task set The set of missions in the satellite resource center of the original mission pool. , This indicates a time window conflict between a task and other tasks when the task remains within its original task pool. ; , representing the normalized value of the mean flight interpolation estimate. This indicates the second preset positive value. The larger the value, the more severe the conflict at the satellite resource center; This indicates the weight of the average flight interpolation estimate. The first weight represents the time window conflict.

[0088] (4) For the scenario of inserting a task from the UAV resource center into a task from the satellite resource center, in this scenario, tasks with lower task clustering indices in the UAV resource center and smaller conflict values ​​after insertion into the satellite resource center are prioritized for transfer. The second comprehensive index is defined as: ; in, Indicates task Local density, Weights representing local density The second weight represents the time window conflict; this is used in the calculation. within At that time, specify the task set The set of tasks for the satellite resource center of the target task pool. This indicates a time window conflict after the task is inserted into the target task pool. Smaller tasks are inserted first.

[0089] 3) Insertion rules for operator tasks based on probability selection Based on the roulette wheel mechanism, a task is selected from the second successful task pool and inserted into the first successful task pool. The probability of selection is: in, Indicates task The interpolated evaluation value, This indicates the second successful task pool. Indicates task The interpolated evaluation value, The larger the value, the greater the probability of it being removed. It is a greedy operator task insertion rule. The interpolation evaluation value obtained in a given scenario, i.e. ,For example .

[0090] For task insertion within the satellite resource center, time window conflict assessment effectively avoids time overlaps, ensuring scheduling feasibility. For task insertion within the UAV resource center, average flight insertion prediction is used to minimize the impact on UAV flight paths, improving UAV resource utilization efficiency. For task insertion across resource types, first and second comprehensive indicators are used to comprehensively consider the constraints and optimization objectives of different resources, making cross-type task allocation more reasonable. Furthermore, the introduction of task insertion rules based on probabilistic selection operators increases the randomness of the search while ensuring a certain optimization direction, helping to escape local optima and discover globally better solutions. These intelligent insertion strategies enable the local optimization process to generate high-quality candidate solutions more efficiently and accurately, significantly improving the optimization efficiency of satellite-UAV collaborative task planning and the quality of the final scheduling scheme.

[0091] S524: Perform task exchange and rescheduling checks on the target individual based on the selected current operator combination, and generate candidate solutions.

[0092] In local optimization, a task reassignment and rescheduling process based on a failed task pool is used to perturb the current allocation scheme. For example... Figure 4 As shown, this step specifically includes the following:

[0093] S5241: Randomly select two resource centers. The tasks that are successfully scheduled in each resource center form a successful task pool, resulting in the first successful task pool and the second successful task pool. All tasks that fail to be scheduled in the resource centers constitute the failed task pool.

[0094] For example, the first successful task pool With the second successful task pool This refers to the set of tasks that have been successfully scheduled in the resource center. This is the pool of failed tasks. It is randomly determined. and .

[0095] S5242: Select a task from the first successful task pool and move it to the failed task pool.

[0096] Specifically, from Tasks selected with the first preset ratio are moved to the failed task pool. .

[0097] S5243: Select a task from the failed task pool and move it into the first successful task pool, and select a task from the second successful task pool and insert it into the first successful task pool.

[0098] Specifically, on the one hand, from Select the second preset ratio for task movement On the other hand, from Select the third preset ratio to insert the task .

[0099] S5244: Perform a rescheduling check on the first successful task pool (perform the corresponding scheduling check), and put the scheduling failure tasks back into the failure task pool.

[0100] Specifically, for Perform a rescheduling and put the failed tasks back into the failed task pool.

[0101] S5245: Select a task from the failed task pool and move it to the second successful task pool.

[0102] Specifically, from Select the fourth preset ratio for task movement .

[0103] S5246: Perform a rescheduling check on the second successful task pool (perform the corresponding scheduling check), and put the scheduling failure tasks back into the failure task pool.

[0104] Specifically, for Perform a rescheduling and put the failed tasks back into the failed task pool.

[0105] S5247: Randomly assign tasks from the failed task pool to all resource centers to obtain candidate solutions.

[0106] By clearly defining task pool divisions and task flow rules, the reinforcement learning module can more accurately guide task adjustments, avoiding blind or inefficient random operations. Strategically moving tasks out, in, and into the successful and failed task pools, supplemented by rescheduling checks, ensures the effectiveness and feasibility of each task adjustment. This not only improves the efficiency of local optimization, enabling more effective exploration of the solution space, but also significantly enhances the quality of generated candidate solutions, thus providing better input for the subsequent acceptance and update mechanism, ultimately contributing to obtaining a superior satellite-UAV collaborative mission planning scheme.

[0107] S525: Determine whether to accept the candidate solution based on the fitness difference between the fitness of the task collaborative planning scheme represented by the candidate solution and the fitness of the task collaborative planning scheme represented by the current solution.

[0108] Specifically, after candidate solutions are generated, candidate solutions are calculated. With the current solution x Poor adaptability .when When, directly accept candidate solutions; when At that time, to simulate the annealing acceptance probability Accept candidate solutions, where This is the current temperature.

[0109] S526: Construct rewards based on the improvement of candidate solutions relative to the current solution, and update the Q value to guide the selection of operator combinations in the next round of local optimization.

[0110] After completing the generation, rescheduling, and evaluation of neighborhood solutions, a reward function is constructed based on the improvement of candidate solutions relative to the current solution: Update the Q value using the Bellman update formula: in, Represents the current state vector In this case, choose an action Execution, and the Q value obtained at the next moment; Represents the current state vector In this case, choose an action Execution, the Q value obtained at the current moment; This represents the reward at the current moment; Represents the vector of the next state Under the condition that the maximum Q value is obtained after executing the actions in action set A; For learning rate, This is the discount factor.

[0111] The introduction of reinforcement learning modules enables local search strategies to adaptively adjust based on the actual state of the current task planning scheme, thus avoiding the limitations of traditional fixed or random local search strategies. This intelligent local optimization process can more effectively escape local optima, accelerate the convergence speed of the algorithm, and ultimately obtain a higher-quality task planning scheme, ensuring that satellite and UAV resources are utilized more fully and rationally, thereby improving the overall efficiency and effectiveness of the observation mission.

[0112] S530: If the current iteration number is less than the preset iteration number, then perform population evolution and local optimization again until the current iteration number equals the preset iteration number, and obtain the optimized population.

[0113] S540: Within the optimized population, the task collaborative planning scheme represented by the individual with the highest fitness is selected as the optimal solution. The tasks are then allocated to the satellite resource center and the UAV resource center according to the optimal solution. Here, fitness is the sum of the execution priority levels of the successfully scheduled tasks within the task collaborative planning scheme represented by the individual after scheduling verification.

[0114] Specifically, during the evolutionary iteration process, a broad solution space is explored through crossover and mutation operations, while in the local optimization stage, elite individuals are finely adjusted to further improve their fitness. Combining global evolutionary search with local optimization improves the accuracy and convergence speed of the solution.

[0115] Fitness is a metric that measures the quality of an individual (i.e., a solution). In mission planning, fitness is typically related to factors such as the sum of execution priorities of successfully scheduled tasks, resource utilization, and mission completion time. An individual with higher fitness represents a better solution. Selecting the individual with the highest fitness represents the best mission planning solution that the current algorithm can find. This optimal solution will then be used to guide the execution of actual satellite and UAV observation missions.

[0116] In summary, compared with existing technologies, it has the following beneficial effects: 1. By using a two-way priority population initialization and multi-point mutation mechanism, the quality of the initial solution, population diversity, and global search capability are improved.

[0117] 2. By constructing a neighborhood evaluation mechanism that takes into account the differentiated characteristics of satellites and UAVs, the targeting of local search and the feasibility of candidate solutions are improved.

[0118] 3. By introducing reinforcement learning into the operator combination selection process in local search, the adaptive capability of local optimization and the overall solution performance of satellite-UAV collaborative mission planning are improved.

[0119] like Figure 5 As shown in the embodiment of this application, a satellite and UAV collaborative mission planning system for Earth observation is provided, comprising: The collection construction module is used to construct task collections and resource center collections based on observation task information and observation resource information. The resource center collections include satellite resource centers and UAV resource centers.

[0120] The satellite priority individual generation module is used to perform initial allocation according to the allocation rules selected from the preset allocation rule set. After satellite scheduling verification, reallocation based on the proximity principle, and random reset, a first preset number of satellite priority initial individuals are obtained. Each allocation rule is selected less than or equal to 1 time. The proximity principle is the principle of allocating the task to the UAV resource center with the smallest distance from the task.

[0121] The drone priority individual generation module is used to perform K-means clustering based on the K value selected from the preset set of cluster categories, and to initially allocate all tasks in each cluster according to the proximity principle. After drone scheduling verification, reallocation according to the allocation rules and random reset, a second preset number of drone priority initial individuals are obtained, and each K value is selected less than or equal to 1 time.

[0122] The population completion module is used to form an initial population by using the obtained satellite-priority initial individuals, UAV-priority initial individuals, and random individuals generated according to the random generation principle. Each individual in the initial population is a task collaborative planning scheme. The optimal solution generation module is used to iteratively optimize the initial population using genetic algorithms and improved local optimization algorithms to obtain the optimal solution, and then allocate the task to the satellite resource center and the UAV resource center according to the optimal solution.

[0123] Through the above technical solution, this system effectively addresses the problem of insufficient global search effectiveness in existing technologies. In the global search phase, the satellite-priority individual generation module and the UAV-priority individual generation module respectively construct high-quality initial solutions based on the wide-area coverage of satellites and the high maneuverability of UAVs. This fully considers the differences in resource characteristics between satellites and UAVs during the initial population construction phase, effectively improving the diversity and quality of the global search and avoiding the low search efficiency caused by random generation. In the local improvement phase, the iterative mechanism, combined with population evolution and local optimization, enables the system to implement precise optimization based on the differences in resource characteristics between satellites and UAVs. This significantly enhances the targeting of local improvements, achieving the effect of efficiently generating high-quality collaborative mission planning schemes under complex constraints, and significantly improving the quality of mission planning schemes. Ultimately, this system achieves efficient collaboration of heterogeneous resources between satellites and UAVs under complex constraints, meeting the increasingly complex, diverse, and time-sensitive needs of Earth observation missions.

[0124] This application aims to address the problems of insufficient search adaptability, limited local optimization capabilities, and difficulty in balancing overall solution efficiency and solution quality in existing satellite-UAV collaborative Earth observation mission planning under complex constraints. By constructing a reinforcement learning-based algorithm, combined with global evolutionary search, local neighborhood improvement, and a learning-driven operator combination selection mechanism, the solution performance, search efficiency, and feasibility of satellite-UAV collaborative mission planning are improved.

[0125] An electronic device provided in this application includes a memory and a processor; the memory is used to store a computer program; the processor is used to implement the satellite and UAV collaborative mission planning method for Earth observation as described above when executing the computer program.

[0126] This application provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the satellite and UAV collaborative mission planning method for Earth observation as described above.

[0127] In this embodiment, the beneficial effects of the electronic device and the computer-readable storage medium are similar to those of the above-described satellite and UAV collaborative mission planning method for Earth observation, and will not be repeated here.

[0128] The present invention describes electronic devices that can serve as servers or clients of this application, which are examples of hardware devices that can be applied to various aspects of this application. Electronic devices are intended to represent various forms of digital electronic computer devices, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. Electronic devices can also represent various forms of mobile devices, such as personal digital assistant devices, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the application described and / or claimed herein.

[0129] Electronic devices include a computing unit that can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) or loaded from a storage unit into random access memory (RAM). The RAM can also store various programs and data required for device operation. The computing unit, ROM, and RAM are interconnected via a bus. Input / output (I / O) interfaces are also connected to the bus.

[0130] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc. In this application, the separately described modules may or may not be physically separate. Some or all of the modules can be selected to achieve the purpose of the embodiments of this application according to actual needs. Furthermore, the functional modules in the various embodiments of this application can be integrated into one processing unit, or each module can exist physically separately, or two or more modules can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0131] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0132] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A satellite and UAV collaborative mission planning method for Earth observation, characterized in that, include: Based on observation mission information and observation resource information, a mission set and a resource center set are constructed. The resource center set includes satellite resource centers and UAV resource centers. The initial allocation is performed according to the allocation rule selected from the preset allocation rule set. Then, after satellite scheduling verification, reallocation based on the proximity principle, and random reset, a first preset number of satellite priority initial individuals are obtained. Each allocation rule is selected less than or equal to 1 time. The proximity principle is the principle of allocating the task to the UAV resource center with the smallest distance from the task. K-means clustering is performed based on the K value selected from the preset set of cluster categories. All tasks in each cluster are initially assigned according to the proximity principle. After drone scheduling verification, reassignment according to the allocation rules, and random reset, a second preset number of drone priority initial individuals are obtained. Each K value is selected less than or equal to 1 time. The initial population is formed by using satellite-priority initial individuals, UAV-priority initial individuals, and random individuals generated according to the random generation principle. Each individual in the initial population is a task collaborative planning scheme. The initial population is iteratively optimized using a genetic algorithm and an improved local optimization algorithm to obtain the optimal solution. The tasks are then allocated to the satellite resource center and the UAV resource center according to the optimal solution. The process of using a genetic algorithm and an improved local optimization algorithm to iteratively optimize the initial population to obtain the optimal solution, and then allocating tasks to the satellite resource center and the UAV resource center according to the optimal solution, includes: The task collaboration planning schemes of individuals within the initial population are evolved using crossover and mutation operations to obtain an evolved population; Select the task collaboration planning scheme of the target individual within the evolutionary population and perform local optimization until the termination condition is met, thus completing one local optimization. If the current iteration number is less than the preset iteration number, then the population evolution and local optimization are performed again until the current iteration number equals the preset iteration number, and an optimized population is obtained. Within the optimized population, the task collaborative planning scheme represented by the individual with the highest fitness is selected as the optimal solution. The tasks are then assigned to the satellite resource center and the UAV resource center according to the optimal solution. Here, fitness is the sum of the execution priority levels of the successfully scheduled tasks within the task collaborative planning scheme represented by the individual after scheduling verification. The process of selecting a target individual within the evolutionary population and performing local optimization of the task coordination planning scheme until the termination condition is met, completing one local optimization cycle, includes: Select elite individuals within the evolutionary population or randomly select individuals according to a preset probability as target individuals. Elite individuals are those with the highest fitness within the evolutionary population. Take the target individual as the current solution and extract the current state vector of the current solution; The current state vector is input into the reinforcement learning module, which then selects the current operator combination from the action set. Based on the selected current operator combination, perform task exchange and rescheduling checks on the target individual to generate candidate solutions; Based on the fitness difference between the fitness of the task collaborative planning scheme represented by the candidate solution and the fitness of the task collaborative planning scheme represented by the current solution, determine whether to accept the candidate solution; Rewards are constructed based on the improvement of candidate solutions relative to the current solution, and the Q value is updated to guide the selection of operator combinations in the next round of local optimization. The set of actions includes: ； in, This represents the move-in / move-out rules for random operator tasks. This represents the move-out and move-in rules for the greedy operator tasks. This represents the insertion rule for random operator tasks. This represents the insertion rule for the greedy operator task. This represents the insertion rule for the operator task based on probability selection.

2. The satellite and UAV collaborative mission planning method for Earth observation as described in claim 1, characterized in that, The process of initial allocation based on allocation rules selected from a set of preset allocation rules, followed by satellite scheduling verification, reallocation based on proximity principle, and random reset to obtain a first preset number of satellite priority initial individuals includes: Randomly select an unused allocation rule from the allocation rule set and allocate tasks from the task set to the satellite resource center to obtain a satellite-priority initial individual; perform satellite scheduling verification on the satellite-priority initial individual; reassign tasks that fail to be scheduled in the satellite-priority initial individual to the UAV resource center according to the proximity principle and perform UAV scheduling verification; randomly reset tasks that still fail to be scheduled, and finally construct a satellite-priority initial individual; The process of constructing satellite-priority initial individuals continues until the first preset number of satellite-priority initial individuals are obtained.

3. The satellite and UAV collaborative mission planning method for Earth observation as described in claim 1, characterized in that, The process involves K-means clustering based on a K value selected from a preset set of cluster categories, initial allocation of all tasks within each cluster according to proximity, followed by drone scheduling verification, reallocation according to allocation rules, and random reset to obtain a second preset number of priority initial drone individuals, including: Based on the spatial location characteristics of the tasks, K-means clustering is performed by selecting unused K values ​​from a preset set of cluster categories to obtain multiple task clusters. All tasks in each task cluster are assigned to the UAV resource center closest to the cluster center to obtain the UAV priority initial individual. The UAV priority initial individual is then checked for UAV scheduling. An allocation rule is selected from the allocation rule set, and tasks that fail to be scheduled in the UAV priority initial individual are assigned to satellite resource centers and checked for satellite scheduling. Tasks that still fail to be scheduled are randomly reset to complete the process and finally construct a UAV priority initial individual. The initial drone priority individuals are constructed again until a second preset number of initial drone priority individuals are obtained.

4. The satellite and UAV collaborative mission planning method for Earth observation as described in claim 1, characterized in that, The process of using crossover and mutation operations to evolve the task collaboration planning schemes of individuals within the initial population results in an evolved population comprising: A binary tournament selection strategy is used to select parent individuals from the current population; By using a single-point crossover method, two task allocation resource centers are exchanged among the parent individuals to form a new allocation scheme, resulting in recombined parent individuals. If the current iteration number is less than or equal to half of the preset iteration number, random single-point mutation is performed on the recombined parent individuals to change the resource center of a task allocation, complete the evolution, and obtain the evolved population. If the current iteration count is greater than half of the preset iteration count, random multi-point mutation is performed on the recombined parent individuals, and the resource centers of multiple task allocations are randomly changed to complete the evolution and obtain the evolved population.

5. The satellite and UAV collaborative mission planning method for Earth observation as described in claim 1, characterized in that, The step of performing task exchange and rescheduling checks on the target individual based on the selected current operator combination to generate candidate solutions includes: Two resource centers are randomly selected. The tasks that are successfully scheduled in each resource center form a successful task pool, resulting in the first successful task pool and the second successful task pool. All tasks that fail to be scheduled in the resource centers constitute the failed task pool. Select tasks from the first successful task pool and move them to the failed task pool; Select tasks from the failed task pool and move them into the first successful task pool; and select tasks from the second successful task pool and insert them into the first successful task pool. Perform a rescheduling check on the first successful task pool and put the tasks that failed to be scheduled back into the failed task pool; Select tasks from the failed task pool and move them to the second successful task pool; Perform a rescheduling check on the second successful task pool and put the tasks that failed to be scheduled back into the failed task pool. The tasks in the failed task pool are randomly assigned to all resource centers, eventually yielding candidate solutions.

6. The satellite and UAV collaborative mission planning method for Earth observation as described in claim 1, characterized in that, The greedy operator task move-out and move-in rules There are two strategies based on the type of resource center; (1) For satellite resource centers, the mission benefit density is defined as: ； in, For the task The execution priority level is a numerical value; For the task Observation duration required; tasks with low benefit density are prioritized for removal, and tasks with high benefit density are prioritized for inclusion. (2) For the UAV resource center, the definition is based on The local density of nearest neighbor distance is: ； in, Indicates task Local density, For the current set of tasks assigned to the drone resource center, Represents a set The number of tasks in the middle, These are distance normalization parameters; Indicates task Observation location and mission The distance between observation positions; tasks with low local density are moved out first, and tasks with high local density are moved in first.

7. The satellite and UAV collaborative mission planning method for Earth observation as described in claim 1, characterized in that, The greedy operator task insertion rule Based on the type of resource center, there are four types of task insertion scenarios, and different task insertion strategies should be selected for different task insertion scenarios. (1) For the scenario of inserting a task from one satellite resource center into another, a time window conflict is used to measure the conflict between the task to be inserted and the specified task set. The degree of conflict between internal tasks, and the time window conflict are as follows: ； in, This indicates a specified set of tasks. For the inserted task set of the satellite resource center, the satellite resource center For the task The set of visible time windows is Satellite Resource Center For the task The set of visible time windows is , Set of visible time windows The y-th time window, Set of visible time windows In the y-th time window, S represents the set of satellite resource centers, within the same satellite resource center The two times for the two different tasks are respectively and , Indicates time window At the beginning of Indicates time window At the end of the day, Indicates time window At the beginning of Indicates time window The end time, combined with the start time Overall End Time ;like If there is overlap, then the task With the task Time window conflict ;otherwise ; For the task The execution priority level; tasks with less time window conflict are inserted first; (2) For the scenario of task insertion from one UAV resource center to another, the average flight insertion estimate is used to estimate the impact of task insertion on the UAV flight path. The average flight insertion estimate is: ； in, This refers to a collection of drones within a drone resource center. For drones Set of scheduled tasks For drones The number of tasks in the scheduled task set. Indicates task Observation location and mission The distance between observation locations For drones Total flight path distance after initial scheduling For the task The distance between the observation location and the UAV base; missions with smaller average flight insertion estimates are inserted first; (3) For the scenario of inserting a task from the satellite resource center into a task from the UAV resource center, the first comprehensive index is defined as: ； in, This represents the time window conflict normalization value. This represents the first preset positive value, at which point the calculation... within At that time, specify the task set The set of missions in the satellite resource center of the original mission pool. , This indicates a time window conflict between a task and other tasks when the task remains within its original task pool. ; , representing the normalized value of the mean flight interpolation estimate. This indicates the second preset positive value. This indicates the weight of the average flight interpolation estimate. The first weight representing time window conflicts; (4) For the scenario of inserting tasks from the UAV resource center to the satellite resource center, the second comprehensive index is defined as: ； in, Indicates task Local density, Weights representing local density The second weight represents the time window conflict; this is used in the calculation. within At that time, specify the task set The set of tasks for the satellite resource center of the target task pool. This indicates a time window conflict after the task is inserted into the target task pool. Insertion rules based on probability-based selection operator tasks Based on the roulette wheel mechanism, a task is selected from the second successful task pool and inserted into the first successful task pool. The probability of selection is: ； in, Indicates task The interpolated evaluation value, This indicates the second successful task pool. Indicates task The interpolated evaluation value.

8. A satellite and UAV collaborative mission planning system for Earth observation, characterized in that, The system implementing the satellite and UAV collaborative mission planning method for Earth observation as described in any one of claims 1-7, the system comprising: The collection construction module is used to construct task collections and resource center collections based on observation task information and observation resource information. The resource center collections include satellite resource centers and UAV resource centers. The satellite priority individual generation module is used to perform initial allocation according to the allocation rules selected from the preset allocation rule set, and then perform satellite scheduling verification, reallocation based on the proximity principle, and random reset to obtain the first preset number of satellite priority initial individuals. Each allocation rule is selected less than or equal to 1 time. The proximity principle is the principle of allocating the task to the UAV resource center with the smallest distance from the task. The drone priority individual generation module is used to perform K-means clustering based on the K value selected from the preset set of cluster categories, and to initially allocate all tasks in each cluster according to the proximity principle. After drone scheduling verification, reallocation according to the allocation rules and random reset, a second preset number of drone priority initial individuals are obtained. Each K value is selected less than or equal to 1 time. The population completion module is used to form an initial population by using the obtained satellite-priority initial individuals, UAV-priority initial individuals, and random individuals generated according to the random generation principle. Each individual in the initial population is a task collaborative planning scheme. The optimal solution generation module is used to iteratively optimize the initial population using genetic algorithms and improved local optimization algorithms to obtain the optimal solution, and then allocate the task to the satellite resource center and the UAV resource center according to the optimal solution.

Images