An aircraft intelligent task allocation method based on classification-evaluation-matching decision tree

Abstract

This invention discloses an intelligent task allocation method for aircraft based on a classification-evaluation-matching decision tree. The method includes: constructing an aircraft system model and a task model; constructing and training a multi-level aircraft task allocation decision tree system; inputting the aircraft system model and task model into the trained decision tree system to generate a preliminary task allocation scheme; detecting time and fuel resource conflicts in the preliminary scheme and resolving conflicts using priority comparison, time shifting, and task segmentation strategies to form an executable task allocation scheme; and finally outputting the task allocation result after conflict resolution. This invention features high interpretability, comprehensive constraints, and high computational efficiency, effectively improving the completion rate and resource utilization efficiency of multi-aircraft collaborative tasks, and is suitable for intelligent task allocation in complex dynamic environments.

Description

Technical Field

[0001] This invention belongs to the field of aircraft mission planning technology, specifically relating to an intelligent mission allocation method for aircraft based on a classification-evaluation-matching decision tree. Background Technology

[0002] With the rapid development of aerospace technology and the ever-increasing demands of space missions, multi-vehicle collaborative operations have become an important mode of space mission execution. In typical applications such as Earth observation, flight surveillance, and communication relay, multiple aircraft are often required to collaboratively complete diverse tasks under complex constraints. How to achieve efficient task allocation and coordination among multiple aircraft while ensuring mission timeliness and resource utilization has become a core issue in multi-vehicle collaborative mission execution.

[0003] Traditional task allocation relies primarily on human experience, with researchers manually assigning tasks based on priority, aircraft performance, and environmental constraints. However, this method is inefficient and error-prone in large-scale, highly dynamic task scenarios, lacking a systematic constraint verification mechanism and prone to issues such as time conflicts and resource overruns. When the number of tasks expands to dozens or even hundreds, manual allocation becomes time-consuming and struggles to guarantee the global optimality and reliability of the allocation scheme. Furthermore, in the event of unexpected tasks or aircraft malfunctions, the manual method suffers from response delays.

[0004] To address the aforementioned issues, scholars both domestically and internationally have proposed various intelligent optimization methods, such as genetic algorithms, particle swarm optimization, and ant colony optimization. These methods have improved allocation efficiency and optimization quality to some extent, but they generally suffer from high computational complexity, susceptibility to local optima, parameter sensitivity, and slow convergence speed. In recent years, reinforcement learning and deep neural networks have been introduced into the task allocation field, enabling adaptive decision-making through interaction with the environment to learn strategies or through end-to-end modeling. However, these methods have limitations such as long training times, poor model interpretability, and insufficient generalization ability, making it difficult to meet the rapid planning needs in complex dynamic environments.

[0005] Furthermore, existing technologies have proposed aircraft decision-making methods based on hierarchical decision trees, such as the technical solution with publication number CN120494727A. This solution constructs a three-level hierarchical decision tree to achieve compliance verification, feasibility assessment, and mission priority adjustment, and combines dynamic airspace modeling and conflict detection to achieve path planning and obstacle avoidance control. While such methods have certain advantages in flight safety and path planning, they still have the following shortcomings: (1) This method mainly focuses on aircraft path planning and conflict avoidance, lacks a systematic task allocation mechanism for multi-task scenarios, does not perform fine modeling of the matching relationship between tasks and aircraft, and is difficult to achieve global optimal allocation under multi-task conditions. (2) Its decision-making process is mainly based on rule-driven hierarchical judgment, lacking a multi-dimensional quantitative evaluation mechanism for aircraft capabilities, mission requirements and resource constraints, making it difficult to comprehensively consider factors such as fuel consumption, time window and mission benefits, resulting in limited optimization of mission allocation results; (3) Although the method adopts a hierarchical decision-making structure, there is a lack of collaborative design between the layers for the task allocation process. It does not form a closed-loop decision-making process of task classification, capability screening and optimal matching, and it is difficult to balance decision efficiency and allocation accuracy. Therefore, existing technologies still struggle to simultaneously balance computational efficiency, global optimization capabilities, and decision interpretability in multi-aircraft task allocation. There is an urgent need for an intelligent task allocation method that can integrate task classification, capability assessment, and matching optimization to achieve efficient collaboration and optimal scheduling of multiple aircraft under complex constraints. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of the existing technology and provide an intelligent mission allocation method for aircraft based on classification-evaluation-matching decision trees.

[0007] The objective of this invention can be achieved through the following technical solutions: This invention provides an intelligent mission allocation method for aircraft based on a classification-evaluation-matching decision tree, comprising the following steps: Obtain information on tasks to be assigned and aircraft information, and construct aircraft system models and task models; A multi-level aircraft mission allocation decision tree system is constructed, and a training dataset is built. The multi-level aircraft mission allocation decision tree system is trained using the training dataset. The multi-level aircraft mission allocation decision tree system includes a mission priority classification decision tree, an aircraft capability assessment decision tree, and a mission-aircraft matching decision tree. The information from the aircraft system model and mission model is input into the trained multi-level aircraft mission allocation decision tree system to generate a preliminary mission allocation scheme. The generated preliminary task allocation scheme is subjected to conflict detection and resolution to form an executable task allocation scheme; Output the task allocation scheme obtained after conflict resolution.

[0008] Furthermore, the aircraft system model includes the state parameters, performance parameters, and mission payload of each aircraft, wherein: The spacecraft's state parameters include three-dimensional position coordinates, orbital velocity, and remaining fuel. The three-dimensional position coordinates are represented in a geocentric inertial coordinate system and are obtained by calculating the spacecraft's orbital elements, including the semi-major axis, eccentricity, orbital inclination, right ascension of the ascending node, argument of perigee, and true anomaly. The orbital velocity of the spacecraft was calculated using Kepler's third law. The remaining fuel of the aircraft is dynamically updated based on the initial fuel amount and the fuel consumed by the missions already performed. When the remaining fuel is lower than a set threshold, the aircraft will no longer accept new mission assignments. The performance parameters include the aircraft's overall capability and maneuverability coefficients; Task load represents the number of tasks assigned to the aircraft, and there is a maximum task load threshold. When the task load reaches the maximum value, it is prohibited to continue assigning new tasks.

[0009] Furthermore, the task model includes feature parameters for each task to be assigned, wherein: The task characteristic parameters include target location coordinates, task priority, task time window, task duration and task type. The target location coordinates are represented in geocentric inertial coordinate system. For ground target observation tasks, they are calculated by converting geographic coordinates to ECI coordinates. Task Priority The range is [0,1], depending on the urgency of the task. and user needs Comprehensive calculations yielded the following results: in, These are the weighting coefficients. Reflecting the importance of the task, Reflects user attention; Task Time Window Define the earliest start time for the task to be executable. and latest deadline ; Task duration The execution time required to complete this task; Task type This indicates the mission category, including remote sensing imaging and flight surveillance. Different types of missions correspond to different aircraft capability requirements.

[0010] Furthermore, the task priority classification decision tree is used to classify the tasks to be assigned based on their priority, ensuring that high-priority tasks are processed first, wherein: The input features of the task priority classification decision tree include task priority. urgency of the task and the latest deadline for the task ; The node splitting of the task priority classification decision tree adopts the information gain criterion, and the formula for calculating information gain is: in, For the current node's dataset, For the splitting features to be selected, To follow the characteristics The subset after value partitioning and These represent the corresponding sample sizes. For dataset Information entropy; Representation of features The number of different values; features Including task priority urgency of the task and the latest deadline for the task ; The information entropy The formula is: in, The total number of task priority categories. For belonging to the first The sample percentage of tasks with different priority levels; the task priority levels include high priority, medium priority, and low priority. The output of the task priority classification decision tree is the priority category of each task.

[0011] Furthermore, the aircraft capability assessment decision tree is used to screen candidate aircraft that meet the execution conditions for each mission, wherein: The input features of the aircraft capability assessment decision tree include the aircraft's overall capabilities. Remaining fuel Task load and the capability matching degree between the aircraft and the mission The overall capabilities of the aforementioned aircraft It represents the overall capability level of an aircraft in performing a mission. It is calculated by weighting the aircraft's remaining fuel, maneuverability, and mission execution capability. The higher the value, the stronger the aircraft's mission execution capability. The output of the aircraft capability assessment decision tree is a binary classification result, used to determine whether the aircraft is suitable for performing the corresponding task. An output of 1 indicates that the aircraft is suitable for performing the task, and an output of 0 indicates that the aircraft is not suitable for performing the task. Based on the screening results of the aircraft capability assessment decision tree, a candidate aircraft set is constructed for each mission. The candidate aircraft set Includes all aircraft deemed suitable for performing the mission; When the candidate aircraft set If the value is empty, the task is determined to be an unassignable task; The node splitting of the aircraft capability assessment decision tree adopts the Gini impurity criterion, and the formula for calculating Gini impurity is as follows: in, For the current node's dataset, The number of categories includes two types: those suitable for performing the task and those unsuitable for performing the task. For belonging to the first The proportion of samples of different classes, Gini impurity is used to measure the uncertainty of sample classification, when... This indicates that the node samples belong to the same category.

[0012] Furthermore, the overall capabilities of the aforementioned aircraft Calculated using a normalized ability scoring model, it is expressed as: in, , Preset weighting coefficients; Indicates aircraft The remaining fuel; Indicates aircraft Total fuel capacity; For aircraft The mobility coefficient; The mobility coefficient Based on the velocity increment of the spacecraft, the corresponding velocity increment is calculated using the orbital transfer model, and is expressed as follows: in, For aircraft The speed increment required to perform orbital maneuvers The initial orbital radius, For the target orbital radius, It is the gravitational constant; This is the preset maximum available speed increment or the upper limit of the aircraft's capabilities.

[0013] Furthermore, the mission-vehicle matching decision tree is used to select candidate aircraft from the mission priority classification result output by the mission priority classification decision tree. Select the optimal aircraft to perform the mission, where: The tasks are sorted according to the task priority categories output by the task priority classification decision tree, and the tasks are assigned in descending order of priority. The remaining fuel of the aircraft and the task load are updated after each task assignment. For each sorted task, based on the corresponding set of candidate aircraft Input the task-aircraft matching decision tree to select the optimal aircraft; The input features of the mission-vehicle matching decision tree include the capability matching degree between the vehicle and the mission. Distance matching degree Time matching degree Fuel cost matching degree ; The output of the mission-vehicle matching decision tree is the assignment probability. Used to characterize the task Assigned to aircraft The degree of matching; For the candidate aircraft set For each aircraft in the array, calculate its assignment probability, and select the aircraft with the highest assignment probability as the aircraft to execute the current mission, as shown below: in, This indicates the optimal allocation of aircraft. For the task A collection of candidate aircraft; The node splitting of the mission-vehicle matching decision tree adopts the information gain ratio criterion: in, For the current node's dataset, For the splitting features to be selected, Features The inherent value of is the entropy of the distribution of feature values.

[0014] Furthermore, the capability matching degree This indicates the degree to which the aircraft's capabilities meet mission requirements: in, Indicates task A score indicating the requirements for aircraft capabilities; the higher the score, the higher the mission requirements. The distance matching degree Indicates the reachability between the aircraft and the mission objective: in, For aircraft With the task Euclidean distance between targets This is the preset maximum reachable distance; The time matching degree This indicates the degree of match between the aircraft's arrival time and the optimal mission execution time. in, For aircraft Expected arrival time The location time is calculated as the current time plus the flight time between the spacecraft and the mission. The midpoint of the task execution time window; Tasks The earliest start time and the latest end time; Fuel cost matching degree This indicates the impact of the fuel consumption required for an aircraft to perform a mission on its remaining fuel: in, For aircraft Execute the task Fuel consumption required For aircraft Remaining fuel, k This is the fuel consumption coefficient.

[0015] Furthermore, the construction of the training dataset specifically includes: For each training sample, the number of aircraft is randomly generated. and number of tasks Within a reasonable range, it randomly generates parameters such as aircraft position, speed, remaining fuel, and overall capabilities, as well as mission target position, priority, time window, and duration; For each training sample, calculate all The capability matching degree, distance matching degree, time matching degree, and fuel cost matching degree of each aircraft-mission pair are evaluated, and the overall matching degree is calculated based on the weighting coefficients: in, Indicates aircraft With the task Overall matching degree; , , , Preset weighting coefficients; based on overall matching degree Based on constraints, including mission time window, remaining vehicle fuel, and mission uniqueness, an assignment label is generated for each vehicle-mission pair. ,in Indicates a feasible allocation. Indicates an infeasible allocation; All generated training samples are mixed to construct the training dataset.

[0016] Furthermore, the training of the multi-level aircraft mission allocation decision tree system using the training dataset specifically includes: When training a multi-level decision tree, a multi-stage adaptive pruning strategy corresponding to the training sample labels and feature vectors is adopted to optimize the decision tree structure. Specifically, this includes setting minimum sample number constraints and spacecraft mission feasibility constraints during node splitting. Splitting stops when a node fails to meet the constraints. The spacecraft mission feasibility constraints are as follows: in, Assign a feasible proportion to the current node; Assign the number of feasible samples to the current node; This represents the total number of samples at the current node. A preset feasibility threshold is set. After the decision tree is constructed, an improved cost complexity pruning algorithm based on task feasibility is used to optimize the tree structure. The pruning cost function is as follows: in, Here is the pruning cost function; This represents the training error; To determine the number of leaf nodes; Assign a penalty to infeasible entities; This is a weighting factor.

[0017] Compared with the prior art, the present invention has the following advantages: (1) In view of the problems that the existing multi-aircraft task allocation mainly relies on human experience or black box optimization methods, the decision-making process is not transparent, has poor interpretability, and is difficult to adjust according to actual needs. This invention constructs a multi-level decision tree system consisting of a task priority classification decision tree, an aircraft capability assessment decision tree, and a task-aircraft matching decision tree. The task allocation process is decomposed into three levels: classification, screening, and matching, and each step of the decision corresponds to a clear input feature and judgment path. This realizes the structured expression and traceable analysis of the task allocation process, which not only significantly improves the transparency and interpretability of the decision-making process, but also enables task planners to intervene and optimize the decision-making process according to actual needs. This effectively overcomes the problems of the existing black box model being difficult to trust and adjust, and improves the reliability and controllability of multi-aircraft task allocation.

[0018] (2) In view of the problems that existing intelligent optimization methods have in dealing with multi-constraint task allocation problems, such as high computational complexity, easy to get trapped in local optima and difficulty in taking multiple constraints into account at the same time, this invention introduces information gain ratio as a splitting criterion in the decision tree construction process, and combines a multi-stage adaptive pruning strategy with minimum sample number constraint and task feasibility constraint. The time window constraint, fuel constraint and task reachability constraint are integrated into the decision tree structure for unified modeling. While ensuring the generalization ability of the model, the decision complexity is effectively reduced, ineffective splitting and overfitting are avoided, and the generated task allocation rules naturally satisfy multiple constraints, thereby improving the computational efficiency and actual executability of the task allocation scheme.

[0019] (3) In view of the problem that the existing technology lacks a systematic constraint verification mechanism in the task allocation process, which easily leads to time conflicts and resource conflicts, resulting in the infeasibility of the allocation scheme in actual execution, the present invention constructs a conflict detection and resource constraint verification mechanism to identify time conflicts and fuel resource conflicts in the task allocation results, and combines a rescheduling strategy based on task priority to redistribute or adjust the time of conflicting tasks. It can proactively discover and eliminate potential conflicts before task execution, ensure the feasibility of the allocation scheme from the task scheduling level, avoid task failure due to conflicts, and ensure that high-priority tasks are executed first, thereby significantly improving the task completion rate and system resource utilization efficiency. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the overall process of the aircraft mission allocation method of the present invention; Figure 2 This is the training convergence curve of the decision tree model in this invention; Figure 3 This is a graph showing the impact of decision tree depth parameters on task completion rate and training time. Figure 4 A comparison chart of fuel consumption for different decision tree structures in various task scenarios; Figure 5 Example diagram of aircraft mission allocation scenario for the present invention; Figure 6 This is a schematic diagram illustrating the time conflict detection and resolution method of the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0022] Example 1: The aircraft mentioned in this application mainly refer to various types of aircraft, including various types of aircraft or spacecraft.

[0023] This embodiment relates to an intelligent mission allocation method for aircraft based on a classification-evaluation-matching decision tree, such as... Figure 1 As shown, it includes the following steps: S1. Obtain information on tasks to be assigned and aircraft information, and construct aircraft system models and task models, specifically including: S11. Establish the aircraft state model: In an embodiment, the aircraft system includes There are 5 to 30 spacecraft in orbit. The state parameters include position coordinates orbital velocity Remaining fuel and current task load .

[0024] The three-dimensional position coordinates of the aircraft The coordinates are represented using an Earth-Centered Inertial (ECI) coordinate system, with the origin at the Earth's center, the Z-axis pointing to the North Pole, and the X-axis pointing to the vernal equinox. The spacecraft's position coordinates are calculated using orbital elements, including the semi-major axis. a eccentricity e Track inclination i Right ascension of ascending node Perigeal argument And true near point angle For a circular orbiter, the eccentricity e ≈0, the orbital radius is approximately 6371km (Earth's radius) plus the orbital altitude (usually 300-1000km).

[0025] orbital velocity of the spacecraft Calculated according to Kepler's third law: in, The gravitational constant of Earth, This refers to the orbital radius of the spacecraft. For low Earth orbit (LEO) spacecraft, the orbital altitude is approximately 500 km, and the orbital radius is... The calculated orbital speed is approximately 7.6 km / s.

[0026] Remaining fuel of the aircraft The initial fuel load is determined based on the aircraft type and mission design. Generally, medium-sized aircraft carry 300-600 kg of fuel, while large aircraft carry 1000-3000 kg. In this embodiment, it is assumed that the initial fuel load is 500 kg, which is continuously consumed as the mission progresses. When the remaining fuel falls below a safety threshold (set at 50 kg), the aircraft will no longer accept new mission assignments.

[0027] Task load This indicates the number of tasks currently assigned to the aircraft; the initial value is 0. When the aircraft is assigned a new task, Add 1. To prevent overload of a single aircraft, set a maximum mission load threshold. In this embodiment, ,when At that time, the aircraft will no longer accept new missions.

[0028] S12, Task System includes M There are 20-100 tasks to be assigned. j The characteristic parameters include the target position coordinates Priority Time window Duration and task type .

[0029] Target location coordinates of the mission The ECI coordinate system is also used. For ground target observation tasks, the geographic coordinates (longitude, latitude, altitude) are first converted to ECI coordinates. The conversion formula involves the Earth's rotation angle and time, specifically: In the above formula, The rotation matrix is ​​about the Z-axis. Greenwich sidereal hour, For the Earth's radius, For the target height, Latitude Longitude.

[0030] Task Priority The range is [0,1], calculated based on a combination of task urgency and user demand: in, These are the weighting coefficients. Reflecting the importance of the task and user needs It reflects the level of user attention to the task and is obtained from user ratings or historical data statistics.

[0031] Task Time Window The time frame within which the task can be executed is defined. The earliest start time is determined by mission requirements and aircraft visibility constraints. This is the latest deadline; tasks exceeding this time will be invalid. Time window width. This reflects the time flexibility of the task; urgent tasks have narrower time windows (e.g., 1-2 hours), while routine tasks have wider time windows (e.g., 6-24 hours). Task duration. The time required for the aircraft to perform the mission.

[0032] Task type These include remote sensing imaging and flight surveillance. Different types of missions have different requirements for the capabilities of the aircraft.

[0033] S2. Construct a decision tree training dataset, calculate the aircraft-mission matching features, including capability matching, distance matching, time matching, and fuel cost matching, and generate training samples; For each pair of aircraft i and tasks j The process requires calculating matching features across four dimensions, which will serve as input to the decision tree. Calculating these matching features is crucial for task allocation and directly impacts the quality of the allocation.

[0034] S21, Ability Matching Degree The formula for calculating the degree of matching between the aircraft's capabilities and mission requirements is as follows: in For aircraft The capability rating indicates that the higher the value, the stronger the aircraft's capabilities. For the task A score is assigned to the aircraft's capabilities. This score uses a simple product form to ensure that a high score is only achieved when both the aircraft's capabilities and mission requirements are high.

[0035] Distance matching degree The reachability between the aircraft and the mission objective is reflected by the following formula: Among them, aircraft To the mission The distance is: To determine the maximum visible distance, this embodiment takes... The formula uses an exponential decay form, where the matching degree is 1 when the distance is 0, and decreases exponentially as the distance increases.

[0036] Time matching degree reflects the degree of matching between the aircraft's arrival time and the optimal execution time of the mission. The calculation formula is as follows: in For aircraft Expected arrival time The location time is For the task j The optimal execution time is determined by taking the midpoint of the time window as the optimal time. This formula ensures a match of 1 when the aircraft's arrival time is exactly at the midpoint of the time window. The match decreases linearly as the arrival time deviates from the midpoint. Beyond the time window When the matching degree is negative or greater than 1, the spacecraft-mission pair is not feasible.

[0037] Fuel cost matching reflects the fuel cost of performing a task, and the calculation formula is as follows: In the above formula, For aircraft Execute the task Required fuel consumption: S22. Calculate the weighted sum of the matching scores across the four dimensions to obtain the overall matching score: In the formula, these weighting coefficients are determined based on actual application needs and expert experience. Capability matching weights. The highest weight is given because aircraft capability is fundamental to mission completion. Distance and time are weighted equally, both involving reachability constraints. Fuel weight is relatively low, but can be increased when fuel is scarce.

[0038] S23. Generate training samples. The generation of training samples is the basis for training the decision tree model. In this embodiment, the following steps are used to generate 1000 sets of training samples.

[0039] Step 1: Randomize scene parameters. For each sample group, randomly generate the number of aircraft. and number of tasks The aircraft's position, speed, fuel, and capabilities are randomly generated within a reasonable range. Mission parameters such as position, priority, and time window are also randomly generated to ensure that the training data covers all possible scenarios.

[0040] Step 2: Calculate the matching degree matrix. For each scenario, calculate all... Four-dimensional matching degree of each aircraft-mission pair and overall matching degree This forms the matching degree matrix.

[0041] Step 3: Construct training sample labels. Based on the matching degree calculation results and constraints, generate labels for the training samples.

[0042] Step 4: Construct sample feature vectors. For each spacecraft-mission pair Extracting feature vectors There are a total of 6 features. The first 4 are matching features, and the last 2 are basic attributes of the mission and the aircraft. These 6 features comprehensively reflect all the information needed for the allocation decision.

[0043] Step 5: Dataset Splitting. Mix all samples generated from the 1000 scenarios and randomly divide them into training and test sets in an 8:2 ratio. The training set is used for training the decision tree model, and the test set is used to evaluate the model's generalization performance. To avoid overfitting, a validation set can be further created for hyperparameter tuning.

[0044] S3. Construct a multi-level aircraft mission allocation decision tree system and train it based on the training dataset; The core innovation of this invention lies in constructing a three-layer decision tree model, which enables layer-by-layer decision-making from task classification to aircraft selection and then to optimal matching. Compared with a single-layer decision tree, the three-layer structure can better model complex allocation logic, improving decision accuracy and efficiency.

[0045] S31, First-level decision tree – Task priority classification decision tree: The first layer of the decision tree prioritizes tasks, ensuring that high-priority tasks are processed first. Input features include: task priority. urgency of the task Task deadline All three characteristics are related to the time urgency of a task and can effectively distinguish tasks of different priorities.

[0046] The first layer of the decision tree uses information gain as the splitting criterion. Information gain measures the contribution of a feature to the classification of a sample, and is calculated using the following formula: in, For the current node's dataset, For the splitting features to be selected, To follow the characteristics The subset after value partitioning, and These represent the corresponding sample sizes. For dataset Information entropy; Representation of features The number of different values; features Including task priority urgency of the task and the latest deadline for the task ; Information entropy The formula is: in, The total number of task priority categories. For belonging to the first The percentage of samples for tasks with different task priorities; task priority categories include high priority, medium priority, and low priority; the greater the information entropy, the more disordered the samples; the greater the information gain, the greater the improvement in sample purity after splitting according to that feature.

[0047] The output of the task priority classification decision tree is the priority category of each task.

[0048] S32, Second-level decision tree – Aircraft capability assessment decision tree: The second-layer decision tree selects suitable candidate aircraft for each task. This layer acts as a coarse filter, eliminating obviously unsuitable aircraft and reducing subsequent computational load. Input features include: overall aircraft capabilities. Remaining fuel Task load Competency matching .

[0049] The output is a binary classification result: whether the aircraft is suitable for performing the task (1 indicates suitable, 0 indicates unsuitable). After filtering through the second-level decision tree, a set of candidate aircraft will be obtained for each task. The set size is typically 2-5 aircraft. If no candidate aircraft passes the screening for a certain task, it means that the current system cannot complete the task, and the task will be marked as unassignable.

[0050] Before screening for skills, the tasks are first quantitatively evaluated, and task priority is calculated using the following formula: in, This indicates the task priority score; This represents the task urgency coefficient, which reflects the time sensitivity of the task. This represents the width of the task time window. The narrower the time window, the more urgent the task. Therefore, its reciprocal is used to increase the weight of urgent tasks. This represents the target value coefficient, used to reflect the task's benefits or importance; Indicates the task complexity coefficient; These are weighting coefficients and satisfy the normalization constraint.

[0051] In the process of aircraft capability assessment, the overall capability score of the aircraft is first calculated using a regression model: in, , Preset weighting coefficients; Indicates aircraft The remaining fuel; Indicates aircraft Total fuel capacity; For aircraft The mobility coefficient; Maneuverability is characterized by orbital maneuverability. Before executing a mission, an aircraft needs to determine its geometric visibility to the target, which is calculated using the following formula: in, Represents the aircraft's position vector. Represents the target position vector. This indicates the angle between the two. When This indicates that the aircraft is capable of observing the target. This determination method, based on spatial geometry, can quickly assess mission reachability, preventing unfeasible missions from entering the subsequent decision-making process.

[0052] For tasks requiring orbital maneuvers, the velocity increment is calculated using the Hohmann transfer: in, For aircraft The speed increment required to perform orbital maneuvers The initial orbital radius, For the target orbital radius, It is the gravitational constant; This is the preset maximum available speed increment or the upper limit of the aircraft's capabilities.

[0053] Further calculate fuel consumption based on speed increments: in, Indicates fuel consumption; Indicates the current mass of the aircraft; Indicates the specific impulse of the propulsion system; This represents standard gravitational acceleration. Based on rocket propulsion equations, this formula accurately reflects the nonlinear relationship between velocity changes and fuel consumption, thus making capability assessments more closely aligned with actual engineering conditions.

[0054] After completing the ability scoring, a regression tree model is used for ability prediction, with node partitioning based on the minimum squared error criterion: in, This represents the two regions after the division. This represents the predicted value for the corresponding region. This criterion aims to make samples within the same region as similar as possible, thereby improving the accuracy of capability prediction and making capability assessment results more stable.

[0055] During the selection phase, a decision tree is used to output a binary classification result, determining whether the aircraft is suitable for the mission, where "1" indicates suitability and "0" indicates unsuitability. After selection, each mission corresponds to a set of candidate aircraft. Typically, the set contains only a small number of aircraft, thus significantly reducing the search space for subsequent matching decisions. When the set is empty, it indicates that the task is not executable under the current resource conditions, thereby avoiding ineffective scheduling.

[0056] Gini impurity is used as a criterion during node splitting: in, For the current node's dataset, The number of categories includes two types: those suitable for performing the task and those unsuitable for performing the task. For belonging to the first The proportion of samples of different classes, Gini impurity is used to measure the uncertainty of sample classification, when... This indicates that the node samples belong to the same category.

[0057] S33, Third-level decision tree – Mission-aircraft matching decision tree: The third-layer decision tree selects the optimal aircraft from the candidate aircraft set. This layer represents a refined decision, requiring comprehensive consideration of all matching features. The input consists of all seven features: In the formula, the output is the allocation probability. This indicates that the task will be completed. j Assigned to aircraft i The confidence level. Each leaf node of a decision tree contains the proportion of positive and negative samples; this proportion is the assignment probability.

[0058] For candidate set For all aircraft in the system, the third-level decision tree calculates their respective assignment probabilities and then selects the aircraft with the highest probability. The third-level decision tree uses the information gain ratio as the splitting criterion: In the formula, the information gain ratio is a normalization of the information gain, preventing the decision tree from biasing towards features with a large number of values. (Intrinsic value) Reflecting attributes The more values ​​a value has and the more evenly distributed it is, the larger the intrinsic value will be, and the lower the information gain ratio will be accordingly.

[0059] S34, Pre-pruning stage - Implement two conditions to stop splitting during the decision tree construction process: Condition 1: Minimum sample size constraint. When the number of samples at a node... Stop splitting when the number of nodes is less than 20 to avoid over-refinement. For a dataset containing 1000 training samples, stop splitting when the number of nodes is less than 20. This is a basic method to prevent overfitting.

[0060] Condition 2: Mission Feasibility Constraint. This is the innovative point of this invention. For each sample (aircraft-mission pair) within a node, check whether the following condition is met: the aircraft can reach the target within the mission time window. The spacecraft has enough fuel remaining to perform the mission. .

[0061] Only samples that meet both of the above conditions are included. The feasibility ratio is calculated as follows: when If the decision path corresponding to that node is likely to produce an infeasible allocation scheme, then the splitting should be stopped.

[0062] Condition 3: Minimum sample size constraint. When the number of samples at a node... Stop splitting when the number of nodes is less than 20 to avoid over-refinement. For a dataset containing 1000 training samples, stop splitting when the number of nodes is less than 20.

[0063] Post-pruning stage - After the decision tree is built, the pruning cost of each subtree is calculated from bottom to top: For each internal node of the decision tree Calculate its subtree The cost: in: The mean squared prediction error of the subtree. This represents the number of leaf nodes in the subtree. The proportion of samples that cannot be assigned within a node; This is the complexity penalty coefficient, with an initial value of 0.01; This is the feasibility penalty coefficient, with a value of 0.5. An infeasible sample is defined as either violating the time window constraint or having insufficient remaining fuel. Samples that are unreachable by orbit.

[0064] If this node is pruned to a leaf node, the cost of pruning is: The benefits of pruning are: .when Pruning should be performed at the appropriate time. The subtree is preserved. The pruning benefit is calculated layer by layer upwards from the bottom-level node. If a node is pruned to a leaf node, it is removed; otherwise, the subtree is preserved until the root node is reached. Nodes close to 0 A comprehensive evaluation is adopted: if the proportion of feasible solutions increases and the decision-making speed increases by more than 20% after pruning, then pruning is performed, and solutions with high feasibility and high efficiency are given priority.

[0065] Dynamically adjust based on system fuel status When fuel is scarce, increase pruning intensity and prioritize the retention of low-fuel-consumption decision paths. This refers to the remaining fuel percentage in the system when fuel is scarce. <0.3), Increase size, prune more aggressively, and prioritize the preservation of low-fuel-consumption pathways.

[0066] After completing the task allocation model construction and conflict resolution process, the performance of the method was verified through multiple sets of simulation experiments. Experimental results show that the constructed multi-level decision tree model exhibits good convergence characteristics during training, such as... Figure 2 As shown, with the increase of training iterations, the model error gradually decreases and tends to stabilize, indicating that the model can effectively learn the mapping relationship between the aircraft and the task, and has good convergence and stability, thus ensuring the reliability of subsequent task allocation results.

[0067] like Figure 3 As shown, different decision tree depths have a significant impact on task completion rate and training time. When the tree depth is small, the model's expressive power is insufficient, resulting in a low task completion rate. As the depth increases, the task completion rate gradually improves, but the training time also increases accordingly. When the depth exceeds a certain threshold, the improvement in task completion rate tends to plateau, while the training time increases significantly, and overfitting may even occur. Therefore, by reasonably selecting the decision tree depth parameter, a balance can be achieved between task completion rate and computational efficiency, improving the overall system performance.

[0068] Regarding fuel consumption, a comparative analysis was conducted on different decision tree structures, such as... Figure 4As shown, under the same task scenario, the multi-level decision tree structure adopted in this method has a significant advantage in fuel consumption compared to single-level or traditional decision models. This is because the method comprehensively considers multiple factors such as distance, time, and fuel during the matching decision process, making task allocation more reasonable, reducing unnecessary orbital maneuvers, thereby reducing overall fuel consumption and improving resource utilization efficiency.

[0069] In addition, verification was conducted by constructing typical task allocation scenarios, such as Figure 5 As shown, multiple aircraft at different orbital positions collaboratively execute multiple tasks. Through task priority classification, aircraft capability screening, and matching decisions, each task is rationally allocated to different aircraft, and after conflict detection and resolution, an execution plan without time conflicts and satisfying fuel constraints is formed. This example intuitively demonstrates the scheduling capability of this method in complex task environments, verifying its effectiveness and practicality in multi-aircraft collaborative task allocation.

[0070] S4. Input the information of the aircraft system model and mission model into the trained multi-level aircraft mission allocation decision tree system to generate a preliminary mission allocation scheme. In S4, the first-level decision tree is used to classify all tasks by priority and sort them from high to low priority. Figure 6 The task ranking results are: {T1(0.95),T2(0.90),T3(0.75),T4(0.70),T5(0.60),T6(0.55),T7(0.50)}. High-priority tasks (black squares) are processed first to ensure that important tasks receive the best aircraft resources.

[0071] S41. Candidate Aircraft Selection and Allocation: For each task, a second-level decision tree is used to select candidate aircraft, and then a third-level decision tree is used to select the optimal aircraft. Taking task T1 as an example: Screening phase: Among the 5 aircraft, S1 (capacity 0.90, fuel 450 kg), S2 (capacity 0.76, fuel 380 kg) and S3 (capacity 0.84, fuel 420 kg) passed the capacity and fuel constraints and entered the candidate set.

[0072] Selection Phase: Calculate the matching degree between the three candidate aircraft and T1. S1 is closest to T1 (approximately 2.2 km), has the highest distance matching degree, and the best overall matching degree. Therefore, T1 is assigned to S1.

[0073] Status update: S1 remaining fuel updated to 450-22.4=427.6 kg, mission load updated to 1.

[0074] Follow this process to process all tasks in sequence. Figure 6Solid lines, dashed lines, and dotted lines represent the lines connecting high, medium, and low priority tasks, respectively. The final allocation result is as follows: S1 allocates T1 and T5 (solid line + dotted line), S2 allocates T2 (solid line), S3 allocates T3 (dashed line), S4 allocates T4 (dashed line), S5 allocates T7 (dotted line), and T6 is not allocated (dashed circle + X mark).

[0075] After completing the initial allocation based on the decision tree, a multi-objective optimization function is introduced to further optimize the allocation scheme: in, Indicate the optimization objective; Indicates task weight; This indicates the task priority score; Indicates the assignment variable, when the task Assigned to aircraft The value is 1 if the condition is met, otherwise it is 0. Indicates fuel consumption; Indicates the task delay time; Indicates a penalty for infeasible allocation; where represents the weighting coefficients. This optimization function achieves multi-objective collaborative optimization by maximizing task rewards while minimizing fuel consumption and time delay.

[0076] The constraints are: In the formula, For aircraft Available time window For aircraft The remaining fuel, To assign variables, when the task Assigned to aircraft The value is 1 if the condition is met, and 0 otherwise. The total number of tasks. This represents the total number of aircraft.

[0077] By combining decision tree prediction results with multi-objective optimization, global optimality can be further improved while ensuring decision-making efficiency. This ensures that task allocation not only meets the constraints but also achieves optimal performance in terms of resource utilization, task completion rate, and execution efficiency.

[0078] S5. Perform conflict detection and resolution on the generated preliminary task allocation scheme, and use three strategies to resolve conflicts: priority comparison, time shifting and task splitting. S51, Collision Detection Figure 6 As shown, time conflict detection and resolution are performed on the preliminary allocation plan. Figure 6(a) shows the timeline of the initial allocation plan. Missions T1 (time window 10-40 minutes) and T2 (time window 30-60 minutes) of aircraft S1 overlap (marked as conflicting by the slash-filled area) and cannot be executed simultaneously. Missions T5 (time window 20-50 minutes) and T6 (time window 45-70 minutes) of aircraft S3 also conflict.

[0079] Time conflicts are determined by the overlap of task execution time intervals, and the determination function is expressed as follows: in, and They represent missions on the same aircraft. and tasks The execution time interval; when two time intervals overlap, it indicates that the spacecraft needs to perform multiple tasks at the same time, thus causing a time conflict. Otherwise, it is a conflict-free state.

[0080] Fuel resource conflicts are determined by the relationship between the total fuel required for the aircraft to perform its mission and the remaining fuel. The determination function is expressed as follows: in, This represents the set of tasks assigned to the same aircraft; Indicates task i Fuel consumption; This indicates the aircraft's current remaining fuel.

[0081] S52. Conflict resolution employs three strategies to resolve conflicts, with the following results: Figure 6 (b) shows: Strategy 1 (priority comparison): Task T1 (priority 0.95) of S1 conflicts with T2 (priority 0.90), so the higher priority T1 is retained and T2 is reassigned to S2. Figure 6 (b) The executing aircraft for T2 is changed from S1 to S2. Strategy 2 (time shift): The time window for mission T6 of S3 is relatively flexible, so its execution time is shifted to after the end of T5. Figure 6 In (b), the time window for T6 was adjusted from 45-70 to 55-80 to avoid conflict with T5. After resolution, Figure 6 (b) Display all aircraft mission time windows without overlap, forming an executable schedule.

[0082] S6. Output the optimal allocation scheme, including allocation results, task completion rate, and system fuel consumption, among other performance indicators.

[0083] Allocation Results: A total of 6 missions were assigned, with 1 mission unassigned (T6). Details of allocation for each aircraft: Aircraft 1: T1 (fuel 22.4 kg), T5 (fuel 76.2 kg), total consumption 98.6 kg, remaining 351.4 kg. Aircraft 2: T2 (fuel 28.3 kg), 351.7 kg remaining. Aircraft 3: T3 (fuel 28.3 kg), 391.7 kg remaining. Aircraft 4: T4 (fuel 36.1 kg), 313.9 kg remaining. Aircraft 5: T7 (fuel 36.1 kg), 363.9 kg remaining. Performance metrics: Task completion rate: Total system fuel consumption: .

[0084] Example 2: This embodiment provides an intelligent mission allocation system for aircraft based on a classification-evaluation-matching decision tree, including: The data modeling module is used to acquire aircraft and mission information, and to build aircraft system models and mission models. The decision tree construction and training module is used to construct a multi-level aircraft mission allocation decision tree system and train it based on training data. The decision tree system includes a mission priority classification decision tree, an aircraft capability assessment decision tree, and a mission-aircraft matching decision tree. The task allocation module is used to input the aircraft system model and the task model into the trained multi-level decision tree system to generate a preliminary task allocation scheme. The conflict detection and resolution module is used to detect time and resource conflicts in the initial task allocation scheme, adjust conflicting tasks, and generate an executable task allocation scheme.

[0085] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0086] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for intelligent mission allocation for aircraft based on classification-evaluation-matching decision trees, characterized in that, Includes the following steps: Obtain information on tasks to be assigned and aircraft information, and construct aircraft system models and task models; A multi-level aircraft mission allocation decision tree system is constructed, and a training dataset is built. The multi-level aircraft mission allocation decision tree system is trained using the training dataset. The multi-level aircraft mission allocation decision tree system includes a mission priority classification decision tree, an aircraft capability assessment decision tree, and a mission-aircraft matching decision tree. The information from the aircraft system model and mission model is input into the trained multi-level aircraft mission allocation decision tree system to generate a preliminary mission allocation scheme. The generated preliminary task allocation scheme is subjected to conflict detection and resolution to form an executable task allocation scheme; Output the task allocation scheme obtained after conflict resolution.

2. The intelligent mission allocation method for aircraft based on classification-evaluation-matching decision trees according to claim 1, characterized in that, The aircraft system model includes the state parameters, performance parameters, and mission payload of each aircraft, wherein: The spacecraft's state parameters include three-dimensional position coordinates, orbital velocity, and remaining fuel. The three-dimensional position coordinates are represented in a geocentric inertial coordinate system and are obtained by calculating the spacecraft's orbital elements, including the semi-major axis, eccentricity, orbital inclination, right ascension of the ascending node, argument of perigee, and true anomaly. The orbital velocity of the spacecraft was calculated using Kepler's third law. The remaining fuel of the aircraft is dynamically updated based on the initial fuel amount and the fuel consumed by the missions already performed. When the remaining fuel is lower than a set threshold, the aircraft will no longer accept new mission assignments. The performance parameters include the aircraft's overall capability and maneuverability coefficients; Task load represents the number of tasks assigned to the aircraft, and there is a maximum task load threshold. When the task load reaches the maximum value, it is prohibited to continue assigning new tasks.

3. The intelligent mission allocation method for aircraft based on classification-evaluation-matching decision trees according to claim 1, characterized in that, The task model includes feature parameters for each task to be assigned, wherein: The task characteristic parameters include target location coordinates, task priority, task time window, task duration and task type. The target location coordinates are represented in geocentric inertial coordinate system. For ground target observation tasks, they are calculated by converting geographic coordinates to ECI coordinates. Task Priority The range is [0,1], depending on the urgency of the task. and user needs Comprehensive calculations yielded the following results: in, These are the weighting coefficients. Reflecting the importance of the task, Reflects user attention; Task Time Window Define the earliest start time for the task to be executable. and latest deadline ; Task duration The execution time required to complete this task; Task type This indicates the mission category, including remote sensing imaging and flight surveillance. Different types of missions correspond to different aircraft capability requirements.

4. The intelligent mission allocation method for aircraft based on classification-evaluation-matching decision trees according to claim 1, characterized in that, The task priority classification decision tree is used to classify the tasks to be assigned, ensuring that high-priority tasks are processed first, wherein: The input features of the task priority classification decision tree include task priority. urgency of the task and the latest deadline for the task ; The node splitting of the task priority classification decision tree adopts the information gain criterion, and the formula for calculating information gain is: in, For the current node's dataset, For the splitting features to be selected, To follow the characteristics The subset after value partitioning and These represent the corresponding sample sizes. For dataset Information entropy; Representation of features The number of different values; features Including task priority urgency of the task and the latest deadline for the task ; The information entropy The formula is: in, The total number of task priority categories. For belonging to the first The sample percentage of tasks with different priority levels; the task priority levels include high priority, medium priority, and low priority. The output of the task priority classification decision tree is the priority category of each task.

5. The intelligent mission allocation method for aircraft based on classification-evaluation-matching decision trees according to claim 1, characterized in that, The aircraft capability assessment decision tree is used to screen candidate aircraft that meet the execution conditions for each mission, wherein: The input features of the aircraft capability assessment decision tree include the aircraft's overall capabilities. Remaining fuel Task load and the capability matching degree between the aircraft and the mission The overall capabilities of the aforementioned aircraft It represents the overall capability level of an aircraft in performing a mission. It is calculated by weighting the aircraft's remaining fuel, maneuverability, and mission execution capability. The higher the value, the stronger the aircraft's mission execution capability. The output of the aircraft capability assessment decision tree is a binary classification result, used to determine whether the aircraft is suitable for performing the corresponding task. An output of 1 indicates that the aircraft is suitable for performing the task, and an output of 0 indicates that the aircraft is not suitable for performing the task. Based on the screening results of the aircraft capability assessment decision tree, a candidate aircraft set is constructed for each mission. The candidate aircraft set Includes all aircraft deemed suitable for performing the mission; When the candidate aircraft set If the value is empty, the task is determined to be an unassignable task; The node splitting of the aircraft capability assessment decision tree adopts the Gini impurity criterion, and the formula for calculating Gini impurity is as follows: in, For the current node's dataset, The number of categories includes two types: those suitable for performing the task and those unsuitable for performing the task. For belonging to the first The proportion of samples of different classes, Gini impurity is used to measure the uncertainty of sample classification, when... This indicates that the node samples belong to the same category.

6. The intelligent mission allocation method for aircraft based on classification-evaluation-matching decision trees according to claim 5, characterized in that, The comprehensive capabilities of the aircraft Calculated using a normalized ability scoring model, it is expressed as: in, , Preset weighting coefficients; Indicates aircraft The remaining fuel; Indicates aircraft Total fuel capacity; For aircraft The mobility coefficient; The mobility coefficient Based on the velocity increment of the spacecraft, the corresponding velocity increment is calculated using the orbital transfer model, and is expressed as follows: in, For aircraft The speed increment required to perform orbital maneuvers The initial orbital radius, For the target orbital radius, It is the gravitational constant; This is the preset maximum available speed increment or the upper limit of the aircraft's capabilities.

7. The intelligent mission allocation method for aircraft based on classification-evaluation-matching decision trees according to claim 1, characterized in that, The mission-vehicle matching decision tree is used to select candidate aircraft from the mission priority classification result output by the mission priority classification decision tree. Select the optimal aircraft to perform the mission, where: The tasks are sorted according to the task priority categories output by the task priority classification decision tree, and the tasks are assigned in descending order of priority. The remaining fuel of the aircraft and the task load are updated after each task assignment. For each sorted task, based on the corresponding set of candidate aircraft Input the task-aircraft matching decision tree to select the optimal aircraft; The input features of the mission-vehicle matching decision tree include the capability matching degree between the vehicle and the mission. Distance matching degree Time matching degree Fuel cost matching degree ; The output of the mission-vehicle matching decision tree is the assignment probability. Used to characterize the task Assigned to aircraft The degree of matching; For the candidate aircraft set For each aircraft in the array, calculate its assignment probability, and select the aircraft with the highest assignment probability as the aircraft to execute the current mission, as shown below: in, This indicates the optimal allocation of aircraft. For the task A collection of candidate aircraft; The node splitting of the mission-vehicle matching decision tree adopts the information gain ratio criterion: in, For the current node's dataset, For the splitting features to be selected, Features The inherent value of is the entropy of the distribution of feature values.

8. The intelligent mission allocation method for aircraft based on classification-evaluation-matching decision trees according to claim 7, characterized in that, The capability matching degree This indicates the degree to which the aircraft's capabilities meet mission requirements: in, Indicates task A score indicating the requirements for aircraft capabilities; the higher the score, the higher the mission requirements. The distance matching degree Indicates the reachability between the aircraft and the mission objective: in, For aircraft With the task Euclidean distance between targets This is the preset maximum reachable distance; The time matching degree This indicates the degree of match between the aircraft's arrival time and the optimal mission execution time. in, For aircraft Expected arrival time The location time is calculated as the current time plus the flight time between the spacecraft and the mission. The midpoint of the task execution time window; Tasks The earliest start time and the latest end time; Fuel cost matching degree This indicates the impact of the fuel consumption required for an aircraft to perform a mission on its remaining fuel: in, For aircraft Execute the task Fuel consumption required For aircraft Remaining fuel, k This is the fuel consumption coefficient.

9. The intelligent mission allocation method for aircraft based on classification-evaluation-matching decision trees according to claim 1, characterized in that, The construction of the training dataset specifically includes: For each training sample, the number of aircraft is randomly generated. and number of tasks Within a reasonable range, it randomly generates parameters such as aircraft position, speed, remaining fuel, and overall capabilities, as well as mission target position, priority, time window, and duration; For each training sample, calculate all The capability matching degree, distance matching degree, time matching degree, and fuel cost matching degree of each aircraft-mission pair are evaluated, and the overall matching degree is calculated based on the weighting coefficients: in, Indicates aircraft With the task Overall matching degree; , , , Preset weighting coefficients; based on overall matching degree Based on constraints, including mission time window, remaining vehicle fuel, and mission uniqueness, an assignment label is generated for each vehicle-mission pair. ,in Indicates a feasible allocation. Indicates an infeasible allocation; All generated training samples are mixed to construct the training dataset.

10. The intelligent mission allocation method for aircraft based on classification-evaluation-matching decision trees according to claim 1, characterized in that, The training of the multi-level aircraft mission allocation decision tree system using a training dataset specifically includes: When training a multi-level decision tree, a multi-stage adaptive pruning strategy corresponding to the training sample labels and feature vectors is adopted to optimize the decision tree structure. Specifically, this includes setting minimum sample number constraints and spacecraft mission feasibility constraints during node splitting. Splitting stops when a node fails to meet the constraints. The spacecraft mission feasibility constraints are as follows: in, Assign a feasible proportion to the current node; Assign the number of feasible samples to the current node; This represents the total number of samples at the current node. A preset feasibility threshold is set. After the decision tree is constructed, an improved cost complexity pruning algorithm based on task feasibility is used to optimize the tree structure. The pruning cost function is as follows: in, Here is the pruning cost function; This represents the training error; To determine the number of leaf nodes; Assign a penalty to infeasible entities; This is a weighting factor.

Images