A method, device, and medium for unmanned aerial vehicle conflict resolution

Abstract

Embodiments of the present application disclose a UAV conflict resolution method, device and medium, and relate to the technical field of UAV traffic control. The method comprises: obtaining a target conflict event to be occurred between UAVs; performing weighted summation on a plurality of conflict indicators of the target conflict event to evaluate a conflict level of the target conflict event, wherein the plurality of conflict indicators comprise at least one of a conflict distance, a conflict duration, a flight speed, airspace complexity and a task priority; performing conflict resolution according to the conflict level, and inputting full parameter features in the conflict resolution into an online decision tree model to evaluate a resolution effect, wherein the full parameter features comprise the conflict level, the conflict indicators and conflict resolution action parameters; and when the resolution effect is lower than a set level, using a SHAP model to locate reasons for poor resolution effect. The embodiments can optimize the conflict resolution effect of the UAV.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) traffic control technology, and more particularly to a UAV conflict resolution method, device, and medium. Background Technology

[0002] With the rapid development of drone technology, drone swarm applications are becoming increasingly widespread. Whether in logistics delivery, power line inspection, environmental monitoring, airport coordination, or emergency rescue, multiple drones are required to operate collaboratively. However, when multiple drones fly in the same airspace, due to factors such as intersecting flight paths, speed differences, environmental interference, and mission scheduling conflicts, mid-air collisions are highly likely to occur. This can not only damage drone equipment and interrupt missions but also potentially lead to safety accidents. Therefore, finding a way to resolve conflicts within drone swarms is crucial.

[0003] Existing solutions for resolving multi-drone conflicts suffer from at least one of the following problems, making it difficult to meet the practical application requirements of drone swarms: 1. After detecting multi-machine conflicts, the severity of the conflicts was not differentiated based on the conflict information, resulting in all conflicts being resolved using the same strategy, which failed to achieve precise conflict management.

[0004] 2. The conflict resolution strategy is fixed and does not take into account dynamic information such as the real-time status of the UAV, mission requirements, and airspace environment to formulate differentiated conflict resolution strategies, thus failing to achieve dynamic and personalized conflict resolution scheduling.

[0005] 3. The lack of attention to the effectiveness of conflict resolution and the reasons for poor resolution results makes it impossible to provide data support for optimizing conflict resolution methods. Summary of the Invention

[0006] This invention provides a method, apparatus, and medium for resolving drone conflicts, to address at least one of the aforementioned problems.

[0007] In a first aspect, embodiments of the present invention provide a method for resolving conflicts between unmanned aerial vehicles (UAVs), comprising: Acquire potential target conflict events between drones; The conflict level of the target conflict event is evaluated by weighted summation of multiple conflict indicators, wherein the multiple conflict indicators include at least one of conflict distance, conflict duration, flight speed, airspace complexity, and mission priority. Conflict resolution is performed based on the conflict level, and the full-parameter features of the conflict resolution are input into an online decision tree model to evaluate the resolution effect. The full-parameter features include the conflict level, various conflict indicators, and conflict resolution action parameters. When the relief effect is lower than the set level, the SHAP model is used to locate the cause of the poor relief effect.

[0008] In a second aspect, embodiments of the present invention provide an electronic device, the electronic device comprising: One or more processors; Memory, used to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the drone conflict resolution method described in any embodiment.

[0009] Thirdly, embodiments of the present invention also provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the UAV conflict resolution method described in any embodiment.

[0010] In summary, this embodiment provides a method for resolving drone conflicts, which can achieve the following beneficial effects: 1. The method in this embodiment adopts a multi-dimensional comprehensive scoring method, which combines multiple indicators such as conflict distance, duration, flight speed, airspace complexity, and mission priority to classify conflicts into three levels: minor, moderate, and severe. This method can accurately distinguish the severity of conflicts and formulate differentiated resolution strategies based on the conflict level, thereby achieving dynamic and personalized resolution scheduling.

[0011] 2. The method in this embodiment evaluates the resolution effect of conflict events using a decision tree model and combines it with a SHAP model to perform root cause analysis on conflict events where the resolution effect is not satisfactory, thereby identifying the parameters affecting the resolution effect. By constructing full-parameter features including those related to conflict level and resolution actions, the main factors that may affect the resolution effect are considered, providing effective data support for the optimization of conflict resolution methods. Attached Figure Description

[0012] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0013] Figure 1 This is a flowchart of a method for resolving drone conflicts provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0014] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0015] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0016] In the description of this invention, it should also be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0017] Figure 1 This is a flowchart of a UAV conflict resolution method provided in an embodiment of the present invention. This method is applicable to situations where a multi-UAV conflict has been detected and resolution is required, and is executed by an electronic device. This electronic device can be a UAV swarm cooperative control system, onboard equipment within the UAV, or other electronic devices; this embodiment does not impose specific limitations. Figure 1 As shown, the method specifically includes: S110: Acquire potential target conflict events between drones.

[0018] If two drones continue flying according to their current flight status and a spatial conflict is expected to occur within a future time window, then a conflict event between the two drones is considered imminent. This embodiment first acquires the detected conflict events as the data source for the entire method.

[0019] In response to the detected conflict event, this embodiment will implement a series of operations to control and intervene in the drones involved in the conflict event, preventing them from actually engaging in spatial conflict until the conflict is resolved. For ease of distinction and description, this embodiment refers to the conflict event to be intervened in during subsequent operations as the target conflict event.

[0020] In one specific implementation, conflict events can be detected in the following manner: First, the trajectory of each UAV in the current airspace is predicted within a future time window. Optionally, the duration of the time window can be between 10 seconds and 300 seconds.

[0021] Then, this time window is divided into multiple time slices. If, based on the predicted trajectory, drones i and k will appear in a future time slice t... s Within, a horizontal distance D appears. h,ik (t s ) <D h,safe And D v,ik (t s ) <D v,safe In this case, it can be determined that drone i and drone k will occupy time slice t. s A conflict occurs internally; record conflict event C. ik , where D h,ik (t s ) and D v,ik (t s ) represent the time slices of drones i and k, respectively. s Horizontal and vertical distances, D h,safe and D v,safe These represent the safety thresholds for the horizontal and vertical distances of the drone, respectively.

[0022] Meanwhile, regarding conflict event C ik It can also record the time slice t of the conflict. s The drones of the two conflicting parties are numbered i and k, and the coordinates of the conflict point are (which can be taken from the time slice t of the two drones). s (midpoint of the predicted location), the speed V of the two drones at the time of the collision. i V k The flight attitude and other parameters are used to create a conflict event information table.

[0023] Furthermore, conflict events can be filtered based on their duration. If drones i and k detect conflicts in multiple consecutive time slices (e.g., within ≥3 time slices), then conflict event C can be classified as such. ik Marked as persistent conflict C cont Ultimately, only those marked with C will be included. cont Add the conflict event to the conflict event information table.

[0024] Optionally, there can be one or more conflict events in the conflict event information table. Each conflict event in the table can be used as a target conflict event and then handled separately for subsequent conflict resolution.

[0025] S120. The conflict level of the target conflict event is evaluated by weighted summation of multiple conflict indicators, wherein the multiple conflict indicators include at least one of conflict distance, conflict duration, flight speed, airspace complexity and mission priority.

[0026] This step combines multiple factors such as the real-time status of the drones involved in the conflict and mission requirements to classify the conflict, providing a basis for the formulation of subsequent conflict resolution strategies.

[0027] In one specific implementation, a multi-dimensional comprehensive evaluation method can be used to classify the conflict levels of each target conflict event. The classification indicators may include at least one of the following: conflict distance, conflict duration, flight speed of both sides, airspace complexity at the conflict point, and mission priority of both sides. In this embodiment, these classification indicators are also referred to as conflict indicators. Taking any target conflict event as an example, the assessment of the conflict level may include the following steps: Step 1: Quantify the various conflict indicators of the target conflict event. Optionally, for quantitative indicators such as conflict distance, conflict duration, and the flight speeds of the conflicting parties, distance, duration, and speed values ​​can be used for quantification; for qualitative indicators such as the airspace complexity of the conflict point and the mission priorities of the conflicting parties, a hierarchical scoring method can be used for quantification. To ensure that the quantification of each conflict indicator maintains a consistent order of magnitude, this embodiment provides the following quantification method: The collision distance can be quantified by the ratio of the horizontal and vertical collision distances to a safety threshold. Taking the target collision event C as an example... ik For example, the quantization of conflict distance S dist as follows: S dist =10-[(D h,ik / D h,safe )×5+(D v,ik / D v,safe )×5](1) Among them, D h,ik and D v,ik D respectively h,ik (t s ) and D v,ik (t s The abbreviations for ) represent target conflict events C. ik The horizontal and vertical collision distances of the drones i and k involved (here, for D)h,ik and D v,ik The meaning of S is limited to this formula. dist The value range is 0~10, and the conflict distance D h,ik D v,ik The smaller, S dist The higher the level, the more severe the conflict.

[0028] Regarding the duration of the conflict, quantification is performed based on the conflict duration and its maximum threshold, resulting in the following quantification representation S. time : S time= min((t conflict / T max )×10, 10)(2) Among them, t conflict The duration of conflict for the target conflict event, T max This is the maximum threshold for conflict duration (default 30s), where min indicates the minimum value. time The value range is 0~10, and the duration of the conflict is t. conflict The longer it goes on, the more serious the conflict becomes.

[0029] Regarding flight speed, based on the average speed V of the two conflicting parties... avg Quantization yields the following quantized representation S. speed : S speed =min((V avg / V max )×10, 10)(3) Among them, the target conflict event C ik For example, V avg =(V i +V k ) / 2; V max This indicates the maximum flight speed of the drone. S speed The value range is also 0~10. The higher the speed, the higher the risk of collision and the more serious the conflict.

[0030] Regarding airspace complexity, it can be graded and scored according to the complexity of the airspace where the conflict point is located, with a score range of 0 to 10. The resulting score is the quantified representation S. airspaceOptionally, airspace complexity can be scored based on the distance from the conflict point to the no-fly zone. For example: if the distance from the conflict point to the no-fly zone boundary is ≥8km, the airspace complexity score is 0; if the distance is in the range [4km, 8km), the airspace complexity score is 3; if the distance is in the range [2km, 4km), the airspace complexity score is 6; if the distance is in the range [0.5km, 2km), the airspace complexity score is 8; if the conflict point enters the no-fly zone, the airspace complexity score is 10, indicating extremely high risk. The more complex the airspace, the more severe the conflict.

[0031] Regarding task priorities, a score can be assigned based on the average priority of the conflicting tasks, ranging from 0 to 10. The resulting score is the quantitative representation S. task_level Optionally, the highest priority is scored as 10 points, the lowest priority as 0 points, and the scores for other priorities are calculated using a linear interpolation method. Higher task priority results in greater losses from conflicts, and higher scores indicate more severe conflicts.

[0032] Step 2: Determine the overall conflict level score based on the quantitative representation of each conflict indicator. Optionally, the overall conflict level score S can be calculated by weighted summation: (4) in, and They represent the first j Quantitative representation and weighting of conflict indicators, =1. With all five conflict indicators included in the conflict level assessment, j When =1,2,3,4,5, S dist , S time , S speed ,S speed , S airspace , S task_level Optionally, the initial weights are 0.4, 0.2, 0.15, 0.1, and 0.15.

[0033] Step 3: Determine the conflict level of the target conflict event based on the comprehensive conflict level score. Optionally, the conflict can be divided into three levels based on the comprehensive conflict level score S: Minor conflict: S∈[3,6), where the conflict distance is relatively far, the duration is short, and the risk of collision is low; Medium-level conflict: S∈[6,8), where the conflict distance is relatively close, the duration is relatively long, and there is a certain risk of collision. Severe conflict: S∈[8,10], where the conflict distance is extremely close, the duration is long, and the risk of collision is extremely high, and a resolution operation must be performed immediately afterward.

[0034] S130. Perform conflict resolution according to the conflict level, and input the full-parameter features of the conflict resolution into the online decision tree model to evaluate the resolution effect. The full-parameter features include the conflict level, each conflict index, and conflict resolution action parameters.

[0035] This step first develops and implements different resolution strategies for different conflict levels to achieve differentiated conflict resolution. Optionally, for the target conflict event: in the case of a minor conflict, adjust the altitude and flight speed of the conflict drone; in the case of a medium conflict, adjust the altitude, flight speed, and flight direction of the conflict drone; in the case of a severe conflict, adjust the altitude, flight speed, and flight direction of the conflict drone; wherein the adjustment range for each conflict level satisfies the following order: severe conflict > medium conflict > minor conflict.

[0036] In one specific implementation, the release action under minor conflict can be further subdivided into the following two cases: Scenario 1: In a target conflict incident, the two drones have different priorities. In this case, the drone with higher priority is designated as U. high Low-priority drones are denoted as U. low In practical applications, the mission priority in the UAV flight plan can be used as the priority here. Under different priorities, the release action can be: control U... low Perform minor maneuvers without altering the flight path, adjusting only speed and altitude. Specific parameters: speed adjusted to 0.85-0.9 times the original speed, altitude adjusted to 50-80 meters above the original altitude, adjustment duration 10-15 seconds, ensuring conflict resolution; after disengagement, U... low Automatically restore the original speed and altitude, and continue executing the original flight plan; Scenario 2: Two drones in a target conflict have the same priority. The resolving action in this case can be: control both drones to simultaneously perform small maneuvers, adjusting their altitude and speed respectively. One drone adjusts its speed to 0.9 times its original speed and altitude to +50m, while the other adjusts its speed to 1.05 times its original speed and altitude to -50m, avoiding altitude overlap. The adjustment should last 10-15 seconds. After the conflict is resolved, return to the original state.

[0037] Similarly, the actions to resolve moderate conflict can be further divided into the following two categories: Scenario 1: In a target conflict incident, the two drones have different priorities. In this case, the drone with the higher priority is still designated as U. high Low-priority drones are denoted as U. low The corresponding release action can be: controlling U lowPerform a moderate maneuver, adjusting flight speed, altitude, and flight direction. Specific parameters: speed adjusted to 0.7-0.8 times the original speed, altitude adjusted to 80-120m above the original altitude, and flight direction deviating from the original trajectory by 15°-30° (avoiding the flight paths of high-priority UAVs and no-fly zones). Adjustment duration is 20-30 seconds to ensure conflict resolution; after resolution, U... low Based on the current location, the shortest path is replanned, and the original flight plan is resumed.

[0038] Scenario 2: Two drones in a target conflict have the same priority. The resolving action in this case can be as follows: Control both drones to perform moderate maneuvers, adjusting their flight direction and altitude. One drone deviates from its original trajectory by 20°~30° (clockwise) and increases its altitude by 100m, while the other drone deviates from its original trajectory by 20°~30° (counter-clockwise) and decreases its altitude by 100m. Both drones adjust their speed to 0.8 times their original speed. This adjustment lasts for 20~30 seconds. After the conflict is resolved, both drones return to their original flight plans.

[0039] Similarly, the actions to resolve severe conflict can be further divided into the following two situations: Scenario 1: In a target conflict incident, the two drones have different priorities. In this case, the drone with the higher priority is still designated as U. high Low-priority drones are denoted as U. low The corresponding release action can be: controlling U low Perform an emergency maneuver, prioritizing rapid disengagement from the conflict zone. Specific parameters: immediately adjust speed to 0.5-0.6 times the original speed (deceleration), immediately increase altitude by 120-150m, and deviate the flight path by 30°-45° (prioritizing open airspace without other drones for deviation), maintaining this adjustment for 30-40 seconds while simultaneously monitoring the conflict resolution status. If the conflict is not resolved within 30 seconds, further adjust altitude and direction until the conflict is resolved. After disengagement, U... low The original flight plan was suspended, the route was replanned to avoid the conflict area, and then the original flight plan was resumed.

[0040] Scenario 2: Two drones in a target conflict have the same priority. The reconciliation maneuver in this case can be as follows: Control both drones to perform emergency maneuvers simultaneously. One drone immediately decelerates to 0.5 times its original speed, increases altitude by 150m, and deviates 45° (clockwise). The other drone immediately accelerates to 1.1 times its original speed, decreases altitude by 150m, and deviates 45° (counter-clockwise), quickly creating distance. This adjustment should last 30-40 seconds, with real-time monitoring of the conflict resolution. Only after ensuring complete conflict resolution should the drones return to their original flight plan.

[0041] The control parameters in the aforementioned resolving maneuver are the conflict resolution parameters, including flight speed, altitude, azimuth, and adjustment duration. Furthermore, during the execution of the resolving maneuver, the UAV's status information (speed, altitude, position, attitude) and conflict resolution status can be collected in real time. Based on the collected information, the resolving maneuver parameters can be dynamically adjusted, and constraint controls can be applied to prevent the resolving maneuver from causing new problems.

[0042] The dynamic adjustment of the release maneuver parameters may include: if the distance between the conflicting parties continues to decrease during the adjustment process (without improving the conflict situation), then immediately increase the magnitude of the release maneuver (such as further increasing the altitude or increasing the directional deviation angle); if the conflict distance increases rapidly during the adjustment process (the conflict has been resolved), then stop the maneuver in advance, restore the original flight state, and reduce the impact on the mission.

[0043] Constraint control can include: Speed ​​constraints: The adjusted speed must not exceed the maximum and minimum speed of the drone to avoid equipment failure due to excessive speed or mission failure due to excessively slow speed. Altitude constraints: The adjusted altitude must not exceed the maximum service ceiling of the drone, nor be lower than the safe flight altitude (avoid getting close to the ground or obstacles). Airspace constraints: The flight direction and adjusted position after the release maneuver must not enter restricted airspace such as no-fly zones or airport take-off and landing zones; Attitude constraints: During the adjustment process, the pitch and roll angles of the drone must not exceed the safety thresholds to avoid loss of attitude control.

[0044] Furthermore, after the liberation is completed, the effectiveness of the liberation action can be evaluated. If the liberation effect is not achieved, the liberation strategy can be readjusted to form a closed-loop optimization. In one specific implementation, the evaluation of the liberation effect may include the following steps: Step 1: During the disengagement process, the distance between the two conflicting drones (including the horizontal distance D) is measured every detection time Δt. h,ik Vertical distance D v,ik Determine whether D is satisfied. h,ik ≥D h,safe And D v,ik ≥D v,safe The conflict resolution condition is met, and the resolution is completed when this condition is met.

[0045] Simultaneously, full-parameter features of the conflict resolution process are collected, including conflict level, various conflict indicators, and conflict resolution action parameters. The conflict level parameter can be a comprehensive conflict level score, and the conflict indicator parameters can be quantitative representations of their respective indicators. Optionally, the sampling period can be Δt, starting from three time slices before the resolution action is triggered and ending ten time slices after the resolution is completed, to complete full-time-series data collection. All data is bound to a unique conflict event ID and UAV ID, forming a traceable resolution process dataset, as shown in Table 1. Table 1 Step 2: Input the full-parameter features from the conflict resolution process into a decision tree model, which will then output a resolution effectiveness score. Optionally, this decision tree model can be pre-trained in the following way: First, the decision tree model is trained using multiple conflict event samples. After the full-parameter features of each sample are input into the model, it can output the labeled relief effect for each sample. Optionally, the full-parameter features of each conflict event constitute a conflict event sample. The relief effect of each conflict event sample can be manually labeled, and the labeling result can be expressed as a comprehensive relief effect score S. effect The score range is [0, 10]. Optionally, the following factors can be considered in the annotation to improve annotation quality: Thoroughness of conflict resolution and ability to prevent secondary risks: For example, pay attention to parameters such as the horizontal and vertical distance between the two aircraft after the resolution strategy is completed, and the minimum distance between the two aircraft during the resolution process; Reduce interference with the mission: for example, monitor parameters such as mission delay duration and flight path deviation; The accuracy of the drone's execution of release commands: for example, focusing on the mean square error between the target action value and the actual feedback value; Additional energy consumption during the release action: For example, pay attention to the additional power consumption caused by the release action, and parameters such as the remaining power of the drone before the release is triggered.

[0046] To improve the accuracy of the annotations, the manual annotations mentioned above must be completed by experienced experts in the field of UAV flight. At the same time, this embodiment provides a variety of core objective parameters as annotation references to ensure the rationality of the annotation results as much as possible.

[0047] The decision tree model is trained using labeled conflict event samples. After inputting the full-parameter features of each sample into the decision tree model, the model's output continuously approximates the comprehensive resolution effectiveness score labeled for each sample. Since there is a non-linear relationship between the comprehensive resolution effectiveness score and the full-parameter features, this embodiment uses a decision tree model to learn this relationship, so that a reasonable comprehensive resolution effectiveness score can be obtained even without expert supervision.

[0048] After the decision tree model is trained, the full parameter features of the target conflict event currently being processed are input into the trained decision tree model, and the model can automatically output a comprehensive score of the resolution effect of the target conflict event.

[0049] S140. When the relief effect is lower than the set level, the SHAP model is used to locate the cause of the poor relief effect.

[0050] Based on the comprehensive score of relief effect S effect The effects of liberation can be divided into multiple levels. For example, Table 2 shows one way to classify the effects of liberation.

[0051] Table 2 To analyze the reasons for unsatisfactory liberation outcomes, this embodiment focuses on conflict events such as critical liberation, liberation failure, and severe anomalies. It utilizes a decision tree model and the SHAP (SHapley Additive exPlanations) model to analyze the causes of substandard liberation outcomes. In general, the reasons for substandard liberation outcomes can be categorized as follows: Root causes of conflict classification: excessively high or quantitatively biased conflict indicators, excessively high or biased overall conflict level score, and unreasonable weighting of conflict indicators. Root causes related to relief strategies include: insufficient / excessive range of motion, unreasonable adjustment direction, and unreasonable adjustment duration. Root causes related to execution and environment: excessive equipment execution deviation, excessive communication latency, failure to avoid airspace constraints, etc.

[0052] Each cause can be reflected in one or more features of the full-parameter features. Decision tree models, while evaluating the effectiveness of conflict resolution, can also filter out factors related to S. effect The most influential parameter feature is the S-value of the target conflict event. Therefore, in this embodiment, when the resolution effect of the target conflict event is lower than a set level, a decision tree model is used to filter the S-values ​​of the target conflict event. effectThe most influential parameters are selected as candidate features for root cause localization. Specifically, the decision tree model performs inference according to the branching logic of "root node → internal node → leaf node," and calculates the importance score of each input feature using the information gain ratio. The higher the score, the greater the influence of the feature on the resolution failure. Features with importance scores greater than a set threshold (e.g., 0.05) can be retained, while redundant features with no significant impact are eliminated to obtain the candidate feature set.

[0053] For each candidate feature, the SHAP value is used to quantify the feature contribution. The SHAP model calculates the contribution of each candidate feature to S. effect The contribution value (i.e., SHAP value) of a feature is calculated as follows: the SHAP value of a feature ≈ the score when the feature is present minus the score when the feature is absent; a negative SHAP value indicates that the feature lowers the relief effect score and is a cause of failure; a positive SHAP value indicates that the feature improves the relief effect score and has no negative impact. The SHAP model outputs the SHAP contribution value of each candidate feature, extracting the negative contribution value. The smaller the value (the larger the negative value), the stronger the impact on failure.

[0054] The top few candidate features with the lowest negative contribution values ​​(below the set negative threshold) are the final root cause localization results. These features will be displayed in the relief effect assessment and root cause localization report for further analysis and judgment by back-office staff.

[0055] In particular, when the root cause localization results include the conflict level (i.e., the comprehensive conflict level score), it indicates that the conflict level has a significant negative impact on the relief effect. However, since the comprehensive conflict level score is obtained by weighting the quantitative representation of each conflict indicator (as shown in formula (4)), it is possible that the conflict indicators have increased the negative impact of the conflict level on the relief effect, or that the unreasonable weighting of each conflict indicator has increased the negative impact of the conflict level on the relief effect. However, the latter reason cannot be identified by the SHAP model, making it difficult to determine whether to improve the conflict indicators / adjust the quantitative representation of the conflict indicators, or to adjust the weighting of each conflict indicator. To compensate for this deficiency, this embodiment constructs the following contribution decomposition model: (5) in, This indicates the contribution of the conflict level score to the effectiveness of the resolution. Indicates the first The contribution of each conflict indicator to the resolution effect has a boundary effect. This indicates the contribution of an unreasonable weighting of various conflict indicators to the effectiveness of conflict resolution.

[0056] Based on the above model, the contribution of unreasonable weighting of each conflict indicator to the effectiveness of conflict resolution can be identified through the following steps: Step 1: Acquire multiple conflict event samples in advance, and use the full parameter features of the conflict level removed from each sample to train another decision tree model for evaluating the relief effect; and use the other decision tree model and the SHAP model to determine the contribution boundary effect of each conflict index to the relief effect, wherein the contribution boundary effect is used to characterize the expected contribution of a unit conflict index to the relief effect.

[0057] In this step, another decision tree model is trained. The input to this model is the full-parameter features of the conflict event after removing the conflict level, and the output is still the resolution effect of the conflict event. The purpose of training this decision tree model is to clarify the contribution boundary effect of each conflict indicator to the resolution effect. Since the conflict level also includes the influence of each conflict indicator, it will interfere with the extraction of the boundary effect of the indicator itself. Therefore, the conflict level is removed from the input features of this decision tree model. For ease of distinction and description, this embodiment refers to the decision tree model used to evaluate the resolution effect of the target conflict event in S130 as the online decision tree model, and the decision tree model used to clarify the contribution boundary effect in this step as the basic decision tree model.

[0058] After the basic decision tree model is trained, a batch of samples with reasonable conflict indicator weight allocations are first identified as high-quality samples. The basic decision tree model and the SHAP model are then used to process these high-quality samples to obtain the contribution of each conflict indicator to the relief effect of each high-quality sample. For example, the reasonableness of the sample weight allocation can be analyzed manually; alternatively, the perfectly relieved samples shown in Table 2 can be directly used as high-quality samples. Assuming that conflict level scores and reasonable weight allocations are removed, the contribution of each conflict indicator output by the SHAP model can be considered as the contribution of the conflict indicator itself to the relief effect.

[0059] Then, based on the ratio of the contribution of each conflict index to the relief effect in the same high-quality sample to the quantified representation of the conflict index, the contribution of the unit quantified representation of each conflict index to the relief effect of the same high-quality sample is calculated. After performing the above operation on all high-quality samples, the expected value of the contribution of the unit quantified representation of each conflict index to the relief effect of each high-quality sample is taken to obtain the boundary effect of the contribution of each conflict index to the relief effect, i.e., the contribution of the conflict index to the relief effect in the above model. : (6) in, Indicates the first j The contribution of each conflict indicator to the resolution effect. This indicates that for all high-quality samples Take the expected value.

[0060] Step 2: For the target conflict event, extract the negative contribution of the conflict level obtained using the online decision tree model and SHAP model to the resolution effect of the target conflict event; and decompose the negative contribution by using the boundary effect of the contribution of each conflict index to the resolution effect, to obtain the negative contribution of the unreasonable weighting of each conflict index to the resolution effect of the target conflict event.

[0061] For ease of distinction and description, this embodiment refers to the negative contribution of conflict level to the resolution effect of the target conflict event as the first negative contribution, and the negative contribution of unreasonable weighting of various conflict indicators to the resolution effect of the target conflict event as the second negative contribution. This step uses the first negative contribution as... Substituting into formula (5), the result obtained from formula (6) is... Specific events in conflict with the target Substituting into formula (5), we can obtain the target conflict event corresponding to... ,Right now: (7) Step 3: If the second negative contribution meets the root cause location conditions (e.g., it is below the set negative threshold and meets the conditions of the few with the smallest negative contribution), determine that the unreasonable weighting of each conflict indicator is one of the reasons for the poor resolution effect of the target conflict event.

[0062] Specifically, if If the value is close to 0, it indicates that the unreasonable weighting of the conflict indicators is not the cause of the poor resolution of the target conflict event, and there is no need to adjust the weighting ratio; however, if... If the weighting is too low, it indicates that the weighting of each conflict indicator is unreasonable, which is the reason for the poor resolution of the target conflict event. It is recommended that the back-end staff further optimize the weighting.

[0063] Furthermore, in another specific implementation, the contribution boundary effect of each conflict index to the resolution effect will differ under different conflict scenarios (such as different spatial complexity, different environmental parameters, etc.). Therefore, this embodiment provides the following method to consider the conflict scenario type when determining the boundary effect: Step 1: Cluster the basic conflict features in the full-parameter features to obtain multiple conflict scenarios. Optionally, remove the conflict event ID and the drone IDs of the conflicting parties from the basic conflict feature data shown in Table 1, and the remaining basic conflict features can be included in the clustering objects.

[0064] Step 2: Quantify the units of each conflict index to estimate the contribution of each high-quality sample to the relief effect in the same type of conflict scenario, thus obtaining the boundary effect of the contribution of each conflict index to the relief effect in the same type of conflict scenario. The model (5) is then improved as follows: (8) in, Indicates the first Several conflict indicators in conflict scenarios The contribution boundary effect to the relief effect. The calculation method is as follows: (9) in, Representing a conflict scenario The first of the high-quality samples below A quantitative representation of conflict indicators. Representing a conflict scenario The first of the high-quality samples below The contribution of each conflict indicator to the resolution effect.

[0065] Having obtained the aforementioned boundary effects, when decomposing the first negative contribution of a target conflict event, we can first determine the target conflict scenario to which the target conflict event belongs based on the fundamental conflict characteristics of the target conflict event; then, the first negative contribution can be used as... Substituting into equation (8), the target conflict scenario obtained from equation (9) Substituting into equation (8), we obtain the second negative contribution of the unreasonable weighting of each conflict indicator to the resolution of the target conflict event. : (10) Therefore, subsequent operations are performed based on the second negative contribution.

[0066] In summary, this embodiment provides a method for resolving drone conflicts, which can achieve the following beneficial effects: 1. Existing technologies, after detecting multi-level conflicts, fail to consider multiple factors such as the drone's flight speed, flight altitude, mission priority, and equipment status, making it impossible to distinguish the severity of the conflict. This results in all conflicts being handled with a uniform resolution strategy, leading to poor flexibility and low resource utilization. For example, treating a minor conflict between a low-priority drone and a high-priority drone as the same as a severe conflict between two high-priority drones not only affects the execution efficiency of high-priority tasks but also fails to achieve precise conflict management.

[0067] Therefore, the method in this embodiment adopts a multi-dimensional comprehensive scoring method, which combines multiple indicators such as conflict distance, duration, flight speed, airspace complexity, and mission priority to classify conflicts into three levels: minor, moderate, and severe. This method can accurately distinguish the severity of conflicts and provides a scientific basis for the formulation of subsequent differentiated resolution strategies, avoiding a "one-size-fits-all" approach to conflict handling.

[0068] 2. Existing conflict resolution strategies are rigid and mostly adopt "fixed avoidance rules," such as "the drone on the left avoids the drone on the right" or "the low-speed drone avoids the high-speed drone." These strategies do not take into account the real-time status of the drones and the dynamic information of the mission requirements, and therefore cannot achieve dynamic and personalized conflict resolution scheduling. At the same time, the conflict resolution actions lack a hierarchical design. Regardless of the severity of the conflict, the same avoidance range is used. This can lead to mission delays due to excessive avoidance ranges or failure to completely resolve the conflict due to insufficient avoidance ranges. Furthermore, the original flight plan cannot be quickly restored after conflict resolution, which affects the overall mission execution efficiency.

[0069] Therefore, the method in this embodiment formulates differentiated resolution strategies based on conflict level and UAV priority to achieve the resolution objective of "prioritizing high priority and minimizing mission impact". At the same time, during the resolution process, the resolution action parameters can be dynamically adjusted according to the real-time status of the UAV, the airspace environment, and the resolution effect, and multi-dimensional constraint control can be applied to avoid generating new conflicts. After the resolution is completed, the original flight plan can be quickly restored, thus improving mission execution efficiency.

[0070] 3. Existing technologies do not address the conflict resolution effect or the reasons for poor resolution, thus failing to provide support for optimizing conflict resolution methods.

[0071] Therefore, the method in this embodiment evaluates the resolution effect of conflict events using a decision tree model and combines it with the SHAP model to perform root cause analysis on conflict events with substandard resolution effects, identifying parameters affecting the resolution effect. By constructing full-parameter features including those related to conflict level and resolution actions, multiple factors that may affect the resolution effect are considered. Furthermore, the influence of conflict indicator weights hidden in conflict grading on the resolution effect is explicitly expressed through boundary effects, distinguishing the impact of the conflict indicators themselves and unreasonable weight allocation on the resolution effect. Additionally, conflict scenarios are divided through clustering of basic conflict features, and the boundary effects of different conflict indicator contributions are solved in different scenarios, providing more effective data support for optimizing conflict resolution methods.

[0072] 4. In the existing technology, conflict detection and conflict resolution are independent of each other. Conflict detection is only responsible for identifying conflicts and cannot accurately synchronize relevant information about the conflict (such as conflict level, conflict time, conflict location, etc.) to the conflict resolution process. This results in a lack of accurate data support for the formulation of resolution strategies. At the same time, the resolution effect is not fed back in real time during the resolution process, and the resolution strategy cannot be dynamically adjusted according to the resolution progress. This may lead to problems such as incomplete resolution and secondary conflicts.

[0073] To this end, this embodiment constructs a closed-loop system of conflict detection, conflict classification, conflict resolution, and feedback optimization to ensure the accurate transmission of conflict information and the efficient execution of resolution strategies. At the same time, the resolution method is continuously optimized through AI algorithms, which improves the system's adaptability and robustness, enabling it to cope with the multi-machine conflict management needs in high-density, high-dynamic, and complex environments.

[0074] It should be noted that all data involved in this application are information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation portals are provided for users to choose to authorize or refuse.

[0075] Figure 2 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention, such as... Figure 2 As shown, the device includes a processor 60, a memory 61, an input device 62, and an output device 63; the number of processors 60 in the device can be one or more. Figure 2 Taking a processor 60 as an example; the processor 60, memory 61, input device 62, and output device 63 in the device can be connected via a bus or other means. Figure 2 Taking the example of a connection between China and Israel via a bus.

[0076] The memory 61, as a computer-readable storage medium, can be used to store software programs, computer-executable programs, and modules, such as the program instructions / modules corresponding to the UAV conflict resolution method in this embodiment of the invention. The processor 60 executes various functional applications and data processing of the device by running the software programs, instructions, and modules stored in the memory 61, thereby implementing the aforementioned UAV conflict resolution method.

[0077] The memory 61 may primarily include a program storage area and a data storage area. The program storage area may store the operating system and at least one application program required for a given function; the data storage area may store data created based on terminal usage. Furthermore, the memory 61 may include high-speed random access memory and non-volatile memory, such as at least one disk storage device, flash memory, or other non-volatile solid-state storage device. In some instances, the memory 61 may further include memory remotely located relative to the processor 60, which can be connected to the device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0078] Input device 62 can be used to receive input digital or character information, and to generate key signal inputs related to user settings and function control of the device. Output device 63 may include display devices such as a display screen.

[0079] This invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the UAV conflict resolution method of any embodiment.

[0080] The computer storage medium of this invention can be any combination of one or more computer-readable media. A computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0081] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, capable of sending, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device.

[0082] Program code contained on a computer-readable medium may be transmitted using any suitable medium, including but not limited to wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0083] Computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof. Programming languages ​​include object-oriented programming languages—such as Java, Smalltalk, and C++—as well as conventional procedural programming languages—such as C or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0084] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention 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 or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the technical solutions of the embodiments of the present invention.

Claims

1. A method for resolving conflicts between unmanned aerial vehicles (UAVs), characterized in that, include: Acquire potential target conflict events between drones; The conflict level of the target conflict event is evaluated by weighted summation of multiple conflict indicators, wherein the multiple conflict indicators include at least one of conflict distance, conflict duration, flight speed, airspace complexity, and mission priority. Conflict resolution is performed based on the conflict level, and the full-parameter features of the conflict resolution are input into an online decision tree model to evaluate the resolution effect. The full-parameter features include the conflict level, various conflict indicators, and conflict resolution action parameters. Multiple conflict event samples are pre-acquired. A basic decision tree model for evaluating the relief effect is trained using all parametric features of the conflict level removed from each sample. Using the basic decision tree model and the SHAP model, high-quality samples with reasonable weighting are processed to obtain the contribution of each conflict index to the relief effect of each high-quality sample. Based on the ratio of the contribution of each conflict index to the relief effect in the same high-quality sample to the quantified representation of the conflict index, the contribution of the unit quantified representation of each conflict index to the relief effect of the same high-quality sample is calculated. The expected value of the contribution of the unit quantified representation of each conflict index to the relief effect of each high-quality sample is then obtained to obtain the boundary effect of the contribution of each conflict index to the relief effect. When the resolution effect of the target conflict event is lower than the set level, the SHAP model is used to solve the first negative contribution of the conflict level to the resolution effect of the target conflict event. Utilizing the boundary effect of the contribution of each conflict indicator to the resolution effect, the first negative contribution is... The decomposition yields the second negative contribution of the unreasonable weighting of each conflict indicator to the resolution of the target conflict event. : in, and They represent the first The quantitative representation and weighted weight of each conflict index in the target conflict event. Indicates the first The contribution of each conflict indicator to the resolution effect has a boundary effect. If the second negative contribution is lower than the set negative threshold, the unreasonable weighting of each conflict indicator is determined to be the reason for the poor resolution effect of the target conflict event, so as to make up for the defect that the unreasonable weighting of each conflict indicator cannot be identified by the SHAP model.

2. The UAV conflict resolution method according to claim 1, characterized in that, The process of performing conflict resolution based on the conflict level includes: In the event of a minor conflict, adjust the altitude and flight speed of the conflict drone. In the event of a medium-level conflict, adjust the altitude, speed, and direction of the conflict drone. In the event of a severe conflict, adjust the altitude, speed, and direction of the conflict drone. The adjustment range for each conflict level is as follows: severe conflict > moderate conflict > minor conflict.

3. The UAV conflict resolution method according to claim 1, characterized in that, The process of performing conflict resolution based on the conflict level includes: Based on the conflict level, adjust at least one of the following: the altitude, flight speed, and flight direction of the conflict drone. If the distance between the conflict points continues to decrease during the adjustment process, increase the adjustment range; If the distance to the collision increases rapidly during the adjustment process, the disengagement maneuver should be stopped in advance and the original flight status restored to minimize the impact on the mission.

4. The method according to claim 1, characterized in that, Before inputting the full-parameter features of conflict resolution into the online decision tree model to evaluate the resolution effect, the following steps are also included: The online decision tree model is trained using multiple conflict event samples. After the full parameter features of each sample are input into the online decision tree model, the model can output the relief effect of each sample label.

5. The UAV conflict resolution method according to claim 1, characterized in that, The step of taking the expected contribution of each conflict index to the relief effect of each high-quality sample by quantifying the unit of each conflict index is used to obtain the boundary effect of the contribution of each conflict index to the relief effect. This includes: clustering the basic conflict features of multiple samples with the best relief effect to obtain multiple conflict scenarios; taking the expected contribution of each conflict index to the relief effect of each high-quality sample in the same conflict scenario by quantifying the unit of each conflict index to the relief effect, and obtaining the boundary effect of the contribution of each conflict index to the relief effect in the same conflict scenario. Accordingly, the step of decomposing the first negative contribution by utilizing the boundary effect of the contribution of each conflict indicator to the resolution effect to obtain the second negative contribution of the unreasonable weighting of each conflict indicator to the resolution effect of the target conflict event includes: determining the target conflict scenario to which the target conflict event belongs based on the basic conflict characteristics of the target conflict event; and determining the second negative contribution of the unreasonable weighting of each conflict indicator to the resolution effect of the target conflict event according to the following decomposition formula. : in, This represents the first negative contribution. and They represent the first The quantitative representation and weighted weight of each conflict index in the target conflict event. Indicates the first The boundary effect of the contribution of each conflict index to the relief effect in the target conflict scenario.

6. An electronic device, characterized in that, include: One or more processors; Memory, used to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the unmanned aerial vehicle (UAV) conflict resolution method according to any one of claims 1-5.

7. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the UAV conflict resolution method according to any one of claims 1-5.

Images