Method for predicting aircraft conflict and method for preventing collision

Abstract

The application discloses an aircraft conflict prediction method and an anti-collision method, relates to the field of aircrafts, and aims to avoid the risk of secondary conflict and a series of problems caused by real-time decision-making. The application predicts future track data based on the track data of cooperative aircrafts and non-cooperative aircrafts, performs conflict prediction on the future track data according to a safety distance, and formulates an anti-collision strategy through an enhanced algorithm. The application avoids the problem that multiple aircrafts involved in a conflict event cannot communicate anti-collision strategies with each other and easily cause secondary conflict, and the conflict risk is predicted in advance, data basis is provided in advance for the formulation of the anti-collision strategy, and sufficient time is reserved.

Description

Technical Field

[0001] This invention relates to the field of aircraft, and in particular to an aircraft conflict prediction method and collision avoidance method. Background Technology

[0002] Aircraft collision avoidance technology has evolved from basic warnings to collaborative avoidance. Early collision avoidance mainly relied on single sensors such as radar to detect threats, which triggered warnings by preset sensing thresholds, resulting in inherent defects such as limited coverage and high false alarm rates.

[0003] With the application of multi-source data fusion technology, collision avoidance mechanisms based on transponder signal interaction and geometric modeling have gradually become widespread. These collision avoidance methods generate vertical avoidance commands by calculating the relative positions between aircraft in real time, achieving good results in simple scenarios. However, this approach still has significant limitations: the decision-making algorithms mostly use static rule bases and fixed priority sorting, making it difficult to adapt to dynamic multi-aircraft game scenarios. Multiple aircraft involved in a conflict cannot communicate their collision avoidance strategies, and when multiple aircraft trigger avoidance commands simultaneously, it may actually exacerbate the collision risk, often leading to secondary conflicts.

[0004] In addition, current known collision avoidance methods all rely on real-time data assessments to determine whether a conflict is imminent, making decisions only when a conflict risk has already materialized. This results in short decision-making times and high safety risks. Summary of the Invention

[0005] The purpose of this invention is to provide an aircraft conflict prediction method and collision avoidance method that addresses all or part of the problems mentioned above. This method employs centralized detection and avoidance to mitigate the risk of secondary conflicts and the series of problems caused by real-time decision-making.

[0006] The technical solution adopted in this invention is as follows:

[0007] An aircraft conflict prediction method, comprising:

[0008] S1. Acquire the first and planned track data of the cooperating aircraft, and acquire the visual and second track data of the non-cooperating aircraft.

[0009] S2. Based on the time and positional deviations between the first trajectory data and the planned trajectory data, the planned trajectory data for a predetermined future time period is corrected to obtain the predicted trajectory data.

[0010] S3. Enhance the visual features of the second track data using the visual data, identify the target type of non-cooperative aircraft based on the visual features, and predict the future flight area of ​​different non-cooperative aircraft based on different target types.

[0011] S4. Based on the minimum distance between the predicted flight path data of cooperative aircraft, or whether the minimum distance between the predicted flight path data of cooperative aircraft and the future flight area of ​​non-cooperative aircraft exceeds the safe distance, whether a conflict is predicted, wherein non-cooperative aircraft of different target types have different safe distances.

[0012] On the other hand, the present invention also provides an aircraft collision avoidance method, which includes: predicting whether a collision will occur using an aircraft conflict prediction method, and performing the following steps when an aircraft conflict is predicted:

[0013] S5. Identify the cooperative aircraft with conflict risk, and sort all cooperative aircraft according to the priority indicated by the planned flight track data, with the principle of prioritizing cooperative aircraft with conflict risk first and cooperative aircraft without conflict risk last.

[0014] S6. Based on the state information at the moment when there is a risk of conflict, determine the cooperative aircraft that need to be scheduled according to the scheduling order from low to high priority using the target value network;

[0015] S7. Based on the state information at the moment when there is a risk of conflict, use the policy network to generate scheduling actions for the cooperative aircraft that need to be scheduled at the moment when there is a risk of conflict.

[0016] S8. The schedule is sent to the corresponding cooperating aircraft so that the scheduled aircraft can avoid conflict at times of conflict risk.

[0017] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0018] This invention employs a centralized conflict prediction and collision avoidance strategy, avoiding the problem of secondary conflicts easily caused by multiple aircraft involved in a conflict failing to communicate their collision avoidance strategies. It also increases the freedom of flight path planning by coordinating multiple cooperating aircraft to reduce conflict risk. Furthermore, by predicting aircraft' future flight paths or areas in advance, this invention anticipates conflict risks, providing data for the development of collision avoidance strategies and allowing sufficient time for the formulation of more comprehensive and reliable long-term flight control strategies. Moreover, based on the different characteristics of cooperating and non-cooperating aircraft in low-altitude conflict scenarios, this invention proposes different conflict prediction methods, rather than arbitrarily designating the area around all types of aircraft as a conflict risk zone. This more efficiently utilizes space resources, reduces the possibility of misjudgment of conflict, and minimizes the need for unnecessary collision avoidance strategy development. Attached Figure Description

[0019] The present invention will be described by way of example and with reference to the accompanying drawings, wherein:

[0020] Figure 1This is a flowchart of one embodiment of the aircraft conflict prediction method.

[0021] Figure 2 This is a data flow diagram of one embodiment of an aircraft collision avoidance method.

[0022] Figure 3 This is a construction diagram of a policy network in one embodiment; Figure 4 This is a construction diagram of a value network in one embodiment.

[0023] Figure 5 This is a flowchart of one embodiment of an aircraft collision avoidance method. Detailed Implementation

[0024] All features disclosed in this specification, or all steps in all disclosed methods or processes, may be combined in any way, except for mutually exclusive features and / or steps.

[0025] Any feature disclosed in this specification (including any appended claims and abstract) may be replaced by other equivalent or similar features, unless specifically stated otherwise. That is, unless specifically stated otherwise, each feature is merely one example of a series of equivalent or similar features.

[0026] To address the problems of current aircraft collision avoidance strategies, such as the susceptibility to secondary conflicts, short decision-making time, and high safety risks associated with decisions to avoid conflict only after the aircraft is already at risk, this application provides an aircraft conflict prediction and collision avoidance method. The aim is to provide a centralized aircraft conflict prediction and collision avoidance strategy to avoid the risk of secondary conflicts and to achieve advanced conflict prediction, allowing sufficient time for decision-making on avoidance schemes and preventing aircraft from being at risk of conflict.

[0027] like Figure 1 As shown, an optional aircraft conflict prediction method in this application includes the following steps:

[0028] S1. Acquire the first and planned track data of the cooperating aircraft, and acquire the visual and second track data of the non-cooperating aircraft.

[0029] Cooperative aircraft are those under unified control, meaning they are all under the unified control of the ground management center; non-cooperative aircraft are those that are not under the unified control of the ground management center.

[0030] Flight path data describes the position and status of an aircraft at different times, arranged in chronological order to form the aircraft's flight path. Flight path data contains two important features: time and position, which characterize the aircraft's location at what time.

[0031] The actual flight path data of an aircraft can typically be captured using sensing / detection / surveillance systems (collectively referred to as detection systems for ease of description). Different detection systems, due to varying accuracy errors, may acquire path data that is not entirely identical. In this application, cooperative aircraft and / or non-cooperative aircraft employ multiple detection systems to acquire diverse path data.

[0032] As an optional implementation, cooperative aircraft employ multiple detection systems to acquire track data in multiple modalities, which are then fused into a first track data. Non-cooperative aircraft may employ one or more detection systems to acquire track data in at least one modality. If only one detection system is used, the acquired single-modal track data becomes the second track data. If multiple detection systems are used, the track data in multiple modalities are fused into the second track data.

[0033] In some alternative implementations, the process of acquiring the first track data includes:

[0034] S11. Obtain multimodal flight path data of cooperative aircraft.

[0035] For example, the data obtained by the ground management center includes:

[0036] 1) Visual data. Visual data refers to video / image data collected through visual devices (such as cameras, infrared detection devices, etc.). Visual data covers both cooperative and non-cooperative aircraft.

[0037] 2) First Mode Track Data. First mode track data is obtained through a single radar system detection. This first mode track data is obtained for both cooperative and non-cooperative aircraft.

[0038] 3) Second-mode track data. Second-mode track data is obtained through secondary radar systems. Second-mode track data is only detected by cooperative aircraft equipped with secondary radar. Each radio signal and each piece of second-mode track data detected by the secondary radar system carries aircraft identification information.

[0039] 4) Third-mode track data. Third-mode track data is obtained through the ADS-B system. Third-mode track data is only detected for cooperative aircraft equipped with ADS-B out devices. Each radio signal and each piece of third-mode track data detected by the ADS-B system carries aircraft identification information.

[0040] 5) Fourth Mode Track Data. Fourth mode track data is obtained through the Remote ID system. Fourth mode track data is only detected for cooperative aircraft equipped with Remote ID devices. Each radio signal and each piece of fourth mode track data detected by the Remote ID system carries aircraft identification information.

[0041] 6) Flight Plan. A flight plan contains planned flight track data, and each track contains aircraft identification information. Furthermore, each track typically indicates its priority, representing the order of the flight mission. Clearly, flight plans are only for cooperating aircraft.

[0042] In the example above, the cooperating aircraft is not required to have all the detection systems. That is, for a given cooperating aircraft, the ground control center may only detect one or more modes of track data other than the first mode track data.

[0043] S12. Weighted fusion of some or all of the multimodal trajectory data is performed to obtain the first trajectory data.

[0044] Each mode of flight track data carries aircraft identification information. Therefore, the flight track data of each mode can be associated based on the aircraft identification information, and it can also be associated with planned flight track data. Assume the number of cooperating aircraft is... Then generate The group is a set of associations identified by aircraft identity information. In form, this... The group association can be a mapping table of "aircraft identity information - track data information - data source". "Track data information" includes the position information, speed information, priority information (if any) carried by the track data. The data source refers to the source from which the track data is obtained, such as a radar system, ADS-B system or flight plan.

[0045] Typically, planned flight path data and actual flight path data of an aircraft deviate from each other in terms of time and location. For example, even if the flight paths are similar, the actual takeoff time may differ from the planned takeoff time by several minutes or even longer; or the flight time may match the planned time, but the flight path may deviate from the planned path; or there may be deviations in both time and path. Actual flight path data is obtained from various detection systems. Due to differences in performance and focus, the accuracy and reliability of the flight path data obtained by different detection systems vary. Furthermore, the flight path data returned by equipment within the aircraft is more accurately matched to its actual flight path. Therefore, it is advisable to consider fusing the flight path data detected by other detection systems besides the primary radar system (i.e., second-mode, third-mode, and fourth-mode flight path data) to combine the advantages of each detection system and obtain more reliable flight path data.

[0046] In some feasible implementations, weights are assigned to the second-mode track data, third-mode track data, and fourth-mode track data based on the historical accuracy and reliability of the secondary radar system, ADS-B system, and Remote ID system. The core principle for assigning weights to these three modes of track data is based on the real-time reliability of each data source at the current moment. The ADS-B system and Remote ID system can actively broadcast their own accuracy and integrity indicators; therefore, the ground management center can obtain the measurement error variances of the third-mode and fourth-mode track data based on the broadcast content, denoted as follows: The variance of the measurement error of the secondary radar is denoted as... The distance, azimuth, and performance parameters of the ground-based secondary radar equipment relative to the target aircraft can be estimated in real time by the ground management center. The weights of the second, third, and fourth mode track data can then be obtained using the following equation:

[0047] .

[0048] The second-mode, third-mode, and fourth-mode track data are weighted and fused according to their assigned weights to obtain the first track data. The first track data can be in the following format: Where ID represents aircraft identity information, and P represents priority. Representing the ( ) 100 cooperative aircraft The three-dimensional (x, y, z) coordinates at any given time, where 'r' in the subscript indicates real (actual trajectory). and Representing the first The start time (star, denoted by subscript s) and end time (end, denoted by subscript e) of an aircraft's actual flight path.

[0049] S2. Based on the time and position deviations between the first trajectory data and the planned trajectory data, the planned trajectory data within a predetermined time period is corrected to obtain the predicted trajectory data.

[0050] As mentioned earlier, there may be time and positional deviations between the actual flight path of an aircraft and the planned flight path data. If these time and positional deviations are eliminated, the planned flight path data can be used as the predicted flight path data for the future aircraft, that is, as the true value for predicting future flight path data.

[0051] Based on the first track data obtained in step S1, it is represented as follows: Assuming the planned flight path data is represented as The subscript 'p' indicates 'plan' (planned trajectory). Representing the A cooperative aircraft The three-dimensional coordinates at time [time]. and Representing the first The start and end times of the planned flight path for each aircraft are determined. The first flight path data is matched with the planned flight path data to calculate the time and positional deviations between the actual and planned paths. Since these deviations involve both time and position, they can be expressed as... ,in This represents the time deviation of the nth cooperating aircraft. This represents the positional deviation of the nth cooperating aircraft (on the x, y, and z axes). The planned trajectory data is then corrected using the calculated time and positional deviations, and the corrected planned trajectory data is used as the predicted trajectory data for the future cooperating aircraft.

[0052] As an optional implementation, a method for calculating the time and position deviations between the first track data and the planned track data includes:

[0053] S21. Calculate the distance between the first track data point at each time point within the historical predetermined time and the planned track data point at each time point, and obtain the planned track data point with the shortest distance to the first track data point at each time point from the planned track data.

[0054] by Indicate the current time, in The duration of a historical scheduled time, or simply historical time, is expressed as: The first track data within this predetermined historical time period can be represented as: Assuming historical time is divided into N (N is a positive integer) time slots, the set of moments within a predetermined historical time period is represented as follows: If the time slot length is fixed, and If the time slot length is not an integer multiple of the time slot length, it is approximately divided into the above set, with the last time slot retained or removed.

[0055] by , representing a moment within a predetermined historical time period, where the coordinates of the first track data point of the nth cooperating aircraft at that moment are denoted as . The set of coordinates of the planned flight path data points of the cooperative aircraft is represented as follows: ,calculate and The distance between each coordinate in the middle, from Find with The coordinates of the shortest distance are represented as... .coordinate It can be used as The coordinates of the location. At every moment in the set, it is possible to obtain the data from the set. Find with Shortest distance .

[0056] The above (This indicates that for the nth cooperative aircraft at all planned times) In, with The moment when the actual coordinate distance is shortest (min) can be determined by the following method:

[0057] .

[0058] Based on the above method, for each historical period Both can be derived from the set of planned flight track data. Get the set .

[0059] S22. Determine the time deviation between the first track data and the planned track data by the time offset between each planned track data point and the corresponding first track data point; determine the positional deviation between the first track data and the planned track data by the positional offset between each planned track data point and the corresponding first track data point.

[0060] Step S11 explains that for each time step... Each of these will correspond to a coordinate. Its corresponding This indicates the planned time to reach that coordinate. and The difference is and The time offset, corresponding to, and The spatial distance between them is the positional offset, which can be represented by coordinates, i.e. The time deviation between the first track data and the planned track data can be calculated by averaging the time offset at each instant, and the position deviation between the first track data and the planned track data can be calculated by averaging the position offset at each instant. The total offset is expressed as:

[0061] .

[0062] Based on this, methods for correcting planned flight path data within a predetermined future timeframe include:

[0063] S23. Determine the correction time range for the planned trajectory data based on the time deviation.

[0064] Assume the end time of the aforementioned future scheduled time is expressed as This represents the upper limit of the correction time range starting from the current time. The upper limit of the correction time is obtained by offsetting the planned end time by the aforementioned time deviation. The calculation method is as follows:

[0065] .

[0066] ,in This indicates a future observation time. It means that it is impossible to make corrections for times outside the current observation period.

[0067] S24. For each moment within the correction time range, based on the position of the planned track data point after offsetting the time deviation at that moment, the position deviation is used for correction to obtain the corrected planned track data point (position) at that moment.

[0068] Specifically, after time correction, the corrected time range is expressed as follows: , This means that at any given moment within the correction time range, in the planned trajectory data of the nth cooperating aircraft, the coordinate at time t should be... Due to time discrepancies, the coordinates of this location have shifted to the specified time. Therefore, for the current time t within the correction time range, it is necessary to use the time in the planned track data. The correction is based on the coordinates, specifically the time in the planned trajectory data. The correction is accomplished by adjusting the coordinate offset of the position by the amount of deviation.

[0069] Specifically, position correction refers to the correction of the three-axis coordinates. The correction method is as follows:

[0070] .

[0071] In the formula, This means the offset at time t. The planned flight path data points afterward x Axis coordinates and Similarly.

[0072] Using the methods described above, predicted flight path data with a smaller error compared to the planned flight path data can be obtained. In practical applications, when the error between the actual flight path and the planned flight path data is within a certain range, the aircraft is allowed to maintain its current flight path. If the deviation of the actual flight path exceeds the allowable error range, the ground control center will control the aircraft to return to the path required by the planned flight path data. Therefore, although the actual flight path may deviate from the planned flight path data, the planned flight path data can still serve as the basis for calculating and predicting future flight paths, and this concept has its rationale.

[0073] S3. Enhance the visual features of the second track data using visual data, identify the target type of non-cooperative aircraft based on the visual features, and predict the future flight area of ​​different non-cooperative aircraft based on different target types.

[0074] Taking the data acquired by the ground management center as an example, for non-collaborative aircraft (represented as non-collaborative), the data acquired by the ground management center includes visual data obtained by visual equipment and first-mode track data detected by a radar system. In this embodiment, the visual data of the non-collaborative aircraft and the first-mode track data are fused at the feature level to add visual features to the track data of the non-collaborative aircraft, such as target type, target size, and motion pattern. The feature-level fusion method specifically includes: on the one hand, the position and speed information of the non-collaborative aircraft can also be obtained from the visual data, which has extremely high authenticity. The position, speed, and other visual information of the non-collaborative aircraft in the first-mode track data can be corroborated by the visual data. On the other hand, other information obtained from the visual data, such as target type, target size, and motion pattern, can be added to the first-mode track data to obtain enhanced first-mode track data. In this embodiment, the enhanced first-mode track data will be used to predict the future flight area of ​​the non-collaborative aircraft.

[0075] Non-cooperative targets are diverse, and their types and movement patterns are difficult to predict solely based on flight track data. In some optional embodiments of this application, the type and movement pattern of non-cooperative aircraft are identified based on visual characteristics, thereby predicting the future flight area of ​​the non-cooperative aircraft based on different identification results. The future flight area of ​​the non-cooperative aircraft is equivalent to the no-fly zone of the cooperative aircraft, and the cooperative aircraft need to avoid the future flight area of ​​the non-cooperative aircraft.

[0076] Based on visual data, existing target recognition algorithms can be used to identify the target type of non-cooperative aircraft, and the position coordinates of non-cooperative aircraft can be obtained through flight track data. In some alternative embodiments of this application, the future flight area of ​​a non-cooperative aircraft is predicted based on its target type and location coordinates.

[0077] Assume that the target type identification results for non-cooperative aircraft include two categories: targets with predictable paths and targets with unpredictable paths. Targets with predictable paths typically exhibit stable motion patterns, such as aircraft with fixed speeds or directions. Targets with unpredictable paths lack fixed motion patterns; for example, their trajectories are chaotic, their speeds fluctuate, or they may move intermittently. For targets with predictable paths, their future flight area is based on their predicted path data, rather than arbitrarily defining a large spatial area as their future flight area. If cooperative aircraft can avoid this predicted path data, it can be considered that there is no risk of conflict, reducing the probability of false collisions and improving space utilization. For targets with unpredictable paths, since their motion patterns are unknown, a certain range around them is defined as their potential future destination; that is, the space around them is designated as their future flight area.

[0078] Specifically, methods for predicting the future flight areas of different non-cooperative aircraft based on different target types include:

[0079] S31. For non-cooperative aircraft with predictable target types, predict the future flight area based on historical flight data within a predetermined time period. When the positional deviation between the actual flight data and the predicted flight data reaches a first threshold, or when the time since the last update of the predicted flight data reaches a second threshold, re-predict the future flight data within a predetermined time period.

[0080] S32. For non-cooperative aircraft with unpredictable flight paths, the future flight area of ​​the non-cooperative aircraft is defined with the observed position as the center of a sphere and a predetermined radius. The flight area is updated again when the distance between the observed position and the center of the flight area reaches a third threshold, or when the time since the last predicted flight path data update reaches a second threshold.

[0081] For example, in some practical implementation schemes, non-cooperative aircraft targets are categorized into three types: hovering targets, constant-speed cruise targets, and irregular-track targets. Hovering targets and irregular-track targets can be classified as unpredictable track targets, and future flight areas are predicted by delineating the space; constant-speed cruise targets are classified as predictable track targets, and future flight areas are determined by predicting track data.

[0082] 1) Hovering targets

[0083] Typical non-cooperative aircraft: reconnaissance drones, tethered balloons.

[0084] Motion characteristics: stationary in a fixed position with no significant displacement (such as hovering monitoring between buildings).

[0085] Predicting future flight areas: using the current coordinates of non-cooperative aircraft A spherical region centered on [center] with a radius ≤ 50 meters (or other value). The future navigation area for hovering targets. It can be represented as:

[0086] .

[0087] The predicted navigation area for hovering targets is represented as .

[0088] Predicted flight area update: When the distance between the position coordinates of a non-cooperative aircraft and its position coordinates at the time of the last update exceeds 10 meters (or other values) or the distance since the last update is... After a certain period of time, the predicted flight area is updated. In addition, if a change in the motion characteristics (motion pattern) of a non-cooperative aircraft is observed, its target type is reclassified, and its future flight area is predicted in a manner corresponding to the updated target type.

[0089] 2) Targets cruising at a constant speed

[0090] Typical non-cooperative aircraft: logistics drones, free balloons.

[0091] Motion characteristics: constant speed / direction.

[0092] Predicting future flight areas: Due to the simple and predictable motion characteristics of this type of non-cooperative aircraft, the processing method for this type of non-cooperative aircraft is similar to that for cooperative aircraft, using predicted future flight path data as the future flight area. For example, historical flight path data can be used to predict future observation times. Based on the current detected speed and direction, and according to the time slot step size, the position at multiple future time steps can be predicted from the trajectory data within the current time slot. This is denoted as... , y and z are the x, y, and z coordinates of the nth uniform cruise non-cooperative aircraft at time t.

[0093] Predicted flight area update: When the distance between the flight position of a non-cooperative aircraft and the corresponding waypoint in the predicted flight track data exceeds 3 meters (or other values), or when the distance from the last updated flight area exceeds a certain threshold, the updated flight area will be updated accordingly. After a certain period, the future flight area is updated. Additionally, if changes in the motion characteristics of non-cooperative aircraft are observed, their target types are reclassified, and their future flight areas are predicted accordingly based on the updated target types.

[0094] 3) Targets with irregular trajectories

[0095] Typical non-cooperative aircraft: serpentine interference drones, runaway kites.

[0096] Motion characteristics: No fixed path (such as spiral, random broken line).

[0097] Predicting future navigation areas: estimating maximum permissible speed based on target type The future flight area will be based on the current coordinates of the non-cooperative aircraft. Centered on, radius A spherical region in meters (or other values). The future navigation area for targets with irregular trajectories. It can be represented as:

[0098] .

[0099] The predicted navigation area for targets with irregular trajectories is represented as: .

[0100] Predicted flight area update: When the distance between the position coordinates of a non-cooperative aircraft and its position coordinates at the time of the last update exceeds 10 meters (or other values) or the distance since the last update is... After a certain period, the predicted flight area is updated. Additionally, if changes in the motion characteristics of non-cooperative aircraft are observed, their target type is reclassified, and their future flight area is predicted accordingly based on the updated target type.

[0101] S4. Based on the minimum distance between the predicted flight paths of cooperative aircraft, or the minimum distance between the predicted flight path data of cooperative aircraft and the future flight area of ​​non-cooperative aircraft, whether a conflict is predicted. Different types of non-cooperative aircraft have different safe distances.

[0102] There may be a risk of conflict between cooperative aircraft and other cooperative aircraft, as well as a risk of conflict with non-cooperative aircraft.

[0103] Taking the classification of non-cooperative aircraft target types in the previous text as an example, we will predict whether cooperative aircraft will face conflict risks.

[0104] Non-cooperative aircraft are classified into targets with predictable trajectories and targets with unpredictable trajectories.

[0105] 1) Risk assessment of conflict with aircraft with predictable trajectories.

[0106] Since the trajectory data of both cooperative aircraft and non-cooperative aircraft of the predictable trajectory target type are predictable, the two types of aircraft can be considered together. Here, the two types of aircraft are collectively referred to as predictable trajectory aircraft, and it is only necessary to consider whether there is a risk of conflict between the predicted trajectory data.

[0107] Specifically, the cooperative aircraft have The number of non-cooperative aircraft of predictable target type is assumed to be [number]. Then you can get The predicted flight path data. For the first... To determine whether a cooperative aircraft will face future conflict risks with other aircraft of predictable trajectories (including other cooperative aircraft and non-cooperative aircraft of predictable target types), it is only necessary to calculate waypoints. With point ,point minimum distance between Is it less than the safe distance? ,in, This represents the position (including the x, y, and z axes) of the predicted trajectory data point of the nth non-cooperative aircraft at time t. If Less than Then there is a risk of conflict.

[0108] .

[0109] 2) Conflict risk assessment with non-cooperative aircraft of the type of target with unpredictable trajectories.

[0110] The predicted flight areas for such non-cooperative aircraft are all considered no-fly zones. Taking the above classification of such non-cooperative aircraft into hovering targets and irregular trajectory targets as an example, the set of no-fly zones includes the set of... and set For the first A cooperative aircraft, for any given time If the following inequality holds, then the cooperating aircraft faces the risk of conflict:

[0111] .

[0112] The above examples demonstrate that the safety distance designed for conflict risk analysis of aircraft with predictable trajectories ( Compared to the safe distance designed for non-cooperative aircraft with unpredictable trajectories ( )big This is because, considering that there may be some deviation between the predicted trajectory data and the actual trajectory data, a certain margin is reserved on the basis of the safety distance when comparing trajectory data; and for non-cooperative aircraft of the target type with unpredictable trajectory, the safety distance margin has been taken into account in their future flight area.

[0113] The above methods enable advance prediction of conflict risks in future time steps, allowing sufficient time and operational space for conflict avoidance strategy planning. Furthermore, classifying non-cooperative aircraft for conflict risk assessment, and using their predicted flight paths as future airspace for non-cooperative aircraft with predictable trajectories, rather than arbitrarily designating large areas as no-fly zones, can reduce the probability of misjudgment of conflict.

[0114] Based on the ideas of this application, and building upon conflict prediction, this application also provides an aircraft collision avoidance method to proactively avoid conflict events.

[0115] As an optional implementation method, the aircraft collision avoidance method includes the following steps:

[0116] First, predict whether there is a risk of conflict between the cooperating aircraft. For example, the aircraft conflict prediction method in any of the previous embodiments can be used to predict whether a conflict will occur. When an aircraft conflict is predicted, perform the following steps:

[0117] S5. Identify the cooperative aircraft with conflict risks, and sort all cooperative aircraft according to the priority indicated by the planned flight track data, with the principle of prioritizing cooperative aircraft with conflict risks over those without conflict risks.

[0118] As mentioned earlier, the flight missions performed by the cooperative aircraft are prioritized. In the embodiments of this application, one or more of the lowest priority missions are scheduled first to avoid conflict risks, thereby ensuring that the relatively important flight missions are guaranteed first.

[0119] Assuming the existence is predicted There is a risk of conflict between the cooperating aircraft; for ease of explanation, this... The cooperating aircraft are sorted in ascending order of priority and then denoted as follows: For the remaining Each cooperating aircraft is sorted individually according to its priority from low to high, and then placed into the already sorted list. Then, it is represented as .

[0120] In the embodiments of this application, the scheduling strategy adopted for aircraft collision avoidance is based on an agent reinforcement learning method. The trained policy network and value network (including a state value network and a target value network) determine the future flight state of cooperative aircraft under different state information. As an optional implementation, the structure diagram of the policy network and value network is as follows: Figure 3 and Figure 4 As shown, the policy network inputs state data into a convolutional network layer to extract feature data, and then connects it to a fully connected layer and a Softmax activation function to obtain the probabilities of different actions; the value network inputs state data into a convolutional layer to extract feature data, and then inputs it into a fully connected layer to obtain the cumulative reward value of the value network.

[0121] Reinforcement learning involves state information, behavioral information, and a reward function; state information includes environmental state information and current state predicted trajectory information; behavioral information includes the aircraft's flight actions; and the reward function defines the reward obtained by the cooperative aircraft in different states.

[0122] The so-called state information includes a set of environmental state and current state predicted trajectory information. The environmental state is represented as... ,in Indicates the first The set of three-dimensional coordinates of all points within a no-fly zone, including the predicted future flight area of ​​targets processed according to the no-fly zone, i.e., the set and set This also includes boundary information. Boundary information includes natural boundaries formed by buildings and the ground, as well as man-made boundaries such as controlled areas. Current state predicted trajectory information includes predicted trajectories for cooperative aircraft and predicted trajectories for non-cooperative aircraft whose trajectories are predictable. Current state predicted trajectory information is represented as... ,in Indicates the time when the nth dispatching aircraft is at time of Axis coordinates Indicates the first The scheduling aircraft at the time of Axis coordinates Indicates the time when the nth dispatching aircraft is at time of Axis coordinates Indicates the time when the nth dispatching aircraft is at time speed, The number of non-cooperative aircraft targeting constant-speed cruise targets is ,satisfy ,in, This indicates the maximum number of non-cooperative aircraft that the model can support for targets of the constant-speed cruise class. This refers to the number of participating aircraft. For convenience, it will be referred to as [missing information] in the following text. , .

[0123] Behavioral information belongs to the behavioral space category. Behavioral information refers to a discrete set of actions, including aircraft speed information. The scheduling action A is defined as follows: ,in The aircraft's first Speed, scheduling action Due to different speeds The combination yields, where Indicates the number of selectable speeds. .

[0124] The reward function can be divided into task completion rewards and safety penalties, represented as follows: .in These are adjustable coefficient parameters. Indicates status information, Indicates a scheduling action. The reward for completing the task can be derived from the following formula:

[0125] ;

[0126] in, Indicates the number of tasks completed. This represents the time difference between the actual time taken to complete the mission and the planned flight time. The parameters are configurable, meaning the more tasks completed, the higher the reward; similarly, the smaller the time difference, the higher the reward.

[0127] The definition of security penalties is as follows:

[0128] ;

[0129] in, Indicates the first Aircraft coordinates and no-fly zone coordinates distance, The three-dimensional coordinates of the nth aircraft need to be calculated during the calculation, including the distance to each coordinate point in the no-fly zone, and the minimum value is taken. To obtain a very small positive value, avoid a denominator of 0. It is an adjustable positive value. This is the set of coordinates for the no-fly zone. The closer to the no-fly zone, the greater the penalty.

[0130] In this embodiment of the application, it is assumed that the observation time period has been divided into equal parts. Each time slot Large enough, in practical applications, if it cannot be divided into If the data is divided into equal parts, then the process is approximated as described throughout the text. For ease of description, the discretization process in this embodiment uses symbols. Discretize the representation. Input time... , The state data is used to train the policy network, which outputs the aircraft's state at time... The flight maneuvers are evaluated using flight maneuver data and state data, and a value network is trained to evaluate these maneuvers.

[0131] As an optional implementation, the training process for the policy network and the target value network includes:

[0132] Starting from the first moment in the historical observation period, obtain the current state information;

[0133] Obtain the scheduling action with the highest probability estimated by each cooperating aircraft through different policy networks under the current state information;

[0134] Calculate the reward at the current moment and the state information at the next moment after executing the scheduling action;

[0135] Estimate the first cumulative reward value at the current moment using a state value network;

[0136] Estimate the second cumulative return value at the next time step based on the state information at the next time step using the target value network;

[0137] The TD error is calculated based on the reward and first cumulative reward value at the current moment, and the second cumulative reward value at the next moment;

[0138] The first network parameters are updated based on the change gradient of the TD error and the first cumulative reward value relative to the first network parameters of the state value network.

[0139] Based on the updated first network parameters, the second network parameters of the target value network are updated;

[0140] For each policy network, the third network parameters are updated based on the gradient of the change of the TD error and the logarithm of the policy network output relative to the third network parameters of the policy network.

[0141] For example, during the training phase, for To operate a cooperative aircraft, training is required. The number of cooperative aircraft scheduled varies, thus requiring different policy and value networks. The policy networks are respectively... , The corresponding number of aircraft to be dispatched is ,implement The time indicates that the lowest priority cooperating aircraft is being scheduled, and the operation is being executed. The time indicates that the two lowest priority cooperating aircraft are being scheduled, and so on, to execute... Indicates to All cooperative aircraft are scheduled, with each strategy network involved. This indicates that in each policy network, one independent action is generated for each scheduled aircraft, and the i-th policy network generates a total of i actions. The value network consists of a state value network. and target value network The structure consists of s, where s represents state information. These are the first network parameters of the state value network for the i-th aircraft (corresponding to scheduling i cooperative aircraft) and the second network parameters of the target value network for the i-th aircraft, respectively, where i represents the index of the value network and represents scheduling i cooperative aircraft. During the decision-making phase, only the policy network and the target value network are needed. The accumulated reward value is obtained by calling the policy network and the value network. If the accumulated reward value is less than a threshold... If the scheduling action generated by the policy network is deemed insufficient to complete the task, the system switches to a policy network and value network that calls upon more aircraft collaborators for recalculation, continuing until a feasible scheduling action is achieved. Specifically, the number of collaborating aircraft required for scheduling is [number missing]. The training process is as follows:

[0142] Gaining Time Status information , No. The cooperating aircraft outputs the probabilities of different scheduling actions through a policy function (i.e., a policy network), and the action corresponding to the maximum probability is denoted as... :

[0143] ,and Indicates scheduling The strategy network for cooperating aircraft, in which Indicates the first The third network parameter of a policy network.

[0144] Gaining Time Rewards and time status .

[0145] Predicting time using state value networks First cumulative return value .

[0146] Predicting time using target value network Second cumulative return value ;

[0147] Calculate TD error ,in, ,in This is the discount factor, representing the weight of future rewards in the current calculation.

[0148] Update the state-value network parameters: Let the loss function be expressed as To minimize To achieve the goal, iteratively update the first network parameters of the i-th state value network and the second network parameters of the target value network. The update methods include:

[0149] ,in, This represents the learning rate, which can be set to values ​​such as 0.1 and 0.01 for training and adjustment based on experience. This represents the gradient of the first cumulative reward value relative to the current first network parameters. , Let represent the current first network parameter and the updated first network parameter of the value network in the i-th state, respectively;

[0150] Update target value network parameters: ,in Let these represent the updated second network parameters and the current second network parameters for the i-th target value network, respectively. This represents the soft update coefficient, which is usually expressed as an empirical value.

[0151] Update policy network parameters: Let the loss function be expressed as ,in Represents the time of the nth aircraft The action. To minimize To achieve the goal, iteratively update the third network parameter of the policy network. The update method is as follows: ,in, Indicates the learning rate. This represents the third network parameter of the i-th policy network. This represents the third network parameter after the i-th policy network update; The policy network represents time. Network parameters Downstream scheduling action The logarithm of the probability is the gradient of the change in the third network parameter relative to the current policy network.

[0152] S6. Based on the state information at the moment when there is a risk of conflict, determine the cooperative aircraft that need to be scheduled according to the scheduling order from low to high priority using the target value network.

[0153] Considering that aircraft have different priorities when performing flight missions, during actual scheduling, based on the state information at the moment when there is a risk of conflict, the lowest priority corresponding number of cooperative aircraft are determined from the ranked cooperative aircraft in an incremental manner. In each round, the cumulative reward value is estimated using the target value network corresponding to the corresponding number of cooperative aircraft, until the cumulative reward value reaches a predetermined threshold. That is, first, the lowest priority aircraft are tried to be scheduled. If the evaluation result (i.e., the cumulative reward value) is less than the threshold, the two lowest priority aircraft are switched to be scheduled, and so on, until the evaluation result is greater than or equal to the threshold. At this point, the scheduling action given by the policy network at the time of stopping is taken.

[0154] For example, such as Figure 5 As shown, at the detection time After the risk of conflict, obtain the time Status information First, the status information Input the target value network that schedules only the lowest priority cooperative aircraft. Output cumulative return value If the cumulative return value is less than the threshold This indicates that scheduling one cooperating aircraft cannot complete the (collision avoidance) task, and in this case, the status information needs to be changed. Input the target value network of the two cooperative aircraft with the lowest scheduling priority. , to obtain cumulative return value Determine if the score is still less than the threshold. This process continues until a cumulative return value greater than or equal to a threshold is found. The algorithm stops when the number of cooperating aircraft is obtained. .

[0155] S7. Based on the state information at the moment when there is a risk of conflict, use the policy network to generate scheduling actions for the cooperative aircraft that need to be scheduled at the moment when there is a risk of conflict.

[0156] Determine the number of aircraft to be dispatched as partners Then, the corresponding policy network is invoked. Output scheduling actions for different cooperating aircraft .

[0157] S8. The schedule is sent to the corresponding cooperating aircraft so that the scheduled aircraft can avoid conflict at times of conflict risk.

[0158] Furthermore, after the conflict is resolved, the flight path should be restored to the predicted path data as soon as possible. Therefore, the methods also include:

[0159] S9. Send a scheduling action to the cooperating aircraft to resume the flight path indicated by the planned path data after avoiding conflict.

[0160] This invention is not limited to the specific embodiments described above. The invention extends to any new feature or combination disclosed in this specification, as well as any new method or process step or combination disclosed herein.

Claims

1. An aircraft conflict prediction method, characterized in that, The method comprises the following steps: S1, acquiring first track data and planned track data of a cooperative aircraft, and acquiring visual data and second track data of a non-cooperative aircraft; S2, correcting the planned track data within a future predetermined time based on the deviation of the first track data and the planned track data in time and position to obtain predicted track data; S3, enhancing the visual features of the second track data by using the visual data, identifying the target type of the non-cooperative aircraft according to the visual features, the target type including a track predictable target and a track unpredictable target, and predicting the future navigation area of different non-cooperative aircrafts according to different target types, comprising: S31, for the non-cooperative aircraft of the track predictable target type, predicting the track data within a future predetermined time as the future navigation area according to the historical track data, and re-predicting the track data within a future predetermined time when the position deviation between the actual track data and the predicted track data reaches a first threshold, or the time from the last predicted track data update reaches a second threshold; S32, for the non-cooperative aircraft of the track unpredictable target type, taking the observed position as the center of a sphere, and defining the future navigation area of the non-cooperative aircraft with a predetermined radius, and re-updating the navigation area when the distance between the observed position and the center position of the sphere reaches a third threshold, or the time from the last predicted track data update reaches a second threshold; S4, predicting whether a conflict will occur based on the minimum distance between the predicted track data of the cooperative aircraft, or whether the minimum distance between the predicted track data of the cooperative aircraft and the future navigation area of the non-cooperative aircraft exceeds a safety distance, wherein different target types of non-cooperative aircrafts have different safety distances.

2. The aircraft conflict prediction method of claim 1, wherein, The method for acquiring the first track data comprises: S11, acquiring multi-modal track data of the cooperative aircraft; S12, performing weighted fusion on part or all of the multi-modal track data to obtain the first track data.

3. The aircraft conflict prediction method of claim 1 or 2, wherein, The method for calculating the deviation of the first track data and the planned track data in time and position comprises: S21, calculating the distance between each first track data point position at each time and each planned track data point position at each time within a historical predetermined time, and obtaining the planned track data point position with the shortest distance from each first track data point position at each time from the planned track data; S22, determining the time deviation of the first track data and the planned track data by the time offset of each planned track data point position and the corresponding first track data point position, and determining the position deviation of the first track data and the planned track data by the position offset of each planned track data point position and the corresponding first track data point position.

4. The aircraft conflict prediction method of claim 3, wherein, The method for correcting the planned track data within a future predetermined time comprises: S23, determining the correction time range of the planned track data according to the time deviation. S24, for each time point in the modified time range, based on the position of the planned flight path data point offset by the time deviation at the time point, the position deviation is used for correction to obtain the modified planned flight path data point at the time point.

5. An aircraft collision avoidance method, characterized by, Comprise: Using the aircraft conflict prediction method of any one of claims 1-4 to predict whether a conflict will occur, when an aircraft conflict is predicted, the following steps are performed: S5, determine the cooperative aircrafts with conflict risk in the cooperative aircrafts, and sort all the cooperative aircrafts according to the priority indicated by the planned flight path data in the principle of cooperative aircrafts with conflict risk in front and cooperative aircrafts without conflict risk behind; S6, based on the state information at the time point with conflict risk, according to the scheduling order from low to high priority, the target value network is used to determine the cooperative aircrafts that need to be scheduled; S7, based on the state information at the time point with conflict risk, the scheduling action for the cooperative aircrafts that need to be scheduled at the time point with conflict risk is generated by using the strategy network; S8, send the scheduling to the corresponding cooperative aircrafts, so that the scheduled aircrafts avoid conflict at the conflict risk time point.

6. The aircraft collision avoidance method of claim 5, wherein, Also include: S9, send the scheduling action to the scheduled cooperative aircrafts to restore to the flight path indicated by the planned flight path data after avoiding conflict.

7. An aircraft collision avoidance method as claimed in claim 5 or 6, characterised in that, The step S6 comprises: Based on the state information at the time point with conflict risk, a corresponding number of cooperative aircrafts with the lowest priority are determined from the sorted cooperative aircrafts in the way of increasing the number round by round, and the cumulative return value is estimated by using the target value network corresponding to the scheduling of the corresponding number of cooperative aircrafts in each round, until the cumulative return value reaches a predetermined threshold.

8. The aircraft collision avoidance method of claim 7, wherein, The training process of the strategy network and the target value network comprises: Starting from the first time point in the historical observation time, the state information at the current time point is obtained; The maximum probability scheduling action of each cooperative aircraft estimated by different strategy networks under the current state information is obtained; The reward at the current time point and the state information at the next time point after executing the scheduling action are calculated; The first cumulative return value at the current time point is estimated by the state value network; The second cumulative return value at the next time point is estimated by the target value network based on the state information at the next time point; The TD error is calculated based on the reward and the first cumulative return value at the current time point, and the second cumulative return value at the next time point; Based on the change gradient of the TD error and the first cumulative return value with respect to the first network parameter of the state value network, the first network parameter is updated; Based on the updated first network parameter, the second network parameter of the target value network is updated; For each strategy network, based on the change gradient of the TD error and the logarithm of the output result of the strategy network with respect to the third network parameter of the strategy network, the third network parameter is updated.

9. The aircraft collision avoidance method of claim 5 or 6, wherein The state information contains the flight path data of the non-cooperative aircrafts of the flight path predictable target type.

Images