Aircraft obstacle avoidance path planning method and system

Abstract

The invention discloses an aircraft obstacle avoidance path planning method and system, relates to the technical field of deep aircraft path planning, artificial intelligence decision making and the like, and is used for solving the problem of difficult path planning of deep aircrafts. Comprising the following steps: determining a starting point and an ending point of an aircraft based on task parameters, and calculating a three-dimensional distance between the aircraft and an obstacle in real time; a flight path is optimized through mixed integer linear programming, and an obstacle avoidance coefficient is predicted in combination with a gradient enhanced decision tree model. The flight space is graded based on obstacle intensity, communication coverage and navigation precision, an obstacle avoidance difficulty coefficient is calculated, and a new obstacle avoidance coefficient is calculated in combination with the obstacle avoidance coefficient. Dynamic data of obstacles are collected through a laser radar and a historical database, dynamic behaviors of the obstacles are predicted through an LSTM model, and accurate analysis of the moving trend of the obstacles in the future is achieved. The invention provides an efficient and safe deep space aircraft obstacle avoidance path planning scheme which is suitable for a complex dynamic environment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of deep space vehicle path planning and artificial intelligence decision-making, and more specifically, to a vehicle obstacle avoidance path planning method and system. BACKGROUND

[0002] Deep space refers to space outside the earth's atmosphere, such as the moon, Mars, and even space outside the solar system. China has realized sampling from the moon to Mars, which is a specific application of deep space technology. As human exploration of deep space outside the earth deepens, there are more and more vehicles in deep space. China plans to establish an international scientific research station on the moon before 2035 to realize data signal relay, navigation, and remote sensing functions. However, the environment in deep space is complex, such as multiple vehicles flying at the same time and satellites operating at high speed. In order to avoid collision accidents during flight, a reasonable, scientific, and efficient flight route (i.e., path) needs to be designed. The existing flight path planning is mainly based on low altitude on the earth, which cannot better meet the complex environment of deep space, and an intelligent path planning algorithm needs to be designed.

[0003] In existing work, Sichuan University in China proposed an autonomous intelligent obstacle avoidance method for vehicles based on deep deterministic policy gradient, but its complexity is high and is not suitable for deep space with high requirements. Harbin Institute of Technology proposed a method for defensive obstacle avoidance of vehicles during combat based on differential equations, which has lower complexity. The Third University of Toulouse in foreign countries studied the influence of sudden changes in flight environment (such as weather) on vehicle obstacle avoidance. It can be seen that the path planning for vehicle obstacle avoidance in deep space needs to be proposed based on multiple conditions.

[0004] In view of the above problems, the present application provides a solution. SUMMARY

[0005] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present application provide a vehicle obstacle avoidance path planning method and system to solve the problems raised in the background art.

[0006] To achieve the above object, the present application provides the following technical scheme:

[0007] In a preferred embodiment,

[0008] Step 1: determining the starting point and the ending point of the vehicle in deep space, the starting point B and the ending point E being a three-dimensional coordinate value;

[0009] Step 2: calculating the real-time coordinates of the obstacles and the vehicle through the orbit parameters, using the Euclidean distance to determine the nearest obstacle, and providing basic data for the migration learning to calculate the obstacle avoidance coefficient Pb;

[0010] Step 3: Solve the obstacle avoidance coefficient Pb by transfer learning, and determine the obstacle avoidance difficulty coefficient by the obstacle density, communication coverage and navigation accuracy of different flight spaces, and optimize the obstacle avoidance coefficient to obtain a new obstacle avoidance coefficient by combining the obstacle avoidance difficulty coefficient;

[0011] Step 4: Predict the dynamic behavior of the obstacle by machine learning, set a unit time Tt, and analyze the changing trajectory of the dynamic behavior of the obstacle and the distance between the aircraft and the obstacle to obtain the optimal solution in the unit time Tt.

[0012] In a preferred embodiment, the task parameters of the input aircraft are input, including: the initial position of the starting point three-dimensional coordinates, the target terminal point; and the three-dimensional coordinates of all reference objects in the deep space environment are obtained and converted into a unified ICRF reference.

[0013] In a preferred embodiment, the obstacle P is found by predictability, and the three-dimensional distance between the obstacle P and the aircraft is calculated according to the satellite's running track and period, and the specific steps are as follows:

[0014] Obtain the orbital data of the obstacle P, including:

[0015] Semi-major axis a;

[0016] Eccentricity e;

[0017] Inclination i;

[0018] Longitude of ascending node Ω;

[0019] Argument of perigee ω;

[0020] Mean anomaly M;

[0021] Obtain the trajectory parameters of the aircraft, including:

[0022] Starting point coordinates (x p ,y q ,z q ) and terminal point coordinates (x z ,y z ,z z );

[0023] Flight time S;

[0024] Velocity vector (V1, V2, V3);

[0025] According to the orbital data of the obstacle, the real-time three-dimensional coordinates (x p ,y p ,z p ) of the obstacle are calculated by establishing an orbital equation, and the specific calculation formula is: Where r represents the distance between obstacle p and the center of the track, and v represents the true anomaly angle; the real-time three-dimensional coordinates of obstacle P are calculated according to the following formula:

[0026] x p =r(cos(Ω)cos(ω+ν)-sin(Ω)sin(ω+ν)cos(i));

[0027] y p =r(sin(Ω)cos(ω+ν)+cos(Ω)sin(ω+ν)cos(i));

[0028] z p = r(sin(ω+ν)sin(i));

[0029] Based on the aircraft's trajectory parameters, the real-time three-dimensional coordinates (x3, y3, z3) of the aircraft are calculated using kinematic formulas, specifically: x3 = V1 × x q y3=V2×y q z3=V3×z q ;

[0030] The distance D between the aircraft and the obstacle P is calculated using Euclidean distance, specifically based on the formula:

[0031] Calculate the Euclidean distance between the three-dimensional coordinates of all aircraft in space and sample data for transfer learning. The flight path planning for aircraft to avoid obstacles from the starting point B to the ending point E is a mixed integer linear programming problem. Its optimization objective and constraints are:

[0032] Minimize flight distance:

[0033] Minimize obstacle avoidance coefficient and flight energy consumption:

[0034] Maximize obstacle avoidance space:

[0035] Where Pb is the obstacle avoidance coefficient, which is related to the size of the obstacle, its speed, and its distance from the aircraft. Its specific value is calculated by an artificial intelligence algorithm. i This is the energy consumed by A when avoiding obstacles. It is the historical average consumption value;

[0036] constraint:

[0037] The endpoint coordinates are much larger than the starting coordinates: E >> B

[0038] The entire flight was powered by POWER. B >POWER E >0

[0039] The coordinates of point A and the coordinates of point P are completely identical in the same time slot:

[0040] The entire flight process A starts from E and flies continuously to S: POWER E >>{POWER1,POWER2,…,POWER i} p >>POWER B With D A-B →∞,D A-E ≈0,D A-P →{0,∞}

[0041] Its coordinates change in different time slots.

[0042] In a preferred embodiment, a gradient-enhanced decision tree is selected as the core model, and training data is loaded from a historical flight mission database, including:

[0043] The distance d between the obstacle and the aircraft;

[0044] The size of the obstacle is ss;

[0045] The relative velocity v between the obstacle and the aircraft;

[0046] The angle θ between the obstacle and the aircraft's path;

[0047] For the loaded training data, extract and construct the feature matrix X and the target vector Y: Feature matrix X = [d,s,v,θ]; Target vector Y = [Pb]; Standardize the feature matrix X according to the formula: Where μ is the mean of the feature and σ is the standard deviation of the feature;

[0048] Real-time orbital data is acquired via CelesTrak, and the spatial environment data fed back by this data is used to determine the obstacle density in the area surrounding the spacecraft, dividing the flight space into three levels.

[0049] Low-density space: There are fewer obstacles, and they are sparsely distributed;

[0050] Medium-density space: Obstacles are moderately distributed, and there are some moving satellites or particles;

[0051] High-density space: obstacles are very dense, with a large number of satellites, floating particles, etc. in high-orbit or low-orbit regions;

[0052] Based on the positioning information of the spacecraft's orbit and the signal interaction between the ground station and the satellite, and according to the communication coverage and navigation accuracy within the flight space, the flight space is divided into two levels:

[0053] High-coverage, high-precision space: strong communication signals, accurate navigation systems

[0054] Low coverage and low precision space: weak communication signals and low navigation accuracy.

[0055] In a preferred embodiment, the obstacle avoidance difficulty coefficient for each level of space is calculated based on the obstacle density, communication coverage, and navigation accuracy of different flight spaces. The specific steps are as follows:

[0056] Step S1: Calculate the obstacle density factor ZMyz, specifically based on the formula: ZMyz = Nz / Area, where Nz represents the number of obstacles in the area and Area represents the area where the aircraft is located; Calculate the communication coverage factor TXyz, specifically based on the formula: TXyz = 1 - (TXzl / TXZLmax), where TXzl is the current communication quality of the aircraft and TXZLmax represents the maximum communication quality; Calculate the navigation accuracy factor DHyz, specifically based on the formula: DHyz = 1 - (DHjd / DHJDmax), where DHjd represents the current navigation accuracy and DHJDmax represents the maximum navigation accuracy.

[0057] Step S2: Calculate the obstacle avoidance difficulty coefficient ODF by weighted combination based on the above factors, specifically according to the formula: ODF=Qz×ZMyz+Qx×TXyz+Qd×DHyz, where Qz represents the weight of the obstacle density factor, Qx represents the weight of the communication coverage factor, and Qd represents the weight of the navigation accuracy factor.

[0058] The obstacle avoidance coefficient Pb and the obstacle avoidance difficulty coefficient ODF are linearly combined with fixed weights, specifically according to the formula: New-Pb=Qbz×Pb+Qbx×ODF; where New-Pb is the new obstacle avoidance coefficient, Qbz is the weight of the obstacle avoidance difficulty coefficient, and Qbx is the weight of the obstacle avoidance coefficient; and a nonlinear adjustment function F(Pb, ODF) is introduced.

[0059] The new obstacle avoidance coefficient is calculated by combining the linear weighting part and the nonlinear adjustment part, specifically according to the formula: New-Pb=Qbz×Pb+Qbx×ODF+γ×F(Pb,ODF), where the weight γ is used to control the influence of nonlinear adjustment on the overall result.

[0060] In a preferred embodiment, obstacle information data is collected using lidar sensors on the aircraft and a historical database, including:

[0061] Historical position: The current and past coordinates (xTt, yTt, zTt) of the obstacle;

[0062] Historical velocity: The current and past velocities of the obstacle (vx, vy, vz);

[0063] Rate of change of direction: Calculate the rate of change of the obstacle's direction Δθ / ΔTt at each time step;

[0064] Environmental factors: including external influences such as wind speed and air pressure;

[0065] The LSTM model based on time-series data in machine learning is used to predict the dynamic behavior of obstacles. The collected obstacle information data is input into the model, and the input data is preprocessed by normalization and time window division.

[0066] The LSTM model is trained based on historical data, and the mean squared error is used as the loss function to measure the model's prediction accuracy. The model's performance on different datasets is evaluated through k-fold cross-validation.

[0067] Based on obstacle information data, the change in obstacle velocity vector is calculated, and the velocity change rate and direction change rate are extracted for each time step; the aircraft's velocity vector vf is then input into the model.

[0068] In a preferred embodiment, basic information about the aircraft and obstacles is obtained using lidar, including:

[0069] Obstacle position: The three-dimensional position of the obstacle (xm, ym, zm);

[0070] Obstacle velocity: The velocity of an obstacle is a vector (vx, vy, vz);

[0071] The speed of the aircraft is vf = (vfx, vfy, vfz);

[0072] The current position of the aircraft (xdq, ydq, zdq);

[0073] Given a time interval T, the relative positions of the aircraft and the obstacle for any unit time T are: (xm(Tt)-xdq(Tt), ym(Tt)-ydq(Tt), zm(Tt)-zdq(Tt)). Calculate the distance d(T) between the aircraft and the obstacle at each unit time T, using the following formula:

[0074] The future trajectory of an obstacle is predicted based on its current velocity and direction changes, specifically using the formula: xm(Tt)=xm(0)+v x ×Tt;ym(Tt)=ym(0)+v y ×Tt;zm(Tt)=zm(0)+v z ×Tt, where (v x vy v z () represents the velocity vector of the obstacle;

[0075] The future trajectory is predicted based on the aircraft's current speed and direction changes, specifically using the formula: xdq(Tt)=xdq(0)+v fx ×Tt;ydq(Tt)=ydq(0)+v fy ×Tt;zdq(Tt)=zdq(0)+v fz ×Tt, where (v fx v fy v fz () represents the velocity vector of the aircraft;

[0076] Substituting the positions of the aircraft and obstacles into the distance formula, we can obtain the unit time T(max) by differentiating the distance function d(T) and finding the extreme point. The second derivative test can verify that the obtained T(max) is the maximum value.

[0077] In a preferred embodiment, it includes: a data acquisition module, a path analysis module, a data storage module, and a future prediction module, with signal connections between the modules;

[0078] The data acquisition module is used to acquire the three-dimensional coordinates of the aircraft's starting and ending points;

[0079] The path analysis module is mainly used to analyze and optimize the flight path of an aircraft by combining the status of obstacles and the status of the aircraft.

[0080] The future prediction module is mainly used to analyze the future dynamic behavior of obstacles and the state of the aircraft, optimize the avoidance route, and determine the best solution by analyzing the trajectory of the obstacle's movement speed and direction and the distance between the aircraft and the obstacle within a unit of time Tt.

[0081] The data storage module is used to store all data processed by the platform.

[0082] This invention discloses a method and system for obstacle avoidance path planning for aircraft, relating to the technical fields of deep-space aircraft path planning and artificial intelligence decision-making, and is used to solve the problem of difficult path planning faced by deep-space aircraft. The method includes the following steps: determining the starting and ending points of the aircraft based on mission parameters, and calculating the three-dimensional distance between the aircraft and obstacles in real time; optimizing the flight path through mixed-integer linear programming, and predicting obstacle avoidance coefficients using a gradient-enhanced decision tree model; calculating the obstacle avoidance difficulty coefficient based on obstacle density, communication coverage, and navigation accuracy-leveled flight space, and combining it with the obstacle avoidance coefficient to calculate a new obstacle avoidance coefficient; collecting dynamic obstacle data through lidar and historical databases, and using an LSTM model to predict the dynamic behavior of obstacles, achieving accurate analysis of future obstacle movement trends. This invention provides an efficient and safe obstacle avoidance path planning scheme for deep-space aircraft, suitable for complex dynamic environments. Attached Figure Description

[0083] Figure 1 This is a schematic diagram of the structure of an obstacle avoidance path planning system for an aircraft according to the present invention.

[0084] Figure 2 This is a schematic diagram illustrating the operation of an obstacle avoidance path planning method for aircraft according to the present invention. Detailed Implementation

[0085] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0086] Example

[0087] This invention discloses an obstacle avoidance path planning method for aircraft, such as... Figure 2 As shown, the specific steps are as follows:

[0088] Step 1: Determine the starting point and ending point of the spacecraft in deep space. The starting point B and the ending point E are three-dimensional coordinate values.

[0089] 1. Input the spacecraft's mission parameters using different data input formats, including: the starting point's three-dimensional coordinates (x1, y1, z1), and the initial position of the destination target. It should be noted that the input mission parameters must conform to the deep space coordinate system. For example, in this embodiment, spacecraft A's mission requires it to travel from near Earth's orbit to Ganymede. The mission parameters are input in JSON format as follows:

[0090] Starting point: (7000, -1000, 3000).

[0091] Final destination: Ganymede.

[0092] 2. Obtain the 3D coordinates of all reference objects in the deep space environment and correctly calculate the relative positional relationship between the starting and ending points. It should be noted that if the coordinate systems of the spacecraft's location or the target area are different, they need to be converted to a unified ICRF reference. For example:

[0093] Ganymede's orbital parameters are expressed in Jupiter's central coordinate system:

[0094] Semi-long wheelbase: 1,070,412 km

[0095] Track inclination: 0.204°

[0096] Orbital period: 7.15 days

[0097] Through orbit calculation and coordinate transformation, the current position of Ganymede in ICRF is obtained as (678200, -150000, 292500).

[0098] Orbit calculation and coordinate transformation are both conventional techniques and will not be elaborated upon here.

[0099] Step 2: Divide the entire flight time T of the aircraft into several time slots s. A Assume that the flight of the aircraft is stable within a time slot.

[0100] In the deep space network, all objects that may affect the spacecraft are called obstacles P = {p(1), p(2), ..., p(n)}, where n is the total number of obstacles (e.g., satellites in operation, space stations, floating particles, etc.). In this embodiment, spacecraft A detects obstacles P and their distances using the following method:

[0101] Obstacle P is detected through predictability, and the three-dimensional distance between obstacle P and spacecraft A is calculated using the satellite's orbit and period. The specific steps are as follows:

[0102] 1. Obtain the trajectory data of obstacle P, including:

[0103] Semi-major axis a (size of the track)

[0104] Eccentricity e (shape of the orbit)

[0105] Inclination angle i (the angle between the track and the reference plane)

[0106] Longitude Ω of the ascending node (direction of the intersection of the orbital plane and the reference plane)

[0107] Perigee argument ω (the position of the satellite's closest point)

[0108] Horizontal anomaly M (the satellite's position in orbit)

[0109] Obtain the trajectory parameters of the aircraft, including:

[0110] Starting coordinates (x) q ,y q ,z q ) and endpoint coordinates (x z ,y z ,z z )

[0111] Flight time S

[0112] Velocity vector (V1, V2, V3)

[0113] Based on the obstacle's trajectory parameters, the real-time three-dimensional coordinates (x, y, z) of the obstacle are calculated by establishing trajectory equations. p ,y p ,z p The specific formula for calculating the orbital position is as follows: Where r represents the distance between the obstacle and the center of the track, and v represents the true anterior angle; the real-time three-dimensional coordinates of obstacle P are calculated according to the following formula:

[0114] x p =r(cos(Ω)cos(ω+ν)-sin(Ω)sin(ω+ν)cos(i));

[0115] y p =r(sin(Ω)cos(ω+ν)+cos(Ω)sin(ω+ν)cos(i));

[0116] z p =r(sin(ω+ν)sin(i)).

[0117] Based on the aircraft's trajectory parameters, the real-time three-dimensional coordinates (x3, y3, z3) at time T are calculated using kinematic formulas, specifically: x3 = V1 × x q y3=V2×y q z3=V3×z q For example: Sample data is as follows:

[0118] Orbital parameters of obstacle P:

[0119] Semi-major axis a = 7000 km

[0120] Eccentricity e = 0.01

[0121] Inclination angle i = 50°

[0122] Ascending node longitude Ω = 120°

[0123] Perigee angle ω = 90°

[0124] The angle of approach to the point M = 0° (at the initial moment).

[0125] Trajectory parameters of aircraft A:

[0126] Starting coordinates: (0, 6378, 0) km

[0127] Flight speed: (7.8,0,0)km / s

[0128] Current time: 100 seconds after the start of the mission.

[0129] Substituting the example data into the calculation, the three-dimensional coordinates of obstacle P are (4212, 5151, 3211), and the real-time three-dimensional coordinates of aircraft A are (780, 6378, 0).

[0130] 2. In this embodiment, the distance between aircraft A and obstacle P is calculated using Euclidean distance. That is, when the three-dimensional coordinates of aircraft A are (x1, y1, z1) and the three-dimensional coordinates of obstacle P are (x2, y2, z2), the Euclidean distance is:

[0131] Calculate the Euclidean distance between the three-dimensional coordinates of all spacecraft A within the space and sample the data for transfer learning. Note that during the first sampling, select the sample with the smallest Euclidean distance to spacecraft A.

[0132] Based on the above, the obstacle avoidance flight path planning for aircraft A from the starting point B to the ending point E is a mixed-integer linear programming problem. Its optimization objective and constraints are:

[0133] Minimize flight distance:

[0134] Minimize obstacle avoidance coefficient and flight energy consumption:

[0135] Maximize obstacle avoidance space:

[0136] Where Pb is the obstacle avoidance coefficient, which is related to the size of the obstacle, its speed, and its distance from A. Its specific value is calculated by an artificial intelligence algorithm. POWER i This refers to the energy consumed by A during obstacle avoidance. Ascent and sharp turns consume more energy than descent, and excessive energy consumption is detrimental to A's long-term flight. This is the historical average consumption value.

[0137] constraint:

[0138] The endpoint coordinates are much larger than the starting coordinates: E >> B;

[0139] The entire flight was powered by POWER. B >POWER E >0;

[0140] The coordinates of point A and the coordinates of point P are completely identical in the same time slot:

[0141] The entire flight process A starts from E and flies continuously to S: POWER E >>{POWER1,POWER2,…,POWER i} p >>POWER B With D A-B →∞,D A-E ≈0,D A-P →{0,∞};

[0142] Its coordinates change in different time slots: For example, by substituting the sample data into the calculation, the Euclidean distance between aircraft A and obstacle P is 5000km; the Euclidean distance between aircraft A and floating particle cloud P2 is 550km. The result is that floating particle cloud P2 is closest to the aircraft and should be avoided first.

[0143] Step 3: Solve for the obstacle avoidance coefficient Pb through transfer learning;

[0144] Specifically, the calculation process is as follows:

[0145] Step 3.1: Model Construction: To effectively predict the obstacle avoidance coefficient Pb, this embodiment uses the Gradient Boosting Decision Tree (XGBoosT) as the core model. First, training data is loaded from the historical flight mission database, including:

[0146] The distance d between the obstacle and the aircraft

[0147] obstacle size ss

[0148] The relative speed v between the obstacle and the aircraft

[0149] The angle θ between the obstacle and the aircraft's path

[0150] For the loaded data, extract and construct the feature matrix X and the target vector Y: Feature matrix X = [d, s, v, θ]; Target vector Y = [Pb]; To avoid model bias caused by differences in the range of different feature values, the feature matrix X is standardized. Specifically, based on the formula: Where μ is the mean of the feature and σ is the standard deviation of the feature. For example, if the range of d is [100, 10000] km, the mean μd is 5050 km, the standard deviation σd is 2900 km, and after normalization, dnorm∈[-1.4,1.7];

[0151] By setting Pb as the prediction target, the model optimizes the loss function by successively fitting the nonlinear relationship between the features and the obstacle avoidance coefficient, so as to minimize the error between the predicted value P^b and the actual value Pb.

[0152] Step 3.2: Application of Transfer Learning Model: Since the specific environment of each deep space mission differs, transfer learning is required for the trained model. The transfer process consists of two parts: freezing and fine-tuning. The former preserves the parameters of the general feature layer, adapting to common feature patterns in historical data, such as the impact of obstacle distance on the threat; the latter fine-tunes the model using current mission data to enhance its adaptability to specific environments. For example, two obstacles are detected in real time: a floating particle cloud P2 and a small satellite P3. P2 has the characteristics of a distance of 550km, a size of 100, and a relative speed of 2km / s. The model predicts Pb = 0.87, indicating priority avoidance. In contrast, P3 has Pb = 0.45, indicating a lower threat that can be ignored.

[0153] It should be noted that in the obstacle avoidance path planning of aircraft, the flight space is not singular. In this embodiment, different flight spaces are divided into categories according to different obstacle densities, different communication coverage and navigation accuracy. The obstacle avoidance difficulty coefficient of each category of flight space is determined according to the different obstacle densities, communication coverage and navigation accuracy of different flight spaces. The new obstacle avoidance coefficient of each category of flight space is determined by combining the obstacle avoidance coefficient and the obstacle avoidance difficulty coefficient analysis.

[0154] Specifically, obstacle density is typically determined by the number and distribution of obstacles in the area surrounding the aircraft. In this embodiment, real-time orbital data is acquired via CelesTrak, and the spatial environment data fed back from the real-time orbital data is used to determine the obstacle density in the area surrounding the aircraft. Based on the obstacle density within the flight space, the flight space is divided into three levels.

[0155] Low-density space: There are fewer obstacles, and they are sparsely distributed;

[0156] Medium-density space: Obstacles are moderately distributed, and there are some moving satellites or particles;

[0157] High-density space: obstacles are very dense, with a large number of satellites, floating particles, etc. in high-orbit or low-orbit regions;

[0158] Based on the positioning information of the spacecraft's orbit and the signal interaction between the ground station and the satellite, the communication signal coverage and navigation accuracy are assessed. Based on the communication coverage and navigation accuracy within the flight space, the flight space is divided into two levels:

[0159] High-coverage, high-precision space: strong communication signals and accurate navigation systems (such as near-Earth orbit or low-Earth orbit).

[0160] Low-coverage, low-precision space: weak communication signals and low navigation accuracy (such as deep space regions).

[0161] The obstacle avoidance difficulty coefficient for each level of space is calculated based on the obstacle density, communication coverage, and navigation accuracy in different flight spaces. The specific steps are as follows:

[0162] Step S1: Calculate the obstacle density factor ZMyz, specifically based on the formula: ZMyz = Nz / Area, where Nz represents the number of obstacles in the area and Area represents the area where the aircraft is located; Calculate the communication coverage factor TXyz, specifically based on the formula: TXyz = 1 - (TXzl / TXZLmax), where TXzl is the current communication quality of the aircraft and TXZLmax represents the maximum communication quality; Calculate the navigation accuracy factor DHyz, specifically based on the formula: DHyz = 1 - (DHjd / DHJDmax), where DHjd represents the current navigation accuracy and DHJDmax represents the maximum navigation accuracy.

[0163] Step S2: Based on the above factors, calculate the obstacle avoidance difficulty coefficient ODF using a weighted combination method, specifically according to the formula: ODF = Qz × ZMyz + Qx × TXyz + Qd × DHyz, where Qz represents the weight of the obstacle density factor, Qx represents the weight of the communication coverage factor, and Qd represents the weight of the navigation accuracy factor. For example: In a specific flight space with 100 obstacles and an area of ​​1000 km², the density factor is calculated to be 0.1; the communication quality is 80% and the maximum quality is 100%, resulting in a communication factor of 0.2; the navigation accuracy is 90% and the maximum accuracy is 100%, resulting in a navigation factor of 0.1. The calculated weights for each factor are Qz (0.3), Qx (0.45), and Qd (0.25). Substituting the example data into the calculation, the obstacle avoidance difficulty coefficient ODF is 0.145.

[0164] The obstacle avoidance performance of the aircraft is evaluated by weighting and nonlinear adjustment, and a new obstacle avoidance coefficient (New-Pb) is calculated. The obstacle avoidance performance of the aircraft includes: the obstacle avoidance coefficient Pb, which describes the threat level of the obstacle itself; and the obstacle avoidance difficulty coefficient ODF, which focuses on the overall complexity of the flight environment. The obstacle avoidance coefficient Pb and the obstacle avoidance difficulty coefficient ODF are weighted by Qbz and Qbx to ensure that the influence of different factors in the comprehensive analysis is consistent with the actual scenario requirements.

[0165] Specifically, the obstacle avoidance coefficient Pb and the obstacle avoidance difficulty coefficient ODF are linearly combined with fixed weights, according to the formula: New-Pbxx=Qbz×Pb+Qbx×ODF; where New-Pbxx is the new obstacle avoidance coefficient, Qbz is the weight of the obstacle avoidance difficulty coefficient, and Qbx is the weight of the obstacle avoidance coefficient.

[0166] It should be noted that when both Pb and ODF are large, the obstacle avoidance risk of the aircraft increases sharply. This nonlinear relationship cannot be accurately expressed by linear weighting. Therefore, a nonlinear adjustment function F(Pb, ODF) is introduced, which can be selected in different forms according to the specific scenario. For example, the product type is: F(Pb, ODF) = Pb × ODF.

[0167] The new obstacle avoidance coefficient is calculated by combining the linear weighting part and the nonlinear adjustment part, specifically according to the formula: New-Pb=Qbz×Pb+Qbx×ODF+γ×F(Pb,ODF), where the weight γ is used to control the influence of nonlinear adjustment on the overall result. For example, in a flight environment with an obstacle density of 0.1, communication coverage of 80%, and navigation accuracy of 90%, the aircraft needs to avoid an obstacle with a size of 500 cubic meters, a speed of 5 km / s, and an angle of 30° with the aircraft's trajectory. Substituting the example data into the calculation, the obstacle avoidance coefficient of the aircraft is 0.267, and the obstacle avoidance difficulty coefficient ODF of the flight space is 0.13. Through calculation, the weight Qbz of the obstacle avoidance difficulty coefficient is 0.5, the weight Qbx of the obstacle avoidance coefficient is 0.3, and the adjustment weight γ is 0.2. Substituting the example data into the formula, the linear part of the new obstacle avoidance coefficient New-Pbxx = 0.1869, the nonlinear adjustment part F(Pb, ODF) = 0.03471, and finally the new obstacle avoidance coefficient New-Pb = 0.193842.

[0168] Step 4: Predict obstacle dynamics through machine learning, including obstacle speed and direction data, and combine obstacle dynamics with aircraft flight speed analysis to predict obstacle movement in advance so as to achieve better obstacle avoidance.

[0169] It should be noted that the speed and direction of the obstacle will not remain constant. In this embodiment, a unit time Tt is set, and the optimal solution is obtained by analyzing two factors: the trajectory of the obstacle's changing speed and direction, and the distance between the aircraft and the obstacle. The solution is the combination of the predicted obstacle dynamic behavior and the aircraft's flight speed within the unit time Tt.

[0170] Step Q1: Collect obstacle information data using the lidar sensors on the aircraft and historical databases, including:

[0171] Historical position: The current and past coordinates (xTt, yTt, zTt) of the obstacle.

[0172] Historical velocity: The current and past velocities of the obstacle (vx, vy, vz).

[0173] Rate of change of direction: Calculate the rate of change of the obstacle's direction Δθ / ΔTt at each time step.

[0174] Environmental factors: including external influences such as wind speed and air pressure.

[0175] In this embodiment, the dynamic behavior of obstacles is predicted based on the time-series data modeling LSTM in the machine learning model. The collected obstacle information data is input into the model, and the input data is preprocessed by normalization and time window division.

[0176] Step Q2: Train the LSTM model using historical data. During training, mean squared error (MSE) is used as the loss function to measure the model's prediction accuracy, and k-fold cross-validation is used to evaluate the model's performance on different datasets to ensure the model's generalization ability.

[0177] Once model training is complete, it predicts the dynamic behavior of obstacles in the near future based on the latest real-time sensor data. For example, the following obstacle information data was obtained from LiDAR and cameras:

[0178]

[0179] Aircraft status: Current speed of the aircraft: vf = (3,2) m / s; Target position of the aircraft: (100, 100);

[0180] Based on obstacle information data, the change in the obstacle velocity vector is calculated, and the velocity change rate and direction change rate at each time step are extracted by calculating the historical position and velocity of the obstacle. The velocity vector vf of the aircraft is input into the model to determine the relative velocity between the aircraft and the obstacle.

[0181] It should be noted that all position and velocity data need to be normalized here to ensure that the data are compared within the same range, thus preventing any single feature from having an excessive impact on the model.

[0182] The normalized obstacle information data is input into the model. The model learns the movement patterns of the obstacles based on historical data and continuously adjusts the weights to reduce errors. For example, when time Tt = 5 seconds, the model predicts that the obstacle will be located at (12, 18) in the next second.

[0183] During flight, both the aircraft and obstacles are constantly changing. Therefore, predicting the dynamic behavior of obstacles requires considering the most suitable time to ensure effective obstacle avoidance. In this embodiment, a unit time Tt is set, and the dynamic behavior of the obstacle is predicted within the unit time T based on the changing trajectory of the obstacle's moving speed and direction, and the distance between the aircraft and the obstacle. This prediction is the most effective way to achieve obstacle avoidance.

[0184] Specifically, basic information about the aircraft and obstacles is obtained through lidar, including:

[0185] Obstacle position: The three-dimensional position of the obstacle (xm, ym, zm);

[0186] Obstacle velocity: The velocity of an obstacle is a vector (vx, vy, vz);

[0187] The speed of the aircraft is vf = (vfx, vfy, vfz);

[0188] The current position of the aircraft (xdq, ydq, zdq);

[0189] Given a time interval T, the relative positions of the aircraft and the obstacle for any unit time T are: (xm(Tt)-xdq(Tt), ym(Tt)-ydq(Tt), zm(Tt)-zdq(Tt)). Calculate the distance d(T) between the aircraft and the obstacle at each unit time T, using the following formula:

[0190] The future trajectory of an obstacle is predicted based on its current velocity and direction changes, specifically using the formula: xm(Tt)=xm(0)+v x ×Tt;ym(Tt)=ym(0)+v y ×Tt;zm(Tt)=zm(0)+v z ×Tt, where (v x v y v z ) represents the velocity vector of the obstacle.

[0191] The future trajectory is predicted based on the aircraft's current speed and direction changes, specifically using the formula: xdq(Tt)=xdq(0)+v fx ×Tt;ydq(Tt)=ydq(0)+v fy ×Tt;zdq(Tt)=zdq(0)+v fz ×Tt, where (v fx v fy v fz ) represents the velocity vector of the aircraft.

[0192] Substituting the aircraft's position and the obstacle's position into the distance formula, we obtain the unit time T(max) by differentiating the distance function d(T) and finding the extreme point. The second derivative test verifies that the calculated T(max) is the maximum value. For example, sample data is as follows:

[0193] Obstacle: Initial position (xm(0), ym(0)) = (0, 0), velocity vm = (2, 2) m / s

[0194] The initial position of the aircraft is (xdq(0), ydq(0)) = (10, 10), and its velocity is vdq = (3, 1) m / s.

[0195] Substituting the performance data into the formula, we obtain the unit time Tt(max) = 4 for the maximum distance, which means that the distance between the aircraft and the obstacle is the maximum at the 4th second.

[0196] It should be noted that the above are all conventional technical methods, and will not be elaborated upon here.

[0197] This invention also relates to an obstacle avoidance path planning system for aircraft, such as... Figure 1 As shown, it includes: a data acquisition module, a path analysis module, a data storage module, and a future prediction module, with signal connections between each module.

[0198] The data acquisition module is used to acquire the three-dimensional coordinates of the aircraft's starting and ending points, and then...

[0199] The path analysis module is mainly used to analyze and optimize the flight path of an aircraft by combining the information of obstacles and the aircraft's status.

[0200] The future prediction module is mainly used to analyze the future dynamic behavior of obstacles and the aircraft's state, optimize avoidance routes, and determine the optimal solution by analyzing the changes in the obstacle's speed and direction and the distance between the aircraft and the obstacle within a unit of time Tt.

[0201] The data storage module is used to store all data processed by the platform.

[0202] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0203] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0204] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and inventive constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0205] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0206] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0207] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for obstacle avoidance path planning for aircraft. Its characteristics include: Step 1: Determine the starting and ending points of the spacecraft in deep space. The starting point B and the ending point E are three-dimensional coordinate values. Step 2: Calculate the real-time coordinates of obstacles and the aircraft using orbital parameters, and use Euclidean distance to determine the nearest obstacle, providing basic data for calculating the obstacle avoidance coefficient Pb through transfer learning; Step 3: Solve the obstacle avoidance coefficient Pb through transfer learning, and determine the obstacle avoidance difficulty coefficient based on the different obstacle density, communication coverage and navigation accuracy in different flight spaces. Combine the obstacle avoidance difficulty coefficient to optimize the obstacle avoidance coefficient and obtain a new obstacle avoidance coefficient. Step 4: Predict the dynamic behavior of obstacles through machine learning. Set a unit time Tt. Analyze the changes in the dynamic behavior of obstacles and the distance between the aircraft and the obstacles to find the optimal solution that is predicted within the unit time Tt.

2. The obstacle avoidance path planning method for an aircraft according to claim 1, characterized in that: Input the mission parameters of the spacecraft, including: the starting point's three-dimensional coordinates and the initial position of the destination target; and obtain the three-dimensional coordinates of all reference objects in the deep space environment and convert them into a unified ICRF benchmark.

3. The obstacle avoidance path planning method for an aircraft according to claim 2, characterized in that: The obstacle P is detected through predictability, and the three-dimensional distance between the obstacle P and the spacecraft is calculated based on the satellite's orbit and period. The specific steps are as follows: Obtain the trajectory data of obstacle P, including: Semi-major axis a; Eccentricity e; Inclination angle i; Longitude of ascending node (Ω); Perimeter argument ω; The angle M is the closest to the target point. Obtain the trajectory parameters of the aircraft, including: Starting coordinates (x) q ,y q ,z q ) and endpoint coordinates (x z ,y z ,z z ); Flight time S; Velocity vector (V1, V2, V3); Based on the obstacle's trajectory data, the real-time three-dimensional coordinates (x, y, z) of the obstacle are calculated by establishing trajectory equations. p ,y p ,z p The specific calculation is based on the following formula: Where r represents the distance between obstacle p and the center of the track, and v represents the true anomaly angle; the real-time three-dimensional coordinates of obstacle P are calculated according to the following formula: x p =r(cos(Ω)cos(ω+ν)-sin(Ω)sin(ω+ν)cos(i)); y p =r(sin(Ω)cos(ω+ν)+cos(Ω)sin(ω+ν)cos(i)); z p =r(sin(ω+ν)sin(i)); Based on the aircraft's trajectory parameters, the real-time three-dimensional coordinates (x3, y3, z3) of the aircraft are calculated using kinematic formulas, specifically: x3 = V1 × x q y3=V2×y q z3=V3×z q ; The distance D between the aircraft and the obstacle P is calculated using Euclidean distance, specifically based on the formula: Calculate the Euclidean distance between the three-dimensional coordinates of all aircraft in space and sample data for transfer learning. The flight path planning for aircraft to avoid obstacles from the starting point B to the ending point E is a mixed integer linear programming problem. Its optimization objective and constraints are: Minimize flight distance: Minimize obstacle avoidance coefficient and flight energy consumption: Maximize obstacle avoidance space: Where Pb is the obstacle avoidance coefficient, which is related to the size of the obstacle, its speed, and its distance from the aircraft. Its specific value is calculated by an artificial intelligence algorithm. i This is the energy consumed by A when avoiding obstacles. It is the historical average consumption value; constraint: The endpoint coordinates are much larger than the starting coordinates: E >> B The entire flight was powered by POWER. B >POWER E >0 The coordinates of point A and the coordinates of point P are completely identical in the same time slot: The entire flight process A starts from E and flies continuously to S: POWER E >>{POWER1,POWER2,…,POWER i } p >>POWER B With D A-B →∞,D A-E ≈0,D A-P →{0,∞} Its coordinates change in different time slots.

4. The obstacle avoidance path planning method for an aircraft according to claim 3, characterized in that... ； Gradient-enhanced decision trees were selected as the core model, and training data was loaded from a historical flight mission database, including: The distance d between the obstacle and the aircraft; The size of the obstacle is ss; The relative velocity v between the obstacle and the aircraft; The angle θ between the obstacle and the aircraft's path; For the loaded training data, extract and construct the feature matrix X and the target vector Y: Feature matrix X = [d,s,v,θ]; Target vector Y = [Pb]; Standardize the feature matrix X according to the formula: Where μ is the mean of the feature and σ is the standard deviation of the feature; Real-time orbital data is obtained through CelesTrak, and the spatial environment data fed back by the real-time orbital data is used to determine the density of obstacles in the area around the aircraft, dividing the flight space into three levels of low-density space: fewer obstacles and sparse distribution. Medium-density space: Obstacles are moderately distributed, and there are some moving satellites or particles; High-density space: obstacles are very dense, with a large number of satellites, floating particles, etc. in high-orbit or low-orbit regions; Based on the positioning information of the spacecraft's orbit and the signal interaction between the ground station and the satellite, and according to the communication coverage and navigation accuracy within the flight space, the flight space is divided into two levels: High-coverage, high-precision space: strong communication signals, accurate navigation systems Low coverage and low precision space: weak communication signals and low navigation accuracy.

5. The obstacle avoidance path planning method for an aircraft according to claim 4, characterized in that: The obstacle avoidance difficulty coefficient for each level of space is calculated based on the obstacle density, communication coverage, and navigation accuracy in different flight spaces. The specific steps are as follows: Step S1: Calculate the obstacle density factor ZMyz, specifically based on the formula: ZMyz = Nz / Area, where, Nz represents the number of obstacles in the area, and Area represents the area where the aircraft is located; the communication coverage factor TXyz is calculated using the formula: TXyz = 1 - (TXzl / TXZLmax), where TXzl is the current communication quality of the aircraft, and TXZLmax represents the maximum communication quality; the navigation accuracy factor DHyz is calculated using the formula: DHyz = 1 - (DHjd / DHJDmax), where DHjd represents the current navigation accuracy, and DHJDmax represents the maximum navigation accuracy; Step S2: Calculate the obstacle avoidance difficulty coefficient ODF by weighted combination based on the above factors, specifically according to the formula: ODF=Qz×ZMyz+Qx×TXyz+Qd×DHyz, where Qz represents the weight of the obstacle density factor, Qx represents the weight of the communication coverage factor, and Qd represents the weight of the navigation accuracy factor. The obstacle avoidance coefficient Pb and the obstacle avoidance difficulty coefficient ODF are linearly combined with fixed weights, specifically according to the formula: New-Pb=Qbz×Pb+Qbx×ODF; where New-Pb is the new obstacle avoidance coefficient, Qbz is the weight of the obstacle avoidance difficulty coefficient, and Qbx is the weight of the obstacle avoidance coefficient; and a nonlinear adjustment function F(Pb, ODF) is introduced. The new obstacle avoidance coefficient is calculated by combining the linear weighting part and the nonlinear adjustment part, specifically according to the formula: New-Pb=Qbz×Pb+Qbx×ODF+γ×F(Pb,ODF), where the weight γ is used to control the influence of nonlinear adjustment on the overall result.

6. The obstacle avoidance path planning method for an aircraft according to claim 5, characterized in that: Obstacle information data is collected using lidar sensors and historical databases on the aircraft, including: Historical position: The current and past coordinates (xTt, yTt, zTt) of the obstacle; Historical velocity: The current and past velocities of the obstacle (vx, vy, vz); Rate of change of direction: Calculate the rate of change of the obstacle's direction Δθ / ΔTt at each time step; Environmental factors: including external influences such as wind speed and air pressure; The LSTM model based on time-series data in machine learning is used to predict the dynamic behavior of obstacles. The collected obstacle information data is input into the model, and the input data is preprocessed by normalization and time window division. The LSTM model is trained based on historical data, and the mean squared error is used as the loss function to measure the model's prediction accuracy. The model's performance on different datasets is evaluated through k-fold cross-validation. Based on obstacle information data, the change in obstacle velocity vector is calculated, and the velocity change rate and direction change rate are extracted for each time step; the aircraft's velocity vector vf is then input into the model.

7. The obstacle avoidance path planning method for an aircraft according to claim 6, characterized in that: Basic information about the aircraft and obstacles is obtained through lidar, including: Obstacle position: The three-dimensional position of the obstacle (xm, ym, zm); Obstacle velocity: The velocity of an obstacle is a vector (vx, vy, vz); The speed of the aircraft is vf = (vfx, vfy, vfz); The current position of the aircraft (xdq, ydq, zdq); Given a time interval T, the relative positions of the aircraft and the obstacle for any unit time T are: (xm(Tt)-xdq(Tt), ym(Tt)-ydq(Tt), zm(Tt)-zdq(Tt)). Calculate the distance d(T) between the aircraft and the obstacle at each unit time T, using the following formula: The future trajectory of an obstacle is predicted based on its current velocity and direction changes, specifically using the formula: xm(Tt)=xm(0)+v x ×Tt;ym(Tt)=ym(0)+v y ×Tt;zm(Tt)=zm(0)+v z ×Tt, where (v x v y v z () represents the velocity vector of the obstacle; The future trajectory is predicted based on the aircraft's current speed and direction changes, specifically using the formula: xdq(Tt)=xdq(0)+v fx ×Tt;ydq(Tt)=ydq(0)+v fy ×Tt;zdq(Tt)=zdq(0)+v fz ×Tt, where (v fx v fy v fz () represents the velocity vector of the aircraft; Substituting the positions of the aircraft and obstacles into the distance formula, we can obtain the unit time T(max) by differentiating the distance function d(T) and finding the extreme point. The second derivative test can verify that the obtained T(max) is the maximum value.

8. The present invention also relates to an obstacle avoidance path planning system for aircraft, characterized in that, include: The system includes a data acquisition module, a path analysis module, a data storage module, and a future prediction module, with signal connections between these modules. The data acquisition module is used to acquire the three-dimensional coordinates of the aircraft's starting and ending points; The path analysis module is mainly used to analyze and optimize the flight path of an aircraft by combining the status of obstacles and the status of the aircraft. The future prediction module is mainly used to analyze the future dynamic behavior of obstacles and the state of the aircraft, optimize the avoidance route, and determine the best solution by analyzing the trajectory of the obstacle's movement speed and direction and the distance between the aircraft and the obstacle within a unit of time Tt. The data storage module is used to store all data processed by the platform.

Images