A fixed-wing unmanned aerial vehicle trajectory planning method and system based on reconstructed airflow field

Abstract

The present application relates to the technical field of trajectory planning, in particular to a fixed-wing unmanned aerial vehicle trajectory planning method and system based on reconstructed airflow field, comprising: obtaining the position of sampling points on the exploration trajectory, the vertical wind speed observation value and the local vertical wind speed gradient direction, and constructing an observation vector set; constructing a multi-core thermal updraft model, and obtaining the response value of the local candidate thermal core base at each sampling point on the exploration trajectory; obtaining the thermal updraft topology prior field of the multi-core thermal updraft model; obtaining the safety target heading instruction meeting the preset flight constraint; obtaining the reconstructed updraft field and performing trajectory planning. The present application can improve the discovery efficiency and reconstruction integrity of the multi-core thermal updraft under the condition of sparse observation, so as to facilitate subsequent trajectory planning.

Description

Technical Field

[0001] This invention relates to the field of trajectory planning technology, and specifically to a method and system for trajectory planning of fixed-wing unmanned aerial vehicles based on reconstructed airflow fields. Background Technology

[0002] Thermal updrafts are common vertical motion structures in the atmospheric boundary layer, providing usable environmental energy for fixed-wing UAVs. For missions such as long-endurance inspections, low-altitude monitoring, emergency communications, and autonomous gliding, timely detection and reconstruction of thermal updraft fields can help UAVs select favorable flight areas, reduce energy consumption, and improve mission safety.

[0003] like Figure 3 As shown, existing methods for sensing and utilizing thermal updrafts are typically based on assumptions of a single center, radial symmetry, or local optimization. When a real thermal updraft is composed of multiple local updraft cores, saddles, and local downdraft compensation regions, the single-core model struggles to describe the multi-core cluster topology, and local optimization strategies are easily attracted by the current strong local updraft region, thus missing other thermal cores or getting stuck in a local suboptimal sampling path.

[0004] Fixed-wing UAVs also have motion constraints such as continuous forward flight, minimum turning radius, maximum yaw rate, limited range, and boundary safety. Unlike rotor platforms, they cannot hover and resample at arbitrary locations. When the target airspace is large and the wind field is unknown a priori, the vertical wind speed observations obtained along the flight path are usually point-like, sparse, and non-uniformly distributed. Traditional coverage scanning requires a long flight path, and traditional pure interpolation methods are prone to generating topological holes or artifacts in unobserved areas. In other words, thermal updraft detection methods are prone to single-core capture, difficulty in detecting multiple local updraft cores, high topological misjudgment rate, and insufficient integrity of continuous vertical wind speed field reconstruction under the conditions of fixed-wing restricted flight paths, unknown multi-core wind fields, and sparse observations. This affects subsequent trajectory planning based on thermal updraft fields.

[0005] Therefore, there is a need to provide a fixed-wing UAV trajectory planning method and system based on reconstructed airflow field to solve the above problems. Summary of the Invention

[0006] To address the problems in existing thermal updraft detection methods, such as the tendency to fall into single-core capture, difficulty in detecting multiple local updraft cores, high topology misjudgment rate, and insufficient integrity of continuous vertical wind speed field reconstruction under fixed-wing restricted trajectories, unknown multi-core wind fields, and sparse observation conditions, thus affecting subsequent trajectory planning based on thermal updraft fields, this invention provides a fixed-wing UAV trajectory planning method and system based on reconstructed airflow fields to solve the existing problems.

[0007] The first aspect of this invention provides a trajectory planning method for a fixed-wing unmanned aerial vehicle (UAV) based on reconstructed airflow field, employing the following technical solution: The flight status data of the fixed-wing UAV in the target airspace is obtained. Based on the flight status data, the vertical wind speed observation value collected along the exploration trajectory is obtained, and the local vertical wind speed gradient direction at the current sampling point on the exploration trajectory is extracted. The sampling point position, vertical wind speed observation value and local vertical wind speed gradient direction on the exploration trajectory are used as the set of observation vectors. A multi-core thermal updraft model is constructed. Based on the multi-core thermal updraft model, multiple candidate thermal core bases are set in the target airspace and a candidate thermal core dictionary is formed. The local candidate thermal core dictionary is obtained based on the Euclidean distance between the sampling points in the historical exploration trajectory and the candidate thermal core bases in the candidate thermal core dictionary and the preset mask radius. The response value of the local candidate thermal core base at each sampling point on the exploration trajectory is obtained according to the position of each local candidate thermal core base in the local candidate thermal core dictionary and the position of each sampling point on the exploration trajectory. The peak values ​​of physical responses on historical exploration trajectories are obtained based on non-negative vertical wind speed observations and response values ​​corresponding to local candidate thermonuclear bases. Local candidate thermonuclear bases with peak physical responses greater than a set physical response threshold are selected as target thermonuclear bases. The intensity of the target thermonuclear bases is iteratively updated using the expectation-maximization iterative algorithm, and the target thermonuclear bases with converged intensity greater than a preset intensity threshold are selected as active thermonuclear bases. The thermal updraft topology prior field of the multi-core thermal updraft model is obtained based on the intensity of all active thermonuclear bases. Based on the preset initial uncertainty, a spatial uncertainty map in the target airspace at the current moment is obtained using a Gaussian decay algorithm. A local existence prior field is constructed based on non-negative vertical wind speed observations. The thermal updraft topological prior field, the local existence prior field, and the spatial uncertainty map are fused to obtain a data acquisition value map. A relative distance cost map is obtained based on the Euclidean distance from each point in the target airspace to the fixed-wing UAV's position at the current moment. A penalty weight that decreases with time step is applied to the positions of historical exploration trajectories passing through the target airspace to generate a dynamic trajectory decay map to suppress local hovering. Based on the data acquisition value map, the relative distance cost map, the dynamic trajectory decay map, the heading angle, the normalized vertical wind speed observations, and the local vertical wind speed gradient direction, a safe target heading command that satisfies preset flight constraints is obtained. The system executes safe target heading instructions and updates the observation vector set until the set flight mission ends to obtain a complete exploration trajectory. Based on the historical observation sample set of vertical wind speed in the observation vector set corresponding to the complete exploration trajectory, a continuous wind speed interpolation field is constructed. The continuous wind speed interpolation field is fused with the thermal updraft topological prior field to obtain a reconstructed updraft field with continuous vertical wind speed in the target airspace. Trajectory planning is then performed based on the reconstructed updraft field.

[0008] A further technical solution of the present invention is that the expression of the multi-core thermal updraft model is:

[0009] In the formula, Indicates the location in the target airspace The actual vertical wind speed at the location; Indicates the first The maximum vertical wind speed at the thermonuclear center; Indicates the first The coordinates of the center position of each thermonucleus; Indicates the first The characteristic radius of each thermonucleus; This represents the total number of thermal nuclei within a multinucleus cluster region in the target airspace; This represents the vertical flow field term in the continuous background.

[0010] A further technical solution of the present invention is that the step of obtaining the local candidate thermonuclear dictionary in the candidate thermonuclear dictionary based on the Euclidean distance between sampling points in the historical exploration trajectory and the candidate thermonuclear substrate in the candidate thermonuclear dictionary and the preset mask radius is as follows:

[0011] In the formula, This represents a local candidate heat kernel dictionary within the candidate heat kernel dictionary; Indicates the first The central location corresponding to each candidate thermonuclear substrate; Indicating the first in the historical exploration trajectory The position of the sampling point is added to the candidate thermonuclear dictionary. The Euclidean distance between the center positions of each candidate thermonuclear basis; Indicates the first on the exploration trajectory Location of each sampling point; Indicates the preset mask radius; This represents the total number of sampling points on the exploration trajectory.

[0012] A further technical solution of the present invention is that the expression for the response value of the local candidate thermonuclear base at each sampling point on the exploration trajectory in the local candidate thermonuclear dictionary is:

[0013] In the formula, Indicates the first in the local candidate thermonuclear dictionary The local candidate thermonuclear basis corresponding to the nth node on the exploration trajectory Each sampling point location The response value at the location; Indicates the first The central location corresponding to each local candidate thermonuclear substrate; The feature radius represents the basis of the candidate heat kernel dictionary; Indicates the first on the exploration trajectory Location of each sampling point; Indicates the first step on the exploration trajectory The location of the sampling point is added to the local candidate thermonuclear dictionary. The Euclidean distance between the locations of the candidate thermonuclear basement.

[0014] A further technical solution of the present invention is that the expression for the peak value of the physical response on the historical exploration trajectory is:

[0015] In the formula, Indicates the first Peak physical response of a local candidate thermonuclear substrate along the historical exploration trajectory; Indicates the first in the local candidate thermonuclear dictionary The candidate thermonuclear substrate corresponding to the nth node on the exploration trajectory Each sampling point location The response value at the location; Indicates the first on the exploration trajectory Non-negative vertical wind speed observations at each sampling point location; This represents the total number of sampling points on the exploration trajectory.

[0016] A further technical solution of the present invention is that the expression for the topological prior field of the thermal updraft in the multi-core thermal updraft model is as follows:

[0017] In the formula, express Position in the target airspace at all times The topological prior estimate of the thermal updraft at a given location is composed of the topological prior estimates of the thermal updraft at all locations in the target airspace, forming the topological prior field of the thermal updraft. An active thermonuclear basement set; The first in the active thermonuclear basement set The intensity corresponding to an active thermonuclear substrate; For the first The central location corresponding to an active thermonuclear basement; The feature radius is the basis of the candidate hot kernel dictionary.

[0018] A further technical solution of the present invention is that the expression for the local existence prior field is:

[0019] In the formula, express Position of target in airspace at any time The local existence prior probability at a location is formed by traversing the local existence prior probabilities of all locations within the target spatial domain. Indicates the forgetting coefficient; express Position of target in airspace at any time The prior probability of local existence at a given location; Indicates an indicator function; Indicates the location of the target in the airspace; Indicates the first The location of each sampling point; Indicates the updated radius; Indicates the effective rising threshold; Indicates the first step in the exploration trajectory Vertical wind speed observations corresponding to each sampling point.

[0020] A further technical solution of the present invention is as follows: the steps for obtaining a safe target heading instruction that satisfies preset flight constraints based on the collected value map, relative distance cost map, dynamic trajectory attenuation map, heading angle, normalized vertical wind speed observation value, and local vertical wind speed gradient direction are as follows: The acquisition value map, relative distance cost map, and dynamic trajectory attenuation map are used as multi-channel image tensors; a state vector is constructed based on heading angle, normalized vertical wind speed observations, and local vertical wind speed gradient direction. Macroscopic spatial geometric features are obtained from image topological flow based on multi-channel image tensors and active sampling strategy; microscopic kinematic features are obtained from local kinematic flow based on state vectors and active sampling strategy. Macroscopic spatial geometric features and microscopic kinematic features are concatenated into a joint state vector. Based on the joint state vector, a discretized relative heading increment is obtained using a two-stream reinforcement learning policy network. The target heading is obtained based on the relative heading increment and the current heading. A safe target heading instruction that satisfies preset flight constraints is constructed based on the direction vector of the target heading and the unit normal vector pointing into the target airspace.

[0021] A further technical solution of the present invention is that the expression for the reconstructed updraft field of continuous vertical wind speed in the target airspace is:

[0022] In the formula, Indicates the location in the target airspace The reconstructed updraft value of continuous vertical wind speed at a location is composed of the reconstructed updraft value of continuous vertical wind speed at all locations in the target airspace, forming a reconstructed updraft field. Indicates the location in the target airspace Continuous wind speed interpolation at the location; Indicates the first Position of target in airspace at any time Topological prior estimates of the thermal updraft at that location; Indicates the location in the target airspace The adaptive fusion coefficient at the location.

[0023] A second aspect of the present invention provides a trajectory planning system for a fixed-wing unmanned aerial vehicle based on reconstructed airflow field, comprising: The parameter acquisition module is used to acquire flight status data of fixed-wing UAV in the target airspace, acquire vertical wind speed observation values ​​collected along the exploration trajectory based on the flight status data, and extract the local vertical wind speed gradient direction at the current sampling point on the exploration trajectory. The sampling point position, vertical wind speed observation value and local vertical wind speed gradient direction on the exploration trajectory are used as the set of observation vectors. The airflow model and response construction module is used to construct a multi-core thermal updraft model. Based on the multi-core thermal updraft model, multiple candidate thermal core bases are set in the target airspace to form a candidate thermal core dictionary. Based on the Euclidean distance between the sampling points in the historical exploration trajectory and the candidate thermal core bases in the candidate thermal core dictionary and the preset mask radius, the local candidate thermal core dictionary is obtained. According to the position of each local candidate thermal core base in the local candidate thermal core dictionary and the position of each sampling point on the exploration trajectory, the response value of each sampling point of the local candidate thermal core base on the exploration trajectory is obtained. The thermal updraft topology prior field calculation module is used to obtain the physical response peak value on the historical exploration trajectory based on the non-negative vertical wind speed observation value and the response value corresponding to the local candidate hot core; the local candidate hot core base when the physical response peak value is greater than the set physical response threshold is taken as the target hot core base; the intensity of the target hot core base is iteratively updated using the expectation-maximization iterative algorithm, and the target hot core base when the intensity after convergence is greater than the preset intensity threshold is taken as the active hot core base; the thermal updraft topology prior field of the multi-core thermal updraft model is obtained based on the intensity of all active hot core bases. The safe target heading instruction acquisition module is used to obtain a spatial uncertainty map in the target airspace at the current moment based on a preset initial uncertainty and using a Gaussian decay algorithm; construct a local existence prior field based on non-negative vertical wind speed observations; fuse the thermal updraft topological prior field, the local existence prior field, and the spatial uncertainty map to obtain a data acquisition value map; obtain a relative distance cost map based on the Euclidean distance from each point in the target airspace to the fixed-wing UAV's position at the current moment; apply a penalty weight that decreases with time step to the positions of historical exploration trajectories passing through the target airspace to generate a dynamic trajectory decay map to suppress local hovering; and obtain a safe target heading instruction that satisfies preset flight constraints based on the data acquisition value map, the relative distance cost map, the dynamic trajectory decay map, the heading angle, the normalized vertical wind speed observations, and the local vertical wind speed gradient direction. The trajectory planning module is used to execute safe target heading instructions and update the observation vector set until the set flight mission ends to obtain a complete exploration trajectory. The complete exploration trajectory is used to construct a continuous wind speed interpolation field. The continuous wind speed interpolation field is fused with the thermal updraft topological prior field to obtain a reconstructed updraft field with continuous vertical wind speed in the target airspace. Trajectory planning is performed based on the reconstructed updraft field.

[0024] The beneficial effects of this invention are: This invention utilizes a candidate thermonuclear dictionary and a non-negative sparse Bayesian learning model with physical response truncation to transform sparse point-state vertical wind speed observation samples obtained along a restricted trajectory into a topological prior field of thermal updraft. This can suppress false thermonuclears and extract the active thermonuclear base even when the number of observation samples is small, noise is strong, and local updraft nuclei are close to each other. This invention integrates the topological prior field, spatial uncertainty map, and local existence prior into the acquisition value, and inputs the multi-channel image tensor composed of the acquisition value, relative distance cost, and dynamic trajectory decay value, along with the state vector of the fixed-wing UAV, into an active sampling strategy. This enables the fixed-wing UAV to actively select a more informative sampling path under the constraints of maximum yaw rate, minimum turning radius, and mission boundary safety, thereby reducing the risk of traditional source-finding strategies locking onto a single strong updraft region for a long time. This invention utilizes both active hot core sets and vertical wind speed observation samples from the complete exploration trajectory for continuous reconstruction after the loop closure is completed. This allows the reconstruction results to retain the topological stability of the multi-core updraft structure while also representing the vertical motion changes in the sinking compensation region and the background. This improves the integrity and engineering usability of the continuous vertical wind speed field in the target airspace, thereby making the trajectory planning results based on the reconstructed updraft field more accurate. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 This is a flowchart illustrating a fixed-wing unmanned aerial vehicle trajectory planning method based on reconstructed airflow field according to the present invention. Figure 2 This is a schematic diagram of the coordinate system of the fixed-wing UAV, the fusion of airborne sensor data, and the formation of vertical wind speed samples in an embodiment of the present invention; Figure 3 This is a schematic diagram of a single-core thermal updraft in an embodiment of the present invention; Figure 4 This is a schematic diagram of the multi-core thermal updraft in an embodiment of the present invention; Figure 5 This is a schematic diagram illustrating the construction of the candidate heat kernel dictionary, local mask, and response matrix in an embodiment of the present invention; Figure 6 The topology inference and global root mean square error convergence curves of the active sampling strategy provided by this invention; Figure 7 The topology inference and global root mean square error convergence curves for existing lawnmower-style baseline strategies are shown. Figure 8 The image shows the reconstruction effect of the exploration trajectory (RL path) generated in the target airspace using the method provided by this invention and the continuous vertical wind speed. Figure 9 The image shows the reconstruction effect of the sampling trajectory (lawnmower path) and continuous vertical wind speed in the target airspace for the existing lawnmower-style baseline strategy. Detailed Implementation

[0027] 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.

[0028] An embodiment of the present invention, based on a reconstructed airflow field, for example... Figure 1 As shown, it includes: S1. Obtain the sampling point locations, vertical wind speed observations, and local vertical wind speed gradient directions on the exploration trajectory, and construct a set of observation vectors. Specifically, the flight status data of the fixed-wing UAV in the target airspace is obtained, the vertical wind speed observation value collected along the exploration trajectory is obtained based on the flight status data, and the local vertical wind speed gradient direction at the current sampling point on the exploration trajectory is extracted. The sampling point position, vertical wind speed observation value and local vertical wind speed gradient direction on the exploration trajectory are used as the set of observation vectors.

[0029] S11. The steps for acquiring flight status data of a fixed-wing UAV in the target airspace are as follows: Specifically, a mission coordinate system is established, and flight status data of the fixed-wing UAV in the target airspace is collected based on the mission coordinate system.

[0030] S111. Establish the task coordinate system: For example, in one specific embodiment, an NED coordinate system (northeast-east coordinate system) is established with the mission starting point as the origin, and the position of the target in the airspace is denoted as... In this embodiment, the position is converted from ENU coordinates (northeast-sky coordinates) to northeast-earth coordinates, and the heading angle is... The north direction is zero and clockwise is positive.

[0031] S112. Collect flight status data: For example, in one specific embodiment, flight state data includes: position, ground speed, attitude, airspeed, angle of attack, sideslip angle, flight control roll control parameters, and mission boundary data; such as Figure 2 As shown, the position and ground speed of the fixed-wing UAV are acquired using GNSS (Global Navigation Satellite System); the attitude of the fixed-wing UAV is acquired using IMU (Inertial Measurement Unit); the airspeed of the fixed-wing UAV is acquired using an airspeed meter; the angle of attack and sideslip angle are measured using a weather vane or a five-hole probe; the flight control computer calculates the flight control roll control quantity based on the target attitude and the current attitude (acquired by the IMU); and the flight control computer integrates the position, attitude, airspeed, angle of attack, sideslip angle, and flight control roll control quantity of the fixed-wing UAV, and combines them with the flight envelope of the aircraft (maximum speed, safe altitude, critical angle of attack) to calculate the mission boundary data used to ensure flight safety.

[0032] S12. The steps for obtaining vertical wind speed observations collected along the exploration trajectory based on flight status data are as follows: Specifically, the flight status data is synchronized in time, and the wind speed vector is inverted to obtain the true value of vertical wind speed. Based on the true value of vertical wind speed, the vertical wind speed observation value is obtained, and isolated outliers in the vertical wind speed observation value are removed to obtain the vertical wind speed observation vector.

[0033] S121, Time Synchronization: For example, in one specific embodiment, the position, ground speed, attitude, airspeed, angle of attack, and sideslip angle are registered according to timestamps based on the flight control master clock; when the time difference is greater than the synchronization threshold, the sampling point is discarded.

[0034] S122: Wind speed vector inversion: For example, in one specific embodiment, the airspeed vector of the aircraft system is constructed from airspeed, angle of attack, and sideslip angle. The ground speed vector in the NE coordinate system is obtained by subtracting the airspeed vector of the aircraft system from the vector after rotating it to the NE coordinate system. Then, the ground component of the wind speed vector in the NE coordinate system is extracted and inverted to obtain the true value of the vertical wind speed. The specific formula for calculating the true value of the vertical wind speed is as follows:

[0035] In the formula, The wind speed vector is given in northeast coordinates. This is the ground velocity vector in northeastern coordinates; Here is the attitude rotation matrix; For the airspeed vector of the machine system; Represents the geodetic unit column vector in the northeast coordinate system. . Representing the true vertical wind speed, the airspeed vector of the aircraft system is:

[0036] In the formula, This indicates the airspeed of a fixed-wing drone; This indicates the angle of attack of a fixed-wing unmanned aerial vehicle (UAV). This indicates the sideslip angle of a fixed-wing unmanned aerial vehicle (UAV).

[0037] S123: Formation of vertical wind speed observations: For example, in one specific embodiment, the first Location of each valid sampling point Compared with vertical wind speed observations If the sample is composed, the observation model is:

[0038] In the formula, Indicates the first step in exploring the trajectory Vertical wind speed observations corresponding to each sampling point; Indicates the first step in exploring the trajectory Location of each sampling point The actual vertical wind speed at the location; This indicates the measurement noise of the vertical wind speed sensor; Indicates a Gaussian normal distribution; The variance of the vertical wind speed observation noise is represented in this embodiment, which is determined based on the sensor accuracy. .

[0039] S124: Obtaining the vertical wind speed observation vector by removing isolated outliers: For example, in one specific embodiment, a robust statistical method based on the absolute deviation of the median is used for outlier detection. When the... Vertical wind speed observations If the following condition is met, the observed vertical wind speed is identified as an isolated anomaly and removed:

[0040] In the formula, This represents the vertical wind speed observation vector within a local observation window. The operator used to calculate the median of the vertical wind speed observation vector; the constant 1.4826 is the scaling factor that ensures the absolute deviation of the median is consistent with the standard deviation under the assumption of a normal distribution; the right side of the equation... The overall characterization represents the median absolute deviation of the vertical wind speed observation vector within the local observation window.

[0041] S13. Extract the local vertical wind speed gradient direction at the current sampling point on the exploration trajectory: For example, in one specific embodiment, considering the engineering difficulties of accurately calculating the global wind speed gradient amplitude of a fixed-wing UAV in a real, complex atmosphere, this embodiment adopts a dimensionality-reduced local perception strategy, directly extracting the local vertical wind speed gradient using aerodynamic response. When a fixed-wing UAV traverses an unevenly distributed updraft, the vertical wind speed difference between the left and right wings induces an aerodynamic roll moment. Based on the roll control parameters (such as aileron deflection) output by the flight control layer to counteract this roll moment, the system directly extracts the lateral wind gradient component in the body coordinate system. Since longitudinal airflow differential is extremely difficult to decouple from the aircraft's own pitch attitude control, this embodiment ignores the longitudinal gradient in local perception (i.e., assumes the longitudinal wind gradient component in the body coordinate system is used). Combined with the drone's current heading angle Transforming the lateral wind gradient components in the body coordinate system to the NED (Northeast Earth) coordinate system yields estimates of the local vertical wind speed gradients in the north and east directions:

[0042] In the formula, This is an estimate of the local vertical wind speed gradient from the north. This is an estimate of the local vertical wind speed gradient heading east. To avoid low signal-to-noise ratio gradient magnitudes directly interfering with the strategy network, this embodiment only retains the local vertical wind speed gradient direction, that is:

[0043] In the formula, The gradient direction threshold is taken as [value missing] in this embodiment. = 0.05 (m / s) / m, this value is determined by the standard deviation of the local vertical wind speed gradient observation noise, and is used to determine the weak gradient direction within the noise scale as unreliable and set it to zero, so as to avoid the frequent flipping of the direction sign caused by weak disturbances. Indicates the first Each sampling point is The direction of the local vertical wind speed gradient; The value can be -1, 0, or +1. Pick hour, for The direction of the local vertical wind speed gradient; when Pick hour, for The direction of the local vertical wind speed gradient; for The estimated local vertical wind speed gradient; if the absolute value of the estimated local vertical wind speed gradient is less than or equal to the gradient direction threshold. If the directional clue is not found, it is considered unreliable and set to 0. Thus, an observation vector can be constructed based on the sampling point locations on the exploration trajectory, the observed vertical wind speed values, and the local vertical wind speed gradient direction. In this embodiment, the [missing information]... The set of observation vectors at each sampling point can be represented as:

[0044] In the formula, For the first The observed vectors at each sampling point. As the fixed-wing UAV continues to fly along the exploration trajectory, the observed vectors at each sampling point... The accumulated historical observation vectors form a set that is used for subsequent sparse Bayesian topology inference and active sampling strategy networks.

[0045] S2. Construct a multi-core thermal updraft model and obtain the response value of each sampling point on the exploration trajectory for the local candidate thermal core substrate; Specifically, a multi-core thermal updraft model is constructed, and multiple candidate thermal core bases are set up in the target airspace based on the multi-core thermal updraft model to form a candidate thermal core dictionary; the local candidate thermal core dictionary is obtained based on the Euclidean distance between the sampling points in the historical exploration trajectory and the candidate thermal core bases in the candidate thermal core dictionary and the preset mask radius; the response value of the local candidate thermal core base at each sampling point on the exploration trajectory is obtained according to the position of each local candidate thermal core base in the local candidate thermal core dictionary and the position of each sampling point on the exploration trajectory.

[0046] S21. Construct a multi-core thermal updraft model: For example, such as Figure 4 As shown, in a specific embodiment, the thermal updraft in the target airspace is regarded as a two-dimensional horizontal field composed of multiple local updraft cores and background vertical motion. The multi-core thermal updraft model is then expressed as:

[0047] In the formula, Indicates the location in the target airspace The actual vertical wind speed at the location; Indicates the first in the target airspace The intensity of a thermonuclear unit is physically defined as the maximum vertical wind speed at the center of that thermonuclear unit. Indicates the first The coordinates of the center position of each thermonucleus (e.g.) Figure 4 middle (Indicates the coordinates of the center position of the third thermonucleus). Indicates the first The characteristic radius of each thermonucleus; This represents the total number of thermal nuclei within a multinucleus cluster region in the target airspace; Indicates position The continuous background vertical flow field term is used to characterize the flow field within the target airspace, except for... In addition to the discrete thermal core, the residual vertical wind field exhibits smooth spatial variation, generated by large-scale weather trends or topographic lifting. It should be noted that this multi-core thermal updraft model is only a "rising core prior" model and not a complete directional atmospheric vertical wind speed field.

[0048] S22. Construct a candidate hot core dictionary: For example, in one specific embodiment, due to the actual number of thermonuclear cells in the objective environment... Both the target space and its state parameters are unknown. In this embodiment, a discrete dictionary basis is used for spatial approximation. During the task initialization phase, the target space is uniformly distributed. Given candidate thermonuclear bases to form a candidate thermonuclear dictionary, then the candidate thermonuclear dictionary set... The mathematical expression is:

[0049] In the formula, Indicates the first in the candidate thermonuclear dictionary The central location corresponding to each candidate thermonuclear substrate.

[0050] S23. Obtain the local candidate heat core dictionary: For example, such as Figure 5 As shown, in one specific embodiment, a local candidate hot core dictionary is obtained based on the Euclidean distance between sampling points in the historical exploration trajectory and the candidate hot core base in the candidate hot core dictionary, and a preset mask radius. That is, in this embodiment, the Euclidean distance between sampling points in the historical exploration trajectory and the candidate hot core base in the candidate hot core dictionary is less than or equal to the preset mask radius. Candidate thermonuclear basis composition local candidate thermonuclear dictionary That is:

[0051] In the formula, This represents a local candidate heat kernel dictionary within the candidate heat kernel dictionary; Indicates the first The central location corresponding to each candidate thermonuclear substrate; Indicating the first in the historical exploration trajectory The position of the sampling point is added to the candidate thermonuclear dictionary. The Euclidean distance between the center positions of each candidate thermonuclear basis; Indicates the first on the exploration trajectory Location of each sampling point; This represents the preset mask radius, which is taken as [value missing] in this embodiment. , which is the feature radius of the candidate hot kernel dictionary basis. Three times that of the candidate thermonuclear substrates with effective response near historical tracks; This represents the total number of sampling points on the exploration trajectory.

[0052] S24. Obtain the response value of each sampling point on the exploration trajectory for the local candidate hot core basis in the local candidate hot core dictionary: For example, in one specific embodiment, for the first on the exploration trajectory The sampling point and the local candidate thermonuclear dictionary of the th sampling point The first local candidate thermonuclear substrate on the exploration trajectory The response value for each sampling point is:

[0053] In the formula, Indicates the first in the local candidate thermonuclear dictionary The local candidate thermonuclear basis corresponding to the nth node on the exploration trajectory Each sampling point location The response value at the location; Indicates the first The central location corresponding to each local candidate thermonuclear substrate; The feature radius represents the basis of the candidate heat kernel dictionary; Indicates the first on the exploration trajectory Location of each sampling point; Indicates the first on the exploration trajectory From the location of the sampling point to the... The Euclidean distances between the locations of local candidate thermonuclear bases are used. In closed-loop detection, the candidate thermonuclear dictionary remains constant; only the response values ​​for newly added sampling points are calculated, and the observation matrix is ​​incrementally updated to reduce onboard computational overhead.

[0054] S3. Obtain the thermal updraft topological prior field of the multi-core thermal updraft model; Specifically, the peak values ​​of physical responses on historical exploration trajectories are obtained based on non-negative vertical wind speed observations and response values ​​corresponding to local candidate thermonuclear bases; local candidate thermonuclear bases with physical response peak values ​​greater than a set physical response threshold are used as target thermonuclear bases; the intensity of the target thermonuclear bases is iteratively updated using the expectation-maximization iterative algorithm, and the target thermonuclear bases with converged intensity greater than a preset intensity threshold are used as active thermonuclear bases; the thermal updraft topology prior field of the multi-core thermal updraft model is obtained based on the intensity of all active thermonuclear bases.

[0055] S31. Obtain non-negative vertical wind speed observations: For example, in one specific embodiment, since the topological inference of multi-core thermal updrafts only applies to non-negative updraft kernels, the acquired vertical wind speed observation samples need to be truncated to extract the non-negative vertical wind speed observation values ​​as valid observation components. When the value is greater than zero, it is directly retained; when it is less than or equal to zero, it is forcibly set to zero. This is used to construct the non-negative vertical wind speed observation vector for the input non-negative sparse Bayesian learning model. The non-negative vertical wind speed observation vector is specifically defined as follows:

[0056] In the formula, This represents the non-negative vertical wind speed observation vector. Indicates the first step on the exploration trajectory The non-negative vertical wind speed observation values ​​corresponding to each sampling point; Indicates the first step on the exploration trajectory Vertical wind speed observations corresponding to each sampling point.

[0057] Furthermore, in order to ensure the physical integrity of the global airflow field reconstruction, the system simultaneously retains the complete original vertical wind speed observation samples while performing non-negative truncation processing; after the closed-loop detection mission is completed, the original vertical wind speed observation samples will be input into the radial basis function (RBF) continuous interpolation branch to express the continuous background vertical flow field changes around the thermal updraft.

[0058] S32. Establish the non-negative sparse Bayesian learning observation equation: For example, in one specific embodiment, a non-negative sparse Bayesian learning observation equation is constructed for subsequent non-negative sparse Bayesian inference. The expression of the non-negative sparse Bayesian learning observation equation is as follows:

[0059] In the formula, This represents the non-negative vertical wind speed observation vector; This represents the response matrix, which consists of the response values ​​at each sampling point on the exploration trajectory of the local candidate thermonuclear base. This represents the intensity vector of the local candidate thermonuclear substrate; This represents the observation noise vector that follows a zero-mean Gaussian distribution. ; Indicates noise accuracy; Indicates a Gaussian normal distribution; Let K represent the identity matrix of dimension K×K; the dimension of the non-negative vertical wind speed observation vector is K×1, the response matrix is ​​K×the number of local basis dimensions, and the intensity of the local candidate thermal core basis is the number of local basis dimensions × 1.

[0060] S33. Obtain the peak physical response along the historical exploration trajectory: For example, in one specific embodiment, for the first A local candidate thermonuclear substrate is defined, and the peak physical response of the local candidate thermonuclear substrate along the historical exploration trajectory is defined. Then we have:

[0061] In the formula, Indicates the first Peak physical response of a local candidate thermonuclear substrate along the historical exploration trajectory; Indicates the first in the local candidate thermonuclear dictionary The candidate thermonuclear substrate corresponding to the nth node on the exploration trajectory Each sampling point location The response value at the location; Indicates the first on the exploration trajectory Non-negative vertical wind speed observations at each sampling point location; This represents the total number of sampling points on the exploration trajectory.

[0062] S34. Obtain the target thermonuclear substrate; For example, in one specific embodiment, when the physical response peak Less than or equal to the set physical response threshold When the local candidate thermonuclear substrate is determined to have no physical support, it is eliminated in subsequent inferences; when the physical response peak value Greater than the set physical response threshold At that time, the local candidate thermonuclear substrate is taken as the target thermonuclear substrate. In this embodiment, the physical response threshold This value is derived from the standard deviation of vertical wind speed observation noise. Sure.

[0063] S35, Obtain an active thermonuclear substrate; For example, in one specific embodiment, based on the observation equation in S32, the expectation-maximization (EM) iterative algorithm in the prior art is used to analyze the target thermonuclear substrate. The intensity is updated iteratively.

[0064] The specific iterative process is as follows: In the expectation step (E step) of each iteration, the posterior mean and covariance of the target thermonuclear base strength are calculated, and a non-negative projection operator is introduced to truncate the posterior mean of the target thermonuclear base strength element by element, forcibly setting elements less than zero to zero, so as to eliminate sinking solutions that do not conform to the true thermodynamic properties; In the maximization step (M step), the strength and observation noise accuracy of each target thermonuclear base are updated according to the existing sparse Bayesian learning rules.

[0065] The above EM iterations are executed repeatedly. When the maximum number of iterations is reached, or when the relative change in intensity of the target thermonuclear substrate that has not been eliminated in two consecutive iterations is less than a preset relative change in intensity threshold (in this embodiment, the relative change in intensity threshold is set to 1), the iterations continue until the maximum number of iterations is reached. (This is the empirical value for iterative convergence of the EM algorithm), stop the iteration and determine convergence. Among the converged target thermonuclear base strength values, those greater than the preset strength threshold are... The local candidate thermonuclear substrate corresponding to the element is taken as the active thermonuclear substrate. In this embodiment, we take... This intensity threshold is consistent with the numerical scale of the convergence criterion of the EM iterative algorithm, and further, the active thermonuclear basis set is composed of all active thermonuclear basis sets. .

[0066] S36. Obtain the thermal updraft topological prior field of the multi-core thermal updraft model: For example, in one specific embodiment, the center positions and convergence intensities corresponding to all active thermonuclear bases are extracted, and then physically superimposed in space using the Gaussian function. This maps the discrete topological inference results to a continuous two-dimensional updraft distribution, generating a multi-nuclear thermo-updraft topological prior field. The expression for the thermo-updraft topological prior field of the multi-nuclear thermo-updraft model is as follows:

[0067] In the formula, Indicates the first Position in the target airspace at all times The topological prior estimate of the thermal updraft at a given location is composed of the topological prior estimates of the thermal updraft at all locations in the target airspace, forming the topological prior field of the thermal updraft. An active thermonuclear basement set; The first in the active thermonuclear basement set The intensity corresponding to an active thermonuclear substrate; For the first The central location corresponding to an active thermonuclear basement; The feature radius is the basis of the candidate hot kernel dictionary.

[0068] S4. Obtain a safe target heading instruction that meets preset flight constraints; Specifically, based on the preset initial uncertainty, a spatial uncertainty map in the target airspace at the current moment is obtained using a Gaussian decay algorithm. A local existence prior field is constructed based on non-negative vertical wind speed observations. The thermal updraft topological prior field, the local existence prior field, and the spatial uncertainty map are fused to obtain a data acquisition value map. A relative distance cost map is obtained based on the Euclidean distance from each point in the target airspace to the fixed-wing UAV's position at the current moment. A penalty weight that decreases with time step is applied to the positions of historical exploration trajectories passing through the target airspace to generate a dynamic trajectory decay map to suppress local hovering. Based on the data acquisition value map, the relative distance cost map, the dynamic trajectory decay map, the heading angle, the normalized vertical wind speed observations, and the local vertical wind speed gradient direction, a safe target heading command that meets preset flight constraints is obtained.

[0069] S41. Obtain the spatial uncertainty map of the target spatial domain at the current moment: For example, in one specific embodiment, the initial spatial uncertainty of all locations within the target airspace is set at the start of the mission. As the flight mission progresses, the Gaussian decay algorithm is used to reduce the spatial uncertainty of the trajectory passing through the target airspace, and the current time is recursively obtained. ,Location Spatial uncertainty for:

[0070] In the formula, Indicates the current time Location in the target airspace The spatial uncertainty is calculated by traversing all locations within the target spatial domain and calculating their spatial uncertainties, thus obtaining the spatial uncertainty map. Indicates fixed-wing unmanned aerial vehicle The position of the target airspace at any given time; express In this embodiment, the observation influence radius is preset based on the effective physical range of the fixed-wing UAV's aerodynamic sensing. Pick This value is determined based on the coverage area that a single aerodynamic sensing operation by a fixed-wing UAV can cover.

[0071] S42. Constructing a local existence prior field based on non-negative vertical wind speed observations: For example, in one specific embodiment, to characterize the probability of a real updraft existing in a local area, a recursive update is performed based on vertical wind speed observations: when a non-negative vertical wind speed (effective updraft) is not observed in a local area for a consecutive period, the existence prior of that local area is reduced; when an effective updraft greater than a set threshold is observed, the existence prior of that area is increased. The recursive formula for the local existence prior field is then:

[0072] In the formula, express Position of target in airspace at any time The local existence prior probability is obtained by traversing all locations within the target spatial domain and calculating their local existence prior probabilities. Indicates the forgetting coefficient; express Position of target in airspace at any time The prior probability of local existence at a given location; Indicates an indicator function; Indicates the location of the target in the airspace; Indicates the first The location of each sampling point; Indicates the updated radius; This represents the effective rising threshold; in this embodiment, the effective rising threshold is... This value is an empirical threshold determined based on the background vertical wind speed level. Indicates the first step on the exploration trajectory Vertical wind speed observations corresponding to each sampling point.

[0073] S43. The topological prior field of the thermal updraft, the local existence prior field, and the spatial uncertainty map are fused to obtain the data acquisition value map: For example, in one specific embodiment, for any location within the target airspace The expression for its collection value is:

[0074] In the formula, Indicates the location in the target airspace The collection value at each location; the collection value map is constructed by traversing the collection values ​​of all locations within the target airspace. Indicates in Position in the prior field of thermal updraft topology corresponding to the target airspace at any given time Topological prior estimates of the thermal updraft at that location; Indicates in Location in the spatial uncertainty diagram corresponding to the target's spatial domain at any given time Spatial uncertainty at that location Indicates in Position in the local existence prior field corresponding to the target's spatial domain at any given time The local existence prior probability at a given location; the first term on the right side of the equation represents the unobserved utilization value of the topological prior support region, and the second term on the right side of the equation represents the exploration value driven by spatial uncertainty; This represents the exploration weight coefficient that is dynamically adjusted according to the mission phase.

[0075] S44. Obtain the relative distance cost map based on the Euclidean distance from each point in the target airspace to the fixed-wing UAV's current position: For example, in one specific embodiment, the system traverses each location point within the target airspace and calculates the normalized relative distance cost from each location point to the current position of the fixed-wing UAV, thereby obtaining a relative distance cost map.

[0076] S45. Apply a penalty weight that decreases with time step to the points in the target airspace that the historical exploration trajectory passes through, and generate a dynamic trajectory decay map to suppress local hovering.

[0077] For example, in one specific embodiment, when a fixed-wing UAV passes through a certain location point in the target airspace, an initial penalty weight is assigned to that location point; as time progresses, the penalty weights of each location point on the historical exploration trajectory are gradually updated according to preset rules; finally, each location point in the target airspace is traversed to obtain the penalty weight corresponding to each location point at the current time, thus obtaining the dynamic trajectory decay map.

[0078] S46. Based on the collected value map, relative distance cost map, dynamic trajectory attenuation map, heading angle, normalized vertical wind speed observation value, and local vertical wind speed gradient direction, obtain the safe target heading command that satisfies the preset flight constraints: Specifically, the acquisition value map, relative distance cost map, and dynamic trajectory decay map are used as multi-channel image tensors, and macroscopic spatial geometric features are obtained using image topology flow with an active sampling strategy. A state vector is constructed using heading angles with sine and cosine decoupling, normalized vertical wind speed, and local vertical wind speed gradient direction. Microscopic kinematic features are obtained based on the state vector and using local kinematic flow with an active sampling strategy. The macroscopic spatial geometric features and microscopic kinematic features are concatenated into a joint state vector. Based on the joint state vector, a discretized relative heading increment is obtained using a two-stream reinforcement learning strategy network. The target heading is obtained based on the relative heading increment and the current heading. A safe target heading instruction that satisfies preset flight constraints is constructed based on the direction vector of the target heading and the unit normal vector pointing into the target airspace.

[0079] S461. Constructing a multi-channel image tensor: For example, in one specific embodiment, the acquisition value map, relative distance cost map and dynamic trajectory attenuation map, each with a size of 100×100, are stitched together in the channel dimension and downsampled by a 5×5 max pooling operation to construct a multi-channel image tensor with a dimension of 3×20×20. S462. Constructing the state vector: For example, in one specific embodiment, a state vector is constructed based on the current heading angle decoupled by sine and cosine, the normalized vertical wind speed observation, the local vertical wind speed gradient direction, the previous action encoding, and the update countdown of the non-negative sparse Bayesian learning model; this state vector serves as the input to the local kinematic flow, i.e., the state vector. for:

[0080] In the formula, Indicates the current heading angle; This represents the normalized vertical wind speed observation value; Indicates the direction of the local vertical wind speed gradient corresponding to the north; This indicates the direction of the local vertical wind speed gradient corresponding to the eastward direction; Indicates the code of the previous action; This indicates the update countdown for the non-negative sparse Bayesian learning model (SBL).

[0081] Among them, normalized vertical wind speed observations The calculation formula is:

[0082] In the formula, This represents the current observed vertical wind speed. This represents the minimum observed vertical wind speed. This represents the observed maximum vertical wind speed.

[0083] S463. Configure the active sampling strategy network: In this embodiment, the image topology flow of the active sampling strategy employs a three-layer 3×3 convolutional network with 16, 32, and 64 channels, while the local kinematics flow of the active sampling strategy employs a two-layer 64-dimensional fully connected network. Based on the multi-channel image tensor, macroscopic spatial geometric features are obtained using the image topology flow of the active sampling strategy. Microscopic kinematic features are obtained based on the state vector and using the input of the local kinematics flow of the active sampling strategy. Then, the joint state vector, obtained by fusing the macroscopic spatial geometric features and the microscopic kinematic features, is input into the active sampling strategy network to obtain the relative heading increment. It should be noted that the active sampling strategy network is existing technology and will not be described in detail in this embodiment. Specifically, the output is a discretized set of relative heading increments:

[0084] In one embodiment, the number of discrete actions is taken as follows: Network output relative heading increment Then, the heading increment is superimposed with the current heading to obtain the target heading:

[0085] In the formula, Indicate the target heading; This represents the angle normalization function, used to transform and restrict the angle to a closed interval between negative and positive pi, and then saturate it with yaw rate to obtain the executable control quantity. for:

[0086] In the formula, This represents the proportional gain coefficient for heading control; This indicates the maximum permissible yaw rate control command limit for a fixed-wing platform. This represents the angle normalization function, used to map the heading error to... Interval.

[0087] S464. Determine the safe target heading that meets preset flight constraints: Specifically, a safe target heading instruction that satisfies preset flight constraints is constructed based on the direction vector of the target heading and the unit normal vector pointing into the target airspace.

[0088] S4641, Preset flight constraints: For example, since fixed-wing UAVs cannot hover for resampling, their nonholonomic kinematic constraint model in the horizontal plane is defined as follows:

[0089] And satisfy the yaw rate and minimum turning radius constraints:

[0090] In the formula, The derivative of the planar flight state vector of a fixed-wing UAV; This is the planar flight state vector of a fixed-wing UAV. Northward coordinates; The coordinates are eastward. For heading angle, Airspeed, This is the yaw rate control value. The maximum permissible yaw rate for a fixed-wing UAV. Minimum turning radius, The maximum allowable roll angle for a fixed-wing UAV. The aforementioned discrete executable control quantity for gravitational acceleration This is the continuous control quantity. In the The specific instructions issued for each discrete sampling step; therefore, the final output maneuver instructions must meet the control limit constraints jointly defined by the maximum yaw rate and the minimum turning radius, so as to ensure that the generated flight exploration trajectory is within the aerodynamic envelope of the fixed-wing UAV and has actual executability.

[0091] S4642, Boundary condition correction: Construct the target direction vector from the target heading:

[0092] In the formula, The target direction vector; Indicates the target heading.

[0093] When the fixed-wing UAV is less than the safe distance threshold from the boundary, extract the unit normal vector pointing into the airspace. And based on the preset rejection weights, a rejection vector is constructed, and then the safe target heading instructions are rewritten by superimposing this rejection vector:

[0094]

[0095] In the formula, This represents the safe target direction vector after boundary condition correction; This represents a preset rejection weight. In this embodiment, the preset rejection weight... It is a dimensionless empirical coefficient used to enhance the boundary correction component pointing towards the mission airspace in the direction of the safety target. Indicates the direction vector of the safe target. Components along the axial direction; Indicates the direction vector of the safe target. Components along the axial direction; This represents the preset guidance control law coefficient, in this embodiment... ; Represents the repulsion vector. This represents the two-parameter arctangent function; based on the safety target direction vector. Calculate the safe target heading The safe yaw rate control command is obtained through yaw rate limiting. .

[0096] At this point, a safe yaw rate control command that meets the preset flight constraints can be obtained.

[0097] S5. Obtain the reconstructed updraft field and perform trajectory planning; Specifically, the safe target heading command is executed and the observation vector set is updated until the set flight mission ends to obtain a complete exploration trajectory. Based on the historical observation sample set of vertical wind speed in the observation vector set corresponding to the complete exploration trajectory, a continuous wind speed interpolation field is constructed. The continuous wind speed interpolation field is fused with the thermal updraft topological prior field to obtain a reconstructed updraft field of continuous vertical wind speed in the target airspace. Trajectory planning is performed based on the reconstructed updraft field.

[0098] S51. Construct a continuous wind speed interpolation field: For example, in one specific embodiment, a continuous wind speed interpolation field is constructed by extracting the historical vertical wind speed observation sample set from the observation vector set corresponding to the complete exploration trajectory. The radial basis function (RBF) explicitly uses a Gaussian kernel, and its mathematical definition is:

[0099] The continuous wind speed interpolation field is defined as the weighted superposition of kernel functions at each vertical wind speed historical observation sample, that is:

[0100] In the formula, Indicates the location in the target airspace Interpolation results of continuous wind speed at the location; Indicates the first step on the exploration trajectory Coordinates of each sampling point; Indicates the first Interpolation weights for each sampling point; The set RBF kernel scale parameters; This represents the total number of sampling points; This represents the Euclidean distance. The interpolation weight column vector is used to represent this. This can be obtained by solving the regularized least squares problem:

[0101] In the formula, for A dimensional kernel matrix whose elements consist of radial basis function values ​​between sampling points; for An identity matrix of order 1; The regularization coefficient is used. The vertical wind speed observation column vector is used to fully explore the observation vector set corresponding to the trajectory.

[0102] S52, Reconstructing the Updraft Field: For example, in one specific embodiment, the reconstructed updraft field is obtained by fusing a continuous wind speed interpolation field and a multi-core thermal updraft topological prior field, that is, the expression for the reconstructed updraft field is:

[0103] In the formula, Indicates the location in the target airspace The reconstructed updraft value of continuous vertical wind speed at a location is composed of the reconstructed updraft value of continuous vertical wind speed at all locations in the target airspace, forming a reconstructed updraft field. Indicates the location in the target airspace Continuous wind speed interpolation at the location; Indicates the first Position of target in airspace at any time Topological prior estimates of the thermal updraft at that location; Indicates the location in the target airspace To avoid introducing abstract confidence levels, the adaptive fusion coefficients at a given location are, in one specific implementation, directly determined by the spatial uncertainty plot:

[0104] In the formula, To set the maximum fusion weight, additional abnormal boundary conditions are defined: if the set of active hot cores extracted after the closed-loop task is empty, then the thermal updraft topological prior field is forced to be empty. ; and the fusion coefficient At this point, the final reconstructed field is completely determined by the continuous wind speed interpolation field.

[0105] This invention provides an embodiment of a fixed-wing UAV trajectory planning system based on reconstructed airflow field, comprising: a parameter acquisition module, an airflow model and response construction module, a thermal updraft topology prior field calculation module, a safe target heading instruction acquisition module, and a trajectory planning module; the parameter acquisition module is used to acquire flight state data of the fixed-wing UAV in the target airspace, acquire vertical wind speed observations collected along the exploration trajectory based on the flight state data, and extract the local vertical wind speed gradient direction at the current sampling point on the exploration trajectory, using the sampling point position, vertical wind speed observations, and local vertical wind speed gradient direction on the exploration trajectory as an observation vector set; the airflow model and response construction module is used to construct a multi-core thermal updraft model, based on the multi-core thermal updraft... The updraft model sets up multiple candidate thermonuclear bases within the target airspace and forms a candidate thermonuclear dictionary. Based on the Euclidean distance between sampling points in the historical exploration trajectory and the candidate thermonuclear bases in the candidate thermonuclear dictionary, and a preset mask radius, a local candidate thermonuclear dictionary is obtained. The response value of each local candidate thermonuclear base at each sampling point on the exploration trajectory is obtained based on the position of each local candidate thermonuclear base in the local candidate thermonuclear dictionary and the position of each sampling point on the exploration trajectory. The thermal updraft topology prior field calculation module is used to obtain the peak physical response on the historical exploration trajectory based on the non-negative vertical wind speed observation value and the response value corresponding to the local candidate thermonuclear base. The local candidate thermonuclear bases whose peak physical response is greater than a set physical response threshold are used as the target thermonuclear bases. The strength of the target thermonuclear base is iteratively updated using the expectation-maximization iterative algorithm, and the target thermonuclear base with a converged strength greater than a preset strength threshold is taken as the active thermonuclear base. The thermal updraft topological prior field of the multi-core thermal updraft model is obtained based on the strength of all active thermonuclear bases. The safe target heading command acquisition module is used to obtain the spatial uncertainty map in the target airspace at the current moment based on a preset initial uncertainty and using a Gaussian decay algorithm. A local existence prior field is constructed based on non-negative vertical wind speed observations. The thermal updraft topological prior field, the local existence prior field, and the spatial uncertainty map are fused to obtain the acquisition value map. The relative distance is obtained based on the Euclidean distance from each point in the target airspace to the fixed-wing UAV's position at the current moment. The system generates a dynamic trajectory decay map to suppress local hovering by applying a penalty weight that decreases with time step to the location points of the historical exploration trajectory passing through the target airspace. Based on the acquisition value map, relative distance cost map, dynamic trajectory decay map, heading angle, normalized vertical wind speed observations, and local vertical wind speed gradient direction, it obtains a safe target heading instruction that meets preset flight constraints. The trajectory planning module executes the safe target heading instruction and updates the observation vector set until the set flight mission ends to obtain a complete exploration trajectory. It uses the complete exploration trajectory to construct a continuous wind speed interpolation field and fuses the continuous wind speed interpolation field with the thermal updraft topological prior field to obtain a reconstructed updraft field with continuous vertical wind speed in the target airspace. The trajectory is then planned based on the reconstructed updraft field.

[0106] The present invention will be described below with reference to the accompanying drawings: This invention collects vertical wind speed observations along a restricted exploration trajectory using a fixed-wing UAV. Combined with a candidate thermonuclear dictionary, physical response truncation, and non-negative sparse Bayesian inference, the system can generate a multi-core thermal updraft topological prior in unknown or weakly prior target airspace. This avoids the traditional single-core source-finding strategy's tendency to repeatedly hover near the first discovered thermonuclear region while ignoring other thermonuclear regions. This invention integrates the thermal updraft topological prior field, spatial uncertainty map, and local existence prior field into a data acquisition value map. Furthermore, it generates relative heading increments that conform to fixed-wing flight constraints through a dual-stream active sampling strategy network. This allows the fixed-wing UAV to balance thermonuclear utilization, unknown region exploration, and boundary safety even under conditions where hovering, turning in place, and limited range budget are not possible. This invention does not directly use sparse topological priors as the final result after the loop closure is completed. Instead, it fuses the thermal updraft topological prior field with the continuous interpolation field constructed based on the vertical wind speed observation samples in the complete exploration trajectory. This preserves the macroscopic topological information of the multi-core updraft structure and uses measured samples to express the local subsidence compensation area and background vertical motion changes, thereby improving the stability and physical consistency of the continuous vertical wind speed reconstruction results in the target airspace.

[0107] To further verify the effectiveness and significant progress of the method proposed in this invention, a comparative verification experiment was conducted in an offline digital twin simulation environment, comparing the active sampling strategy proposed in this invention with the traditional lawnmower uniform baseline coverage strategy.

[0108] like Figure 6 and Figure 7 As shown, Figure 6 This is a graph showing the convergence curves of topology inference and global root mean square error of the active sampling strategy in this embodiment of the invention on the task execution time axis; Figure 7 The topology inference and global root mean square error convergence curves of the existing lawnmower-style baseline strategy on the task execution time axis are shown. Figure 6 and Figure 7 The solid gray line represents the inferred number of active heat cores, and the dashed gray line represents the actual number of heat cores (the actual value is 4); the solid blue line represents the trend of the root mean square error (RMSE) of the global wind field reconstruction. Figure 6 It can be seen that the active sampling strategy of this invention stably converges the inferred number of heat cores to the true value of 4 after 37.2 seconds of flight, and the global root mean square error rapidly decreases to below 0.5 m / s after 38.8 seconds. (Comparison) Figure 7It is known that the traditional lawnmower baseline strategy, due to its rigid geometric path, takes 44.5 seconds to achieve convergence of the number of hot cores, and the time to reduce the global error to below 0.5 m / s is as high as 80.0 seconds. This fully demonstrates that the maneuver commands generated by this invention can guide fixed-wing UAVs to explore high-value areas more efficiently, significantly shortening the multi-core topology discovery time.

[0109] like Figure 8 and Figure 9 As shown, Figure 8 The image shows the reconstruction effect of the exploration trajectory (RL path) generated in the target airspace using the method provided by this invention and the continuous vertical wind speed. Figure 9 The image shows the reconstruction results of the sampling trajectory (lawnmower path) and continuous vertical wind speed in the target airspace for the existing lawnmower-based baseline strategy; comparison. Figure 8 and Figure 9 The sampling trajectory shows that, due to the constraints of minimum turning radius and maximum yaw rate, fixed-wing UAVs... Figure 9 Traditional lawnmower strategies often result in sparse and untargeted track distribution, easily missing small or remote updraft cores, leading to topological omissions in wind field reconstruction. However, using... Figure 8 The active sampling strategy of this invention, as illustrated, demonstrates strong environmental adaptability under the drive of value map acquisition by the fixed-wing UAV. Its flight path not only achieves effective airspace exploration but also performs precise penetration and high-frequency sampling near the inferred thermonuclear region. The final reconstruction results show that the method of this invention best matches the actual wind field distribution, with a significantly lower global root mean square error (RMSE) than traditional strategies, greatly improving the completeness and engineering usability of reconstructing continuous vertical wind speed fields in unknown airspace.

[0110] 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 trajectory planning method for a fixed-wing unmanned aerial vehicle (UAV) based on reconstructed airflow field, characterized in that, include: The flight status data of the fixed-wing UAV in the target airspace is obtained. Based on the flight status data, the vertical wind speed observation value collected along the exploration trajectory is obtained, and the local vertical wind speed gradient direction at the current sampling point on the exploration trajectory is extracted. The sampling point position, vertical wind speed observation value and local vertical wind speed gradient direction on the exploration trajectory are used as the set of observation vectors. A multi-core thermal updraft model is constructed, and multiple candidate thermal core bases are set up in the target airspace based on the multi-core thermal updraft model to form a candidate thermal core dictionary; The local candidate thermonuclear dictionary is obtained based on the Euclidean distance between the sampling points in the historical exploration trajectory and the candidate thermonuclear base in the candidate thermonuclear dictionary and the preset mask radius. The response value of each sampling point on the exploration trajectory is obtained based on the location of each local candidate thermonuclear base in the local candidate thermonuclear dictionary and the location of each sampling point on the exploration trajectory. The peak values ​​of physical responses on the historical exploration trajectory are obtained based on the non-negative vertical wind speed observation values ​​and the response values ​​corresponding to the local candidate thermonuclear bases; the local candidate thermonuclear bases when the peak values ​​of physical responses are greater than the set physical response thresholds are taken as the target thermonuclear bases; The strength of the target thermonuclear substrate is iteratively updated using the expectation-maximization iterative algorithm, and the target thermonuclear substrate with a strength greater than a preset strength threshold after convergence is taken as the active thermonuclear substrate. The thermal updraft topological prior field of the multi-core thermal updraft model is obtained based on the intensity of all active thermonuclear substrates. Based on the preset initial uncertainty, a spatial uncertainty map in the target airspace at the current moment is obtained using a Gaussian decay algorithm. A local existence prior field is constructed based on non-negative vertical wind speed observations. The thermal updraft topological prior field, the local existence prior field, and the spatial uncertainty map are fused to obtain a data acquisition value map. A relative distance cost map is obtained based on the Euclidean distance from each point in the target airspace to the fixed-wing UAV's position at the current moment. A penalty weight that decreases with time step is applied to the positions of historical exploration trajectories passing through the target airspace to generate a dynamic trajectory decay map to suppress local hovering. Based on the data acquisition value map, the relative distance cost map, the dynamic trajectory decay map, the heading angle, the normalized vertical wind speed observations, and the local vertical wind speed gradient direction, a safe target heading command that satisfies preset flight constraints is obtained. The system executes safe target heading instructions and updates the observation vector set until the set flight mission ends to obtain a complete exploration trajectory. Based on the historical observation sample set of vertical wind speed in the observation vector set corresponding to the complete exploration trajectory, a continuous wind speed interpolation field is constructed. The continuous wind speed interpolation field is fused with the thermal updraft topological prior field to obtain a reconstructed updraft field with continuous vertical wind speed in the target airspace. Trajectory planning is then performed based on the reconstructed updraft field.

2. The fixed-wing UAV trajectory planning method based on reconstructed airflow field according to claim 1, characterized in that, The expression for the multinuclear thermal updraft model is: In the formula, Indicates the location in the target airspace The actual vertical wind speed at the location; Indicates the first The maximum vertical wind speed at the thermonuclear center; Indicates the first The coordinates of the center position of each thermonucleus; Indicates the first The characteristic radius of each thermonucleus; This represents the total number of thermal nuclei within a multinucleus cluster region in the target airspace; This represents the vertical flow field term in the continuous background.

3. The fixed-wing UAV trajectory planning method based on reconstructed airflow field according to claim 1, characterized in that, The steps for obtaining the local candidate heat core dictionary based on the Euclidean distance between sampling points in the historical exploration trajectory and the candidate heat core substrate in the candidate heat core dictionary, and the preset mask radius are as follows: In the formula, This represents a local candidate heat kernel dictionary within the candidate heat kernel dictionary; Indicates the first The central location corresponding to each candidate thermonuclear substrate; Indicating the first in the historical exploration trajectory The position of the sampling point is added to the candidate thermonuclear dictionary. The Euclidean distance between the center positions of each candidate thermonuclear basis; Indicates the first step on the exploration trajectory Location of each sampling point; Indicates the preset mask radius; This represents the total number of sampling points on the exploration trajectory.

4. The fixed-wing UAV trajectory planning method based on reconstructed airflow field according to claim 1, characterized in that, The expression for the response value of the local candidate thermonuclear base at each sampling point on the exploration trajectory in the local candidate thermonuclear dictionary is: In the formula, Indicates the first in the local candidate thermonuclear dictionary The local candidate thermonuclear basis corresponding to the nth node on the exploration trajectory Each sampling point location The response value at the location; Indicates the first The central location corresponding to each local candidate thermonuclear substrate; The feature radius represents the basis of the candidate heat kernel dictionary; Indicates the first step on the exploration trajectory Location of each sampling point; Indicates the first step on the exploration trajectory The location of the sampling point is added to the local candidate thermonuclear dictionary. The Euclidean distance between the locations of the candidate thermonuclear basement.

5. The fixed-wing UAV trajectory planning method based on reconstructed airflow field according to claim 1, characterized in that, The expression for the peak physical response along the historical exploration trajectory is: In the formula, Indicates the first Peak physical response of a local candidate thermonuclear substrate along the historical exploration trajectory; Indicates the first in the local candidate thermonuclear dictionary The candidate thermonuclear substrate corresponding to the nth node on the exploration trajectory Each sampling point location The response value at the location; Indicates the first step on the exploration trajectory Non-negative vertical wind speed observations at each sampling point location; This represents the total number of sampling points on the exploration trajectory.

6. The fixed-wing UAV trajectory planning method based on reconstructed airflow field according to claim 1, characterized in that, The expression for the topological prior field of the thermal updraft in the multi-core thermal updraft model is as follows: In the formula, express Position in the target airspace at all times The topological prior estimate of the thermal updraft at a given location is composed of the topological prior estimates of the thermal updraft at all locations in the target airspace, forming the topological prior field of the thermal updraft. An active thermonuclear basement set; The first in the active thermonuclear basement set The intensity corresponding to an active thermonuclear substrate; For the first The central location corresponding to an active thermonuclear basement; denoted as the feature radius of the candidate hot kernel dictionary basis.

7. The fixed-wing UAV trajectory planning method based on reconstructed airflow field according to claim 1, characterized in that, The expression for the local existence prior field is: In the formula, express Position of target in airspace at any time The local existence prior probability at a location is formed by traversing the local existence prior probabilities of all locations within the target spatial domain. Indicates the forgetting coefficient; express Position of target in airspace at any time The prior probability of local existence at a given location; Indicates an indicator function; Indicates the location of the target in the airspace; Indicates the first The location of each sampling point; Indicates the updated radius; Indicates the effective rising threshold; Indicates the first step in the exploration trajectory Vertical wind speed observations corresponding to each sampling point.

8. The fixed-wing UAV trajectory planning method based on reconstructed airflow field according to claim 1, characterized in that, The steps for obtaining a safe target heading command that satisfies preset flight constraints based on the collected value map, relative distance cost map, dynamic trajectory attenuation map, heading angle, normalized vertical wind speed observations, and local vertical wind speed gradient direction are as follows: The acquisition value map, relative distance cost map, and dynamic trajectory attenuation map are used as multi-channel image tensors; a state vector is constructed based on heading angle, normalized vertical wind speed observations, and local vertical wind speed gradient direction. Macroscopic spatial geometric features are obtained from image topological flow based on multi-channel image tensors and active sampling strategies. Microscopic kinematic features are obtained from local kinematic flow based on state vectors and utilizing an active sampling strategy; Macroscopic spatial geometric features and microscopic kinematic features are concatenated into a joint state vector. Based on the joint state vector, a discretized relative heading increment is obtained using a two-stream reinforcement learning policy network. The target heading is obtained based on the relative heading increment and the current heading. A safe target heading instruction that satisfies preset flight constraints is constructed based on the direction vector of the target heading and the unit normal vector pointing into the target airspace.

9. The fixed-wing UAV trajectory planning method based on reconstructed airflow field according to claim 1, characterized in that, The expression for the reconstructed updraft field of continuous vertical wind speeds in the target airspace is: In the formula, Indicates the location in the target airspace The reconstructed updraft value of continuous vertical wind speed at a location is composed of the reconstructed updraft value of continuous vertical wind speed at all locations in the target airspace, forming a reconstructed updraft field. Indicates the location in the target airspace Continuous wind speed interpolation at the location; Indicates the first Position of target in airspace at any time Topological prior estimates of the thermal updraft at that location; Indicates the location in the target airspace The adaptive fusion coefficient at the location.

10. A trajectory planning system for a fixed-wing unmanned aerial vehicle (UAV) based on reconstructed airflow field, characterized in that, include: The parameter acquisition module is used to acquire flight status data of fixed-wing UAV in the target airspace, acquire vertical wind speed observation values ​​collected along the exploration trajectory based on the flight status data, and extract the local vertical wind speed gradient direction at the current sampling point on the exploration trajectory. The sampling point position, vertical wind speed observation value and local vertical wind speed gradient direction on the exploration trajectory are used as the set of observation vectors. The airflow model and response construction module is used to construct a multi-core thermal updraft model, and based on the multi-core thermal updraft model, multiple candidate thermal core bases are set in the target airspace and a candidate thermal core dictionary is formed. The local candidate thermonuclear dictionary is obtained based on the Euclidean distance between the sampling points in the historical exploration trajectory and the candidate thermonuclear base in the candidate thermonuclear dictionary and the preset mask radius. The response value of each sampling point on the exploration trajectory is obtained based on the location of each local candidate thermonuclear base in the local candidate thermonuclear dictionary and the location of each sampling point on the exploration trajectory. The thermal updraft topology prior field calculation module is used to obtain the physical response peak value on the historical exploration trajectory based on the non-negative vertical wind speed observation value and the response value corresponding to the local candidate heat core; and to take the local candidate heat core base when the physical response peak value is greater than the set physical response threshold as the target heat core base. The strength of the target thermonuclear substrate is iteratively updated using the expectation-maximization iterative algorithm, and the target thermonuclear substrate with a strength greater than a preset strength threshold after convergence is taken as the active thermonuclear substrate. The thermal updraft topological prior field of the multi-core thermal updraft model is obtained based on the intensity of all active thermonuclear substrates. The safe target heading instruction acquisition module is used to obtain a spatial uncertainty map in the target airspace at the current moment based on a preset initial uncertainty and using a Gaussian decay algorithm; construct a local existence prior field based on non-negative vertical wind speed observations; fuse the thermal updraft topological prior field, the local existence prior field, and the spatial uncertainty map to obtain a data acquisition value map; obtain a relative distance cost map based on the Euclidean distance from each point in the target airspace to the fixed-wing UAV's position at the current moment; apply a penalty weight that decreases with time step to the positions of historical exploration trajectories passing through the target airspace to generate a dynamic trajectory decay map to suppress local hovering; and obtain a safe target heading instruction that satisfies preset flight constraints based on the data acquisition value map, the relative distance cost map, the dynamic trajectory decay map, the heading angle, the normalized vertical wind speed observations, and the local vertical wind speed gradient direction. The trajectory planning module is used to execute safe target heading instructions and update the observation vector set until the set flight mission ends to obtain a complete exploration trajectory. The complete exploration trajectory is used to construct a continuous wind speed interpolation field. The continuous wind speed interpolation field is fused with the thermal updraft topological prior field to obtain a reconstructed updraft field with continuous vertical wind speed in the target airspace. Trajectory planning is performed based on the reconstructed updraft field.

Images