Method, system and storage medium for flight trajectory planning based on random event capture

Abstract

The present invention discloses a flight trajectory planning method based on random event capture, which includes: building a relationship model between flight trajectory and energy consumption based on the historical flight data of the fixed-wing drone; building a random event emulation model in the cruise task based on the camera parameters on the fixed-wing drone and the distribution situation of the event interest points in the cruise area; constraining the three-dimensional trajectory machine of the fixed-wing drone based on the relationship model and the random event model, and performing variable discretization based on the cruise task period, thereby building a target optimization model based on the acceleration in the three-dimensional direction; inputting the cruise task into the target optimization model, and optimizing the flight trajectory of the fixed-wing drone using a reinforcement learning method to obtain a planning result. The present invention further provides a flight trajectory planning system and a storage medium. The method of the present invention can effectively improve the energy efficiency of the event capture of the fixed-wing drone, thereby improving the automation level of the patrol of the fixed-wing drone.

Description

[Technical field]

[0001] The present invention relates to the technical field of drone flight planning, and in particular to a flight trajectory planning method, system, and storage medium based on random event capture. [Background technology]

[0002] In recent years, the development of the drone (Unmanned Aerial Vehicle, UAV) industry has accelerated, and the application of drones has expanded from military to civilian use. According to the "Guiding Opinions of the Ministry of Industry and Information Technology on Promoting and Standardizing the Development of the Civilian Drone Manufacturing Industry", the production value of civilian drones will reach about 180 billion yuan in 2025. In 2018, the China Academy of Communications published the "5G Drone Application White Paper", which details the application scenarios and communication demands of Internet-connected drones in logistics, agriculture and plant protection, patrol, surveying and mapping, live streaming, etc., of which patrol has the largest future market space. At present, patrol drones are emerging in areas that ensure people's personal safety, such as power, roads, and urban safety. Drones are low cost, highly flexible, safe, and less susceptible to the effects of natural environments and terrains, so they can replace people's work in harsh environments and provide better monitoring perspectives and patrol quality.

[0003] Patent document CN116113025A discloses a method for trajectory design and power allocation in a drone cooperative communication network, which includes constructing a power allocation model with the goal that the final virtual surplus energy of a node is closest to the maximum capacity of the battery of the node, and using a reinforcement learning distance reward and punishment algorithm to solve the power allocation model and obtain an optimization scheme for node resource allocation and surplus energy. The method allocates the communication trajectory and power of drones using a reinforcement learning method.

[0004] Patent document CN115877871A discloses a non-zero-sum game drone formation control method based on reinforcement learning, the specific steps are as follows: S1: construct a drone dynamics model, S2: construct a non-zero-sum game formation model, S3: solve the non-zero-sum game formation model constructed in step S2 using reinforcement learning method, S4: design a non-zero-sum game formation controller. The method adopts reinforcement learning method to organize the flight trajectory of drones. Summary of the Invention

[0005] The objective of the present invention is to provide a flight trajectory planning method, system, and storage medium based on random event capture, which can effectively improve the energy efficiency of event capture of fixed-wing drones, thereby improving the automation level of fixed-wing drone patrols.

[0006] In order to achieve the first object of the present invention, a flight trajectory planning method based on random event capture is provided, which includes: Constructing a relationship model between flight trajectory and energy consumption based on historical flight data of a fixed-wing drone; Constructing a random event emulation model in a cruise task based on camera parameters on the fixed-wing drone and a distribution situation of event interest points in a cruise area; Constraining the three-dimensional trajectory of the fixed-wing drone based on the relationship model and the random event model, and performing variable discretization based on the cruise task period, thereby constructing a target optimization model based on the acceleration in the three-dimensional direction; The method includes inputting a cruise task into the target optimization model and optimizing a flight trajectory of the fixed-wing drone using a reinforcement learning method to obtain a planning result.

[0007] The present invention obtains optimal flight planning results by constructing a relationship model between trajectory and energy consumption and a random event model, and employing a reinforcement learning method to solve a target function formed by their combination.

[0008] Specifically, the relationship model specifically calculates the total energy consumption of a fixed-wing drone during a cruise task time as the integral of electric power over the cruise task time.

[0009] Specifically, the relationship model integrates the total energy consumption after the fixed-wing drone completes a given flight trajectory based on the flight time, and the formula is as follows:

number

[0010] Specifically, the generation of each event interest point within the cruise area follows a Poisson distribution.

[0011] Specifically, the random event capture emulation model obtains all events captured by the fixed-wing drone within a cruise task period by judging whether the event of the event interest point in the cruise area is within the camera shooting range of the fixed-wing drone when the event occurs, and the formula is as follows:

number

[0012] Specifically, the reinforcement learning method selects each flight action of the fixed-wing drone to generate a corresponding flight trajectory with the lowest internal energy consumption within a cruise task period under a predetermined state parameter as a reward, the state parameter including the dynamic characteristics of the event interest point, the current position and the current speed of the fixed-wing drone, and the flight action includes a three-dimensional acceleration of the fixed-wing drone at each time, and the formula is as follows:

number

number

[0013] Specifically, the flight operations include turning left at zero pitch angle, turning right at zero pitch angle, flying straight at zero pitch angle, turning left at positive pitch angle, turning right at positive pitch angle, flying in a straight line at positive pitch angle, turning left at negative pitch angle, turning right at negative pitch angle, and flying in a straight line with negative pitch angle.

[0014] Based on the angle between the acceleration direction of the fixed-wing drone and the current flight direction, a flight action is selected and a specific flight trajectory is generated, and the acceleration direction is obtained by fitting with the three-dimensional acceleration obtained by solving the reinforcement learning method.

[0015] Specifically, the direction of travel of the fixed-wing drone is the positive X-axis direction, the right turn direction is the positive Y-axis direction, and the upward direction is the positive Z-axis direction. When the acceleration in the z-axis direction is zero, if the acceleration in the y-axis direction is less than zero, turn left with zero pitch angle.

[0016] When the acceleration in the z-axis direction is zero, if the acceleration in the y-axis direction is greater than zero, then turn right with zero pitch angle.

[0017] When the acceleration in the z-axis direction and the acceleration in the y-axis direction are both equal to zero, the robot moves straight with a zero pitch angle.

[0018] If the acceleration in the z direction is greater than zero and the acceleration in the y direction is less than zero, then turn left at a positive pitch angle.

[0019] If the acceleration in the z direction is greater than zero and the acceleration in the y direction is greater than zero, then turn right at a positive pitch angle.

[0020] When the acceleration in the z-axis direction is greater than zero and the acceleration in the y-axis direction is equal to zero, the aircraft will fly in a straight line with a positive pitch angle.

[0021] If the acceleration in the z direction is less than zero and the acceleration in the y direction is less than zero, then turn left with a negative pitch angle.

[0022] If the acceleration in the z direction is less than zero and the acceleration in the y direction is greater than zero, then turn right with a negative pitch angle.

[0023] When the acceleration in the z-axis direction is less than zero and the acceleration in the y-axis direction is equal to zero, a straight line is flown with a negative pitch angle.

[0024] To achieve the second objective of the present invention, a flight trajectory planning system is provided that is implemented based on the above flight trajectory planning method, and the system includes a random event simulation module, a drone control module, a reinforcement learning module, and a selection and decision module.

[0025] The random event simulation module is used to generate the occurrence and disappearance time points of random events for each event point of interest.

[0026] The reinforcement learning module updates a Q-table of the relationship between each flight operation of the fixed-wing drone and an expected bonus based on the occurrence and disappearance times of random events and the state and position of the fixed-wing drone.

[0027] The selection and decision module selects a flight action for the fixed-wing drone based on a maximizing bonus and constructs a corresponding flight path.

[0028] The drone control module is used to control a fixed-wing drone based on a selected flight maneuver and flight path.

[0029] In order to achieve the third object of the present invention, there is provided a storage medium including a program stored thereon, which, when executed, causes a processor to execute the above-mentioned random event capture based flight trajectory planning method.

[0030] Compared with the prior art, the beneficial effects of the present invention are as follows: Plan a flight trajectory for a fixed-wing drone based on real propulsive energy and three-dimensional free space motion relationships, as well as the occurrence and disappearance of random events.

[0031] At the same time, reinforcement learning methods are used to optimize and analyze the flight behavior of the fixed-wing drone to obtain the optimal flight trajectory. [Brief description of the drawings]

[0032] [Figure 1] FIG. 2 is a schematic diagram of a flight trajectory planning method based on random event capture according to the present embodiment. [Diagram 2] FIG. 2 is a schematic diagram of flight operation regulations in reinforcement learning according to the present embodiment. [Diagram 3] 1 is a diagram showing instantaneous power when the fixed-wing drone of this embodiment flies in a straight line at a constant speed. [Figure 4] This is a relationship diagram of the angle between the speed and acceleration vector of a fixed-wing drone in a turning state in this embodiment. [Diagram 5] FIG. 13 is a diagram showing the relationship between the magnitude of acceleration and power when the fixed-wing drone of this embodiment is in a turning state. [Figure 6] 13 shows training curves for nine flight operations in reinforcement learning for a fixed-wing drone according to this embodiment. [Figure 7] FIG. 13 is a linear relationship diagram between event point of interest density and energy efficiency according to an embodiment; DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0033] The present invention will be further described below with reference to specific examples. It should be understood that these examples are only for the purpose of illustrating the present invention, but do not limit the scope of the present invention. It should also be understood that after reading the teachings of the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent forms are also within the scope defined by the appended claims of this application.

[0034] As shown in FIG. 1, a flight trajectory planning method based on random event capture, comprising: Building a relationship model between trajectory and energy consumption based on historical flight data of a fixed-wing drone; Constructing a random event emulation model in a cruise task based on camera parameters on the fixed-wing drone and a distribution situation of event interest points in a cruise area; Constraining the three-dimensional trajectory of the fixed-wing drone based on the relationship model and the random event model, and performing variable discretization based on the cruise task period, thereby constructing a target optimization model based on the acceleration in the three-dimensional direction; The method includes inputting a cruise task into the target optimization model and optimizing a flight trajectory of the fixed-wing drone using a reinforcement learning method to obtain a planning result.

[0035] More specifically, a set of event interest points

number

[0036] Fixed-wing drone trajectory and energy consumption model: The total energy consumption of a fixed-wing drone includes its propulsion energy and communication energy. Note that in practice, the propulsion energy of a fixed-wing drone is usually much larger than its signal processing energy. Therefore, in this invention, the communication energy is ignored.

[0037] The 3D trajectory of a fixed-wing drone is q(t)=[x(t)y(t)h(t)] T ∈R 3×1 , 0≦t≦T. The velocity and acceleration vectors of a fixed-wing drone are

number

[0038] The fixed-wing drone instantaneous power model is as follows:

number

[0039] Calculate the total energy consumption of a fixed-wing drone over time T as the integral of power P(v,a) over time T:

number

[0040] Event occurrence and capture: Assuming that each event interest point i ∈ O is static on the ground, its position is l i =(x i ,y i ,0). Specifically, each event interest point can represent a hotspot of traffic accidents, a point where jungle fires are likely to occur, etc. As shown in Figure 1, at any given time t ∈ (0,T), events at any event interest point i ∈ O occur independently in space and time. Under this assumption, the event generation process of each event interest point follows a Poisson process. Every event interest point has a random event arrival rate λ and disappearance rate μ, which represents the frequency with which an event occurs and disappears at that location. Let X and Y describe the event dynamic characteristics of the event interest point. X denotes the dwell time of the event, and Y denotes the event arrival time. Note that X and Y follow exponential distribution with mean values ​​of 1 / λ and 1 / μ.

[0041] During the task time T, event interest point i receives a set of n iEvent C i ={C i 1 ,C i 2 ,…,C i ni Consider that event c occurs. i q occurs at event interest point i during time t ∈ T. i q Let (t)∈0,1, which gives us the following function:

number

[0042] The fixed-wing drone is equipped with an image sensor and a stabilized camera can guarantee a circular coverage area, denote by ρ the field of view (FoV) of the image sensor, and the fixed-wing drone at time t has a camera image footprint on the ground with radius r(t) = h(t) tan(ρ / 2). σ i t Let be a binary variable, and let it be 1 if the fixed-wing drone is within the coverage area of ​​the fixed-wing drone at time t. The variable σ i t can be calculated with the following constraints:

number

number

number

[0043] ζ(c i q ,T) is the event capture indicator, i.e., ζ(c i q , T)=1 if the task is captured during time T.

[0044] So the constraints for event capture are:

number

[0045] All events captured by the fixed-wing drone during task period T are

number

[0046] Based on the above two models, a corresponding target optimization model is constructed.

[0047] The surveillance and monitoring performance of fixed-wing drones is primarily limited by the power budget of their onboard batteries. Fixed-wing drones have the advantage of longer cruise times than rotary-wings, but inefficient ballistic designs can lead to rapid battery depletion, i.e., instantaneous power P → ∞, defined as the ratio T of the number of captured events to the total energy consumption within the task time, as v → 0.

[0048] The optimization problem is modeled as a linear split-plot program:

number

[0049] Constraints (8b) and (8c) provide a bounded surveillance region. Constraint (8d) provides a lower altitude limit for communication restrictions and an upper limit for image quality degradation. Given the flight altitude h(t) of a fixed-wing drone, this altitude is the minimum height h(t) to achieve high-quality LoS channel conditions. min According to 3GPP, a minimum flight altitude of 40 meters is required for rural macro scenes and 100 meters for urban macro scenes. (8e) ensures that fixed-wing drones fly at a certain minimum speed to stay airborne and do not exceed a maximum speed.

[0050] On the other hand, since the above optimization problem is complicated to solve directly, to alleviate the problem, the task period T is discretized into equal slots, with n=1,2,...,N as the index. The length of each time step T / N is selected to be sufficient time, and the position of the fixed-wing drone is constant at each time t=nT / N. Therefore, the speed and trajectory are

number

[0051] The above optimization problem can be transformed into:

number

[0052] That is, for each slot n=1,2,...,N, the three-dimensional acceleration a x [n], a y [n] and a z [n] needs to be determined.

[0053] To better solve the above optimization problem, we use reinforcement learning to reduce its complexity, i.e., by executing an action a∈A to move to another state, the bonus function calculates the number of such state-action pairs taken, records it in a Q-table, and the Q-table is initialized to a particular goal state. The model is denoted as a tuple {S, A, {R}, γ}, as follows:

[0054] State, S: Q-learning is used to derive a long-term policy in different environments (e.g., random occurrences and arrivals of events). At each slot n, action (i.e., the vector acceleration of the fixed-wing drone) is calculated as the current system state s n The system state vector s n includes: i) the dynamic features δ of the event interest point s; i q (nT / N);ii) the current position q[n] of the fixed-wing drone;iii) the current velocity vector v[n] of the fixed-wing drone.

[0055] Action, A: For each slot n, action a n changes the speed of the fixed-wing drone. The trajectory of the fixed-wing drone changes with the action (presenting the fixed-wing drone acceleration). The 3D Dubin-inspired path model describes the optimal path as a series of analogies with the fixed-wing drone: "turn right (R)", "turn left (L)", or "go straight (S)" and pitch angles, as shown in Figure 2. A sharp turn angle at high speed for a fixed-wing drone means dissipation of energy. Therefore, let Φ and ψ be small constant steering and pitch angles.

[0056] The fixed-wing drone can randomly take one of nine actions each time: i) turn left with zero pitch angle, ii) turn right with zero pitch angle, iii) go straight with zero pitch angle, iv) turn left with positive pitch angle, v) turn right with positive pitch angle, vi) fly straight with positive pitch angle, vii) turn left with negative pitch angle, viii) turn right with negative pitch angle, ix) fly straight with negative pitch angle. When the fixed-wing drone changes flight direction, it also adjusts its speed. Based on the instantaneous power model of the fixed-wing drone, the energy consumption rate is minimized.

[0057] Reward, R: Current state of reward bonus nWe evaluated the impact of a fixed-wing drone's maneuver plan on the action taken at time step n. The bonus at time n is calculated as the ratio of the number of newly detected events to the propulsive energy at time n:

number

[0058] The proposed Q-learning policy needs to select an action for a particular state to maximize the average reward within a certain time horizon. The UAV selects an action Q(s n ,a n The Q-table is updated after completing each transition and observing the current state-action pair (s,a), and the optimal policy is learned by maintaining a Q-table value, denoted as

number

[0059] where α represents the learning rate that controls the convergence speed, and γ∈(0,1) is the discount factor. In the above model, the agent (UAV) needs to observe the geographical location of the event interest point and the occurrence of that event. At each time step n, based on the Q-table, the agent changes the state s n Select the action that causes the maximum Q value above,

number

[0060] Finally, a fixed-wing drone flight trajectory is planned based on the motion corresponding to the output maximum Q value.

[0061] The present invention further provides a flight trajectory planning system based on the flight trajectory planning method provided by the above embodiment, the system including a random event simulation module, a drone control module, a reinforcement learning module and a selection and decision module.

[0062] The random event simulation module is used to generate the occurrence and disappearance time points of random events for each event point of interest.

[0063] The reinforcement learning module updates a Q-table of the relationship between each flight operation of the fixed-wing drone and an expected bonus based on the occurrence and disappearance times of random events and the state and position of the fixed-wing drone.

[0064] The selection and decision module selects a flight action for the fixed-wing drone based on a maximizing bonus and constructs a corresponding flight path.

[0065] The drone control module is used to control a fixed-wing drone based on a selected flight maneuver and flight path.

[0066] At the same time, a storage medium containing a program stored thereon is further provided, which, when executed, causes a processor to perform the above flight trajectory planning method.

[0067] These computer program instructions may be stored in a computer readable memory that can direct a computer or other programmable data processing device to operate in a particular manner, whereby the instructions stored in the computer readable memory create an article of manufacture that includes a command apparatus that implements the functions specified in one or more flows of the flowcharts and / or one or more blocks of the block diagrams. These computer program instructions may be loaded into a computer or other programmable data processing device, whereby the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more flows of the flowcharts and / or one or more blocks of the block diagrams by causing the computer or other programmable device to perform a series of operational steps to create a computer-implemented process.

[0068] In order to better illustrate the technical proposal according to this embodiment, a simulation evaluation result analysis is provided.

[0069] In the scenario, a fixed-wing drone is tasked with cruising an area of ​​3 × 3 square kilometers and capturing events of interest that occur at 10 locations within a task time of 10 minutes. Each event interest point has an event arrival rate of 0.05 and a disappearance rate of 0.05. The maximum and minimum heights of the fixed-wing drone are set to 500 feet and 40 meters, respectively. The energy consumption parameters c1 and c2 are set to 9.26 × 10 -4 and 2250. The mass of the fixed-wing drone including the payload is 10 Kg. For the Q-learning method, we use a discount factor of 0.1 and a learning rate of 0.4. The length of each time step is 2 seconds.

[0070] We compare the simulation results with a circular trajectory, which is the most common patrol trajectory for fixed-wing drones. We consider each point in the data and calculate the average result over 100 runs.

[0071] As shown in Figure 3, this is the instantaneous power when a fixed-wing drone flies in a straight line at a constant speed. When the speed changes from 5 meters / second to 65 meters / second, the energy consumption rate of the fixed-wing drone first drops from 450 watts to 100 watts. The power reaches its lowest point at 30 meters / second and slowly increases to 194 watts at 65 meters / second. A reasonable guess is that when the moving speed of the fixed-wing drone is lower than 5 meters / second, the output power may be much higher than 450 watts. The battery consumption of the fixed-wing drone is quick. This proves that a different algorithm is required for fixed-wing drones than for rotary-wing drones that normally sense the environment and hover.

[0072] As shown in Figure 4, the angular relationship between the velocity and acceleration vector of a fixed-wing drone in a turning state. This figure shows that the choice of acceleration direction is important, especially for high-speed fixed-wing drones. When ||v||=45 and ||a||=4, the instantaneous power can reach a maximum of 1900 watts and a minimum of 200 watts (10.5% of the maximum power). When the angle between the velocity and the acceleration direction z is 90 degrees, the power consumption of the fixed-wing drone is minimum. When ||v||=15 and ||a||=2, the difference between the minimum and maximum power consumption is not very large. More specifically, the minimum power is 8 watts (when $θ=120$) and the maximum power is 453 watts.

[0073] As shown in Figure 5, the relationship between the magnitude of acceleration and power when the fixed-wing drone is turning. In the simulation, the speed is set to 30 m / s. When θ=0 or θ=180, the fixed-wing drone accelerates or decelerates in the same direction. When θ=90, the fixed-wing drone turns. The results show that high acceleration usually requires more power. Also, turning requires less energy than accelerating in the same direction.

[0074] As shown in Figure 6, the training curves of nine flight operations of fixed-wing drones are: zero pitch angle left turn, zero pitch angle right turn, zero pitch angle straight, positive pitch angle left turn, positive pitch angle right turn, positive pitch angle straight, negative pitch angle left turn and negative pitch angle right turn. Within the first 100 seconds, the nine operations are performed with approximately the same probability of 11%. Then, during learning, each operation increases or decreases gradually until it reaches a stable state. Specifically, when t=100 seconds, S-positive, R-positive, and L-positive increase to about 13.5%, which is more than twice as much as S-positive. In other words, fixed-wing drones are encouraged to fly at higher altitudes. The reason is that fixed-wing drones at higher altitudes cover a larger area. In training, R-negative and L-negative show similar trends. That is, the Q-learning method selects R-negative or S-negative in each 9.1% slot. Fixed-wing drones chose S-negative behavior the least frequently, at 6%. This is because it consumes a lot of energy to adjust speed while keeping the same direction. Also, a negative pitch angle reduces the area that a fixed-wing drone sees.

[0075] As shown in Figure 7, we show how the energy efficiency changes when the event interest point density is from 5 to 50. Experiments are conducted using the proposed Q-learning method and compared with a reference circular trajectory with an altitude of 152 meters and a radius of 500 meters. The average power efficiency within 10 minutes of task time is calculated. Referring to the figure, there is a linear relationship between the event interest point density and the energy efficiency. The reason is that when the density of event interest points is higher, the fixed-wing drone can encounter more events. For the circular trajectory, when there are 5 event interest points, the energy efficiency is 2×10 -4 With 50 event interest points, this results in a 1.6×10 -3 The proposed Q-learning method shows that the energy efficiency is 43% higher than that of the circular trajectory.

[0076] In summary, the present invention aims to deeply study the path optimization problem of fixed-wing drones for patrolling specific events, and design a 3D trajectory construction algorithm under dynamic environments, thereby improving the energy efficiency of event capture and improving the automation level of patrol aircraft.

Claims

1. A flight trajectory planning method based on random event capture, comprising: Constructing a relationship model between flight trajectory and energy consumption based on historical flight data of a fixed-wing drone; Constructing a random event emulation model in a cruise task based on camera parameters on the fixed-wing drone and a distribution situation of event interest points in a cruise area; Constraining the three-dimensional trajectory of the fixed-wing drone based on the relationship model and the random event model, and performing variable discretization based on the cruise task period, thereby constructing a target optimization model based on the acceleration in the three-dimensional direction; and inputting a cruise task into the target optimization model and optimizing the flight trajectory of the fixed-wing drone using a reinforcement learning method to obtain a planning result.

2. The flight trajectory planning method based on random event capture according to claim 1, wherein the historical flight data includes a 3D trajectory of a fixed-wing drone, a weight and wing area of ​​the fixed-wing drone, a speed and acceleration vector during a cruise task, and environmental factors.

3. The relationship model is constructed by integrating the total energy consumption after a fixed-wing drone completes a given flight trajectory based on flight time, and the formula is as follows: [0010] In the formula, q(t) represents the flight trajectory of the fixed-wing drone, v(t) represents the flight speed of the fixed-wing drone at time t, a(t) represents the acceleration of the fixed-wing drone at time t, P(v, a) represents the instantaneous power of the fixed-wing drone, v represents the instantaneous speed of the fixed-wing drone, a represents the instantaneous acceleration of the fixed-wing drone, g represents the gravitational acceleration, m represents the mass of the fixed-wing drone, and c １ and ２ denotes the constant term, 【number】 2. The method of claim 1 , wherein: Δt denotes the time integral of total energy consumption at the completion of a flight trajectory of a fixed-wing drone.

4. The flight trajectory planning method based on random event capture according to claim 1 , wherein the generation of each event interest point in the cruise area all follows a Poisson distribution.

5. The random event capture emulation model obtains all events captured by the fixed-wing drone within a cruise task period by determining whether the event of the event interest point in the cruise area is within the camera shooting range of the fixed-wing drone when the event occurs, and the formula is as follows: [0025] In the formula, N denotes all events captured by the fixed-wing drone within the cruise task period, and ζ(c ｉ ｑ , T) denotes the event capture indicator, i.e., ζ(c ｉ ｑ , T)=1, if the event interest point is captured during task time T, then σ ｉ ｔ indicates the camera's shooting range, and c ｉ ｑ indicates the event occurrence of the event interest point, and (x ｉ ｑ , y ｉ ｑ 2. The method for flight trajectory planning based on random event capture according to claim 1, wherein: t denotes a time, T denotes a cruise task period, and t∈T.

6. The reinforcement learning method selects each flight action of the fixed-wing drone to generate a corresponding flight trajectory with the lowest internal energy consumption within a cruise task period as a reward under a predetermined state parameter, the state parameter including the dynamic characteristics of the event interest point, the current position and the current speed of the fixed-wing drone, and the flight action including the three-dimensional acceleration of the fixed-wing drone at each time, the formula of which is as follows: [0030] In the formula, α denotes the learning rate that controls the convergence speed, and s ｎ indicates the state parameter at time n, and a ｎ indicates the three-dimensional acceleration at time n, a indicates the acceleration, A indicates the flight motion, [0045] indicates the speed of the fixed-wing drone, and a[n] = {a ｘ [n], a ｙ [n], a ｚ 2. The method for flight trajectory planning based on random event capture according to claim 1, wherein {n} denotes a three-dimensional acceleration of the fixed-wing drone, n=1, 2, ... N denotes a time index, T denotes a cruise task period, and T / N denotes a time step.

7. The flight operations include turning left at zero pitch angle, turning right at zero pitch angle, flying straight at zero pitch angle, turning left at positive pitch angle, turning right at positive pitch angle, flying in a straight line at positive pitch angle, turning left at negative pitch angle, turning right at negative pitch angle, and flying in a straight line with negative pitch angle; The flight trajectory planning method based on random event capture as described in claim 6, characterized in that: a flight action is selected and a specific flight trajectory is generated based on the angle between the acceleration direction of the fixed-wing drone and the current flight direction, and the acceleration direction is obtained by fitting with the three-dimensional acceleration obtained by solving with a reinforcement learning method.

8. The target optimization model relaxes the problem by discretizing the task period T into equal slots, and the equation is: [0050] In the formula, a[n]={a ｘ [n], a ｙ [n], a ｚ [n]} indicates the three-dimensional acceleration of the fixed-wing drone, [006] 2. The method for flight trajectory planning based on random event capturing according to claim 1, wherein h[n] denotes the speed of the fixed-wing drone, h[n] denotes the flight height of the fixed-wing drone, T denotes the cruise task period, n=1, 2, ... N denotes the index of time, and T / N denotes the time step.

9. A flight trajectory planning system implemented by the flight trajectory planning method based on random event capture according to any one of claims 1 to 8, comprising: a random event simulation module, a drone control module, a reinforcement learning module, and a selection and decision module; The random event simulation module is used to generate the occurrence and disappearance time points of random events for each event interest point; The reinforcement learning module updates a Q-table of the relationship between each flight operation of the fixed-wing drone and an expected bonus based on the occurrence and disappearance times of the random event and the state and position of the fixed-wing drone; The selection and determination module selects a flight operation of the fixed-wing drone based on a maximization bonus and constructs a corresponding flight path; The drone control module is used to control a fixed-wing drone based on a selected flight motion and flight path.

10. 10. A storage medium containing a program stored thereon, the program, when executed, causing a processor to execute a flight trajectory planning method based on random event capture according to any one of claims 1 to 8.

Images