A UAV path optimization method based on bionic growth optimization process

Abstract

The method for optimizing the path of a UAV based on the bionic growth optimization process comprises the following steps: establishing a perception model of the UAV; establishing a control model of the UAV; and planning the path of the UAV based on the bionic growth optimization process. The present invention integrates artificial intelligence, machine learning, cognitive computing, and event-related potential loop development patterns that imitate the human brain. A perception model is established to enable the UAV to perceive and identify surrounding obstacle information; a control model is established to allow the UAV to select eight basic flight directions on a two-dimensional plane to form an obstacle avoidance strategy; path planning combines the artificial potential field method and the bionic growth mechanism to enable the UAV to perceive obstacles in real time during flight and optimize the navigation path; as flight experience accumulates, the UAV knowledge base is continuously updated to form a more efficient and safe path planning. The present invention significantly improves the adaptability and decision-making ability of the UAV in an unknown environment, and enhances the autonomous obstacle avoidance ability and task execution efficiency of the UAV in a complex environment.

Description

Technical Field

[0001] The present invention relates to the field of intelligent obstacle avoidance technology for unmanned aerial vehicles, and more specifically to a method for optimizing the path of unmanned aerial vehicles based on a bionic growth optimization process. The technical field includes, but is not limited to, artificial intelligence, machine learning, cognitive computing, robot autonomous navigation, and perception and decision-making systems for unmanned aerial vehicles. Specifically, the present invention combines human behavior, neuroscience, and robotics engineering to improve the autonomous obstacle avoidance capability and task execution efficiency of unmanned aerial vehicles in complex environments. Background Art

[0002] Drones are widely used in modern military and civilian fields, including but not limited to reconnaissance, surveillance, logistics, agriculture, search and rescue, etc. However, drones face complex and changing environments when performing tasks, and how to effectively avoid collisions and plan the optimal path has become a key technical issue.

[0003] Traditional path planning methods, such as geometry-based methods and graph search algorithms, usually rely on prior knowledge of the environment and have poor adaptability in dynamic environments. With the development of artificial intelligence technology, path planning methods based on machine learning have gradually become a research hotspot. These methods can optimize path planning by learning historical data.

[0004] However, existing technologies are still limited in simulating the human brain's ability to handle complex environments. As humans grow up, they gradually develop the ability to adapt to complex environments through cognitive functions such as perception, learning, memory and decision-making. Summary of the invention

[0005] In view of the shortcomings of the existing technical methods, the present invention proposes an innovative UAV path optimization method based on a bionic growth optimization process, which specifically includes the following steps:

[0006] Step 1: Establishing UAV perception model;

[0007] The drone is assumed to have a certain perception radius, which is a spherical area. The drone can identify all obstacle information within this range.

[0008] At the initial stage of the mission, the drone needs to perceive the surrounding obstacles and its own status information. The drone perception model is as follows: Assume that the drone's perception range is a spherical area with itself as the origin in a three-dimensional coordinate system, where the x-axis of the three-dimensional coordinate system is the north direction of the geographic coordinate axis, the y-axis is the east direction of the geographic coordinate axis, and the z-axis is the upward direction. The drone's perception radius is defined as d safe , the current position of the drone in space is [x uav ,y uav ,z uav ], the drone can sense the radius area dsafe All obstacle information within [x o ,y o ,z o ]for:

[0009]

[0010] Step 2: Establish the UAV control model;

[0011] Assume that the drone only needs to consider obstacles in a two-dimensional plane when flying in the air; the drone can choose eight basic directions to perform flight maneuvers and form a safe path for obstacle avoidance strategies;

[0012] Assuming that the drone is flying in the air, the horizontal plane is intercepted from the current height of the drone. The drone only needs to avoid all obstacles in the two-dimensional plane to complete safe flight; assuming that the drone is a particle control model, when each flight direction is selected, the drone's flight action behavior Act has a total of 8 directions to choose from:

[0013] Act={UL,U,UR,L,R,DL,D,DR} (2)

[0014] Among them: UL, U, UR, L, R, DL, D, DR represent the eight directions on the two-dimensional plane with the current drone as the origin, namely upper left, upper, upper right, left, right, lower left, lower, and lower right;

[0015] During the flight of the drone, by combining the sequences of these basic flight actions, a safe path of the obstacle avoidance strategy is obtained; that is, the safe flight path obtained by the drone's anti-collision strategy is the sequence A of the drone's flight action behaviors Act on the timeline:

[0016] A={Act1, Act2, ..., Act n} (3)

[0017] Among them: Act1, Act2, ..., Act n The UAV flight actions taken at time 1 to n on the time axis;

[0018] After planning the drone’s path, the drone needs to fly along the established trajectory;

[0019] Step 3: UAV path planning based on bionic growth optimization process;

[0020] The details are as follows:

[0021] Step 1: UAV path planning;

[0022] Assume that the UAV perceives the surrounding obstacle information in real time during flight as the nearest coordinate point of the static obstacle SO = [x SO ,y SO ,z SO ] and the closest coordinate point of the dynamic obstacle MO = [x MO ,y MO ,z MO ], assuming that the target coordinate point of the UAV mission is p tar =[x p ,y p ,z p ], then define the target attraction U of the artificial potential field a and obstacle repulsion U r They are:

[0023] U a =-k a ·[(x p -x uav ) 2 +(y p -y uav ) 2 +(z p -z uav ) 2 ] (4)

[0024]

[0025] Where: k a is the weight coefficient of the target attraction, k r1 and k r2 are the repulsive force weight coefficients of static obstacle SO and dynamic obstacle MO, σ r1 and σ r2 are the exclusion coefficients of the static barrier SO and the dynamic barrier MO, respectively, and exp is a power function;

[0026] Setting||p tar -[x uav ,y uav ,z uav ]||=[(x p -x uav ) 2 +(y p -y uav ) 2 +(z p -z uav ) 2 ],||MO-[x uav ,y uav ,z uav ]||=[(x MO -x uav )2 +(y MO -y uav ) 2 +(z MO -z uav ) 2 ],||SO-[x uav ,y uav ,z uav ]||=[(x SO -x uav ) 2 +(y SO -y uav ) 2 +(z SO -z uav ) 2 ], and the attractive gradient of the potential field is calculated With the repulsive force gradient function They are:

[0027]

[0028] in: Represents the gradient symbol;

[0029] The navigation vector for controlling the UAV is obtained is the sum of functions of the potential field gradient:

[0030]

[0031] Where: i is the number of obstacles within the drone’s sensing area, Represents the gradient function of all repulsive forces within the drone's sensing range Perform superposition and summation, is the repulsive force gradient function of the i-th obstacle;

[0032] Based on the previously calculated gravitational gradient function and repulsive gradient function, the action sequence A on the time axis is obtained according to step one. The UAV performs flight actions according to the action sequence A, and the anti-collision strategy L in the obstacle environment can be obtained.

[0033] STEP 2: Match obstacle pattern with anti-collision strategy mapping;

[0034] When the UAV constructs the mapping relationship in the traditional anti-collision algorithm, there is a one-to-one mapping relationship between the obstacle pattern and the anti-collision strategy; for any obstacle pattern MS = [SO, MO], there is a unique anti-collision strategy L;

[0035] The UAV initially plans its path in an obstacle environment. Due to its unfamiliarity with environmental obstacle information, it uses an anti-collision path planning algorithm of "obstacle discovery - algorithm calculation - reasonable avoidance". At this time, the optimal path can be selected according to the path cost function, and the UAV flies according to the path.

[0036] When the UAV is continuously trained in the environment and accumulates knowledge of the mapping relationship between various obstacle modes MS and anti-collision strategies, the method of "obstacle discovery - reasonable avoidance" is adopted to directly and quickly obtain the safe path L, thereby controlling the flight of the UAV, as follows:

[0037] Within the perception range of the drone, the method of "obstacle discovery - reasonable avoidance" is a matching selection mechanism. According to the threat mode MS = [SO, MO], through the mapping relationship F, based on the previously selected anti-collision strategy L, a more optimized path L' is obtained according to the current obstacle mode MS;

[0038] L'=F(MS,L,KB(MS),KB(L),KB(F)),MS∈Ω,L∈Ω (9)

[0039] Among them: KB(MS), KB(L), KB(F) represent the obstacle pattern knowledge base, the anti-collision strategy knowledge base, and the mapping relationship knowledge base, respectively, which are the gradual accumulation and recording process of various obstacle patterns, anti-collision strategies and the mapping relationship between the two obtained during the continuous training of the UAV in the environment. Ω represents the unknown range, and F(·) represents the mapping relationship;

[0040] When the knowledge base is enriched to a certain extent, the current drone has a high probability of encountering a threat pattern MS′ that is the same or similar to the threat pattern MS in the threat pattern knowledge base. In this case, the drone can directly generate an anti-collision strategy, i.e., a safe flight path, by mapping the optimal path L′ obtained directly based on the current obstacle pattern MS, without long-term algorithm calculation, i.e., the method of “obstacle discovery-reasonable avoidance” shown in formula (9);

[0041] Step 3: Cognitive development through bionic growth optimization pathway;

[0042] A feedback mechanism is introduced to gradually expand the drone’s knowledge base as follows:

[0043] Formula (10) shows the mathematical model of bionic growth optimization path development. According to the current threat mode MS, the UAV obtains a more optimized path L' based on the previously selected safe path L through the mapping relationship F. In this process, the new threat mode increment ΔKB is added each time n (MS), path increment ΔKB n (L), and the mapping relationship increment ΔKBn (F) Add incrementally to the knowledge base KB n (MS),KB n (L),KB n (F) is thus expanded into a richer knowledge base KB n+1 (MS),KB n+1 (L),KB n+1 (F)

[0044]

[0045] Where: L' is the more optimized path obtained by MS based on the current obstacle mode, KB n (MS),KB n (L),KB n (F) represents the increment of obstacle pattern knowledge base, anti-collision strategy knowledge base, and mapping relationship knowledge base, ΔKB n (MS),ΔKB n (L),ΔKB n (F) represents the current threat pattern increment, path increment and mapping relationship increment respectively; under the incremental expansion, the knowledge base is updated and enriched one by one to obtain KB n+1 (MS),KB n+1 (L),KB n+1 (F).

[0046] In step 3, step 2, of an embodiment of the present invention, the drone is in an obstacle environment, and the current threat mode it faces is MS = [SO, MO]. At this time, the drone calculates the anti-collision strategy L through the drone path planning method of Step 1, and a mapping relationship F(·) is formed between the threat mode MS and the safe path L; the obstacle mode, the anti-collision strategy L, and the mapping relationship F(·) are stored in the corresponding obstacle mode knowledge base KB (MS), the anti-collision strategy knowledge base KB (L), and the mapping relationship knowledge base KB (F), respectively. By repeating the above steps in the obstacle environment, the three knowledge bases can be enriched incrementally, and the three relationship knowledge bases can be gradually accumulated.

[0047] The present invention is based on a bionic growth optimization process, imitating the developmental pattern of the event-related potential (ERP) loop of the human brain to achieve efficient and intelligent navigation of UAVs in complex environments.

[0048] To address the limitations of the prior art, the present invention achieves the following three key technical steps: establishing a UAV perception model, establishing a UAV control model, and planning a UAV path based on a bionic growth optimization process:

[0049] These steps together constitute the core of the present invention, which aims to improve the autonomous obstacle avoidance capability and task execution efficiency of drones in complex environments by simulating the developmental pattern of the human brain. Compared with the prior art, the drone path optimization method of the present invention has higher intelligence, adaptability and efficiency, and can significantly improve the operation capability of drones in complex environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 The flow of the UAV path optimization method based on the bionic growth optimization process of the present invention is shown;

[0051] Figure 2 A schematic diagram of the UAV perception model is shown;

[0052] Figure 3 Shows the UAV control strategy block diagram;

[0053] Figure 4 The block diagram of the principle of cognitive development of anti-collision mode in the process of bionic growth optimization of UAV is shown;

[0054] Figure 5 The simulation results of multiple training of UAV collision avoidance based on ERP algorithm are shown. Figure 5 (a) to 5(i) respectively show 9 routes planned by the UAV path optimization method based on the bionic growth optimization process;

[0055] Figure 6 The optimal UAV path after ERP algorithm is shown. DETAILED DESCRIPTION

[0056] The present invention will be described in detail below with reference to the accompanying drawings.

[0057] The UAV path planning method based on bionic growth optimization process mainly includes three steps: establishing the UAV perception model, mapping and matching the obstacle pattern and anti-collision strategy, and UAV path planning based on bionic growth optimization process.

[0058] Step 1: Establish the UAV perception model.

[0059] The present invention first establishes a perception model of the drone, so that the drone can perceive obstacles and its own state in the surrounding environment. The drone is assumed to have a spherical area with a certain perception radius, and can identify all obstacle information within this range.

[0060] At the initial stage of the mission, the drone needs to perceive the surrounding obstacle information and its own status information, such as Figure 1The following is a drone perception model. Assume that the drone’s perception range is a spherical area in a stereo coordinate system with itself as the origin, where the x-axis of the stereo coordinate system is the north direction of the geographic coordinate axis, the y-axis is the east direction of the geographic coordinate axis, and the z-axis is the upward direction. The drone’s perception radius is defined as d safe , the current position of the drone in space is [x uav ,y uav ,z uav ], the drone can sense the radius area d safe All obstacle information within [x o ,y o ,z o ]for:

[0061]

[0062] Step 2: Establish the UAV control model.

[0063] Next, the present invention constructs a control model for the drone, assuming that the drone only needs to consider obstacles within a two-dimensional plane when flying in the air. The drone can choose eight basic directions for flight actions to form a safe path for the obstacle avoidance strategy.

[0064] Assuming that the drone is flying in the air, if we intercept the horizontal plane from the current height of the drone, we can know that the drone only needs to avoid all obstacles in the two-dimensional plane to complete safe flight. Assuming that the drone is a particle control model, when each flight direction is selected, the drone's flight action behavior Act has a total of 8 directions to choose from:

[0065] Act={UL,U,UR,L,R,DL,D,DR} (2)

[0066] Among them: UL, U, UR, L, R, DL, D, and DR represent the eight directions on the two-dimensional plane with the current drone as the origin, namely, upper left, upper, upper right, left, right, lower left, lower, and lower right.

[0067] During the flight of the drone, a safe path of an obstacle avoidance strategy can be obtained by combining the sequence of these basic flight actions. In other words, the safe flight path obtained by the drone's anti-collision strategy is the sequence A of the drone's flight action behaviors Act on the timeline:

[0068] A={Act1, Act2, ..., Act n} (3)

[0069] Among them: Act1, Act2, ..., Act n It is the UAV flight action taken at time 1 to n on the timeline.

[0070] After planning the path of the drone, the drone needs to fly according to the predetermined trajectory. At this time, the control strategy is as follows Figure 3 As shown (the control strategy is well known to those skilled in the art), the planned UAV trajectory passes through the navigation control layer and the action execution layer, and finally transmits the control signal to the servo circuit to control the flight command of the UAV.

[0071] Step 3: UAV path planning based on bionic growth optimization process.

[0072] The present invention realizes the path planning of UAV through the process of bionic growth optimization. The UAV perceives the surrounding obstacle information in real time, and uses the artificial potential field to calculate the target attraction and obstacle repulsion to generate the navigation vector. Through the mapping and matching of obstacle patterns and anti-collision strategies, the UAV can quickly obtain a safe flight path. With the accumulation of flight experience, the knowledge base of the UAV will be continuously updated and optimized, thereby improving the efficiency and safety of path planning.

[0073] Step 1: Drone path planning.

[0074] Assume that the UAV perceives the surrounding obstacle information in real time during flight as the nearest coordinate point of the static obstacle SO = [x SO ,y SO ,z SO ] and the closest coordinate point of the dynamic obstacle MO = [x MO ,y MO ,z MO ], assuming that the target coordinate point of the UAV mission is p tar =[x p ,y p ,z p ], then the target attraction U of the artificial potential field can be defined a and obstacle repulsion U r They are:

[0075] U a =-k a ·[(x p -x uav ) 2 +(y p -y uav ) 2 +(z p -z uav ) 2 ] (4)

[0076]

[0077] Where: k a is the weight coefficient of the target attraction, k r1 and kr2 are the repulsive force weight coefficients of static obstacle SO and dynamic obstacle MO, σ r1 and σ r2 are the exclusion coefficients of static barrier SO and dynamic barrier MO respectively, and exp is a power function.

[0078] Setting||p tar -[x uav ,y uav ,z uav ]||=[(x p -x uav ) 2 +(y p -y uav ) 2 +(z p -z uav ) 2 ],||MO-[x uav ,y uav ,z uav ]||=[(x MO -x uav ) 2 +(y MO -y uav ) 2 +(z MO -z uav ) 2 ],||SO-[x uav ,y uav ,z uav ]||=[(x SO -x uav ) 2 +(y SO -y uav ) 2 +(z SO -z uav ) 2 ], from which the attractive gradient of the potential field can be calculated With the repulsive force gradient function They are:

[0079]

[0080] in: Represents the gradient symbol.

[0081] From this, we can get the navigation vector for controlling the drone is the sum of functions of the potential field gradient:

[0082]

[0083] Where: i is the number of obstacles within the drone’s sensing area, Represents the gradient function of all repulsive forces within the drone's sensing range Perform superposition and summation, is the repulsive force gradient function of the i-th obstacle.

[0084] STEP 2: Match the obstacle pattern with the anti-collision strategy mapping.

[0085] In the initial stage of the algorithm of the UAV path optimization method based on the bionic growth optimization process constructed by the present invention, as Figure 4 As shown in the figure, when the drone constructs the mapping relationship in the traditional anti-collision algorithm, there is a one-to-one mapping relationship between the obstacle pattern and the anti-collision strategy. For any obstacle pattern MS = [SO, MO], there is a unique anti-collision strategy L. With the gradual accumulation of mapping relationship knowledge, in the process of obstacle pattern recognition and pattern matching, an obstacle pattern similar to the current perception can be matched in the obstacle pattern knowledge base. At this time, the anti-collision route can be obtained without going through the traditional path planning algorithm.

[0086] exist Figure 4 In the process, the UAV initially plans a path in an obstacle environment. Due to the unfamiliarity with environmental obstacle information, the anti-collision path planning algorithm of "obstacle discovery - algorithm calculation - reasonable avoidance" is generally adopted (this method is well known to people in this field). The optimal path can be selected according to the path cost function, and the UAV flies according to the path. This method is the operating principle of the algorithm in the upper part of the figure. When the UAV is continuously trained in the environment and accumulates knowledge of the mapping relationship between various obstacle modes MS and anti-collision strategies, it can continuously enrich its own path planning experience knowledge through a bionic growth optimization process similar to that of humans. Then, the method of "obstacle discovery - reasonable avoidance" can be adopted (such as Figure 4 The operation principle of the algorithm in the middle and lower parts is shown), and the safe path L is obtained directly and quickly, thereby controlling the flight of the drone, as follows.

[0087] Within the perception range of the drone, the "obstacle detection - reasonable avoidance" approach is a matching selection mechanism. According to the threat pattern MS = [SO, MO], through the mapping relationship F, based on the previously selected anti-collision strategy L, a more optimized path L' is obtained according to the current obstacle pattern MS.

[0088] L'=F(MS,L,KB(MS),KB(L),KB(F)),MS∈Ω,L∈Ω (9)

[0089] Among them: KB(MS), KB(L), KB(F) represent the obstacle pattern knowledge base, the anti-collision strategy knowledge base, and the mapping relationship knowledge base, respectively, which are the gradual accumulation and recording process of various obstacle patterns, anti-collision strategies and the mapping relationship between them obtained during the continuous training of the UAV in the environment. Ω represents the unknown range, and F(·) represents the mapping relationship.

[0090] For example, when the drone is in an obstacle environment, the current threat pattern it faces is MS = [SO, MO]. At this time, the drone calculates the anti-collision strategy L through the route planning algorithm in Step 1, and a mapping relationship F(·) is formed between the threat pattern MS and the safe path L. The obstacle pattern, anti-collision strategy L, and mapping relationship F(·) are stored in the corresponding obstacle pattern knowledge base KB(MS), anti-collision strategy knowledge base KB(L), and mapping relationship knowledge base KB(F) respectively. By repeating the above steps in an obstacle environment, the three knowledge bases can be enriched incrementally and the three relationship knowledge can be gradually accumulated.

[0091] When the knowledge base is enriched to a certain extent, the current drone has a high probability of encountering a threat pattern MS' that is the same or similar to the threat pattern MS in the threat pattern knowledge base. In this case, it is not necessary to use a long-term algorithm calculation, but directly use the more optimized path L' obtained based on the current obstacle pattern MS, that is, the "obstacle discovery-reasonable avoidance" method shown in formula (9) to directly generate an anti-collision strategy, that is, a safe flight path, through mapping.

[0092] Step 3: Bionic growth optimization path for cognitive development.

[0093] Although the above steps can obtain a safe route through step 3, this does not mean that the mapping relationship can be constructed step by step, nor does it mean that the obtained collision avoidance strategy is the optimal strategy. Therefore, it is necessary to introduce a feedback mechanism to gradually expand the knowledge base of the drone, such as Figure 4 This incremental feedback mechanism allows the drone to update and optimize its knowledge base based on actual experience and results after each flight, thereby gradually improving the efficiency and safety of path planning.

[0094] Formula (10) shows the mathematical model of bionic growth optimization path development. According to the current threat mode MS, the UAV obtains a more optimized path L' based on the previously selected safe path L through the mapping relationship F. In this process, the new threat mode increment ΔKB is added each time n (MS), path increment ΔKB n (L), and the mapping relationship increment ΔKB n (F) Add incrementally to the knowledge base KB n (MS),KBn (L),KB n (F) is thus expanded into a richer knowledge base KB n+1 (MS),KB n+1 (L),KB n+1 (F)

[0095]

[0096] Where: L' is the more optimized path obtained by MS based on the current obstacle mode, KB n (MS),KB n (L),KB n (F) represents the increment of obstacle pattern knowledge base, anti-collision strategy knowledge base, and mapping relationship knowledge base, ΔKB n (MS),ΔKB n (L),ΔKB n (F) represents the current threat pattern increment, path increment, and mapping relationship increment respectively. Under the incremental expansion, the knowledge base is updated and enriched one by one to obtain KB n+1 (MS),KB n+1 (L),KB n+1 (F).

[0097] Based on the introduction of the incremental expansion knowledge base, the mapping relationship between the obstacle pattern of the drone and the anti-collision strategy is many-to-many, that is, the same obstacle pattern may correspond to multiple anti-collision strategies. At this time, the route {L1, L2, L3, ..., L K}It needs to be optimized according to the actual obstacle environment of the drone. After completing the anti-collision behavior, the three knowledge bases will be updated in real time, and the knowledge base will be expanded in an incremental manner (the increments are ΔKB(MS), ΔKB(L), and ΔKB(F)), so as to realize the recognition of obstacle patterns, anti-collision strategies, and mapping relationship networks, as well as the developmental bionic growth of the knowledge base.

[0098] In unfamiliar environments, drones generally use the anti-collision planning algorithm of "obstacle discovery - algorithm calculation - reasonable avoidance", which results in a long safe path planning time, affecting flight safety. After continuous anti-collision training and learning in the environment, the idea of ​​bionic growth optimization is adopted. After anti-collision training, three knowledge bases are identified and enriched through incremental feedback development. On the basis of the incremental expansion of the knowledge base method, drones can form a "obstacle discovery - reasonable avoidance" mode, and finally realize the drone path planning method based on the bionic growth optimization process, ensuring that drones can quickly obtain safe flight paths in dense obstacle environments and ensure the flight safety of drones. Specific embodiments

[0100] The starting point of the drone is set as the coordinate origin, and the target point it needs to execute is (200,200). During the flight, the drone needs to continuously avoid obstacles to ensure that it can perform the mission safely. Figure 5 The paper shows nine routes planned by the UAV path optimization method based on the bionic growth optimization process. It can be seen that the paths planned each time are different. This is because random variables are introduced into the algorithm, which allows the UAV's route to have a certain degree of random disturbance.

[0101] After 100 cycles of the growth optimization method, the optimal path is as follows Figure 6 As shown in the figure, it can be observed that the optimal path is basically to fly along the edge of the obstacle, which is similar to the behavior of a person quickly avoiding collisions in an obstacle environment. In order to ensure flight safety and efficiency of mission completion, the drone adopts a balancing strategy.

[0102] Based on such biological inspiration, this paper proposes a UAV path optimization method based on the bionic growth optimization process, which aims to simulate the development process of humans in a multi-obstacle environment by building an ERP loop similar to the human brain, from frequent collisions in infancy to skilled obstacle avoidance in adulthood, and then to the optimal path selection in old age. The UAV learns and optimizes its obstacle avoidance strategy through a similar process. This method not only improves the obstacle avoidance efficiency of the UAV, but also enhances its adaptability and decision-making ability in unknown environments.

Claims

1. A method for optimizing the path of an unmanned aerial vehicle based on a bionic growth optimization process, characterized in that: The specific steps include: Step 1: Establishing the UAV perception model; The drone is assumed to have a certain perception radius, which is a spherical area. The drone can identify all obstacle information within this range. At the initial stage of the mission, the drone needs to perceive the surrounding obstacles and its own status information. The drone perception model is as follows: Assume that the drone's perception range is a spherical area with itself as the origin in a three-dimensional coordinate system, where the x-axis of the three-dimensional coordinate system is the north direction of the geographic coordinate axis, the y-axis is the east direction of the geographic coordinate axis, and the z-axis is the upward direction. The drone's perception radius is defined as d safe , the current position of the drone in space is [x uav ,y uav ,z uav ], the drone can sense the radius area d safe All obstacle information within [x o ,y o ,z o ]for: Step 2: Establish the UAV control model; Assume that when a drone is flying in the air, it only needs to consider obstacles within a two-dimensional plane; The drone can choose eight basic directions for flight maneuvers, forming a safe path for obstacle avoidance strategies; Assuming that the drone is flying in the air, the horizontal plane is intercepted from the current height of the drone. The drone only needs to avoid all obstacles in the two-dimensional plane to complete safe flight; assuming that the drone is a particle control model, when each flight direction is selected, the drone's flight action behavior Act has a total of 8 directions to choose from: Act={UL,U,UR,L,R,DL,D,DR} (2) Among them: UL, U, UR, L, R, DL, D, DR represent the eight directions on the two-dimensional plane with the current drone as the origin, namely upper left, upper, upper right, left, right, lower left, lower, and lower right; During the flight of the drone, by combining the sequences of these basic flight actions, a safe path of the obstacle avoidance strategy is obtained; that is, the safe flight path obtained by the drone's anti-collision strategy is the sequence A of the drone's flight action behaviors Act on the timeline: A={Act1,Act2,...,Act n } (3) Among them: Act1, Act2, ..., Act n The UAV flight actions taken at time 1 to n on the time axis; After planning the drone’s path, the drone needs to fly along the established trajectory; Step 3: UAV path planning based on bionic growth optimization process; The details are as follows: Step 1: UAV path planning; Assume that the UAV perceives the surrounding obstacle information in real time during flight as the nearest coordinate point of the static obstacle SO = [x SO ,y SO ,z SO ] and the closest coordinate point of the dynamic obstacle MO = [x MO ,y MO ,z MO ], assuming that the target coordinate point of the UAV mission is p tar =[x p ,y p ,z p ], then define the target attraction U of the artificial potential field a and obstacle repulsion U r They are: U a =-k a ·[(x p -x uav ) 2 +(y p -y uav ) 2 +(z p -z uav ) 2 ] (4) Where: k a is the weight coefficient of the target attraction, k r1 and k r2 are the repulsive force weight coefficients of static obstacle SO and dynamic obstacle MO, σ r1 and σ r2 are the exclusion coefficients of the static barrier SO and the dynamic barrier MO, respectively, and exp is a power function; Setting||p tar -[x uav ,y uav ,z uav ]||=[(x p -x uav ) 2 +(y p -y uav ) 2 +(z p -z uav ) 2 ],||MO-[x uav ,y uav ,z uav ]||=[(x MO -x uav ) 2 +(y MO -y uav ) 2 +(z MO -z uav ) 2 ],||SO-[x uav ,y uav ,z uav ]||=[(x SO -x uav ) 2 +(y SO -y uav ) 2 +(z SO -z uav ) 2 ], and the attractive gradient of the potential field ▽U is calculated a and the repulsive force gradient function ▽U r They are: ▽U a =-2·k a ·(p tar -[x uav ,y uav ,z uav ]) (6) Among them: ▽ represents the gradient symbol; The navigation vector for controlling the UAV is obtained is the sum of functions of the potential field gradient: Where: i is the number of obstacles within the drone’s sensing area, Represents the gradient function of all repulsive forces within the drone's sensing range ▽U r Perform superposition and summation, is the repulsive force gradient function of the i-th obstacle; Based on the previously calculated gravitational gradient function and repulsive gradient function, the action sequence A on the time axis is obtained according to step 1. The UAV performs flight actions according to the action sequence A, and the anti-collision strategy L in the obstacle environment can be obtained; STEP 2: Match obstacle pattern with anti-collision strategy mapping; When the UAV constructs the mapping relationship in the traditional anti-collision algorithm, there is a one-to-one mapping relationship between the obstacle pattern and the anti-collision strategy; for any obstacle pattern MS = [SO, MO], there is a unique anti-collision strategy L; The UAV initially plans its path in an obstacle environment. Due to its unfamiliarity with environmental obstacle information, it uses an anti-collision path planning algorithm of "obstacle discovery - algorithm calculation - reasonable avoidance". At this time, the optimal path can be selected according to the path cost function, and the UAV flies according to the path. When the UAV is continuously trained in the environment and accumulates knowledge of the mapping relationship between various obstacle modes MS and anti-collision strategies, the method of "obstacle discovery - reasonable avoidance" is adopted to directly and quickly obtain the safe path L, thereby controlling the flight of the UAV, as follows: Within the perception range of the drone, the method of "obstacle discovery - reasonable avoidance" is a matching selection mechanism. According to the threat mode MS = [SO, MO], through the mapping relationship F, based on the previously selected anti-collision strategy L, a more optimized path L' is obtained according to the current obstacle mode MS; L'=F(MS,L,KB(MS),KB(L),KB(F)),MS∈Ω,L∈Ω (9) Among them: KB(MS), KB(L), KB(F) represent the obstacle pattern knowledge base, the anti-collision strategy knowledge base, and the mapping relationship knowledge base, respectively, which are the gradual accumulation and recording process of various obstacle patterns, anti-collision strategies and the mapping relationship between the two obtained during the continuous training of the UAV in the environment. Ω represents the unknown range, and F(·) represents the mapping relationship; When the knowledge base is enriched to a certain extent, the current drone has a high probability of encountering a threat pattern MS′ that is the same or similar to the threat pattern MS in the threat pattern knowledge base. In this case, the optimal path L′ can be directly obtained based on the current obstacle pattern MS without long-term algorithm calculation, that is, the "obstacle discovery-reasonable avoidance" method shown in formula (9). The anti-collision strategy, that is, the safe flight path, can be directly generated through mapping; Step 3: Cognitive development through bionic growth optimization pathway; A feedback mechanism is introduced to gradually expand the drone’s knowledge base as follows: Formula (10) shows the mathematical model of bionic growth optimization path development. According to the current threat mode MS, the UAV obtains a more optimized path L' based on the previously selected safe path L through the mapping relationship F. In this process, the new threat mode increment ΔKB is added each time n (MS), path increment ΔKB n (L), and the mapping relationship increment ΔKB n (F) Add incrementally to the knowledge base KB n (MS),KB n (L),KB n (F) is thus expanded into a richer knowledge base KB n+1 (MS),KB n+1 (L),KB n+1 (F) Where: L' is the more optimized path obtained by MS based on the current obstacle mode, KB n (MS),KB n (L),KB n (F) represents the obstacle pattern knowledge base, anti-collision strategy knowledge base, and mapping relationship knowledge base of the nth step, ΔKB n (MS),ΔKB n (L),ΔKB n (F) represents the current threat pattern increment, path increment and mapping relationship increment respectively; under the incremental expansion, the knowledge base is updated and enriched one by one to obtain KB n+1 (MS),KB n+1 (L),KB n+1 (F).

2. The method for optimizing the path of a UAV based on a bionic growth optimization process according to claim 1, characterized in that: In step 3, step 2, the UAV is in an obstacle environment and the current threat mode it faces is MS = [SO, MO]. At this time, the UAV calculates the anti-collision strategy L through the UAV path planning method in Step 1, and a mapping relationship F(·) is formed between the threat mode MS and the safe path L; the obstacle mode, anti-collision strategy L and mapping relationship F(·) are stored in the corresponding obstacle mode knowledge base KB (MS), anti-collision strategy knowledge base KB (L), and mapping relationship knowledge base KB (F), respectively. By repeating the above steps in the obstacle environment, the three knowledge bases can be enriched incrementally and the three relationship knowledge can be gradually accumulated.

Images