Unmanned aerial vehicle safety path calculation method based on comprehensive risk balance strategy

Abstract

The application discloses a kind of based on comprehensive risk balance strategy's unmanned plane safety path calculation method, belong to the technical field of unmanned plane path planning;It includes: calculating the three-dimensional ground safety risk level of unmanned plane;Based on the three-dimensional ground safety risk level of unmanned plane, and in combination with three-dimensional airspace constraint condition, the comprehensive cost of unmanned plane path is optimized, and the flight path of unmanned plane is obtained;In combination with risk drop priority A* algorithm and dynamic risk weight coefficient optimization unmanned plane flight path, the optimal flight path of unmanned plane is obtained;The application uses improved A* algorithm, can improve search efficiency, and ensure that path safety meets target safety level;By introducing dynamic risk weight coefficient, the trade-off between risk and distance is dynamically adjusted in the algorithm, to ensure that in different risk areas can adaptively adjust path search strategy, to obtain the best path.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of unmanned aerial vehicle path planning, and particularly relates to a safe path calculation method for unmanned aerial vehicles based on a comprehensive risk balancing strategy. BACKGROUND

[0002] With the continuous expansion of the application scenarios of unmanned aerial vehicles, how to ensure public safety during the flight of unmanned aerial vehicles has become one of the important issues, especially in the complex airspace environment of cities, the planning of flight paths and the management of risks to the ground are particularly important.

[0003] Existing research has made positive exploration in the field of safe path planning for unmanned aerial vehicles in urban airspace, but there are still deficiencies. Some research only considers collision risks and does not reduce the loss of ground personnel casualties caused after a failure from the perspective of path planning; some research can plan effective paths that comprehensively consider risks, but it is difficult to balance safety and efficiency at the same time; some research only verifies and compares unmanned aerial vehicle path planning methods in simulation scenarios, which cannot guarantee the applicability of the actual complex flight environment in cities. In fact, future urban low-altitude airspace operations will face more complex operating environments, more flexible flight tasks, more diverse flight subjects, and higher operating densities, and many other difficulties and challenges, which further pose great challenges to unmanned aerial vehicle path planning. It is necessary to consider typical elements of urban airspace, complex operating environments, flight risk costs, and unmanned aerial vehicle flight performance, and design unmanned aerial vehicle operation planning methods for urban safety to meet the safety operation needs of unmanned aerial vehicles.

[0004] Existing path planning algorithms, such as the A* algorithm, usually search for paths in a two-dimensional plane, but for complex three-dimensional urban airspace, especially when considering ground risk factors, the performance of traditional algorithms is limited and it is difficult to balance flight efficiency and flight safety. Therefore, how to consider multiple constraint conditions such as flight altitude, turning angle, path length, ground risk, and available airspace in a complex three-dimensional airspace to find an optimal flight path that is both safe and efficient has become a technical problem to be solved in the field of unmanned aerial vehicle path planning. SUMMARY

[0005] The purpose of the present application is to solve the problem that traditional unmanned aerial vehicle path planning algorithms are difficult to balance flight efficiency and flight safety by providing a safe path calculation method for unmanned aerial vehicles based on a comprehensive risk balancing strategy.

[0006] To achieve the above-mentioned purpose, the technical solution adopted by the present application is:

[0007] A safe path calculation method for unmanned aerial vehicles based on a comprehensive risk balancing strategy, comprising the following steps:

[0008] S1, calculate a three-dimensional-to-ground safety risk level of the unmanned aerial vehicle;

[0009] S2, based on the three-dimensional-to-ground safety risk level of the unmanned aerial vehicle, and in combination with a three-dimensional airspace constraint condition, optimize a comprehensive cost of a flight path of the unmanned aerial vehicle, to obtain the flight path of the unmanned aerial vehicle;

[0010] S3, in combination with a risk reduction priority A* algorithm and a dynamic risk weight coefficient, optimize the flight path of the unmanned aerial vehicle, to obtain an optimal flight path of the unmanned aerial vehicle.

[0011] Further, in the S1, the three-dimensional-to-ground safety risk level of the unmanned aerial vehicle is calculated, including:

[0012]

[0013] In the formula, R n,m,l represents a three-dimensional-to-ground risk matrix of the unmanned aerial vehicle; R(i,j,h) represents an equivalent safety level of a risk of the unmanned aerial vehicle to the ground at a coordinate (i,j,h) position:

[0014] R(i,j,h)=P(LOC)·R f (i,j,h)

[0015] In the formula, P(LOC) represents a probability of failure and crash of the unmanned aerial vehicle when flying in the air, R f (i,j,h) represents a final risk value of the unmanned aerial vehicle to the ground when crashing on the ground in a grid with a position (i,j,h).

[0016] Further, in the S2, the comprehensive cost of the flight path of the unmanned aerial vehicle is optimized, including:

[0017]

[0018] In the formula, Cost Path represents a comprehensive cost of the flight path of the unmanned aerial vehicle, Path={RP1,RP2,…,RP k} represents a path of the unmanned aerial vehicle, RP k is a node; Cost r r=1,…,k is a comprehensive cost of the path of the unmanned aerial vehicle on RP r-1 RP r ; represents a normalized distance cost; represents a normalized risk cost; h low , h high respectively represent a lower limit and an upper limit of a flight height of the unmanned aerial vehicle, h UAV represents the flight height of the unmanned aerial vehicle; θ UAV represents a flight direction change angle of the unmanned aerial vehicle in the three-dimensional space; represents a vector from the route node RP r-1 to RP r , represents the Euclidean distance that the UAV flies from the route node RP r-1 to RP r ; d norm represents a normalized distance; r norm represents a normalized risk value of the path; v UAV represents a cruising speed of the UAV; R ELOS represents an equivalent safety level of the planned path; t r represents the time that the UAV flies from the route node RP r-1 to RP r ; R TLOS represents a target safety level; U n,m,l represents a spatial set used by the UAV to fly.

[0019] Further, the S3 specifically comprises the following sub-steps:

[0020] S31, initializing a UAV data set;

[0021] S32, creating an open list and a closed list, and adding a starting point of a UAV safety path to the closed list;

[0022] S33, adding a node adjacent to the starting point to the open list, selecting a node corresponding to a minimum total value function value to expand, adding the node to the closed list, and determining a search direction;

[0023] S34, according to the search direction, expanding a node by using a method of dynamically adjusting a risk weight coefficient;

[0024] S35, cyclically executing S34; until when a target point is selected as a current node, outputting an optimal path, and stopping executing S34; or when the open list is empty, there is no effective optimal path from the starting point to the target point, and stopping executing S34.

[0025] Further, in the S33, the total value function is:

[0026] F'(r) = G'(r) + H'(r)

[0027] wherein,

[0028] In the formula, F'(r) represents a total value function; G'(r) represents an actual cost function; H'(r) represents a heuristic function; G'(r-1) represents an actual cost function from a starting point to a parent node RP r-1 of a current node; A G(r) represents the dynamic risk weight coefficient of the actual cost function G'(r); A H (r) represents the dynamic risk weight coefficient of the heuristic function H'(r); H dist (r) represents the normalized expected distance cost function; H risk (r) represents the normalized expected risk cost function.

[0029] Further, the normalized expected distance cost function H dist (r) and the normalized expected risk cost function H risk (r) are respectively:

[0030]

[0031] wherein,

[0032]

[0033] wherein, RP r and GOAL represent the three-dimensional coordinates of the current node and the target node respectively; R esti (r) represents the expected risk evaluation from the current node RP r to the target node GOAL, t esti is the expected flight time; R represents the average risk value of all nodes within the search space range; represents the median of the risk distribution of all nodes within the range; n P represents the number of all nodes within the search space range; R(x, y, z) represents the risk value set within the search space range; (i r , j r , h r ) and (i g , j g , h g ) represent the three-dimensional coordinates of the node RP r and the node GOAL respectively.

[0034] Further, the S34 specifically comprises the following steps:

[0035] S341, adding several potential nodes of the current node RP r in the direction to the open list;

[0036] S342, calculating the dynamic risk weight coefficient corresponding to the actual cost function of the potential node;

[0037] S343, if the node RP q (q = 1, …, r) in the open list, then determining the node RP qthe optimal parent node of the current node RP r the actual cost function G'(r) of the actual parent node to the node RP q q the actual cost function G'(r) of the actual parent node to the node RP q the cost information and the parent node index are updated;

[0038] S344, the dynamic risk weight coefficient corresponding to the heuristic function of the potential node is calculated;

[0039] S345, it is judged whether the expected path equivalent safety level corresponding to the potential node meets the target safety level, if yes, S346 is executed, and if not, the corresponding potential node is added to the closed list;

[0040] S346, the total cost function value of all potential nodes in the open list is calculated, the node with the minimum total cost function value is selected and expanded, and is recorded as RP r+1 , and RP r+1 is added to the closed list;

[0041] S347, the node count parameter r is updated as r = r + 1.

[0042] Further, in the S342, the dynamic risk weight coefficient Α G (r) corresponding to the actual cost function of the potential node is calculated as follows:

[0043]

[0044] In the formula, α low and α high represent the lower limit and the upper limit of the value range interval of Α G (r); and λ represents the correction coefficient of the dynamic weight with r norm .

[0045] Further, in the S344, the dynamic risk weight coefficient Α H (r) corresponding to the heuristic function of the potential node is calculated as follows:

[0046]

[0047] Further, in the S345, the expected path equivalent safety level corresponding to the potential node meets the target safety level, including:

[0048] R ELOS ≤ R TLOS

[0049]

[0050] The unmanned aerial vehicle safety path calculation method based on the comprehensive risk balance strategy has the following beneficial effects:

[0051] The application combines building distribution, population density and risk mitigation measures to accurately calculate the risk to the ground and take it as an important constraint condition in path planning; in the three-dimensional space, considering various constraint conditions, including flight height, turning angle, path length, risk to the ground and other factors, the safest and most efficient flight path is found by optimizing the comprehensive cost of the path, including distance cost and risk cost; and the traditional path planning method is improved based on the improved A* algorithm, a composite cost function and heuristic function suitable for three-dimensional risk space are designed, so as to improve the search efficiency and ensure that the path safety meets the target safety level; at the same time, by introducing a dynamic risk weight coefficient, the trade-off between risk and distance is dynamically adjusted in the algorithm, so as to adaptively adjust the path search strategy in different risk areas to obtain the best path. BRIEF DESCRIPTION OF DRAWINGS

[0052] Figure 1 The building point cloud set in the discrete three-dimensional space of the application;

[0053] Figure 2 The flight path in the discrete three-dimensional space of the application;

[0054] Figure 3 The three-dimensional risk-to-ground matrix representation of the application;

[0055] Figure 4 The potential waypoint node of the current node in the discrete three-dimensional space of the application;

[0056] Figure 5 The expected risk cost statistical range of the improved heuristic function of the application;

[0057] Figure 6 The improved RDP-A* algorithm flow of the application;

[0058] Figure 7 The path planning algorithm result of the application in the three-dimensional space;

[0059] Figure 8 The path planning algorithm result of the application in the risk distribution diagram.

[0060] Figure 9 The flowchart of the unmanned aerial vehicle safety path calculation method of the application based on the comprehensive risk balancing strategy. DETAILED DESCRIPTION

[0061] The specific embodiments of the present application are described below to facilitate the understanding of the present application for those skilled in the art, but it should be clear that the present application is not limited to the scope of the specific embodiments, and for those skilled in the art, it is obvious that various changes are within the spirit and scope of the present application defined and determined by the appended claims, and all the inventions utilizing the concept of the present application are within the scope of protection.

[0062] Embodiment 1

[0063] The unmanned aerial vehicle safety path calculation method based on the comprehensive risk balance strategy provided in this embodiment can optimize the comprehensive cost of the flight path under the premise of meeting the safety requirements when ensuring the unmanned aerial vehicle to fly in the urban airspace; by considering the flight height, turning angle, path length, risk to the ground and available airspace and other factors, based on the risk descent priority A* algorithm (Risk Descent Priority A*, hereinafter referred to as RDP-A*), following the principle of priority risk reduction, using the strategy of dynamically adjusting the risk weight coefficient, priority is given to avoiding high-risk areas to reduce the risk of injury to ground personnel caused by the unmanned aerial vehicle falling, efficient and safe path planning is realized, and Figure 6 and Figure 9 , which specifically includes the following contents:

[0064] Step S1, calculating the three-dimensional risk level of the unmanned aerial vehicle to the ground, which specifically includes the following contents:

[0065] The urban airspace is the basis for the operation of the unmanned aerial vehicle. In order to characterize the airspace operation environment, the urban three-dimensional airspace is discretized by using a grid, the grid intersection points in the range of n x m x l are taken as nodes and numbered. The discretization of the three-dimensional space is represented as a set of uniformly distributed three-dimensional point clouds, and the matrix P n,m,l is described, where P(i,j,h) (i=1,…,n; j=1,…,m; h=1,…,l) represents a point in the three-dimensional space with coordinates (i,j,h).

[0066]

[0067] The buildings in the urban low-altitude environment are regarded as static obstacles for the operation of the unmanned aerial vehicle. For high-rise buildings, the unmanned aerial vehicle can fly around, and for low-rise buildings, the unmanned aerial vehicle can fly across. In the three-dimensional discrete space, P(i,j,h)=1 represents the points occupied by the urban buildings, and P(i,j,h)=0 represents the space points not occupied by the urban buildings. The matrix B n,m,l represents the node set occupied by the urban buildings, as shown in Figure 1 .

[0068]

[0069] where B(i,j,h) represents a point in the building node set B n,m,l with coordinates (i,j,h).

[0070] Taking the complement of matrix B n,m,l , we get the set of space that can be used for UAV flight, denoted as matrix U n,m,l . All the nodes in set U n,m,l are the potential flight path nodes that can be used by UAV. Let a series of path nodes RP r represent the flight path of UAV, then RP r ∈ U n,m,l (r = 1, …, k). For set U n,m,l , let U(i,j,h) = 1 represent that the path of UAV passes through this node, and U(i,j,h) = 0 represent that the path of UAV does not pass through this node, so the path of UAV can be represented as Path = U n,m,l (for all U(i,j,h) = 1 points), as shown in Fig. 2. Figure 2

[0071] Path = {RP1, RP2, …, RP k} (3)

[0072] where Path represents the path of UAV, and RP k represents the kth path node.

[0073] The ground risk of UAV represents the expected number of casualties caused by the failure of UAV system in the air and crashing on the ground of the city, which is the most important impact of UAV flight in urban airspace on public safety. Referring to Figure 3 , in the three-dimensional urban airspace P n,m,l , a three-dimensional ground risk matrix R n,m,l of the same size is constructed:

[0074]

[0075] where R(i,j,h) represents the equivalent level of safety (ELOS) of UAV at the coordinate (i,j,h), which means that the expected number of casualties caused by the crash of UAV at this coordinate is R(i,j,h), and the unit is the number of casualties per flight hour, which can be calculated by formula (5):

[0076] R(i,j,h) = P(LOC) · R f (i,j,h) (5)​

[0077] In the formula: P(LOC) represents the probability of the UAV failing and crashing in the air, R f (i,j,h) represents the final risk-to-ground value of the UAV crashing on the ground in the grid with position (i,j,h), which means the expected casualties of the ground personnel in the grid when they are hit by the crashed UAV in the normal operating environment, which can be represented by the final risk-to-ground matrix R f i and the risk mitigation effect matrix M of the UAV:

[0078] R f = R i ⊙M (6)

[0079] The initial risk-to-ground R i of the UAV is an inherent property of the UAV operation, which is only related to the characteristic parameters of the UAV (maximum characteristic size, maximum take-off mass, cruise speed) and the ground population density of the crash area. The risk mitigation matrix M represents the effect of risk mitigation measures, which is divided into M1 and M2 types. M1 type is to reduce the number of people directly exposed to the risk area, which can be calculated by quantifying the protection effect of ground buildings and other elements on people hit by the UAV. M2 type is to reduce the impact of the UAV hitting the ground. The widely concerned method at present is to install a parachute on the UAV. By analyzing the kinematics process of the parachute system after the UAV fails in the air, the mitigation effect of the M2 measure is represented.

[0080] Step S2, based on the three-dimensional safety risk level of the UAV, and combined with the three-dimensional airspace constraint conditions, the path comprehensive cost of the UAV is optimized to obtain the flight path of the UAV, which specifically includes the following contents:

[0081] The problem of UAV safety path planning based on risk constraints is essentially to find an optimal path with the minimum path comprehensive cost from the starting point to the target point in the available three-dimensional space containing risk factors. The goal is to optimize the path comprehensive cost of the UAV path, which is represented by formula (7). The factors that constrain the path comprehensive cost of the UAV flight path include flight height, turning angle, risk-to-ground, path length, and available airspace, etc.

[0082]

[0083] In the formula: Cost Path is the path comprehensive cost of the UAV flight path, Cost r (r = 1, …, k) is the path comprehensive cost of the UAV on the RP r-1 RP r segment.​

[0084] The constraints for the UAV to find the optimal path in the three-dimensional risk space in this embodiment include:

[0085] 1) Flight altitude constraints: In this invention, the flight altitude range that can be used in urban airspace is set as [h low ,h high During the operational phase (excluding takeoff and landing), the drone should fly within this altitude range, while needing to bypass or fly over buildings of different heights.

[0086] 2) Turning Angle Constraint: This invention takes the much-discussed multi-rotor UAV as the research object. Considering the performance limitations of the UAV, the search angle range of the path planning algorithm is constrained, that is, the angle θ of the UAV's flight direction change in three-dimensional space. UAV ∈[0,π / 3].

[0087] In discrete three-dimensional space, for each waypoint RP r It has 26 neighboring nodes that can be selected as the next potential waypoint. To improve algorithm performance and avoid the path planning algorithm getting stuck in local optima at certain special nodes, while also considering the performance limitations of the UAV, the search angle range of the path planning algorithm is constrained, i.e., the angle θ of the UAV's flight direction change. UAV ∈[0,π / 3], in the model used in this invention, the angle of change of the UAV's flight direction can be calculated by formula (8).

[0088]

[0089] In the formula: and These represent the directions from waypoint RP. r-1 To RP r and waypoints RP r To RP r+1 Two vectors.

[0090] At the same time, refer to Figure 4 By restricting the UAV to search only within the angle range [0, π / 3] in front of the current waypoint, the algorithm can search only a few candidate points in front of the current waypoint when selecting the next expansion node, thus improving search efficiency.

[0091] 3) Path Length Constraint: In discrete 3D space, the UAV's flight path (Path) can be represented by connecting all adjacent waypoint nodes. The total flight path length can be obtained by calculating and summing the Euclidean distances of each pair of adjacent waypoint nodes. For any flight segment RP... r-1 RP r Its path length is expressed in Euclidean distance:

[0092]

[0093] wherein: denotes the normalized risk cost of a leg, t r-1 and t r denote the three-dimensional coordinates of two nodes, d norm denotes the normalized distance, used to normalize the path distance data.

[0094] 4) The risk of ground constraint: based on the risk of ground matrix, the risk cost (i.e. the expected value of casualties) caused by the UAV flying on a unit path leg can be calculated by formula (10) and (11):

[0095]

[0096] wherein: denotes the normalized risk cost of a leg, t r denotes the time taken by the UAV to fly from the waypoint node RP r-1 to RP r , r norm is the normalized risk value of this path, by r norm the risk data of the path can be normalized. denotes the Euclidean distance moved by the UAV to fly from the waypoint node RP r-1 to RP r , v UAV denotes the cruising speed of the UAV.

[0097] On this basis, R ELOS denotes the equivalent safety level of the planned path, calculated as formula (12), the actual meaning is that when the UAV runs on this path, the expected casualties is R ELOS casualties per flight hour, and the expected casualties is not greater than the target safety level R TLOS (Target Level of Safety, TLOS) required for the operation of this UAV type, i.e. formula (13).

[0098]

[0099] R ELOS ≤ R TLOS (13)

[0100] The target safety level is a quantitative aviation safety index, used to measure whether the risk of the operation of the aviation system is within an acceptable range, usually expressed as the target value of the risk level, and in the present application, it is the safety index that the UAV needs to meet to be able to safely operate in the low-altitude traffic system.

[0101] 5) Available airspace constraints: The search space of the flight path planning algorithm should be the set of spatial points U that the UAV can fly in n,m,l In order to avoid the terrain buildings in the path planning algorithm search process, it is necessary to constrain Path∈U n,m,l .

[0102] Taking the path synthesis cost of the UAV flight path in the urban airspace as the objective function, the flight height, turning angle, ground risk, path length and available airspace of the UAV as the constraint conditions, the path planning problem model of the UAV in three-dimensional space can be obtained as:

[0103]

[0104] In the three-dimensional airspace, considering the above-mentioned various constraint conditions, including flight height, turning angle, path length, ground risk and other factors, by optimizing the path synthesis cost, including distance cost and risk cost, the safest and most efficient flight path is found.

[0105] Step S3, combine the risk reduction priority A* algorithm and dynamic risk weight coefficient to optimize the UAV flight path, and obtain the optimal UAV flight path;

[0106] The risk reduction priority A* algorithm is used to calculate the total value function value F'(r) in this embodiment, which specifically includes the following contents:

[0107] A* algorithm is a classic heuristic search algorithm, which is widely used in two-dimensional plane shortest path planning problems. By using heuristic function, A* algorithm can effectively guide the search direction, especially in discrete space path planning problems. The total cost function f(r) of the standard A* algorithm is as follows:

[0108] f(r)=g(r)+h(r) (15)

[0109] In the formula: g(r) represents the actual cost from the starting point to the node r, and h(r) is the heuristic function of the estimated cost from the node r to the target point. In two-dimensional path planning problems, h(r) usually uses Manhattan distance or Euclidean distance.

[0110] When introducing A* algorithm in the three-dimensional UAV path planning problem, the risk cost can be allocated to different components in the path synthesis cost, i.e. g(r) and h(r). This requires improvement of the actual cost function and heuristic function of A* algorithm in order to consider distance and risk cost at the same time.

[0111] In a three-dimensional space containing ground-related risk factors, it is necessary to extend the actual cost function g(r) to incorporate risk information into the calculation. A composite actual cost function G(r) is designed to combine distance cost and risk cost, as follows:

[0112]

[0113] In the formula: G(r-1) represents the parent node RP from the starting point to the current node. r-1 The actual cumulative cost. α1 and α2 are the weight coefficients of the normalized distance cost and the normalized risk cost, respectively, and satisfy α1+α2=1. By adjusting the weight coefficients, the bias of the algorithm's actual cost in terms of minimizing distance cost and minimizing risk cost can be adjusted.

[0114] When calculating the actual risk cost, the risk cost and path cost are not on the same order of magnitude. If the weighting coefficients are on the same order of magnitude, direct addition will cause the risk cost to be ignored. Therefore, it is necessary to normalize the distance cost and the risk cost separately. Normalized distance d norm The value is generally the step size of the algorithm's exploration; the normalized risk value r norm The value is set according to the safety requirements of the urban transportation system in the region.

[0115] At the same time, we can obtain the algorithm's search from the starting node START to the current node RP. r Cumulative normalized risk cost

[0116]

[0117] The heuristic function h(r) is used to estimate the expected cost from the current node to the target node. The Euclidean distance heuristic function used in the standard A* algorithm only reflects the distance cost, so the calculation method of h(r) needs to be improved to incorporate the risk estimate on the target path into the expected cost estimation.

[0118] The expected cost of the path from the current node to the target node is divided into two parts: the expected distance cost H. dist (r) and expected risk cost H risk (r). Construct the expected distance cost function H dist (r), using Euclidean distance to represent the current node RP r The straight-line distance to the target node GOAL is given in formula (18).

[0119] To estimate the risk cost on the target path, construct the current node's RP. r The normalized expected risk cost function H to the target node GOAL risk(r), which can be calculated by multiplying the normalized expected risk value and the expected flight time, as shown in equation (19). The normalized expected risk value can be calculated according to the risk distribution of the region where the target path is located, and the expected flight time can be calculated according to the expected distance, as shown in equation (20).

[0120]

[0121] wherein RP r and GOAL represent the three-dimensional coordinates of the current node and the target node, respectively, and R esti (r) represents the expected risk value from the current node RP r to the target node GOAL, and t esti is the expected flight time.

[0122] Because the risk is unevenly distributed in different airspace units and difficult to predict, it is difficult to obtain a suitable heuristic function value. If the heuristic function value is too large, the lower limit of the risk cost may be increased, thereby reducing the performance of the algorithm. Conversely, if the heuristic function value is too small, the search space within the lower limit range becomes larger, and the calculation efficiency of the algorithm is significantly reduced. Finding a suitable heuristic function value can achieve a good balance between path quality and calculation efficiency, and therefore, selecting a suitable method to calculate the expected risk value is an important step in improving the heuristic function.

[0123] A cube range with a line segment from the current node RP r to the target node GOAL as a diagonal is selected, and the six faces of the cube are parallel or perpendicular to the coordinate planes of the grid space, as shown in Figure 5 The search space range of the expected risk cost is represented by the cube space, denoted as a set P(x, y, z), and P(x, y, z) ∈ U n,m,l , and the following equation is obtained:

[0124]

[0125] wherein (i r ,j r ,h r ) and (i g ,j g ,h g ) represent the three-dimensional coordinates of the node RP r and the node GOAL, respectively, and x, y, and z are positive integers.

[0126] The risk value distribution of all space points in the search space range P(x, y, z) is counted, and the expected risk level R esti (r) of the heuristic function is calculated.

[0127]

[0128] wherein: represents the average risk value of all nodes in the search space range, represents the median of the risk distribution of all nodes in the range, n P represents the number of all nodes in the search space range, and R(x, y, z) represents the risk value set in the search space range.

[0129] In the risk data set in the search space range, the average value and the median both can represent the central tendency of the distribution of the risk data, and the smaller value is taken because the purpose of the algorithm is to make the risk distribution of the path fall in a smaller interval.

[0130] When , it means that there is an extremely low risk value, making the risk data distribution tend to a lower trend.

[0131] When , it means that the high-risk extreme value pulls the risk distribution trend to a higher position, while most of the data is concentrated in a lower risk interval, so that the search of the algorithm can be inclined to a lower risk.

[0132] According to formula (13), the equivalent safety level of the path risk also needs to meet the target safety level requirement of the urban air transportation system, so that the target safety level is taken as the boundary condition of the average risk of the path, and the risk cost of the expected path in the heuristic function can be limited, that is:

[0133]

[0134] wherein: Cost TLOS represents the normalized target risk cost corresponding to the target safety level, which can be calculated by formula (25), and t curr represents the cumulative flight time from the starting point to the current node, which can be calculated using formula (26).

[0135]

[0136] In order to make the algorithm search the optimal path that meets the target safety level, the total risk cost F risk (r) of the current node in the heuristic function is judged:

[0137] 1) When F irsk (r)≤Cost TLOS , it is considered that the path obtained by continuing to search at this node meets the target safety level, and the algorithm continues to work.

[0138] 2) when F risk (r) > Cost TLOS , the path which continues to search based on this node is considered to have an equivalent safety level of risk that does not satisfy the TLOS, and this node is added to the closed list.

[0139] This approach can make the heuristic function value fall within a more reasonable interval, guiding the algorithm to find a better quality path in the search space range. The improved heuristic function is constructed by combining the expected distance cost and the expected risk cost as follows:

[0140] H(r) = a1H dist (r) + a2H risk (r) (27)

[0141] In the study of risk constraints of UAV path planning, the equivalent safety level R ELOS of the path Path is the most important indicator of path safety, and it is necessary to find a path with the optimal comprehensive cost as much as possible under the condition of satisfying the TLOS. According to formulas (10) and (12), the expression of the equivalent safety level R ELOS of the path can be derived as follows:

[0142]

[0143] According to the expression (14) of the optimization model, the relationship between the total risk expectation and the normalized risk value can be obtained, as shown in formula (29). In the case where the algorithm can find an effective path, the search step d norm of the algorithm and the cruise speed v UAV of the UAV are determined values, so the equivalent safety level R ELOS of the path will be affected by the normalized risk value r norm and the ratio of the path risk cost to the distance cost, and it satisfies the target safety level:

[0144]

[0145] Formula (29) discusses the influence of the normalized risk value on the total risk expectation value, while the ratio of the path risk cost to the distance cost is another factor affecting the equivalent safety level of the path. For each leg RP r-1 RP r , the path comprehensive cost can be obtained according to expression (14) as follows:

[0146]

[0147] For this leg, the normalized distance cost is a determined value, which is only related to the search space of the algorithm, that is, in the search range of the next node of the current node, only Three values. So the path synthesis cost Cost r is determined by the normalized risk cost and the risk weight coefficient α2. The value of the normalized risk cost is related to the actual risk matrix, according to formula (40), here we need to classify the cases of and , from formula (9-11) we can get the size relationship of and , essentially it is the comparison of r norm and R(i,j,h), which can be expressed as:

[0148] 1) When , that is, r norm < d norm R(i,j,h) / v UAV , the value of the path synthesis cost Cost r decreases with the decrease of the risk weight coefficient α2.

[0149] 2) When , that is, At this time, the path synthesis cost Cost is independent of the value of the risk weight coefficient α2.

[0150] 3) When , that is, r norm > d norm R(i,j,h) / v UAV , the value of the path synthesis cost Cost r decreases with the increase of the risk weight coefficient α2.

[0151] From the above analysis, it can be seen that the equivalent safety level R ELOS of the path is a variable about the risk weight coefficient α2 and the normalized risk value r norm . The risk weight coefficient α2 determines the risk avoidance ability of the algorithm, the higher α2 is, the stronger the risk avoidance ability of the algorithm is. The normalized risk value r norm to some extent represents the average risk level of the path finally searched by the algorithm, the lower r norm is, the lower the overall risk level of the path finally selected by the algorithm is.

[0152] Based on the above derivation, this invention employs a risk reduction priority strategy using a dynamic risk weight coefficient A(r). Based on the relationship between segment risk cost and distance cost, the value of the risk weight coefficient is dynamically adjusted, making the algorithm more inclined to search in the direction of path risk reduction. The dynamic risk weight coefficient A(r) is used to adjust the algorithm's local optimization capability. When the overall risk level of the region is high, a larger A(r) value is used to improve the algorithm's risk avoidance capability; when the overall risk level of the region is low, a smaller A(r) value is used to improve the algorithm's optimization capability in terms of distance cost.

[0153] Establish a dynamic risk weight function A(r):

[0154]

[0155] In the formula: A G (r) represents the risk weight coefficient of the actual cost function G′(r), α low and α high The value range of the dynamic weight is represented by r, where λ represents the variation of the dynamic weight with respect to r. norm The correction factor. Typically, α... low The value of α is 0.1. high The value of λ is 0.9, and the value of 1 is 1.

[0156] This strategy can be applied in the interval [α] low ,α high The risk weight coefficient is dynamically adjusted within the range of α, while within the risk weight interval [α]. low ,α high In addition to [other values], the normalized risk value r is used. norm This allows for the overall control of the risk level of the flight path. By adjusting the normalized risk value, the normalized risk cost of flight segments can be controlled. The range of variation ensures that the equivalent security level of the path is within the acceptable target security level TLOS range.

[0157] Similarly, for heuristic functions, we have:

[0158]

[0159] In the formula: A H (r) represents the risk weight coefficient of the heuristic function H′(r).

[0160] Combining the practical cost function and the heuristic function, the total cost function F′(r) of the improved risk decline-first A*(RDP-A*) algorithm is expressed as follows:

[0161] F′(r)=G′(r)+H′(r) (34)

[0162] The actual cost function G′(r) and the heuristic function H′(r) are as follows:

[0163]

[0164] The specific steps of optimizing the UAV flight path using the risk descent-first A* algorithm and dynamic risk weighting coefficients in this embodiment are as follows:

[0165] Step S31: Initialize the UAV dataset, including UAV performance parameters, terrain function, static ground risk function, flight path constraints, available search space, etc.

[0166] Step S32: Create an Open List and a Closed List. Add the starting point START of the drone's safe path to the Closed List to store the nodes that have been explored; and use the Open List to store the nodes to be explored.

[0167] Step S33: Add the 26 nodes adjacent to the starting point START to the Open List. Calculate the total value function value F′(r) of each of the 26 nodes using the risk decline priority A* algorithm. Select the node with the smallest total value function value F′(r) for expansion, denoted as RP. r (r=1), and RP r Add to the closed list and determine the search direction.

[0168] The calculation of the total value function value F′(r) in this step is detailed in formula (34).

[0169] Step S34: Based on the search direction, node expansion is performed using a method that dynamically adjusts the risk weight coefficient. This specifically includes the following sub-steps:

[0170] Step S341: Set the current node RP r exist (When r = 1, RP) r-1 This means adding several potential nodes (nodes in the non-closed list Close Llist) in the direction of the START node to the open list;

[0171] Step S342: Calculate the normalized distance cost and normalized risk cost of the node, then calculate the dynamic risk weight coefficient of the actual cost function, and determine the dynamic risk weight coefficient A. G The value of (r) is given in formula (32);

[0172] Step S343: If there is a node RP among the potential nodes... q(q = 1,..., r) in the open list, then it indicates that the node RP q has been expanded before, and then the optimal parent node of the node RP q is determined; if the actual cost function G'(r) of the current node RP r to the node RP q is lower than the actual cost function G'(r) of the original parent node to the node RP q , then the cost information and the parent node index are updated, and it is ensured that the path found is always optimal;

[0173] Step S344, the normalized expected distance cost and the normalized expected risk cost of the current node are calculated, and then the dynamic risk weight coefficient A(r) of the potential node corresponding to the heuristic function is calculated H (r), and then the value of the dynamic risk weight coefficient A(r) is determined, and the specific value is shown in formula (33); H

[0174] Step S345, it is determined whether the expected path equivalent safety level R ELOS of the potential node meets the target safety level R TLOS , if yes, step S346 is executed; if no, the corresponding potential node is added to the closed list;

[0175] Step S346, the total cost function value F'(r) of all potential nodes in the open list Open List is calculated, the node with the minimum total cost function value F'(r) is selected and expanded, and is recorded as RP r+1 , and the RP r+1 is added to the closed list Close List;

[0176] Step S347, the node count parameter r = r + 1 is updated

[0177] Step S35, the step S34 is executed in a loop; until when the target point is selected as the current node, the optimal path is output, and the step S34 is stopped; or when the open list is empty, there is no effective optimal path from the starting point to the target point, and the step S34 is stopped.

[0178] Reference Figure 7 and Figure 8 ​By comparing the figure, the application compares the path planning ability of the RRT* algorithm based on graph structure search, the standard A* algorithm, the improved risk constraint A* algorithm (Risk Constraint A*, hereinafter referred to as RC-A*) and the proposed risk descent priority A* algorithm (Risk Descent Priority A*, hereinafter referred to as RDP-A*) in the same simulation scene. The algorithm RDP-A* proposed in the application effectively balances the path length and risk cost, explores a flight path with the optimal comprehensive cost, and has significant progress compared with the traditional algorithm.

[0179] Although the specific embodiments of the application are described in detail with reference to the accompanying drawings, it should not be understood as limiting the protection scope of the patent. Various modifications and variations made by those skilled in the art within the scope described in the claims are still within the protection scope of the patent.

Claims

1. A method for calculating a safe path of a UAV based on a comprehensive risk balancing strategy, characterized in that, The method comprises the following steps: S1, calculating a three-dimensional ground safety risk level of the unmanned aerial vehicle; S2, based on the three-dimensional ground safety risk level of the unmanned aerial vehicle, and in combination with a three-dimensional airspace constraint condition, optimizing a comprehensive cost of a flight path of the unmanned aerial vehicle to obtain a flight path of the unmanned aerial vehicle; S3, in combination with a risk reduction priority A* algorithm and a dynamic risk weight coefficient, optimizing the flight path of the unmanned aerial vehicle to obtain an optimal flight path of the unmanned aerial vehicle; In the S1, the three-dimensional ground safety risk level of the unmanned aerial vehicle is calculated, comprising: In the formula, represents the three-dimensional risk matrix of the UAV to the ground; represents the three-dimensional risk matrix of the UAV to the ground; Equivalent safety level of the position to the ground: wherein, represents the probability of the drone failing and crashing while flying in the air, represents the probability of the drone crashing on the ground at a location the final risk-to-ground value within the grid; In the S2, the comprehensive cost of the flight path of the unmanned aerial vehicle is optimized, comprising: In the formula, The total path cost represents the total cost of the drone's flight path. Indicates the path of the drone. For nodes; For drones Comprehensive path cost on the flight segment; This represents the normalized distance cost; This indicates the risk and cost of normalization; , These represent the lower and upper limits of the drone's flight altitude, respectively. Indicates the drone's flight altitude; This indicates the angle of change in the flight direction of the drone in three-dimensional space; Indicates from route node arrive The vector, Indicates that the drone departs from the waypoint Fly to The Euclidean distance moved; Represents the normalized distance; This represents the normalized risk value of the path; Indicates the drone's cruising speed; Indicates the equivalent safety level of the planned path; Indicates that the drone departs from the waypoint Fly to The time taken; Indicates the target safety level; This refers to the set of spaces used by drones for flight; The S3 specifically comprises the following sub-steps: S31, initializing a data set of the unmanned aerial vehicle; S32, creating an open list and a closed list, and adding a starting point of a safe path of the unmanned aerial vehicle into the closed list; S33, adding a node adjacent to the starting point into the open list, selecting a node corresponding to a minimum total value function value to expand, adding the node into the closed list, and determining a search direction; S34, according to the search direction, using a method of dynamically adjusting a risk weight coefficient to expand the node; S35, repeatedly executing the S34; until when a target point is selected as a current node, outputting an optimal path, and stopping the execution of the S34; or when the open list is empty, there is no effective optimal path from the starting point to the target point, and the execution of the S34 is stopped.

2. The method of claim 1, wherein, In the S33, the total value function is: wherein In the formula, Represents the total value function; Represents the actual cost function; Represents a heuristic function; This represents the parent node from the starting point to the current node. The actual cost function; Represents the actual cost function Dynamic risk weighting coefficient; Representing a heuristic function Dynamic risk weighting coefficient; This represents the normalized expected distance cost function; This represents the normalized expected risk cost function. 3.The method of claim 2, wherein, the normalized expected distance cost function and the normalized expected risk cost function respectively. wherein wherein, and represent the three-dimensional coordinates of the current node and the target node, respectively; represents the estimated risk value from the current node to the target node , is the estimated flight time; represents the average risk value of all nodes within the search space range; represents the median of the risk distribution of all nodes within the range; represents the number of all nodes within the search space range; represents the risk value set within the search space range; and represent the three-dimensional coordinates of the node and the node , respectively.

4. The method of claim 2, wherein, The S34 specifically comprises the following steps: S341, Set the current node exist Several potential nodes in the direction are added to the open list; S342, calculating a dynamic risk weight coefficient of a potential node corresponding to an actual cost function; S343, If there are nodes in the potential nodes In the open list, the node is determined. The optimal parent node; if the current node Reaching the node Actual cost function Nodes reached from the original parent node Actual cost function If the cost is lower, then update its cost information and parent node index; S344, calculating a dynamic risk weight coefficient of the potential node corresponding to a heuristic function; S345, judging whether an expected path equivalent safety level corresponding to the potential node meets a target safety level, if yes, executing S346; if not, adding the potential node corresponding to the potential node into the closed list; S346、calculate the total cost function value of all potential nodes in the open list, select the node with the minimum total cost function value to expand, denoted as , and add to the closed list; S347、updating the node count parameter .

5. The method of claim 4, wherein, In the S342, the dynamic risk weight coefficient corresponding to the actual cost function of the potential node is calculated is: In the formula, and denote the lower limit and the upper limit of the value range interval of denote the correction coefficient of the dynamic weight with the change of the time.

6. The method of claim 5, wherein, In the S344, the potential node corresponding to the dynamic risk weight coefficient of the heuristic function is calculated is: 。 7. The method of claim 4, wherein the method further comprises: In the S345, the expected path equivalent safety level corresponding to the potential node meets the target safety level, comprising: 。

Images