Single-track rotation reference adaptive three-dimensional path planning method

Abstract

The application provides a single-track rotation reference adaptive three-dimensional route planning method, and relates to the technical field of unmanned aerial vehicle autonomous task planning.The application takes a single current route as a planning object, generates a candidate route through bidirectional disturbance of each cooperative adjustment direction along an orthogonal adjustment reference, selectively receives an improved route according to an engineering task cost evaluation result, and adaptively adjusts a search step length of each direction; after each planning period ends, a net waypoint-height cooperative deformation formed by the received route is extracted, an orthogonalization treatment is performed to update an adjustment reference of the next period; when a route channel lacks substantial improvement, route reconstruction is triggered; after the planning is terminated, an optimal route is independently and safely reviewed and converted into a flight control task instruction.The application can reduce the number of invalid evaluations, improve the route planning efficiency under a limited evaluation budget, and is suitable for unmanned aerial vehicle intelligent inspection tasks in power line, oil and gas pipeline and factory site scenes.

Description

Technical Field

[0001] This invention relates to the field of autonomous mission planning technology for unmanned aerial vehicles (UAVs), specifically to an adaptive three-dimensional route planning method based on a single-track rotation reference. Background Technology

[0002] Intelligent drone inspection typically requires drones to depart from a designated starting point, complete tasks such as photography, temperature measurement, laser measurement, or other payload operations along the corridor containing the inspected object, and then reach a designated endpoint. The inspection area may include mountains, buildings, no-fly zones, communication blind spots, strong electromagnetic fields, or other areas that need to be avoided. Drones are also subject to flight condition limitations such as maximum range, permissible turning angles, flight altitude corridors, geofencing, and remaining battery power. Therefore, the route planning device needs to generate a three-dimensional route that simultaneously meets inspection coverage requirements and flight safety conditions within the permissible planning time before takeoff or the replanning time window during flight.

[0003] Three-dimensional inspection routes are typically determined by multiple intermediate waypoints and the corresponding flight altitudes of flight segments. Changes in the position of any intermediate waypoint simultaneously alter the length, turning angle, distance to the threat area, and corresponding energy consumption of adjacent flight segments. Changes in the altitude of any segment also affect terrain clearance, payload observation distance, and climb or descent. Evaluating a candidate route typically requires map lookup, segment sampling, threat exposure calculation, energy consumption calculation, and checks of range, turning angle, and altitude constraints. When using high-precision terrain data or flight dynamics simulation, the time required for a single evaluation further increases, thus limiting the number of candidate route evaluations that can be completed within the planned time window.

[0004] In existing route planning schemes, one type of scheme maintains multiple candidate routes simultaneously and updates the search direction based on the differences between candidate routes, group centers, or guide routes. This type of scheme typically requires performing map queries and flight constraint evaluations on multiple candidate routes separately within an update cycle. When the number of evaluations and computation time are limited, the evaluation budget available for fine-tuning a single route is correspondingly reduced. Another type of scheme adjusts a single route according to coordinate direction or a preset fixed direction. Its search direction does not change with the route improvement results already obtained. When flyable channels are tilted relative to the coordinate direction, or when multiple adjacent waypoints and segment altitudes need to be coordinated, duplicate and invalid evaluations in multiple directions are prone to occur, and it is difficult to utilize existing improvement processes to form subsequent coordinated adjustment directions.

[0005] Some existing route planning schemes use the waypoint set obtained during the planning phase or candidate routes with lower mission costs as the calculation results. However, the discrete sampling density, constraint penalties, and coordinate boundaries used in the route planning phase are not entirely the same as the terrain clearance, range, adjacent segment turning angles, geofencing, and waypoint format checks required by the flight controller before executing the mission. Even if a candidate route meets the variable boundaries of waypoint coordinates and altitude, there may still be situations where the route crosses a no-fly zone within a segment, exceeds the total range limit, or the turning angle of adjacent segments does not meet flight performance requirements. In addition, the mission data required by the flight controller also includes the waypoint sequence after coordinate system transformation, flight altitude, cruise speed, and inspection payload actions. If the planning result output, independent safety review, and flight control mission command generation are separated, they need to be checked and transformed again by other modules or manually, which is not conducive to forming inspection tasks that can be directly issued within the limited replanning time.

[0006] Therefore, existing UAV intelligent inspection route planning has at least the following interrelated technical problems: reducing the budget occupied by parallel evaluation of multiple routes when the number of candidate route evaluations is limited; forming a coordinated adjustment direction of multiple waypoints and segment altitudes based on the actual improvement results already achieved for a route, reducing redundant and invalid evaluations in fixed directions; timely transfer to other flyable channels when local routes continue to lack substantial improvement; and ensuring that the planning results are reviewed and converted into flight control mission commands according to unified flight safety conditions.

[0007] In view of the above, this application is hereby submitted. Summary of the Invention

[0008] This invention provides an adaptive three-dimensional route planning method based on a single-track rotation reference, which can at least partially improve the above-mentioned problems.

[0009] To achieve the above objectives, the present invention adopts the following technical solution: A single-track rotation datum adaptive three-dimensional route planning method, comprising: Step S1: Obtain the input dataset and perform initialization processing based on the input dataset; Step S2: After initialization, within the current planning cycle, sequentially perform bidirectional perturbation on the current route along each multi-waypoint-altitude coordinated adjustment direction to generate candidate routes, and calculate the engineering task cost for each candidate route; exclude candidate routes that fall into the terrain safety envelope or hard no-fly zone; based on the comparison of the engineering task cost between the candidate routes and the current route, select one of the candidate routes with reduced cost as the new current route, and adjust the search step size of the current direction and update the previous optimal routes according to whether the new current route is successfully received; after traversing all executable directions in this cycle, complete the direction traversal of the current planning cycle; Step S3: Based on the end-of-cycle route and the beginning-of-cycle route, determine the net waypoint-altitude coordinated deformation within the current planning cycle; construct the reference matrix to be orthogonalized based on the net waypoint-altitude coordinated deformation and the current orthogonal route adjustment reference; selectively perform orthogonalization processing on the reference matrix to be orthogonalized to update the orthogonal route adjustment reference, or retain the current orthogonal route adjustment reference and accumulate the stall count, based on the numerical rank state of the reference matrix to be orthogonalized and the determination result of whether an effective net deformation has been formed in this cycle; enter the next planning cycle and update the historical window data of the previous optimal costs; Step S4: When it is determined that the current route channel meets the preset reconstruction trigger conditions, route reconstruction is initiated; the reconstruction search range is set according to the proportion of the consumed assessment budget, and a reconstructed route is generated within the reconstruction search range based on the previous best routes, and flight safety constraint checks and engineering task cost assessments are performed on the reconstructed routes; the reconstructed routes that pass the checks are selectively updated to the current route and the previous best routes; after reconstruction is completed, the adjustment step size and adjustment benchmark in each direction are reset, the stall count is cleared, and the optimal cost history window is refreshed with the engineering task cost of the current previous best routes; Step S5: When the planning termination condition is met, the optimal routes of each iteration are independently reviewed, and the optimal routes that pass the review are converted into flight control mission commands and issued.

[0010] Preferably, step S1 specifically includes: Obtain the input dataset, which includes the lower bound of the route variables, the upper bound of the route variables, a fixed starting point, and a fixed ending point; Based on the lower bound and upper bound of the route variables, the allowable route span vector, the maximum value of the allowable route span vector, the maximum adjustment scale, and the minimum adjustment scale are set sequentially. The initial route is generated by coarse planning based on a fixed starting point, fixed ending point, existing route or A-star heuristic graph. After feasible region projection and hard safety check of the initial route, the corresponding engineering task cost is calculated. The current route and the previous best routes are set as the initial route. The engineering task cost of the current route and the previous best routes are set as the engineering task cost of the initial route. The number of engineering evaluations used is set to 1. Initialize the D-dimensional adjustment scale vector, orthogonal route adjustment benchmark, planning cycle number as 0, consecutive invalid cycle count as 0, and optimal cost historical window.

[0011] Preferably, step S2 specifically includes: Step S201: When it is determined that the number of engineering evaluations used is less than the total number of joint evaluations, proceed to the next planning cycle, save the starting route of the cycle, and set the value of the receiving judgment logic flag to no; Step S202: Determine whether the number of engineering evaluations used is equal to the total number of joint evaluations; if yes, stop generating new candidate routes; if no, calculate the forward and backward candidate routes for the d-th multi-waypoint-altitude coordinated adjustment direction, and calculate the engineering task cost of the forward candidate route and the engineering task cost of the backward candidate route respectively. Each time the engineering task cost is calculated, the number of engineering evaluations used is incremented by 1.

[0012] Step S203: When it is determined that the forward or backward candidate route of the d-th multi-waypoint-altitude coordinated adjustment direction intersects with the terrain safety envelope or hard no-fly zone, record its engineering task cost as +∞ and reject it; Step S204: Select the engineering task cost of the forward candidate route and the backward candidate route that has the smaller value among the engineering task costs of the d-th multi-waypoint-altitude coordinated adjustment direction. When it is determined that the engineering task cost is less than the engineering task cost of the current route, enter the route receiving branch. Step S205: Select the candidate route with the smaller engineering task cost as the new current route. When the engineering task cost of the current candidate route is equal to that of the backward candidate route, the forward candidate route is always selected as the new current route. Step S206: Select the smaller value from the product of the scale expansion coefficient and the D-dimensional adjustment scale vector corresponding to the d-th multi-waypoint-altitude coordinated adjustment direction, and the maximum adjustment scale, and use it as the new D-dimensional adjustment scale vector corresponding to the d-th multi-waypoint-altitude coordinated adjustment direction, and set the value of the receiving judgment logic flag to yes. Step S207: When it is determined that the engineering task cost of the current route is less than the engineering task cost of the previous optimal routes, write all waypoints, segment altitudes and constraint states of the current route into the previous optimal routes, and use the engineering task cost of the current route as the engineering task cost of the new previous optimal routes. Step S208: When it is determined that the engineering task cost of the current route is not less than the engineering task cost of the previous optimal routes, keep the current route unchanged, and take the product of the contraction coefficient and the D-dimensional adjustment scale vector corresponding to the d-th multi-waypoint-altitude coordinated adjustment direction as the new D-dimensional adjustment scale vector corresponding to the d-th multi-waypoint-altitude coordinated adjustment direction. Step S209: When it is determined that there is still a remaining assessment budget, for the (d+1)th multi-waypoint-altitude coordinated adjustment direction, execute steps S201 to S208, and repeat the above steps until there is no remaining assessment budget or all multi-waypoint-altitude coordinated adjustment directions within this planning period have completed steps S201 to S208.

[0013] Preferably, step S3 specifically includes: The current route obtained after completing step S2 is taken as the end route of the cycle, and the net waypoint-altitude cooperative deformation is calculated based on the end route of the cycle and the start route of the cycle. Based on the net waypoint-altitude coordinated deformation and orthogonal route adjustment benchmark, construct the benchmark matrix to be orthogonalized for the current planning cycle; When it is determined that the numerical rank of the reference matrix to be orthogonalized in the current planning cycle is insufficient, the current orthogonal route adjustment reference is retained; When it is determined that the numerical rank of the reference matrix to be orthogonalized in the current planning cycle is sufficient, the value of the receiving judgment logic flag is yes, and the vector length of the net waypoint-altitude coordinated deformation is greater than the preset effective deformation threshold, QR orthogonal-upper triangular decomposition is performed on the reference matrix to be orthogonalized to obtain the updated D-dimensional orthogonal reference matrix and D-dimensional upper triangular matrix. The updated D-dimensional orthogonal reference matrix is ​​then used as the new orthogonal route adjustment reference, and the continuous invalid cycle count is set to 0.

[0014] When it is determined that no effective net deformation has been formed in the current planning cycle, the current orthogonal route adjustment benchmark is retained, and the count of consecutive invalid cycles is incremented by 1; Enter the next planning cycle and update the historical data of the best cost. When it is determined that the index number of the current planning cycle is greater than or equal to the progress window length, calculate the best relative improvement ratio of each cycle.

[0015] Preferably, step S4 specifically includes: When it is determined that the number of consecutive invalid cycles is greater than or equal to the threshold of consecutive invalid cycles, or the index of the current planning cycle is greater than or equal to the progress window length and the best relative improvement ratio of the current planning cycle is less than the relative improvement tolerance, or the maximum component value in the D-dimensional adjustment scale vector corresponding to each multi-waypoint-altitude coordinated adjustment direction is less than the preset minimum adjustment scale, it is determined that the current route channel is no longer worth allocating the assessment budget and route reconstruction is initiated. Based on the number of engineering assessments used and the total number of joint evaluations, set the budget consumption ratio, and set the local reconstruction radius vector based on the route allowable span vector; Based on the local reconstruction probability and the previous optimal routes, a reconstructed route is generated. The reconstructed route is then checked for terrain, no-fly zones, range, turning angle, and altitude, and the corresponding engineering task cost is calculated. When it is determined that the reconstructed route has passed the inspection, the reconstructed route will be used as the new current route. When it is determined that the engineering task cost of the reconstructed route is less than the engineering task cost of the previous best route, the reconstructed route will be used as the new previous best route, and the engineering task cost of the reconstructed route will be used as the engineering task cost of the new previous best route. When it is determined that the reconstructed route fails the inspection, the reconstructed route is regenerated within the remaining assessment budget. When the budget is insufficient, the best route from previous attempts is retained as the current route. For all multi-waypoint-altitude coordinated adjustment directions, the larger of the local reconstruction radius vector corresponding to the d-th multi-waypoint-altitude coordinated adjustment direction and the minimum adjustment scale is taken as the new D-dimensional adjustment scale vector corresponding to the d-th multi-waypoint-altitude coordinated adjustment direction. The orthogonal route adjustment benchmark is set to equal the D-dimensional identity matrix, the consecutive invalid cycle count is 0, and the optimal cost history window is refreshed with the engineering task cost of the current optimal route.

[0016] Preferably, step S5 specifically includes: Recheck whether the number of engineering assessments used is less than the total number of joint assessments. If so, proceed to the next planning cycle. If not, end the planning cycle and send the optimal routes of each time to the independent flight constraint review. The threat zone, hard no-fly zone, terrain clearance, total flight distance, adjacent segment turning angle and altitude corridor of the previous optimal flight route are re-verified with more dense segment sampling than in the optimization phase. If any constraint is not met, flight control execution commands shall not be generated. The verified optimal flight paths are converted into a sequence of mission commands in the flight control coordinate system, and waypoint sequence, altitude, speed and payload actions are sent to the flight controller. The flight controller controls the UAV to perform intelligent inspection based on positioning and altitude feedback.

[0017] In summary, this invention addresses the following technical challenges: Given the time constraints on map queries, threat calculations, and flight constraint evaluations in UAV intelligent inspection, how can a search direction be obtained that simultaneously adjusts multiple intermediate waypoints and segmented altitudes, based solely on a currently flyable route and its received improvement processes? Furthermore, how can the route be re-established when substantial improvements to the corridor lack sufficient space, thereby achieving a 3D route that meets inspection requirements within a limited evaluation budget? This invention processes actual waypoint-altitude data, including terrain, threat zones, no-fly zones, range, turning angles, and altitude limitations, to generate inspection commands that have been verified against engineering constraints and can be executed by the flight controller.

[0018] Specifically, taking a single current route as the planning state, the orthogonal adjustment benchmark is updated using the net waypoint-altitude collaborative deformation of the received routes relative to the starting route of the cycle within a route adjustment cycle. The decision to rebuild the current route is based on the window improvement amount of the optimal mission cost in each iteration, the number of consecutive ineffective adjustment cycles, and the adjustment scale. After planning termination, independent flight constraint verification is performed using segment sampling at a higher density than in the planning evaluation phase. Routes that pass the verification are converted into flight control mission commands containing waypoint sequence, flight altitude, cruise speed, and inspection payload actions. This ensures seamless integration of route adjustment, stall handling, safety verification, and inspection execution within a limited evaluation budget.

[0019] Compared with existing technologies, this method has the following advantages: This invention maintains only a single current route, eliminating the need to maintain a population of candidate routes, thus avoiding the problem of dispersed evaluation budgets caused by parallel evaluation of multiple routes. This allows a limited number of evaluation attempts to be concentrated on the effective adjustment of the current route. By extracting the net waypoint-altitude co-deformation within the route adjustment cycle and updating the orthogonal adjustment benchmark, subsequent search directions can be adjusted collaboratively along the combined directions that have successfully reduced mission costs, reducing redundant and invalid evaluations in fixed coordinate directions. The search range is reconstructed through adaptive adjustment of the budget consumption ratio, combined with global random branching, balancing early-stage channel exploration with later-stage fine-tuning. After planning termination, flight safety constraints are independently reviewed with denser sampling, ensuring consistency between the planning results and actual flight execution conditions. This invention can significantly reduce the overall mission cost and improve route planning efficiency and engineering reliability under a limited route evaluation budget. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating the single-track rotation reference adaptive three-dimensional route planning method provided in this embodiment of the invention.

[0021] Figure 2 This is a two-dimensional route projection diagram of different planning methods under reference conditions provided in the embodiments of the present invention.

[0022] Figure 3 This is a three-dimensional route and altitude trajectory diagram of different planning methods under reference working conditions provided in the embodiments of the present invention.

[0023] Figure 4 This is a convergence curve of the optimal task cost under the reference working condition provided in the embodiments of the present invention.

[0024] Figure 5 This is a distribution diagram of the final task cost obtained from ten independent planning operations under the reference working conditions provided in this embodiment of the invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0026] refer to Figure 1As shown, the first embodiment of the present invention discloses a single-track rotation reference adaptive three-dimensional route planning method, which can be executed by a single-track rotation reference adaptive three-dimensional route planning device (hereinafter referred to as the planning device). This device consists of a task configuration interface, an environmental data interface, a flight parameter interface, a route planning computer, a flight controller, an unmanned aerial vehicle (UAV) platform, and positioning and altitude sensors. The task configuration interface receives the starting point, ending point, inspection object, allowed planning time, or maximum number of route assessments; the environmental data interface receives a two-dimensional or three-dimensional map, no-fly zone, threat source location, threat radius, and threat level; the flight parameter interface receives cruise speed, maximum range, maximum turning angle, allowed altitude corridor, preferred cruise altitude, and energy consumption coefficient. Specifically, the method is executed by one or more processors within the planning device to achieve the following: Step S1: Obtain the input dataset and perform initialization processing based on the input dataset; Specifically, step S1 further includes: obtaining the input dataset, which includes the lower bound of the route variable lb, the upper bound of the route variable ub, and the fixed starting point. Fixed endpoint ; In addition, it may include: the number of flight segments M ≥ 2; consisting of M The route variable (i.e., the current route) consists of one two-dimensional intermediate waypoint and M segment altitudes. ,in, This is the 0th two-dimensional intermediate waypoint. This is the first two-dimensional intermediate waypoint. For the (M-1)th two-dimensional intermediate waypoint, The flight altitude for the first segment. The flight altitude for the Mth segment. Let S be the set of real numbers, and let S represent the set of three-dimensional flyable routes that satisfy the boundary conditions of the variables. For transpose, D=2(M) 1) + M = 3M 2 (this is both the dimension of the variable space and the column number of the orthogonal reference matrix; both have the same value because the matrix is ​​generated based on the dimension); lower bound lb and upper bound ub for route variables; cost function for route engineering tasks. Maximum number of route engineering assessments (MaxFEs); Scale expansion coefficient Shrinkage coefficient Progress window length W, relative improvement tolerance ε rel Continuous invalid period threshold τ s Local reconstruction probability p hat and reconstruction radius coefficient c rThe upper and lower boundaries of plane waypoints are determined by the inspection area, geofence, and the outer boundary of the no-fly zone, while the upper and lower boundaries of altitude components are determined by the permissible altitude corridor. For any candidate route, component width limiting or boundary projection is first performed, and then the intersection of the flight segment with the no-fly zone or terrain safety envelope is checked based on map geometry.

[0027] Based on the lower and upper bounds of the route variables, the allowable route span vector is set sequentially. Maximum allowable span vector of air route Maximum adjustment scale Minimum adjustment scale , With an infinite norm, each component of the route's allowable span vector represents an adjustable range of plane waypoint coordinates or segment heights; Based on a fixed starting point, a fixed ending point (i.e., straight-line interpolation between the starting and ending points), existing routes, or A-star ( Heuristic graph search coarse programming generates an initial route z0 (i.e., a complete initial route is generated with fixed origin and destination as boundaries), and the feasible region P is projected onto the initial route. S After hard safety checks, calculate the corresponding engineering task cost, and compare the current route z with the previous optimal routes. The values ​​are all set to the initial route z0, and the engineering task cost F(z) = F(z0) of the current route is compared with the engineering task cost of the previous optimal routes. All are set as the engineering task cost of the initial route, and the number of engineering evaluations used is equal to 1, i.e., FEs=1; Initialize D-dimensional adjustment scale vector Orthogonal route adjustment benchmark The planning cycle number is 0, i.e., t=0; the continuous invalid cycle count is 0, i.e., c stag =0, optimal cost historical window, where... For the first multi-waypoint—highly coordinated course adjustment The corresponding D-dimensional scale adjustment vector Adjust direction for the Dth multiwaypoint - altitude coordination The corresponding D-dimensional scale adjustment vector Let D be the D-dimensional adjustment scale vector corresponding to the d-th multiwaypoint-altitude coordinated adjustment direction. It is a D-dimensional identity matrix.

[0028] In this embodiment, an input dataset is obtained through a task configuration interface. This input dataset includes at least a lower bound, an upper bound, a fixed start point, and a fixed end point for the route variables. The route variables consist of the coordinates of multiple intermediate plane waypoints and the flight altitudes of each segment. Their lower and upper bounds are used to limit the legal range of changes in the coordinates and altitudes of each waypoint during subsequent route adjustments. The fixed start point and fixed end point correspond to the takeoff point of the UAV performing the inspection mission and the end point after the mission is completed, respectively. After obtaining the above input data, the route planning computer calculates the allowable span vector and its maximum value based on the lower and upper bounds of the route variables. This maximum value reflects the maximum amplitude within the allowable adjustment range of all route variables and serves as a global benchmark for setting the upper and lower limits of the search step size. Based on this, the maximum adjustment scale is set to 20% of the maximum allowable span, and the minimum adjustment scale is set to 10 to the power of -8 times the maximum allowable span, used to constrain the upper and lower limits of the adjustment step size in each direction, ensuring that the search step size always matches the overall scale of the mission airspace.

[0029] An initial route is then generated. This initial route can be generated using linear interpolation based on a fixed origin and destination, or it can be obtained by calling existing route data or through coarse planning using the A-satellite heuristic graph search. A feasible region projection is performed on the generated initial route to ensure that all waypoint coordinates and segment altitudes fall within their respective allowable ranges. Simultaneously, a hard safety check is performed to ensure that the route does not enter no-fly zones and does not intersect with the terrain safety envelope. The engineering cost of the initial route that passes the check is calculated. This cost comprehensively reflects threat exposure, energy consumption, altitude preference, turn agility, and the satisfaction of various constraints. The current route and all previous optimal routes are set as this initial route, and their engineering cost is also set as the engineering cost of this initial route. The number of engineering evaluations used is counted as one.

[0030] Based on this, the control variables are further initialized. An adjustment scale vector with the same dimension as the route variables is initialized, with each component corresponding to the adjustment step size of each coordinated adjustment direction. Each component is constrained between the minimum and maximum adjustment scales. The orthogonal route adjustment datum is initialized as an identity matrix, with each adjustment direction aligned with the coordinate axes. The planning cycle number is set to zero, the consecutive invalid cycle count is set to zero, and an optimal cost history window is established to record changes in optimal costs over time. Through this initialization, the system establishes a complete planning state, preparing for subsequent cyclical adjustments.

[0031] Step S2: After initialization, within the current planning cycle (executed sequentially for direction numbers d=1,…,D), the current route is bidirectionally perturbed along each multi-waypoint-altitude coordinated adjustment direction to generate candidate routes, and the engineering task cost is calculated for each candidate route; candidate routes that fall into the terrain safety envelope or hard no-fly zone are excluded; based on the comparison of the engineering task cost between the candidate routes and the current route, one of the candidate routes with reduced cost is selected as the new current route, and the search step size of the current direction is adjusted accordingly and the optimal routes of each iteration are updated based on whether the new current route is successfully received; after traversing all executable directions in this cycle, the direction traversal of the current planning cycle is completed. Specifically, step S2 further includes: Step S201: When it is determined that the number of engineering assessments used (FEs) is less than the total number of joint assessments (MaxFEs) (MaxFEs represents the total number of joint assessments of maps, threats, energy consumption, and constraints that are allowed to be completed, rather than ordinary cycle counting), proceed to the next planning cycle and save the cycle's starting route (z). start =z, and set the value of the success logic flag to false (success is the logic flag indicating whether a route that satisfies the constraints and reduces the mission cost has been received in this cycle). Step S202: Determine whether the number of engineering evaluations used is equal to the total number of joint evaluations; if yes, stop generating new candidate routes; if no, calculate the forward candidate route z of the d-th multi-waypoint-altitude coordinated adjustment direction. d+ =P S (z+σ d ·q d ) and backward candidate routes z d- =P S (z σ d ·q d ), and calculate the engineering task cost F(z) of the forward candidate routes respectively. d+ The engineering task cost F(z) of the backward candidate route d- An F-evaluation includes: the threat cost of the seven-point segment (C). m th (th indicates threat), segment energy consumption cost C m eng (eng represents energy consumption), high preference cost C m alt (alt indicates altitude), turning cost C m turn (turn indicates turning) and constraint penalty P con(con denotes a constraint); wherein, each time the calculation of the engineering task cost is completed, the used engineering evaluation count is incremented by 1, PS denotes the feasible region projection operator, y denotes the route variable before projection, P S (y)=min{max{y,lb},ub} is executed component-wise, min denotes taking the smaller value, max denotes taking the larger value, to ensure that the waypoint coordinates and altitude do not exceed the allowed upper and lower bounds, q d is the d-th multi-waypoint-altitude cooperative adjustment direction.

[0032] Step S203: When it is determined that the forward candidate route or the backward candidate route of the d-th multi-waypoint-altitude cooperative adjustment direction intersects with the terrain safety envelope or the hard no-fly zone, the engineering task cost thereof is recorded as +∞ and rejected; Step S204: Screen out the engineering task cost with the smaller value among the engineering task cost of the forward candidate route and the engineering task cost of the backward candidate route of the d-th multi-waypoint-altitude cooperative adjustment direction, and when it is determined that this engineering task cost is less than the engineering task cost of the current route (min{F(z d+ ),F(z d- )}<F(z)), enter the route acceptance branch; Step S205: Take the candidate route with the smaller engineering task cost as the new current route, wherein when the engineering task cost of the forward candidate route is equal to the engineering task cost of the backward candidate route, the forward candidate route is fixedly selected as the new current route, so as to ensure that the same input obtains a determined result; Step S206: Select the smaller value from the product of the scale expansion coefficient and the D-dimensional adjustment scale vector corresponding to the d-th multi-waypoint-altitude cooperative adjustment direction, and the maximum adjustment scale, and take it as the new D-dimensional adjustment scale vector σ corresponding to the d-th multi-waypoint-altitude cooperative adjustment direction d ←min{ασ d ,σ max}, and set the value of the acceptance judgment logic identifier success to true (true indicates that at least one flyable route with reduced task cost has been accepted in this cycle); Step S207: When it is determined that the engineering task cost of the current route is less than the engineering task cost of the previous optimal route (F(z)<F ), write all waypoints, segment altitudes and constraint states of the current route into the previous optimal route z , and take the engineering task cost of the current route as the new engineering task cost F of the previous optimal route ←F(z); Step S208: When it is determined that the engineering task cost of the current route is not less than the engineering task cost of the previous optimal routes (i.e., neither of the two candidate routes reduces the current task cost), keep the current route unchanged, and use the product of the contraction coefficient and the D-dimensional adjustment scale vector corresponding to the d-th multi-waypoint-altitude coordinated adjustment direction as the new D-dimensional adjustment scale vector σ corresponding to the d-th multi-waypoint-altitude coordinated adjustment direction. d ←βσ d This is to refine the subsequent waypoint-altitude modification range for this collaborative direction; Step S209: When it is determined that there is still a remaining assessment budget, for the (d+1)th multi-waypoint-altitude coordinated adjustment direction, execute steps S201 to S208, and repeat the above steps until there is no remaining assessment budget or all multi-waypoint-altitude coordinated adjustment directions in this planning cycle have completed steps S201 to S208 (i.e. until the evaluation of executable directions in this cycle is completed).

[0033] It should be noted that, let δ m =r m r m-1 d m =||δ m ||2, where δ m Let d represent the planar displacement vector of the m-th flight segment. m This indicates the horizontal length of the flight segment. Let m-1 be the (m-1)th two-dimensional intermediate waypoint; m For adjacent flight segments δ m and δ m+1 The turning angle. L in the route constraint formula. max Indicates the maximum allowed total range, Θ max H represents the maximum allowed turning angle. min and H max These represent the lower and upper limits of the permissible flight altitude, respectively, and cos represents the cosine function. Let m be the flight altitude for the m-th segment. The route must at least satisfy the following range, turning angle, and altitude constraints: , , .

[0034] The route engineering assessment module sets up multiple internal sampling points along each flight segment to query the degree of proximity of that segment to threat sources or terrain hazards. For the nth threat source, n represents the threat source number, u represents the flight segment sampling point, and ρ... n (u) represents the distance from sampling point u to the threat source, ξ n Indicates the threat level of the threat source, R n Indicates the threat radius; α thβ represents the near-field threat enhancement coefficient. th This represents a multiple of the transition zone radius, where the subscript 'th' indicates the threat level. Formula The near-field enhancement inverse square distance term, the transition region inverse square distance term, and the zero value outside the transition region are adopted, and ρ is... n (u)=R n Explicitly classify it into the near-field branch to ensure the piecewise condition is complete; when ρ n When (u)=0, the sampling point is determined to overlap with the threat source center and the corresponding candidate route is rejected based on hard constraint conflict. The threat cost function value generated by the nth threat source at sampling point u is used to quantify the threat contribution of the threat source to that sampling point.

[0035] In this embodiment, seven internal sampling points are taken for each flight segment, and the q-th sampling point is r. m-1 +(2q / 15)(r m r m-1 ), q is the sampling point number within the flight segment and q=1,…,7; N represents the total number of threat sources, and the corresponding threat cost for the flight segment is: Furthermore, the formula , , , , This reflects the calculation process of the cost of the engineering task.

[0036] Among them, C m eng This represents the energy cost of the m-th segment, where the subscript eng indicates energy consumption, η indicates the energy consumption coefficient, and v indicates the cruising speed; C m alt This represents the altitude preference cost for the m-th flight segment, where the subscript alt indicates altitude, μ represents the altitude preference coefficient, and H represents the altitude preference cost. Indicates the preferred cruising altitude; C m turn This represents the cost of turning at point m, where the subscript "turn" indicates a turn. m κ represents the sharp turn gain coefficient, where λ1, λ2, λ3, and λ4 represent the non-negative weights of threat, energy consumption, altitude, and turn cost, respectively. J(z) represents the basic engineering task cost composed of threat, energy consumption, altitude, and turn cost; P con (z) represents the constraint penalty added in addition to the basic cost, used to quantify the degree of exceeding the range limit, exceeding the turning limit, exceeding the altitude limit, entering the no-fly zone, or intersecting with the terrain safety envelope; F(z)=J(z)+P con(z) represents the total engineering cost of the candidate route comparison in this invention. In actual deployment, no-fly zones and terrain collisions are directly rejected as hard constraints, while other recoverable constraint deviations can be addressed with continuous penalties during the pre-planning stage, and hard constraint verification is performed again during the output stage.

[0037] The route planning computer only maintains the current route z and the previous optimal routes z. Adjustment scales in each direction and orthogonal route adjustment benchmarks are used, but the candidate route population is not maintained. Forward and backward candidate routes are generated along the d-th direction: , , P S This involves projecting waypoint coordinates and altitudes onto the corresponding allowable range, and using map queries to repair or reject obviously unflyable flight segments. The planning computer calls the same route engineering evaluation module to calculate the task cost of two candidate routes respectively; if at least one candidate route reduces the task cost, the one with the lower cost is accepted, and the corresponding adjustment scale is multiplied by an expansion factor greater than 1 and limited to the maximum scale; if neither reduces the task cost, the corresponding adjustment scale is multiplied by a contraction factor between 0 and 1.

[0038] Specifically, in this embodiment, the main planning loop is entered, and the direction traversal within the current planning cycle begins. At the beginning of each planning cycle, it is first determined whether the number of engineering evaluations used is less than the total number of joint evaluations. If the total number of joint evaluations has been reached or exceeded, the generation of new candidate routes is stopped and the current cycle is exited; if there is still remaining evaluation budget, the current planning cycle is entered, the current route is saved as the starting route of the cycle, and the reception judgment logic flag is set to no to record whether an improved route has been successfully received within this cycle.

[0039] Subsequently, the system sequentially generates and evaluates candidate routes along each multi-waypoint-altitude coordinated adjustment direction in the orthogonal route adjustment benchmark. For the currently processed d-th direction, the system determines whether the number of engineering assessments used equals the total number of joint assessments. If so, it immediately stops generating new candidate routes; otherwise, it performs bidirectional probing in that direction. Specifically, the system applies offsets to the current route in both the forward and reverse directions of that direction, generating forward and backward candidate routes, and calculates the engineering task cost for each of these two candidate routes. Each time the engineering task cost calculation for a candidate route is completed, the number of engineering assessments used increases by one. Since this method processes a single route, each evaluation performs map lookup, threat calculation, and constraint checks on a complete candidate route. Therefore, the use of the evaluation budget directly corresponds to the number of candidate routes generated, which is beneficial for achieving optimal allocation of evaluation times under a limited budget.

[0040] After generating candidate routes, the system performs flight safety constraint checks on them. If a forward or backward candidate route intersects with a terrain safety envelope or a hard no-fly zone, the engineering cost of that candidate route is recorded as infinite and it is rejected, preventing it from being selected in subsequent comparisons. This hard constraint check mechanism ensures that all candidate routes entering the subsequent evaluation process possess basic flight safety, avoiding wasting evaluation budget on no-fly routes.

[0041] For candidate routes that pass the hard constraint check, the system filters and accepts them based on their engineering task costs. Specifically, it selects the route with the smaller engineering task cost from both forward and backward candidate routes. If this value is less than the engineering task cost of the current route, it enters the route acceptance branch. The candidate route with the smaller engineering task cost is then adopted as the new current route. If the engineering task costs of the two candidate routes are equal, the forward candidate route is always selected as the new current route. This deterministic acceptance rule ensures the reproducibility of the planning results under the same input conditions. If the engineering task cost of the new current route is better than that of any previous optimal route, all waypoints, segment altitudes, and constraint states of the current route are written into the previous optimal route list, and the engineering task cost of the current route is used as the engineering task cost of the new previous optimal route.

[0042] After accepting or rejecting a route, the system adjusts the search step size in the current direction based on whether a new route was successfully accepted. If an improved route is successfully accepted, the smaller value is selected from the product of the scale expansion coefficient and the adjustment scale vector corresponding to the current direction, and the maximum adjustment scale, as the new adjustment scale vector. This expands the step size to accelerate exploration efficiency in that direction, and the acceptance judgment logic flag is set to "yes". If the engineering task cost of the current route is not less than the engineering task cost of the previous best routes, the current route remains unchanged, and the product of the contraction coefficient and the adjustment scale vector corresponding to the current direction is used as the new adjustment scale vector. This shrinks the step size for more refined exploration in that direction. This adaptive step size adjustment mechanism maintains an active exploration posture in successful directions while gradually reducing the step size in ineffective directions to avoid wasting evaluation budget.

[0043] After processing the current direction, the system determines whether there is any remaining budget. If there is, for the next multi-waypoint-altitude coordinated adjustment direction, the entire process of candidate route generation, engineering task cost calculation, hard constraint checking, route acceptance judgment, and step size adjustment is repeated. This process is repeated until there is no remaining budget or all multi-waypoint-altitude coordinated adjustment directions within the current planning cycle have been processed, thus completing the direction traversal for the current planning cycle.

[0044] Step S3: Based on the end-of-cycle route and the beginning-of-cycle route, determine the net waypoint-altitude coordinated deformation within the current planning cycle; construct the reference matrix to be orthogonalized based on the net waypoint-altitude coordinated deformation and the current orthogonal route adjustment reference; selectively perform orthogonalization processing on the reference matrix to be orthogonalized to update the orthogonal route adjustment reference, or retain the current orthogonal route adjustment reference and accumulate the stall count, based on the numerical rank state of the reference matrix to be orthogonalized and the determination result of whether an effective net deformation has been formed in this cycle; enter the next planning cycle and update the historical window data of the previous optimal costs; Specifically, step S3 further includes: taking the current route obtained after completing step S2 as the end route of the cycle, i.e., z end =z, and based on the end-of-cycle route and the beginning-of-cycle route, calculate the net waypoint-altitude coordinated deformation, i.e., v=z end z start The components of v are the net changes in the x-coordinate, y-coordinate, and segmental altitude of intermediate waypoints caused by actual reception during this period. Based on the net waypoint-altitude coordinated deformation and orthogonal route adjustment benchmark, construct the benchmark matrix to be orthogonalized for the current planning cycle, namely B. t =[v|q1|…|q D 1], that is, the reference matrix to be orthogonalized for the planning period t, q D 1 represents the (D-1)th multiwaypoint—altitude-coordinated direction adjustment; When it is determined that the numerical rank of the reference matrix to be orthogonalized in the current planning cycle is insufficient, the current orthogonal route adjustment reference is retained to avoid uncertain references caused by degradation and deformation. When it is determined that the numerical rank of the reference matrix to be orthogonalized in the current planning period is sufficient (rank(B)... t =D), the value of the receiving judgment logic flag is yes (success=true), and the vector length of the net waypoint-altitude coordinated deformation is greater than the preset effective deformation threshold (||v||2>ε). v When performing QR orthogonal-upper triangular decomposition (B) on the reference matrix to be orthogonalized, t =Q′ t ·R t ), thus obtaining the updated D-dimensional orthogonal reference matrix Q′ t and D-dimensional upper triangular matrix R t (It is a non-negative diagonal element), and the updated D-dimensional orthogonal reference matrix is ​​used as the new orthogonal route adjustment reference, i.e., Q←Q′. t To ensure that the first direction of the next cycle is consistent with the most recent successful net coordinated deformation v, a continuous invalid cycle count c is set. stag ε is 0 v=10 -12 Δ max The threshold value is non-zero, and rank represents the matrix rank.

[0045] When it is determined that no effective net deformation has been formed in the current planning cycle (i.e., the three judgment conditions in the previous step have failed), the current orthogonal route adjustment benchmark is adjusted, and the count of consecutive invalid cycles is incremented by 1, i.e., c. stag ←c stag +1, where stag represents stagnation; Enter the next planning cycle (i.e., t←t+1) and update the historical data of the optimal cost. When it is determined that the index number of the current planning cycle is greater than or equal to the progress window length (t≥W), calculate the optimal relative improvement ratio (RIR) for each iteration. t =[F t W F t ] / max{1,|F t W |},ε rel In this context, rel represents a relative quantity, and F t Let F be the historical time point corresponding to the t-th planning period, representing the cost value of each optimal route engineering task that the algorithm has obtained and saved. t W The algorithm has obtained and saved the cost of each optimal route engineering task up to the historical time corresponding to the tWth planning period.

[0046] Among them, z is saved at the beginning of a complete route adjustment cycle. start After processing the predetermined directions and receiving several routes in sequence, save z. end From v=z end z start The net waypoint-altitude coordinated deformation of successful routes within this period is obtained. This vector simultaneously contains changes in the x and y coordinates of multiple waypoints and the altitude of multiple flight segments, thus recording the low-risk corridor directions that actually reduce threat, energy consumption, and turning costs.

[0047] When the net co-deformation v satisfies the condition of being non-zero and B t When the rank is full, follow the formula , , , Construct a D×D reference matrix B to be orthogonalized tAnd perform QR orthogonal-upper triangular decomposition. Q′ t R represents the orthogonal route adjustment benchmark after the update in period t. t Represents the corresponding upper triangular matrix; by convention R t The diagonal elements are non-negative, which determines the sign of the decomposition result. In the next period, let Q←Q′. t When no cost-reducing routes or B are received in this cycle t When the numerical rank is insufficient, the original orthogonal reference is retained and the invalid period count is increased.

[0048] To prevent a route from undergoing minor improvements that fail to improve the optimal results of inspection tasks over the long term, the planning computer calculates the relative improvement in the cost of each optimal task within a window of W planning periods: .

[0049] In this embodiment, the current route at the end of the cycle is obtained as the cycle-ending route. This cycle-ending route reflects the final state after the cumulative superposition of all successfully received route adjustments within this cycle. Combining the cycle-starting route—that is, the route state saved before step S2—the system calculates the difference between the two to obtain the net waypoint-altitude coordinated deformation within the current planning cycle. This net deformation is a vector, whose components represent the net changes in the coordinates of each waypoint and the segment altitude within this cycle. It comprehensively reflects the overall displacement direction and magnitude of all successfully received route adjustments within this cycle in the multivariable space, and is the core basis for subsequent adjustment benchmark updates.

[0050] After obtaining the net waypoint-altitude coordinated deformation, the system constructs a reference matrix to be orthogonalized for the current planning cycle based on this net deformation and the current orthogonal route adjustment reference. Specifically, the system places the net deformation vector at the front, followed by the remaining directional vectors from the current orthogonal reference except for the first column, thus forming a new matrix with the same dimension as the route variable space. This matrix construction method ensures that successful coordinated deformation information for the current cycle is incorporated into the reference update process, while retaining the directional information from the original reference except for the first column.

[0051] After constructing the reference matrix to be orthogonalized, the system determines the numerical rank of the matrix and, based on the result of whether a valid net deformation has been formed in the current cycle, selectively performs a reference update or a stall count. When it is determined that the numerical rank of the reference matrix to be orthogonalized is insufficient, i.e., there is linear dependence between the columns of the matrix that makes it impossible to construct a complete orthogonal basis, the system retains the current orthogonal route adjustment reference unchanged and does not perform any update operation.

[0052] When the system determines that the numerical rank of the reference matrix to be orthogonalized is sufficient, the value of the receiving judgment logic flag is "yes"—meaning that at least one improved route has been successfully received in the current cycle—and the vector length of the net waypoint-altitude coordinated deformation is greater than the preset effective deformation threshold, the system determines that an effective net deformation has been formed in the current cycle, meeting the conditions for updating the reference. At this point, orthogonalization processing is performed on the reference matrix to be orthogonalized, decomposing it into the product of an orthogonal matrix and an upper triangular matrix through numerical decomposition. The resulting orthogonal matrix is ​​the updated orthogonal route adjustment reference. Its first column is precisely aligned with the direction of the net waypoint-altitude coordinated deformation in the current cycle, while the remaining columns retain the spatial information of the original reference as much as possible while maintaining orthogonality to the first column. The system uses this updated orthogonal matrix as the new orthogonal route adjustment reference for direction traversal in the next planning cycle, while setting the consecutive invalid cycle count to zero to indicate that the current channel still has the potential for further optimization. The aforementioned baseline rotation mechanism ensures that the preferred adjustment direction for the next cycle aligns with the cooperative deformation direction that has been validated in the current cycle. This allows subsequent searches to simultaneously adjust multiple waypoints and segment altitudes along the combined directions that have successfully reduced mission costs. This helps reduce redundant and invalid evaluations in fixed coordinate directions and improves adaptability to inclined or curved flyable channels.

[0053] When it is determined that no effective net deformation has been formed in the current planning cycle—including situations such as insufficient rank of the reference matrix to be orthogonalized, a negative value of the receiving judgment logic flag, or the vector length of the net waypoint-altitude coordinated deformation not exceeding the preset effective deformation threshold—the system retains the current orthogonal route adjustment reference unchanged, does not perform any update operation on the reference, and increments the count of consecutive invalid cycles. This count is used to subsequently track the number of consecutive cycles in which the current route channel fails to produce effective improvement, providing a data basis for the reconstruction trigger judgment in step S4.

[0054] After completing the baseline update or stall count operation, the system enters the next planning cycle and updates the historical window data of previous optimal costs. Specifically, the system records the engineering task cost of the current optimal route in the historical window to maintain the trajectory of the optimal cost change over the most recent cycles. When it is determined that the index number of the current planning cycle is greater than or equal to the preset progress window length, the system calculates the relative improvement ratio of previous optimal costs, that is, the relative decrease of the current optimal cost relative to the optimal cost before the window length. This relative improvement ratio reflects the actual progress of route optimization in the recent period and is an important indicator for subsequently judging whether the current route still has optimization potential. If the relative improvement ratio of the optimal cost within the window period is too small, it indicates that the current route has become saturated, and the system will trigger route reconstruction in step S4 accordingly.

[0055] Step S4: When it is determined that the current route channel meets the preset reconstruction trigger conditions, route reconstruction is initiated; the reconstruction search range is set according to the proportion of the consumed assessment budget, and a reconstructed route is generated within the reconstruction search range based on the previous best routes, and flight safety constraint checks and engineering task cost assessments are performed on the reconstructed routes; the reconstructed routes that pass the checks are selectively updated to the current route and the previous best routes; after reconstruction is completed, the adjustment step size and adjustment benchmark in each direction are reset, the stall count is cleared, and the optimal cost history window is refreshed with the engineering task cost of the current previous best routes; Specifically, step S4 further includes: when it is determined that a continuous invalid cycle count c is obtained... stag Greater than or equal to the continuous invalid period threshold τ s Or, the index number t of the current planning period is greater than or equal to the progress window length W and the best relative improvement ratio (RIR) of the current planning period is... t Less than the relative improvement tolerance ε rel Or, the maximum component value ||σ| in the D-dimensional adjustment scale vector corresponding to the multi-waypoint-altitude coordinated adjustment direction. |∞ Less than the preset minimum adjustment scale σ min When it is determined that the current airway is no longer worth allocating an assessment budget for and airway reconstruction is initiated; Based on the number of engineering assessments used and the total number of joint evaluations, the budget consumption ratio p = FEs / MaxFEs is set, and the local reconstruction radius vector r(p) = c is set according to the route allowable span vector. r (1 p) 2 The components of Δ and r(p) have the same dimensions as the allowable range of the corresponding waypoint coordinates or altitudes; Based on the local reconstruction probability p hat Each optimal flight path is used to generate a reconstructed flight path z′. The reconstructed flight path undergoes terrain, no-fly zone, range, turning angle, and altitude checks, and its corresponding engineering task cost F(z′) is calculated. For each joint check and cost calculation, FEs←FEs+1 is set; where p is the probability... hat Let z′=P S [z +r(p)⊙ξ],ξ~N(0,I D () is a D-dimensional standard Gaussian perturbation; with probability 1 p hat Take z′=lb+Δ⊙u, u~Unif([0,1] D ) is a D-dimensional independent uniform random vector, and ⊙ represents multiplication by components; When the reconstructed route passes the (hard constraint) check, it is used as the new current route. When the engineering task cost of the reconstructed route is less than the engineering task cost of the previous optimal route, it is used as the new previous optimal route, and its engineering task cost is used as the engineering task cost of the new previous optimal route. That is, let the current route z←z′, and when F(z′) <F When again let z ←z′、F ←F(z′).

[0056] If the reconstructed route fails the check, a new route is generated within the remaining assessment budget. When the budget is insufficient, the best route from previous attempts is retained as the current route; that is, when the budget is insufficient, the best route from previous attempts is retained. As the current route, it is forbidden to write unflyable routes into the current route status; For all multi-waypoint-altitude coordinated adjustment directions, the larger of the local reconstruction radius vector corresponding to the d-th multi-waypoint-altitude coordinated adjustment direction and the minimum adjustment scale is taken as the new D-dimensional adjustment scale vector corresponding to the d-th multi-waypoint-altitude coordinated adjustment direction (i.e., for d=1,…,D, let σ...). d ←max{[r(p)] d, σ min}), and set the orthogonal route adjustment reference to equal the D-dimensional identity matrix (Q←I). D ), Continuous invalid cycle count c stag The cost F is 0, based on the engineering task cost of the current optimal route. Refresh the historical window for optimal cost.

[0057] Among them, when RIR t Triggering occurs when the relative improvement tolerance is below, the number of consecutive periods without effective adjustment reaches a threshold, or the maximum adjustment scale is below the minimum scale. , , , The route reconstruction is shown. p represents the proportion of the number of engineering evaluations used relative to the maximum number of engineering evaluations, and r(p) represents the local reconstruction radius vector that shrinks as the evaluation budget is consumed. hat (Right now ) represents the probability of selecting a local reconstruction branch; ξ represents a D-dimensional standard Gaussian random vector; u represents a D-dimensional independent uniform random vector; and ⊙ represents component multiplication. Local reconstruction revolves around the optimal path z from each iteration. One branch generates the route variable globally within the allowed range, and both branches re-execute projection and flight constraint verification.

[0058] The local reconstruction radius decreases quadratically with the assessment budget consumption, allowing for wider hazard corridors in the early stages of planning and fine-tuning around favorable flight paths in the later stages. The global reconstruction branch retains the opportunity to restart from other flyable corridors. After reconstruction, the orthogonal datum is reset to the unit datum, and the adjustment scales in each direction are reset according to the larger of the local reconstruction radius and the minimum scale.

[0059] In this embodiment, the reconstruction triggering conditions include three scenarios: First, the consecutive invalid cycle count reaches a preset consecutive invalid cycle threshold, indicating that the current channel has failed to produce effective net deformation for a long period. Second, the index number of the current planning cycle has reached the progress window length, and the calculated optimal relative improvement ratio is less than the preset relative improvement tolerance, indicating that the improvement of the optimal route in the most recent window period has stagnated. Third, the maximum component value in the adjustment scale vector corresponding to all multi-waypoint-altitude coordinated adjustment directions is less than the preset minimum adjustment scale, indicating that the search step size in all directions has shrunk to the point where it cannot produce effective displacement. When any of the above three scenarios is met, the system determines that the current route channel is no longer worth allocating evaluation budget to and then initiates the route reconstruction process. The above three judgment conditions comprehensively evaluate the channel status from three dimensions: short-term stagnation, long-term stagnation, and step size failure, avoiding the reliance solely on whether a single candidate route has improved as the basis for continuing local adjustments, making the reconstruction decision more robust and reliable.

[0060] After initiating route reconstruction, the system calculates the budget consumption ratio based on the number of engineering assessments and the total number of joint evaluations used so far, to determine the reconstruction search range for the current stage. Based on this, a local reconstruction radius vector is set according to the route's allowable span vector, with each component of this reconstruction radius vector having the same dimension as the allowable range of the corresponding waypoint coordinates or altitude. Since the budget consumption ratio monotonically increases with the planning progress, while the reconstruction radius decreases quadratically with the budget consumption ratio, the reconstruction range is larger in the early stages of planning, enabling the crossing of wider flyable channels for channel transitions; in the later stages of planning, the reconstruction range gradually shrinks, allowing for refined adjustments around existing excellent routes, balancing the diversity of early exploration with the stability of later optimization.

[0061] After determining the reconstruction search range, the system generates a reconstructed route based on the previous optimal routes. Specifically, the generation method includes two branches: in the branch executed with local reconstruction probability, the system applies a Gaussian perturbation with the same dimensions as the reconstruction radius vector around the previous optimal routes, generating a locally perturbed route near the previous optimal routes; in the global branch executed with another probability, the system performs completely uniform random generation within the allowed range of route variables, allowing the algorithm to retain the opportunity to start over from other flyable corridors. Regardless of which branch is used to generate the reconstructed route, the system performs a complete flight safety constraint check on it, including terrain check, no-fly zone check, range check, angle check, and altitude corridor check, and calculates the corresponding engineering task cost. Each time a joint check and cost calculation is completed, the number of engineering evaluations increases by one.

[0062] After checking the reconstructed route, the system selectively updates it based on the check results. When a reconstructed route passes the hard constraint check, it is designated as the new current route. Simultaneously, when the engineering cost of the reconstructed route is less than the engineering cost of the previously optimal route, it is designated as the new previously optimal route, and its engineering cost is used as the engineering cost of the new previously optimal route. When a reconstructed route fails the hard constraint check, the system regenerates the reconstructed route within the remaining evaluation budget. When the evaluation budget is insufficient, the system retains the previously optimal route as the current route and prohibits writing unflyable routes into the current route status. This rigorous checking and updating mechanism ensures that the current route is always flyable, preventing subsequent adjustments from becoming invalid due to the introduction of unflyable routes.

[0063] After generating and updating the reconstructed routes, the system resets the control variables. For all multi-waypoint-altitude coordinated adjustment directions, the larger of the local reconstruction radius vector and the minimum adjustment scale for each direction is used as the new adjustment scale for that direction, restoring the step size of each direction to a level that matches the current reconstruction search range. Simultaneously, the orthogonal route adjustment datum is reset to the identity matrix, allowing the search direction for the next cycle to restart along the coordinate axes. The count of consecutive invalid cycles is cleared, and the optimal cost history window is refreshed using the engineering task cost of the current optimal route. After the above reset operations are completed, the system enters the next planning cycle in a completely new state and continues the direction traversal in step S2.

[0064] Step S5: When the planning termination condition is met, the optimal routes of each iteration are independently reviewed, and the optimal routes that pass the review are converted into flight control mission commands and issued.

[0065] Specifically, step S5 further comprises: rechecking whether the used engineering evaluation count is less than the total joint evaluation count (i.e., rechecking FEs<MaxFEs after the reconstruction judgment), if yes, entering the next planning cycle, if no, terminating the planning loop and sending all previous optimal routes to independent flight constraint review; After the iterative loop ends, all previous optimal routes z are re-verified with denser waypoint segment sampling than in the optimization phase for threat zones, hard no-fly zones, terrain clearance, total range, adjacent waypoint segment turning angles and altitude corridors, wherein if any constraint is not satisfied, the flight control execution instruction shall not be generated; All previous optimal routes that have passed the review are converted into a sequence of task instructions in the flight control coordinate system, and the waypoint sequence, altitude, speed and payload actions are issued to the flight controller, wherein the flight controller controls the unmanned aerial vehicle to perform intelligent inspection based on positioning and altitude feedback.

[0066] After the planning is terminated, the flight constraint review and waypoint instruction generation module conducts independent review on all previous optimal routes, comprising: recalculating the total range; checking the upper and lower altitude limits and terrain clearance point by point; checking the turning angle at each adjacent waypoint segment; checking the relationship between the waypoint segments and the threat zones, no-fly zones and geofences with denser sampling; converting the planar waypoints and segmented altitudes that have passed the review into latitude-longitude-altitude or local north-east-down coordinates in the flight control coordinate system, and appending marks for passing sequence, tolerance radius, cruise speed and inspection payload action. Only the instruction sequence that has passed the review is sent to the flight controller through the communication interface.

[0067] In simple terms, the final outputs are: all previous optimal 3D routes, the corresponding optimal task costs, and the flight control task instruction sequence obtained through independent safety review and coordinate conversion; the flight control task instruction sequence includes the waypoint sequence, flight altitude, cruise speed and inspection payload action.

[0068] In this embodiment, after completing the reconstruction judgment and processing in step S4, the system rechecks whether the used engineering evaluation count is less than the total joint evaluation count. If the used engineering evaluation count is still less than the total joint evaluation count, it indicates that the evaluation budget has not been exhausted, and the system returns to execute the next planning cycle, and continues to execute the cyclic adjustment from step S2 to step S4. If the used engineering evaluation count has reached the total joint evaluation count, it indicates that the evaluation budget has been exhausted, and the system terminates the planning loop, and sends all previous optimal routes to the independent flight constraint review phase. This termination condition ensures that the planning process is completed within a limited evaluation budget, and avoids the planning time exceeding the limit caused by an infinite loop.

[0069] During the independent flight constraint verification phase, the system re-verifies the various flight safety constraints of previous optimal routes with a denser segment sampling density than in the planning and optimization phase. Specifically, the verification includes at least the following: threat area avoidance check, verifying whether each segment of the route maintains a safe distance from preset threat sources; hard no-fly zone avoidance check, verifying whether the route enters any no-fly zones; terrain clearance check, verifying whether each segment of the route maintains sufficient vertical safety margin with the terrain safety envelope; total range check, verifying whether the cumulative flight distance of the entire route is within the maximum allowable range; adjacent segment turning angle check, verifying whether the turning angle between adjacent segments meets the maximum allowable turning angle limit of the aircraft; and altitude corridor check, verifying whether the flight altitude of each segment is always between the lower and upper limits of the allowable altitude. If any of the above constraints are not met, the route is considered to have failed the verification and flight control execution commands cannot be generated. This independent review phase is independent of the constraint checks in the planning phase. Because the review phase uses a higher sampling density than the optimization phase, it can effectively discover local safety hazards that may have been missed in the optimization phase due to sparse sampling, ensuring the safety and feasibility of the final output route in actual flight.

[0070] Once the optimal route has passed all the aforementioned independent flight constraints, the system converts the route into a sequence of mission commands in the flight control coordinate system and sends it to the flight controller. This sequence of mission commands includes at least the coordinates of waypoints arranged in sequence, the flight altitude of each segment, the cruise speed, and the corresponding inspection payload action markers for each waypoint or segment. Upon receiving this sequence of mission commands, the flight controller, based on real-time feedback information from the onboard positioning and altitude sensors, controls the UAV to fly sequentially along the planned route and perform corresponding inspection actions, thus completing the intelligent inspection task.

[0071] During actual flight, when the flight controller detects abnormal situations such as insufficient remaining battery power, degraded positioning quality, geofence updates, or the operator issuing new threat zones, it can suspend subsequent mission execution and send a replanning request to the route planning computer. When replanning is triggered, the system uses the UAV's current position as the new starting point, and the incomplete inspection objects and safety endpoints as new task inputs, re-executing the complete planning process from steps S1 to S5. This replanning mechanism ensures that the UAV can adjust its flight plan promptly in the face of dynamically changing environments or unexpected situations, improving the robustness and safety of mission execution.

[0072] Specifically, to further illustrate the effectiveness of this method, an inspection of UAV facilities within a three-dimensional area is used. The route planning program runs on a ground station computer, with the flight controller mounted on the UAV platform. The ground station and flight controller communicate via a data transmission link. The mission configuration interface can be provided by the inspection management software, and the environmental data interface can read data from a geographic information system, digital elevation model, manually marked no-fly zones, and threat source databases. When onboard computing power allows, the route planning program can also be deployed on the onboard computer and communicate with the flight controller via serial port, Controller Area Network (CAN) bus, or Ethernet. The route planning computer includes a processor, memory, and communication interface. The memory stores the environmental map index, mission parameters, upper and lower bounds of route variables, the current route, previous optimal routes, orthogonal route adjustment benchmarks, adjustment scales in each direction, progress windows, and the maximum number of route evaluations. A candidate route evaluation can invoke geometric distance calculation, map grid query, threat field calculation, energy consumption model, or flight dynamics simulation; regardless of the evaluation method used, it returns comparable mission costs and constraint states, as shown in Table 1.

[0073] Table 1. Engineering inputs and outputs of each module in the route planning system:

[0074] The reference working condition's two-dimensional working area is a normalized planar coordinate range from 0 to 100, with the starting point s located at (0, 0) and the ending point e located at (100, 100). Three threat sources with different threat levels and radii are set within this area. The flight path consists of 15 segments, therefore including 14 two-dimensional intermediate waypoints and 15 segment altitudes, resulting in a flight path variable dimension of 43. The UAV's maximum turning angle is 60°, the permissible altitude is 100 to 200 meters, the preferred cruising altitude is 125 meters, and the maximum range is 2000 meters. See Table 2 for details.

[0075] Table 2 Reference Inspection Condition Parameters:

[0076] Before the mission begins, the ground station standardizes map coordinates to local northeast coordinates or other planar coordinates, calculates the upper and lower bounds of each route variable, and generates a list of threat source locations, threat radii, and threat levels. The initial route can be obtained by interpolating and projecting straight lines from the origin and destination, or by using existing routes, the results of coarse planning from the A-satellite heuristic map search, or by a corridor specified by the operator. This invention does not limit the method of generating the initial route, but the initial route must pass basic altitude and boundary checks before entering the rotation reference adjustment phase.

[0077] The route planning program sets the current route as the initial route, initializes the previous optimal routes and mission costs to the current values, initializes the orthogonal route adjustment benchmark to the unit benchmark, initializes the adjustment scale of each direction to the preset proportion of the allowable range of the corresponding route variables, and sets the number of times the used route has been evaluated to the number of times the initial evaluation has been performed.

[0078] Within each route adjustment cycle, the program generates forward and backward candidate routes for each direction of the orthogonal datum. The candidate routes first pass through waypoints and altitude projections, then through no-fly zones, terrain, and zero-length segment checks. Candidate routes that fail the hard constraint checks can be directly assigned high penalty values ​​without entering high-fidelity simulation, thus saving evaluation time. For candidate routes that pass the basic checks, the length of each segment, adjacent segment turning angles, seven-point threat exposure, energy consumption, and altitude preference are calculated sequentially, and the mission cost is summarized.

[0079] If a lower-cost candidate route exists among the forward and backward candidate routes, the program accepts the lower-cost route as the new current route and updates the previous best routes simultaneously if it is superior to the previous best route. If neither candidate route shows improvement, the current route remains unchanged. After completing all directions, the net cooperative deformation is calculated based on the cycle start and end routes, and the orthogonal baseline is refreshed. Thus, success in one direction can include lateral obstacle avoidance, longitudinal target approach, and gradual altitude change at multiple waypoints. The next cycle will first continue evaluation along this combination of directions.

[0080] The optimization window is updated at the end of each cycle. If a local reconstruction is triggered, the program generates a disturbed route near the previous optimal route and performs projection and hard constraint checks on all variables. If a global reconstruction is triggered, the program generates a new route within the feasible box and then performs projection, geometric repair, and hard constraint checks. Only the reconstructed route that passes the hard constraint check becomes the new current route; the optimal record is updated when it is better than the previous optimal route, otherwise the optimal record remains unchanged.

[0081] After reaching the maximum number of route evaluations, the ground station re-examines the threat zone and geofence of each optimal route using a denser sampling of segments than in the optimization phase, and independently calculates the segment length, total distance, turning angle, and altitude. If the review fails, the previous optimal route that has passed the review is reverted, or a local reconstruction is performed centered on the failed segment within the remaining time.

[0082] After verification, the ground station converts local waypoints into latitude, longitude, and relative or absolute altitude for flight control, forming task items according to flight segment sequence. For line inspections, hovering, gimbal orientation, photography, or temperature measurement actions can be added to waypoints approaching towers or facilities; for pipeline inspections, constant speed and continuous load sampling markers can be added. After receiving the task items, the flight controller performs waypoint tracking and attitude control, with positioning and altitude sensors providing closed-loop feedback.

[0083] During flight, if the flight control system detects insufficient remaining battery power, degraded positioning quality, geofence updates, or the operator issues a new threat zone, it can pause subsequent tasks and send a replanning request to the route planning computer. During replanning, the UAV's current location is used as the new starting point, and the unfinished inspection targets and safety endpoints are used as new task inputs. The system then re-executes the flyable airspace establishment, route assessment, rotation reference adjustment, and command verification processes of this invention.

[0084] Figure 2 and Figure 3 The two-dimensional and three-dimensional routes of the planning module of this invention and several comparative planning methods are given under reference conditions. Figure 2 and Figure 3 r in x This indicates the longitudinal position coordinates of the UAV in the horizontal plane. Figure 2 and Figure 3 r in y This indicates the lateral position coordinates of the UAV in the horizontal plane. Figure 3 In this context, 'h' represents the UAV's flight altitude. Figures 2 to 5 In this context, RIPO represents the rotation information tracking planning module of the present invention; NL-SHADE-LBC represents the adaptive differential evolution comparison method with linear bias correction for successful history of nonlinear population reduction; WOA represents the whale optimization algorithm; MFPSO represents the hybrid moth-flame particle swarm optimization comparison method; GA-PSO represents the genetic algorithm-particle swarm hybrid optimization method; and QSMODE represents the multi-search strategy hybrid evolutionary optimization comparison method. These comparison abbreviations are for illustrative purposes only and are not essential modules of the present invention.

[0085] Ten independent planning iterations were performed for each method. Table 3 lists the optimal, median, mean, standard deviation, and average ranking of the final task cost. Lower task cost indicates better overall threat, energy consumption, altitude, and turning performance. This invention achieved the lowest optimal value of 377.00 and the lowest mean of 489.15, and tied for first place with an average ranking of 1.60. The median of WOA is slightly lower, but the standard deviation is significantly larger; the terminal results of this invention are more concentrated in the low-cost region.

[0086] Table 3. Final task cost of 10 independent planning simulations:

[0087] Figure 4 This invention demonstrates that it can continuously reduce the cost of previous optimal missions within a limited evaluation budget, and can perform fine-tuning in later stages through smaller route adjustment scales. Figure 4In this context, Function Evaluations represents the number of function evaluations, and Median Best-So-Far F(z) represents the median engineering task cost F(z) of all independently obtained historical best routes under each number of function evaluations. Figure 5 This indicates that most of the independent planning results of this invention are concentrated in the low-cost range. Figure 5 F(z) represents the engineering cost of the route. In the control group with random initialization and no route adjustments, the optimal cost is 7.96 × 10⁻⁶. 6 This indicates that the aforementioned low-cost routes are not generated naturally from variable boundaries, but are formed jointly by engineering assessment, collaborative direction extraction, and route reconstruction.

[0088] In summary, the rotation information tracking planning process of this invention takes a current 3D route that has passed basic boundary checks as the processing object, and sequentially performs bidirectional candidate route generation in orthogonal directions, evaluation of engineering task costs and flight constraints, cost-reduced route reception, extraction of net waypoint-altitude cooperative deformation, orthogonal benchmark updating, and route reconstruction under stationary conditions. Each evaluation in the planning process is for a candidate route that includes actual plane waypoints and segmented altitudes. After planning, only routes that have passed independent safety verification are converted into mission commands executable by the flight controller.

[0089] Compared with existing technologies, this method has the following advantages: First, this invention encodes inspection task data, environmental data, and UAV flight parameters into a three-dimensional route that includes intermediate plane waypoints and segmented altitudes. It also uses the same route engineering evaluation module to calculate threat exposure, energy consumption, turning, altitude preference, and constraint status, so that each route improvement based on the orthogonal benchmark update corresponds to an actual inspectable change in the inspection route.

[0090] Second, the present invention uses the net waypoint-altitude coordinated deformation formed by the received routes within a route adjustment cycle as the first direction of the orthogonal adjustment benchmark for the next cycle, so that subsequent candidate routes can simultaneously modify multiple waypoints and segment altitudes along the combined direction that has reduced the cost of engineering tasks. This helps to reduce redundant and invalid evaluations in fixed coordinate directions and improve the adaptability to inclined or curved flyable channels.

[0091] Third, this invention saves the current route, the previous optimal routes, the orthogonal adjustment benchmark, and the adjustment scale in each direction during the planning process. It does not require maintaining a candidate population consisting of multiple routes at the same time, thus reducing the storage of the planning state and allowing the evaluation budget to be concentrated on the two-way exploration of the current route, benchmark updates, and necessary route reconstruction. It is suitable for inspection scenarios where map queries, collision detection, or flight dynamics simulations are time-consuming.

[0092] Fourth, this invention determines the timing of route reconstruction based on the relative improvement of the optimal task cost within a preset window, the number of consecutive periods without effective adjustment, and the adjustment scale, avoiding the reliance solely on whether a single candidate route has improved as the basis for continuing local adjustments; the local reconstruction radius decreases with the proportion of evaluation budget consumption, while retaining global reconstruction branches, which is beneficial for balancing channel conversion in the early planning stage and route fine-tuning in the later planning stage.

[0093] Fifth, before generating flight control mission commands, this invention independently verifies the total flight distance, adjacent flight segment turning angles, flight altitude, terrain clearance, no-fly zones, and geofences using more frequent flight segment sampling than in the planning and evaluation phase. Only the verified routes undergo flight control coordinate transformation and are supplemented with speed and inspection payload actions. When geofences, remaining battery power, or environmental data change during execution, replanning is triggered, which helps improve the consistency between the planning results and the actual execution conditions of the UAV.

[0094] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A single-track rotation datum adaptive three-dimensional route planning method, characterized in that, include: Step S1: Obtain the input dataset and perform initialization processing based on the input dataset; Step S2: After the initialization process is completed, within the current planning cycle, bidirectional disturbances are performed on the current route along each multi-waypoint-altitude coordinated adjustment direction to generate candidate routes, and the engineering task cost is calculated for each candidate route. Candidate routes that fall within the terrain safety envelope or hard no-fly zones will be excluded; Based on the comparison of the engineering task costs between the candidate routes and the current route, one of the candidate routes with reduced costs is selected as the new current route. The search step size of the current direction is adjusted accordingly and the optimal routes of each previous cycle are updated based on whether the new current route is successfully received. After traversing all executable directions in this cycle, the direction traversal of the current planning cycle is completed. Step S3: Based on the end route of the cycle and the beginning route of the cycle, determine the net waypoint-altitude coordinated deformation within the current planning cycle; Based on the net waypoint-altitude coordinated deformation and the current orthogonal route adjustment benchmark, construct the benchmark matrix to be orthogonalized; Based on the numerical rank state of the reference matrix to be orthogonalized and the determination result of whether an effective net deformation has been formed in the current period, the reference matrix to be orthogonalized is selectively orthogonalized to update the orthogonal route adjustment reference, or the current orthogonal route adjustment reference is retained and the stall count is accumulated. Enter the next planning cycle and update the historical window data of the previous optimal costs; Step S4: When it is determined that the current route channel meets the preset reconstruction trigger conditions, route reconstruction is initiated; the reconstruction search range is set according to the proportion of the consumed assessment budget, and a reconstructed route is generated within the reconstruction search range based on the previous best routes, and flight safety constraint checks and engineering task cost assessments are performed on the reconstructed routes; the reconstructed routes that pass the checks are selectively updated to the current route and the previous best routes; after reconstruction is completed, the adjustment step size and adjustment benchmark in each direction are reset, the stall count is cleared, and the optimal cost history window is refreshed with the engineering task cost of the current previous best routes; Step S5: When the planning termination condition is met, the optimal routes of each iteration are independently reviewed, and the optimal routes that pass the review are converted into flight control mission commands and issued.

2. The single-track rotation reference adaptive three-dimensional route planning method according to claim 1, characterized in that, Step S1 is as follows: Obtain the input dataset, which includes the lower bound of the route variables, the upper bound of the route variables, a fixed starting point, and a fixed ending point; Based on the lower bound and upper bound of the route variables, the allowable route span vector, the maximum value of the allowable route span vector, the maximum adjustment scale, and the minimum adjustment scale are set sequentially. The initial route is generated by coarse planning based on a fixed starting point, fixed ending point, existing route or A-star heuristic graph. After feasible region projection and hard safety check of the initial route, the corresponding engineering task cost is calculated. The current route and the previous best routes are set as the initial route. The engineering task cost of the current route and the previous best routes are set as the engineering task cost of the initial route. The number of engineering evaluations used is set to 1. Initialize the D-dimensional adjustment scale vector, orthogonal route adjustment benchmark, planning cycle number as 0, consecutive invalid cycle count as 0, and optimal cost historical window.

3. The single-track rotation reference adaptive three-dimensional route planning method according to claim 2, characterized in that, Step S2 is as follows: Step S201: When it is determined that the number of engineering evaluations used is less than the total number of joint evaluations, proceed to the next planning cycle, save the cycle start route, and set the value of the receiving judgment logic flag to no; Step S202: Determine whether the number of engineering evaluations used is equal to the total number of joint evaluations; if yes, stop generating new candidate routes; if no, calculate the forward and backward candidate routes for the d-th multi-waypoint-altitude coordinated adjustment direction, and calculate the engineering task cost of the forward candidate route and the engineering task cost of the backward candidate route respectively. Each time the engineering task cost is calculated, the number of engineering evaluations used is incremented by 1.

4. The single-track rotation reference adaptive three-dimensional route planning method according to claim 3, characterized in that, Also includes: Step S203: When it is determined that the forward or backward candidate route of the d-th multi-waypoint-altitude coordinated adjustment direction intersects with the terrain safety envelope or hard no-fly zone, record its engineering task cost as +∞ and reject it; Step S204: Select the engineering task cost of the forward candidate route and the backward candidate route that has the smaller value among the engineering task costs of the d-th multi-waypoint-altitude coordinated adjustment direction. When it is determined that the engineering task cost is less than the engineering task cost of the current route, enter the route receiving branch. Step S205: Select the candidate route with the smaller engineering task cost as the new current route. When the engineering task cost of the current candidate route is equal to that of the backward candidate route, the forward candidate route is always selected as the new current route. Step S206: Select the smaller value from the product of the scale expansion coefficient and the D-dimensional adjustment scale vector corresponding to the d-th multi-waypoint-altitude coordinated adjustment direction, and the maximum adjustment scale, and use it as the new D-dimensional adjustment scale vector corresponding to the d-th multi-waypoint-altitude coordinated adjustment direction, and set the value of the receiving judgment logic flag to yes. Step S207: When it is determined that the engineering task cost of the current route is less than the engineering task cost of the previous optimal routes, write all waypoints, segment altitudes and constraint states of the current route into the previous optimal routes, and use the engineering task cost of the current route as the engineering task cost of the new previous optimal routes. Step S208: When it is determined that the engineering task cost of the current route is not less than the engineering task cost of the previous optimal routes, keep the current route unchanged, and take the product of the contraction coefficient and the D-dimensional adjustment scale vector corresponding to the d-th multi-waypoint-altitude coordinated adjustment direction as the new D-dimensional adjustment scale vector corresponding to the d-th multi-waypoint-altitude coordinated adjustment direction. Step S209: When it is determined that there is still a remaining assessment budget, for the (d+1)th multi-waypoint-altitude coordinated adjustment direction, execute steps S201 to S208, and repeat the above steps until there is no remaining assessment budget or all multi-waypoint-altitude coordinated adjustment directions within this planning period have completed steps S201 to S208.

5. The single-track rotation reference adaptive three-dimensional route planning method according to claim 4, characterized in that, Step S3 is as follows: The current route obtained after completing step S2 is taken as the end route of the cycle, and the net waypoint-altitude cooperative deformation is calculated based on the end route of the cycle and the start route of the cycle. Based on the net waypoint-altitude coordinated deformation and orthogonal route adjustment benchmark, construct the benchmark matrix to be orthogonalized for the current planning cycle; When it is determined that the numerical rank of the reference matrix to be orthogonalized in the current planning cycle is insufficient, the current orthogonal route adjustment reference is retained; When it is determined that the numerical rank of the reference matrix to be orthogonalized in the current planning cycle is sufficient, the value of the receiving judgment logic flag is yes, and the vector length of the net waypoint-altitude coordinated deformation is greater than the preset effective deformation threshold, QR orthogonal-upper triangular decomposition is performed on the reference matrix to be orthogonalized to obtain the updated D-dimensional orthogonal reference matrix and D-dimensional upper triangular matrix. The updated D-dimensional orthogonal reference matrix is ​​then used as the new orthogonal route adjustment reference, and the continuous invalid cycle count is set to 0.

6. The single-track rotation reference adaptive three-dimensional route planning method according to claim 5, characterized in that, Also includes: When it is determined that no effective net deformation has been formed in the current planning cycle, the current orthogonal route adjustment benchmark is retained, and the count of consecutive invalid cycles is incremented by 1; Enter the next planning cycle and update the historical data of the best cost. When it is determined that the index number of the current planning cycle is greater than or equal to the progress window length, calculate the best relative improvement ratio of each cycle.

7. The single-track rotation reference adaptive three-dimensional route planning method according to claim 6, characterized in that, Step S4 is as follows: When it is determined that the number of consecutive invalid cycles is greater than or equal to the threshold of consecutive invalid cycles, or the index of the current planning cycle is greater than or equal to the progress window length and the best relative improvement ratio of the current planning cycle is less than the relative improvement tolerance, or the maximum component value in the D-dimensional adjustment scale vector corresponding to each multi-waypoint-altitude coordinated adjustment direction is less than the preset minimum adjustment scale, it is determined that the current route channel is no longer worth allocating the assessment budget and route reconstruction is initiated. Based on the number of engineering assessments used and the total number of joint evaluations, set the budget consumption ratio, and set the local reconstruction radius vector based on the route allowable span vector; Based on the local reconstruction probability and the previous optimal routes, a reconstructed route is generated. The reconstructed route is then checked for terrain, no-fly zones, range, turning angle, and altitude, and the corresponding engineering task cost is calculated. When it is determined that the reconstructed route has passed the inspection, the reconstructed route will be used as the new current route. When it is determined that the engineering task cost of the reconstructed route is less than the engineering task cost of the previous best route, the reconstructed route will be used as the new previous best route, and the engineering task cost of the reconstructed route will be used as the engineering task cost of the new previous best route. When it is determined that the reconstructed route fails the inspection, the reconstructed route is regenerated within the remaining assessment budget. When the budget is insufficient, the best route from previous attempts is retained as the current route. For all multi-waypoint-altitude coordinated adjustment directions, the larger of the local reconstruction radius vector corresponding to the d-th multi-waypoint-altitude coordinated adjustment direction and the minimum adjustment scale is taken as the new D-dimensional adjustment scale vector corresponding to the d-th multi-waypoint-altitude coordinated adjustment direction. The orthogonal route adjustment benchmark is set to equal the D-dimensional identity matrix, the consecutive invalid cycle count is 0, and the optimal cost history window is refreshed with the engineering task cost of the current optimal route.

8. The single-track rotation reference adaptive three-dimensional route planning method according to claim 1, characterized in that, Step S5 is as follows: Recheck whether the number of engineering assessments used is less than the total number of joint assessments. If so, proceed to the next planning cycle. If not, end the planning cycle and send the optimal routes of each time to the independent flight constraint review. The threat zone, hard no-fly zone, terrain clearance, total flight distance, adjacent segment turning angle and altitude corridor of the previous optimal flight route are re-verified with more dense segment sampling than in the optimization phase. If any constraint is not met, flight control execution commands shall not be generated. The verified optimal flight paths are converted into a sequence of mission commands in the flight control coordinate system, and waypoint sequence, altitude, speed and payload actions are sent to the flight controller. The flight controller controls the UAV to perform intelligent inspection based on positioning and altitude feedback.

Images