A scene-driven multi-strategy optimization unmanned aerial vehicle path planning method and system

Abstract

This invention provides a scenario-driven multi-strategy optimization method and system for UAV path planning, belonging to the field of UAV path planning technology. Based on DEM elevation benchmark data and airborne sensor data, it constructs wind direction and asymmetric vertical energy consumption models, sensor field-of-view coverage models, airspace safety constraint models, and no-fly zone constraint models. It then constructs a scenario-driven weighted linear scalar composite fitness function to normalize multiple objectives to the same dimension space, coupled with an exponential outlier penalty mechanism to achieve zero tolerance for rigid safety constraints. Finally, it employs an improved multi-strategy chaotic evolutionary optimization algorithm (MSCEO) integrating five mechanisms, including adaptive chaotic sampling and mirror reflection boundary repair, to perform optimization calculations. Finally, it generates a cubic B-spline smoothed trajectory using the optimal trajectory control points and periodically initiates online replanning based on airborne data. This invention can effectively reduce UAV flight energy consumption and significantly improve the convergence robustness, environmental adaptability, and engineering implementation reliability of the planning method.

Description

Technical Field

[0001] This invention belongs to the field of UAV path planning technology, and specifically refers to a scenario-driven multi-strategy optimization method and system for UAV path planning. Background Technology

[0002] With the rapid iteration of drone technology and the continuous expansion of engineering application scenarios, drones have become the core execution carrier for various low-altitude operations. With their outstanding advantages such as mobility, high deployment efficiency, and no restrictions on ground access conditions, they are widely adaptable to diverse operational needs.

[0003] In real-world operational scenarios, drones often face complex and ever-changing airspace environments and multiple mission constraints. On the one hand, the vertical ascent and descent processes of drones exhibit significant asymmetric energy consumption characteristics, and the spatiotemporal differences in airspace wind fields directly impact horizontal flight power, resulting in multi-dimensional coupling of energy consumption factors. On the other hand, the operational process must simultaneously meet multiple constraints, including airspace restrictions, safe clearance distances, and mission coverage effectiveness. These complex conditions place higher technical demands on the environmental modeling accuracy, multi-constraint coordinating capabilities, online computation real-time performance, and engineering deployment robustness of drone path planning methods.

[0004] Existing drone inspection path planning technology still faces the following prominent problems in practical applications.

[0005] First, the environmental modeling lacks fidelity. Most schemes are based on simplified two-dimensional terrain models or static three-dimensional models, only considering geometric constraints such as terrain collision and trajectory curvature, without incorporating actual physical constraints such as vertical asymmetric flight energy consumption, horizontal wind disturbance, no-fly zones, and effective sensor coverage into a unified modeling framework. The lack of these physical factors leads to a serious disconnect between the planned path and actual flight conditions, manifesting as significantly higher actual flight energy consumption, insufficient safety margins, and in severe cases, even rendering the planned path unenforceable in real-world environments. Second, the real-time performance of multi-objective optimization is difficult to guarantee. Although the Pareto-based optimization framework can effectively balance the conflict relationships among multiple objectives, the computational cost of its non-dominated sorting and crowding calculations increases dramatically with the population size and the number of objectives, resulting in excessively long iteration times. This computational load is difficult to handle on resource-constrained embedded microcontrollers, making it impossible to achieve high-frequency online trajectory replanning, and thus failing to meet the real-time requirements of UAV flight control.

[0006] Third, existing chaotic evolutionary optimization algorithms have inherent flaws when applied to constrained optimization, resulting in insufficient engineering robustness. These algorithms leverage the ergodicity of chaotic sequences to enhance global search capabilities to some extent, but existing improvements often employ static boundary absorption strategies to handle the problem of search individuals crossing boundaries. When a search individual touches a terrain constraint boundary or a no-fly zone boundary, this strategy forces feasibility recovery by directly pruning the outbound individual to the boundary or randomly resetting its position. However, this simplistic approach leads to a large number of individuals clustering near the constraint boundary, causing a sharp decline in population diversity, loss of search vitality, and ultimately trapping the algorithm in local optima or in inescapable stagnation areas such as terrain blind valleys or ridges.

[0007] Therefore, developing a path planning method that can accurately reproduce complex mountainous terrain with multiple physical conditions, efficiently solve multi-constrained 3D paths, and be lightweight enough to be deployed on embedded MCUs, in order to overcome the technical defects of the existing technologies, such as insufficient environmental modeling fidelity, poor real-time performance of multi-objective optimization, and the tendency of chaotic evolutionary optimization algorithms to get stuck in topological stagnation, has become a key technical problem that urgently needs to be solved in this field. Summary of the Invention

[0008] To address the technical problems existing in the prior art, the present invention provides a scenario-driven multi-strategy optimization method and system for UAV path planning, the technical solution of which is as follows: On the one hand, a scenario-driven, multi-strategy optimization method for UAV path planning is provided, which includes: S1. Acquire DEM elevation benchmark data and airborne sensor data; S2. Based on the DEM elevation benchmark data and airborne sensor data, construct the wind direction and asymmetric vertical energy consumption model, sensor field of view coverage model, airspace safety constraint model and no-fly zone constraint model to form a complete three-dimensional environmental constraint system. S3. Construct a scenario-driven weighted linear scalar composite fitness function to uniformly map coverage gap, wind field-oriented energy consumption cost, and spatial path length to a comparable scalar space. At the same time, treat terrain boundary crossing and no-fly zone intrusion as rigid constraints and achieve zero tolerance through high extra points penalty. S4. The improved multi-strategy chaotic evolutionary optimization algorithm MSCEO is used to perform path optimization. MSCEO integrates five mechanisms: adaptive chaotic sampling, linear decay dual-strategy trial vector generation, successful history memory and external archiving, mirror reflection boundary repair and greedy selection, and linear population size reduction. It performs optimization calculations according to the complete time sequence of population initialization, iterative update and convergence termination, and dynamically balances global exploration and local development capabilities. S5. After iterative convergence, the optimal trajectory control point sequence obtained by optimization is used as the control vertex of the cubic B-spline curve to generate a smooth three-dimensional trajectory and send it to the UAV flight control platform. At the same time, online replanning is periodically initiated in combination with the environmental data transmitted back in real time by the airborne sensors.

[0009] On the other hand, a scenario-driven multi-strategy optimized UAV path planning system is provided. The system adopts a two-level architecture of airborne execution carrier + control and computing core, and is composed of two parts: UAV flight control platform and flight control management device. The UAV flight control platform is the aerial physical execution carrier for the inspection mission, and integrates an environmental perception module, a wireless communication module, a flight control execution module, and a mission sensor module. The environmental perception module is used to collect real-time data on the terrain elevation, UAV real-time flight altitude, airspace wind field vector, and no-fly zone information of the inspection area, providing raw data support for subsequent environmental modeling and constraint verification; the wireless communication module is used to realize bidirectional data transmission between the UAV flight control platform and the flight control management device, ensuring stable interaction between downlink flight path commands and uplink sensor data; the flight control execution module is used to receive and execute flight path control commands issued by the flight control management device, achieving high-precision tracking flight of the planned trajectory through attitude calculation and power output control; the mission sensor module is used to collect optical, visible light, and infrared images of the inspection area during cruise flight. The flight control and management device is the ground or airborne computing and control core of the system, including a data receiving module, a central control module, a three-dimensional environment modeling module, and a path optimization module; The data receiving module receives raw sensor data from the UAV flight control platform and performs standardized preprocessing. The central control module coordinates task scheduling, operational status management, and workflow control, triggering model updates and trajectory replanning based on environmental changes and task requirements. The 3D environment modeling module constructs wind direction and asymmetric vertical energy consumption models, sensor field-of-view coverage models, airspace safety constraint models, and no-fly zone constraint models based on preprocessed multi-source sensor data, forming a complete 3D environmental constraint system. The path optimization module is equipped with an improved multi-strategy chaotic evolutionary optimization algorithm, MSCEO. Based on various physical models output by the 3D environment modeling module and a preset scene-driven weighted linear scalar composite fitness function, it sequentially performs a complete time-series calculation of population initialization, iterative optimization, and convergence termination, ultimately outputting a 3D trajectory smoothed by cubic B-spline processing. This trajectory is then transmitted to the flight control execution module of the UAV flight control platform via a wireless communication module, and online replanning is periodically initiated in conjunction with real-time environmental data transmitted from airborne sensors.

[0010] On the other hand, an electronic device is provided, comprising a processor and a memory, wherein the memory stores at least one instruction, which is loaded and executed by the processor to implement the scenario-driven multi-strategy optimized UAV path planning method described above.

[0011] On the other hand, a computer-readable storage medium is provided, wherein at least one instruction is stored in the storage medium, the at least one instruction being loaded and executed by a processor to implement the above-described scenario-driven multi-strategy optimized UAV path planning method.

[0012] The beneficial effects of the technical solution provided by this invention include at least the following: 1. Improve the fidelity of environmental modeling and plan routes to match real mountain conditions. This invention constructs a unified modeling framework encompassing vertical asymmetric flight energy consumption, horizontal wind field disturbance, no-fly zones, dynamic sensor field of view coverage, and terrain safety constraints. It incorporates all the core physical constraints in mountain patrol scenarios into the planning system, solving the problem that traditional simplified models only consider geometric constraints and are severely out of touch with real flight conditions. It can accurately quantify the energy consumption differences under different wind conditions and vertical flight states, while rigidly ensuring terrain clearance and no-fly zone safety, effectively reducing actual flight energy consumption, improving flight safety margin, and ensuring that the planned path can be implemented in real mountain environments.

[0013] 2. Reduce and optimize computational load to meet the needs of embedded deployment and real-time replanning. This invention employs a dimensionless normalization and weighted linear scalar single-objective fitness aggregation method to replace the traditional Pareto multi-objective optimization framework. This eliminates the high-complexity computations of non-dominated sorting and crowding calculation, significantly reducing the computational load and memory consumption of a single iteration. The algorithm can be lightweightly deployed on resource-constrained embedded microcontrollers, supports high-frequency online trajectory replanning, and fully meets the real-time requirements of UAV flight control.

[0014] 3. A multi-mechanism collaborative optimization of the chaotic evolution algorithm comprehensively improves solution performance and engineering robustness. This invention achieves systematic optimization of the chaotic evolutionary optimization algorithm by integrating five mechanisms: adaptive chaotic sampling, linear decay dual-strategy trial vector generation, successful history memory and external archiving, mirror reflection boundary repair, and linear population size reduction. Adaptive chaotic sampling dynamically adjusts the search intensity, balancing global exploration capability with computational overhead; dual-strategy trial vector generation enables a smooth transition from early-stage wide-area search to later-stage elite refinement, avoiding premature convergence; successful history memory adaptively optimizes core parameters, and external archiving enriches search directions; mirror reflection boundary repair preserves search momentum, solving the problem of a sharp drop in population diversity and topological stagnation caused by boundary clustering; and linear population size reduction optimizes computational power allocation, improving later-stage refinement efficiency. Through the synergistic effect of these multiple mechanisms, the algorithm's convergence stability, solution accuracy, and operational efficiency are comprehensively improved. It is adaptable to embedded lightweight deployments, significantly enhancing the reliability of engineering implementation in complex constraint scenarios. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a flowchart of a scenario-driven multi-strategy optimization UAV path planning method provided by an embodiment of the present invention; Figure 2 This is a schematic diagram of scene model construction provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the execution path optimization using the improved multi-strategy chaotic evolutionary optimization algorithm MSCEO provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0017] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0018] This invention provides a scenario-driven, multi-strategy optimized UAV path planning method, which can be implemented by an electronic device, such as a terminal or a server. Figure 1 The diagram shown is a flowchart of the method. The processing flow may include the following steps: S1. Acquire DEM elevation benchmark data and airborne sensor data; Optionally, embodiments of the present invention can acquire DEM elevation benchmark data and airborne sensor data through an environmental perception module. This environmental perception module is composed of four types of sensing units: a lidar, a barometric altimeter, a wind speed and direction sensor, and a thermal imaging sensor. The lidar is installed below the UAV fuselage, covering the ground directly below and the forward flight direction. During operation, it collects 3D point cloud data of the inspection area in real time at a preset frequency. After point cloud denoising, ground point filtering, and coordinate calculation, the 3D coordinate information of the terrain surface is obtained, providing high-precision basic data for 3D terrain modeling. The barometric altimeter is integrated into the UAV's flight control and navigation unit, collecting atmospheric pressure data in real time at a preset frequency. The elevation calculation model converts the absolute altitude and relative ground altitude of the UAV, providing altitude feedback data for terrain clearance safety constraint verification and trajectory altitude closed-loop control. The wind speed and direction sensor is installed in an unobstructed position on the top of the UAV to avoid interference from the airflow of the fuselage with the measurement accuracy. It collects the wind speed and wind direction angle of the inspected airspace in real time at a preset frequency, and calculates the two-dimensional wind field vector, providing real-time wind field input for asymmetric energy consumption modeling and wind disturbance flight compensation. The thermal imaging sensor adopts a downward-looking wide-angle infrared thermal imaging device, which is installed in the mission pod under the UAV fuselage. It performs infrared scanning of the inspected area at a preset frequency, and identifies high-temperature fire points and hot updraft areas in the area through thermal imaging image processing algorithms. It outputs the center coordinates and spatial influence range parameters of the thermal threat area, providing a data foundation for the construction of the no-fly zone model.

[0019] S2. Based on the DEM elevation benchmark data and airborne sensor data, construct the wind direction and asymmetric vertical energy consumption model, sensor field of view coverage model, airspace safety constraint model and no-fly zone constraint model to form a complete three-dimensional environmental constraint system. A schematic diagram of scene model construction in an embodiment of the present invention is shown below. Figure 2 As shown.

[0020] Optionally, the construction of the wind direction and asymmetric vertical energy consumption model specifically includes: Based on discrete wind field data measured by airborne wind speed and direction sensors, a continuous wind field distribution across the entire domain is generated through spatial interpolation. The cosine similarity between the horizontal direction of the flight path and the wind field direction for each flight segment is calculated. The differential impact of tailwind, headwind, and crosswind on the UAV's propulsion power is quantified, and a wind field correction factor is obtained. This provides a calculation basis for subsequent wind field disturbance correction of horizontal flight energy consumption. Among these factors: For the m-th discrete track segment, its horizontal displacement vector is defined as: The corresponding horizontal segment length is: Let the horizontal wind field vector in the planning time domain be: Cosine similarity is used to describe the azimuth relationship between the flight direction and the wind field direction during the flight path segment: In the formula, To prevent extremely small positive numbers from being divided by zero; The value range is approximately [-1, 1]: a positive value indicates that the flight path is in the same direction as the wind field, corresponding to a tailwind condition; a negative value indicates that the flight path is in the opposite direction to the wind field, corresponding to a headwind condition; a value close to zero indicates that the flight direction is perpendicular to the wind direction, corresponding to a crosswind condition. Wind field correction factor constructed based on cosine similarity. To achieve differentiated calculation of horizontal energy consumption under different wind conditions: In the formula, For wind direction sensitivity coefficient, satisfying The wind field correction factor is used to characterize the impact of wind direction changes on the horizontal propulsion power of UAVs. Under this definition, the wind field correction factor is less than 1 in the tailwind flight segment, which corresponds to a reduction in the horizontal energy consumption value; the wind field correction factor is greater than 1 in the headwind flight segment, which corresponds to an increase in the horizontal energy consumption value; and the correction factor is close to 1 in the crosswind condition, and the energy consumption level is the same as the no-wind baseline. The planned 3D flight path is discretized into M equal-length flight segments. The energy consumption differences between climb, level flight, and descent are differentiated. Combined with a wind field correction factor, the total flight energy consumption is calculated to accurately characterize the asymmetric energy consumption characteristics of vertical flight in mountainous environments. The calculation formula is as follows: In the formula, E is the total flight energy consumption cost after wind field correction; m is the index of the discrete segment of the track; M is the total number of discrete sampling points of the track; The baseline level flight energy consumption coefficient characterizes the energy consumption per unit horizontal distance during level flight under windless conditions. Let m be the horizontal displacement length of the m-th segment of the trajectory; is the wind field correction factor corresponding to the m-th segment of the trajectory; The climbing energy consumption coefficient represents the additional energy consumed per unit vertical climbing height. Let be the vertical climb altitude of the m-th segment of the trajectory, where This represents the vertical displacement of the flight path segment; The energy consumption coefficient is used to characterize the energy consumption per unit vertical descent height. Let m be the vertical descent altitude of the m-th segment of the trajectory, and the model satisfies... The constraint relationship, namely that the climbing energy consumption coefficient is significantly greater than the descending energy consumption coefficient, is consistent with the actual energy consumption law of UAV vertical flight and can accurately reflect the additional energy consumption burden caused by repeated climbing in mountainous areas.

[0021] Optionally, constructing the sensor field-of-view coverage model specifically includes: A mapping relationship is established between the UAV's flight altitude and the sensor's instantaneous ground detection radius to reflect the dynamic impact of flight altitude changes on the inspection coverage. Simultaneously, a hardware limitation on the sensor's maximum effective detection radius is introduced. The calculation formula is as follows: In the formula, Represents the ground projection radius of the sensor at the m-th sampling time; This represents the real-time flight altitude of the UAV relative to the ground at the m-th sampling time; This represents the half-angle of the sensor's field of view, which is determined by the sensor's hardware parameters. This represents the maximum effective sensing radius of the sensor, which is the upper limit of hardware performance.

[0022] Optionally, the construction of the airspace safety constraint model specifically includes: Based on DEM elevation benchmark data, the real-time ground clearance of each trajectory sampling point relative to the terrain surface is calculated. A minimum safe clearance height threshold is preset, and the terrain penetration violation depth of the trajectory points is quantified to provide a basis for subsequent safety penalty calculations. The calculation formula is as follows: In the formula, This represents the terrain clearance violation depth at the m-th trajectory sampling point; This is the preset minimum safe ground clearance height for drones; The real-time ground clearance of the UAV relative to the terrain surface at the m-th sampling point; when the trajectory point meets the airspace safety requirements, the violation depth is 0; when the trajectory is below the safe height, the violation depth increases with the degree of terrain penetration, adapting to the undulating characteristics of mountainous terrain, avoiding planning deviations such as compliance in valleys and violations on ridges at the same absolute height, and rigidly ensuring the terrain safety constraints of inspection flights.

[0023] Optionally, the construction of the no-fly zone constraint model specifically includes: The spatial boundary of the danger zone is described using a three-dimensional ellipsoid equation. This determines whether a trajectory point intrudes into the no-fly zone and calculates the intrusion depth for subsequent safety penalty calculations. For the j-th no-fly zone, its spatial boundary satisfies the following ellipsoid equation: In the formula, Indicates the center coordinates of the j-th no-fly zone; These represent the semi-axis lengths of the no-fly zone in the three coordinate axes (x, y, z). When the coordinates of the trajectory point are substituted into the above formula and the condition is less than 1, it is determined to be an intrusion into the no-fly zone, and the intrusion depth is the distance value inside the ellipsoid; otherwise, it is determined to be in a safe area.

[0024] S3. Construct a scenario-driven weighted linear scalar composite fitness function to uniformly map coverage gap, wind field-oriented energy consumption cost, and spatial path length to a comparable scalar space. At the same time, treat terrain boundary crossing and no-fly zone intrusion as rigid constraints and achieve zero tolerance through high extra points penalty. Optionally, S3 specifically includes: S3.1 Multi-objective dimensionless normalization processing: To address the differences in physical dimensions and numerical magnitudes among three categories of physical quantities—coverage gap, wind field guiding energy consumption cost, and spatial path length—normalization mapping is performed to uniformly transform them into the [0,1] scalar interval, thus eliminating dimensional differences. (1) Normalization of coverage missing value Using the complement of the global inspection coverage rate as an indicator of coverage gap, it directly reflects the proportion of blind spots in the inspection coverage of candidate tracks. The calculation formula is as follows: In the formula, X is the control point decision vector corresponding to the candidate trajectory; The global inspection coverage of the candidate track is defined, with a value range of [0,1]. The normalized coverage gap is [0,1]. The smaller the value, the better the inspection coverage of the candidate track and the lower the proportion of blind spots. Among them, the overall inspection coverage rate In conventional modeling of UAV inspection path planning, the discrete grid statistical method is generally used. The core is to calculate the proportion of target grids effectively covered by airborne sensors within the statistical task area to the total number of target grids. The formula is as follows: In the formula, N is the total number of target grids to be inspected within the task area (usually the target inspection area is discretized into two-dimensional ground grids at a fixed resolution, which can be replaced by voxels in a three-dimensional scene). For the j-th target grid cell to be inspected; This is an indicator function that takes the value 1 when the condition is met and 0 when the condition is not met. (2) Wind farm directional energy consumption normalization The ratio of the total flight energy consumption after wind field correction to the preset baseline energy consumption is used as the normalized energy consumption index to eliminate the interference of the absolute magnitude of energy consumption on the weighting configuration. The calculation formula is as follows: In the formula, E(X) represents the total energy consumption cost of wind field guidance corresponding to the candidate trajectory; The baseline energy consumption reference value is used as the theoretical energy consumption value under the straight-line level flight condition between the start and end points of the flight path, and it is used as the unified benchmark for energy consumption evaluation. This represents the normalized energy consumption cost; the smaller the value, the higher the energy efficiency of the candidate trajectory. (3) Path length normalization The ratio of the total spatial length of the trajectory to the straight-line distance between the origin and destination points is used as the normalized path length index to reflect the degree of detour of the candidate trajectory relative to the theoretical shortest path. The calculation formula is as follows: In the formula, L(X) is the total spatial path length corresponding to the candidate track; The straight-line distance between the start and end points of the track serves as a unified benchmark for evaluating path length. The normalized path length is represented by a smaller value, indicating a more compact candidate path and a lower degree of spatial detour. S3.2 Construction of the external point penalty term for rigid safety constraints: For two types of rigid safety constraints—terrain clearance violations and no-fly zone incursions—an out-of-point penalty mechanism of "baseline penalty + depth weighting" is adopted to construct the total safety penalty term. The magnitude of the penalty term is much larger than the sum of the values ​​of all task objective terms, ensuring that candidate tracks with safety violations are directly eliminated during the evolution process, achieving zero-tolerance screening of safety violations. At the same time, the depth weighting term provides a continuous penalty gradient to guide the search towards the feasible region. (1) Penalties for violations of terrain clearance For each trajectory sampling point, the clearance violation is simultaneously overlaid with baseline penalty and depth penalty, calculated as follows: In the formula, This is an indicator function that takes the value 1 when the condition inside the parentheses is met, and takes the value 0 when the condition is not met. The terrain clearance violation depth for the m-th trajectory sampling point; The baseline penalty value for terrain violations is a very large constant that is much larger than the sum of the mission objectives. This ensures that as long as there is a terrain clearance violation, the fitness of the candidate track will be significantly degraded and it will be directly eliminated in the evolutionary selection. is the terrain violation depth weighting coefficient, used to quantify the penalty gradient of violation depth. The more severe the violation, the higher the penalty value, guiding the search to converge in the direction that meets the clearance requirements; M is the total number of trajectory sampling points; (2) Penalties for intruding into the no-fly zone For each trajectory sampling point, the no-fly zone intrusion behavior is simultaneously superimposed with baseline penalty and intrusion depth penalty, calculated as follows: In the formula, J represents the total number of no-fly zones for thermal threats; The algebraic distance between the m-th sampling point and the j-th ellipsoidal no-fly zone; To normalize the depth of intrusion, a larger value indicates a deeper intrusion into the no-fly zone; The penalty value for intruding into the baseline of the no-fly zone is a very large constant that is much larger than the sum of the mission objectives, ensuring that as long as there is a no-fly zone intrusion, the candidate track will be marked as infeasible; These are the intrusion depth weighting coefficients, used to quantify the penalty gradient of the intrusion depth; (3) Total safety penalty items The total safety penalty is the sum of the penalty for terrain clearance violations and the penalty for intrusion into thermal threat no-fly zones, calculated using the following formula: When a candidate track fully satisfies all safety constraints, the total safety penalty term is 0; when any safety violation exists, the penalty term value increases sharply, directly determining the degree of fitness degradation of the candidate track and rigidly ensuring the highest priority of safety constraints. S3.3 Construction of Weighted Linear Scalar Composite Fitness Function A scenario-specific weight vector is set, and the three types of normalized task objectives are weighted and aggregated. Then, a rigid safety penalty term is added to obtain the final single scalar fitness function, calculated as follows: In the formula, , , These are task weight coefficients for coverage gap, energy consumption cost, and path length, respectively, satisfying... The weighting coefficients are pre-configured based on the inspection task scenario: for daily inspection scenarios covering the entire area, the coverage weight is increased to ensure inspection coverage; for fixed-point fire verification scenarios, the energy consumption weight is increased to extend the endurance time; for emergency rapid response scenarios, the path length weight is increased to shorten the arrival time. The optimization objective of path optimization is to minimize the fitness function value J(X).

[0025] S4. The improved multi-strategy chaotic evolutionary optimization algorithm MSCEO is used to perform path optimization. MSCEO integrates five mechanisms: adaptive chaotic sampling, linear decay dual-strategy trial vector generation, successful history memory and external archiving, mirror reflection boundary repair and greedy selection, and linear population size reduction. It performs optimization calculations according to the complete time sequence of population initialization, iterative update and convergence termination, and dynamically balances global exploration and local development capabilities. Optionally, such as Figure 3 As shown, the improved multi-strategy chaotic evolutionary optimization algorithm MSCEO for execution path optimization specifically includes: S4.1 Algorithm Initialization Preset core parameters such as maximum number of iterations, upper and lower limits for population size, upper and lower limits for chaotic sampling, length of historical memory pool, population diversity threshold, maximum number of stalled generations, and convergence threshold; in the decision space corresponding to the three-dimensional trajectory control points. Initial population is generated randomly within the population. Each individual within the population A set of 3D track control point coordinates corresponds to a candidate path scheme; initialize the scaling factor memory pool. Cross rate memory pool The array is composed entirely of 0.5 elements, and the memory pool has a length of H. The external archive A is initialized to be empty. The fitness values ​​of all individuals in the initial population are calculated, and the initial globally optimal individual is selected. And set the iteration counter t to 0 and the continuous stagnation algebra counter s to 0.

[0026] S4.2 Iterative Optimization The loop is centered on the dynamic balance between global exploration and local development. It operates in concert through five core mechanisms, updates the population generation by generation, and repeats the process until the convergence condition is met. S4.3 Convergence Termination Judgment: The termination condition is checked in real time during the iteration process. The iterative optimization process is terminated when any of the following conditions are met: First, the number of iterations reaches the preset maximum number of iterations. ; Second, continuous The change in the fitness value of the globally optimal individual is less than the set convergence threshold.

[0027] After the iteration terminates, the sequence of 3D track control points corresponding to the current global optimal solution is output as the input for the subsequent three B-spline track smoothing processes.

[0028] Optionally, the specific implementation of the five mechanisms of MSCEO is as follows: (1) Adaptive chaotic sampling mechanism In each iteration, population diversity is monitored in real time, and the number of chaotic samples is dynamically adjusted to balance global exploration capability and computational overhead. The specific steps are as follows: ① Calculate the mean vector of the population in generation t. The average Euclidean distance from individuals within a population to the mean vector is used as an indicator of diversity. In the formula, For the population size of generation t, Let be the decision vector of the i-th individual in the t-th generation; The diversity ratio is obtained by normalizing the diversity index: In the formula, , These represent the upper and lower boundaries of the decision space, respectively. To prevent extremely small positive numbers from being divided by zero; ② Update the number of chaotic samples based on the diversity ratio and the stagnation state of the population: when Below the preset threshold or global optimal fitness continuous When the situation is not improved, increasing the number of chaotic samples enhances global exploration and helps the population escape local optima; when Higher than the preset threshold Furthermore, as the global optimal fitness continues to improve, the number of chaotic samples is reduced to lower computational overhead, focusing on local development; the update rule is: In the formula, As the expansion factor, It is a contraction factor; , , respectively, represent the lower and upper limits of the number of chaotic samples; s(t) is the stagnation algebra counter for the t-th generation; ③ Generate chaotic samples through a two-dimensional exponential discrete chaotic mapping: In the formula, k is the chaotic mapping control coefficient. This represents element-wise multiplication, which normalizes the generated chaotic variables and maps them to the decision space to obtain chaotic samples. , used to construct the probing direction of a chaotic exploration strategy; (2) Linear decay dual-strategy trial vector generation mechanism Two strategies, chaotic exploration and archive-assisted elite refinement, are used to generate trial vectors. The selection probability of the chaotic strategy decreases linearly with iteration, achieving a smooth transition from wide-area exploration in the early stage to local refinement in the later stage. The specific steps are as follows: ① The probability of choosing a chaotic strategy decreases linearly with the number of iterations: In the formula, Choose probabilities for the initial chaotic policy. To determine the maximum number of iterations, for each parent individual, with probability... Choose a chaotic exploration strategy, based on probability. In the early stages of iteration, the probability of using the elite strategy is high, focusing on global search; in the later stages of iteration, the proportion of the elite strategy increases, focusing on local convergence. ② The elite refinement strategy employs an archive-assisted mutation method from the current to the optimal population, where the proportion of elite individuals decreases linearly with iteration: In the formula, , These represent the initial and final elite ratios, respectively, and the mutation formula is: In the formula, Let be the scaling factor for the i-th individual; From the top of fitness ranking Randomly selected from individuals; Randomly select from the current population; Randomly selected from the union of the current population and the external archive; and the three individual indices are all different and none of them are equal to the parent individual index i; ③ The chaotic exploration strategy generates candidate vectors based on chaotic samples: In the formula, The chaos step size coefficient, Chaotic samples generated by an adaptive chaotic sampling mechanism; ④ Perform a binomial crossover between the mutated candidate vector and the parent individual to generate a trial vector: In the formula, Let be the crossover rate of the i-th individual; A uniform random number within the interval ([0,1]); The dimension index is randomly selected to ensure that at least one dimension of mutation information enters the trial vector, thus avoiding complete replication of the parent individual; (3) Successful historical memory and external archiving mechanism The process involves recording parameter combinations that generate high-quality solutions during evolution, adaptively adjusting the distribution of scaling factors and crossover rates, and reusing replaced parent individuals to maintain population diversity. The specific steps are as follows: ① Scaling factor for each individual and cross rate Samples were taken from the successful historical memory pool: In the formula, M_F and M_{CR} are the scaling factor and the successful history memory pool of the crossover rate, respectively, and both have a length of H; Let H be a memory pool index randomly selected from ([1,H]). If the sampled index is... Then resample, if Then truncate to 1; Cut off to the interval ([0,1]); ② If the fitness of the trial vector is better than that of its corresponding parent, the evolution is considered successful, and the scaling factor, crossover rate, and fitness improvement amount corresponding to the trial vector are recorded; after each generation of evolution, the parameters at the corresponding position in the memory pool are updated according to the fitness improvement amount. In the formula, S is the set of all successfully evolved individuals in the current generation; Let be the fitness improvement amount for the i-th individual; For extremely small positive numbers; memory pool index k is... The parameters are incremented sequentially to achieve rolling updates. ③ External archives are used to store parent individuals that are replaced during evolution; the maximum archive capacity is [missing information]. When the number of individuals in the archive exceeds the capacity limit, the excess individuals are randomly deleted. The individuals in the archive only participate in the mutation operation of the elite refinement strategy and do not participate in the population fitness ranking and selection. By introducing differential information of historical inferior solutions, the search direction is enriched and the population diversity is maintained. (4) Mirror reflection boundary repair and greedy selection mechanism Mirroring and reflecting outbounded trial vectors helps prevent diversity loss caused by boundary clustering. Greedy selection is used to retain high-quality solutions, driving the population to evolve towards better fitness. The specific steps are as follows: ① Perform mirror reflection repair on the dimensional components of the trial vector that exceed the decision boundary: In the formula, , Let be the lower and upper boundaries of the d-th dimension decision variable, respectively. Let d be the d-th component of the i-th trial vector in the t-th iteration. These are the corresponding dimensional components after a single mirror reflection process; If the reflected component still exceeds the boundary, then clip to the nearest boundary value: In the formula This is the final feasible component after mirror reflection and boundary clipping; This approach preserves the momentum information of the search, avoiding the problems of large-scale aggregation of individuals at the boundary and rapid decline in population diversity caused by traditional boundary absorption strategies.

[0029] ② For each parent individual, generate We use a set of trial vectors to select the individual with the best fitness as a candidate; then, we update the next generation of the population through greedy selection. In the formula, This is the optimal trial vector corresponding to the parent generation. The fitness function is such that if the trial vector survives, the parent individual to be replaced is stored in an external archive. (5) Linear population size reduction mechanism The population size is linearly reduced as the iteration progresses, and computational resources are concentrated on high-quality individuals to improve the efficiency of later local refinement. The next generation population size is updated according to the following formula: In the formula, , These are the preset maximum and minimum population sizes, respectively; This indicates rounding to the nearest integer. If the updated population size is smaller than the current population size, the individual with the worst fitness in the current population will be deleted until the target size is reached, and the capacity limit of the external archive will be adjusted accordingly.

[0030] S5. After iterative convergence, the optimal trajectory control point sequence obtained by optimization is used as the control vertex of the cubic B-spline curve to generate a smooth three-dimensional trajectory and send it to the UAV flight control platform. At the same time, online replanning is periodically initiated in combination with the environmental data transmitted back in real time by the airborne sensors.

[0031] Optionally, S5 specifically includes: S5.1 Smoothing of three consecutive B-spline tracks The optimal control point sequence obtained through iterative convergence is used as the control vertices of a cubic B-spline curve to generate a continuous, smooth 3D trajectory. The calculation formula is as follows: In the formula, (T(t)) represents the three-dimensional flight path of the UAV during the actual inspection mission; Let i be the i-th track control point; The basis functions are cubic B-spline functions. The starting point of the corresponding flight path. The endpoint of the corresponding flight path; S5.2 Track Command Generation and Issuance The smoothed continuous trajectory is discretized into a sequence of trajectory points with a preset time interval. Each trajectory point contains three-dimensional coordinates, desired velocity, and desired heading angle information. The trajectory command is then sent to the UAV flight control platform via a wireless communication module. S5.3 Periodic Online Replanning A fixed-cycle online replanning mechanism is adopted to achieve dynamic environment adaptation. The improved multi-strategy chaotic evolutionary optimization algorithm MSCEO is called every 5 seconds to update four types of physical models based on the latest wind field, terrain, and thermal zone data collected by airborne sensors. During online replanning, the flight path of the already flown part is retained unchanged, and only the remaining unflown segment of the flight path is locally optimized. The algorithm has a single run time of less than 50ms and a peak memory usage of less than 50KB, and can run stably on embedded microcontrollers.

[0032] This invention also provides a scenario-driven multi-strategy optimized UAV path planning system. The system adopts a two-level architecture of airborne execution carrier + control and computing core, and is composed of two parts: UAV flight control platform and flight control management device. The UAV flight control platform is the aerial physical execution carrier for the inspection mission, and integrates an environmental perception module, a wireless communication module, a flight control execution module, and a mission sensor module. The environmental perception module is used to collect real-time data on the terrain elevation, UAV real-time flight altitude, airspace wind field vector, and no-fly zone information of the inspection area, providing raw data support for subsequent environmental modeling and constraint verification; the wireless communication module is used to realize bidirectional data transmission between the UAV flight control platform and the flight control management device, ensuring stable interaction between downlink flight path commands and uplink sensor data; the flight control execution module is used to receive and execute flight path control commands issued by the flight control management device, achieving high-precision tracking flight of the planned trajectory through attitude calculation and power output control; the mission sensor module is used to collect optical, visible light, and infrared images of the inspection area during cruise flight. The flight control and management device is the ground or airborne computing and control core of the system, including a data receiving module, a central control module, a three-dimensional environment modeling module, and a path optimization module; The data receiving module receives raw sensor data from the UAV flight control platform and performs standardized preprocessing. The central control module coordinates task scheduling, operational status management, and workflow control, triggering model updates and trajectory replanning based on environmental changes and task requirements. The 3D environment modeling module constructs wind direction and asymmetric vertical energy consumption models, sensor field-of-view coverage models, airspace safety constraint models, and no-fly zone constraint models based on preprocessed multi-source sensor data, forming a complete 3D environmental constraint system. The path optimization module is equipped with an improved multi-strategy chaotic evolutionary optimization algorithm, MSCEO. Based on various physical models output by the 3D environment modeling module and a preset scene-driven weighted linear scalar composite fitness function, it sequentially performs a complete time-series calculation of population initialization, iterative optimization, and convergence termination, ultimately outputting a 3D trajectory smoothed by cubic B-spline processing. This trajectory is then transmitted to the flight control execution module of the UAV flight control platform via a wireless communication module, and online replanning is periodically initiated in conjunction with real-time environmental data transmitted from airborne sensors.

[0033] The scenario-driven multi-strategy optimization UAV path planning system provided in this embodiment of the invention has a functional structure that corresponds to the scenario-driven multi-strategy optimization UAV path planning method provided in this embodiment of the invention, and will not be described again here.

[0034] Figure 4 This is a schematic diagram of the structure of an electronic device 400 provided in an embodiment of the present invention. The electronic device 400 may vary considerably due to different configurations or performance. It may include one or more central processing units (CPUs) 401 and one or more memories 402. The memory 402 stores at least one instruction, which is loaded and executed by the processor 401 to implement the steps of the above-mentioned scenario-driven multi-strategy optimized UAV path planning method.

[0035] In an exemplary embodiment, a computer-readable storage medium is also provided, such as a memory including instructions that can be executed by a processor in a terminal to complete the scenario-driven multi-strategy optimized UAV path planning method described above. For example, the computer-readable storage medium may be a ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, or optical data storage device, etc.

[0036] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by hardware or by a program instructing related hardware. The program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk.

[0037] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A scenario-driven, multi-strategy optimization method for UAV path planning, characterized in that, The method includes: S1. Acquire DEM elevation benchmark data and airborne sensor data; S2. Based on the DEM elevation benchmark data and airborne sensor data, construct the wind direction and asymmetric vertical energy consumption model, sensor field of view coverage model, airspace safety constraint model and no-fly zone constraint model to form a complete three-dimensional environmental constraint system. S3. Construct a scenario-driven weighted linear scalar composite fitness function to uniformly map coverage gap, wind field-oriented energy consumption cost, and spatial path length to a comparable scalar space. At the same time, treat terrain boundary crossing and no-fly zone intrusion as rigid constraints and achieve zero tolerance through high extra points penalty. S4. The improved multi-strategy chaotic evolutionary optimization algorithm MSCEO is used to perform path optimization. MSCEO integrates five mechanisms: adaptive chaotic sampling, linear decay dual-strategy trial vector generation, successful history memory and external archiving, mirror reflection boundary repair and greedy selection, and linear population size reduction. It performs optimization calculations according to the complete time sequence of population initialization, iterative update and convergence termination, and dynamically balances global exploration and local development capabilities. S5. After iterative convergence, the optimal trajectory control point sequence obtained by optimization is used as the control vertex of the cubic B-spline curve to generate a smooth three-dimensional trajectory and send it to the UAV flight control platform. At the same time, online replanning is periodically initiated in combination with the environmental data transmitted back in real time by the airborne sensors.

2. The method according to claim 1, characterized in that, The construction of the wind direction and asymmetric vertical energy consumption model specifically includes: Based on discrete wind field data measured by airborne wind speed and direction sensors, a continuous wind field distribution across the entire domain is generated through spatial interpolation. The cosine similarity between the horizontal direction of the flight path and the wind field direction for each flight segment is calculated. The differential impact of tailwind, headwind, and crosswind on the UAV's propulsion power is quantified, and a wind field correction factor is obtained. This provides a calculation basis for subsequent wind field disturbance correction of horizontal flight energy consumption. Among these factors: For the m-th discrete track segment, its horizontal displacement vector is defined as: The corresponding horizontal segment length is: Let the horizontal wind field vector in the planning time domain be: Cosine similarity is used to describe the azimuth relationship between the flight direction and the wind field direction during the flight path segment: In the formula, To prevent extremely small positive numbers from being divided by zero; The value range is approximately [-1, 1]: a positive value indicates that the flight path is in the same direction as the wind field, corresponding to a tailwind condition; a negative value indicates that the flight path is in the opposite direction to the wind field, corresponding to a headwind condition; a value close to zero indicates that the flight direction is perpendicular to the wind direction, corresponding to a crosswind condition. Wind field correction factor constructed based on cosine similarity. To achieve differentiated calculation of horizontal energy consumption under different wind conditions: In the formula, For wind direction sensitivity coefficient, satisfying The wind field correction factor is used to characterize the impact of wind direction changes on the horizontal propulsion power of UAVs. Under this definition, the wind field correction factor is less than 1 in the tailwind flight segment, which corresponds to a reduction in the horizontal energy consumption value; the wind field correction factor is greater than 1 in the headwind flight segment, which corresponds to an increase in the horizontal energy consumption value; and the correction factor is close to 1 in the crosswind condition, and the energy consumption level is the same as the no-wind baseline. The planned 3D flight path is discretized into M equal-length flight segments. The energy consumption differences between climb, level flight, and descent are differentiated. Combined with a wind field correction factor, the total flight energy consumption is calculated to accurately characterize the asymmetric energy consumption characteristics of vertical flight in mountainous environments. The calculation formula is as follows: In the formula, E is the total flight energy consumption cost after wind field correction; m is the index of the discrete segment of the track; M is the total number of discrete sampling points of the track; The baseline level flight energy consumption coefficient characterizes the energy consumption per unit horizontal distance during level flight under windless conditions. Let m be the horizontal displacement length of the m-th segment of the trajectory; is the wind field correction factor corresponding to the m-th segment of the trajectory; The climbing energy consumption coefficient represents the additional energy consumed per unit vertical climbing height. Let be the vertical climb altitude of the m-th segment of the trajectory, where This represents the vertical displacement of the flight path segment; The energy consumption coefficient is used to characterize the energy consumption per unit vertical descent height. Let m be the vertical descent altitude of the m-th segment of the trajectory, and the model satisfies... The constraint relationship, namely that the climbing energy consumption coefficient is significantly greater than the descending energy consumption coefficient, is consistent with the actual energy consumption law of UAV vertical flight and can accurately reflect the additional energy consumption burden caused by repeated climbing in mountainous areas.

3. The method according to claim 1, characterized in that, The construction of the sensor field-of-view coverage model specifically includes: A mapping relationship is established between the UAV's flight altitude and the sensor's instantaneous ground detection radius to reflect the dynamic impact of flight altitude changes on the inspection coverage. Simultaneously, a hardware limitation on the sensor's maximum effective detection radius is introduced. The calculation formula is as follows: In the formula, Represents the ground projection radius of the sensor at the m-th sampling time; This represents the real-time flight altitude of the UAV relative to the ground at the m-th sampling time; This represents the half-angle of the sensor's field of view, which is determined by the sensor's hardware parameters. This represents the maximum effective sensing radius of the sensor, which is the upper limit of hardware performance.

4. The method according to claim 1, characterized in that, The construction of the airspace safety constraint model specifically includes: Based on DEM elevation benchmark data, the real-time ground clearance of each trajectory sampling point relative to the terrain surface is calculated. A minimum safe clearance height threshold is preset, and the terrain penetration violation depth of the trajectory points is quantified to provide a basis for subsequent safety penalty calculations. The calculation formula is as follows: In the formula, This represents the terrain clearance violation depth at the m-th trajectory sampling point; This is the preset minimum safe ground clearance height for drones; The real-time ground clearance of the UAV relative to the terrain surface at the m-th sampling point; when the trajectory point meets the airspace safety requirements, the violation depth is 0; when the trajectory is below the safe height, the violation depth increases with the degree of terrain penetration, adapting to the undulating characteristics of mountainous terrain, avoiding planning deviations such as compliance in valleys and violations on ridges at the same absolute height, and rigidly ensuring the terrain safety constraints of inspection flights.

5. The method according to claim 1, characterized in that, The construction of the no-fly zone constraint model specifically includes: The spatial boundary of the danger zone is described using a three-dimensional ellipsoid equation. This determines whether a trajectory point intrudes into the no-fly zone and calculates the intrusion depth for subsequent safety penalty calculations. For the j-th no-fly zone, its spatial boundary satisfies the following ellipsoid equation: In the formula, Indicates the center coordinates of the j-th no-fly zone; These represent the semi-axis lengths of the no-fly zone in the three coordinate axes (x, y, z). When the coordinates of the trajectory point are substituted into the above formula and the condition is less than 1, it is determined to be an intrusion into the no-fly zone, and the intrusion depth is the distance value inside the ellipsoid; otherwise, it is determined to be in a safe area.

6. The method according to claim 1, characterized in that, S3 specifically includes: S3.1 Multi-objective dimensionless normalization processing: To address the differences in physical dimensions and numerical magnitudes among three categories of physical quantities—coverage gap, wind field guiding energy consumption cost, and spatial path length—normalization mapping is performed to uniformly transform them into the [0,1] scalar interval, thus eliminating dimensional differences. (1) Normalization of coverage missing value Using the complement of the global inspection coverage rate as an indicator of coverage gap, it directly reflects the proportion of blind spots in the inspection coverage of candidate tracks. The calculation formula is as follows: In the formula, X is the control point decision vector corresponding to the candidate trajectory; The global inspection coverage of the candidate track is defined, with a value range of [0,1]. The normalized coverage gap is [0,1]. The smaller the value, the better the inspection coverage of the candidate track and the lower the proportion of blind spots. (2) Wind farm directional energy consumption normalization The ratio of the total flight energy consumption after wind field correction to the preset baseline energy consumption is used as the normalized energy consumption index to eliminate the interference of the absolute magnitude of energy consumption on the weighting configuration. The calculation formula is as follows: In the formula, E(X) represents the total energy consumption cost of wind field guidance corresponding to the candidate trajectory; The baseline energy consumption reference value is used as the theoretical energy consumption value under the straight-line level flight condition between the start and end points of the flight path, and it is used as the unified benchmark for energy consumption evaluation. This represents the normalized energy consumption cost; the smaller the value, the higher the energy efficiency of the candidate trajectory. (3) Path length normalization The ratio of the total spatial length of the trajectory to the straight-line distance between the origin and destination points is used as the normalized path length index to reflect the degree of detour of the candidate trajectory relative to the theoretical shortest path. The calculation formula is as follows: In the formula, L(X) is the total spatial path length corresponding to the candidate track; The straight-line distance between the start and end points of the track serves as a unified benchmark for evaluating path length. The normalized path length is represented by a smaller value, indicating a more compact candidate path and a lower degree of spatial detour. S3.2 Construction of the external point penalty term for rigid safety constraints: For two types of rigid safety constraints—terrain clearance violations and no-fly zone incursions—an out-of-point penalty mechanism of "baseline penalty + depth weighting" is adopted to construct the total safety penalty term. The magnitude of the penalty term is much larger than the sum of the values ​​of all task objective terms, ensuring that candidate tracks with safety violations are directly eliminated during the evolution process, achieving zero-tolerance screening of safety violations. At the same time, the depth weighting term provides a continuous penalty gradient to guide the search towards the feasible region. (1) Penalties for violations of terrain clearance For each trajectory sampling point, the clearance violation is simultaneously overlaid with baseline penalty and depth penalty, calculated as follows: In the formula, This is an indicator function that takes the value 1 when the condition inside the parentheses is met, and takes the value 0 when the condition is not met. The terrain clearance violation depth for the m-th trajectory sampling point; The baseline penalty value for terrain violations is a very large constant that is much larger than the sum of the mission objectives. This ensures that as long as there is a terrain clearance violation, the fitness of the candidate track will be significantly degraded and it will be directly eliminated in the evolutionary selection. is the terrain violation depth weighting coefficient, used to quantify the penalty gradient of violation depth. The more severe the violation, the higher the penalty value, guiding the search to converge in the direction that meets the clearance requirements; M is the total number of trajectory sampling points; (2) Penalties for intruding into the no-fly zone For each trajectory sampling point, the no-fly zone intrusion behavior is simultaneously superimposed with baseline penalty and intrusion depth penalty, calculated as follows: In the formula, J represents the total number of no-fly zones for thermal threats; The algebraic distance between the m-th sampling point and the j-th ellipsoidal no-fly zone; To normalize the depth of intrusion, a larger value indicates a deeper intrusion into the no-fly zone; The penalty value for intruding into the baseline of the no-fly zone is a very large constant that is much larger than the sum of the mission objectives, ensuring that as long as there is a no-fly zone intrusion, the candidate track will be marked as infeasible; These are the intrusion depth weighting coefficients, used to quantify the penalty gradient of the intrusion depth; (3) Total safety penalty items The total safety penalty is the sum of the penalty for terrain clearance violations and the penalty for intrusion into thermal threat no-fly zones, calculated using the following formula: When a candidate track fully satisfies all safety constraints, the total safety penalty term is 0; when any safety violation exists, the penalty term value increases sharply, directly determining the degree of fitness degradation of the candidate track and rigidly ensuring the highest priority of safety constraints. S3.3 Construction of Weighted Linear Scalar Composite Fitness Function A scenario-specific weight vector is set, and the three types of normalized task objectives are weighted and aggregated. Then, a rigid safety penalty term is added to obtain the final single scalar fitness function, calculated as follows: In the formula, , , These are task weight coefficients for coverage gap, energy consumption cost, and path length, respectively, satisfying... The weighting coefficients are pre-configured based on the inspection task scenario: for daily inspection scenarios covering the entire area, the coverage weight is increased to ensure inspection coverage; for fixed-point fire verification scenarios, the energy consumption weight is increased to extend the endurance time; for emergency rapid response scenarios, the path length weight is increased to shorten the arrival time. The optimization objective of path optimization is to minimize the fitness function value J(X).

7. The method according to claim 1, characterized in that, The specific implementation of the five mechanisms of MSCEO is as follows: (1) Adaptive chaotic sampling mechanism In each iteration, population diversity is monitored in real time, and the number of chaotic samples is dynamically adjusted to balance global exploration capability and computational overhead. The specific steps are as follows: ① Calculate the mean vector of the population in generation t. The average Euclidean distance from individuals within a population to the mean vector is used as an indicator of diversity. In the formula, For the population size of generation t, Let be the decision vector of the i-th individual in the t-th generation; The diversity ratio is obtained by normalizing the diversity index: In the formula, , These represent the upper and lower boundaries of the decision space, respectively. To prevent extremely small positive numbers from being divided by zero; ② Update the number of chaotic samples based on the diversity ratio and the stagnation state of the population: when Below the preset threshold or global optimal fitness continuous When the situation is not improved, increasing the number of chaotic samples enhances global exploration and helps the population escape local optima; when Higher than the preset threshold Furthermore, as the global optimal fitness continues to improve, the number of chaotic samples is reduced to lower computational overhead, focusing on local development; the update rule is: In the formula, As the expansion factor, It is a contraction factor; , , respectively, represent the lower and upper limits of the number of chaotic samples; s(t) is the stagnation algebra counter for the t-th generation; ③ Generate chaotic samples through a two-dimensional exponential discrete chaotic mapping: In the formula, k is the chaotic mapping control coefficient. This represents element-wise multiplication, which normalizes the generated chaotic variables and maps them to the decision space to obtain chaotic samples. , used to construct the probing direction of a chaotic exploration strategy; (2) Linear decay dual-strategy trial vector generation mechanism Two strategies, chaotic exploration and archive-assisted elite refinement, are used to generate trial vectors. The selection probability of the chaotic strategy decreases linearly with iteration, achieving a smooth transition from wide-area exploration in the early stage to local refinement in the later stage. The specific steps are as follows: ① The probability of choosing a chaotic strategy decreases linearly with the number of iterations: In the formula, Choose probabilities for the initial chaotic policy. To determine the maximum number of iterations, for each parent individual, with probability... Choose a chaotic exploration strategy, based on probability. In the early stages of iteration, the probability of using the elite strategy is high, focusing on global search; in the later stages of iteration, the proportion of the elite strategy increases, focusing on local convergence. ② The elite refinement strategy employs an archive-assisted mutation method from the current to the optimal population, where the proportion of elite individuals decreases linearly with iteration: In the formula, , These represent the initial and final elite ratios, respectively, and the mutation formula is: In the formula, Let be the scaling factor for the i-th individual; From the top of fitness ranking Randomly selected from individuals; Randomly select from the current population; Randomly selected from the union of the current population and the external archive; and the three individual indices are all different and none of them are equal to the parent individual index i; ③ The chaotic exploration strategy generates candidate vectors based on chaotic samples: In the formula, The chaos step size coefficient, Chaotic samples generated by an adaptive chaotic sampling mechanism; ④ Perform a binomial crossover between the mutated candidate vector and the parent individual to generate a trial vector: In the formula, Let be the crossover rate of the i-th individual; A uniform random number within the interval ([0,1]); The dimension index is randomly selected to ensure that at least one dimension of mutation information enters the trial vector, thus avoiding complete replication of the parent individual; (3) Successful historical memory and external archiving mechanism The process involves recording parameter combinations that generate high-quality solutions during evolution, adaptively adjusting the distribution of scaling factors and crossover rates, and reusing replaced parent individuals to maintain population diversity. The specific steps are as follows: ① Scaling factor for each individual and cross rate Samples were taken from the successful historical memory pool: In the formula, M_F and M_{CR} are the scaling factor and the successful history memory pool of the crossover rate, respectively, and both have a length of H; Let H be a memory pool index randomly selected from ([1,H]). If the sampled index is... Then resample, if Then truncate to 1; Cut off to the interval ([0,1]); ② If the fitness of the trial vector is better than that of its corresponding parent, the evolution is considered successful, and the scaling factor, crossover rate, and fitness improvement amount corresponding to the trial vector are recorded; after each generation of evolution, the parameters at the corresponding position in the memory pool are updated according to the fitness improvement amount. In the formula, S is the set of all successfully evolved individuals in the current generation; Let be the fitness improvement amount for the i-th individual; For extremely small positive numbers; memory pool index k is... The parameters are incremented sequentially to achieve rolling updates. ③ External archives are used to store parent individuals that are replaced during evolution; the maximum archive capacity is [missing information]. When the number of individuals in the archive exceeds the capacity limit, the excess individuals are randomly deleted. The individuals in the archive only participate in the mutation operation of the elite refinement strategy and do not participate in the population fitness ranking and selection. By introducing differential information of historical inferior solutions, the search direction is enriched and the population diversity is maintained. (4) Mirror reflection boundary repair and greedy selection mechanism Mirroring and reflecting outbounded trial vectors helps prevent diversity loss caused by boundary clustering. Greedy selection is used to retain high-quality solutions, driving the population to evolve towards better fitness. The specific steps are as follows: ① Perform mirror reflection repair on the dimensional components of the trial vector that exceed the decision boundary: In the formula, , Let be the lower and upper boundaries of the d-th dimension decision variable, respectively. Let d be the d-th component of the i-th trial vector in the t-th iteration. These are the corresponding dimensional components after a single mirror reflection process; If the reflected component still exceeds the boundary, then clip to the nearest boundary value: In the formula This is the final feasible component after mirror reflection and boundary clipping; ② For each parent individual, generate We use a set of trial vectors to select the individual with the best fitness as a candidate; then, we update the next generation of the population through greedy selection. In the formula, This is the optimal trial vector corresponding to the parent generation. The fitness function is such that if the trial vector survives, the parent individual to be replaced is stored in an external archive. (4) Linear population size reduction mechanism The population size is linearly reduced as the iteration progresses, and computational resources are concentrated on high-quality individuals to improve the efficiency of later local refinement. The next generation population size is updated according to the following formula: In the formula, , These are the preset maximum and minimum population sizes, respectively; This indicates rounding to the nearest integer. If the updated population size is smaller than the current population size, the individual with the worst fitness in the current population will be deleted until the target size is reached, and the capacity limit of the external archive will be adjusted accordingly.

8. A scenario-driven multi-strategy optimized UAV path planning system, characterized in that, The system adopts a two-level architecture of airborne execution carrier + control and computing core, and is composed of two parts: UAV flight control platform and flight control management device. The UAV flight control platform is the aerial physical execution carrier for the inspection mission, and integrates an environmental perception module, a wireless communication module, a flight control execution module, and a mission sensor module. The environmental perception module is used to collect real-time data on the terrain elevation, UAV real-time flight altitude, airspace wind field vector, and no-fly zone information of the inspection area, providing raw data support for subsequent environmental modeling and constraint verification; the wireless communication module is used to realize bidirectional data transmission between the UAV flight control platform and the flight control management device, ensuring stable interaction between downlink flight path commands and uplink sensor data; the flight control execution module is used to receive and execute flight path control commands issued by the flight control management device, achieving high-precision tracking flight of the planned trajectory through attitude calculation and power output control; the mission sensor module is used to collect optical, visible light, and infrared images of the inspection area during cruise flight. The flight control and management device is the ground or airborne computing and control core of the system, including a data receiving module, a central control module, a three-dimensional environment modeling module, and a path optimization module; The data receiving module receives raw sensor data from the UAV flight control platform and performs standardized preprocessing. The central control module coordinates task scheduling, operational status management, and workflow control, triggering model updates and trajectory replanning based on environmental changes and task requirements. The 3D environment modeling module constructs wind direction and asymmetric vertical energy consumption models, sensor field-of-view coverage models, airspace safety constraint models, and no-fly zone constraint models based on preprocessed multi-source sensor data, forming a complete 3D environmental constraint system. The path optimization module is equipped with an improved multi-strategy chaotic evolutionary optimization algorithm, MSCEO. Based on various physical models output by the 3D environment modeling module and a preset scene-driven weighted linear scalar composite fitness function, it sequentially performs a complete time-series calculation of population initialization, iterative optimization, and convergence termination, ultimately outputting a 3D trajectory smoothed by cubic B-spline processing. This trajectory is then transmitted to the flight control execution module of the UAV flight control platform via a wireless communication module, and online replanning is periodically initiated in conjunction with real-time environmental data transmitted from airborne sensors.

9. An electronic device comprising a processor and a memory, wherein the memory stores at least one instruction, characterized in that, The at least one instruction is loaded and executed by the processor to implement the scenario-driven multi-strategy optimized UAV path planning method as described in any one of claims 1-7.

10. A computer-readable storage medium storing at least one instruction, characterized in that, The at least one instruction is loaded and executed by the processor to implement the scenario-driven multi-strategy optimized UAV path planning method as described in any one of claims 1-7.

Images