A method and system for adaptive path planning of unmanned aerial vehicle based on double-layer search

Abstract

The present invention discloses a method and system for adaptive path planning of unmanned aerial vehicles (UAVs) based on double-layer search, and relates to the field of UAV inspection technology. Currently, the inspection safety and autonomy of UAVs on substations are poor, and the efficiency is low. The present invention includes: using sensors carried by UAVs to collect the location, type and safety distance data of electrical equipment in substations in real time; dynamically calculating the three-dimensional safety distance between the UAV and the equipment based on the type and location of the equipment, combined with the real-time status information and environmental data of the UAV; using a double-layer path planning algorithm based on edge computing nodes to generate the optimal safety path; sending the planned path to the UAV, continuously monitoring the flight status, and triggering dynamic path adjustment or hovering instructions if deviation from the path or sudden obstacles are detected; evaluating flight efficiency and safety indicators after the task is completed, optimizing algorithm parameters and updating system settings. This technical solution achieves precise obstacle avoidance, efficient planning, real-time response and adaptive evolution.

Description

Technical Field

[0001] The present invention relates to the field of unmanned aerial vehicle (UAV) inspection technology, and in particular to a UAV adaptive path planning method and system based on double-layer search. Background Art

[0002] With the rapid development of drone technology, its application value in the inspection of key facilities in the power system has become increasingly prominent, especially in scenarios with high precision and high safety requirements such as substations. Drones have gradually become the core means of equipment status monitoring due to their flexibility and efficiency.

[0003] However, traditional drone path planning technology faces the following prominent problems in substation inspections:

[0004] Existing technologies often rely on preset paths or single algorithms for planning, and are unable to dynamically adjust paths based on real-time environmental changes (such as sudden strong winds or rainfall) or equipment abnormalities. For example, when a drone encounters an unexpected obstacle or deteriorating weather conditions, traditional methods require manual intervention or a return to the preset path, significantly reducing inspection efficiency and even causing flight accidents.

[0005] Substations are densely populated with high-voltage electrical equipment. Existing path planning technologies typically calculate safe distances based on two-dimensional planes or simplified models, failing to consider the impact of equipment 3D shapes (such as cylindrical transformers and rectangular switchgear) on flight paths. While some technologies incorporate safe distance constraints, they lack dynamic adjustment capabilities, potentially causing drones to stray into high-risk areas depending on flight direction or equipment type.

[0006] Traditional global search methods like particle swarm optimization (PSO) are prone to falling into local optima and have slow convergence. Local optimization algorithms like tabu search rely on the quality of the initial solution and struggle to cope with large-scale, complex environments. Real-time data processing delays make it difficult to meet the timeliness requirements of substation inspections.

[0007] Existing technologies for obstacle identification often rely on static maps or low-precision sensors (such as single GPS), making it difficult to construct a high-precision 3D model of the substation environment in real time. Path planning methods based on grid maps result in high computational complexity and cannot dynamically update obstacle information.

[0008] To address the above issues, there is an urgent need for a drone path planning method and system that can dynamically adapt to the complex environment of substations, accurately calculate three-dimensional safety distances, and have an efficient algorithm architecture to improve inspection safety, efficiency, and autonomy. Summary of the Invention

[0009] The technical problem to be solved and the technical task to be addressed by this invention are to improve and enhance existing technical solutions by providing a method and system for adaptive path planning for unmanned aerial vehicles (UAVs) based on a dual-layer search, with the goal of improving inspection safety, efficiency, and autonomy. To this end, this invention adopts the following technical solutions.

[0010] A method for adaptive path planning of a UAV based on a double-layer search comprises the following steps:

[0011] Step 1: Use the drone’s onboard sensors to collect real-time data on the location, type, and safety distance of electrical equipment in the substation, and clean and organize the data;

[0012] Step 2: Dynamically calculate the three-dimensional safe distance between the drone and the device based on the device type and location, combined with the drone's real-time status information and environmental data. The drone's real-time status information includes latitude and longitude, altitude, flight time, and remaining battery power; environmental data includes temperature, humidity, and wind speed.

[0013] Step 3: Based on the edge computing node, a two-layer path planning algorithm is used to generate the optimal safe path, where:

[0014] The upper layer uses a particle swarm algorithm to search the global solution space. Latin hypercube sampling is used to initialize the particle swarm, and virtual obstacle particles are embedded in the particle swarm to simulate the location of real obstacles. The quality of particles is evaluated by a composite fitness function that integrates path mileage, energy consumption cost, and safety cost.

[0015] The lower layer performs local optimization on the solution space output by the upper layer based on the tabu search algorithm, establishes a tabu table with a short-term memory mechanism, and breaks through the local optimum through a neighborhood search strategy;

[0016] Step 4: Convert the planned path into flight control commands and send them to the drone. Continuously monitor the flight status. If deviation from the path or sudden obstacles are detected, dynamic path adjustment or hovering commands are triggered.

[0017] Step 5: After the mission is completed, evaluate flight efficiency and safety indicators, optimize algorithm parameters and update system settings.

[0018] This technical solution utilizes dynamic three-dimensional safety modeling, a layered collaborative optimization algorithm, real-time edge computing, and a closed-loop feedback mechanism to achieve precise obstacle avoidance, efficient planning, real-time response, and adaptive evolution, without relying on weight parameters or other subsequent claim details. Specifically, this technical solution dynamically calculates the three-dimensional safety distance between the drone and electrical equipment by combining equipment type, location, and real-time environmental data (temperature, humidity, and wind speed), overcoming the limitations of traditional two-dimensional planar models. It accurately calculates the Euclidean distance between the nearest points on their surfaces, ensuring the drone maintains safe flight in complex three-dimensional configurations. Latin hypercube sampling is used to initialize the particle swarm, ensuring uniform coverage of the global solution space and avoiding search blind spots caused by traditional random initialization. By embedding virtual obstacle particles (e.g., simulating the locations of high-voltage equipment), high-risk areas are avoided in advance, enhancing path prediction capabilities. A short-term memory mechanism (tabu table) prevents repeated visits to local suboptimal solutions. Neighborhood search strategies (such as path node swapping and insertion) are combined to fine-tune the path, effectively avoiding local optimality traps and improving solution quality. This two-layer collaborative mechanism achieves a "global rapid exploration → local refined optimization" division of labor, balancing computational efficiency and path safety. A composite fitness function that integrates path mileage, energy consumption cost, and safety cost comprehensively evaluates particle quality from multiple dimensions. For example, while shortening path mileage, it automatically balances energy consumption with safety distance constraints, avoiding the one-sidedness caused by single-objective optimization. The path planning module is deployed at the edge node and directly processes real-time sensor data (such as drone status and environmental information), reducing cloud transmission latency and response time. When path deviation or sudden obstacles are detected, it dynamically triggers path adjustments or hovering commands, avoiding the response lag caused by global replanning in traditional methods. After the mission is completed, algorithm parameters (such as particle swarm population size and taboo table length) can be optimized based on flight efficiency and safety indicators (such as path deviation rate and number of emergency obstacle avoidances), and system settings can be updated to achieve adaptive iterative improvement. For example, based on historical data of a specific substation layout, the initial solution generation strategy can be automatically optimized to reduce the convergence time of subsequent tasks.

[0019] As a preferred technical means: the sensor in step 1 includes:

[0020] RTK positioning system, used to obtain high-precision latitude, longitude and altitude of the UAV;

[0021] Inertial measurement unit, used to measure the attitude angle and linear acceleration of the drone. The attitude angle includes pitch angle, yaw angle and roll angle;

[0022] Vision cameras, used to build 3D contour models of substation equipment;

[0023] Meteorological sensors are used to collect wind speed, temperature and humidity data in real time.

[0024] This technical solution utilizes real-time kinematic positioning (RTK) technology to provide centimeter-level accuracy in latitude, longitude, and altitude, surpassing traditional GPS. This ensures precise positioning of drones within densely populated substation equipment, avoiding collision risks caused by positioning drift. Furthermore, RTK's robustness against multipath ensures stable positioning data in environments with strong electromagnetic interference (such as near high-voltage equipment), preventing drones from straying into high-risk areas due to signal distortion. The inertial measurement unit (IMU) measures the drone's attitude angles (pitch, yaw, and roll) and linear acceleration in real time, providing high-precision motion state feedback to the flight control system. In the event of a temporary GPS signal failure (e.g., equipment obstruction), the IMU's inertial navigation capabilities maintain stable flight for at least 30 seconds, significantly enhancing system robustness. High-resolution images of substation equipment are captured by binocular vision cameras and combined with SLAM (Simultaneous Localization and Mapping) technology to construct a 3D contour model of the equipment in real time. The vision cameras can identify subtle surface structures (such as cracked insulators and loose bolts) and dynamically update the locations of obstacles (such as temporarily stacked tools), addressing the static limitations of pre-set maps. Meteorological sensors can monitor wind speed, temperature, and humidity in real time, providing early warning of strong winds, high temperatures, or condensation risks. The drone's flight altitude and speed thresholds are dynamically adjusted based on wind speed to prevent attitude instability. This technical solution ensures flight safety through data fusion. RTK positioning data and IMU attitude data complement each other, enabling accurate positioning around the clock in complex electromagnetic environments. Visual modeling combined with meteorological monitoring helps proactively avoid dynamic obstacles and sudden environmental changes. This technical solution achieves full-dimensional environmental perception, covering four dimensions of data: spatial positioning (RTK), motion status (IMU), device configuration (vision), and meteorological conditions (wind speed / temperature and humidity), providing highly reliable input for subsequent path planning.

[0025] As a preferred technical means: the calculation method of the three-dimensional safety distance in step 2 is:

[0026] Model the electrical equipment as a regular geometric body and calculate the Euclidean distance between the drone and the closest point on the equipment surface;

[0027] Combined with the flight direction of the drone, the safety distance threshold is dynamically adjusted. The formula is:

[0028]

[0029] Where: Respectively represent drones Longitude, latitude and altitude of the time period; The coordinates of the closest point between the electrical equipment and the drone; The minimum safety radius corresponding to the device type.

[0030] This solution utilizes 3D geometric modeling, dynamic directional adaptive thresholds, and device-type-specific radiuses to achieve precise obstacle avoidance, dynamic adaptability, and risk-tiered management. Specifically, substation electrical equipment (such as transformers and circuit breakers) is modeled as regular geometric shapes (cylinders, cuboids, etc.), accurately reflecting the equipment's 3D physical contours. This avoids the simplification errors of traditional point models or 2D plane models and reduces the risk of scratches. Dynamically adjusting the safety distance threshold based on the drone's real-time flight direction addresses the limited adaptability of traditional fixed thresholds in dynamic scenarios and prevents overly conservative or risky path planning. The 3D Euclidean distance formula is used to calculate the spatial distance between the drone and the nearest point on the equipment surface, encompassing longitude, latitude, and altitude data. This addresses the problem of missed detections often associated with traditional 2D distance calculations in the 3D substation environment (e.g., the risk of a drone flying over the top of equipment). The formula also boasts high computational efficiency and meets real-time requirements. Dynamically setting the minimum safety radius based on equipment type matches the electromagnetic field strength and safety requirements of equipment at different voltage levels, avoiding inspection blind spots and wasted resources.

[0031] As a preferred technical means: In step 2, a linear regression model is used to predict the correlation between environmental factors and safe flight of drones, and the correlation is used as a weight coefficient to dynamically adjust the safety cost.

[0032] This technical solution uses a linear regression model to analyze the mathematical correlation between environmental factors (wind speed, temperature, and humidity) and drone flight safety in real time. This correlation is quantified into weight coefficients, dynamically adjusting the safety cost calculation. For example, when wind speed increases from 5m / s to 15m / s, the safety cost weight automatically increases by 40%, forcing the path planning algorithm to generate routes away from high-risk areas, avoiding the rigidity of traditional fixed-weight strategies. Dynamically assigning safety cost weights based on environmental risk levels avoids overly conservative path planning, reduces energy consumption and inspection time, and ensures that risks are manageable.

[0033] As a preferred technical means: the fitness function of the particle swarm algorithm in step 3 is defined as:

[0034] ;

[0035] Where: is the path mileage, Energy consumption costs and For security costs; 、 、 are the corresponding weight ratios, .

[0036] This solution achieves flexible adaptation, efficient computation, and manageable risk through multi-objective linear fusion and normalized weight constraints. Specifically, the fitness function integrates three objectives: path length, energy cost, and safety cost. Dynamically assigning priorities through weight coefficients avoids the biased optimization of single objectives. For example, different weights can be set for densely populated substation areas (high-risk scenarios) and open areas (low-risk scenarios). This weight combination flexibly adapts to different mission requirements, achieving a global optimal balance between safety, efficiency, and energy consumption. Enforcing the sum of weight coefficients to 1 ensures uniformity across objectives and prevents any one objective from dominating the optimization process due to excessively large or small absolute weights. The safety cost term is directly related to the distance between the drone and obstacles; closer distances increase the cost, forcing the particle swarm algorithm to actively avoid high-risk areas. A mathematical penalty mechanism transforms safety constraints into optimization objectives, avoiding the fragmentation of the solution space caused by traditional hard constraints (such as distance thresholds).

[0037] As a preferred technical means: the specific implementation of the tabu search algorithm in step 3 includes:

[0038] The path output by the upper particle swarm is used as the initial solution;

[0039] Set the length of the taboo table and the maximum number of iterations to prohibit repeated access to recent solutions;

[0040] Generate new solutions through neighborhood operations, and select the solution with the best fitness and not tabooed to update the global optimal path.

[0041] This solution achieves efficient convergence, improved path quality, and real-time performance through an initial solution, a tabu table anti-loop mechanism, and dynamic expansion of neighborhood operations. Specifically, the path output by the upper-layer particle swarm algorithm is used as the initial solution, leveraging the results of the global search and avoiding the inefficient exploration caused by the tabu search starting with a random solution. The particle swarm algorithm has already screened out optimal paths globally, and the tabu search performs local optimization based on this, improving convergence speed and reducing the number of iterations. By setting the tabu table length and maximum number of iterations, the algorithm is prohibited from repeatedly revisiting recent solutions, effectively avoiding local optima. Compared to traditional tabu search, this approach significantly improves the avoidance rate of local optima and path quality in substation scenarios. Neighborhood operations (such as node swapping, insertion, or deletion) generate diverse new solutions, combined with a fitness function to select optimal solutions that are not tabooed, ensuring that the search direction approaches the global optimum. Neighborhood operations cover the majority of potential optimal paths, shortening path length and reducing energy costs with the same number of iterations. The combination of a hard constraint on the maximum number of iterations and a flexible tabu table strategy achieves efficient optimization within limited computing resources. On the edge computing node, the time required for a single taboo search is shortened, meeting the real-time requirements of substation inspections.

[0042] As a preferred technical means: the dynamic path adjustment in step 4 includes:

[0043] If the UAV deviates from the path by more than a threshold, the two-layer search algorithm is re-invoked to generate a detour path;

[0044] If a sudden obstacle is detected, the device model is updated based on the real-time data from the visual camera, and a local taboo search is triggered for rapid obstacle avoidance.

[0045] This technical solution achieves precise obstacle avoidance, efficient response, and resource optimization through global and local collaborative optimization, visual real-time modeling, and a hierarchical response mechanism. Specifically, when the drone deviates from the path by more than a threshold, the two-layer search algorithm (upper-layer particle swarm + lower-layer taboo search) is re-called to generate a detour path, ensuring that the new path comprehensively optimizes path mileage, energy consumption, and safety costs on a global scale, avoiding suboptimal solutions caused by local adjustments. The device model is updated based on real-time data from the visual camera (such as the position and shape of dynamic obstacles), triggering a local taboo search to quickly generate an obstacle avoidance path. The local taboo search only performs neighborhood operations (such as node insertion and detours) on the affected path segments, reducing computational time and improving the efficiency of obstacle avoidance path generation. The visual camera captures the three-dimensional contours of the obstacle in real time. After updating the device model, the error in safety distance calculation is improved, the risk of misjudgment is avoided, and dynamic recognition of temporary obstacles (such as maintenance tools and vehicles) is supported. The hierarchical response mechanism balances efficiency and safety. Minor deviations can only fine-tune the flight attitude without triggering re-planning; severe deviations (such as those caused by strong winds) can be treated with global re-planning; and sudden obstacles can be quickly avoided through local taboo search. The hierarchical strategy reduces computing resource consumption while ensuring priority response in high-risk scenarios.

[0046] As a preferred technical means: the optimization algorithm parameters in step 5 include:

[0047] Adjust the inertia weight of the particle swarm algorithm and the length of the tabu table of the tabu search based on historical flight data;

[0048] Update the safety distance weight coefficient through the linear regression model , to reflect the risk level of different devices.

[0049] This technical solution dynamically adjusts the inertia weight of the particle swarm algorithm based on historical flight data (such as convergence speed records of different substation layouts) to achieve an intelligent balance of "global exploration → local convergence". In complex equipment layout scenarios (such as multi-transformer areas), the inertia weight is increased (such as from 0.4 to 0.8) to enhance global search capabilities and avoid premature convergence; in simple open areas, the inertia weight is reduced (such as 0.2) to speed up convergence and reduce path planning time. Based on historical obstacle avoidance success rate data, the length of the taboo table is dynamically set to avoid excessive restriction or redundancy in the search space. In high-risk areas (such as areas with dense high-voltage equipment), the taboo table is extended to prevent repeated visits to suboptimal solutions; in low-risk areas, it is shortened to 5 times to accelerate the generation of neighborhood solutions and reduce local optimization time; after the taboo table is dynamically adjusted, the average path length of the local search is shortened and the obstacle avoidance response speed is improved. The quantitative relationship between environmental factors (wind speed, humidity) and safety events (such as the number of emergency hovering) in historical flight data is analyzed through linear regression, and the safety distance weight coefficient is dynamically updated. This enables risk-driven path planning, reducing the probability of straying into high-risk areas and improving inspection efficiency. After each mission is completed, flight data (such as path deviation rate and energy consumption) is automatically fed back to the parameter optimization module, forming a closed loop of "planning → execution → learning → optimization." This gradually adapts to the specific needs of substations, reducing the average mission time in the same scenario.

[0050] Another technical solution of the present invention is: a UAV adaptive path planning system based on double-layer search, the system comprising:

[0051] The data acquisition module is used to collect the location, type and safety distance data of electrical equipment in the substation in real time through the sensors carried by the drone, and to clean and organize the data;

[0052] A safety calculation module is used to dynamically calculate the three-dimensional safety distance between the drone and the device based on the device type and location, combined with the drone's real-time status information and environmental data; the drone's real-time status information includes latitude and longitude, altitude, flight time, and remaining battery power; and the environmental data includes temperature, humidity, and wind speed;

[0053] The path planning module, deployed on edge computing nodes, uses a two-layer path planning algorithm to generate the optimal and safe path, including:

[0054] The upper-level planning unit uses a particle swarm algorithm to search the global solution space. When initializing the particle swarm, a uniformly distributed initial solution is generated through Latin hypercube sampling, and virtual obstacle particles used to simulate the positions of real obstacles are embedded in the particle swarm. The fitness function of the particle swarm algorithm integrates path mileage, energy consumption cost, and safety cost, and its formula is: Where: is the path mileage, Energy consumption cost and For security costs; 、 、 are the corresponding weight ratios, ;

[0055] The lower-level optimization unit performs local optimization on the solution space output by the upper-level layer based on the tabu search algorithm, breaking through the local optimum by establishing a tabu table with a short-term memory mechanism and a neighborhood search strategy;

[0056] The flight control module converts the planned path into flight control commands and sends them to the drone. It also continuously monitors the flight status and triggers dynamic path adjustments or hovering commands if it detects deviation from the path or sudden obstacles.

[0057] The feedback optimization module is used to evaluate flight efficiency and safety indicators after the mission is completed, optimize algorithm parameters and update system settings.

[0058] This technical solution's data acquisition module utilizes multiple sensors onboard drones to collect rich data in real time and then cleans and organizes it, providing an accurate and reliable data foundation for subsequent steps and ensuring the system's precise perception of the substation environment and drone status. The safety calculation module dynamically calculates three-dimensional safety distances based on device type, location, drone real-time status, and environmental data. Taking multiple factors into account, it more accurately measures the safety status between drones and electrical equipment, effectively mitigating safety risks. The upper-level planning unit of the path planning module utilizes a particle swarm algorithm, initializing the particle swarm through Latin hypercube sampling to ensure a uniform distribution of initial solutions, improving search efficiency and quality. Virtual obstacle particles are embedded to simulate real-world scenarios, enhancing the algorithm's adaptability to complex environments. A multi-cost fitness function is integrated to comprehensively consider path mileage, energy consumption, and safety factors, ensuring that the planned path is more aligned with actual needs. The lower-level optimization unit, based on a tabu search algorithm, utilizes a tabu table with a short-term memory mechanism and a neighborhood search strategy to avoid local optima. It then refines the upper-level solution space for local optimization, further improving path quality and safety. The flight control module converts the planned path into commands to control the drone's flight and continuously monitors flight status. If the drone strays from the path or encounters an unexpected obstacle, it can promptly trigger dynamic path adjustments or hover commands to ensure safe flight and smooth mission execution. After the mission is completed, the feedback optimization module evaluates flight efficiency and safety indicators, optimizing algorithm parameters and updating system settings accordingly, continuously improving system performance and better adapting to different operational scenarios and changing requirements.

[0059] As a preferred technical means: the data acquisition module includes the following sensors: an RTK positioning system for obtaining high-precision latitude, longitude and altitude of the UAV; an inertial measurement unit for measuring the attitude angle and linear acceleration of the UAV, wherein the attitude angle includes pitch angle, yaw angle and roll angle; a visual camera for constructing a three-dimensional contour model of the substation equipment; and a meteorological sensor for collecting wind speed, temperature and humidity data in real time.

[0060] The safety calculation module models the electrical equipment as a regular geometric body and calculates the Euclidean distance between the UAV and the nearest point on the equipment surface. If the calculated distance is greater than or equal to the minimum safety radius corresponding to the equipment type, it is considered safe. The safety calculation module includes a correlation prediction unit, which is used to predict the correlation between environmental factors and the safe flight of the UAV through a linear regression model and use the correlation as a weight coefficient to dynamically adjust the safety cost.

[0061] The specific implementation of the taboo search algorithm in the lower optimization unit includes: taking the path output by the upper particle swarm as the initial solution; setting the length of the taboo table and the maximum number of iterations to prohibit repeated access to recent solutions; generating new solutions through neighborhood operations, and selecting the solution with the best fitness and not tabooed to update the global optimal path;

[0062] The dynamic path adjustment of the flight control module includes: if the drone deviates from the path by more than a preset threshold, re-calling the path planning module to generate a detour path; if a sudden obstacle is detected, updating the device model based on the real-time data of the visual camera, and triggering the lower-level optimization unit to perform local taboo search for rapid obstacle avoidance.

[0063] The data acquisition module integrates multiple sensors. The RTK positioning system acquires high-precision position data, the inertial measurement unit accurately measures attitude angles and linear acceleration, the visual camera constructs a 3D contour model, and the meteorological sensor collects real-time weather data. This comprehensive and highly accurate information acquisition provides a reliable basis for system decision-making and ensures the drone's accurate perception of its environment and its own status. The safety calculation module models electrical equipment as regular geometric bodies and uses Euclidean distance to calculate the distance to the nearest point on the equipment surface. This method is compared with the minimum safety radius to determine safety. This intuitive and easy-to-implement method can quickly and effectively assess flight safety distances. The relevance prediction unit uses a linear regression model to predict the relevance of environmental factors to safe flight. This factor is used as a weighting factor to dynamically adjust the safety cost, making the safety assessment more accurate, enhancing the system's adaptability to complex environmental changes, and ensuring flight safety. The lower-level optimization unit's tabu search algorithm uses the particle swarm output path as the initial solution. By setting the tabu table length and maximum number of iterations to avoid revisiting recent solutions, it generates new solutions using neighborhood operations and selects the optimal solution to update the global path. This effectively avoids local optima, improves the quality and efficiency of path planning, and ensures the drone obtains a more optimal flight path. If the drone deviates from its path by more than a preset threshold, the flight control module calls the path planning module to generate a detour. This ensures the drone can promptly correct its flight trajectory and complete its mission as planned, enhancing flight stability and reliability. When an unexpected obstacle is detected, the device model is updated based on real-time data from the visual camera, triggering the lower-level optimization unit to perform a local taboo search for rapid obstacle avoidance. This allows the drone to quickly respond to emergencies, effectively avoid obstacles, ensure flight safety, and enhance the system's emergency response capabilities.

[0064] Beneficial effects:

[0065] 1. Calculate the safe flight distance required between the drone and the equipment, comprehensively assessing the safe flight relevance of various factors such as drone status information, environmental data, and safe distances from electrical equipment. The drone adaptively adjusts its flight path based on real-time data, enhancing the system's flexibility and adaptability.

[0066] 2. Design a two-layer path search scheme to quickly find the optimal solution space in the upper-layer particle swarm algorithm. Then, based on the lower-layer local search algorithm solution space, further search for the optimal path. This enhances the interpretability of the algorithm, reduces the time cost of the algorithm search, and further improves the quality of the solution.

[0067] 3. The planning model is adjusted and optimized based on actual flight results, enabling the self-learning and evolutionary capabilities of the planning path model. Simultaneously, users can monitor the drone's flight status in real time, adaptively adjusting the planned path based on inspection needs, thereby improving user satisfaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 It is a flow chart of the present invention.

[0069] Figure 2 It is a flow chart of the safety path generation of the present invention. DETAILED DESCRIPTION

[0070] The technical solution of the present invention is further described in detail below with reference to the accompanying drawings.

[0071] Example 1:

[0072] like Figure 1 As shown, a UAV adaptive path planning method based on double-layer search includes the following steps:

[0073] S1: The drone-mounted sensors collect real-time data on the location, type, and safety distance of electrical equipment in the substation, and clean and organize the data;

[0074] S2: Dynamically calculates the three-dimensional safe distance between the drone and the device based on the device type and location, combined with the drone's real-time status information and environmental data. The drone's real-time status information includes latitude and longitude, altitude, flight time, and remaining battery power; environmental data includes temperature, humidity, and wind speed.

[0075] S3: Based on edge computing nodes, a two-layer path planning algorithm is used to generate the optimal safe path, where:

[0076] The upper layer uses a particle swarm algorithm to search the global solution space. Latin hypercube sampling is used to initialize the particle swarm, and virtual obstacle particles are embedded in the particle swarm to simulate the location of real obstacles. The quality of particles is evaluated by a composite fitness function that integrates path mileage, energy consumption cost, and safety cost.

[0077] The lower layer performs local optimization on the solution space output by the upper layer based on the tabu search algorithm, establishes a tabu table with a short-term memory mechanism, and breaks through the local optimum through a neighborhood search strategy;

[0078] S4: Converts the planned path into flight control commands and sends them to the drone. It continuously monitors the flight status and triggers dynamic path adjustment or hovering commands if deviation from the path or sudden obstacles are detected.

[0079] S5: After the mission is completed, evaluate flight efficiency and safety indicators, optimize algorithm parameters and update system settings.

[0080] This embodiment achieves precise obstacle avoidance, efficient planning, real-time response, and adaptive evolution through dynamic three-dimensional safety modeling, layered collaborative optimization algorithm, edge real-time computing, and closed-loop feedback mechanism, without relying on subsequent claim details such as weight parameters.

[0081] Specifically, step 1:

[0082] The drone system boots up and loads the required hardware and algorithm modules, including sensor drivers, communication modules, and path planning modules. Simultaneously, it obtains the drone's own status information, such as its current latitude and longitude, altitude, flight time, remaining battery power, and the required safe distance for inspection equipment. It then establishes a communication connection with the ground control station to ensure stable and reliable data transmission.

[0083] The drone's onboard sensors are used to initialize the drone's status information and environmental data. The drone's location information, including longitude, latitude, and altitude, is obtained through the RTK (Real-Time Kinematic) positioning system. The drone's attitude angles (pitch, yaw, and roll) and linear acceleration are measured through the inertial measurement unit. High-precision three-dimensional terrain and object contours are obtained through the visual camera. The surrounding weather conditions (temperature, humidity, wind speed, etc.) are collected in real time through meteorological sensors.

[0084] Useful features are extracted from sensor data, such as the device's location coordinates, drone status data, weather conditions, and obstacle information. Statistical methods are used to delete, replace, or interpolate outliers in the collected data to facilitate subsequent processing and analysis.

[0085] Specifically, step 2:

[0086] By combining pre-entered equipment data with the visual recognition system on the drone, the location, shape, and type of electrical equipment in the substation are identified to determine the minimum safety radius length. . Calculate the actual distance between the drone and the electrical equipment based on their three-dimensional position coordinates:

[0087] Assume that the position coordinates of the UAV are ,in, Respectively represent drones The longitude, latitude, and altitude of the time period. Considering that electrical equipment is usually large and regular in shape, a conventional rectangular block or cylinder is usually used for modeling analysis when calculating the closest point between the drone and the equipment. Then, the coordinates of the position on the equipment closest to the drone are determined based on the flight direction, and the shortest safe distance is calculated using the Euclidean method based on this position. :

[0088]

[0089] and,

[0090] in, is the coordinate of the closest point between the electrical equipment and the drone, The radius length of the circular cross section for modeling the drone;

[0091] The collected environmental data and drone status information are processed, meteorological data is analyzed, and the factors affecting safe flight in the current weather environment are determined, including visibility, humidity, wind speed, etc. For example, strong winds require adjustment of the drone's flight altitude and speed. According to the topography and obstacle distribution, mathematical models are used in simulation experiments to conduct realistic simulations to determine potential flight risk areas, including altitude, longitude, and latitude. Finally, considering the safety distance requirements for electrical equipment, the safe inspection distance of drones is calculated. Evaluate the relevance of different factors considered in drone safety operations:

[0092] This paper uses a linear regression model to predict the correlation between these factors and safe UAV flight. This model can not only handle complex nonlinear relationships, but also continuously improve the prediction accuracy as data accumulates:

[0093]

[0094] in, is the predicted value, is the intercept term, is the coefficient of each feature, is the input feature;

[0095] Specifically, step 3:

[0096] The safe path generation process is as follows: Figure 2 As shown in the figure, the drone collects flight status information, environmental data, and the safety distance required for equipment inspection, transmits this information to the edge computing node, and determines the correlation coefficient of each factor affecting the safe flight of the drone based on the linear regression training model.

[0097] A two-layer search approach is used to solve the drone path planning problem. The upper layer uses a particle swarm algorithm for a coarse search, while the lower layer uses a more localized search method for refined optimization. A flight path is generated based on the drone's current position, target location, and device safety distance requirements, taking into account environmental data, obstacle information, and drone status information.

[0098] The particle swarm optimization algorithm is a heuristic search method that simulates the foraging behavior of a flock of birds. Each particle represents a potential solution. The particle's velocity and position are updated by tracking individual and global extrema, and the optimal solution is then searched for in the solution space.

[0099] To enhance the uniformity and universality of the initial particle distribution, Latin hypercube sampling is used to initialize the population. The sampling method in statistics is used to generate a sample set so that the particle samples are evenly distributed in all dimensions, thereby reducing the correlation between samples and improving the representativeness of the initial particle group:

[0100] Assume that Dimensional space generation particles, each dimension interval is Then, each dimension is divided into intervals of equal width , for the Dimension, Create uniformly distributed intervals: ,in, In each interval Randomly select a particle ,in, Indicates that in the interval Finally, the sample points of each dimension are combined to form a sample point ,in, It is from An integer randomly selected from the , to ensure that the particles of each dimension From different intervals.

[0101] Furthermore, in order to enhance the robustness of the path planning algorithm, some virtual particles representing obstacles are added to the particle swarm. The positions of these particles are fixed and used to simulate the positions of real obstacles. , energy consumption cost and safety costs As the fitness function of the particle swarm, the evaluation criteria for the quality of the path are defined:

[0102]

[0103] in, 、 、 are the corresponding weight ratios respectively.

[0104] Virtual obstacle particles can affect safety cost calculations through spatial position encoding, and each virtual particle generates a repulsive field. In FPGA hardware testing, virtual particle data can be directly written into the safety cost calculation unit through the DMA channel, while the formula structure remains unchanged. The added virtual particles have little impact on the calculation, but can effectively improve the path safety margin. The implicit mapping of the parameter space maintains the simplicity of the formula while realizing the physical influence of the virtual particles, which conforms to the cohesive principle of the control logic.

[0105] Based on the local search algorithm Tabu Search, the solution space found by the improved particle swarm algorithm is refined to obtain a more optimal path result. In the UAV path planning problem, Tabu Search introduces a short-term memory mechanism (i.e., a tabu table) to prevent the algorithm from falling into the local optimal solution during the search process. The specific implementation scheme is as follows:

[0106] The preliminary path obtained from the improved particle swarm algorithm search above is used as the initial solution. The tabu length and maximum number of iterations are set. At the same time, an empty tabu table is initialized to store recently visited solutions or move operations to prevent the algorithm from repeating the same content during the search process. The fitness function defined by the particle swarm algorithm is used as the evaluation function of Tabu Search. The optimal solution is selected from the neighborhood of the current solution and the tabu condition is determined to be met. When a solution is selected, it is added to the tabu table and the tabu length is set. If a solution is already in the tabu table, it is skipped. When the new solution is better than the currently known best solution, the best solution is updated. When the preset maximum number of iterations is reached, the search is stopped, and the final result found during the precise search is output as the best path.

[0107] Specifically, step 4:

[0108] The planned flight path is broken down into a series of specific instructions, such as takeoff, turning, forward, and descent, which are sent to the drone's flight control system to control the drone's flight direction, speed, altitude, and other aspects. During flight, the drone's status data, including position (latitude and longitude, altitude), speed, and attitude (pitch, yaw, and roll angles), is continuously collected. Wireless communication modules are used to maintain communication with the ground control station, transmitting flight data and status information in real time for ground personnel to monitor and make decisions.

[0109] The drone's flight status is continuously monitored, and the RTK data is compared with the planned path to ensure the drone adheres to the planned route. If deviation or unexpected obstacles are detected, adjustments are made immediately. If the drone still cannot safely avoid obstacles despite the dynamically adjusted flight path based on real-time data, the drone will hover in place, awaiting further instructions and triggering an alarm mechanism to notify the ground control station for manual intervention.

[0110] Specifically, step 5:

[0111] The return-to-home process begins when the inspection mission is completed, the preset flight time is reached, or the battery is low. Based on the current location and environmental information, a safe return path is planned to ensure the drone returns smoothly to its takeoff point or designated landing location. During the return process, environmental awareness and flight control are continuously maintained to ensure a safe landing.

[0112] The flight quality of the planned path is comprehensively evaluated using flight efficiency, safety, and inspection quality indicators. Based on any identified issues, path planning algorithms, such as the particle swarm optimization algorithm, are adjusted. Simultaneously, the ground control station's software interface and operational procedures are optimized to improve the efficiency and accuracy of manual intervention. The optimized algorithms and system settings are applied to subsequent flight missions, continuously monitoring and evaluating various flight indicators.

[0113] Example 2:

[0114] A UAV adaptive path planning system based on double-layer search, the system comprising:

[0115] 1. Data acquisition module, which uses sensors carried by drones to collect real-time data on the location, type, and safety distance of electrical equipment in substations, and cleans and organizes the data;

[0116] The data acquisition module includes the following sensors:

[0117] RTK positioning system, used to obtain high-precision latitude, longitude and altitude of the UAV;

[0118] An inertial measurement unit (IMU) is used to measure the attitude angle and linear acceleration of the drone, where the attitude angles include pitch, yaw, and roll angles.

[0119] Vision cameras, used to build 3D contour models of substation equipment;

[0120] Meteorological sensors are used to collect wind speed, temperature and humidity data in real time.

[0121] Second, a safety calculation module for dynamically calculating the three-dimensional safety distance between the drone and the device based on the device type and location, combined with the drone's real-time status information and environmental data; the drone's real-time status information includes latitude and longitude, altitude, flight time, and remaining battery power; and the environmental data includes temperature, humidity, and wind speed;

[0122] The calculation method of the three-dimensional safety distance in the safety calculation module is:

[0123] Model the electrical equipment as a regular geometric body and calculate the Euclidean distance between the drone and the closest point on the equipment surface;

[0124] The safety distance threshold is dynamically adjusted based on the flight direction of the drone. The calculation formula is:

[0125] The formula is:

[0126]

[0127] Where: Respectively represent drones Longitude, latitude and altitude of the time period; The coordinates of the closest point between the electrical equipment and the drone; The minimum safety radius corresponding to the device type.

[0128] The safety calculation module also includes: a correlation prediction unit, which is used to predict the correlation between environmental factors and the safe flight of the drone through a linear regression model, and dynamically adjust the safety cost by using the correlation as a weight coefficient.

[0129] Path planning module, deployed on edge computing nodes, uses a two-layer path planning algorithm to generate the optimal and safe path, including:

[0130] The upper-level planning unit uses a particle swarm algorithm to search the global solution space. When initializing the particle swarm, a uniformly distributed initial solution is generated through Latin hypercube sampling, and virtual obstacle particles used to simulate the positions of real obstacles are embedded in the particle swarm. The fitness function of the particle swarm algorithm integrates path mileage, energy consumption cost, and safety cost, and its formula is:

[0131] ;

[0132] Where: is the path mileage, Energy consumption costs and For security costs; 、 、 are the corresponding weight ratios, ;

[0133] The lower-level optimization unit performs local optimization on the solution space output by the upper-level layer based on the tabu search algorithm, breaking through the local optimum by establishing a tabu table with a short-term memory mechanism and a neighborhood search strategy;

[0134] The specific implementation of the taboo search algorithm in the lower-level optimization unit includes:

[0135] The path output by the upper particle swarm is used as the initial solution; the length of the taboo table and the maximum number of iterations are set to prohibit repeated access to recent solutions; new solutions are generated through neighborhood operations, and the solution with the best fitness and not tabooed is selected to update the global optimal path.

[0136] The flight control module converts the planned path into flight control commands and sends them to the drone. It also continuously monitors the flight status and triggers dynamic path adjustments or hovering commands if it detects deviation from the path or unexpected obstacles.

[0137] The dynamic path adjustment of the flight control module includes: if the drone deviates from the path by more than a preset threshold, re-calling the path planning module to generate a detour path; if a sudden obstacle is detected, updating the device model based on the real-time data of the visual camera, and triggering the lower-level optimization unit to perform local taboo search for rapid obstacle avoidance.

[0138] 5. Feedback optimization module, which is used to evaluate flight efficiency and safety indicators after the mission is completed, optimize algorithm parameters and update system settings.

[0139] The optimization algorithm parameters of the feedback optimization module include: adjusting the inertia weight of the particle swarm algorithm and the taboo table length of the taboo search according to historical flight data; updating the safety distance weight coefficient through the linear regression model , to reflect the risk level of different devices.

[0140] This technical solution's data acquisition module utilizes multiple sensors onboard drones to collect rich data in real time and then cleans and organizes it, providing an accurate and reliable data foundation for subsequent steps and ensuring the system's precise perception of the substation environment and drone status. The safety calculation module dynamically calculates three-dimensional safety distances based on device type, location, drone real-time status, and environmental data. Taking multiple factors into account, it more accurately measures the safety status between drones and electrical equipment, effectively mitigating safety risks. The upper-level planning unit of the path planning module utilizes a particle swarm algorithm, initializing the particle swarm through Latin hypercube sampling to ensure a uniform distribution of initial solutions, improving search efficiency and quality. Virtual obstacle particles are embedded to simulate real-world scenarios, enhancing the algorithm's adaptability to complex environments. A multi-cost fitness function is integrated to comprehensively consider path mileage, energy consumption, and safety factors, ensuring that the planned path is more aligned with actual needs. The lower-level optimization unit, based on a tabu search algorithm, utilizes a tabu table with a short-term memory mechanism and a neighborhood search strategy to avoid local optima. It then refines the upper-level solution space for local optimization, further improving path quality and safety. The flight control module converts the planned path into commands to control the drone's flight and continuously monitors flight status. If the drone strays from the path or encounters an unexpected obstacle, it can promptly trigger dynamic path adjustments or hover commands to ensure safe flight and smooth mission execution. After the mission is completed, the feedback optimization module evaluates flight efficiency and safety indicators, optimizing algorithm parameters and updating system settings accordingly, continuously improving system performance and better adapting to different operational scenarios and changing requirements.

[0141] It is understandable that the detailed functional implementation of each of the above modules can be found in the introduction of the aforementioned method embodiment, and no further details are given here.

[0142] The above-mentioned method and system for adaptive path planning of a drone based on double-layer search is a specific embodiment of the present invention, which has reflected the substantial characteristics and progress of the present invention. According to actual use needs, under the guidance of the present invention, equivalent modifications in shape, structure, etc. can be made to it, which are all within the scope of protection of this scheme.

Claims

1. A method for adaptive path planning of unmanned aerial vehicle based on double-layer search, characterized in that: The following steps are involved: Step 1: Use the drone-mounted sensors to collect real-time information about the location and type of electrical equipment in the substation. Clean and organize the data to determine the minimum safety radius based on the type of electrical equipment. Step 2: Dynamically calculate the three-dimensional safe distance between the drone and the device based on the device type and location, combined with the drone's real-time status information and environmental data. The drone's real-time status information includes latitude and longitude, altitude, flight time, and remaining battery power; environmental data includes temperature, humidity, and wind speed. Step 3: Based on the edge computing node, a two-layer path planning algorithm is used to generate the optimal safe path, where: The upper layer uses a particle swarm algorithm to search the global solution space. Latin hypercube sampling is used to initialize the particle swarm, and virtual obstacle particles are embedded in the particle swarm to simulate the location of real obstacles. The quality of particles is evaluated by a composite fitness function that integrates path mileage, energy consumption cost, and safety cost. The lower layer performs local optimization on the solution space output by the upper layer based on the tabu search algorithm, establishes a tabu table with a short-term memory mechanism, and breaks through the local optimum through a neighborhood search strategy; Step 4: Convert the planned path into flight control commands and send them to the drone. Continuously monitor the flight status. If deviation from the path or sudden obstacles are detected, dynamic path adjustment or hovering commands are triggered. Step 5: After the mission is completed, evaluate flight efficiency and safety indicators, optimize algorithm parameters and update system settings; The sensor in step 1 includes: RTK positioning system, used to obtain high-precision latitude, longitude and altitude of the UAV; Inertial measurement unit, used to measure the attitude angle and linear acceleration of the drone. The attitude angle includes pitch angle, yaw angle and roll angle; Vision cameras, used to build 3D contour models of substation equipment; Meteorological sensors for collecting wind speed, temperature and humidity data in real time; The calculation method of the three-dimensional safety distance in step 2 is: Model the electrical equipment as a regular geometric body and calculate the Euclidean distance between the drone and the closest point on the equipment surface; Combined with the flight direction of the drone, the safety distance threshold is dynamically adjusted. The formula is: Where: Respectively represent drones Longitude, latitude and altitude of the time period; The coordinates of the closest point between the electrical equipment and the drone; The minimum safety radius corresponding to the device type.

2. The method for adaptive path planning for a UAV based on a dual-layer search according to claim 1, characterized in that: In step 2, a linear regression model is used to predict the correlation between environmental factors and safe UAV flight, and the correlation is used as a weight coefficient to dynamically adjust the safety cost.

3. The method for adaptive path planning for a UAV based on a dual-layer search according to claim 1, characterized in that: The fitness function of the particle swarm algorithm in step 3 is defined as: ； Where: is the path mileage, Energy consumption cost and For security costs; 、 、 are the corresponding weight ratios, .

4. The method for adaptive path planning for a UAV based on a dual-layer search according to claim 3, characterized in that: The specific implementation of the tabu search algorithm in step 3 includes: The path output by the upper particle swarm is used as the initial solution; Set the length of the taboo table and the maximum number of iterations to prohibit repeated access to recent solutions; Generate new solutions through neighborhood operations, and select the solution with the best fitness and not tabooed to update the global optimal path.

5. The method for adaptive path planning of a UAV based on a dual-layer search according to claim 4, characterized in that: The dynamic path adjustment in step 4 includes: If the UAV deviates from the path by more than a threshold, the two-layer search algorithm is re-invoked to generate a detour path; If a sudden obstacle is detected, the device model is updated based on the real-time data from the visual camera, and a local taboo search is triggered for rapid obstacle avoidance.

6. The method for adaptive path planning of a UAV based on a dual-layer search according to claim 5, characterized in that: The optimization algorithm parameters in step 5 include: Adjust the inertia weight of the particle swarm algorithm and the length of the tabu table of the tabu search based on historical flight data; Update the safety cost weight coefficient through the linear regression model , to reflect the risk level of different devices.

7. A UAV adaptive path planning system based on double-layer search, characterized in that: include: The data acquisition module is used to collect the location and type of electrical equipment in the substation in real time through sensors carried by drones, clean and organize the data, and obtain the corresponding minimum safety radius based on the type of electrical equipment; A safety calculation module is used to dynamically calculate the three-dimensional safety distance between the drone and the device based on the device type and location, combined with the drone's real-time status information and environmental data; the drone's real-time status information includes latitude and longitude, altitude, flight time, and remaining battery power; and the environmental data includes temperature, humidity, and wind speed; The path planning module, deployed on edge computing nodes, uses a two-layer path planning algorithm to generate the optimal and safe path, including: The upper-level planning unit uses a particle swarm algorithm to search the global solution space. When initializing the particle swarm, a uniformly distributed initial solution is generated through Latin hypercube sampling, and virtual obstacle particles used to simulate the positions of real obstacles are embedded in the particle swarm. The fitness function of the particle swarm algorithm integrates path mileage, energy consumption cost, and safety cost, and its formula is: Where: is the path mileage, Energy consumption cost and For security costs; 、 、 are the corresponding weight ratios, ; The lower-level optimization unit performs local optimization on the solution space output by the upper-level layer based on the tabu search algorithm, breaking through the local optimum by establishing a tabu table with a short-term memory mechanism and a neighborhood search strategy; The flight control module converts the planned path into flight control commands and sends them to the drone. It also continuously monitors the flight status and triggers dynamic path adjustments or hovering commands if it detects deviation from the path or sudden obstacles. The feedback optimization module is used to evaluate flight efficiency and safety indicators after the mission is completed, optimize algorithm parameters and update system settings.

8. The UAV adaptive path planning system based on dual-layer search according to claim 7, characterized in that: The data acquisition module includes the following sensors: an RTK positioning system for obtaining high-precision latitude, longitude, and altitude of the drone; an inertial measurement unit for measuring the drone's attitude angles and linear acceleration, including pitch, yaw, and roll angles; a visual camera for building a three-dimensional contour model of the substation equipment; and a meteorological sensor for collecting wind speed, temperature, and humidity data in real time. The safety calculation module models the electrical equipment as a regular geometric body and calculates the Euclidean distance between the UAV and the nearest point on the equipment surface. If the calculated distance is greater than or equal to the minimum safety radius corresponding to the equipment type, it is considered safe. The safety calculation module includes a correlation prediction unit, which is used to predict the correlation between environmental factors and the safe flight of the UAV through a linear regression model and use the correlation as a weight coefficient to dynamically adjust the safety cost. The specific implementation of the taboo search algorithm in the lower optimization unit includes: taking the path output by the upper particle swarm as the initial solution; setting the length of the taboo table and the maximum number of iterations to prohibit repeated access to recent solutions; generating new solutions through neighborhood operations, and selecting the solution with the best fitness and not tabooed to update the global optimal path; The dynamic path adjustment of the flight control module includes: if the drone deviates from the path by more than a preset threshold, re-calling the path planning module to generate a detour path; if a sudden obstacle is detected, updating the device model based on the real-time data of the visual camera, and triggering the lower-level optimization unit to perform local taboo search for rapid obstacle avoidance.

Images