Multi-unmanned aerial vehicle cooperative rail transit inspection method and system based on improved dung beetle optimization algorithm

Abstract

The invention provides a multi-unmanned aerial vehicle cooperative rail transit inspection method and system based on an improved dung beetle optimization algorithm, and the method comprises the steps: constructing a rail transit network, marking an inspection task point, a no-fly zone and a charging station, and obtaining the parameters of an unmanned aerial vehicle fleet; on the basis of geographical distribution and priority of inspection task points, the inspection task is divided into a plurality of sub-tasks / responsibility areas, and initial roles or responsible areas are set for all unmanned aerial vehicles in combination with unmanned aerial vehicle fleet parameters; the method comprises the following steps: initializing each individual representing a complete inspection task scheme of task allocation-flight path-charging plan by adopting chaotic Tent mapping based on an initial role or a responsible area set by each unmanned aerial vehicle to form an initial dung beetle population; performing iterative optimization circulation on the initial dung beetle population based on an improved ERA-DBO algorithm to obtain an optimal individual; the optimal individual is an optimal inspection scheme, and the optimal inspection scheme is issued to each unmanned aerial vehicle to execute an inspection task, so that multi-unmanned aerial vehicle cooperative rail transit inspection is realized.

Description

Technical Field

[0001] This invention belongs to the field of intelligent operation and maintenance technology for rail transit infrastructure, specifically relating to a multi-UAV collaborative rail transit inspection method and system based on an improved dung beetle optimization algorithm. Background Technology

[0002] As a crucial transportation infrastructure in modern society, rail transit has a wide distribution and complex structure, placing extremely high demands on the real-time, comprehensive, and accurate nature of its inspection and maintenance. Traditional manual inspection methods are labor-intensive, inefficient, and highly subjective, and pose significant safety risks in adverse weather or complex terrain, making them unsuitable for the urgent needs of current rail transit safety operations. The application of drone technology has brought a completely new solution to rail transit inspection. Relying on the integration of multiple sensors such as high-definition cameras and LiDAR, drones can quickly, efficiently, and comprehensively complete automated inspections of track lines and their ancillary facilities, greatly improving the timeliness of problem detection and the quality of inspections.

[0003] While drone technology has significantly improved the efficiency and safety of rail transit inspections, a series of key technical challenges remain in practical applications. With the extension of inspection routes and the dense distribution of task points, multi-drone collaborative operations have become commonplace. However, achieving efficient drone collaboration, balanced energy consumption, and optimized endurance under deeply coupled task allocation, path planning, and energy constraints has become a major challenge in practical engineering. Inappropriate collaboration strategies will lead to insufficient task coverage, resource waste, or endurance bottlenecks, affecting overall operational efficiency.

[0004] Furthermore, the rail transit environment is highly dynamic; train traffic, weather changes, and drone malfunctions can all affect the real-time execution of inspection tasks. Existing algorithms struggle to achieve rapid adaptation to environmental changes and task replanning, making it difficult to guarantee operational continuity and safety. The coupling between path planning and energy management is also very tight; flight distance, mission duration, and battery life are closely related. Without unified modeling and optimization, practical problems such as mission interruptions or uneven energy utilization can easily occur.

[0005] Meanwhile, as the operation and maintenance system along the railway line evolves towards an edge computing architecture, the method faces multiple challenges in scenarios with limited computing power and high response time, including convergence speed, solution quality, and resource constraints. Existing intelligent optimization methods still struggle to balance global search capability, local convergence efficiency, and engineering practicality.

[0006] To address the aforementioned challenges, researchers have attempted to apply swarm intelligence optimization methods such as genetic algorithms, particle swarm optimization, and ant colony optimization. Dung beetle optimization (DBO), as an emerging swarm intelligence method, has shown certain advantages in path optimization and other problems by simulating dung beetle behavior. However, standard DBO and its existing improvements still have shortcomings: (1) the population initialization method is singular, making it difficult to guarantee the global exploration capability of the solution space; (2) the individual update mechanism and behavioral parameters are fixed, making it impossible to dynamically adjust according to the environment, resulting in unstable convergence speed and solution quality; (3) there is a lack of integrated mechanisms for the actual needs of rail transit, such as multi-UAV task allocation, energy constraints, and real-time dynamic replanning, making it difficult to balance global optimization and engineering practicality.

[0007] Therefore, there is an urgent need for an intelligent optimization method that can address the characteristics of multi-UAV collaborative inspection in rail transit, integrate task allocation, path planning and energy management, adapt to edge computing environments, and possess efficient global search, fast local convergence and real-time dynamic adaptive capabilities, in order to break through existing technical bottlenecks and achieve efficient, safe and intelligent upgrades to rail transit inspection. Summary of the Invention

[0008] To address the problems existing in the prior art, this invention provides a multi-UAV collaborative rail transit inspection method and system based on an improved dung beetle optimization algorithm. The aim is to deeply modify and expand the mechanism and functions of the standard DBO method, so that it can not only optimize the path, but also adapt to the specific needs of rail transit inspection, such as dynamic task and role allocation, and intelligent decision-making based on energy state.

[0009] To achieve the above objectives, the present invention provides the following solution: A multi-UAV cooperative rail transit inspection method based on an improved dung beetle optimization algorithm, the method comprising: S1. Construct a rail transit network, mark inspection task points, no-fly zones and charging stations based on the rail transit network, and obtain drone fleet parameters; S2. Based on the geographical distribution and priority of inspection task points, the inspection task is divided into several sub-tasks / responsibility areas through clustering or heuristic rules, and the initial role or responsible area is set for each drone in combination with the drone fleet parameters. S3. Based on the initial roles or areas of responsibility set for each drone, use chaotic Tent mapping to initialize each individual representing the complete inspection task plan of "task allocation-flight path-charging plan" to form the initial dung beetle population. S4. Based on the improved ERA-DBO algorithm, the initial dung beetle population is iteratively optimized to obtain the optimal individual. S5. The optimal individual is the optimal inspection plan. The optimal inspection plan is distributed to each drone to perform the inspection task, so as to realize the collaborative inspection of rail transit by multiple drones.

[0010] Preferably, the chaotic Tent mapping includes: ; in, For the first The dimensional chaotic variable corresponds to the 3rd dimension in the optimization algorithm. k The dung beetle individual in the first Normalized positional values ​​in the solution space. These are control parameters.

[0011] Preferably, the method for S4 to iteratively optimize the initial dung beetle population based on the improved ERA-DBO algorithm to obtain the optimal individual includes: S4.1. Based on the dung beetle, explore energy sensing paths and generate a global path framework for sustainable energy. S42. Based on the global path skeleton, a breeding ball mechanism is used to dynamically delineate high-priority collaborative operation areas or task clusters for each drone team. S43. Based on the dung beetle's local path and parameter optimization for a subset of tasks assigned to a specific UAV within its breeding ball; S44. When the drone is low on energy, malfunctions, or encounters an urgent new mission, activate the dung beetle stealing mechanism to take over and dynamically redistribute inspection tasks based on the dung beetle stealing mechanism, and obtain a new inspection plan. S45. The fitness of the new inspection scheme is evaluated based on the multi-objective fitness function to obtain the fitness value; S46. Based on the fitness value, update the population according to the principle of survival of the fittest, and retain the best solution to promote optimization until the best individual is obtained.

[0012] Preferably, the method in S42 for dynamically allocating high-priority collaborative work areas or task clusters for each UAV team using the breeding ball mechanism includes: ; ; in, and They represent the first The lower and upper boundaries of the reproductive sphere in 3D space; This indicates that the globally optimal individual in the current iteration is at the [number]th ...]. The position value of the dimension; This indicates that the worst individual in the current iteration is at the [number]th position. The position value of the dimension; It is a range adjustment factor for the breeding ball, which can be dynamically adjusted according to task density, priority and the distribution of charging stations in the area to ultimately determine the range of the breeding ball and attract dung beetles for subsequent optimization.

[0013] Preferably, the method in S44 for taking over and dynamically redistributing inspection tasks based on dung beetles to obtain a new inspection plan includes: Lévy flight enables large-step search to escape local optima: ; in, For Lévy's stride length during flight, and All are random variables that follow a normal distribution. , , This represents the power-law exponent of Lévy's flight; Except for the first Every other drone besides the one , Calculate the cost of taking over the task. : ; in, The increased energy consumption to take over the mission To increase time costs, For drones Current load, These are the corresponding weight coefficients; Select the optimal drone with the lowest cost : .

[0014] Preferably, the multi-objective fitness function in S45 is: ; in, For multi-objective fitness functions, For the inspection plan, The maximum task completion time. To ensure coverage of high-priority tasks, This refers to the total energy consumption or remaining electricity. For task load balancing, To constrain penalties, , , , All weights are adjustable.

[0015] This invention also provides a multi-UAV collaborative rail transit inspection system based on an improved dung beetle optimization algorithm. The system is used to implement the aforementioned method and includes: an environment modeling module, a task allocation and role assignment module, a population initialization module, an ERA-DBO optimization module, and an inspection execution module. The environmental modeling module is used to construct a rail transit network, mark inspection task points, no-fly zones and charging stations based on the rail transit network, and obtain drone fleet parameters. The task allocation and role assignment module is used to divide the inspection task into several sub-tasks / responsibility areas based on the geographical distribution and priority of the inspection task points, and to set the initial role or responsibility area for each drone in combination with the drone fleet parameters. The population initialization module is used to initialize each individual representing the complete inspection task plan of "task allocation-flight path-charging plan" based on the initial role or area of ​​responsibility set for each UAV, using chaotic Tent mapping, to form the initial dung beetle population; The ERA-DBO optimization module is used to iteratively optimize the initial dung beetle population based on the improved ERA-DBO algorithm to obtain the optimal individuals. The inspection execution module is used to determine the optimal inspection plan for each individual drone and distribute the optimal inspection plan to each drone to carry out the inspection task, thereby realizing multi-drone collaborative rail transit inspection.

[0016] Preferably, the ERA-DBO optimization module includes: an energy-sensing path generation unit, a key region identification unit, a local path optimization unit, a task dynamic reallocation unit, a fitness evaluation unit, and an optimal individual acquisition unit; An energy-sensing path generation unit is used to explore energy-sensing paths based on the dung beetle and generate a global path framework for sustainable energy. The key area identification unit is used to dynamically delineate high-priority collaborative operation areas or task clusters for each drone team based on the global path skeleton and using a breeding ball mechanism. The local path optimization unit is used to optimize the local path and parameters of a subset of tasks assigned to a specific UAV based on the dung beetle's behavior within its breeding ball. The task dynamic redistribution unit is used to activate the dung beetle mechanism when the drone is low on energy, malfunctions, or an urgent new task occurs. Based on the dung beetle, the inspection task is replaced and dynamically redistributed to obtain a new inspection plan. The fitness evaluation unit is used to evaluate the fitness of new inspection schemes based on a multi-objective fitness function and obtain fitness values. The optimal individual acquisition unit is used to update the population based on fitness values ​​according to the principle of survival of the fittest, and retain the optimal solution to promote optimization until the optimal individual is obtained.

[0017] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the aforementioned method.

[0018] The present invention also provides a computer-readable storage medium storing a computer program that, when executed, implements the aforementioned method.

[0019] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention deeply modifies and expands the mechanism of the standard DBO method, enabling it not only to optimize paths but also to adapt to the specific needs of rail transit inspection, such as dynamic task and role allocation and intelligent decision-making based on energy state.

[0020] (1) This invention innovatively introduces a dynamic role allocation mechanism. By redefining various behavioral patterns in the standard dung beetle optimization method, the "breeding ball" is used to divide key inspection areas and allocate task sets, so that each UAV or team can dynamically assume a specific role and responsibility area according to actual needs. The position and range of the breeding ball can be adjusted in real time according to task priority, density and UAV capabilities. For local path optimization, the foraging behavior of "small dung beetles" is used as a reference to realize the autonomous path fine optimization of a single UAV in the allocated area, which fully considers environmental constraints and operational requirements. At the same time, combined with the "stealing dung beetle" behavior, a dynamic task redistribution and resource scheduling mechanism is proposed. When some UAVs cannot continue to work due to insufficient energy, failure or encountering emergency tasks, their tasks can be dynamically transferred to UAVs with capabilities, ensuring the continuity and efficiency of the overall system operation.

[0021] (2) This invention further proposes an energy perception optimization mechanism. It deeply integrates factors such as the UAV's energy consumption model, battery status, and charging stations along the route into the ERA-DBO method process. The path generation stage fully incorporates energy cost evaluation, prioritizing energy-efficient and energy-feasible paths, and rationally planning charging behavior, dynamically optimizing charging timing and charging station selection. The fitness assessment comprehensively considers key indicators such as total task energy consumption, remaining battery power, and charging accessibility, achieving synergistic optimization of inspection operations and energy management, thereby improving the endurance and operational efficiency of multi-UAV systems. Attached Figure Description

[0022] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments are briefly introduced below. Obviously, the 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.

[0023] Figure 1 This is an overall flowchart of the ERA-DBO method proposed in this invention; Figure 2This is a flowchart of the dung beetle energy sensing path generation module in the ERA-DBO proposed in this invention; Figure 3 This is a flowchart of the dynamic role / task redistribution mechanism for dung beetle theft in ERA-DBO proposed in this invention; Figure 4 This is a schematic diagram of a rail transit simulation environment for verifying the ERA-DBO method proposed in this invention; Figure 5 This invention compares the performance of ERA-DBO with four comparison methods in energy balance contrast index under large-scale complex scenes and dynamic scenes. Figure 6 This invention compares the performance of ERA-DBO with four other methods in high-priority task coverage metrics under large-scale complex and dynamic scenarios. Figure 7 This invention compares the performance of ERA-DBO with four comparative methods in both large-scale and dynamic scenarios in terms of maximum completion time. Figure 8 This invention compares the performance of ERA-DBO in dynamic scenarios with four comparative methods in terms of dynamic task recovery time. Figure 9 This is a radar chart comparing the overall performance of the proposed ERA-DBO algorithm in dynamic scenarios with four other methods. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0026] Example 1 like Figure 1 As shown, this invention provides a multi-UAV cooperative rail transit inspection method based on an improved dung beetle optimization algorithm, comprising: S1. Construct a rail transit network, mark inspection task points, no-fly zones and charging stations based on the rail transit network, and obtain drone fleet parameters; S2. Based on the geographical distribution and priority of inspection task points, the inspection task is divided into several sub-tasks / responsibility areas through clustering or heuristic rules, and the initial role or responsible area is set for each drone in combination with the drone fleet parameters. S3. Based on the initial roles or areas of responsibility set for each drone, use chaotic Tent mapping to initialize each individual representing the complete inspection task plan of "task allocation-flight path-charging plan" to form the initial dung beetle population. S4. Based on the improved ERA-DBO algorithm, the initial dung beetle population is iteratively optimized to obtain the optimal individual. S5. The optimal individual is the optimal inspection plan. The optimal inspection plan is distributed to each drone to perform the inspection task, so as to realize the collaborative inspection of rail transit by multiple drones.

[0027] like Figure 1 As shown, the specific implementation process of the present invention is as follows: S1. Construct a rail transit network, mark inspection task points, no-fly zones, and charging stations based on the rail transit network, and obtain drone fleet parameters, specifically including: System initialization and rail transit environment modeling are the foundation for achieving effective route planning and resource allocation.

[0028] S11, Define the track network ( ), drone fleet ( ) and its various attributes (such as battery capacity) Average energy consumption rate ), task set ( This includes the location of each inspection task. Priority ), charging station ( ) and no-fly zones ( Basic variables such as )

[0029] S12. Define multiple constraints, including the maximum / minimum flight speed of the UAV ( , ), maximum climb and descent rate, minimum and maximum flight altitude ( , Maximum driving range or maximum battery life (Based on current battery status), etc. Simultaneously, it is essential to ensure the effective working distance and field of view of the sensors to cover all target points, and to ensure that the drone does not enter the no-fly zone at any time. This means setting a "no entry into the no-fly zone" constraint for all flight waypoints: for all waypoints All must meet .

[0030] S13. Simultaneously, hard constraints are defined for the minimum safe distance between drones and the remaining battery power safety margin: the minimum safe distance between drones throughout the entire flight is defined. For any time and different drones There is always ,in and Representing drones and drones The location coordinates.

[0031] The energy consumption of a drone can be expressed as the sum of its flight power and sensor power over a given time period: ,in Flight power is determined by factors such as flight speed, weight, and air density. This indicates the sensor power. For flight time, This refers to the sensor's operating time.

[0032] For different flight states (hovering, level, climb, etc.), power is modeled in segments. The power model for the UAV in level flight is as follows: ; in, For speed, , and These are all aerodynamic correlation coefficients. Specifically: The parasitic power coefficient is related to air density and the frontal area of ​​the fuselage, and represents the characteristic of the UAV to overcome air resistance. The induced power coefficient is related to the UAV's gravity and rotor / wing area, and represents the induced power characteristics required to generate lift. The drag power coefficient is related to the shape and rotational speed of the rotor blades, and represents the power required for the blades to overcome their own rotational drag.

[0033] S14, Define Total Flight Energy Consumption The cumulative energy consumption for each path segment is calculated using the following formula: ; in, For flight power; for The speed of flight at any given moment; for The flight mode at any given time includes hovering, level flight cruise, climb or descent, and different modes correspond to different power calculation parameters.

[0034] In practical planning, it is necessary to ensure that the total energy consumption of any path plus the energy required to reach a charging station or return does not exceed the currently available power, with a safety margin reserved. The energy constraint formula is as follows: ; in: Indicates drone The estimated energy consumption to complete the current planned route; This indicates the energy that the drone is expected to consume from the end of its path to the nearest charging station or return point. Indicates drone The current battery level displayed shows the remaining available power. This indicates the preset safety margin of power (used to cope with uncertainties such as changes in wind resistance or emergency obstacle avoidance).

[0035] Furthermore, S2, based on the geographical distribution and priority of inspection task points, divides the inspection task into several sub-tasks / responsibility areas through clustering or heuristic rules, and assigns an initial role or area of ​​responsibility to each drone in conjunction with drone fleet parameters, specifically including: In the initial task decomposition and generalized UAV role or area assignment phase, the inspection targets of rail transit first need to be spatially divided. The K-means++ algorithm is used to divide them into several inspection segments based on distance or segment characteristics. If the inspection task consists of multiple discrete points, clustering methods can be used to group them according to their geographical distribution. Specifically, all points to be inspected are clustered into... Clusters ( (This represents the number of drones), and then each cluster is directly assigned to one drone. The final result is the number of drones per cluster. initial task set or responsibility area .

[0036] Furthermore, S3, based on the initial roles or areas of responsibility set for each drone, uses chaotic Tent mapping to initialize each individual representing the complete inspection task plan of "task allocation-flight path-charging plan," forming the initial dung beetle population, specifically including: S31. During the ERA-DBO population initialization phase, each individual (i.e., the first) The dung beetle is coded into a complete multi-drone collaborative inspection solution. For A drone, It can be represented as: ; in, For the first The ordered mission sequence of the drone is , For the first The first drone One task, For the first The number of missions involving drones To and The corresponding list of path parameters, such as flight speed, altitude, or critical waypoints between each mission point. express The roles within the solution can be predefined task types (such as wide-area search, fine-grained detection, and emergency response) or dynamically generated job descriptions. Charging tasks can be inserted as special tasks. middle, i =1,2,..., N .

[0037] S32. Initialization is performed using Tent mapping. The formula for chaotic Tent mapping is: ; in, For the first The chaotic variable of dimension corresponds to the first dimension in the optimization algorithm. k The dung beetle individual in the first Normalized positional values ​​in the solution space; This is a control parameter (usually set to 2).

[0038] The generated chaotic sequence ( (The decision variable dimension) is transformed into an individual solution through the following mapping rule. Specific decision variables in the process: (1) For continuous decision variables (such as the initial position coordinates of the UAV, flight speed parameters, etc.), linear carrier mapping is used to transform them into the range of values ​​defined by the variables. Inside: ; in, For the first The first individual Dimensional actual parameter values, and These are the lower and upper bounds of the parameter, respectively. (2) For discrete decision variables (such as task allocation) The task execution priority in the sequence is transformed using a sorting mapping method: the generated chaotic sequence is then transformed. Sort in ascending or descending order to obtain the sorted index sequence. This index sequence corresponds to the task set. The execution order is arranged to transform chaotic variables into specific task allocation and execution order schemes.

[0039] Furthermore, S4, based on the improved ERA-DBO algorithm, iteratively optimizes the initial dung beetle population to obtain the optimal individual, specifically including: During the ERA-DBO iterative optimization cycle, the focus is on the inspection scheme. The fitness of the multi-objective system is updated iteratively. The fitness function is defined as a weighted sum of multiple optimization objectives, in the following form: ; in, For multi-objective fitness functions, For the inspection plan, The maximum task completion time. To ensure coverage of high-priority tasks, This refers to the total energy consumption or remaining electricity. For task load balancing, To constrain penalties, , , , All weights are adjustable.

[0040] S41. Based on the dung beetle, explore energy sensing paths and generate a global path framework for energy sustainability: like Figure 2 As shown, the dung beetle is responsible for energy-sensing optimization of the path of a single drone or a group of drones. First, it obtains the current state input parameters, including: Indicates the first Individuals (drones) in the first Current position at the next iteration This represents the drone's current remaining battery power. For unit energy consumption rate, For the next objective task, Find the nearest charging station to your current location. This is the preset low battery safety threshold.

[0041] Next, energy prediction calculations are performed. This predicts the total energy required to complete the next task. The calculation formula is as follows: ; in, The estimated flight energy consumption from the current location to the next mission point. The energy consumption required to complete this task.

[0042] The next step is to determine the energy threshold: assess the predicted energy. Is it below the low battery safety threshold? .

[0043] Based on the judgment results, the location update strategy falls into two categories: (1) If the judgment is "yes" (i.e.) If the path is 0, then the path will enter branch a "energy priority direction": the path exploration direction will prioritize the nearest charging station to ensure the energy feasibility of the path. The direction calculation formula is: ; in, Represents normalization. Represents the position coordinates of an object.

[0044] (2) If the result is "no", then proceed to branch b "Rolling Behavior Direction": perform a regular path search according to the rolling behavior of the standard DBO algorithm.

[0045] After determining the direction, a sequence of feasible waypoints is generated and energy consumption is recorded. The energy consumption of candidate path segments is calculated as follows: ; in, This represents a candidate path segment, which is the flight vector or trajectory segment from the current position of the UAV to the next candidate waypoint generated by the search. This variable contains information on the Euclidean distance and elevation changes of the path segment. This represents the energy calculation function, specifically meaning: based on path segments Physical properties (such as distance or estimated flight time) and drones unit energy consumption rate Perform a product operation to obtain the estimated energy consumption of this path segment.

[0046] At this point, a second verification is required. ( If the minimum remaining battery power is required, the path is deemed high-risk or infeasible. The final output is the drone's new location and remaining battery power, completing this iteration.

[0047] S42. Strategic Area / Task Priority Definition for the Breeding Sphere: Based on the global path skeleton, the breeding sphere mechanism dynamically allocates high-priority collaborative work areas or task clusters for each drone team. Specifically: The breeding sphere mechanism dynamically allocates high-priority collaborative work areas or task clusters for each drone team. The boundaries of these areas can be calculated based on the globally optimal and worst-case solutions. Its region boundary can be calculated based on the global optimal and worst solutions: ; ; in: and They represent the first The lower and upper boundaries of the reproductive sphere (i.e., the high-priority collaborative work area) in 3D space; This indicates that the globally optimal individual in the current iteration is at the [number]th ...]. The position value of the dimension; This indicates that the worst individual in the current iteration is at the [number]th position. The position value of the dimension; It is the range adjustment factor (or radius coefficient) of the breeding ball. This parameter is a dimensionless scalar that can be dynamically adjusted according to task density, priority and the distribution of charging stations in the area to ultimately determine the range of the breeding ball and attract dung beetles for subsequent optimization.

[0048] S43. Fine-grained path optimization within the dung beetle's role: Based on the dung beetle's local path and parameter optimization within the breeding sphere for a subset of tasks assigned to a specific drone, specifically: The dung beetle optimizes local paths and parameters for a subset of tasks assigned to a specific drone within its breeding sphere. Its typical position update format is as follows: ; in, and They represent the first Second and third The individual's position at the next iteration; To show obedience Uniformly distributed random numbers are used to introduce random perturbations to enhance the diversity of local searches; This represents the local search step size coefficient (or deflection coefficient), used to control the rate at which the dung beetle approaches the target location; Indicates the range of the breeding ball The selected guidance target location represents a potential better solution or task guidance point within the current high-priority responsibility area.

[0049] S44. Dynamic Role / Task Reassignment of Dung Beetles: When the drone is low on energy, malfunctions, or an urgent new task arises, the dung beetle mechanism is activated. Based on the dung beetles, inspection tasks are replaced and dynamically reassigned to obtain a new inspection plan, as detailed below: like Figure 3 As shown, the dung beetle mechanism is activated when the system malfunctions or changes. First, it collects the current system status information, including the... Remaining battery power of the drone Fault status flag bit (1 indicates a fault, 0 indicates normal operation), and newly arrived emergency task queues. .

[0050] Next, we proceed to the judgment stage: determining whether the conditions are met. or or (i.e., the battery level is below the threshold, a malfunction has occurred, or a new task has been assigned). If the result is "No", the process ends; if the result is "Yes", the task to be reassigned is extracted.

[0051] Set of tasks to be reassigned It consists of two parts: the first Unfinished mission set of drones With the new task queue ,Right now .

[0052] Subsequently, Lévy flight was used to generate position perturbation candidates. To escape local optima. Lévy's flight step size. The calculation formula is: ; in, and All are random variables that follow a normal distribution, used to approximate the random step size of the Lévy stable distribution using the Mantegna algorithm, specifically satisfying... , . The power-law exponent (or stability exponent) representing Lévy flight is used to control the long-tail property of the random step size distribution (i.e., to control the probability of generating large step sizes). In this embodiment... The value is usually 1.5.

[0053] Based on the perturbed state, except for the first Every other drone besides the one Calculate the cost of taking over the task. The cost function is defined as: ; in, The increased energy consumption to take over the mission To increase time costs, For drones Current load, These are the corresponding weighting coefficients.

[0054] After the calculation is completed, select the optimal drone with the lowest cost. : ; Finally, the set of tasks to be assigned Insert into the optimal drone In the path, and update its energy state. After allocation is complete, fitness is updated and recalculated. Fitness The value is then used to end the current redistribution process and obtain a new inspection plan.

[0055] S45. Fitness Assessment: The fitness of the new inspection plan is assessed based on a multi-objective fitness function to obtain a fitness value. Specifically: For each newly generated or modified inspection plan The aforementioned fitness function is used to conduct a multi-objective comprehensive evaluation, quantifying its global and local performance.

[0056] S46. Population Update and Selection: Based on fitness values, the population is updated according to the principle of survival of the fittest, and the optimal solution is retained to promote optimization until the best individual is obtained. Specifically: Based on fitness values, the DBO standard selection and optimization strategy is adopted to update the position of individuals, form the next generation of population, and retain the optimal solution to promote optimization.

[0057] Furthermore, S5, the optimal individual, represents the optimal inspection plan. This optimal inspection plan is then distributed to each drone to execute the inspection task, achieving collaborative multi-drone rail transit inspection. Specifically, this includes: When the algorithm terminates, the globally optimal individual This is the final output of the collaborative inspection plan. The plan includes the detailed flight path of each drone (including waypoints, speed, and altitude), task allocation (inspection point order), roles, and the expected charging schedule.

[0058] In summary, this invention proposes a multi-UAV collaborative rail transit inspection method based on an improved dung beetle optimization algorithm. It deeply modifies and expands the mechanism of the standard DBO method, enabling it not only to optimize paths but also to adapt to the specific needs of rail transit inspection, such as dynamic task and role allocation and intelligent decision-making based on energy state.

[0059] Specifically, this invention innovatively introduces a dynamic role allocation mechanism. By redefining various behavioral patterns in the standard dung beetle optimization method, a "breeding ball" is used to divide key inspection areas and allocate task sets, allowing each drone or team to dynamically assume specific roles and responsibility areas according to actual needs. The position and range of the breeding ball can be adjusted in real time based on task priority, density, and drone capabilities. For local path optimization, the foraging behavior of "small dung beetles" is used as a reference to achieve fine-grained autonomous path optimization for a single drone within the allocated area, fully considering environmental constraints and operational requirements. Simultaneously, combined with the behavior of "stealing dung beetles," a dynamic task reallocation and resource scheduling mechanism is proposed. When some drones cannot continue operating due to insufficient energy, malfunction, or encountering emergency tasks, their tasks can be dynamically transferred to drones with the capability, ensuring the continuity and efficiency of the overall system operation.

[0060] Furthermore, this invention proposes an energy-sensing optimization mechanism. It deeply integrates factors such as the UAV's energy consumption model, battery status, and charging stations along the route into the ERA-DBO method flow. The path generation stage fully incorporates energy cost evaluation, prioritizing energy-efficient and energy-feasible paths, and rationally planning charging behavior, dynamically optimizing charging timing and charging station selection. The fitness assessment comprehensively considers key indicators such as total mission energy consumption, remaining battery power, and charging accessibility, achieving synergistic optimization of inspection operations and energy management, thereby improving the endurance and operational efficiency of multi-UAV systems.

[0061] Example 2 To verify the effectiveness and superiority of the multi-UAV cooperative rail transit inspection method based on an improved dung beetle optimization algorithm proposed in this invention, a high-fidelity simulation environment was constructed and multiple sets of experiments were designed. Tests were conducted focusing on the key innovations proposed in the invention, such as task allocation, path planning, energy sensing, and dynamic reallocation. The experimental scheme covered both static and dynamic scenarios, and the technical effects of the method were comprehensively evaluated through performance comparisons with comparative algorithms.

[0062] 1. Simulation Environment Setup In the verification process of the multi-UAV cooperative rail transit inspection method based on the improved dung beetle optimization algorithm described in the foregoing embodiments, this embodiment constructs a parameterized and high-fidelity digital twin simulation environment, such as... Figure 4As shown, the effectiveness of the ERA-DBO optimization mechanism is rigorously evaluated. This simulation environment centers on a rail transit network, abstracting railway lines and related infrastructure into a directed graph G(V,E). Vertex set V represents stations, interchanges, and inspection task points, while edge set E represents feasible flight corridors and their attributes, including physical parameters such as length, altitude limits, and turning radius, thus closely mirroring actual track conditions. Inspection tasks are designed as independent objects with multiple attributes, including geographic coordinates, priority (levels 1–5), and estimated energy consumption, ensuring a close match to real-world operations. To meet the path constraints of flight path points in the aforementioned embodiments, the simulation environment also simulates no-fly zones, using three-dimensional geometry to represent obstacles or airspace restrictions, such as cylinders, cuboids, and polygons, to verify the compliance of UAV path planning. For energy replenishment needs, multiple charging stations are pre-defined in the simulation, with their locations, charging rates, and capacities defined to support dynamic charging scheduling for multiple UAVs. Regarding the UAV model itself, the system provides detailed characterization of its flight, sensor, and energy consumption parameters, including horizontal flight power. hovering power The climbing / descent power model conforms to the energy constraint principle described in the preceding embodiments and is superimposed when the sensor is working. Power Consumption. The system monitors the remaining power in real time and enforces feasibility constraints to ensure mission completion and sufficient return or charging capacity. Furthermore, considering the uncertainties of actual railway scenarios, the simulation system integrates dynamic events, including sudden UAV malfunctions, temporary insertion of high-priority tasks, changes in no-fly zones, and the unavailability of some charging stations. A detection delay parameter is introduced to simulate the impact of information transmission lag on dynamic task reallocation, thereby comprehensively verifying the robustness and dynamic adaptability of the ERA-DBO method. Table 1 introduces the simulation environment elements and related parameters.

[0063] Table 1. Description of Simulation Environment Elements and Parameters 2. Experimental Design (1) Scenario 1: Large-scale complex scenario The orbital dimensions are 20 km × 20 km, with 200 mission points (20% high, 40% medium, and 40% low), 10 drones, 15 no-fly zones (high complexity), and 8 charging stations. The scalability and computational efficiency of the methodology are examined.

[0064] (2) Scene 2: Dynamic Scene Based on large-scale scenarios, drone malfunctions, temporary no-fly zones, and high-priority task emergencies were introduced to test the dynamic task reallocation capability under the dung beetle mechanism.

[0065] 3. Comparison Methods The following methods were selected as the benchmark for comparison: (1) Standard Dung Beetle Optimization Algorithm (DBO) to evaluate the improvement effect of ERA-DBO in this invention.

[0066] (2) Multi-objective particle swarm optimization (MOPSO) and genetic algorithm (GA) represent traditional swarm intelligence optimization methods.

[0067] (3) Multi-agent reinforcement learning (MARL, using the MAPPO framework) represents a decentralized learning scheme.

[0068] All algorithms have undergone uniform parameter tuning to ensure fair comparison.

[0069] 4. Evaluation Indicators The following performance indicators were selected in the experiment to evaluate the effectiveness of the method of the present invention: (1) Task efficiency index, i.e., maximum task completion time; (2) Task quality indicators, namely, coverage of high-priority tasks; (3) Energy index, namely energy balance (variance of remaining power). (4) Robustness index, namely dynamic task recovery time.

[0070] 5. Simulation Results Figure 5 This section compares the ERA-DBO method with four other methods in terms of energy balance contrast (remaining battery power variance). This index reflects the balance of each method in task allocation and energy consumption scheduling. The smaller the variance, the closer the remaining battery power of each UAV (or robot), and the more balanced the energy utilization. This helps extend the continuous operation capability of the overall system and reduce the risk of single-point failure. In large-scale complex scenarios, ERA-DBO's energy balance score is only 15.7, far lower than other algorithms, indicating that it can effectively achieve balanced energy consumption distribution and prevent some UAVs from prematurely exiting the mission due to energy depletion. In dynamic scenarios, ERA-DBO's energy balance score is 17.5, still better than other algorithms, and it can maintain good energy balance even when the mission changes dynamically.

[0071] Figure 6This section compares the ERA-DBO method with four other methods in terms of high-priority task coverage. In large-scale complex scenarios, ERA-DBO achieves a high-priority task coverage of 98%, significantly higher than the other algorithms. This indicates that it prioritizes the completion of critical tasks during task allocation, improving the system's responsiveness to important tasks. In contrast, DBO, MOPSO, GA, and MARL have coverage rates of 89%, 91%, 92%, and 95%, respectively, all lower than ERA-DBO. DBO and MOPSO, in particular, show weaker performance in ensuring high-priority tasks in complex environments. In dynamic scenarios, ERA-DBO maintains its leading position with a high-priority task coverage of 97%, demonstrating its ability to prioritize resource allocation for critical tasks even when facing dynamic task changes and environmental uncertainties.

[0072] Figure 7 This section compares the performance of ERA-DBO with four other methods in terms of maximum completion time. In large-scale complex scenarios, ERA-DBO's maximum completion time is 112 minutes, significantly lower than the other algorithms (DBO 129 minutes, MOPSO 135 minutes, GA 140 minutes, MARL 121 minutes), indicating that ERA-DBO can complete all tasks more efficiently and shorten the overall system execution cycle. In dynamic scenarios, ERA-DBO's maximum completion time is 128 minutes, still better than the other algorithms (DBO 155 minutes, MOPSO 162 minutes, GA 170 minutes, MARL 140 minutes), demonstrating its ability to maintain fast task completion efficiency even when facing dynamic task changes and environmental uncertainties.

[0073] Figure 8 This section compares the performance of ERA-DBO with four other methods in terms of dynamic task recovery time. Figure 8 It can be seen that ERA-DBO's dynamic task recovery time is only 3.8 minutes, significantly lower than other algorithms (DBO 7.5 minutes, MOPSO 6.9 minutes, GA 7.2 minutes, and MARL 5.1 minutes). This indicates that ERA-DBO can complete scheduling and resource allocation faster when facing sudden tasks or dynamic changes in tasks, enabling the system to respond quickly to new tasks. In contrast, the recovery times of other algorithms are generally longer, indicating their insufficient adaptability and scheduling efficiency in dynamic environments.

[0074] Figure 9This is a radar chart comparing the overall performance of ERA-DBO with four other methods in dynamic scenarios. The radar chart clearly shows that the ERA-DBO algorithm ranks at the outermost edge in all four dimensions, demonstrating its superior performance. This indicates significant advantages in task completion efficiency, critical task assurance, energy distribution balance, and dynamic response capabilities. Specifically, ERA-DBO achieves the best results in both maximum completion time and dynamic task recovery time, reflecting its efficient task scheduling and rapid dynamic adaptation. In terms of high-priority task coverage, ERA-DBO also outperforms other algorithms, prioritizing the completion of critical tasks. Regarding energy balance, ERA-DBO achieves balanced energy distribution, improving the overall robustness of the system. In contrast, other algorithms have significant shortcomings in certain metrics, resulting in overall performance inferior to ERA-DBO.

[0075] Example 3 Based on the same inventive concept, the present invention also provides a multi-UAV collaborative rail transit inspection system based on an improved dung beetle optimization algorithm, used to implement the method described in the foregoing embodiments. The system includes: an environment modeling module, a task allocation and role assignment module, a population initialization module, an ERA-DBO optimization module, and an inspection execution module. The environmental modeling module is used to construct a rail transit network, mark inspection task points, no-fly zones and charging stations based on the rail transit network, and obtain drone fleet parameters. The task allocation and role assignment module is used to divide the inspection task into several sub-tasks / responsibility areas based on the geographical distribution and priority of the inspection task points, and to set the initial role or responsibility area for each drone in combination with the drone fleet parameters. The population initialization module is used to initialize each individual representing the complete inspection task plan of "task allocation-flight path-charging plan" based on the initial role or area of ​​responsibility set for each UAV, using chaotic Tent mapping, to form the initial dung beetle population; The ERA-DBO optimization module is used to iteratively optimize the initial dung beetle population based on the improved ERA-DBO algorithm to obtain the optimal individuals. The inspection execution module is used to determine the optimal inspection plan for each individual drone and distribute the optimal inspection plan to each drone to carry out the inspection task, thereby realizing multi-drone collaborative rail transit inspection.

[0076] Furthermore, the ERA-DBO optimization module includes: an energy-sensing path generation unit, a key region identification unit, a local path optimization unit, a task dynamic reallocation unit, a fitness evaluation unit, and an optimal individual acquisition unit; An energy-sensing path generation unit is used to explore energy-sensing paths based on the dung beetle and generate a global path framework for sustainable energy. The key area identification unit is used to dynamically delineate high-priority collaborative operation areas or task clusters for each drone team based on the global path skeleton and using a breeding ball mechanism. The local path optimization unit is used to optimize the local path and parameters of a subset of tasks assigned to a specific UAV based on the dung beetle's behavior within its breeding ball. The task dynamic redistribution unit is used to activate the dung beetle mechanism when the drone is low on energy, malfunctions, or an urgent new task occurs. Based on the dung beetle, the inspection task is replaced and dynamically redistributed to obtain a new inspection plan. The fitness evaluation unit is used to evaluate the fitness of new inspection schemes based on a multi-objective fitness function and obtain fitness values. The optimal individual acquisition unit is used to update the population based on fitness values ​​according to the principle of survival of the fittest, and retain the optimal solution to promote optimization until the optimal individual is obtained.

[0077] Example 4 Based on the same inventive concept, corresponding to the methods of any of the above embodiments, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the methods described in any of the above embodiments.

[0078] Example 5 Based on the same inventive concept, corresponding to the methods of any of the above embodiments, the present invention also provides a non-transitory computer-readable storage medium storing computer instructions for causing the computer to perform the methods described in any of the above embodiments.

[0079] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A multi-UAV cooperative rail transit inspection method based on an improved dung beetle optimization algorithm, characterized in that, The method includes: S1. Construct a rail transit network, mark inspection task points, no-fly zones and charging stations based on the rail transit network, and obtain drone fleet parameters; S2. Based on the geographical distribution and priority of inspection task points, the inspection task is divided into several sub-tasks / responsibility areas through clustering or heuristic rules, and the initial role or responsible area is set for each drone in combination with the drone fleet parameters. S3. Based on the initial role or area of ​​responsibility set for each drone, use chaotic Tent mapping to initialize each individual representing the complete inspection task plan of "task allocation-flight path-charging plan" to form the initial dung beetle population. S4. Based on the improved ERA-DBO algorithm, the initial dung beetle population is iteratively optimized to obtain the optimal individual. S5. The optimal individual is the optimal inspection plan. The optimal inspection plan is distributed to each drone to perform the inspection task, so as to realize the collaborative inspection of rail transit by multiple drones.

2. The method according to claim 1, characterized in that, The chaotic Tent mapping includes: ； in, For the first The dimensional chaotic variable corresponds to the 3rd dimension in the optimization algorithm. k The dung beetle individual in the first Normalized positional values ​​in the solution space. These are control parameters.

3. The method according to claim 1, characterized in that, The method described in S4, which iteratively optimizes the initial dung beetle population using an improved ERA-DBO algorithm to obtain the optimal individual, includes: S4.

1. Based on the dung beetle, explore energy sensing paths and generate a global path framework for sustainable energy. S42. Based on the global path skeleton, a breeding ball mechanism is used to dynamically delineate high-priority collaborative operation areas or task clusters for each drone team. S43. Based on the dung beetle's local path and parameter optimization for a subset of tasks assigned to a specific UAV within its breeding ball; S44. When the drone is low on energy, malfunctions, or encounters an urgent new mission, activate the dung beetle stealing mechanism to take over and dynamically redistribute inspection tasks based on the dung beetle stealing mechanism, and obtain a new inspection plan. S45. The fitness of the new inspection scheme is evaluated based on the multi-objective fitness function to obtain the fitness value; S46. Based on the fitness value, update the population according to the principle of survival of the fittest, and retain the best solution to promote optimization until the best individual is obtained.

4. The method according to claim 3, characterized in that, The method used in S42 to dynamically allocate high-priority collaborative work areas or task clusters for each UAV team using the breeding ball mechanism includes: ； ； in, and They represent the first The lower and upper boundaries of the reproductive sphere in 3D space; This indicates that the globally optimal individual in the current iteration is at the [number]th ...]. The position value of the dimension; This indicates that the worst individual in the current iteration is at the [number]th position. The position value of the dimension; It is a range adjustment factor for the breeding ball, which can be dynamically adjusted according to task density, priority and the distribution of charging stations in the area to ultimately determine the range of the breeding ball and attract dung beetles for subsequent optimization.

5. The method according to claim 4, characterized in that, The method in S44 for taking over and dynamically redistributing inspection tasks based on dung beetles to obtain a new inspection plan includes: Lévy flight enables large-step search to escape local optima: ； in, For Lévy's stride length during flight, and All are random variables that follow a normal distribution. , , This represents the power-law exponent of Lévy's flight; Except for the first Every other drone besides the one , Calculate the cost of taking over the task. : ； in, The increased energy consumption to take over the mission To increase time costs, For drones Current load, These are the corresponding weight coefficients; Select the optimal drone with the lowest cost : 。 6. The method according to claim 5, characterized in that, The multi-objective fitness function in S45 is: ； in, For multi-objective fitness functions, For the inspection plan, The maximum task completion time. To ensure coverage of high-priority tasks, This refers to the total energy consumption or remaining electricity. For task load balancing, To constrain penalties, , , , All weights are adjustable.

7. A multi-UAV cooperative rail transit inspection system based on an improved dung beetle optimization algorithm, the system being used to implement the method described in any one of claims 1-6, characterized in that, The system includes: an environment modeling module, a task allocation and role assignment module, a population initialization module, an ERA-DBO optimization module, and an inspection execution module; The environmental modeling module is used to construct a rail transit network, mark inspection task points, no-fly zones and charging stations based on the rail transit network, and obtain drone fleet parameters. The task allocation and role assignment module is used to divide the inspection task into several sub-tasks / responsibility areas based on the geographical distribution and priority of the inspection task points, and to set the initial role or responsibility area for each drone in combination with the drone fleet parameters. The population initialization module is used to initialize each individual representing the complete inspection task scheme of "task allocation-flight path-charging plan" based on the initial role or area of ​​responsibility set for each drone, using chaotic Tent mapping, to form the initial dung beetle population; The ERA-DBO optimization module is used to iteratively optimize the initial dung beetle population based on the improved ERA-DBO algorithm to obtain the optimal individuals. The inspection execution module is used to determine the optimal inspection plan for each individual drone and distribute the optimal inspection plan to each drone to carry out the inspection task, thereby realizing multi-drone collaborative rail transit inspection.

8. The system according to claim 7, characterized in that, The ERA-DBO optimization module includes: an energy-sensing path generation unit, a key area identification unit, a local path optimization unit, a task dynamic reallocation unit, a fitness evaluation unit, and an optimal individual acquisition unit; An energy-sensing path generation unit is used to explore energy-sensing paths based on the dung beetle and generate a global path framework for sustainable energy. The key area identification unit is used to dynamically delineate high-priority collaborative operation areas or task clusters for each drone team based on the global path skeleton and using a breeding ball mechanism. The local path optimization unit is used to optimize the local path and parameters of a subset of tasks assigned to a specific UAV based on the dung beetle's behavior within its breeding ball. The task dynamic redistribution unit is used to activate the dung beetle mechanism when the drone is low on energy, malfunctions, or an urgent new task occurs. Based on the dung beetle, the inspection task is replaced and dynamically redistributed to obtain a new inspection plan. The fitness evaluation unit is used to evaluate the fitness of new inspection plans based on a multi-objective fitness function and obtain fitness values. The optimal individual acquisition unit is used to update the population based on fitness values ​​according to the principle of survival of the fittest, and retain the optimal solution to promote optimization until the optimal individual is obtained.

9. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the method as described in any one of claims 1-6.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed, implements the method as described in any one of claims 1-6.

Images