Path optimization method for cleaning unmanned aerial vehicle swarm and application

Abstract

The invention provides a path optimization method for cleaning an unmanned aerial vehicle swarm and application. The method comprises the steps that position initialization is conducted on multiple unmanned aerial vehicles in a task area, a fitness function is set, and a variable Trials is recorded; the stages of employed bees, observation bees and reconnaissance bees are executed in sequence through an artificial bee colony algorithm, and path searching, probability selection, boundary processing and position updating are carried out; a reverse learning strategy, disturbance control frequency and a global factor are introduced into track initialization to improve search performance, and meanwhile, a nectar source regeneration mechanism in a scout bee stage is set. Through dynamic task allocation optimization, global collaborative path planning and a distributed elastic control architecture, the cleaning resource utilization rate, the operation efficiency and the energy consumption balance are improved, the fault-tolerant capability of the system is enhanced, and the cleaning requirement of the complex building facade is met.

Description

Technical Field

[0001] The present invention belongs to the technical field of unmanned aerial vehicles, and particularly relates to a path optimization method and application for a swarm of cleaning unmanned aerial vehicles. Background Technique

[0002] With high-rise buildings standing in cities, glass curtain walls, as an important part of modern architecture, their cleaning and maintenance have become a heavy and dangerous task. At present, traditional glass curtain wall cleaning mainly relies on manual labor or large-scale mechanical equipment, which not only has high costs, but also has problems such as low operation efficiency and great potential safety hazards. The application of a single unmanned aerial vehicle in cleaning operations is restricted by the equipment's load-bearing capacity and operating space, and it is difficult to efficiently complete large-area and high-difficulty cleaning tasks.

[0003] In recent years, the development of artificial intelligence (AI) technology has provided new possibilities for the cooperation of unmanned aerial vehicle swarms. Through the optimization of AI algorithms, unmanned aerial vehicle swarms can achieve complex and precise flight and operation modes, and cooperate to complete large-scale and high-efficiency cleaning tasks.

[0004] The current cooperation of unmanned aerial vehicle swarms in cleaning operations faces the following technical bottlenecks: First, the task allocation mechanism is rigid, and the differences in stain distribution and dynamic load changes are not considered, resulting in uneven distribution of cleaning resources, low operation efficiency and unbalanced battery energy consumption; second, the path planning lacks global cooperation, resulting in path overlap and obstacle avoidance conflicts, increasing the risk of collision and forming cleaning blind spots, reducing the operation coverage rate; third, the system relies on a centralized control architecture and a fixed communication topology, and there is a response delay when facing obstacles or failures, and the task interruption frequency is relatively high, making it difficult to adapt to the dynamic environment of complex building facades.

[0005] Therefore, the present invention provides a path optimization method and application for a swarm of cleaning unmanned aerial vehicles. Summary of the Invention

[0006] The present invention designs a path optimization method and application for a swarm of cleaning unmanned aerial vehicles to solve the problems existing in the above background technique.

[0007] In order to achieve the above technical effects, the present invention is realized through the following technical solutions: A path optimization method for a swarm of cleaning unmanned aerial vehicles, characterized by comprising the following steps:

[0008] S1: First, initialize the positions of the unmanned aerial vehicles: Select N unmanned aerial vehicles to participate in the task in the task area, and the position of each unmanned aerial vehicle is initialized by generating uniformly distributed random numbers within the corresponding interval; next, initialize the fitness function; finally, set a variable Trials for each unmanned aerial vehicle to record the number of times without improvement.

[0009] S2: In the employed bee stage, each drone searches for a new position within the neighborhood of its current location; in the observing bee stage, drones are selected by calculating the selection probability and using the roulette wheel selection method, followed by neighborhood search and boundary processing to update the position and fitness value; in the scout bee stage, the position is reinitialized by judging the abandonment condition, and the positions of the drones are reassigned.

[0010] S3: In the track initialization stage, a strategy based on reverse learning is adopted, introducing the disturbance control frequency and the global factor; in the scout bee stage, a way to regenerate the nectar source is defined.

[0011] Further, in S1, the two-dimensional plane area where the task area is located ranges from [0, L x in the x-axis direction and from [0, L y in the y-axis direction, where L x and L y respectively represent the side length and other dimension range values of the curtain wall in the corresponding directions.

[0012] Further, in S1, the initial value of the variable Trials is 0.

[0013] The present invention further protects the application of the above-mentioned cleaning drone path optimization method in cleaning the glass curtain wall of high-rise buildings.

[0014] The beneficial effects of the present invention are as follows:

[0015] The present invention adjusts the task weight in real time based on the stain characteristics and load status, optimizes the task allocation, improves the utilization rate of cleaning resources, and improves the operation efficiency and energy consumption balance; adopts global collaborative path planning, combines multi-aircraft state perception and obstacle prediction, effectively avoids path conflicts, reduces the collision risk and improves the cleaning coverage uniformity; through the distributed elastic control architecture, adopts the adaptive communication topology and redundant decision-making mechanism, enhances the fault tolerance ability of the system to dynamic interference, ensures the continuity of tasks, and meets the high-precision cleaning requirements of complex building facades. Description of the Drawings

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0017] Figure 1 It is a schematic diagram of the model process of the present invention; Detailed Embodiments

[0018] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0019] Embodiment 1

[0020] This embodiment aims to illustrate the actual application process of the present invention in the cleaning task of high-rise building glass curtain walls. The method plans the flight paths of multiple drones through intelligent algorithms and dynamically coordinates the spraying and scrubbing operations, so as to achieve efficient, intelligent, and collaborative cleaning task execution.

[0021] I. System construction and preparation

[0022] The experimental platform includes three major parts: a cleaning drone group, a central control system, and a sensing and feedback system:

[0023] The cleaning drone group is divided into a spraying group and a scrubbing group. The drones in the spraying group are equipped with water tanks and spraying devices for evenly covering the cleaning liquid on the curtain wall surface; the drones in the scrubbing group are equipped with drum scrubbing devices for mechanically scrubbing the sprayed areas.

[0024] Each drone is equipped with a navigation and perception module, including GPS, IMU, cameras, lidar, temperature and humidity sensors, etc., and has the capabilities of flight navigation, environment recognition, and operation feedback.

[0025] The central control system integrates a path optimization algorithm module (based on the artificial bee colony algorithm) for uniformly scheduling the task allocation, path generation, and dynamic adjustment of all drones.

[0026] Before the task starts, perform software and hardware checks on all drones, confirm that the water tanks, spraying systems, and scrubbing devices are working properly, and complete sensor calibration and communication tests. Subsequently, the central control system models the operation area according to the building curtain wall structure, divides the cleaning grid, and generates an initial path planning task.

[0027] II. Path optimization and task allocation

[0028] The central control system uses the artificial bee colony algorithm to optimize the paths of the drone swarm. The main process is as follows:

[0029] 1. Position initialization:

[0030] The system randomly generates an initial position for each drone in the operation area and initializes the fitness function and algorithm parameters.

[0031] 2. Bee colony optimization stage:

[0032] The employed bee phase conducts a neighborhood search on the current path;

[0033] The observing bee phase selects an optimized path according to the fitness value using the roulette wheel mechanism;

[0034] The scout bee phase re-initializes the path that has not been improved for a long time; during the search process, a perturbation control factor and a global factor are introduced to enhance the path diversity and the algorithm convergence effect.

[0035] 3. Task Assignment and Job Sequence Generation:

[0036] After optimization, the system automatically assigns the path to the spraying group and the scrubbing group, and reasonably arranges the start time and operation area of each drone to ensure the continuity of the cleaning process and no interference between different groups.

[0037] III. Cleaning Task Execution Process

[0038] 1. Flight and Operation:

[0039] All drones take off in sequence according to the generated path. The spraying group first evenly sprays the cleaning liquid according to the planned path, and then the scrubbing group drones take off and perform scrubbing operations along the spraying trajectory.

[0040] 2. Task Coordination and Obstacle Avoidance:

[0041] The central control system monitors the flight status, relative position and cleaning progress of each drone in real time, avoids the intersection or conflict of flight paths, and dynamically adjusts the flight paths and operation timings of some drones.

[0042] 3. Environment Adaptive Adjustment:

[0043] The system automatically adjusts the flight altitude, speed and task strategy according to environmental parameters such as temperature, humidity and wind speed to ensure the cleaning quality and flight safety.

[0044] IV. Data Feedback and Task Evaluation

[0045] During the cleaning process, the drones continuously collect data related to the operation, including the flight path, spraying volume, cleaning progress and visual images, etc. All data are transmitted back to the central control system in real time, and the cleaning effect is evaluated and the strategy is optimized by the AI analysis module.

[0046] After the task is completed, the system judges the cleaning coverage and uniformity according to the visual detection results, and generates a cleaning report to summarize the operation process and optimize the subsequent scheduling parameters.

[0047] Embodiment 2

[0048] Algorithm Details:

[0049] I. Artificial Bee Colony Algorithm Process

[0050] 1. Initialization phase:

[0051] Suppose there are N drones participating in the cleaning task. The two-dimensional plane area where the glass curtain wall is located (the range of the two-dimensional plane area where the task area is located) ranges from [0, L x in the x-axis direction and from [0, L y in the y-axis direction (L x , L y respectively represent the dimension range values such as the side length of the curtain wall in the corresponding direction). For the i-th drone (i = 1, 2,..., N), the initialization formulas for its initial abscissa x i0 and ordinate y i0 are:

[0052]

[0053] where rand(0, L x ) and ran⊙(0, L y ) represent uniformly distributed random numbers generated within the corresponding intervals. In this way, each drone is randomly placed within the glass curtain wall area as the initial position (food source position).

[0054] 2. Initialization of fitness function

[0055] Define the fitness function F(x i , y i ) to measure the contribution degree of the drone at the position (x i , y i ) to the overall cleaning task, considering the following factors:

[0056] F(x i , y i ) = k1 × Coverage(x i , y i ) - k2 × Distance(x i , y i ) + k3 × CleaningEfficiency(x i , y i ) where: · k1, k2, k3 are weight coefficients, which are set according to actual needs and the importance of each factor, and k1, k2, k3 ≥ 0. Usually, k1 + k2 + k3 = 1.

[0057] Coverage(x i , y i ) represents that the drone is located at (x i , y i) The area of the glass curtain wall that can be covered by a single cleaning when in a certain position can be calculated based on actual physical parameters such as the spraying range of the drone cleaning equipment and the extension range of the cleaning arm. For example, assuming that the cleaning equipment takes the position of the drone as the center of a circle, and a circular area with a radius of r can be covered, then Coverage(x i ,y i ) = πr 2 (If the coverage range is in other shapes such as a rectangle, the calculation formula needs to be changed accordingly).

[0058] Distance(x i ,y i ) measures the distance of this position from other factors. There are two common situations:

[0059] Firstly, the sum of the distances from other drones. The calculation formula is The purpose is to avoid the risk of drones getting too close and colliding and to reduce the overlapping cleaning areas. Or the distance from the already cleaned area (if there is a coordinate set representing the already cleaned area, it can be calculated according to the distance from a point to a set in a similar way). Here, the sum of the distances from other drones is taken as an example for now.

[0060] CleaningEfficiency(x i ,y i ) represents the cleaning efficiency at this position, which can be related to factors such as the attitude of the drone and the working angle of the cleaning equipment. For example, an efficiency value in the range [0, 1] can be set according to actual test data. Here, it is assumed to be a simple linear function related to the distance from the curtain wall center (for example only, can be adjusted as needed), such as

[0061] where d max is a pre-set maximum distance (such as the distance from one corner of the curtain wall to the opposite corner, etc.). The closer to the curtain wall center, the higher the cleaning efficiency.

[0062] Secondly, initialize the number of non-improving times. Set a variable Trials for each drone i to record the number of non-improving times, and the initial value is set to 0, which is used to judge whether to re-find a position later.

[0063] 3. Employed Bee Phase

[0064] (1) Neighborhood Search Formula

[0065] For the i-th employed bee (drone), search for a new position within the neighborhood of its current position (x i ,y i ) The neighborhood search is carried out through the following formula:

[0066]

[0067] Among them, k is the index of a drone randomly selected from the N - 1 drones other than i, that is and are random numbers generated in the interval [-1, 1] respectively, used to control the search step size and direction, so that the new position varies within a certain neighborhood range of the current position.

[0068] (2) Boundary processing formula

[0069]

[0070] Among them: To ensure that the new position is still within the glass curtain wall area, boundary processing is required. The formula is as follows:

[0071]

[0072] That is, if then If then For the ordinate Do the same processing to limit it within the effective range of the curtain wall area.

[0073] (3) Fitness comparison and update formula

[0074] Calculate the fitness value of the new position If then update the position and fitness value of drone i and the number of non - improvement times. The formula is as follows:

[0075]

[0076] If then only update the number of non - improvement times: Trials i =

[0077] Trials i + 1.

[0078] 4. Scout bee stage

[0079] (1) Selection probability calculation formula

[0080] First, calculate the probability P i that each drone i is selected by the scout bee. The calculation formula is:

[0081]

[0082] That is, the fitness value of the UAV accounts for the proportion of the sum of the fitness values of all UAVs, so as to reflect its relative advantage degree. The higher the fitness value of the UAV, the greater the probability of being selected.

[0083] (2) Roulette wheel selection formula (select UAVs by taking roulette wheel as an example)

[0084] Determine which UAV the observing bee selects to operate through the roulette wheel selection method. Assume that a random number r generated in the interval [0, 1] is taken, and start accumulating P from i = 1 i When it satisfies then select the i-th UAV for subsequent operations.

[0085] (3) Neighborhood search and subsequent processing (similar to the employed bee stage)

[0086] Assume that the UAV index selected by the roulette wheel is m, and calculate the new position according to the same neighborhood search formula as in the employed bee stage and perform boundary processing, then calculate the fitness value of the new position If then update:

[0087]

[0088] If then update the number of non-improved times:

[0089] Trials m =Trials m +1.

[0090] 5. Scout bee stage

[0091] (1) Judgment abandonment condition formula

[0092] For each UAV i, set a threshold limit. When Trials i ≥limit, it is considered that the position of this UAV has not been improved for a long time, the abandonment condition is met, and a new position needs to be found.

[0093] (2) Re-initialization position formula

[0094] For the UAV i that meets the abandonment condition, randomly generate position coordinates again within the glass curtain wall area. The formula is similar to the position initialization in the initialization stage:

[0095]

[0096] Then recalculate its fitness value F(x i ,y i ), and reset the number of non-improved times to 0:

[0097] F(x i ,y i ) = k1×Coverage(x i ,y i ) - k2×Distance(x i ,y i ) + k3

[0098] Trials i = 0。

[0099] 6. Algorithm Improvement and Optimization

[0100] (1) Introduce perturbation control frequency and global factor:

[0101] To accelerate the algorithm convergence, a strategy based on reverse learning can be adopted in the track initialization stage, introducing perturbation control frequency and global factor to balance the global search performance and local exploitation performance of the artificial bee colony.

[0102] For example, perturbation control frequency and global factor can be introduced into the search formula in the employed bee stage, such as v ij = x ij + φ ij (x ij - x kj ) + ω * global_factor, where ω is the perturbation control frequency and global_factor is the global factor. By adjusting these two parameters, the global search and local search capabilities of the algorithm can be balanced, improving the performance of the algorithm.

[0103] (2) Method of regenerating nectar sources:

[0104] In the scout bee stage, a method of regenerating nectar sources can be defined to improve the search efficiency of the algorithm.

[0105] For example, according to the characteristics of the problem and prior knowledge, a specific method can be designed to generate the location of the new food source. For example, in the task of cleaning the glass curtain wall of high-rise buildings, a new path plan can be generated according to factors such as the shape and size of the glass curtain wall.

[0106] Application of UAV Swarms in Cleaning Glass Curtain Walls of High-rise Buildings

[0107] (1) Path planning:

[0108] In the task of cleaning the curtain wall of high-rise building slopes, the drone needs to plan an efficient cleaning path. The cleaning path can be represented as a series of coordinate points, and each coordinate point corresponds to a position on the glass curtain wall. When using the artificial bee colony algorithm for path planning, each food source can be represented as a path plan, and the fitness value can be determined according to factors such as the length of the path, cleaning time, and energy consumption. Through continuous iterative search, the optimal path plan is found.

[0109] For example, the fitness function can be defined as: where length is the path length, time is the cleaning time, energy is the energy consumption, and α and β are weight coefficients used to adjust the importance of different factors.

[0110] (2) Task allocation:

[0111] For a drone swarm, it is necessary to reasonably allocate cleaning tasks to improve cleaning efficiency. The task allocation problem can be transformed into an optimization problem and solved using the artificial bee colony algorithm.

[0112] Each food source can be represented as a task allocation plan, and the fitness value can be determined according to factors such as the task completion time and the balance among drones).

[0113] For example, the fitness function can be defined as: where total_time is the total time for all drones to complete the tasks, balance is the balance of task allocation among drones, and γ is the weight coefficient.

[0114] (3) Cooperative control:

[0115] During the process of a drone swarm cleaning the glass curtain wall of a high-rise building, it is necessary to ensure the cooperative work among drones and avoid collisions. The artificial bee colony algorithm can be used for cooperative control.

[0116] Each food source can be represented as a cooperative control plan, and the fitness value can be determined according to factors such as the distance between drones and the collision risk.

[0117] For example, the fitness function can be defined as: where distance is the average distance between drones, collision_risk is the collision risk, and δ is the weight coefficient.

[0118] Example 3

[0119] I. Detailed implementation process of the simulation experiment:

[0120] 1. Simulation environment and tools

[0121] (1) Simulation platform: A three-dimensional simulation environment is built using MATLAB / Simulink, integrating the dynamic model of the unmanned aerial vehicle, the geometric model of the curtain wall, and the environmental interference model (such as wind speed disturbance).

[0122] (2) Algorithm comparison: The genetic algorithm (GA) and the particle swarm optimization algorithm (PSO) are selected as the comparison benchmarks to verify the superiority of the artificial bee colony algorithm (ABC).

[0123] (3) Parameter settings:

[0124] Number of unmanned aerial vehicles: 10 (5 in the spraying group and 5 in the scrubbing group).

[0125] Curtain wall size: 50m × 40m (Lx = 50, Ly = 40).

[0126] Parameters of the artificial bee colony algorithm: The proportion of employed bees is 50%, the proportion of observing bees is 50%, the threshold of scouting bees limit = 10, and the weight coefficients are k1 = 0.5, k2 = 0.3, and k3 = 0.2.

[0127] Parameters of the comparison algorithms: The GA crossover rate is 0.8 and the mutation rate is 0.1; the PSO inertia weight is 0.6, the learning factor c1 = 1.5, and c2 = 1.7.

[0128] 2. Simulation process

[0129] (1) Scene modeling: The curtain wall is divided into grids of 1m × 1m to simulate the stain distribution (randomly generate a 10% polluted area).

[0130] (2) Path generation: The ABC algorithm initializes the positions of the unmanned aerial vehicles and generates a path covering the entire curtain wall through iterative optimization ( Figure 1 ).

[0131] (3) Dynamic adjustment: Introduce random wind speed disturbance (0 - 5m / s) to trigger real-time path correction of the unmanned aerial vehicle.

[0132] (4) Performance evaluation: Record the algorithm convergence time, total path length, energy consumption (mAh), and cleaning coverage rate.

[0133] 3. Simulation results and analysis

[0134] Index ABC algorithm GA algorithm PSO algorithm Path length (km) 3.8 4.5 4.2 Convergence time (s) 120 200 180 Energy consumption (mAh) 1850 2200 2100 Cleaning coverage rate 99.2％ 95.4％ 96.8％

[0135] Conclusion: The ABC algorithm performs optimally in path optimization and energy consumption control, with the cleaning coverage rate increased by 3% - 4% and the dynamic adjustment response time shortened by more than 40%.

[0136] II. Test environment and equipment configuration

[0137] 1. Test Building: A commercial building with a curtain wall height of 60m and an area of 2,500 ㎡, with oil stains and dust distributed on the surface.

[0138] 2. Drone Models:

[0139] (1) Spraying Group: DJIAgras T30 modified version, with a 5kg water tank and a spraying radius of 3m.

[0140] (2) Scrubbing Group: Customized quadcopter drone equipped with a rotating scrubbing head with adjustable pressure.

[0141] 3. Sensor System:

[0142] (1) RTK-GPS positioning (accuracy ±2cm), lidar obstacle avoidance, and 4K camera for real-time monitoring.

[0143] (2) Control Center: ROS-based distributed control system, supporting multi-robot collaboration and data fusion.

[0144] 4. Implementation Process

[0145] (1) Task Zoning: Divide the curtain wall into 5 vertical areas, with each area responsible for 1 spraying + 1 scrubbing drone.

[0146] (2) Path Planning: The ABC algorithm generates a spiral ascending path. The spraying group sprays first, and the scrubbing group starts working 5 minutes later.

[0147] (3) Dynamic Adjustment Cases:

[0148] Sudden Wind Speed: The 3rd area encounters a gust (8m / s), and the drone automatically switches to the obstacle avoidance mode, and the path is adjusted to an "S" shape for detouring.

[0149] Insufficient Cleaning Liquid: The liquid volume of the 3rd spraying drone is lower than 20%, and the system schedules a backup drone to take over, ensuring seamless task connection.

[0150] (4) Data Collection: Record flight trajectories, cleaning liquid consumption, scrubbing pressure, and environmental parameters (temperature, humidity, wind speed).

[0151] 5. Measured Data

[0152] Index Measured result Total cleaning time 40 minutes Average energy consumption per drone 2100 mAh Cleaning liquid consumption 18L Detergency 97.5％ Path overlap rate <2％

[0153] Effect Comparison: Compared with traditional manual cleaning using hanging baskets (taking 3 hours with a decontamination rate of 82%), the efficiency is increased by 75%, and 40% of the cleaning liquid is saved.

[0154] III. In-depth Description of Algorithm Verification and Analysis

[0155] 1. Parameter Tuning Process

[0156] (1) Determination of weight coefficients: Through the orthogonal experiment method, the combinations of k1 (coverage rate), k2 (distance), and k3 (efficiency) are tested, and finally k1 = 0.5, k2 = 0.3, and k3 = 0.2 are selected to balance the cleaning quality and energy consumption.

[0157] (2) Optimization of the scout bee threshold: Experiments show that when limit = 10, the global search ability of the algorithm is the best, avoiding premature convergence to local optima.

[0158] 2. Dynamic adjustment strategy

[0159] Collision avoidance logic: When the distance between UAVs < 2m, the repulsion formula is triggered: (d ij is the distance between UAVs) to adjust the flight direction in real time to ensure a safe distance.

[0160] 3. Robustness test

[0161] (1) Extreme environment: Under the conditions of a wind speed of 10m / s and a humidity of 90%, the ABC algorithm still maintains a cleaning coverage rate of 85%, while the GA / PSO algorithm drops below 70%.

[0162] (2) Equipment failure simulation: Randomly shut down 1 UAV, and the system reallocates tasks within 30 seconds, with the overall efficiency loss < 5%.

[0163] IV. Analysis of experimental results

[0164] 1. Cleaning efficiency

[0165] The AI-enhanced UAV swarm system has significant advantages in cleaning efficiency. Compared with the manual cleaning method, it can not only complete tasks faster (35 minutes compared with 2 hours of manual cleaning), but also greatly improve the cleaning degree, with a decontamination rate reaching 98%.

[0166] 2. Energy efficiency performance

[0167] The energy consumption of the UAV system is relatively low. The energy consumption of each UAV is 2000mAh, which is more energy-efficient than the 3000mAh of manual equipment, showing obvious advantages for long-term use.

[0168] 3. Path optimization and task allocation

[0169] The path planning and task allocation of the artificial bee colony algorithm effectively avoid the overlap of flight paths and the repetition of operations, enabling each UAV to efficiently complete its respective tasks and dynamically adjust the path according to real-time feedback to adapt to changing environmental factors.

[0170] 4. Operation coordination and safety

[0171] Compared with the manual cleaning method, the AI-enhanced drone swarm system has higher operation coordination. Drones can share operation progress and environmental data in real time, avoiding the problems of repeated cleaning areas and omissions that may occur during manual cleaning.

[0172] In addition, the drone cleaning system avoids the safety risks of high-altitude operations and has higher safety.

[0173] 5. Cleaning liquid usage

[0174] The cleaning liquid consumption of the spray group drone system is relatively small, only 50% of that of manual cleaning, which reflects its advantages in operation efficiency and resource utilization.

[0175] V. Conclusion

[0176] 1. Technical feasibility

[0177] The experiment verified the feasibility of the AI-enhanced autonomous drone swarm described in the present invention in the cleaning of high-rise building glass curtain walls. Through the optimization of the artificial bee colony algorithm, the drone swarm can efficiently cooperate to complete complex cleaning tasks.

[0178] 2. Efficiency improvement

[0179] Compared with the traditional manual cleaning method, the cleaning time is greatly shortened (from 2 hours to 35 minutes), the cleaning liquid consumption is reduced, and the operation efficiency is significantly improved.

[0180] 3. Cleaning effect

[0181] Through the analysis of experimental data, the drone swarm can achieve efficient cleaning effects, and the stain removal rate on the glass surface reaches more than 98%, meeting the cleaning standards of high-rise building glass curtain walls.

[0182] 4. Safety and economy

[0183] The operation of the drone swarm reduces the safety risks of manual operations and reduces potential accidents in high-altitude operations. At the same time, by optimizing task allocation and path planning, the energy consumption and operation cost are significantly reduced.

[0184] 5. Application prospects

[0185] The AI-enhanced drone swarm collaboration of the present invention is not only applicable to the cleaning of high-rise building glass curtain walls, but also can be extended to other fields such as the maintenance of building facades, the cleaning of photovoltaic panels, and urban environmental monitoring, with broad market prospects and application values.

Claims

1. A path optimization method for cleaning a drone swarm, characterized in that, It includes the following steps: S1: First, initialize the positions of the drones: Select N drones to participate in the task in the mission area, and the position of each drone is initialized by generating random numbers uniformly distributed within the corresponding interval; next, initialize the fitness function; finally, set a variable Trials for each drone to record the number of times without improvement; S2: In the employed bee stage, each drone searches for a new position within the neighborhood of its current position; In the observing bee stage, drones are selected by calculating the selection probability and using the roulette wheel selection method, followed by neighborhood search and boundary processing to update the position and fitness value; in the scout bee stage, the position is reinitialized by judging the abandonment condition, and the positions of the drones are reallocated; S3: In the track initialization stage, adopt a strategy based on reverse learning, introduce the disturbance control frequency and the global factor; in the scout bee stage, define a way to regenerate the nectar source.

2. The path optimization method for cleaning a drone swarm according to claim 1, wherein, In S1, the two-dimensional plane area where the task area is located ranges from [0, L x in the x-axis direction and from [0, L y in the y-axis direction, where L x and L y respectively represent the dimension range values such as the side lengths of the curtain wall in the corresponding directions.

3. The path optimization method for a swarm of cleaning drones according to claim 1, wherein In the above S1, the initial value of the variable Trials is 0.

4. Application of the cleaning drone path optimization method according to any one of claims 1-3 in cleaning the glass curtain wall of high-rise buildings.

Images