Adaptive control distribution method for multiple unmanned aerial vehicles based on mobile nest

Abstract

The invention discloses a multi-unmanned aerial vehicle adaptive control distribution method based on a mobile nest, and belongs to the technical field of unmanned aerial vehicle control, and the method comprises the following steps: S1, information collection and fusion, S2, dynamic task priority evaluation, S3, task matching, S4, cooperative control strategy generation, S5, detection and adjustment, and S6, management and optimization. By constructing a system architecture with a mobile nest as a core and combining an adaptive task allocation algorithm and reinforcement learning cooperative control, efficient cooperation of multiple unmanned aerial vehicles in a complex scene can be realized, and the mobile nest senses task requirements and environmental changes in real time, dynamically allocates tasks and adjusts a control strategy by virtue of strong calculation and communication capabilities, so that the system is more efficient and efficient. The unmanned aerial vehicle resources are accurately matched, and idling or overload is avoided.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) control technology, and more specifically, to an adaptive control allocation method for multiple UAVs based on a mobile nest. Background Technology

[0002] Mobile drone nests are mobile devices that provide autonomous take-off and landing, charging, and mission execution support for drones. They are mainly used in scenarios requiring rapid deployment, such as urban security and power grid security.

[0003] The prior art patent document with authorization announcement number CN118276600B discloses "A UAV Inspection System and Method Based on Mobile Nests", which includes: a remote control platform, a UAV, and two mobile nests, namely a first mobile nest and a second mobile nest. It also includes an alternating control module and a take-off and landing arrangement module. Compared with fixed nests, the present invention uses mobile nests to control the UAV, which can greatly reduce the limitation on the inspection range of the UAV caused by the inability of the nest to move, thereby expanding the inspection route of the UAV to the range that the mobile nests can reach.

[0004] The patent document with authorization announcement number CN116301059A discloses "A mobile drone nest and drone takeoff control method", including: a takeoff and landing platform and an electromagnetic adsorption component and magnetic feet located on the takeoff and landing platform; the magnetic feet are used to connect with the drone and are used to move in cooperation with the electromagnetic adsorption component. The method includes: determining multiple unobstructed sub-regions based on image data of a set area above the drone; obtaining multiple verified unobstructed sub-regions based on ultrasonic data of the multiple unobstructed sub-regions; obtaining the projection points of the center points of the multiple verified unobstructed sub-regions on the takeoff and landing platform, and taking the area where the projection point with the smallest distance from the current center point of the drone is located as the takeoff area; and moving the drone to the takeoff area by cooperating with the electromagnetic adsorption component and the magnetic feet to control the drone to take off.

[0005] While existing technologies can enable UAVs to operate according to the initial plan and respond to new targets to a certain extent through specific algorithms and mechanisms, such as adjusting the set of operational targets based on UAV feedback and optimizing decisions using fusion probability enhancement algorithms and specific models, the centralized control in existing technologies is prone to information interaction conflicts and algorithm complexity, while the decentralized control lacks effective interaction and has poor control performance. Furthermore, task allocation relies on static parameters, making it difficult to adapt to environmental changes in real time and resulting in unreasonable resource utilization. At the same time, the fixed UAV collaboration method makes it difficult to flexibly respond to sudden situations in dynamic environments, and communication is easily interrupted by interference. Summary of the Invention

[0006] This invention mainly provides an adaptive control and allocation method for multiple UAVs based on a mobile nest, which can solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: an adaptive control and allocation method for multiple UAVs based on a mobile nest, comprising: S1. The mobile drone nest uses a communication link and employs sensor data acquisition technology and multi-source data fusion algorithms to collect drone status, environmental dynamics data and mission requirements, and processes them in a unified manner to prepare for control allocation. S2. A deep neural network model based on machine learning, combined with real-time task characteristics and environmental changes, dynamically calculates task priorities and outputs a hierarchical task sequence. S3. The mobile nest constructs a resource status matrix and a task requirement matrix. It uses resource status modeling technology and the Hungarian algorithm to dynamically match tasks and UAV resources and generate an initial allocation scheme. S4. Based on a deep Q-network reinforcement learning framework, the UAV uses a global environment model constructed by multi-sensor fusion and octree map algorithm, combined with state perception information, to autonomously make flight decisions and optimize collaborative strategies in homing. S5. Employs real-time monitoring technology and an event-driven adjustment mechanism to monitor task execution and UAV status in real time, triggering the event-driven adjustment mechanism to ensure task continuity; S6. After the mission is completed, the drone returns to the mobile nest. The nest uses automatic charging technology and fault diagnosis algorithms to perform energy replenishment and data review, and optimizes the control strategy for the next mission.

[0008] Furthermore, in S1, the sensor data acquisition technology is implemented through the lidar, inertial measurement unit and GPS carried by the UAV, which respectively collect obstacle distance and position, UAV attitude acceleration and UAV position information. The multi-source data fusion algorithm adopts the extended Kalman filter algorithm to improve the accuracy of state perception.

[0009] Furthermore, in S2, the deep neural network model for machine learning is specifically a long short-term memory network. The input features include the urgency, importance, and required resources of the task. In specific scenarios, the fire spread rate, the changing trend of the number of trapped people, and the importance of the task location are also included. The backpropagation algorithm is used to adjust the parameters to minimize the priority prediction error, and finally outputs a hierarchical task sequence sorted by priority.

[0010] Furthermore, in S3, the resource status modeling technology uses the Kalman filter algorithm to estimate and predict the power consumption of the UAV in real time. The resource status matrix records the current power and load capacity of each UAV, and the task requirement matrix clarifies the power and load required for each task. The optimal match is found in the two matrices through the Hungarian algorithm to generate an initial allocation scheme that adapts resources to tasks, thereby avoiding resource overload or idleness.

[0011] Furthermore, in S4, the global environment model is constructed using an octree map algorithm, which divides the three-dimensional space into cubic units and marks the state of obstacle occupancy. The state space of the deep Q-network reinforcement learning framework contains the UAV's own state and global environment information, and the action space includes flight control actions such as acceleration, deceleration, turning, ascent, and descent.

[0012] Furthermore, in S5, the real-time monitoring technology obtains the drone status and task execution progress in real time through the communication link between the mobile nest and the drone. The event-driven adjustment mechanism establishes an event library containing "task cancellation, task addition, priority change, and drone failure". Each event corresponds to a specific adjustment strategy, and adjustment notifications are sent to all drones to update the task plan and path.

[0013] Furthermore, in S6, the automatic charging technology is combined with an intelligent charging management algorithm to dynamically adjust the charging current and voltage according to the battery type and power level. The fault diagnosis algorithm is based on a neural network fault diagnosis model, which judges potential faults by monitoring the speed, vibration and propeller wear of the drone motor. The data review and statistics of task execution efficiency and resource utilization data provide a basis for the next round of control strategy optimization.

[0014] Furthermore, the UAV communication and collaboration mechanism adopts a self-organizing network communication protocol to achieve real-time communication between UAVs, and the mobile UAV nests coordinate communication through time division multiple access or code division multiple access technology.

[0015] The beneficial effects of the adaptive control allocation method for multiple UAVs based on mobile nesting in this invention are as follows: By constructing a system architecture centered on a mobile drone nest, and combining adaptive task allocation algorithms with reinforcement learning collaborative control, efficient collaboration among multiple drones in complex scenarios can be achieved. With its powerful computing and communication capabilities, the mobile drone nest can perceive task requirements and environmental changes in real time, dynamically allocate tasks and adjust control strategies, ensuring precise matching of drone resources and avoiding idleness or overload. Whether facing sudden fires during fire rescue or complex terrain during urban inspections, it can respond quickly and collaborate flexibly, significantly improving task execution efficiency and adaptability, and breaking through the limitations of traditional control methods in information interaction and task adaptation.

[0016] Through the intelligent management function and dynamic adjustment mechanism of the mobile drone nest, a reliable guarantee is built for drone operations. The automatic energy replenishment and fault detection and maintenance of the nest can restore the drone's operational capability in a timely manner and reduce the impact of failures on the mission. During mission execution, event-driven dynamic adjustment can quickly respond to mission changes, drone failures and other situations, ensuring continuous mission execution. At the same time, enhanced learning helps drones to coordinate precisely. In scenarios such as formation flight and collaborative transport, precise action coordination is achieved, improving operational reliability and coordination accuracy, and solving the shortcomings of traditional technologies in resource guarantee, mission resilience and coordination effect. Attached Figure Description

[0017] The present invention will now be described in further detail with reference to the accompanying drawings and specific implementation methods.

[0018] Figure 1 This is a schematic diagram of the method flow for the adaptive control and allocation method for multiple UAVs based on a mobile nest, as described in this invention. Detailed Implementation

[0019] To make the technical solution of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Example 1

[0020] like Figure 1 As shown, a technical solution is provided: an adaptive control and allocation method for multiple UAVs based on a mobile nest, including: Step 1: Information Collection and Integration The mobile drone nest uses a communication link and employs sensor data acquisition technology and multi-source data fusion algorithms to collect drone status, environmental dynamics data, and mission requirements, and processes them in a unified manner to prepare for control allocation. Specifically, sensor data acquisition technology is achieved through lidar, inertial measurement unit and GPS carried by UAV, which respectively collect obstacle distance and position, UAV attitude acceleration and UAV position information. The multi-source data fusion algorithm adopts extended Kalman filter algorithm to improve the accuracy of state perception. Among them, the UAV communication and collaboration mechanism adopts a self-organizing network communication protocol (such as the MANET protocol for mobile ad hoc networks) to realize real-time communication between UAVs. The mobile UAV nest coordinates communication through time division multiple access (TDMA) or code division multiple access (CDMA) technology to ensure that various types of data collected by the sensors can be stably and in real-time transmitted to the mobile UAV nest, avoiding data loss or delay caused by communication interruption. Meanwhile, the lidar focuses on collecting three-dimensional distance and position information of obstacles around the drone's flight path, the inertial measurement unit (IMU) captures the drone's attitude angles (such as roll, pitch, and yaw) and linear acceleration data in real time, and the GPS accurately locates the drone's latitude, longitude, and altitude. The data from these three types of sensors complement each other, covering the core needs of the drone's own state and external environment perception. In addition, during the multi-source data fusion process, the Extended Kalman Filter (EKF) algorithm first preprocesses the raw data from each sensor to filter out noise caused by changes in ambient light in the lidar, cumulative drift error of the inertial measurement unit, and positioning jump caused by GPS signal blockage. Then, through state estimation and data fusion, it outputs unified and accurate comprehensive state information and environmental feature data of the UAV, providing a reliable data foundation for subsequent control allocation. Finally, after receiving data transmitted from multiple drones, the mobile data center classifies and integrates the data, classifying and integrating drone status data (battery level, load, current location, attitude), environmental dynamics data (obstacle distribution, terrain features), and task requirement data (task type, target area, required resources) into corresponding data modules. Abnormal data is eliminated through a data verification mechanism to ensure that the data used for subsequent control and allocation is authentic and valid. At the same time, the data after unified processing is standardized in format, transforming drone status, environmental dynamics, and task requirement data into structured data formats suitable for subsequent task priority evaluation and resource matching (for example, converting drone battery level into "remaining flight time" parameter, and obstacle location into "octree map coordinates").

[0021] Step 2: Dynamic Task Priority Assessment Based on a deep neural network model using machine learning, and combined with real-time task characteristics and environmental changes, the system dynamically calculates task priorities and outputs a hierarchical task sequence. Specifically, the deep neural network model used in machine learning is a Long Short-Term Memory (LSTM) network. The input features include the urgency and importance of the task, the resources required, and in specific scenarios, the fire spread rate, the changing trend of the number of trapped people, and the importance of the task location are also included. The backpropagation algorithm is used to adjust the parameters to minimize the priority prediction error, and the final output is a hierarchical task sequence sorted by priority. The urgency of a task is quantified by the "task deadline" and the "rate of environmental change". For example, in a fire rescue scenario, the "fire spread rate" is converted into the "fire coverage area increase per unit time". If this value exceeds a preset threshold (e.g., 50㎡ / min), the urgency of the corresponding task is increased by one level. The trend of the number of trapped people is calculated by the "increase in the number of trapped people every 10 minutes". The greater the increase, the higher the priority weight of the task, ensuring that life rescue tasks are executed first. Meanwhile, the importance of a task is determined based on the "value of the task objective". For example, in urban security scenarios, the "key business district inspection task" has a high importance weight (e.g., 0.8) because it involves a high density of people and high property value, while the "ordinary street inspection task" has a low weight (e.g., 0.4). The resources required for a task are comprehensively evaluated by "the number of drones required to complete the task", "the minimum flight time of a single drone", and "whether special equipment (e.g., infrared thermal imagers) is required". The scarcer the required resources, the more priority should be given to matching drones with sufficient idle resources when calculating the priority. When dynamically calculating priorities, the LSTM model receives the environmental dynamic data processed in step 1 in real time. If the environment changes abruptly (for example, a temporary obstacle suddenly appears in the originally unobstructed task area, or an emergency requirement is added to the task target area), the model will recalculate the task priority within 1 second (for example, the "suburban power tower inspection task" with a priority of 3 is immediately upgraded to 1 due to the sudden fire at the tower base, which activates the "fire spread speed" parameter in the environmental data). In addition, the output hierarchical task sequence adopts a "three-level, nine-tier" structure. Level 1 corresponds to "urgent and important" tasks (such as life rescue and troubleshooting major facility failures), Level 2 corresponds to "routine and critical" tasks (such as daily inspections of key areas), and Level 3 corresponds to "ordinary auxiliary" tasks (such as data verification and path exploration). Each level is further subdivided into high, medium, and low tiers (for example, high-level tasks in Level 1 need to be assigned to drones within 5 minutes, while medium-level tasks in Level 1 can be assigned within 10 minutes, ensuring that the task execution pace matches the urgency level). Finally, the mobile nest will perform a preliminary matching and verification between the hierarchical task sequence and the standardized drone resource data in step 1. If a certain level of task does not have corresponding resource support (for example, a first-level high-end task requires a drone equipped with an infrared thermal imager, but all drones of this type are currently busy), then "priority dynamic fine-tuning" will be triggered, temporarily reducing the task to the next higher level, and at the same time sending a resource shortage warning to the remote management platform to prevent the task from stalling due to lack of resources.

[0022] Step 3: Task Matching The mobile drone nest constructs a resource status matrix and a task requirement matrix. It uses resource status modeling technology and the Hungarian algorithm to dynamically match tasks and drone resources and generate an initial allocation scheme. Specifically, the resource status modeling technology uses the Kalman filter algorithm to estimate and predict the power consumption of drones in real time. The resource status matrix records the current power and load capacity of each drone, and the task requirement matrix clarifies the power and load required for each task. The Hungarian algorithm is used to find the optimal match between the two matrices to generate an initial allocation scheme that adapts resources to tasks, thus avoiding resource overload or idleness. The Kalman filter algorithm, when estimating and predicting drone power consumption, combines historical power data (such as power change curves over the past 30 minutes) and current flight status (such as flight speed, altitude, and payload weight) to build a power consumption model. For example, when a drone flies at 8 m / s carrying a 5 kg payload, the algorithm predicts its power consumption per kilometer based on historical data and further calculates the remaining battery power's support for flight time and distance, providing accurate power parameter support for the resource status matrix. Simultaneously, the resource status matrix uses "drone number" as the row index and, for example, "current battery power (remaining flight time in minutes), maximum payload weight (kg), current payload weight (kg), and executable task type (such as inspection / rescue / transportation)," quantifies the real-time resource status of each drone into specific values. For instance, the resource status of drone 1 is recorded as "25 minutes, 10 kg, 3 kg, inspection + rescue." The task requirement matrix uses "task number" as the row index and "minimum required endurance minutes, minimum required load weight, and required equipment type (e.g., LiDAR / infrared camera)" as columns to clearly define the resource threshold for each task. For example, the requirement record for task A (fire rescue sub-task) is "20 minutes, 5 kg, infrared camera". Furthermore, before using the Hungarian algorithm for task and resource matching, the mobile nest first performs compatibility screening on the two matrices, eliminating matching combinations where the resource status cannot meet the minimum requirements of the task (for example, excluding drones with a maximum load of only 3kg from the matching range for tasks requiring a 5kg load, thus reducing the computational load of the algorithm). During the algorithm's operation, it uses "maximizing resource utilization" and "minimizing task execution risk" as dual objective functions. Resource utilization is calculated using the ratio of "actual drone load / maximum load" and "actual drone range usage / remaining range," while task execution risk is calculated using the difference between "required range for the task - remaining drone range" and "maximum drone load - required load for the task." The algorithm prioritizes matching combinations with high resource utilization and low execution risk. (For example, for task A, if both drone 1 (remaining flight time 25 minutes, maximum payload 10kg) and drone 3 (remaining flight time 22 minutes, maximum payload 8kg) meet the requirements, the algorithm compares their resource utilization (drone 1: 5kg / 10kg=50%, 20 minutes / 25 minutes=80%; drone 3: 5kg / 8kg=62.5%, 20 minutes / 22 minutes≈90.9%) and execution risk (drone 1: flight time difference 5 minutes, payload difference 5kg; drone 3: flight time difference 2 minutes, payload difference 3kg), and ultimately selects drone 3 to match task A, achieving more efficient resource utilization.) After generating the initial allocation scheme, the mobile drone nest will also perform a feasibility check on the scheme, simulating resource fluctuations that may occur during task execution (for example, assuming that a drone consumes 10% more power due to increased wind resistance during a task, check whether the drone can still complete the assigned task). If the check finds that a certain matching combination has execution risks (for example, the remaining range of the drone after adjustment cannot cover the round-trip distance of the task), the Hungarian algorithm will be called again to find a suboptimal match among the remaining drones and tasks until all tasks are matched with feasible drone resources. At the same time, the initial allocation scheme will include information such as the "task-drone" correspondence, the expected take-off time of the drone to perform the task, the start and end points of the flight path, and key nodes of task execution (such as key detection points in inspection tasks and target location points in rescue tasks), and synchronize them to the onboard computer of the corresponding drone to provide clear instructions for the drone's subsequent task execution. Finally, the mobile drone nest will store the initial allocation plan in the task management database and mark the plan generation time, algorithm parameters used (such as the weight coefficient of the Hungarian algorithm), number of drones and tasks involved in the matching, etc. This makes it easy to trace the basis for the initial plan when adjustments are made during subsequent task execution. If new tasks are added or the drone resource status changes, the initial allocation plan will be used as the basic data and updated quickly through incremental calculation without the need for full matching again, thus improving the efficiency of task adjustment.

[0023] Step 4: Generation of collaborative control strategy Based on a reinforcement learning framework using deep Q-networks, the UAVs share a global environment model constructed using multi-sensor fusion and octree map algorithms, combined with state perception information, to autonomously make flight decisions and optimize collaborative strategies in homing. Among them, the global environment model is constructed using the octree map algorithm, which divides the three-dimensional space into cubic units and marks the obstacle occupancy status. The state space of the deep Q network reinforcement learning framework contains the UAV's own state and global environment information, and the action space includes flight control actions such as acceleration, deceleration, turning, ascent, and descent. The core of the Deep Q-Network (DQN) reinforcement learning framework lies in guiding the drone to optimize action decisions through a reward function. The reward function design combines the task objective and execution rules. For example, a positive reward (e.g., +10 points) is given when the drone successfully avoids obstacles and maintains a safe formation distance, while a negative reward (e.g., -20 points) is given when the drone exhibits abnormal flight attitude, gets too close to other drones, or does not fly along the task route. A final positive reward (e.g., +100 points) is given when the task objective is achieved. This drives the drone to continuously learn the optimal action strategy during training and execution. Furthermore, when constructing the global environment model, the mobile nest continuously receives environmental data acquired by each UAV through multi-sensor fusion (e.g., fusion of LiDAR, inertial measurement unit, and GPS data using extended Kalman filter algorithm), and updates the obstacle occupancy status of the cube cells in the octree map in real time. If a cell was originally marked as "empty," but a UAV newly detects a temporary obstacle in that area (e.g., a sudden building debris fall), the mobile nest immediately updates the cell to "occupied" and synchronizes it to the global environment model of all UAVs, ensuring that the environmental information on which the UAV makes decisions is consistent with reality. At the same time, before autonomously deciding on flight actions, the UAV first extracts environmental information within a certain range (e.g., 50 meters) around itself from the global environment model, including obstacle distribution, the real-time position and flight direction of other cooperating UAVs, and combines it with its current position, attitude, speed, and other state information, inputting it into the policy network of the deep Q-network. The policy network then outputs the optimal action for the current state (e.g., "turn right 15° + ascend 2 meters"). Finally, the mobile nest will synchronously monitor the action decisions and execution of all drones. If it finds that the actions of multiple drones conflict (for example, two drones may arrive in the same airspace at the same time), the nest will fine-tune the action strategies of some drones based on the global collaborative goal. For example, by sending a command to one of the drones to "delay the turning action by 2 seconds" to avoid airspace conflict, the adjustment logic will be fed back to the training process of the deep Q network to optimize the coordination of subsequent autonomous decision-making. This ensures that multiple drones can autonomously and flexibly respond to local environmental changes in complex environments (such as mountain rescue and dense urban building inspection) while maintaining overall collaborative efficiency.

[0024] Step 5: Testing and Adjustment Employing real-time monitoring technology and an event-driven adjustment mechanism, the system monitors mission execution and UAV status in real time, triggering the event-driven adjustment mechanism to ensure mission continuity. Specifically, the real-time monitoring technology obtains the drone status and mission execution progress in real time through the communication link between the mobile drone nest and the drone. The event-driven adjustment mechanism establishes an event library containing "mission cancellation, mission addition, priority change, and drone failure". Each event corresponds to a specific adjustment strategy and sends adjustment notifications to all drones to update the mission plan and path. The core data monitored in real time includes the drone's real-time battery level, remaining flight time, current load, flight attitude stability, and mission progress (e.g., the number of target points covered in an inspection mission, the area searched in a rescue mission). The mobile pod receives a status data packet uploaded by the drone every 0.5 seconds and analyzes the data to determine if there are any anomalies. For example, when the battery consumption rate of a drone suddenly exceeds the historical average by 15%, the system will mark the drone as "in a state requiring attention" and increase the battery monitoring frequency to once every 0.2 seconds. At the same time, it will assess whether the drone can complete the current mission in conjunction with the mission progress. For different types of events in the event database, the dedicated adjustment strategy will specify the specific execution process: If a "drone malfunction" event is triggered, the mobile nest will first obtain the precise location of the malfunctioning drone and the completion rate of the executed tasks (e.g., 70% of the inspection route has been completed) through the communication link. Then, the resource information of the drone will be removed from the resource-task matching matrix, and its unfinished tasks (e.g., the remaining 30% of the inspection route) will be added back to the task pool. If a "task addition" event is triggered (e.g., a sudden new rescue need), the LSTM model in step 2 will be called first to evaluate the priority of the new task. Then, based on the current drone resource status, the Hungarian algorithm will be used to quickly match a suitable drone to ensure that the new task is started within the time window corresponding to the priority. Furthermore, during incident handling, the mobile drone nest will send adjustment notifications to all affected drones via Time Division Multiple Access (TDMA) technology. These notifications include updated mission lists, flight path coordinates (e.g., coordinates of the replanned inspection route inflection points), and coordination requirements with other drones (e.g., adjusted formation spacing and rendezvous times). Upon receiving the notification, the drones will immediately update their onboard mission plans and synchronize the adjustment information with relevant collaborating drones via a self-organizing network communication protocol. For example, when a drone takes over a mission from a malfunctioning drone, it will send a "coordination adjustment request" to nearby drones responsible for missions in adjacent areas to synchronize flight speed and path, avoiding overlapping mission areas. In order to ensure the continuity of the adjusted task execution, the mobile nest will conduct short-term simulation verification of the adjustment plan. Taking the "drone failure" event as an example, after the task is reassigned, the system will simulate the flight path and task execution time of the newly assigned drone to verify whether it can cover the unfinished area of ​​the failed drone with the remaining power and without conflicting with the flight paths of other drones. If the simulation finds that there is a risk of path intersection (for example, two drones may enter the same 50-meter airspace in 1 minute), the flight speed of one of the drones will be finely adjusted (for example, from 8m / s to 6m / s) or the path turning point (for example, the turning point will be shifted by 10 meters) until the conflict is eliminated. Finally, after the event is processed, the mobile nest records information such as event type, trigger time, processing flow, and adjustment results (e.g., task delay duration, resource re-matching time). This data will serve as an important basis for subsequent data review. For example, it will calculate the average processing time for "drone malfunction" events. If the processing time exceeds a preset threshold (e.g., 5 minutes), it will analyze whether the problem is caused by insufficient number of available drones in the resource pool, providing direction for resource configuration optimization in the next round of tasks. At the same time, it will solidify the effective strategies in the processing flow into standard operation templates to improve the processing efficiency of similar events in the future.

[0025] Step 6: Management and Optimization After completing the mission, the drone returns to its mobile nest, where it uses automatic charging technology and fault diagnosis algorithms to replenish energy and perform data analysis to optimize control strategies for the next mission. Among them, the automatic charging technology is combined with the intelligent charging management algorithm to dynamically adjust the charging current and voltage according to the battery type and power. The fault diagnosis algorithm is based on the fault diagnosis model of the neural network. It judges potential faults by monitoring the speed, vibration and propeller wear of the drone motor. The data review and statistics of task execution efficiency and resource utilization data provide a basis for the next round of control strategy optimization. Furthermore, the neural network fault diagnosis model upon which the fault diagnosis algorithm relies needs to be trained first using a large amount of UAV fault sample data (such as abnormal motor speed data and vibration data caused by propeller wear). The mobile drone, through sensors installed on key parts of the UAV, collects real-time data on motor speed (e.g., normal range 1500-3000 rpm), motor vibration frequency (e.g., normal range 50-100 Hz), and propeller thickness wear (e.g., normal wear threshold ≤ 0.5 mm). This data is then input into the trained model, which, through comparative analysis, determines whether the UAV has potential faults—if a motor speed abnormality is detected... If the speed remains below 1400 rpm and the vibration frequency exceeds 120 Hz, it is determined to be a motor failure. If the propeller wear exceeds 0.5 mm, it is determined that the propeller needs to be replaced. For cases determined to be simple failures (such as propeller wear), the mobile pod will automatically deploy a spare propeller and matching repair tools, and the propeller will be disassembled and replaced by a robotic arm. For complex failures (such as internal motor failures), the pod will record detailed failure information (including the time of failure and failure parameter data). After the mission is completed, the drone will be taken back to the base for in-depth repair, and the failure data will be added to the model training sample library to continuously optimize the model's diagnostic accuracy.

[0026] In summary, the above six steps achieved full-process adaptive control and allocation of multiple UAVs centered on a mobile UAV nest: from building a precise data foundation in the information collection and fusion stage, to determining the task execution order through dynamic task priority assessment, to achieving optimal resource and task matching through task matching, to relying on reinforcement learning to generate collaborative control strategies to ensure flexible UAV operation, to combining real-time detection and adjustment mechanisms to deal with emergencies to maintain task continuity, and finally to completing energy replenishment, fault maintenance and strategy iteration through management and optimization, forming a closed loop of "data-driven - dynamic decision-making - collaborative execution - continuous optimization".

[0027] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for adaptive control and allocation of multiple UAVs based on a mobile nest, characterized in that: S1. The mobile drone nest uses a communication link and employs sensor data acquisition technology and multi-source data fusion algorithms to collect drone status, environmental dynamics data and mission requirements, and processes them in a unified manner to prepare for control allocation. S2. A deep neural network model based on machine learning, combined with real-time task characteristics and environmental changes, dynamically calculates task priorities and outputs a hierarchical task sequence. S3. The mobile nest constructs a resource status matrix and a task requirement matrix. It uses resource status modeling technology and the Hungarian algorithm to dynamically match tasks and UAV resources and generate an initial allocation scheme. S4. Based on a deep Q-network reinforcement learning framework, the UAV uses a global environment model constructed by multi-sensor fusion and octree map algorithm, combined with state perception information, to autonomously make flight decisions and optimize collaborative strategies in homing. S5. Employs real-time monitoring technology and an event-driven adjustment mechanism to monitor task execution and UAV status in real time, triggering the event-driven adjustment mechanism to ensure task continuity; S6. After the mission is completed, the drone returns to the mobile nest. The nest uses automatic charging technology and fault diagnosis algorithms to perform energy replenishment and data review, and optimizes the control strategy for the next mission.

2. The adaptive control and allocation method for multiple UAVs based on a mobile nest as described in claim 1, characterized in that: In S1, sensor data acquisition technology is achieved through the lidar, inertial measurement unit and GPS carried by the UAV, which respectively collect obstacle distance and position, UAV attitude acceleration and UAV position information. The multi-source data fusion algorithm adopts the extended Kalman filter algorithm to improve the accuracy of state perception.

3. The adaptive control and allocation method for multiple UAVs based on a mobile nest as described in claim 1, characterized in that: In S2, the deep neural network model for machine learning is specifically a long short-term memory network. The input features include the urgency and importance of the task, the resources required, and in specific scenarios, the fire spread rate, the changing trend of the number of trapped people, and the importance of the task location are also included. The backpropagation algorithm is used to adjust the parameters to minimize the priority prediction error, and finally outputs a hierarchical task sequence sorted by priority.

4. The adaptive control and allocation method for multiple UAVs based on a mobile nest as described in claim 1, characterized in that: In S3, the resource status modeling technology uses the Kalman filter algorithm to estimate and predict the power consumption of the UAV in real time. The resource status matrix records the current power and load capacity of each UAV, and the task requirement matrix clarifies the power and load required for each task. The optimal match is found in the two matrices through the Hungarian algorithm to generate an initial allocation scheme that adapts resources and tasks, avoiding resource overload or idleness.

5. The adaptive control and allocation method for multiple UAVs based on a mobile nest as described in claim 1, characterized in that: In S4, the global environment model is constructed using an octree map algorithm, which divides the three-dimensional space into cubic units and marks the state of obstacles. The state space of the deep Q-network reinforcement learning framework contains the UAV's own state and global environment information, and the action space includes flight control actions such as acceleration, deceleration, turning, ascent, and descent.

6. The adaptive control and allocation method for multiple UAVs based on a mobile nest as described in claim 1, characterized in that: In S5, the real-time monitoring technology obtains the drone status and task execution progress in real time through the communication link between the mobile nest and the drone. The event-driven adjustment mechanism establishes an event library containing "task cancellation, task addition, priority change, and drone failure". Each event corresponds to a specific adjustment strategy and sends adjustment notifications to all drones to update the task plan and path.

7. The adaptive control and allocation method for multiple UAVs based on a mobile nest as described in claim 1, characterized in that: In S6, the automatic charging technology is combined with an intelligent charging management algorithm to dynamically adjust the charging current and voltage according to the battery type and power level. The fault diagnosis algorithm is based on a neural network fault diagnosis model. It identifies potential faults by monitoring the speed, vibration and propeller wear of the drone motor. The data review and statistics of task execution efficiency and resource utilization provide a basis for the next round of control strategy optimization.

8. The adaptive control and allocation method for multiple UAVs based on a mobile nest as described in claim 1, characterized in that: The UAV communication and collaboration mechanism adopts a self-organizing network communication protocol to achieve real-time communication between UAVs, and the mobile UAV nests coordinate communication through time division multiple access or code division multiple access technology.

Images