Unmanned aerial vehicle low-altitude logistics distribution system and method based on AI path planning

Abstract

The application relates to the technical field of intelligent logistics, in particular to an unmanned aerial vehicle low-altitude logistics distribution system and method based on AI path planning, which comprises an unmanned aerial vehicle cluster module, a ground edge computing device cluster module, a central control module and a communication transmission module. The method comprises three core steps of macroscopic path planning, path dynamic adjustment and edge-terminal collaborative calculation, a mapping model between an unmanned aerial vehicle motor current waveform and a minimum energy consumption action instruction is constructed, a flight path is optimized, and the wind field interference resistance is improved; a dynamic edge computing device selection method is adopted, the optimal edge computing device best matching a task is effectively selected, and the problem of low task offloading efficiency caused by the single basis of traditional edge computing device selection is overcome.

Description

Technical Field

[0001] This invention relates to the field of smart logistics technology, and more specifically, to a drone low-altitude logistics delivery system and method based on AI path planning. Background Technology

[0002] As a core application scenario of the low-altitude economy, low-altitude logistics delivery has made significant progress in technological innovation and commercialization in recent years. With the popularization and development of AI technology, its development is also gradually integrating with AI technology, but some technologies still have some shortcomings:

[0003] Existing solutions to address the energy loss caused by low-altitude winds on drones generally employ the method of equipping them with wind sensors to monitor environmental data in real time and perform real-time calculations. This approach not only directly increases the payload weight of the drone and raises its own energy consumption, but is also limited by the latency in the process of collecting information in real time and performing complex calculations to generate scheduling instructions, affecting the timeliness of decision-making and failing to minimize the energy loss caused by wind fields.

[0004] Meanwhile, in existing technologies, the selection methods for edge devices in drone task offloading schemes are mostly general strategies (such as proximity principle and load balancing), lacking customized selection schemes for logistics scenarios. This results in a situation where even if edge computing devices for task offloading are selected according to general strategies, the results are not received for a long time.

[0005] To address the aforementioned shortcomings, this invention proposes an AI-based path planning-based drone low-altitude logistics delivery system and method, aiming to optimize and resolve the aforementioned technical deficiencies. Summary of the Invention

[0006] In view of the shortcomings of existing technologies, the purpose of this invention is to provide a drone low-altitude logistics delivery system and method based on AI path planning.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] The AI-based path planning-based drone low-altitude logistics delivery system includes a drone swarm module, a ground edge computing device swarm module, a central control module, and a communication transmission module.

[0009] Unmanned Aerial Vehicle (UAV) Swarm Module: Composed of multiple UAVs, each equipped with a data acquisition unit, flight control unit, mission payload unit, and wireless communication unit;

[0010] Ground edge computing device cluster module: It consists of edge computing devices distributed in different geographical locations. Each edge computing device has a computing processing unit, a storage unit and a communication interface unit.

[0011] Central control module: The core control hub of the entire system, which includes a task scheduling unit, an equipment management unit, and a data processing unit;

[0012] Communication transmission module: Establishes communication links between the UAV and the central control module, the UAV and the ground edge computing device cluster module, and the central control module and the ground edge computing device cluster module;

[0013] A drone-based low-altitude logistics delivery method based on AI path planning, the process of which is as follows:

[0014] Step 1: Collection and classification of delivery tasks.

[0015] Delivery task collection: Collect relevant information about delivery tasks through online order platforms, customer self-ordering applications, or logistics outlets, and transmit it directly to the task scheduling unit of the central control module through an interface;

[0016] Delivery task classification: Store delivery tasks according to timeliness, geographical location, and cargo characteristics to establish a comprehensive task classification database;

[0017] Step 2: Macro-path planning.

[0018] Preliminary path planning is performed using the traditional path planning algorithm A*.

[0019] Step 3: Drone resource scheduling.

[0020] S31. Define the basic variables related to UAV and mission characteristics;

[0021] S32. Construct a dynamic weight model;

[0022] S33, Computational Task Suitability The formula is as follows:

[0023] ;

[0024] in, The weighting coefficients correspond to the drone status, drone health, historical performance, mission urgency, mission spatial complexity, and cargo risk, respectively. These represent the drone's status, drone health, historical performance, urgency level, spatial complexity, and cargo risk level, respectively.

[0025] Select the optimal drone to perform the mission based on F;

[0026] Step 4: Dynamically adjust the path.

[0027] Data acquisition: Build a simulation experimental platform to acquire flight data of UAVs under the optimal energy consumption strategy and store it in a database;

[0028] Mapping relationship construction: Using deep learning algorithms, a mapping relationship model between motor current waveform and the minimum energy consumption action command of the UAV is established, while the electrode current waveform is correlated with the environmental data of the UAV through data acquisition;

[0029] Real-time waveform matching: During the drone's flight, the collected motor current waveform is matched with existing waveform data in the database;

[0030] Waveform trend prediction: For successfully matched template sequences, obtain their corresponding environmental information and subsequent waveform data. Based on time series prediction algorithms, predict the motor current waveform trend over a future period. Using the predicted waveform data, infer the environment (wind field) the drone may face and the action commands it will take, and directly execute the predicted action commands in special scenarios.

[0031] Real-time path optimization: The flight path is optimized in real time by monitoring the cosine similarity between the UAV motor current and the template sequence.

[0032] Step 5: Edge-Terminal Collaborative Computing.

[0033] By leveraging edge computing technology and communication transmission modules, computational tasks are offloaded to edge devices closer to the drone, improving task execution efficiency. A multi-dimensional evaluation index system is constructed, and a reinforcement learning decision model is used to dynamically select the optimal edge computing device.

[0034] Furthermore, in step three, the basic variables related to the drone and mission characteristics include drone status, drone health, historical performance, and mission characteristics, among which mission characteristics include urgency, spatial complexity, and cargo risk.

[0035] Furthermore, in step four, the flight data of the UAV under the optimal energy consumption strategy is obtained, and the specific method is as follows:

[0036] A simulation platform was built to construct flight environments with different wind layers, wind forces, and wind directions. Based on a reinforcement learning framework, a policy network was used to directly output UAV control commands. With the goal of minimizing energy consumption, a value network was used to evaluate the state value to optimize the policy, thereby generating an energy-optimal flight path. During the experiment, wind field data and motor current waveform data around the UAV were collected simultaneously, and continuous UAV action commands were marked on the current waveforms as action nodes. While storing the above data in a database, the database was continuously supplemented and improved using actual flight data.

[0037] Furthermore, in step four, the acquired motor current waveform is matched with existing waveform data in the database. The specific method is as follows:

[0038] A matching method combining Dynamic Time Warping (DTW) algorithm and cosine similarity calculation is adopted. Real-time motor current waveform data collected during UAV flight is used as the query sequence, and existing waveform data in the database is used as the template sequence. The DTW algorithm is used to calculate the optimal alignment between the query sequence and the template sequence on the time axis. Based on this, the cosine similarity between the aligned two waveform sequences is calculated. The closer the cosine similarity is to 1, the more similar the shapes and features of the two waveforms are. A similarity threshold is set; when the cosine similarity between the query sequence and a certain template sequence exceeds the threshold, the two are considered to be successfully matched.

[0039] Furthermore, in step four, the flight path is optimized in real time, and the specific process is as follows:

[0040] The cosine similarity between the actual UAV motor current waveform and the matched template sequence is monitored in real time. When the similarity is continuously higher than the set threshold and the proportion of similar sequences reaches the similarity threshold, the control command of the target action node is called in advance to adjust the UAV flight trajectory and avoid potential adverse environments (strong wind environment or cross wind disturbance, etc.). If the calculated cosine similarity is lower than the threshold before reaching the next action node, the template sequence re-matching mechanism is triggered to re-search and match the appropriate template sequence.

[0041] Furthermore, in step five, a multi-dimensional evaluation indicator system is constructed, with the specific steps as follows:

[0042] S1. The computing task adaptation dimension includes the computing timeliness requirement score, computing data volume score, and data transmission volume score.

[0043] S2, the device resource dimension includes computing resource score, communication quality score, and device stability score.

[0044] Furthermore, in step five, the optimal edge computing device is dynamically selected, and the specific steps are as follows:

[0045] S1. Construct a dynamic weight matrix;

[0046] S2. Dynamically adjust the weights based on actual task requirements;

[0047] S3. Use a reinforcement learning decision model to select the optimal edge computing device.

[0048] Furthermore, in step five, a reinforcement learning decision model is used to select the optimal edge computing device. The specific steps are as follows:

[0049] S1. Define the state space;

[0050] S2, Define the action space;

[0051] S3. Design the reward function;

[0052] S4. Iterative optimization;

[0053] S5. Output the optimal solution.

[0054] Compared with the prior art, the present invention has the following beneficial effects:

[0055] 1. Significant energy consumption optimization: By constructing a mapping relationship between motor current waveform and action commands with the lowest energy consumption, not only can the increased load on the drone due to the wind sensor be reduced, but also the wind field can be avoided in advance by real-time prediction, or the computational complexity can be reduced by waveform matching, and the action commands can be directly called to react quickly and reduce the duration of wind field interference.

[0056] 2. By constructing a multi-dimensional edge device selection index system and adopting a dynamic selection method, combined with the scenario of low-altitude logistics delivery, we have deeply considered the adaptability of computing tasks and edge device resources, effectively avoiding the situation where traditional edge device selection strategies cannot effectively unload computing tasks. Attached Figure Description

[0057] Figure 1 This is a block diagram of a drone low-altitude logistics delivery system based on AI path planning.

[0058] Figure 2 A flowchart of a drone low-altitude logistics delivery method based on AI path planning;

[0059] Figure 3 This is a flowchart for selecting the optimal edge computing device in step five. Detailed Implementation

[0060] Example, refer to Figure 1 The UAV low-altitude logistics delivery system based on AI path planning in this embodiment includes a UAV swarm module, a ground edge computing device swarm module, a central control module, and a communication transmission module.

[0061] The drone swarm module consists of multiple drones, each equipped with a data acquisition unit, a flight control unit, a payload unit, and a wireless communication unit. The data acquisition unit collects real-time data on the drone's position, speed, attitude, remaining battery power, and other status information, as well as environmental and obstacle information. The flight control unit receives and processes multi-source data collected by the data acquisition unit, integrating AI path planning, real-time obstacle avoidance, and path optimization algorithms to control the drone's flight attitude, speed, and path. The payload unit loads the cargo to be delivered. The wireless communication unit is responsible for data interaction with the central control unit module and the ground edge computing device cluster module, uploading collected data and receiving task commands and calculation results.

[0062] Ground-based edge computing device cluster module: This module consists of multiple edge computing devices distributed across different geographical locations. Each edge computing device has a computing processing unit, a storage unit, and a communication interface unit. The computing processing unit is responsible for executing computing tasks offloaded from the UAV; the storage unit stores task data, intermediate computing results, and related algorithm models; and the communication interface unit enables data communication with the UAV and the central control module, ensuring stable data transmission.

[0063] Central Control Module: As the core control hub of the entire system, it includes a task scheduling unit, an equipment management unit, and a data processing unit. The task scheduling unit receives logistics delivery orders and assigns delivery tasks to each drone based on the status information of the drone swarm module. The equipment management unit monitors the resource status of each edge computing device in the ground edge computing device cluster module in real time, including CPU utilization, memory usage, remaining storage space, and communication link quality, and performs registration, deregistration, and maintenance management of the edge computing devices. The data processing unit is responsible for providing the computing power required by the task scheduling unit and the equipment management unit.

[0064] Communication transmission module: Constructs communication links between the UAV and the central control module, the UAV and the ground edge computing device cluster module, and the central control module and the ground edge computing device cluster module. It adopts wireless communication technologies such as 5G and Wi-Fi to ensure fast, stable and secure data transmission between the modules.

[0065] like Figure 2 As shown, the method for low-altitude logistics delivery by drones based on AI path planning is as follows:

[0066] Step 1: Collection and classification of delivery tasks.

[0067] Delivery task collection: Collect relevant information about delivery tasks through online order platforms, customer self-ordering applications, or logistics outlets, including shipping address, delivery address, cargo weight, dimensions, estimated delivery time, cargo type, etc., and transmit it directly to the task scheduling unit of the central control module through an interface;

[0068] Delivery task classification: Delivery tasks are classified and stored according to timeliness (urgent tasks, secondary urgent tasks, ordinary tasks), geographical tasks (high-frequency delivery areas, low-frequency delivery areas, cross-regional tasks), and cargo characteristics (fragile items, dangerous items, ordinary items). A comprehensive task classification database is established, and the task data can be easily queried, statistically analyzed, and processed through the database management system.

[0069] Step 2: Macro-path planning.

[0070] Initial path planning is performed using the traditional A* algorithm. First, the environment is modeled as a grid or graph structure, clearly defining the start point, end point, and obstacle locations. Next, an Open list (containing nodes to be expanded) and a Close list (containing expanded nodes) are initialized. The start point is added to the Open list, and its g-value (the actual cost from the start point to the current node) and h-value (the heuristically estimated cost from the current node to the end point) are calculated. The sum of these two values, f, is used for sorting. Each time, the node with the smallest f-value is removed from the Open list and expanded. It is then checked whether this is the end point; if so, the path is generated by backtracking the parent node; otherwise, adjacent passable nodes are traversed, and the g-value and h-value of the new node are calculated. If the new node is not in the Open list, it is added and its parent node is recorded; if it is already in the list, the g-value is compared for better value, and if better, the node information is updated. This process is repeated until the end point is found or it is confirmed that there is no solution. Finally, the optimal path from the start point to the end point is reconstructed using the parent node chain.

[0071] Step 3: Drone resource scheduling.

[0072] S31. Definition of basic variables:

[0073] Drone status defined Among them, the idle state Delivery status In this embodiment ;

[0074] Drone health is defined as ,in (Normalization processing), in this embodiment Hardware health is measured by years of use and number of logistics deliveries, software health is obtained by testing the same code running time, and battery life health is calculated by battery charging and discharging time.

[0075] Historical performance is defined as ,in For on-time rate, For task completion rate, The abnormality rate corresponds to the proportion of on-time delivery, completion, and number of abnormalities in recent (3 months in this example) delivery tasks.

[0076] Task characteristics are defined as ,in For urgency (within 30 minutes) =1, within 30 minutes to one hour =0.7, others =0.3), indicating the urgency of the task needing to be delivered within a certain timeframe. For space complexity (high-frequency region) =0.3, low frequency region =0.7, cross-regional =1), indicating the frequency of historical delivery tasks reaching the destination. Risk level of goods (ordinary goods) =0.2, fragile items =0.6, dangerous goods This indicates the risk attributes of the delivered goods;

[0077] S32, Dynamic Weight Model:

[0078] Basic weight matrix: , The weighting coefficients correspond to the drone's status, health, historical performance, mission urgency, mission spatial complexity, and cargo risk, respectively. This embodiment sets the weighting coefficients according to the principle of "status priority, health as a minimum, historical data as an aid, and mission adaptation." In practice, the weights are dynamically adjusted according to the task scenario; for example, the weights are increased in emergency task scenarios. Increased risk of hazardous materials scenarios ;

[0079] S33, Computational Task Suitability The formula is as follows:

[0080] ;

[0081] When selecting drones for a mission, prioritize Drones. If they exist The idle drones are selected from the group with the highest F value; otherwise, the drone with the highest F value and the earliest estimated completion time is selected from all drones. Finally, the task scheduling unit of the central control module sends the task to the corresponding drone through the communication transmission module.

[0082] Step 4: Dynamically adjust the path.

[0083] Data Acquisition: A simulation platform was built to construct flight environments with different wind layers, wind forces, and wind directions. Based on a reinforcement learning framework, a policy network was used to directly output UAV control commands. With the goal of minimizing energy consumption, a value network was used to evaluate state value to optimize the policy, thereby generating an energy-optimal flight path. During the experiment, wind field data and motor current waveform data around the UAV were collected simultaneously, and continuous UAV action commands were marked on the current waveforms as action nodes. While storing the above data in a database, the database was continuously supplemented and improved using actual flight data.

[0084] Mapping Relationship Construction: Using deep learning algorithms, the drone's motor current waveform data is used as input, and action commands are used as labels. These labels are associated with the drone's operational decisions in the given environment. Through deep training on a large number of data samples, a mapping relationship model between the motor current waveform and the drone's lowest energy consumption action commands is established. Meanwhile, the electrode current waveform is correlated with the environmental data of the drone through data acquisition.

[0085] Real-time waveform matching: A matching method combining Dynamic Time Warping (DTW) algorithm and cosine similarity calculation is adopted. The motor current waveform data collected in real-time by the data acquisition unit during UAV flight is used as the query sequence (the sequence length is determined by the decision efficiency required in the actual flight scenario; in this embodiment, the sequence length is a 0.5-second time span sequence). Existing waveform data in the database is used as the template sequence. The DTW algorithm is used to calculate the optimal alignment between the query sequence and the template sequence on the time axis, solving the problem of inconsistent waveform time scales caused by changes in UAV motor operating speed. Based on this, the cosine similarity between the two aligned waveform sequences is calculated. The closer the cosine similarity is to 1, the more similar the shapes and features of the two waveforms are. A similarity threshold is set (0.9 in this embodiment). When the cosine similarity between the query sequence and a certain template sequence exceeds the threshold, the two are considered to be successfully matched.

[0086] Waveform trend prediction: For successfully matched template sequences, obtain their corresponding environmental information and subsequent waveform data. Based on time series prediction algorithms (such as Long Short-Term Memory networks LSTM), predict the motor current waveform trend over a future period based on the matched historical waveform data. Since a mapping relationship has been established between the motor current waveform and the UAV's minimum energy consumption action commands, the predicted waveform data can be used to infer the environment (wind field) the UAV may face next and the action commands it will take. In special scenarios (scenarios particularly sensitive to wind fields), the predicted action commands can be executed directly, reducing the risk of interference.

[0087] Real-time path optimization: The cosine similarity between the actual UAV motor current waveform and the matched template sequence is monitored in real time. When the similarity is consistently higher than the set threshold and the proportion of similar sequences reaches the similarity threshold (95% in this embodiment), the control command of the target action node is called in advance. The flight trajectory of the UAV is adjusted through the UAV flight control unit to avoid potential adverse environments (strong wind environment or cross wind disturbance, etc.). If the calculated cosine similarity is lower than the threshold before reaching the next action node, the template sequence re-matching mechanism is triggered to re-search and match the appropriate template sequence.

[0088] Step 5: Edge-Terminal Collaborative Computing.

[0089] Logistics drones operate in complex and ever-changing environments. Tasks such as path planning and obstacle avoidance demand significant computing power. However, logistics drones have limited computing resources and capabilities. Adding an additional data processing module or chip increases logistics and delivery costs. Therefore, utilizing edge computing technology to offload computational tasks to edge devices via communication transmission modules can effectively reduce the computational load on drones and improve task execution efficiency. The optimal edge computing device is selected through the device management unit of the central control module, and the specific steps are as follows:

[0090] S1. Construct a multi-dimensional evaluation index system.

[0091] S11. The computing task adaptation dimension includes the computing timeliness requirement score, computing data volume score, and data transmission volume score.

[0092] S111, Calculate the timeliness requirement score: Let... The maximum processing time allowed for drones. The score S is the difference between the actual processing time and the ideal processing time. The formula for calculating S is:

[0093] ;

[0094] in, This means rounding up x, with a score range of 1 to 10.

[0095] S112. Calculate the amount of data required for the task: For the number of floating-point operations (FLOPs) required for the task, use the theoretical calculation method (calculate FLOPs from the formula level based on the expression) or the model pre-evaluation method (for a known machine learning model, the required FLOPs can be calculated based on the model structure) to obtain the number of floating-point operations required for the task.

[0096] Calculate the score To reflect the degree to which a task consumes the device's computing resources, the formula is as follows:

[0097] ;

[0098] in, This represents the floor function, with a score range of 1 to 10;

[0099] S113. Data Transmission Volume Scoring: Let the data volume be x (MB), then the score for this item is calculated as follows:

[0100] ;

[0101] Where x is obtained by calculating the average of the amount of data transmitted and the estimated amount of data returned;

[0102] S12, the device resource dimension includes computing resource score, communication quality score, and device stability score.

[0103] S121. Computing Resource Score: Based on CPU utilization, remaining memory, and current status, a computing resource score is calculated using the following formula. :

[0104] ;

[0105] in, This indicates CPU utilization, in percentage. This indicates memory usage, expressed as a percentage (%).

[0106] S122. Communication Quality Score: The score for this dimension is calculated based on the scoring formula and mapped to a score of 1 to 10, as detailed below:

[0107] ;

[0108] in, The weighting coefficients representing signal strength, transmission rate, and packet loss rate are used in this embodiment. ;

[0109] S123. Equipment Stability Rating: This rating assesses the overall failure probability of the equipment throughout its entire lifespan. and the duration of this run The equipment stability score is calculated using the following formula:

[0110] ;

[0111] S2, Dynamic selection of target equipment.

[0112] S21. Constructing a dynamic weight matrix In this embodiment, the initial values ​​are 0.3, 0.25, 0.15, 0.15, 0.1, and 0.05 respectively.

[0113] S22. Dynamically adjust weights when the task timeliness score is applied. hour, Increased to 0.4, Reduced to 0.2; if the device communication quality score , Increased to 0.2, Reduced to 0.1; for dangerous goods transportation tasks, Improved by 0.15; In practical applications, based on the actual needs generated, the dynamic weight adjustment rules are continuously improved and supplemented to make the model's selection of edge devices more in line with actual needs;

[0114] S23. Use a reinforcement learning decision model to select the optimal edge computing device, such as... Figure 3 As shown.

[0115] S231. State space definition: ;

[0116] S232, Action Space Definition: , which corresponds to the set of optional edge devices, where n is the total number of devices;

[0117] S233, Reward Function Design: ;

[0118] in, This embodiment represents the weight of consideration for processing time, computational load, and number of task errors. ; This refers to the actual processing time. This is the maximum allowed duration; For ideal computational load, This represents the actual amount of computation. The number of task errors. Total number of tasks;

[0119] S234, Algorithm Iteration Process:

[0120] 1. Initialize the Q-value table Q(S,A) to 0;

[0121] 2. For each task's unloading cycle:

[0122] 2.1. Collect the current status ;

[0123] 2.2 Action selection based on a greedy strategy ;

[0124] 2.3 Execution of Actions Get rewards With the next state ;

[0125] 2.4 Update the Q value. ;in, The learning rate is dynamically adjusted from 0.1 to 0.3. The discount factor is 0.9 in this embodiment; This indicates that the maximum value of action 'a' is taken for all possible actions in action space A;

[0126] S24. Reinforcement learning models use an "exploration-exploitation" balance mechanism. In the early stages, they need to perform tasks to accumulate experience. However, after the model matures, they mainly use the existing Q-value table to select the optimal edge computing device. Only when necessary (such as environmental changes) do they perform a small number of tasks to optimize the strategy.

[0127] Through the detailed description of the above embodiments, the AI-based path planning-based drone low-altitude logistics delivery system and method proposed in this invention demonstrate significant technical advantages: On the one hand, by constructing a mapping model between motor current waveforms and action commands with minimum energy consumption, it not only avoids the problem of increased drone load caused by carrying wind sensors, but also proactively avoids wind field influences through real-time prediction, or achieves rapid response by directly calling action commands through waveform matching, significantly shortening the duration of wind field interference; on the other hand, by constructing a multi-dimensional edge device selection index system, adopting a dynamic selection strategy and deeply integrating it with the low-altitude logistics delivery scenario, it fully considers the adaptability of computing tasks and edge device resources, effectively solving the industry pain point of low computing task unloading efficiency under the traditional general edge device selection strategy. This solution achieves systematic optimization of drone logistics delivery in terms of energy consumption control, anti-interference capability, and edge computing resource scheduling through technological innovation, providing a practical and forward-looking solution for the intelligent development of low-altitude logistics.

[0128] The above formulas are all dimensionless calculations, and the preset parameters in the formulas should be set by those skilled in the art according to the actual situation.

[0129] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0130] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0131] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0132] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0133] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0134] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0135] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A drone-based low-altitude logistics delivery system based on AI path planning, characterized in that, The system includes an unmanned aerial vehicle (UAV) swarm module, a ground edge computing device swarm module, a central control module, and a communication transmission module. Central control module: The core control hub of the entire system, which includes a task scheduling unit, an equipment management unit, and a data processing unit. It is responsible for assigning logistics and delivery tasks to UAVs and managing ground edge computing devices. Drone swarm module: Composed of multiple drones, it is scheduled by the task scheduling unit of the central control module to perform logistics and delivery tasks; Ground edge computing device cluster module: It consists of edge computing devices distributed in different geographical locations. The device management unit of the central control unit manages them uniformly and is responsible for offloading the UAV computing tasks. Communication transmission module: Establishes communication links between the UAV and the central control module, the UAV and the ground edge computing device cluster module, and the central control module and the ground edge computing device cluster module; A method for implementing an AI-based path planning-based drone low-altitude logistics delivery system, the method flow is as follows: Step 1: Collection and Classification of Delivery Tasks: Collect and classify delivery tasks generated in real time; Step 2, Macro-path planning: Using A The algorithm performs preliminary path planning; Step 3: Drone Resource Scheduling S31. Define the basic variables related to the drone and the mission; S32. Construct a dynamic weight model for each basic variable in S31; S33. Calculate the task suitability based on the basic variables and their corresponding weights, and select the optimal UAV to perform the task based on the task suitability. Step 4, Dynamic Path Adjustment: Establish a mapping model between the UAV motor current waveform and the minimum energy consumption action command; through motor current waveform matching, monitor the cosine similarity between the UAV motor current and the template sequence in real time to optimize the flight path; Step 5, Edge-Terminal Collaborative Computing: By constructing a multi-dimensional evaluation index system, a reinforcement learning decision model is used to dynamically select the optimal edge computing device to offload the computing tasks of the drone.

2. The UAV low-altitude logistics delivery system based on AI path planning according to claim 1, characterized in that, In step three, the basic variables related to the drone and the mission include drone status, drone health, historical performance, and mission characteristics, among which mission characteristics include urgency, spatial complexity, and cargo risk.

3. The drone low-altitude logistics delivery system based on AI path planning according to claim 1, characterized in that, In step four, the method for obtaining the drone's minimum energy consumption action command is as follows: A simulation experimental platform was built to construct flight environments with different wind layers, wind force, and wind direction. Based on the reinforcement learning framework, the UAV control commands were directly output using the policy network. With the goal of minimizing energy consumption, the value of the state was evaluated by the value network to generate the flight path with the lowest energy consumption. During the experiment, wind field data and motor current waveform data around the UAV were collected simultaneously, and the continuous action commands of the UAV were marked on the current waveform as action nodes.

4. The drone low-altitude logistics delivery system based on AI path planning according to claim 1, characterized in that, In step four, the motor current waveform is matched, and the specific method is as follows: A matching method combining dynamic time warping algorithm and cosine similarity calculation is adopted. The real-time motor current waveform data collected during the flight of the UAV is used as the query sequence, and the existing waveform data is used as the template sequence. The dynamic time warping algorithm is used to calculate the best alignment between the query sequence and the template sequence on the time axis. Based on this, the cosine similarity between the two waveform sequences is calculated. When the cosine similarity between the query sequence and a certain template sequence exceeds the set threshold, the two are considered to be successfully matched.

5. The drone low-altitude logistics delivery system based on AI path planning according to claim 1, characterized in that, In step four, the flight path is optimized in real time. The specific process is as follows: The system monitors the cosine similarity between the actual drone motor current waveform and the matched template sequence in real time. When the similarity is consistently higher than the set threshold and the proportion of similar sequences reaches the similarity threshold, the system calls the control command of the target action node in advance to adjust the drone's flight trajectory. If the cosine similarity is lower than the set threshold before reaching the next action node, the system re-searches and matches the appropriate template sequence.

6. The drone low-altitude logistics delivery system based on AI path planning according to claim 1, characterized in that, Step five involves constructing a multi-dimensional evaluation indicator system. The specific steps are as follows: S1. Construct computing task adaptation dimension indicators, including computing timeliness requirement score, computing data volume score, and data transmission volume score; S2. Construct device resource dimension indicators, including computing resource score, communication quality score, and device stability score.

7. The drone low-altitude logistics delivery system based on AI path planning according to claim 1, characterized in that, In step five, the optimal edge computing device is dynamically selected. The specific steps are as follows: S1. Construct a dynamic weight matrix; S2. Dynamically adjust the weight matrix according to actual task requirements; S3. Use a reinforcement learning decision model to select the optimal edge computing device.

8. The drone low-altitude logistics delivery system based on AI path planning according to claim 7, characterized in that, The specific steps for selecting the optimal edge computing device using a reinforcement learning decision model are as follows: S1. Define the state space; S2, Define the action space; S3. Design the reward function; S4. Iterative optimization; S5. Output the optimal solution.

Images