Edge computing task allocation method for air-ground cooperative inspection of electric power storage

Abstract

The invention discloses an edge computing task allocation method for electric power storage air-ground cooperative inspection, and belongs to the technical field of robot cooperative control and mobile edge computing. The method comprises the following steps: establishing an air-ground cooperative mobile edge computing system consisting of an edge server, a ground inspection robot and a micro unmanned aerial vehicle; dividing a task period into discrete time slots, and constructing a dynamic channel gain model and a task transmission rate model considering obstacle shielding; defining a task transfer distribution proportion, and respectively establishing a calculation time delay model and a system energy consumption model of the master robot and the slave robot; and constructing a utility function with the purpose of minimizing the total cost of the system based on a time delay and energy consumption model, and solving the optimal task transfer distribution proportion through an optimization algorithm. Through collaborative optimization of communication, calculation and energy consumption resources, balanced improvement of task execution timeliness and overall energy efficiency of the system in a complex dynamic storage environment is realized, and adaptability, stability and task completion rate of the inspection system are effectively improved.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of robot collaborative control and mobile edge computing technology. Specifically, it relates to a method for dynamic scheduling and resource optimization of computing tasks in a master-slave robot collaborative system for intelligent inspection of power storage, and in particular, a method for allocating edge computing tasks for air-ground collaborative inspection robots based on utility function optimization. Background Technology

[0002] With the continuous development of smart grid construction, the demand for intelligent and refined inspection of power storage materials is becoming increasingly prominent. Traditional manual inspection methods have inherent drawbacks such as low efficiency, high cost, and limited coverage, making it difficult to meet the management requirements of modern large-scale warehousing. In recent years, the integration of robotics technology and edge computing has provided a new solution to this problem. Among them, the air-ground collaborative robot system, which is a master-slave system composed of a ground robot with global mobility (the master robot) and a drone with flexible vision and maneuverability (the slave robot), has shown great potential. By dynamically migrating and distributing computing tasks between the robot's local location and the drone's edge node, it is expected to achieve efficient collaborative processing of inspection data.

[0003] However, applying such systems to real-world power storage environments with complex structures and dynamic changes still faces a series of severe technical challenges, making it difficult to achieve the expected overall inspection efficiency. Existing technologies mainly suffer from the following problems: 1. Insufficient adaptability to dynamic environments: The numerous shelving units and frequent equipment movement within power storage facilities lead to severe non-line-of-sight obstructions and signal fluctuations in communication links. Most existing collaborative frameworks employ static or quasi-static task allocation and channel models, failing to adequately consider the real-time impact of robot mobility and environmental obstacles on communication quality. This results in uncontrollable task transmission delays and decreased system reliability.

[0004] 2. Lack of Multi-Dimensional Resource Coordination Optimization: Air-ground collaborative inspection is a systemic problem involving the coupling of multi-dimensional resources such as communication, computing, and energy consumption. Existing research often optimizes a single indicator in isolation (such as minimizing only latency or energy consumption), or uses overly complex models such as deep reinforcement learning, which makes practical deployment difficult. There is a lack of a lightweight and solvable equilibrium strategy that can coordinate the optimization of task execution timeliness and overall system energy consumption under complex constraints.

[0005] 3. Edge computing load conflicts and energy efficiency imbalances: When distributing tasks among heterogeneous master-slave robots, edge nodes such as drones may experience computational overload, while the master robot's computing resources may be idle. Meanwhile, drones consume significant energy during hovering and flight. Existing methods fail to effectively embed refined modeling and constraints on drone hovering states and total system energy consumption into task allocation strategies, leading to low energy efficiency.

[0006] Therefore, there is an urgent need for an intelligent strategy that can adapt to the dynamic and complex environment of power storage, collaboratively optimize multi-dimensional resources of communication, computing and energy consumption, and effectively guide the transfer and allocation of edge computing tasks between master and slave robots, so as to significantly improve the overall energy efficiency and task completion rate of the system while ensuring the real-time performance of inspection tasks. Summary of the Invention

[0007] This invention aims to provide an edge computing task allocation method for collaborative air-ground inspection of power storage facilities, addressing the problems of high latency, high energy consumption, and low resource coordination efficiency in master-slave robot systems under dynamic and complex storage environments. This invention achieves coordinated optimization of task execution timeliness and overall system energy consumption by constructing a multi-dimensional fusion system model and optimization framework.

[0008] The technical solution adopted in this invention is as follows: An edge computing task allocation method for collaborative air-ground inspection of power storage facilities includes the following steps: S1. Construct an air-ground collaborative mobile edge computing system consisting of an edge server, a ground inspection robot, and a micro drone; the ground inspection robot acts as the master robot, receiving and executing inspection tasks issued by the edge server; the micro drone acts as the slave robot, receiving and processing some edge computing tasks transferred by the master robot while hovering. S2 divides the total inspection task cycle T into N discrete time slots and models the position coordinates of the master and slave robots in each time slot; based on the time-varying Euclidean distance between the two robots and the influence of obstacles, a dynamic channel gain model is constructed, and then the uplink transmission rate of the master robot to transmit task data to the UAV is calculated. S3, Define the task transfer allocation ratio parameter Based on the uplink transmission rate and the local computing capabilities of the master and slave robots, a computational latency model for the master robot to execute tasks and a computational latency model for the UAV to process transfer tasks are established respectively. S4. Establish the mobile energy consumption, computing energy consumption and communication energy consumption models of the main robot, as well as the flight energy consumption and computing energy consumption models of the UAV, to obtain the total energy consumption model of the system in a single time slot. S5. Based on the aforementioned time delay model and energy consumption model, construct a utility function with the objective of minimizing the total system cost. The optimal task transfer allocation ratio is obtained by optimizing the algorithm. .

[0009] Furthermore, in step S2, the dynamic channel gain Represented as: ; in, For the first The dynamic channel gain within each time slot reflects the quality of the signal propagation path; The reference channel gain per unit distance; For the first The three-dimensional Euclidean distance between the master and slave robots within a time slot is defined as: ; in, Main robot coordinates; The hovering coordinates of the drone; To maintain a constant flight altitude for the drone.

[0010] Furthermore, in step S2, the uplink transmission rate Represented as: ; in, Indicates the first Within each time slot, the main robot To drones Uplink transmission rate of data transmission task; Channel bandwidth; The uplink transmission power of the main robot; Noise power; Environmental obstacle occlusion factor; This is the baseline value for non-line-of-sight path loss; This is the additional loss caused by multipath scattering.

[0011] Furthermore, in step S3, the main robot calculates the latency. Includes local computing and task transfer components: ; in, The main robot in the The computation latency within each time slot includes both local computation latency and task transmission latency. The proportion of tasks executed locally by the main robot indicates the portion of the total tasks that were not transferred to the drone. The task transfer allocation ratio, with a value range of [0,1], represents the proportion of the total task transferred to the drone; The amount of data to be processed within a single time slot; The number of CPU cycles required to process a unit bit of task; The main robot's CPU calculates the frequency; Indicates the first Within each time slot, the main robot To drones Uplink transmission rate of data for transmitting tasks.

[0012] Furthermore, in step S3, the drone calculates the latency. for: ; in, Human Machine In the The computational latency within each time slot represents the time required for the drone to process the edge computing tasks transferred from the main robot, including only the computational portion. The computing frequency of a drone's CPU represents the number of computation cycles that the drone processor can execute per second, and is used to quantify its computing power.

[0013] Furthermore, in step S4, the main robot in the... Energy consumption within a time slot for: ; in, For the robot in the The total energy consumption within each time slot includes three parts: mobile energy consumption, computing energy consumption, and communication energy consumption, which is used to quantify system resource consumption. The payload mass of the main robot represents the total mass of the main robot, including the robot body and onboard equipment, and is used to calculate the energy consumption during movement. The main robot's moving speed represents the average speed at which the main robot walks or moves during the inspection process, and is used to calculate the kinetic energy-related moving power. The effective switching capacitor coefficient of the CPU is an integrated parameter used to quantify the proportionality constant of the relationship between the CPU's dynamic power consumption and the cube of the frequency. The main robot's CPU calculation frequency represents the number of calculation cycles executed per second by the main robot's processor, and is used to calculate dynamic calculation power. The main robot in the The computational latency within each time slot includes local computation and task transmission latency, used to convert power into energy consumption; The uplink transmission power of the main robot represents the transmission power used by the main robot when transmitting mission data to the drone, and is used to calculate communication energy consumption; The task transfer allocation ratio, with a value range of [0,1], represents the proportion of the total task transferred to the drone; The amount of data to be processed within a single time slot; Indicates the first Within each time slot, the main robot To drones Uplink transmission rate of data for transmitting tasks.

[0014] Furthermore, in step S4, the drone... Energy consumption within a time slot for: ; in, For drones in the The total energy consumption within a time slot includes only two parts: flight / hovering energy consumption and computing energy consumption. For the effective payload mass of the UAV; The average flight speed of the UAV during the hovering position adjustment process within this time slot; The CPU's calculation frequency for the drone; For drones in the The actual time spent executing computational tasks within a time slot.

[0015] Furthermore, in step S5, the utility function Defined as the weighted sum of total system delay and total energy consumption, and solved using a genetic algorithm or particle swarm optimization algorithm to obtain the optimal task transfer allocation ratio. This proportion varies with environmental barrier factors. Dynamic adjustment of channel status and computing load: ; in, The overall system optimization objective function; The task transfer allocation ratio, with a value range of [0,1], represents the proportion of the total task transferred to the drone; This indicates that the entire inspection cycle is discretized into One time slot; This is the time delay weighting coefficient. ; Energy consumption weighting coefficient ; For the first The computational delay of the time-slot master robot itself; For the first The computational latency of time-slotted UAVs in handling transfer tasks; For the first Total task completion delay of the time-slot system; For the first Total energy consumption of the time-slotted main robot; For the first Total energy consumption of time-slot drones; For the first Total energy consumption of the time-slot system; This is the optimal task transfer ratio obtained from the final solution; For channel status, computational load, affecting Dynamically adjusted key environmental parameters The larger the value, the worse the channel performance.

[0016] A master-slave robot edge computing task transfer and allocation system for collaborative air-ground inspection of power storage facilities, characterized in that it includes: Edge servers are used to distribute inspection tasks and manage the system's operational status. The ground inspection robot, as the main robot, is equipped with an edge computing unit to perform local computing tasks and transfer some tasks to the drone; Micro drones, as robots, are equipped with edge computing units to receive and process transferred computing tasks while hovering. The system is configured to perform the edge computing task allocation method described above for collaborative air-ground inspection of power storage facilities.

[0017] Compared with existing technologies, the edge computing task allocation method for collaborative air-ground inspection of power storage facilities provided by this invention has the following significant advantages: 1. Implemented adaptive task allocation in dynamic environments: This invention discretizes the inspection cycle into multiple time slots and combines them with a dynamic channel gain model that considers obstacle obstruction. This enables real-time sensing and response to communication quality fluctuations caused by shelf obstruction and equipment movement in power storage. This allows the task allocation strategy to be adjusted according to real-time communication conditions, significantly improving the system's robustness and adaptability in complex dynamic environments. It effectively solves the problems of uncontrollable latency and poor transmission reliability in dynamic scenarios caused by traditional static strategies.

[0018] 2. Multi-dimensional collaborative optimization of latency and energy consumption has been achieved: This invention innovatively integrates models of three key resources—communication, computing, and energy consumption—and constructs an optimization framework with the weighted sum of total task completion latency and total system energy consumption as the objective function. By introducing adjustable weight coefficients, real-time performance and energy efficiency can be flexibly balanced according to specific task requirements. This lightweight and solvable equilibrium optimization strategy overcomes the shortcomings of existing methods that rely on single-index optimization or overly complex models that are difficult to deploy, significantly reducing overall system energy consumption while ensuring real-time task completion.

[0019] 3. Effectively avoids edge node overload, improving overall system efficiency and stability: By introducing a task transfer allocation ratio parameter ρ and using an optimization algorithm to solve for its optimal value, this invention can intelligently determine the ratio of tasks to be executed locally and offloaded to the drone based on the heterogeneous computing capabilities and real-time load of the master and slave robots. This avoids the drone becoming a performance bottleneck due to task overload as an edge node, and also makes full use of the idle computing resources of the master robot. At the same time, the detailed modeling of the drone's hovering state and energy consumption further optimizes its energy efficiency when performing computing tasks, thereby improving the system's task completion rate and operational stability at the global level.

[0020] 4. Provides a scalable and easily deployable system solution. The method proposed in this invention is based on a clear mathematical model and a standard optimization algorithm. The model is clear and the steps are well-defined, making it easy to verify in simulations and providing clear guidance for engineering deployment in real-world systems. This framework possesses good scalability, laying the foundation for further research into task allocation problems in more complex collaborative scenarios such as multi-robot and multi-edge-node environments.

[0021] In summary, this invention effectively solves the core challenges of dynamic task allocation, multi-objective trade-offs, and resource conflicts in the collaborative air-ground inspection of power storage facilities through system modeling and collaborative optimization, providing a practical technical solution for achieving efficient, reliable, and energy-saving intelligent inspection. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0023] Figure 1 This is a schematic diagram of the overall process of the edge computing task allocation method for air-ground collaborative inspection of power storage provided in this embodiment of the invention; Figure 2 This is a schematic diagram of the structure of the air-ground collaborative master-slave robot mobile edge computing task system constructed in an embodiment of the present invention; Figure 3 This is a schematic diagram illustrating the discretization of the inspection task time period T in an embodiment of the present invention. Detailed Implementation

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

[0025] To address the problems of high latency, high energy consumption, and low resource coordination efficiency in task allocation of master-slave robot systems in the dynamic and complex environment of power storage, this embodiment provides an edge computing task allocation method for air-ground collaborative inspection of power storage. This method constructs an air-ground collaborative mobile edge computing system and establishes a multi-dimensional model integrating communication, computing, and energy consumption. Finally, an optimization algorithm is used to solve the task transfer and allocation strategy, achieving comprehensive optimization of task execution timeliness and overall system energy consumption.

[0026] like Figure 1 As shown, the edge computing task allocation method for the collaborative air-ground inspection of power storage facilities is implemented in the following steps: S1, constructing an air-ground collaborative mobile edge computing system consisting of edge servers, ground inspection robots, and micro drones: The air-ground collaborative mobile edge computing system in this embodiment is a heterogeneous robot collaborative architecture designed to meet the needs of three-dimensional and dynamic inspection of power storage facilities. For example... Figure 2 As shown, the system consists of three layers: an edge server, a ground quadruped inspection robot (i.e., the main robot), and a micro inspection drone (i.e., the slave robot). It achieves collaborative optimization of computing resources through a dynamic task transfer mechanism.

[0027] 1. System composition and functional division: Edge Server: Deployed in the warehouse control center, serving as the task management and scheduling hub. Its main functions include: generating and distributing global inspection tasks; receiving processed inspection data; and monitoring the operating status and resource load of master and slave robots. The server does not participate in real-time computation task processing.

[0028] The ground inspection robot, or main robot R, utilizes a ground platform with strong obstacle-crossing and stable movement capabilities to handle large-scale inspections and overall task coordination. Its onboard edge computing unit performs path planning, environmental perception, preliminary target detection, and collaborative scheduling with drones. Due to its physical height limitations, it cannot directly perform detailed observations of high-rise shelving.

[0029] Micro unmanned aerial vehicles (UAVs), also known as robotic UAVs, act as mobile edge nodes in the air. Their core function is to receive and process tasks transferred from the main robot that require high-altitude visibility or sophisticated calculations. A key operational constraint is that the UAV only activates its computing units and processes tasks in a stable hovering state to ensure flight safety and computational stability. Its mobility is used to quickly reach optimal communication and observation positions.

[0030] 2. Task Transfer and Collaboration Mechanism: Within each discretized inspection time slot, the main robot R receives a total of L computational tasks to be processed. The system utilizes a core optimization variable: the task transfer allocation ratio. The execution location of tasks is dynamically determined: the amount of tasks executed locally by the main robot is: The amount of tasks transferred to drones is .

[0031] The collaborative workflow is briefly described below: a. The main robot determines the optimal time slot for the current time slot based on the current environment and its own load using an optimization algorithm. value.

[0032] b. The main robot will transfer the data that needs to be transferred. The message is sent wirelessly to the drone that has reached the cooperative position and is hovering.

[0033] c. The drone completes the calculation task while hovering and sends the simplified results back to the main robot.

[0034] d. The main robot integrates the processing results from the local machine and the drone, and finally reports them to the edge server.

[0035] The system architecture established in this step is the cornerstone of this method. Its resources are heterogeneous and complementary, combining the long endurance and heavy payload of ground robots with the flexible field of view and rapid deployment advantages of UAVs. Competency is achieved through computational task transfer. Computation and communication are decoupled, and the constraints of "hovering computation" are clearly defined, separating the energy-intensive flight phase of the UAV from the computation-intensive processing phase, facilitating detailed energy consumption and latency modeling. A dynamic adaptive framework is established by introducing an optimizable task transfer ratio. This provides a core control mechanism for the system to cope with dynamically changing communication environments, heterogeneous computing capabilities, and real-time task requirements. This system model provides a clear physical context and constraints for subsequent steps in constructing accurate communication transmission models, computational latency models, system energy consumption models, and ultimately, target optimization, enabling the entire method to closely align with the actual application scenarios of power storage inspection.

[0036] S2, Constructing an air-to-ground collaborative master-slave communication transmission model: The core objective of this step is to establish a communication transmission model that accurately reflects the complex and dynamic environment of power storage facilities, providing a precise data transmission rate basis for subsequent latency calculations, energy consumption analysis, and task allocation strategy optimization. Due to the presence of dense, multi-level metal shelving, frequently moving equipment, and a complex electromagnetic reflection environment within power storage facilities, traditional free-space or simple non-line-of-sight channel models can lead to a significant overestimation of the transmission rate, causing optimized task allocation strategies to fail in actual deployment. Therefore, this method makes crucial modifications to the classic model specifically tailored to the characteristics of the storage scenario.

[0037] 1. Spatiotemporal Discretization and Location Modeling: To achieve dynamic analysis, the total inspection cycle T is discretized into N equal-length time slots, such as... Figure 3 As shown. The time slot length is In each time slot Assuming the positions of the master and slave robots are relatively stable, they can be represented by coordinates respectively: Main robot R position: UAV hovering position: ;in, The constant hovering altitude for drones when performing collaborative tasks.

[0038] Based on this, the first The three-dimensional Euclidean distance between the two robots within the time slot is: .

[0039] 2. Dynamic Channel Gain Model: Channel power gain under line-of-sight (LoS) conditions. It is inversely proportional to the square of the distance: ; in, For the first The dynamic channel gain within each time slot reflects the quality of the signal propagation path; The reference channel gain per unit distance is determined by the carrier frequency and antenna characteristics. For the first The three-dimensional Euclidean distance between the master and slave robots within the time slot.

[0040] 3. Warehouse Environment-Modified Transmission Rate Model: To characterize the complex impact of the warehouse environment on communication quality, this method introduces an environmental perception correction term into the Shannon formula, resulting in an actual uplink transmission rate model from the main robot to the drone: ; in, Indicates the first Within each time slot, the main robot To drones Uplink transmission rate of data transmission task; Channel bandwidth; The uplink transmission power of the main robot; Noise power; This represents the environmental obstacle occlusion factor, where 0 represents no obstruction and 1 represents severe occlusion. This serves as a baseline value for non-line-of-sight path loss, characterizing the penetration loss through obstacles such as shelves. The additional loss caused by multipath scattering characterizes signal interference in metallic environments.

[0041] Continuously adjustable obstacle factor It breaks through the traditional binary model of "connected / disconnected" and can accurately represent the gradual degree of occlusion in different areas, such as open aisles and densely packed shelving areas. It can be dynamically assigned values ​​through environmental maps or real-time perception.

[0042] Environment-specific loss items and This feature is specifically designed to address the physical characteristics of power storage warehouses, which are characterized by numerous metal shelves and strong surface reflections. This allows the model to better match the actual channel characteristics, significantly improving the accuracy of transmission rate prediction.

[0043] This step, by constructing a communication transmission model that integrates spatiotemporal discretization, three-dimensional distance, and warehouse-specific correction factors, provides a realistic and reliable communication foundation for the entire task allocation method. It ensures that all subsequent operations dependent on the transmission rate... The calculation of latency and energy consumption, as well as the final task allocation optimization results, can effectively reflect the actual operating conditions of power storage, thereby ensuring the engineering feasibility and superiority of the strategy of this invention.

[0044] S3, Constructing an air-to-ground collaborative master-slave computing latency model: The goal of this step is to accurately quantify the execution time of tasks in a master-slave robot collaborative system, in order to evaluate the system's real-time performance and ultimately optimize the task allocation ratio. Provides core evidence. The model is built upon the communication model of step S2, by introducing decision variables. It clearly depicts the trade-off between local computation on the main robot and computation transferred to the drone, directly addressing the contradiction between the limited computing power of the main robot and the real-time requirements of the inspection task.

[0045] 1. Definition of Task Load and Transfer Mechanism: Assume that in the first... Within a time slot, the main robot R receives a total of L edge computing tasks to be processed. These tasks may include high-resolution image analysis, infrared thermal image processing, RFID data parsing, etc. A core decision variable is defined: the task transfer allocation ratio. Used for dynamically allocating tasks: The amount of tasks executed locally by the main robot R is: The workload transferred to UAVs is: .

[0046] 2. Main robot computation latency model: Total task processing latency of the main robot. It consists of two parts: Local computation latency, processing The computation time required for a bit task.

[0047] Task transmission latency will The time required for bit task data to be transmitted to the drone via the wireless channel is included in the delay on the main robot side, since the transmission process is initiated and controlled by the main robot.

[0048] Therefore, as Figure 3 As shown, the main robot's time delay model is: ; in, The main robot in the The computation latency within each time slot includes both local computation latency and task transmission latency. The proportion of tasks executed locally by the main robot indicates the portion of the total tasks that were not transferred to the drone. The task transfer allocation ratio, with a value range of [0,1], represents the proportion of the total task transferred to the drone; The amount of data to be processed within a single time slot; The number of CPU cycles required to process a unit bit of task; The CPU calculates the frequency for the main robot.

[0049] 3. UAV latency calculation model: latency of unmanned aerial vehicles (UAVs) It only includes the time required to process the transfer tasks it receives. For example... Figure 3 As shown, since the drone receives tasks while hovering, its power consumption for receiving data is extremely low and negligible, and the amount of task data is also very small. The amount of data returned is much larger than the computational latency of the returned data, so the model focuses on its computational latency. ; in, Human Machine In the The computational latency within each time slot represents the time required for the drone to process the edge computing tasks transferred from the main robot, including only the computational portion. The computing frequency of a drone's CPU represents the number of computation cycles that the drone processor can execute per second, and is used to quantify its computing power.

[0050] This step, by introducing the number of computation cycles *s* per bit, achieves a unified quantification of the complexity of heterogeneous computing tasks, making the model applicable to diverse inspection tasks in power storage. The model uses... As a bridge, it will enhance communication performance. With heterogeneous computing power ( , The dynamic correlation provides a mathematical foundation for finding the optimal task splitting point. It clearly distinguishes between the combined "computation + transmission" latency of the master robot and the "pure computation" latency of the drone, conforming to the actual operation process of master-slave collaboration and ensuring the practicality of subsequent optimization results. This computational latency model is a key component in constructing the system's total cost function and achieving coordinated optimization of latency and energy consumption.

[0051] S4: Constructing an energy consumption model for an air-ground collaborative master-slave system This step aims to establish a systematic energy consumption model to accurately quantify the total energy expenditure of the master and slave robots during collaborative inspection tasks. This model is crucial for balancing task execution timeliness with overall system energy consumption and achieving sustainable inspection. By decomposing energy consumption into multiple dimensions such as mobile flight, computation, and communication, and modeling it based on the latency results from step S3, it provides a basis for subsequent optimization of task allocation ratios. It provides a basis for energy consumption constraints, thereby avoiding premature battery depletion of robots or low system energy efficiency due to improper task allocation.

[0052] 1. System energy consumption composition: The system in the first... Total energy consumption per time slot The sum of the energy consumption of the main robot and the drone: ; In the formula, For the robot in the The total energy consumption within each time slot includes three parts: mobile energy consumption, computing energy consumption, and communication energy consumption, which is used to quantify system resource consumption. For drones in the The total energy consumption within a time slot includes only two parts: flight / hovering energy consumption and computing energy consumption.

[0053] The energy sources for each robot are as follows: Main robot Energy is consumed for movement: overcoming friction and maintaining speed; calculation: the CPU executes instructions; and communication: transmitting data uplink.

[0054] drones Energy is consumed in flight / hovering adjustments to maintain airborne position and attitude, and in computation to handle transfer tasks. Since it acts as the task receiver, its communication reception power consumption is much lower than other components, and therefore it is ignored in this model to simplify the model and focus on the main energy sources.

[0055] 2. Main robot energy consumption model: The main robot in the... Total energy consumption of time slot The model is as follows: ; in, The payload mass of the main robot represents the total mass of the main robot, including the robot body and onboard equipment, and is used to calculate the energy consumption during movement. The main robot's moving speed represents the average speed at which the main robot walks or moves during the inspection process, and is used to calculate the kinetic energy-related moving power. is the CPU effective switching capacitor coefficient, which is an integrated parameter used to quantify the proportional constant of the relationship between CPU dynamic power consumption and frequency cube.

[0056] Model Term Explanation: Mobile and Computation Power Consumption Term: Represents maintaining average moving speed Approximate kinetic energy power required; This is a power calculation based on the classic CMOS circuit dynamic power consumption model. Power consumption time: The power consumption time during task execution. Internal continuous consumption. Communication energy consumption: Indicates sending The energy consumed by bit task data.

[0057] 3. Drone Energy Consumption Model: Human-machine interaction in the first... Total energy consumption of time slot The model is as follows: ; in, For the effective payload mass of the UAV; The average flight speed of the UAV during the hovering position adjustment process within this time slot; The calculation frequency of the CPU on the drone.

[0058] Model Term Explanation: Flight Hovering and Calculation Power Consumption Terms: This represents the average flight kinetic energy power of the drone during the hovering process; This refers to its computational power consumption. Power consumption time: the delay during which the above power is consumed while the drone is performing computational tasks. Internal continuous consumption. The model implicitly assumes that the power required for the UAV to maintain its position during hovering calculations is already included in the flight power term.

[0059] This step decomposes the robot's energy consumption into independent, quantifiable components such as movement, computation, and communication, making the energy consumption analysis more accurate and helping to identify energy bottlenecks. The energy consumption model directly relies on the latency calculated in step S3. and A clear transmission chain of "task allocation → latency → energy consumption" was established. By simplifying aspects such as ignoring the energy consumption of UAV receivers, the complexity was reduced while ensuring the core accuracy of the model, facilitating its integration into subsequent optimization algorithms. This model, together with the computational latency model, constitutes the two core inputs to the objective utility function in step S5, providing a quantitative basis for optimally balancing latency and energy consumption. This system energy consumption model ensures the optimal task allocation strategy is ultimately solved. It not only considers the speed of task completion, but also fully takes into account the system's energy efficiency and endurance, which is crucial for long-term inspection tasks of power storage in actual deployment.

[0060] S5: Establish the objective utility function and solve for the optimal task transfer allocation ratio. : This step is the final decision-making stage of the invention, aiming to integrate the aforementioned communication model S2, latency model S3, and energy consumption model S4 into a global optimization problem, and to solve for the optimal task transfer allocation ratio using mathematical methods. This ratio will serve as the core control command, guiding the main robot to dynamically adjust the amount of tasks unloaded to the drone in each time slot, thereby achieving the optimal balance between task execution timeliness and overall system energy consumption in complex warehousing environments.

[0061] 1. Construct a global objective utility function: First, the definition in the... The total cost of the system within a time slot is a weighted sum of latency and energy consumption, reflecting the execution efficiency and resource consumption of that time slot: ; in: The total system delay is determined by the S3 model.

[0062] The total system energy consumption is determined by the S4 model.

[0063] Delay weighting coefficient, 0≤ ≤1. Parameter The settings can be configured by the operations and maintenance personnel according to the actual task requirements: When the value approaches 1, optimization tends to minimize latency, i.e., to complete the task as quickly as possible. When the value is close to 0, optimization tends to minimize power consumption, i.e., pursue the longest possible system battery life; =0.5 indicates that latency and energy consumption are equally important.

[0064] Considering that the entire inspection cycle contains N time slots, the global optimization objective is to minimize the total cost of all time slots. Therefore, the objective utility function is constructed as follows: ; in, The overall system optimization objective function; This indicates that the entire inspection cycle is discretized into One time slot; For the first Total task completion delay of the time-slot system; For the first Total energy consumption of the time-slot system.

[0065] here, It is the only decision variable. Based on the problem description, It can be a globally uniform value, or it can be optimized independently for each time slot. To simplify online calculations, this embodiment preferably employs a globally unified approach. Optimize.

[0066] 2. Optimize the solution algorithm: objective function It is about The function is a one-dimensional function, but it contains complex communication, latency, and energy consumption models, and may be non-convex. To ensure robustness and efficiency of the solution, this embodiment uses the Particle Swarm Optimization (PSO) algorithm. Its specific configuration is as follows: Search space: ∈[0,1].

[0067] Particle dimension: 1.

[0068] Population size: 20-50.

[0069] Maximum number of iterations: 100.

[0070] Convergence condition: The improvement of the global optimal solution after 10 consecutive iterations is less than [a certain value]. .

[0071] Output: Make smallest .

[0072] The PSO algorithm has the advantages of few parameters, fast convergence, and ease of implementation. Tests have shown that it can complete the solution within milliseconds on embedded edge computing platforms such as the NVIDIA Jetson series, meeting the requirements for online real-time decision-making. It can also be replaced by genetic algorithms, simulated annealing, or one-dimensional direct search methods, depending on the specific circumstances.

[0073] This step successfully integrated a multi-dimensional, nonlinear, complex system model into a single decision variable. This optimization problem significantly reduces the computational complexity of online decision-making. This is achieved by introducing weight coefficients. This provides a clear and intuitive "control knob," enabling the system to flexibly adjust its strategy based on different operational goals, such as prioritizing timeliness or battery life, thus enhancing the method's practicality and operability. The lightweight and efficient optimization algorithm allows the optimal task allocation strategy to be calculated within milliseconds, meeting the real-time response requirements of power storage inspection, a stark contrast to deep reinforcement learning methods that require extensive offline training. The final output... It is a specific numerical value that can be directly used for control commands, guiding the main robot to accurately unload tasks, thus completing the closed loop from model to application.

[0074] In summary, step S5, by establishing and solving the objective utility function, provides an efficient, adaptive, and interpretable optimal task allocation decision mechanism for the air-ground collaborative inspection system proposed in this method, which is key to the superior performance of the entire method.

[0075] To further verify the practical effect and optimization performance of the edge computing task allocation method for power storage air-ground collaborative inspection proposed in this invention, this embodiment designed and executed the following complete test process, demonstrating the entire process from system modeling to optimal strategy solution through specific numerical calculations: 1. Test system and parameter initialization: System configuration: The system model established strictly follows this method and consists of a warehouse edge server, a ground quadruped inspection robot (i.e., the master robot R), and a micro unmanned aerial vehicle (i.e., the slave robot UAV).

[0076] Task settings: The total inspection cycle is divided into N=3 time slots, and the amount of edge computing tasks that the main robot needs to process in each time slot is fixed at L=50 bits.

[0077] Location and communication parameters: The drone hovers at a constant height of H=3m.

[0078] The test is aimed at the first Time slot, set the position of the main robot =(0,0,0), drone position =(3,4,3).

[0079] Channel parameters: Channel gain per unit distance =1, bandwidth =50Hz, uplink transmit power =0.1W, noise power =0.001W.

[0080] Environmental parameters simulating severe occlusion: obstacle factor =1, non-line-of-sight path loss =0.01W, scattering and reflection loss =0.005W.

[0081] 2. Model calculation process and results: Step S2, Calculation of communication transmission model: Channel gain: ; Uplink transmission rate: .

[0082] Step S3, calculate the delay model Calculation parameter: CPU cycle count Main robot CPU frequency Drone CPU frequency .

[0083] Main robot latency: ; The simplified expression is: ; Drone latency: ; Use the simplified expression: .

[0084] Step S4, system energy consumption model calculation: Energy consumption parameters: mass , ;speed , CPU capacitance coefficient .

[0085] Main robot power consumption: Use the simplified expression: .

[0086] Drone energy consumption: ; Use the simplified expression: .

[0087] 3. Objective function construction and optimal policy solution Step S5, utility function optimization: Total delay per time slot: ; Total energy consumption per time slot: ; Global objective function, time delay and energy consumption weights : ; Optimization solution: Objective function It is about A linearly decreasing function. Under constraints. In the context of [0,1], to make Minimize, should take The maximum value. Considering potential constraints in the model, such as the maximum processing capacity of the drone or practical engineering limitations, the optimal task transfer allocation ratio for this test scenario is obtained by solving the problem using optimizers such as genetic algorithms / particle swarm optimization: .

[0088] 4. Test Conclusion: This test, using a specific set of parameters, fully demonstrates the implementation steps of the method of this invention: A dynamic communication model was successfully established, and the actual transmission rate (12.17 bps) under specific obstruction conditions was calculated.

[0089] Constructed with A time delay and energy consumption model with variables as variables.

[0090] A clear linear objective utility function has been formed. .

[0091] The optimal task transfer ratio can be obtained by solving the problem. .

[0092] Test results show that, in the specific scenario of severe occlusion, low bandwidth, and low task load set in this embodiment, offloading approximately 45% of the computational task to UAV collaborative processing can minimize the weighted sum of the overall system cost, i.e., latency and energy consumption. This verifies the effectiveness of the method of the present invention in making optimal decisions under given conditions, demonstrating its complete process and practical value in guiding actual task allocation through mathematical modeling and optimization solutions. This method can be applied to different parameters in the actual scenario, such as task load L, channel bandwidth B, and occlusion factor. Wait, adaptively adjust the optimal This makes it widely applicable to the diverse inspection needs of power storage facilities.

[0093] It should be noted that some parameters used in the above tests should be determined based on actual hardware performance data and environmental measurement results in real-world applications. The core value of this method lies in the proposed system architecture, modeling approach, and optimization framework; its effectiveness does not depend on the specific values ​​shown in the examples.

[0094] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. An edge computing task allocation method for power storage yard collaborative inspection, characterized in that: The method comprises the following steps: S1, constructing an air-ground collaborative mobile edge computing system composed of an edge server, a ground inspection robot and a micro unmanned aerial vehicle; the ground inspection robot serves as a master robot, receives and executes an inspection task issued by the edge server; the micro unmanned aerial vehicle serves as a slave robot, receives and processes part of the edge computing task transferred by the master robot in a hovering state; S2, dividing a total inspection task cycle T into N discrete time slots, and modeling the position coordinates of the master and slave robots in each time slot; based on the time-varying Euclidean distance between the two robots and the influence of obstacles, a dynamic channel gain model is constructed, and then the uplink transmission rate of the master robot to the unmanned aerial vehicle for transmitting task data is calculated; S3, defining a task transfer allocation proportion parameter Based on the uplink transmission rate and the local computing capacity of the master and slave robots, a computing time delay model for the master robot to execute a task and a computing time delay model for the UAV to process a transfer task are respectively established. S4, respectively establishing the mobile energy consumption, computing energy consumption and communication energy consumption models of the master robot, and the flight energy consumption and computing energy consumption models of the unmanned aerial vehicle, to obtain a total energy consumption model of the system in a single time slot; S5, constructing an utility function aiming at minimizing the total cost of the system based on the time delay model and the energy consumption model , obtaining the optimal task transfer allocation proportion by solving the optimization algorithm .

2. The edge computing task allocation method for power warehouse yard collaborative inspection according to claim 1, characterized in that: In step S2, the dynamic channel gain is represented as: ； wherein, is the dynamic channel gain in the i-th time slot, reflecting the quality of the signal propagation path; is the dynamic channel gain in the i-th time slot, reflecting the quality of the signal propagation path; is the unit distance reference channel gain; is the dynamic channel gain in the i-th time slot, reflecting the quality of the signal propagation path; is the three-dimensional Euclidean distance between the master and slave robots in the i-th time slot, defined as: ； wherein, is the host robot coordinate; is the drone hover coordinate; is the drone constant flight height.

3. The edge computing task allocation method for power warehouse yard collaborative inspection according to claim 2, characterized in that: The uplink transmission rate in step S2 is expressed as: ； wherein, represents the host robot transmits task data to the UAV uplink transmission rate; is the channel bandwidth; is the host robot uplink transmission power; is the noise power; is the environmental obstacle blockage factor; is the non-line-of-sight path loss reference value; is the additional loss caused by multipath scattering.

4. The edge computing task allocation method for power warehouse yard collaborative inspection according to claim 1, characterized in that: In step S3, the master robot calculates the time delay comprising a local computation part and a task transfer part: ； in, The main robot in the The computation latency within each time slot includes both local computation latency and task transmission latency. The proportion of tasks executed locally by the main robot indicates the portion of the total tasks that were not transferred to the drone. The task transfer allocation ratio, with a value range of [0,1], represents the proportion of the total task transferred to the drone; The amount of data to be processed within a single time slot; The number of CPU cycles required to process a unit bit of task; The main robot's CPU calculates the frequency; Indicates the first Within each time slot, the main robot To drones Uplink transmission rate of data for transmitting tasks.

5. The edge computing task allocation method for power warehouse yard collaborative inspection according to claim 4, characterized in that: In step S3, the UAV calculates the time delay is: ； wherein, human-computer In the first time delay in the time slot, representing the time required for the UAV to process the edge computing task transferred from the host robot, only including the calculation part; The computing processing frequency of the UAV CPU represents the number of computing cycles that the UAV processor can execute per second, which is used to quantify its computing power.

6. The edge computing task allocation method for power warehouse yard collaborative inspection according to claim 1, characterized in that: In step S4, the master robot calculates the energy consumption of the slave robot in the time slot of the first time slot is: ； in, For the robot in the The total energy consumption within each time slot includes three parts: mobile energy consumption, computing energy consumption, and communication energy consumption, which is used to quantify system resource consumption. The payload mass of the main robot represents the total mass of the main robot, including the robot body and onboard equipment, and is used to calculate the energy consumption during movement. The main robot's moving speed represents the average speed at which the main robot walks or moves during the inspection process, and is used to calculate the kinetic energy-related moving power. The effective switching capacitor coefficient of the CPU is an integrated parameter used to quantify the proportionality constant of the relationship between the CPU's dynamic power consumption and the cube of the frequency. The main robot's CPU calculation frequency represents the number of calculation cycles executed per second by the main robot's processor, and is used to calculate dynamic calculation power. The main robot in the The computational latency within each time slot includes local computation and task transmission latency, used to convert power into energy consumption; The uplink transmission power of the main robot represents the transmission power used by the main robot when transmitting mission data to the drone, and is used to calculate communication energy consumption; The task transfer allocation ratio, with a value range of [0,1], represents the proportion of the total task transferred to the drone; The amount of data to be processed within a single time slot; Indicates the first Within each time slot, the main robot To drones Uplink transmission rate of data for transmitting tasks.

7. The edge computing task allocation method for power warehouse yard collaborative inspection according to claim 6, characterized in that: In step S4, the UAV consumes energy in the time slot of the first time slot is: ； in, For drones in the The total energy consumption within a time slot includes only two parts: flight / hovering energy consumption and computing energy consumption. For the effective payload mass of the UAV; The average flight speed of the UAV during the hovering position adjustment process within this time slot; The CPU's calculation frequency for the drone; For drones in the The actual time spent executing computational tasks within a time slot.

8. The edge computing task allocation method for power warehouse yard collaborative inspection according to claim 1, characterized in that: In step S5, the utility function defined as the weighted sum of total latency and total energy consumption of the system, is solved by genetic algorithm or particle swarm optimization algorithm to obtain the optimal task transfer allocation ratio which dynamically adjusts with environmental obstacle factor , channel state and computing load: ； wherein, is the total optimization objective function of the system; is the task transfer allocation ratio, taking the value range [0, 1], representing the proportion of the total task transferred to the unmanned aerial vehicle; represents the entire inspection cycle being discretized into time slots; is the delay weight coefficient, ; is the energy consumption weight coefficient, ; is the computing delay of the master robot itself in the time slot; is the computing delay of the unmanned aerial vehicle processing the transferred task in the time slot; is the total task completion delay of the system in the time slot; is the total energy consumption of the master robot in the time slot; is the total energy consumption of the unmanned aerial vehicle in the time slot; is the total energy consumption of the system in the time slot; is the optimal task transfer ratio obtained by final solution; is the channel state, computing load, and influences key environmental parameters for dynamic adjustment, the worse the channel.

9. A master-slave robot edge computing task transfer allocation system for power storage yard collaborative inspection, characterized in that, It comprises: an edge server configured to issue an inspection task and manage the system running state; a ground inspection robot serving as a master robot, carrying an edge computing unit, configured to execute a local computing task and transfer part of the task to a micro unmanned aerial vehicle; a micro unmanned aerial vehicle serving as a slave robot, carrying an edge computing unit, configured to receive and process the transferred computing task in a hovering state; The system is configured to perform the edge computing task allocation method for power warehouse air-ground collaborative inspection according to any one of claims 1 to 8.

Images