Multi-unmanned aerial vehicle cooperative inspection planning and scheduling method and device in smart power grid scene
Through cluster analysis and deep reinforcement learning algorithms, the problem of large amount of task planning and poor robustness in smart grids is solved, and the rapid and efficient inspection and resource optimization of smart grids are achieved.
Patent Information
- Application Number
- CN202410590767.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-05-13
- Publication Date
- 2024-11-08
AI Technical Summary
In smart grid scenarios, the existing UAV task planning methods are large in computing, difficult to converge results, and low applicability, resulting in unreasonable task allocation, long time, poor robustness, and unable to effectively cope with UAV performance and tasks. Dynamic changes in resources.
The cluster analysis method is used to determine the starting position of the drone, and combined with the deep reinforcement learning algorithm, scheduling instructions are generated based on the task priority and drone status information, and task allocation is optimized to ensure the overall optimization and robustness of task execution.
Improve task allocation efficiency, shorten calculation time, realize fast and efficient inspection of smart grids, reduce time and energy consumption costs, avoid resource waste caused by repeated tasks, and improve task completion rate and drone load balanced.
Smart Images

Figure CN118916125A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of drone scheduling, and in particular to a method and device for planning and scheduling collaborative inspections of multiple drones in a smart grid scenario. Background Art
[0002] Autonomous decision-making and control enable drones to perform tasks independently, and mission planning is the key to achieving autonomous decision-making and control of multiple drones. The collaborative mission planning of multiple drones must not only consider the requirements of the mission itself, but also the constraints of the collaborative execution of multiple drones. According to the characteristics of the mission, the collaborative flight path for the drones should be designed to optimize or approach the best overall performance.
[0003] In the actual situation, during the planning of drone inspection tasks, due to the limited number of drones, the number of tasks is usually much greater than the number of drones. In this case, each drone needs to be assigned multiple tasks, but at the same time, it can only complete a single task, so it needs to complete the assigned tasks in a certain order. For each drone, it may not be possible to guarantee the local optimization of its task execution process, but for all drones, the overall optimization of task execution should be guaranteed as much as possible. When inspecting with drones, it is necessary not only to consider the tasks that the drone needs to perform, but also to prioritize certain tasks due to different task states, and to plan the order in which the drones perform tasks.
[0004] In the related technology, the UAV mission planning is realized based on the traditional intelligent optimization algorithm. It is necessary to select the appropriate algorithm to allocate the tasks according to the different constraints or threats required by different scenarios. However, the intelligent optimization algorithm will have the problem of a sharp increase in the amount of calculation or the result is difficult to converge, and the intelligent optimization algorithm has certain limitations for the problem of dynamic changes in constraints and objective functions. Therefore, the introduction of deep reinforcement learning algorithm as the direction of solving UAV mission planning can be applied in a variety of different environments and can better solve the problem. However, there are the following problems in the current research method: 1. The existing technology often makes the UAV start from the same starting point, and does not consider the balance between the different capabilities of multiple UAVs and the resources required by the target, resulting in problems such as long time and poor robustness. 2. The robustness to different use environments and scenarios is poor, that is, the designed system is only for fixed special scenarios, but in actual situations, such as the number of tasks and the performance of UAVs are not the same, the current research method needs to be adjusted according to the scene changes, so the applicability is low. Summary of the invention
[0005] The present application provides a method, device, vehicle and storage medium for planning and scheduling multi-UAV collaborative inspections in a smart grid scenario, so as to solve the problem that UAV task planning in related technologies needs to select appropriate algorithms to allocate tasks according to the constraints or threats required by different scenarios, resulting in a sharp increase in the amount of calculation, difficulty in converging results, low applicability, and unreasonable task allocation, resulting in long time and poor robustness.
[0006] The first aspect of the present application provides a method for multi-UAV collaborative inspection planning and scheduling in a smart grid scenario, and the method is applied to a server, wherein the method includes the following steps: obtaining at least one inspection task in the smart grid scenario and multiple UAVs that perform the inspection tasks; identifying the status information and task priority of each inspection task and the status information of each UAV; generating a scheduling instruction for each UAV based on the status information and task priority of each inspection task and the status information of each UAV, and scheduling the multiple UAVs to perform the at least one inspection task based on the scheduling instruction of each UAV.
[0007] Optionally, the status information of each inspection task includes the task location and task time, the status information of each drone includes the drone location and flight time, and the task priority is determined based on the dynamic characteristics of power load fluctuations in the smart grid scenario.
[0008] Optionally, the scheduling instructions for each UAV are generated based on the status information and task priority of each inspection task and the status information of each UAV, including: determining the starting position of the UAV based on clustering the status information of each inspection task; inputting the starting position of the UAV, the status information and task priority of each inspection task and the status information of each UAV into a task planning model, and the task planning model outputs the scheduling instructions for each UAV.
[0009] Optionally, the task planning module is used to: traverse each inspection task in turn, generate a state vector based on the status information of the current inspection task, the task priority and the status information of each drone; generate the action probability of each drone selecting the current inspection task based on the state vector and the reward function, and generate the scheduling instructions for the drone executing the current inspection task based on the action probability until all inspection tasks are traversed.
[0010] Optionally, the reward function r t It is expressed as:
[0011] r t =r s +r E +r p
[0012] Among them, r S is the time reward, r S =-T total ×ε,0<ε<1,T total The total task completion time, ε represents the impact of task completion time on time reward; r E is the energy consumption reward, r E =-E total ×μ, 0<μ<1, E total is the total communication energy consumption, μ represents the influence of task energy consumption on energy consumption reward; r p As a reward for the task, Among them, n represents the total number of tasks, C represents the basic reward of each task, PIi represents the priority index of the task, k represents the number of completed tasks, and J represents the degree of task completion.
[0013] Optionally, before the starting position of the drone, the status information and task priority of each inspection task, and the status information of each drone are input into the task planning model, it also includes: obtaining a training data set, wherein the training data set includes the status information and task priority of multiple task samples, and the status information of the drone; traversing the multiple task samples in sequence, generating a state vector of the current task sample based on the status information and task priority of the current task sample, and the status information of the drone, and generating an action of the current task sample based on the state vector, wherein the action of the current task sample represents the probability of selecting the current task sample for each drone; obtaining the reward and next-moment state vector of each drone performing the action, generating a training sample based on the state vector, action, reward and next-moment state vector of the current task sample, and placing the training sample into the experience pool until all task samples are traversed.
[0014] Optionally, the determining the starting point position of the UAV according to the clustering of the status information of each inspection task includes: determining the task distance according to the task position of each inspection task; generating a feature matrix based on the task distance of each inspection task, and generating a new feature matrix according to the feature matrix and the task time; performing distance clustering on the new feature matrix based on a distance loss function to obtain a clustering result, and determining the starting point position of the UAV according to the clustering result.
[0015] The second aspect of the present application provides a multi-UAV collaborative inspection planning and scheduling device in a smart grid scenario, and the device is applied to a server, wherein the device includes: an acquisition module, used to acquire at least one inspection task in the smart grid scenario and multiple UAVs that perform the inspection tasks; an identification module, used to identify the status information and task priority of each inspection task and the status information of each UAV; a scheduling module, used to generate a scheduling instruction for each UAV based on the status information and task priority of each inspection task and the status information of each UAV, and schedule the multiple UAVs to perform the at least one inspection task based on the scheduling instruction of each UAV.
[0016] The third aspect of the present application provides a server, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the program to implement the multi-UAV collaborative inspection planning and scheduling method in the smart grid scenario as described in the above embodiment.
[0017] The fourth aspect of the present application provides a computer-readable storage medium having a computer program stored thereon, which is executed by a processor to implement the multi-UAV collaborative inspection planning and scheduling method in a smart grid scenario as described in the above embodiment.
[0018] Therefore, this application has at least the following beneficial effects:
[0019] (1) Before performing task allocation, the embodiment of the present application first uses a cluster analysis method to cluster multiple drones, and comprehensively considers distance and task time to determine the starting position of the drone, thereby shortening the calculation time of task allocation, improving allocation efficiency, and saving time and resources.
[0020] (2) The embodiment of the present application introduces task priority, distinguishes different tasks and drones according to the different endurance of drones and the time required for tasks, comprehensively considers the impact of drone performance and task execution time, and realizes the allocation of all tasks.
[0021] (3) Using the UAV task allocation method based on deep reinforcement learning, for high-priority tasks, the task planning model is trained by combining the path length and comprehensively considering the inspection sequence; adding UAVs and task states in the state space can quickly adapt to changes in quantity. After training, a task allocation plan for inspection is finally given to complete the priority inspection of the power grid by UAVs, realize fast and efficient inspection of smart grids, minimize time and energy consumption costs, avoid waste of resources caused by repeated execution of tasks, and at the same time improve the task completion rate and balance the UAV load, providing better support for power grid inspection.
[0022] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through the practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS
[0023] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:
[0024] Figure 1 A flowchart of a method for planning and scheduling multi-UAV collaborative inspection in a smart grid scenario according to an embodiment of the present application;
[0025] Figure 2 A framework diagram of a method for scheduling inspection of unmanned aerial vehicles provided according to an embodiment of the present application;
[0026] Figure 3 A flowchart of DRL model training provided according to an embodiment of the present application;
[0027] Figure 4 is a block diagram of a multi-UAV collaborative inspection planning and scheduling device in a smart grid scenario according to an embodiment of the present application;
[0028] Figure 5 Schematic diagram of the structure of a server according to an embodiment of the present application. DETAILED DESCRIPTION
[0029] The embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.
[0030] With the continuous development of smart grids, power grid equipment is widely distributed in various scenarios such as mountainous areas, forests, rural areas and towns. In order to ensure the effective operation of huge infrastructure, regular inspections of various types of equipment are essential. However, the traditional manual inspection method has a low level of informatization, making it difficult to effectively integrate data, which limits the improvement of inspection efficiency. By utilizing the advantages of drone inspections such as wide field of view and high flexibility, equipment defects and hidden dangers can be quickly detected. Through the automated drone inspection process, not only can labor costs be significantly reduced, but work efficiency can also be improved.
[0031] At the same time, as users continue to improve the reliability of power supply, the reliability and safety standards of the power supply system are also constantly improving. In order to ensure the safe operation and stability of power grid equipment in various environments, systematic management and accurate inspections are particularly critical. In this process, the clarity of tasks and the reasonable order become crucial links. By formulating clear inspection goals and priority plans, combined with appropriate technical means, potential problems can be more effectively discovered and preventive maintenance measures can be taken. Considering the high requirements of power grid equipment for safety and stability, the clarity and order of tasks become the key to achieving this goal. Clear inspection tasks and reasonable execution order can not only improve the reliability of power grid equipment, but also provide a solid foundation for responding to the continuous evolution of power demand and environmental changes. In the future, as the level of power supply continues to improve, power grid managers and operation and maintenance teams need to pay close attention to the clarification of tasks and the formulation of order to ensure that power grid equipment operates under high safety and stability requirements and provide users with more reliable power supply services.
[0032] Autonomous decision-making and control enable drones to perform tasks independently, and mission planning is the key to achieving autonomous decision-making and control of multiple drones. The collaborative mission planning of multiple drones must not only consider the requirements of the mission itself, but also the constraints of the collaborative execution of tasks by multiple drones, and design collaborative flight paths for drones based on the characteristics of the mission to optimize or approach the optimal overall performance. Therefore, drone mission planning is a multi-objective optimization and decision-making problem with many constraints, high complexity, and strong coupling. When assigning tasks to multiple drones, all tasks need to be divided into individual drones, and the order in which each drone performs tasks must be determined. During the task scheduling process, if all tasks cannot be completed due to certain constraints, the number of tasks in the execution task list may be less than the number of tasks.
[0033] There are currently a variety of algorithms to solve the problem of drone mission planning. Traditional algorithms perform well in simple environments, but are less effective in complex scenarios. Using deep reinforcement learning algorithms for mission planning has strong versatility and can better solve problems. For example, researchers have proposed an intelligent method for task allocation based on deep reinforcement learning algorithms, which allows agents to learn how to allocate tasks on site in order to optimize the execution time of tasks under energy constraints.
[0034] In order to solve the multi-objective task allocation problem in smart grids, multiple UAVs are allocated and executed in parallel and synchronously according to constraints, which reduces the overall time and avoids unnecessary waste of resources caused by repeated execution of tasks.
[0035] Therefore, this application aims at the problem of UAV task scheduling and allocation in power grid scenarios, and proposes a method for optimizing the allocation of power grid equipment inspection tasks based on a deep reinforcement learning algorithm. In this application, by considering the starting point of the UAV and the initial placement of the UAV, it is hoped to reduce the flight distance and time of the UAV, improve the inspection efficiency, and save time and resources. This method can comprehensively consider the task priority, the performance of the UAV and the impact of the task execution time, and realize the allocation of all tasks. The goal of this application is to achieve fast and efficient inspection of smart grids, minimize time and energy costs, avoid waste of resources caused by repeated execution of tasks, and at the same time improve the task completion rate and balance the UAV load, so as to provide better support for power grid inspection.
[0036] The following describes the method, device, vehicle and storage medium for planning and scheduling multi-UAV collaborative inspection in a smart grid scenario in an embodiment of the present application with reference to the accompanying drawings. Specifically, Figure 1 A flowchart of a method for multi-UAV collaborative inspection planning and scheduling in a smart grid scenario provided in an embodiment of the present application.
[0037] like Figure 1 As shown, the multi-UAV collaborative inspection planning and scheduling method in the smart grid scenario is applied to the server, wherein the method includes the following steps:
[0038] In step S101, at least one inspection task of a smart grid scenario and a plurality of drones that perform the inspection task are obtained.
[0039] It can be understood that the embodiments of the present application can obtain at least one inspection task in a smart grid scenario and multiple drones that perform inspection tasks, so as to facilitate the subsequent identification of the status information and task priority of each inspection task and the status information of each drone.
[0040] It should be noted that the inspection task can be for drones to take photos or videos of regional power grid equipment.
[0041] In step S102, the status information and task priority of each inspection task and the status information of each drone are identified.
[0042] Among them, the status information of each inspection task includes the task location and task time, the status information of each drone includes the drone location and flight time, and the task priority is determined based on the dynamic characteristics of power load fluctuations in the smart grid scenario.
[0043] It can be understood that the embodiments of the present application can identify the status information and task priority of each inspection task and the status information of each drone, so as to generate scheduling instructions for each drone based on the status information and task priority of each inspection task and the status information of each drone.
[0044] It should be noted that before task allocation, a cluster analysis method is first used to cluster multiple drones, and the starting position of the drone is determined by comprehensively considering the distance and task time, so as to shorten the calculation time of task allocation and improve the allocation efficiency. At the same time, the priority of the task is introduced to distinguish different tasks and drones according to the different endurance of the drones and the time required for the task; a drone task allocation method based on deep reinforcement learning is used, and for high-priority tasks, the inspection order is comprehensively considered in combination with the path length.
[0045] Specifically, this application combines the dynamic characteristics of power load fluctuations to determine the priority of tasks. First, missing values and outliers are processed. Considering that power load data has temporal continuity as a time series, it is necessary to supplement missing values and eliminate outliers. This application uses the mean filling method to supplement missing and eliminated data.
[0046] The level of inspection tasks is determined based on the current real-time data of power load. In smart grids, power load may change rapidly due to the continuous access of renewable energy and the dynamic nature of the power system. Adjusting the priority of equipment based on power load conditions helps to respond more flexibly to changes in energy demand. At the same time, user experience is closely related to power load.
[0047] By setting the priority of power load, users' satisfaction with power services can be improved. The greater the fluctuation, the higher the level. UAV inspections prioritize the power grid according to the degree. The tasks are divided into three levels, based on the real-time load data of the power system; when the power load fluctuation value ΔP fluctuates by ΔP≤±5%, ±5%<ΔP≤±10%, ΔP>±10% compared with the previous time period, corresponding to levels Ⅰ, Ⅱ, and Ⅲ respectively, the greater the fluctuation, the higher the level. Equipment or areas with large load fluctuations are listed as high priority, indicating that priority inspections are needed to determine whether the equipment status in the area can support large power load fluctuations.
[0048] In step S103, a scheduling instruction for each drone is generated according to the status information and task priority of each inspection task and the status information of each drone, and multiple drones are scheduled to perform at least one inspection task based on the scheduling instruction of each drone.
[0049] It can be understood that the embodiments of the present application can generate scheduling instructions for each drone based on the status information and task priority of each inspection task and the status information of each drone, and schedule multiple drones to perform at least one inspection task based on the scheduling instructions of each drone, comprehensively considering the impact of drone performance, task priority and task execution time, to achieve the allocation of all tasks and fast and efficient inspection of smart grids, minimize time and energy costs, avoid waste of resources caused by repeated task execution, while improving the task completion rate and balancing the drone load, providing better support for power grid inspection.
[0050] In an embodiment of the present application, a scheduling instruction for each UAV is generated based on the status information and task priority of each inspection task and the status information of each UAV, including: determining the starting position of the UAV based on clustering the status information of each inspection task; inputting the starting position of the UAV, the status information and task priority of each inspection task and the status information of each UAV into a task planning model, and the task planning model outputs the scheduling instruction for each UAV.
[0051] It can be understood that the embodiments of the present application can determine the starting position of the UAV based on the status information cluster of each inspection task, and input the starting position of the UAV, the status information and task priority of each inspection task, and the status information of each UAV into the task planning model. The task planning model outputs the scheduling instructions for each UAV, comprehensively considering the impact of UAV performance, task priority, and task execution time, to achieve the allocation of all tasks and fast and efficient inspection of the smart grid, minimize time and energy consumption costs, while improving the task completion rate and balancing the UAV load.
[0052] In an embodiment of the present application, the starting position of the drone is determined based on clustering of status information of each inspection task, including: determining the task distance based on the task position of each inspection task; generating a feature matrix based on the task distance of each inspection task, and generating a new feature matrix based on the feature matrix and the task time; performing distance clustering on the new feature matrix based on a distance loss function to obtain a clustering result, and determining the starting point position of the drone based on the clustering result.
[0053] It can be understood that the clustering method of the embodiment of the present application comprehensively considers the task execution time and task distance to determine the optimal drone starting point, so as to subsequently output the scheduling instructions for each drone based on the starting position of the drone, the status information and task priority of each inspection task, and the status information of each drone input into the task planning model.
[0054] Specifically, the starting position of the drone is determined by clustering the status information of each inspection task, and the clustering method that comprehensively considers the task execution time and task distance is used to determine the optimal drone starting point. Specifically:
[0055] K-means is used to cluster the mission targets, and the starting point of the drone is determined by combining the mission location distance and the time required to execute the mission. Assuming that there are n samples in the sample space, recorded as X = {x 1 ,x 2 ,…,x n}; There are k cluster centers in total, each cluster is denoted by C = {c 1 ,c 2 ,…,c k}.
[0056] Integrate the distance features of the task points into a feature matrix.
[0057]
[0058] Each row represents a sample (task point), and there are n tasks in total; each column represents a feature, and there are m distance features in total. i,j Represents the distance feature between the i-th sample point and the j-th sample point.
[0059] In order to consider the execution time of the task itself, a feature column is added to the feature matrix to indicate the execution time required for each task point; thus, the feature matrix becomes:
[0060]
[0061] Since the locations and execution time units of the task points are different, the data needs to be standardized to ensure that they are on the same scale, which helps to avoid a single feature from having too much influence on the clustering results. The elbow rule is used to select the optimal number of clusters; then the K-means clustering algorithm and the determined number of clusters are used to cluster the task points.
[0062] Finally, the samples are clustered into k clusters so that the loss function of the clustering result reaches the minimum value; the loss function can be expressed as:
[0063]
[0064] in, represents the centroid of the ith cluster, which is the average value of all samples in the cluster, ω is the weight of the task execution time, and t i is the execution time of the task point, It is cluster c i The average execution time of the task points.
[0065] The whole algorithm process is to randomly initialize k centroids first, and then iterate continuously until the centroids no longer change. Each iteration requires recalculating the clusters to which each sample point belongs and the centroids of each cluster. The final cluster division is the result with the minimum loss function; that is, the determined starting point of the drone.
[0066] In an embodiment of the present application, the task planning module is used to: traverse each inspection task in turn, generate a state vector based on the status information of the current inspection task, the task priority and the status information of each drone; generate the action probability of each drone selecting the current inspection task based on the state vector and the reward function, and generate the scheduling instructions for the drone executing the current inspection task based on the action probability until all inspection tasks are traversed.
[0067] It can be understood that the embodiments of the present application can generate a state vector based on the status information of the current inspection task, the task priority and the status information of each drone, and generate the action probability of each drone selecting the current inspection task based on the state vector and the reward function, and generate the scheduling instructions for the drone executing the current inspection task based on the action probability, combine the frequency of decision-making with the task completion status, optimize the key time steps, reduce resource consumption, and finally output the optimal task planning plan, which can enable the drone to complete the priority task with the minimum time and energy consumption.
[0068] Specifically, the DRL (Deep Reinforcement Learning) module observes the current state, obtains the drone state information and task state information to generate the state vector s t ,s t =(s U ,Q i ,s P ,PI i ,T pi ,s T ), s U Contains all drone locations Q i is the flight time of the drone, s P Contains all drone locations PI i Indicates the priority of each task, T pi Indicates the time required to perform the task, s T Represents the status of each task s T ={s 1 ,…,s i ,…,s n}, the executed tasks are set to 1, and the unexecuted ones are set to 0.
[0069] The output of the module is the action a performed by DRLt ={a 1 ,…,a i ,…,a m}, where a i represents the task selected by the i-th UAV {P 1 ,P 2 ,…,P n}, each drone U i The task with the highest output probability is selected for execution until all tasks are traversed, and the UAV scheduling strategy is finally output, and the task is completed. After the inspection task is completed, samples are randomly selected from the experience pool to train the DRL model; ultimately, the UAV can complete the priority tasks with the minimum time and energy consumption, and no tasks are missed.
[0070] In the embodiment of the present application, the reward function r t It is expressed as:
[0071] r t =r s +r E +r p
[0072] Among them, r S is the time reward, r S =-T total ×ε,0<ε<1,T total The total task completion time, ε represents the impact of task completion time on time reward; r E is the energy consumption reward, r E =-E total ×μ, 0<μ<1, E total is the total communication energy consumption, μ represents the influence of task energy consumption on energy consumption reward; r p As a reward for the task, Among them, n represents the total number of tasks, C represents the basic reward of each task, PIi represents the priority index of the task, k represents the number of completed tasks, and J represents the degree of task completion.
[0073] It can be understood that the task reward in the embodiment of the present application is based on a reward function that changes as the task starts to be executed. The reward function also changes with the execution of the task according to the different priorities, so as to facilitate the subsequent comprehensive consideration of the influencing factors of priority and flight path to determine the order of drone inspections.
[0074] Specifically, this application designs a reward function r based on the priority inspection scenario t , including time reward r s , Energy consumption reward E And the task reward p , specifically:
[0075] Time Reward s It measures the completion time of all tasks. The time required for a drone to complete an inspection task can be divided into three types: flight time, inspection time, and communication time. The flight time of a drone starting from the starting point and reaching the task point to be performed is the flight time. d i is the flight distance of the drone, v max is the flight speed of the drone; inspection time T pi The time for the drone to inspect the mission point.
[0076] Assuming that during the UAV inspection data transmission process, the data receiving rate is R, then the transmission rate D ni Communication time of data volume A single UAV U i The total time to complete a task is: T si =T fi +T comi +T pi , the time when the last UAV returns to the starting point is taken as the total mission completion time T total , then the time reward r S =-T total ×ε, 0<ε<1, ε represents the influence of task completion time on time reward.
[0077] Energy consumption reward E is the total energy consumption of all drones during the drone inspection process, and the energy consumption per time slot is e f , time consuming T fi , then the flight energy consumption is E f =e f ×T fi ; Assume the energy consumption of taking photos is e p / photos, need to take x photos i The energy consumption of taking photos is E pi =e p × i ; The energy consumption when transmitting data is e com / Mb, image size is S comi , then the communication energy consumption is E comi =e com ×S comi ; Single UAV U i The total energy consumed in one mission is: E si =E fi +E comi +E pi ; The total energy consumption of all drones is recorded as E total , r E =-E total×μ, 0<μ<1, μ represents the influence of task energy consumption on energy consumption reward.
[0078] Mission Rewards p :Evaluate the coverage of all target points by the drone during inspection missions, and give corresponding rewards based on the priority and inspection order, which can change based on the execution time of the mission.
[0079]
[0080] Among them, n represents the total number of tasks, C represents the basic reward of each task, PI i represents the priority index of the task, k represents the number of completed tasks, and J represents the task completion degree. The task completion degree J is expressed as the proportion of the tasks completed by all drones in all tasks, which is used to evaluate the coverage of all target points by drones when completing inspection tasks.
[0081]
[0082] Among them, m is the number of drones, n is the total number of tasks, P i Represents the amount of tasks completed by the i-th UAV.
[0083] In summary, the reward function r t Expressed as: r t =r s +r E +r p .
[0084] In an embodiment of the present application, before the starting position of the drone, the status information and task priority of each inspection task, and the status information of each drone are input into the task planning model, it also includes: obtaining a training data set, wherein the training data set includes the status information and task priority of multiple task samples, and the status information of the drone; traversing multiple task samples in sequence, generating a state vector of the current task sample based on the status information and task priority of the current task sample, and the status information of the drone, and generating an action of the current task sample based on the state vector, wherein the action of the current task sample represents the probability of selecting each drone to select the current task sample; obtaining the reward and next-moment state vector of each drone performing the action, generating a training sample based on the state vector, action, reward and next-moment state vector of the current task sample, and placing the training sample into the experience pool until all task samples are traversed.
[0085] It can be understood that the embodiments of the present application can traverse multiple task samples in sequence, generate a state vector of the current task sample based on the state information and task priority of the current task sample and the state information of the drone, generate the action of the current task sample based on the state vector, obtain the reward and the next-moment state vector of each drone performing the action, generate training samples based on the state vector, action, reward and next-moment state vector of the current task sample, and put the training samples into the experience pool until all task samples are traversed, thereby training the DRL model based on the data in the experience pool to achieve autonomous learning of the DRL scheduling module to improve performance.
[0086] Specifically, (1) the DRL model obtains the starting point of the drone determined by the clustering results; (2) the DRL model obtains the inspection time T of the current task p , task location Mission priority PI, UAV endurance Q; (3) Form the input vector s of the DRL model at time t t =(s U ,Q i ,s P ,PI i ,T pi ,s T );(4) Generates t Action t ={a 1 ,…,a i ,…,a m}; (5) Execute the action and obtain the reward r of the environment feedback t , and enter the next state s t+1 ; (6) Forming four-tuple training samples t ,a t ,r t ,s t+1 >Save to the experience pool; (7) Determine whether the task is completed. If so, proceed to the next step; otherwise, return to step (1); (9) Randomly sample from the experience pool for DRL model training.
[0087] According to the collaborative inspection planning and scheduling of multiple drones in a smart grid scenario proposed in an embodiment of the present application, at least one inspection task in the smart grid scenario and multiple drones that perform the inspection tasks are obtained, the status information and task priority of each inspection task and the status information of each drone are identified, and a scheduling instruction for each drone is generated based on the status information and task priority of each inspection task and the status information of each drone. Multiple drones are scheduled to perform at least one inspection task based on the scheduling instruction of each drone, and the impact of drone performance, task priority and task execution time are comprehensively considered to achieve the allocation of all tasks and fast and efficient inspection of the smart grid, minimize time and energy costs, avoid waste of resources caused by repeated execution of tasks, and at the same time improve the task completion rate and balance the drone load, thereby providing better support for power grid inspection.
[0088] The following will be combined Figure 2 and Figure 3 The planning and scheduling of multi-UAV collaborative inspection in the smart grid scenario of this application are described in detail as follows:
[0089] Assume that the smart grid currently needs drones to inspect the equipment in the area, and the following conditions are met during the inspection process: the equipment is randomly distributed in the drone service area, all equipment can be inspected by drones, and the drones take photos or videos of the equipment during the inspection to monitor the equipment; the control center uniformly allocates the task list to be performed by the drone throughout the process, and the drone immediately returns the data after completing the inspection task (taking photos or videos) of the current task point; during the entire inspection process, the communication status in the area is good, and the problem of being unable to perform inspection work and data transmission due to network problems is not considered. Therefore, this application comprehensively considers the drone and task status of the drone inspection scheduling method, specifically:
[0090] (1) The overall structure of this application: Figure 2 As shown in the figure, first, before allocating tasks, the cluster analysis method is used to cluster multiple drones, and the starting position of the drone is determined by combining the distance and task time to shorten the calculation time of task allocation and improve the allocation efficiency. At the same time, the priority of the task is introduced, and different tasks and drones are distinguished according to the different endurance of the drones and the different time required for the task; the drone task allocation method based on deep reinforcement learning is used. For high-priority tasks, the inspection order is comprehensively considered in combination with the path length; drones issue tasks and upload status through the same operation control center; adding drones and task status in the state space can quickly adapt to changes in quantity. After training, a task allocation plan for inspection is finally given to complete the priority inspection of the power grid by drones.
[0091] (2) Determine the priority of tasks based on the dynamic characteristics of power load fluctuations
[0092] First, the missing values and outliers are processed. Considering that the power load data as a time series has temporal continuity, it is necessary to supplement the missing values and remove the outliers. The present invention adopts the mean filling method to supplement the missing and removed data; according to the current real-time data of the power load, the level of the inspection task is determined. In the smart grid, due to the continuous access of renewable energy and the dynamic nature of the power system, the power load may change rapidly; adjusting the priority of the equipment according to the power load situation helps to respond to changes in energy demand more flexibly;
[0093] At the same time, user experience is closely related to power load; by setting the priority of power load, the user's satisfaction with power service is improved; the greater the fluctuation, the higher the level, and the drone inspection will prioritize the power grid according to the degree; the task is divided into three levels, according to the real-time load data of the power system; when the power load fluctuation value ΔP fluctuates by ΔP≤±5%, ±5%<ΔP≤±10%, ΔP>±10% compared with the previous time period, corresponding to levels Ⅰ, Ⅱ, and Ⅲ respectively, the greater the fluctuation, the higher the level, and equipment or areas with a larger degree of load fluctuation are listed as high priority, indicating that priority inspection is needed to determine whether the equipment status in the area can support larger power load fluctuations.
[0094] (3) A clustering method that comprehensively considers mission execution time and mission distance to determine the optimal UAV starting point
[0095] K-means is used to cluster the mission targets, and the starting point of the drone is determined by combining the mission location distance and the time required to execute the mission. Assuming that there are n samples in the sample space, recorded as X = {x 1 ,x 2 ,…,x n}; There are k cluster centers in total, each cluster is denoted by C = {c 1 ,c 2 ,…,c k}.
[0096] Integrate the distance features of the task points into a feature matrix.
[0097]
[0098] Each row represents a sample (task point), and there are n tasks in total; each column represents a feature, and there are m distance features in total; x i,j Represents the distance feature between the i-th sample point and the j-th sample point.
[0099] In order to consider the execution time of the task itself, a feature column is added to the feature matrix to indicate the execution time required for each task point; the feature matrix is:
[0100]
[0101] Since the locations and execution time units of the task points are different, the data needs to be standardized to ensure that they are on the same scale, which helps to avoid a single feature from having too much influence on the clustering results. The elbow rule is used to select the optimal number of clusters; then the K-means clustering algorithm and the determined number of clusters are used to cluster the task points.
[0102] Finally, the samples are clustered into k clusters so that the loss function of the clustering result reaches the minimum value. The loss function can be expressed as:
[0103]
[0104] in, represents the centroid of the ith cluster, which is the average value of all samples in the cluster. ω is the weight of the task execution time, t i is the execution time of the task point, It is cluster c i The average execution time of the task points.
[0105] The whole algorithm process is to randomly initialize k centroids first, and then iterate continuously until the centroids no longer change. Each iteration requires recalculating the clusters to which each sample point belongs and the centroids of each cluster. The final cluster division is the result with the minimum loss function; that is, the determined starting point of the drone.
[0106] (4) Application of DRL’s task planning module, which is used to allocate inspection tasks and plan the inspection sequence
[0107] Assume there are m drones, and the drone set U = {U 1 ,U 2 ,…,U m ,…,U M}, UAV U m Location coordinates UAV maximum linear speed v max .
[0108] There are n mission points in the drone’s service area, and the mission set P = {P 1 ,P 2 ,…,P n ,…,P N}, Task P n Location coordinates Task inspection time T pi , the priority of the inspection task PI i It is assumed that there is no interdependency between tasks, i.e. each task can be executed independently of other tasks.
[0109] UAV i Comprehensive evaluation index Q i , which is an indicator of UAV performance and reliability, is used to evaluate the working ability and flight safety of UAVs, resulting in different abilities to perform tasks. Whether a UAV can perform the tasks that need to be completed is determined based on comprehensive evaluation indicators.
[0110] At each time step t, the DRL model obtains the drone state information and task state information to generate a state vector s t , generates action a according to its own strategy t , when performing action a t Then get reward r from the environment t , and transfer to the next state s t+1 .
[0111] This application designs a reward function r based on the priority inspection scenario t , including time reward r s , Energy consumption reward E And the task reward p , specifically:
[0112] Time Reward s It measures the completion time of all tasks. The time required for a drone to complete an inspection task can be divided into three types: flight time, inspection time, and communication time. The flight time of a drone starting from the starting point and reaching the task point to be performed is the flight time. d i is the flight distance of the drone, v max is the flight speed of the drone; inspection time T pi The time for the drone to inspect the mission point.
[0113] Assuming that during the UAV inspection data transmission process, the data receiving rate is R, then the transmission rate D ni Communication time of data volume A single UAV U i The total time to complete a task is: T si =T fi +T comi +T pi , the time when the last UAV returns to the starting point is taken as the total mission completion time T total , then the time reward r S =-T total ×ε, 0<ε<1, ε represents the influence of task completion time on time reward.
[0114] Energy consumption reward E is the total energy consumption of all drones during the drone inspection process, and the energy consumption per time slot is e f, time consuming T fi , then the flight energy consumption is E f =e f ×T fi ; Assume the energy consumption of taking photos is e p / photos, need to take x photos i The energy consumption of taking photos is E pi =e p × i ; The energy consumption when transmitting data is e com / Mb, image size is S comi , then the communication energy consumption is E comi =e com ×S comi ; Single UAV U i The total energy consumed in one mission is: E si =E fi +E comi +E pi ; The total energy consumption of all drones is recorded as E total , r E =-E total ×μ, 0<μ<1, μ represents the influence of task energy consumption on energy consumption reward.
[0115] Mission Rewards p :Evaluate the coverage of all target points by the drone during inspection missions, and give corresponding rewards based on the priority and inspection order, which can change based on the execution time of the mission.
[0116]
[0117] Among them, n represents the total number of tasks, C represents the basic reward of each task, PI i represents the priority index of the task, k represents the number of completed tasks, and J represents the task completion degree. The task completion degree J is expressed as the proportion of the tasks completed by all drones in all tasks, which is used to evaluate the coverage of all target points by drones when completing inspection tasks:
[0118]
[0119] Among them, m is the number of drones, n is the total number of tasks, P i Represents the amount of tasks completed by the i-th UAV.
[0120] In summary, the reward function r t Expressed as: r t =r s +r E +r p .
[0121] (5) DRL model training process in power grid inspection scenario
[0122] First, the starting position of each UAV generated by clustering is obtained, as well as the task priority determined according to the real-time power load data.
[0123] The DRL module observes the current state, obtains the drone state information and mission state information, and generates a state vector s t ,s t =(s U ,Q i ,s P ,PI i ,T pi ,s T ), s U Contains all drone locations Q i is the flight time of the drone, s P Contains all drone locations PI i Indicates the priority of each task, T pi Indicates the time required to perform the task, s T Represents the status of each task s T ={s 1 ,…,s i ,…,s n}, the executed tasks are set to 1, and the unexecuted ones are set to 0.
[0124] The output of the module is the action a performed by DRL t ={a 1 ,…,a i ,…,a m}, where a i represents the task selected by the i-th UAV {P 1 ,P 2 ,…,P n}, each drone U i The task with the highest output probability is selected for execution until all tasks are traversed, and the UAV scheduling strategy is finally output, and the task is completed. After completing the inspection task, samples are randomly selected from the experience pool to train the DRL model. Ultimately, the UAV can complete the priority tasks with the minimum time and energy consumption, and no tasks are missed.
[0125] The training process is as follows Figure 3 As shown in Figure 1, (1) the DRL model obtains the starting point of the drone determined by the clustering results; (2) the DRL model obtains the inspection time T of the current task. p , task location Mission priority PI, UAV endurance Q; (3) Form the input vector s of the DRL model at time tt =]s U ,Q i ,s P ,PI i ,T pi ,s T );(4) Generates t Action t ={a 1 ,…,a i ,…,a m}; (5) Execute the action and obtain the reward r of the environment feedback t , and enter the next state s t+1 ; (6) Forming four-tuple training samples t ,a t ,r t ,s t+1 >Save to the experience pool; (7) Determine whether the task is completed. If so, proceed to the next step; otherwise, return to step (1); (9) Randomly sample from the experience pool for DRL model training.
[0126] In summary, the present invention fully considers the starting point position and task priority of the drone, and combines the DRL technology to achieve the optimal inspection plan through the task allocation module, and improves the inspection efficiency and reliability of the smart grid by optimizing the drone power grid inspection scheduling strategy; when considering the task priority, the system will prioritize the tasks according to their urgency and importance. This helps to ensure that under limited resources, the tasks with the greatest impact on the operation of the power grid are processed first, thereby improving the overall operating efficiency. For the starting point position of the drone, the system will comprehensively consider the equipment distribution and drone differences to minimize the flight distance and time of the drone. The DRL task allocation module first obtains the previous starting point results, and adjusts the inspection plan by analyzing the real-time status information of the drone and the task, including the remaining battery capacity, task completion progress, etc. This enables the system to cope with different environmental conditions and task changes, and improve the adaptability and robustness of the smart grid.
[0127] The entire solution of the present application will be described below in conjunction with specific embodiments, specifically:
[0128] With the continuous development of smart grids, the expansion of scale and the increase of task complexity, drone power grid inspections have gradually replaced traditional manual inspection methods. Drones are widely used in various fields due to their high flexibility, small size, and strong environmental adaptability. Drone task planning technology is the key to achieving autonomous decision-making and control of multiple drones. Multi-drone collaborative task planning must also consider the constraints of multiple drones in coordinating and performing tasks together. The present invention designs a drone priority inspection scheduling method based on deep reinforcement learning. This method first reduces the flight time and distance of the drone by determining the task priority and determining the drone starting point by clustering, thereby improving inspection efficiency and saving time and resources.
[0129] Assume that within a certain smart grid area, drones need to inspect the equipment in the area. The grid equipment in the area is randomly distributed, and all equipment can be inspected by drones. All equipment only needs to be inspected once. Drones can perform multiple inspection tasks, but can only complete the inspection of a single device at the same time. The initial position of the drone is random. The inspection process is to detect the grid equipment by taking photos or videos of the equipment by the drone. Task allocation first requires the drone's operation control center to obtain the status information of the drone and the task, and then the task allocation module performs model training and finally outputs the task planning plan. The drone immediately transmits data back after completing the photo or video task of the current task point. Assuming that the communication status in the inspection area is good, the situation where the drone cannot complete the task due to network problems is not considered.
[0130] During the inspection process, the time required for the drone to complete the inspection task is mainly divided into three types: flight time T fi , Inspection time T pi , communication time T comi . Single UAV U i The total time to complete a task is: T si =T fi +T comi +T pi The longest time to complete the task among all drones is taken as the total completion time of the task T total , the total completion time of the UAV should be kept small.
[0131] Correspondingly, the energy consumption at each stage is: Flight energy consumption E fi , Photographing energy consumption E pi , communication energy consumption E com . Single UAV U i The total energy consumed in one mission is: E si =E fi +E comi +E pi The total energy consumption of all drones is recorded as E total , The final task allocation plan enables the drone to complete all inspection tasks with minimal time and energy consumption.
[0132] The task priority and the relevant status information of the drone and the task are integrated to form a state vector, and the corresponding action is output according to the current state vector, that is, the task planning plan for the current task. In order to achieve the priority task allocation of drones under the conditions of minimizing energy consumption and time, DRL is used to train the model; the reward function of this module takes into account the impact of priority on the inspection order, and uses the autonomous learning mechanism of the DRL algorithm. After training, the model will adjust the parameters in the direction of maximizing the reward function, that is, shorten the overall task completion time as much as possible, and be able to better inspect the operating status of the power grid, achieving the expected effect of the present invention.
[0133] In combination with the above embodiments, the present application (1) integrates the UAV clustering process of distance and task time and the UAV task allocation based on DRL. The inspection operation control center collects historical environmental data of power grid equipment, determines the starting point of the UAV and the task priority, and performs task planning through DRL according to the status information of the UAV and the task, and determines the inspection task and inspection order;
[0134] (2) Prioritize equipment based on power load so that priority can be taken into account when assigning UAV tasks. Due to the dynamic nature of the power system, power load may change rapidly. Adjusting the priority of equipment based on power load conditions helps to respond more flexibly to changes in energy demand.
[0135] (3) The drone clustering process that integrates distance and task time. Before assigning drone tasks, the tasks are clustered to determine the starting point. Then, the drones are placed at the cluster center point as the starting point of the drone inspection according to the principle of matching the number of tasks. This can effectively reduce the flight distance and time of the drones, improve the inspection efficiency, and save time and resources.
[0136] (4) A DRL-based task planning method is used to allocate inspection tasks and plan the inspection order. It uses drone and task information to form a state vector, comprehensively considers the influencing factors of priority and flight path, and determines the order of drone inspection. The present invention designs a reward function for priority-based drone inspection, including time reward, energy consumption reward and task reward. Among them, the task reward is a reward function that changes with the execution of the task; the reward function that changes with the execution of the task is generated according to the priority;
[0137] (5) For the state space of drone task allocation, by adding drone and task status information to the input state of deep reinforcement learning, the execution status of the task is marked to indicate whether the task has been assigned to a drone for execution, thereby avoiding resource waste and time consumption caused by task duplication. And the drone's mark position marks whether the drone is currently idle. During the process, when it is observed that a drone has completed the task, the next decision is triggered. In this way, the frequency of decision-making can be combined with the task completion status, the critical time step can be optimized, and resource consumption can be reduced. The final output generates the optimal task planning solution;
[0138] (6) DRL model training process. This process describes the training process of the task allocation module. It includes obtaining the clustered drone starting point and task priority information, the current task and drone status information, combining them to form a state vector, and using DRL to generate corresponding actions; the environment feedbacks rewards based on the action and transfers to the next state, completing a time step, and generating a training sample and storing it in the experience pool. When the task is completed, the module extracts samples from the experience pool for model training, enabling autonomous learning of the DRL scheduling module to improve performance.
[0139] Secondly, refer to the accompanying drawings to describe a multi-UAV collaborative inspection planning and scheduling device in a smart grid scenario proposed according to an embodiment of the present application.
[0140] Figure 4 It is a block diagram of a multi-UAV collaborative inspection planning and scheduling device in a smart grid scenario in an embodiment of the present application.
[0141] like Figure 4 As shown, a multi-UAV collaborative inspection planning and scheduling device 10 in the smart grid scenario is applied to a server, wherein the device includes: an acquisition module 100, an identification module 200 and a scheduling module 300.
[0142] Among them, the acquisition module 100 is used to acquire at least one inspection task in the smart grid scenario and multiple drones that perform the inspection tasks; the identification module 200 is used to identify the status information and task priority of each inspection task and the status information of each drone; the scheduling module 300 is used to generate a scheduling instruction for each drone based on the status information and task priority of each inspection task and the status information of each drone, and schedule multiple drones to perform at least one inspection task based on the scheduling instruction of each drone.
[0143] In an embodiment of the present application, the status information of each inspection task includes the task location and task time, the status information of each drone includes the drone location and flight time, and the task priority is determined based on the dynamic characteristics of power load fluctuations in the smart grid scenario.
[0144] In an embodiment of the present application, the scheduling module 300 is further used to: determine the starting position of the UAV based on the status information clustering of each inspection task; input the starting position of the UAV, the status information and task priority of each inspection task, and the status information of each UAV into the task planning model, and the task planning model outputs the scheduling instructions for each UAV.
[0145] In an embodiment of the present application, the task planning module is used to: traverse each inspection task in turn, generate a state vector based on the status information of the current inspection task, the task priority and the status information of each drone; generate the action probability of each drone selecting the current inspection task based on the state vector and the reward function, and generate the scheduling instructions for the drone executing the current inspection task based on the action probability until all inspection tasks are traversed.
[0146] In the embodiment of the present application, the reward function r t It is expressed as:
[0147] r t =r s +r E +r p
[0148] Among them, r S is the time reward, r S =-T total ×ε,0<ε<1,T total The total task completion time, ε represents the impact of task completion time on time reward; r E is the energy consumption reward, r E =-E total ×μ, 0<μ<1, E total is the total communication energy consumption, μ represents the influence of task energy consumption on energy consumption reward; r p As a reward for the task, Among them, n represents the total number of tasks, C represents the basic reward of each task, PIi represents the priority index of the task, k represents the number of completed tasks, and J represents the degree of task completion.
[0149] In an embodiment of the present application, it also includes: a traversal module, which is used to obtain a training data set, wherein the training data set includes status information and task priorities of multiple task samples, and status information of the drone; traversing multiple task samples in sequence, generating a state vector of the current task sample based on the status information and task priority of the current task sample, and the status information of the drone, and generating an action of the current task sample based on the state vector, wherein the action of the current task sample represents the probability of selecting each drone to select the current task sample; obtaining the reward and next-moment state vector of each drone performing the action, generating a training sample based on the state vector, action, reward and next-moment state vector of the current task sample, and placing the training sample into the experience pool until the traversal of all task samples is completed.
[0150] In an embodiment of the present application, the scheduling module 300 is further used to: determine the task distance based on the task location of each inspection task; generate a feature matrix based on the task distance of each inspection task, and generate a new feature matrix based on the feature matrix and the task time; perform distance clustering on the new feature matrix based on the distance loss function to obtain a clustering result, and determine the starting point position of the drone based on the clustering result.
[0151] It should be noted that the above explanation of the embodiment of the multi-UAV collaborative inspection planning and scheduling method in the smart grid scenario is also applicable to the multi-UAV collaborative inspection planning and scheduling device in the smart grid scenario of this embodiment, and will not be repeated here.
[0152] According to the multi-UAV collaborative inspection planning and scheduling device in the smart grid scenario proposed in the embodiment of the present application, at least one inspection task in the smart grid scenario and multiple UAVs that perform the inspection tasks are obtained, the status information and task priority of each inspection task and the status information of each UAV are identified, and the scheduling instructions for each UAV are generated according to the status information and task priority of each inspection task and the status information of each UAV. Based on the scheduling instructions of each UAV, multiple UAVs are scheduled to perform at least one inspection task, and the impact of UAV performance, task priority and task execution time are comprehensively considered to achieve the allocation of all tasks and the fast and efficient inspection of the smart grid, minimize time and energy consumption costs, avoid waste of resources caused by repeated execution of tasks, and at the same time improve the task completion rate and balance the UAV load, thereby providing better support for power grid inspection.
[0153] Figure 5 A schematic diagram of the structure of a server provided in an embodiment of the present application. The server may include:
[0154] A memory 501 , a processor 502 , and a computer program stored in the memory 501 and executable on the processor 502 .
[0155] When the processor 502 executes the program, the multi-UAV collaborative inspection planning and scheduling method in the smart grid scenario provided in the above embodiment is implemented.
[0156] Furthermore, the server also includes:
[0157] The communication interface 503 is used for communication between the memory 501 and the processor 502 .
[0158] The memory 501 is used to store computer programs that can be executed on the processor 502 .
[0159] The memory 501 may include a high-speed RAM (Random Access Memory) memory, and may also include a non-volatile memory, such as at least one disk memory.
[0160] If the memory 501, the processor 502 and the communication interface 503 are implemented independently, the communication interface 503, the memory 501 and the processor 502 can be connected to each other through a bus and communicate with each other. The bus can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 5 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.
[0161] Optionally, in a specific implementation, if the memory 501, the processor 502 and the communication interface 503 are integrated on a chip, the memory 501, the processor 502 and the communication interface 503 can communicate with each other through an internal interface.
[0162] The processor 502 may be a CPU (Central Processing Unit), or an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the embodiments of the present application.
[0163] An embodiment of the present application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-mentioned multi-UAV collaborative inspection planning and scheduling method in a smart grid scenario.
[0164] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms are not necessarily directed to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or N embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.
[0165] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of the features. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise clearly and specifically defined.
[0166] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, fragment or portion of code comprising one or N executable instructions for implementing the steps of a custom logical function or process, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in reverse order depending on the functions involved, which should be understood by technicians in the technical field to which the embodiments of the present application belong.
[0167] It should be understood that the various parts of the present application can be implemented by hardware, software, firmware or a combination thereof. In the above embodiment, the N steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one or a combination of multiple of the following technologies known in the art: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array, a field programmable gate array, etc.
[0168] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.
[0169] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations on the present application. Ordinary technicians in this field can change, modify, replace and modify the above embodiments within the scope of the present application.
Claims
1. A multi-UAV collaborative inspection planning and scheduling method in a smart grid scenario, characterized in that: The method is applied to a server, wherein the method comprises the following steps: Obtain at least one inspection task of a smart grid scenario and multiple drones that perform the inspection tasks; Identify the status information and task priority of each inspection task and the status information of each drone; A scheduling instruction for each drone is generated according to the status information and task priority of each inspection task and the status information of each drone, and the multiple drones are scheduled to perform the at least one inspection task based on the scheduling instruction of each drone.
2. The method for planning and scheduling multi-UAV collaborative inspection in a smart grid scenario according to claim 1 is characterized in that: The status information of each inspection task includes the task location and task time, the status information of each drone includes the drone location and flight time, and the task priority is determined based on the dynamic characteristics of power load fluctuations in the smart grid scenario.
3. The method for planning and scheduling multi-UAV collaborative inspection in a smart grid scenario according to claim 1 is characterized in that: The generating of the scheduling instruction for each UAV according to the status information and the task priority of each inspection task and the status information of each UAV includes: Determine the starting position of the UAV according to the clustering of the status information of each inspection task; The starting position of the UAV, the status information and task priority of each inspection task, and the status information of each UAV are input into the task planning model, and the task planning model outputs the scheduling instructions for each UAV.
4. The method for planning and scheduling multi-UAV collaborative inspection in a smart grid scenario according to claim 3 is characterized in that: The mission planning module is used to: Traversing each inspection task in turn, generating a state vector according to the state information of the current inspection task, the task priority and the state information of each drone; The action probability of each drone selecting the current inspection task is generated according to the state vector and the reward function, and the scheduling instruction of the drone executing the current inspection task is generated according to the action probability until all inspection tasks are traversed.
5. The method for planning and scheduling multi-UAV collaborative inspection in a smart grid scenario according to claim 4 is characterized in that: The reward function r t It is expressed as: r t =r s +r E +r p Among them, r S is the time reward, r S =-T total ×ε,0<ε<1,T total The total task completion time, ε represents the impact of task completion time on time reward; r E is the energy consumption reward, r E =-E total ×μ, 0<μ<1, E total is the total communication energy consumption, μ represents the influence of task energy consumption on energy consumption reward; r p As a reward for the task, Among them, n represents the total number of tasks, C represents the basic reward of each task, PI i represents the priority index of the task, k represents the number of completed tasks, and J represents the degree of task completion.
6. The method for planning and scheduling multi-UAV collaborative inspection in a smart grid scenario according to claim 3 is characterized in that: Before inputting the starting position of the drone, the status information and the task priority of each inspection task, and the status information of each drone into the task planning model, the method further includes: Acquire a training data set, wherein the training data set includes status information and task priorities of multiple task samples, and status information of a drone; Traversing the multiple task samples in sequence, generating a state vector of the current task sample according to the state information and task priority of the current task sample and the state information of the drone, and generating an action of the current task sample based on the state vector, wherein the action of the current task sample represents the probability of selecting each drone to select the current task sample; Obtain the reward and next-moment state vector of each drone performing the action, generate training samples according to the state vector, action, reward and next-moment state vector of the current task sample, and put the training samples into the experience pool until all task samples are traversed.
7. The method for planning and scheduling multi-UAV collaborative inspection in a smart grid scenario according to claim 3 is characterized in that: The step of determining the starting position of the drone according to the clustering of the status information of each inspection task includes: Determine the task distance according to the task location of each inspection task; Generate a feature matrix based on the task distance of each inspection task, and generate a new feature matrix based on the feature matrix and the task time; The new feature matrix is clustered based on the distance loss function to obtain a clustering result, and the starting point position of the drone is determined according to the clustering result.
8. A multi-UAV collaborative inspection planning and scheduling device in a smart grid scenario, characterized in that: The device is applied to a server, wherein the device comprises: An acquisition module, used to acquire at least one inspection task of a smart grid scenario and multiple drones that perform the inspection task; An identification module is used to identify the status information and task priority of each inspection task and the status information of each drone; A scheduling module is used to generate a scheduling instruction for each drone according to the status information and task priority of each inspection task and the status information of each drone, and schedule the multiple drones to perform the at least one inspection task based on the scheduling instruction of each drone.
9. A server, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the multi-UAV collaborative inspection planning and scheduling method in a smart grid scenario as described in any one of claims 1 to 7.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: The program is executed by a processor to implement a multi-UAV collaborative inspection planning and scheduling method in a smart grid scenario as described in any one of claims 1-7.
Citation Information
Cited By
Oil depot tank field inspection robot task allocation method and system based on Internet of Things
CN119417192A
Inspection task allocation method and device, processor and readable storage medium
CN119514978A
Multi-unmanned aerial vehicle cooperative power reinspection method and system based on communication technology
CN119597018A
Electric power inspection unmanned aerial vehicle autopilot deployment scheduling method, system, device and medium
CN120163307A
Power inspection drone autopilot deployment and scheduling method, system, equipment and medium
CN120163307B
