Planning method for unmanned aerial vehicle to monitor, operate and maintain power transmission line

Abstract

The invention relates to a planning method for monitoring, operating and maintaining a power transmission line by an unmanned aerial vehicle, and relates to the technical field of unmanned aerial vehicle planning. A scene model of a scene where a power transmission line is located is created; modeling the unmanned aerial vehicle; constructing a planner by using a language model, and controlling the planner through cue words to generate an action for controlling the unmanned aerial vehicle to go to a specified power transmission line area in a scene model according to current and preorder unmanned aerial vehicle states, the scene model, actions of each unmanned aerial vehicle in a preorder unmanned aerial vehicle group and experience of a filtering experience pool; the planner is controlled to optimize a local inspection route of unmanned aerial vehicle inspection in each power transmission line area with the purpose of minimizing the inspection flight cost; unreasonable task actions and polling paths formed by illusion of the planner are filtered through the planning filter to avoid unsafe or practical behaviors, and filtering experience generated by the planning filter is provided to the language model to allow the language model to continuously improve decisions based on the filtering experience.

Description

Technical Field

[0001] The present invention relates to the technical field of drone inspection and scheduling, and in particular to a planning method for drone monitoring and maintenance of power transmission lines. Background Art

[0002] With the rapid development of drone technology, drone inspection, as an emerging technology, has been widely used in the field of transmission line monitoring and operation and maintenance.

[0003] Existing drone inspection systems typically consist of a drone platform, a ground control station, and image recognition modules. The drone platform, equipped with sensors such as a high-definition camera, inspects power lines using pre-set routes or manual control. The ground control station is responsible for drone takeoff and landing, route planning, and data transmission. The image recognition module automatically identifies and detects anomalies in images captured by the drone.

[0004] However, existing drone inspection systems still have some problems and shortcomings. First, drone inspection routes are relatively fixed, regardless of whether normal operations or abnormalities occur. When a drone detects an anomaly around a transmission line during an inspection, it typically needs to record and transmit the anomaly type and complete the subsequent inspection route. The entire drone system lacks an intelligent linkage mechanism for abnormalities, requiring human intervention in drone inspection control based on the recorded anomaly. This results in insufficient flexibility and adaptability to abnormalities caused by intelligent planning. To address these issues, some researchers have proposed transmission line inspection systems based on multi-drone collaboration. These systems utilize multi-drone collaboration technology to enable information sharing and collaborative operations among drones. However, these systems still have some shortcomings. For example, in terms of anomaly detection and response, although information sharing is achieved among drones, the response mechanism after an anomaly detection remains relatively simple, lacking intelligent path planning and autonomous control capabilities.

[0005] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present disclosure, and therefore may include information that does not constitute prior art known to ordinary technicians in the field. Summary of the Invention

[0006] In order to solve the above technical problems or at least partially solve the above technical problems, the present invention provides a planning method for drone monitoring and operation and maintenance of power transmission lines.

[0007] The present invention provides a planning method for monitoring and maintaining power transmission lines by using a drone, comprising: Use raster maps to create digital scene models that describe the environment in which the transmission lines are located; Modeling of UAVs used for transmission line monitoring and maintenance from four aspects: function, operating status, energy consumption, and battery. A planner is constructed using a language model. Prompt words control the planner to generate conditional probabilities for controlling drones to move toward a specific transmission line area in the scenario model based on the current and previous drone states, the scenario model, the actions of each drone in the previous drone group, and the experience of a filtered experience pool. The control planner optimizes the local inspection routes of drones within each transmission line area with the goal of minimizing inspection flight costs. A reinforcement learning-based planning filter is used to filter out unreasonable task actions and inspection paths formed by the planner's hallucinations to avoid behaviors that are not safe or practical. The filtered experience content generated by the planning filter is provided to the language model, allowing the language model to refer to past filtering experience to continuously improve its decision-making.

[0008] Furthermore, the grid map of the scene model contains several grids, and each grid is configured with the following attributes: altitude attribute, no-fly zone, entity attribute, and entity status attribute. The environmental information is represented by the grid and the grid attributes. Among them, the altitude attribute configured by the grid describes the altitude information of the environmental entity corresponding to the grid, and the grid position and its altitude attribute fully describe the structure of the environmental entity; the no-fly zone attribute configured by the grid describes whether the environmental area corresponding to the grid is no-fly zone; the entity attribute configured by the grid describes the type of environmental entity corresponding to the grid; and the entity status attribute configured by the grid can describe the status of the entity corresponding to the grid.

[0009] Furthermore, the UAV for transmission line monitoring and operation is modeled from four aspects: function, operating status, energy consumption and battery. Among them, the function description attribute of the UAV model describes the transmission line monitoring and operation function supported by the UAV; the UAV operating status attribute describes the UAV position and the UAV in charging, working or idle state; energy consumption modeling is performed for the two states of hovering and propulsion; the battery model includes: battery discharge power and SOC constraint; according to the energy conservation principle and energy conversion efficiency, , x=1,2, where is the battery energy conversion efficiency, is the battery discharge power, is the non-driving power consumption, when x=1, is the power of the drone in hovering state, when x=2, The UAV propulsion power.

[0010] Furthermore, the power in hovering state is: ; in, Indicates the hovering drag power, which is caused by the friction resistance between the blades and the air when the rotor rotates, and is related to the blade drag coefficient , air density , number of blades N, blade chord length , speed Cubic and blade radius Proportional to the fourth power; It represents the hovering induced power, which is the energy required to offset the gravity by accelerating the air downward to generate a reaction force. mg is the gravity on the drone. is the induced power correction factor.

[0011] Furthermore, the power in the propulsion state is: ; in, is the drag power correction factor for propulsion compared to hovering state, is the induced power correction factor for propulsion compared to hovering state, is the hovering induced speed; The drag power of level flight is the power required to overcome the aerodynamic resistance of the fuselage when the UAV is in level flight, and is related to the fuselage drag coefficient , air density , number of blades N, blade chord length Blade radius Proportional to the cube of the speed.

[0012] Furthermore, the prompt words include: The meaning description of the scenario model and the current scenario model; it is required to understand the information of the scenario in which the transmission line is located based on the provided scenario model and the meaning description of the scenario model; UAV location, UAV status, understanding the UAV situation based on the UAV location and UAV status; Filtering the experience pool; Responsible for drone planning in a scene. The task is to determine the drone's mission action based on the experience of the filtered experience pool, the current scene model, the drone's position and the drone's status; ensure that the selected mission action complies with the no-fly and drone energy constraints; Minimize inspection flight costs and plan inspection routes based on the experience pool, current transmission line area conditions, and drone status. Ensure the planned routes comply with no-fly restrictions and drone energy constraints. If the task action and path may violate any constraints, return a null decision instead of an unsafe decision; and interact with the planning filter to filter the task actions and paths.

[0013] Furthermore, planning filters based on reinforcement learning include critic networks and actor networks; During training, the critic network and actor network are trained according to the drone planning process. The drone planning is formulated as a constrained Markov decision process, defined as follows: (S, A, r, P, γ, Ω), where S represents the state space, including: scene model, drone position, battery power, drones currently assigned to the power line area, and accumulated inspection distance; A represents the action space, including commands to assign drones to the power line area, return to the drone station, and switch between charging and idle states; r is the reward function based on state and action, P corresponds to the state transition probability, γ is the discount factor, and Ω is a set of constraints that the drone must meet; constraints are mandatory requirements for the drone to perform operations, and a cost function is constructed for each constraint. The planning filter maximizes the reward and adheres to the constraints through the following Lagrangian dual optimization: ; in, characterizes the decision made, λ ≥ 0 as the Lagrange multiplier, is the safety threshold; Actor Network Generating Actions , the critic network evaluates the expected return and associated constraint costs , the policy gradient is updated as follows: ; Where θ is the policy parameter; The actions formed by the rule filter interact with the scene model and the drone model to generate a new state space, and store (S, A, r, P, γ, Ω) in the filtered experience pool.

[0014] Furthermore, during training, if the cost of an action selected by the actor network exceeds a safety threshold, a safe alternative action will replace the original action.

[0015] Furthermore, during the filtering process, the actions and paths generated by the planner are directly evaluated by the trained critic network with respect to the associated constraint costs. , compared with the safety threshold to achieve action and path filtering generated by the planner.

[0016] Furthermore, the experience corresponding to the filtered action paths is stored in the filtered experience pool, and the planner makes decision improvements based on the experience in the filtered experience pool.

[0017] The above technical solution provided by the embodiment of the present invention has the following advantages compared with the prior art: The present application creates a scenario model of the scenario in which the transmission line is located; models the drone; uses a language model to build a planner, and controls the planner through prompt words to generate actions to control the drone to go to the specified transmission line area in the scenario model based on the current and previous drone states, scenario models, the actions of each drone in the previous drone group, and the experience of the filtered experience pool; the control planner optimizes the local inspection route of the drone inspection in each transmission line area with the goal of minimizing the inspection flight cost; uses a planning filter to filter the unreasonable task actions and inspection paths formed by the planner's hallucinations to avoid behaviors that are not in line with safety or reality, and provides the filtered experience generated by the planning filter to the language model, allowing the language model to continuously improve its decision-making based on the filtered experience. The present application uses a planning filter based on reinforcement learning to filter the unreasonable task actions and inspection paths formed by the planner's hallucinations to avoid behaviors that are not in line with safety or reality, and provides the filtered experience content generated by the planning filter to the language model, allowing the language model to refer to past filtered experience to continuously improve its decision-making. Combining the generalization advantages of the language model with the adaptability and real-time learning capabilities of reinforcement learning, it ensures that intelligent drone planning is generated in the dynamic environment of transmission line inspection. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0019] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0020] Figure 1 This is a flowchart of a planning method for monitoring and maintaining power transmission lines using drones, as provided in the disclosed embodiment of the present invention.

[0021] Figure 2 A schematic diagram of a scene model provided in accordance with an embodiment of the present invention.

[0022] Figure 3 A schematic diagram of a network architecture provided for an embodiment of the present invention.

[0023] Figure 4 A schematic diagram of a planning device for monitoring and maintaining power transmission lines using a drone according to an embodiment of the present invention. DETAILED DESCRIPTION

[0024] To make the purpose, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0025] It should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or apparatus comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or apparatus comprising the element.

[0026] Example 1 See Figure 1 As shown, in order to solve the problem that conventional fixed path planning cannot adapt to environmental dynamics and abnormal situations, an embodiment of the present invention provides a planning method for UAV monitoring and maintenance of power transmission lines, comprising: Create a digital scenario model describing the environment surrounding a power transmission line. This environment includes the transmission line, transmission equipment, drone stations, and environmental entities that could potentially obstruct drones. The drone stations are equipped with drone charging equipment for landing and recharging. An example scenario model is created based on a grid map. The grid map of the scenario model contains several grids. To better describe the environment, each grid is assigned the following attributes: altitude, no-fly zone, entity attributes, and entity status attributes. Grids and their attributes represent environmental information.

[0027] Specifically, the height attribute configured by the grid describes the altitude information of the environmental entity corresponding to the grid. The grid position and its height attribute fully describe the structure of the environmental entity, which is of great significance for UAV planning and obstacle avoidance; the no-fly attribute configured by the grid describes whether the environmental area corresponding to the grid is no-fly; the entity attribute configured by the grid describes the type of environmental entity corresponding to the grid. The entity attribute can be used to effectively describe the area where the transmission line, transmission equipment, and UAV station are located. In the specific implementation process, the entity attribute is used to divide the scene where the transmission line is located into multiple transmission line areas containing the points to be measured. For any transmission line area i, it is set to contain The entity status attributes configured by the grid can describe the status of the entity corresponding to the grid, especially the entities that can interact with the drone, such as whether the charging station entity in the drone station is occupied, whether the transmission line has completed the inspection, and the abnormal status of the transmission equipment of the transmission line.

[0028] Compared with traditional grid maps, the scene model of this application introduces new attributes to the grid. In the specific practice process, the scene model is represented by a multi-channel high-dimensional matrix. The height and width of the matrix correspond to the grid arrangement. Each channel represents the situation of an attribute, one channel represents the height, one channel represents whether it is a no-fly zone, one channel represents the entity type, and multiple channels represent the entity status. In the specific implementation process, whether it is a no-fly zone and the entity status use 0 and 1 to represent logical yes and no, and the entity type and entity status are digitally encoded. The matrixed scene model can be subsequently perceived and understood by the language model of the planner.

[0029] In the transmission line monitoring and operation and maintenance scenario, a variety of drones with different functions may be used. In order to achieve intelligent planning and scheduling, the drones are modeled.

[0030] The constructed drone model includes the following: Functional description attributes: The functional description attributes of the drone model describe the transmission line monitoring and operation and maintenance functions supported by the drone. When dispatching drones based on exceptions, drones with corresponding functional description attributes are dispatched based on the abnormal state in the scenario model.

[0031] Drone operation status attributes,Drone operation status attributes describe the drone location and whether the drone is charging, working or idle.

[0032] For a system consisting of M drones, an M-dimensional three-channel tensor is constructed to model the functional description and operating status of each drone. One channel consists of the functional description encoding of the M drones, one channel consists of the location encoding of the M drones, and one channel consists of the charging, working, or idle status encoding of the M drones.

[0033] UAV energy consumption model, in order to accurately describe the energy consumption of the UAV under different working conditions, this application divides the working conditions into two states: hovering and propulsion, and builds energy consumption models for the two states.

[0034] The power model in the hovering state is: ; in, Indicates the hovering drag power, which is caused by the friction resistance between the blades and the air when the rotor rotates, and is related to the blade drag coefficient , air density , number of blades N, blade chord length , speed Cubic and blade radius It is proportional to the fourth power, reflecting the influence of air, rotor geometry and rotation speed on energy consumption in the hovering state.

[0035] It represents the hovering induced power, which is the energy required to generate a reaction force downward by accelerating the air to offset the gravity. mg is the gravity on the drone. is the induced power correction coefficient, used to correct non-ideal flow field losses, k∈[0.1,0.3].

[0036] The power in propulsion state is: ; in, The drag power correction factor for the propulsion system is used to correct the drag power of the rotor in level flight, taking into account the effect of forward speed on the blade load distribution. , The speed in is ignored and becomes , which is consistent with the drag power in the hovering state. At high speeds, the drag power increases significantly with increasing speed, and the drag increases nonlinearly due to the local speed differences of the blades.

[0037] It is the correction factor of the induced power in the propulsion ratio compared to the hovering state. Based on the forward ratio theory, it corrects the change of the induced power with the speed v in the level flight of the UAV. is the hovering induction speed. , Approximately degenerates into hovering induced power . At high speed : The induced power decreases as the speed v increases because the forward speed reduces the air flow that needs to be accelerated.

[0038] is the drag power of level flight, which is the power required to overcome the aerodynamic resistance of the fuselage when the UAV is in level flight. , air density , number of blades N, blade chord length Blade radius Proportional to the cube of the speed.

[0039] Battery model: The drone is powered by batteries, and the battery discharge power follows: ;in, is the open circuit voltage related to SOC, The internal resistance related to SOC; when the battery is charging and discharging, SOC follows: .

[0040] According to the law of conservation of energy and energy conversion efficiency, , x=1,2. is the battery energy conversion efficiency, is the battery discharge power, is the non-driving power consumption, when x=1, is the power of the drone in hovering state, when x=2, The UAV propulsion power.

[0041] The drone model combines the energy consumption model and the battery model to provide effective constraints for drone planning in transmission line monitoring and operation scenarios.

[0042] A planner is built using language models to support flexible drone planning in power line monitoring and maintenance scenarios. Examples of these language models include Qwen32B, gpt-4o, and deepseek.

[0043] The planner is controlled by prompt words to generate the conditional probability of the task action of controlling the drone to go to the specified transmission line area in the scene model based on the current and previous drone states, the scene model, the actions of each drone in the previous drone group, and the experience of the filtered experience pool: .in, represents the mission actions of each drone in the preceding drone group before time t, represents the predicted task action at time t, and X is the current and previous drone states, scene model, and filtered experience pool experience.

[0044] For any transmission line area i, set it to contain points to be tested, forming an arrangement sequence The drone inspects the test points in the order they are arranged. A planner is then required to optimize the local inspection routes of the drone within each transmission line area with the goal of minimizing the inspection flight cost. The inspection flight cost is calculated using the drone's energy consumption model, battery model, and planned routes.

[0045] In a specific implementation process, the content of the prompt words used by the control planner to perform planning includes: "The meaning description of the scenario model and the current scenario model; it is required to understand the information of the scenario in which the transmission line is located based on the provided scenario model and the meaning description of the scenario model.

[0046] UAV functions and status: it is required to understand the UAV situation based on the UAV functions and status; Filtering the experience pool; Responsible for drone planning within a scenario, this task involves determining the drone's action based on the filtered experience pool, the current scenario model, the drone's position, and its state. This task ensures that the selected action complies with the no-fly requirements and energy constraints in the scenario model. Based on the filtered experience pool, the current transmission line area, and the drone's state, the inspection route is planned by minimizing the flight cost, ensuring that the planned route complies with the no-fly requirements and energy constraints in the scenario model. If the action and path may violate any constraints, a null decision is returned. This task also interacts with planning filters to filter the action and path. This application uses a reinforcement learning-based planning filter to filter out unreasonable task actions and inspection paths generated by the planner's hallucinations, avoiding unsafe or unrealistic behaviors. The filter's filtered experience is then fed into the language model, allowing the language model to reference past filtering experience to continuously improve its decisions. Combining the generalization advantages of the language model with the adaptability and real-time learning capabilities of reinforcement learning, this ensures intelligent drone planning in the dynamic environment of power transmission line inspections.

[0047] In the specific implementation process, the planning filter based on reinforcement learning includes a critic network and an actor network.

[0048] During training, the critic and actor networks are trained based on the drone plan. Drone planning is formulated as a constrained Markov decision process, defined as follows: (S, A, r, P, γ, Ω), where S represents the state space, A represents the action space, r is the reward function based on the state and action, P corresponds to the state transition probability, γ ∈ [0, 1] is the discount factor, and Ω is a set of constraints that the drone must satisfy. Specifically, the state space includes the scene model, drone state, battery charge, drones currently assigned to the power line area, and accumulated inspection distance. The action space includes commands for assigning drones to power line areas, returning to drone stations, and switching between charging and idle states. Rewards are calculated based on flight costs. Constraints enforce operational requirements on the drone, such as battery conservation and route efficiency. A cost function c(s, a) is constructed for each constraint.

[0049] The planning filter maximizes the reward and adheres to the constraints through the following Lagrangian dual optimization: ; in, characterizes the decision made, λ ≥ 0 as the Lagrange multiplier, is the safety threshold.

[0050] Actor Network Generating Actions , the critic network evaluates the expected return and associated constraint costs The policy gradient update is performed as follows: ; where θ is the policy parameter. If the cost of the selected action exceeds a safety threshold, a safe alternative action will replace the original action.

[0051] The actions formed by the rule filter interact with the scene model and the drone model to generate a new state space, and store (S, A, r, P, γ, Ω) in the filtered experience pool.

[0052] During the filtering process, the actions and paths generated by the planner are directly evaluated by the critic network with the associated constraint costs. , compared with the safety threshold and filtered. The (S, A, r, P, γ, Ω) corresponding to the filtered action path is stored in the filtering experience pool.

[0053] Example 2 See Figure 4 As shown, the present invention provides a planning device for monitoring and maintaining power transmission lines by drones, comprising: at least one processing unit, the processing unit being connected to a storage unit via a bus unit, the storage unit storing a computer program, and the computer program, when executed by the processing unit, implementing the multi-level power optimization allocation and equipment selection method for the energy and power system, comprising: Use raster maps to create digital scene models that describe the environment in which the transmission lines are located; Modeling of UAVs used for transmission line monitoring and maintenance from four aspects: function, operating status, energy consumption, and battery. A planner is constructed using a language model. Prompt words control the planner to generate conditional probabilities for controlling drones to move toward a specific transmission line area in the scenario model based on the current and previous drone states, the scenario model, the actions of each drone in the previous drone group, and the experience of a filtered experience pool. The control planner optimizes the local inspection routes of drones within each transmission line area with the goal of minimizing inspection flight costs. A reinforcement learning-based planning filter is used to filter out unreasonable task actions and inspection paths formed by the planner's hallucinations to avoid behaviors that are not safe or practical. The filtered experience content generated by the planning filter is provided to the language model, allowing the language model to refer to past filtering experience to continuously improve its decision-making.

[0054] Of course, the computer program stored in the storage unit of the planning device for monitoring and maintaining transmission lines by drones provided in an embodiment of the present invention is not limited to the operations of the method described above, and can also execute the relevant operations in the planning method for monitoring and maintaining transmission lines by drones provided in any embodiment of the present invention.

[0055] Example 3 An embodiment of the present invention provides a computer-readable storage medium storing computer instructions. When the computer instructions are executed by a processor, the method for planning the operation and maintenance of a power transmission line using a drone is implemented, including: Use raster maps to create digital scene models that describe the environment in which the transmission lines are located; Modeling of UAVs used for transmission line monitoring and maintenance from four aspects: function, operating status, energy consumption, and battery. A planner is constructed using a language model. Prompt words control the planner to generate conditional probabilities for controlling drones to move toward a specific transmission line area in the scenario model based on the current and previous drone states, the scenario model, the actions of each drone in the previous drone group, and the experience of a filtered experience pool. The control planner optimizes the local inspection routes of drones within each transmission line area with the goal of minimizing inspection flight costs. A reinforcement learning-based planning filter is used to filter out unreasonable task actions and inspection paths formed by the planner's hallucinations to avoid behaviors that are not safe or practical. The filtered experience content generated by the planning filter is provided to the language model, allowing the language model to refer to past filtering experience to continuously improve its decision-making.

[0056] Of course, the computer-readable storage medium provided by an embodiment of the present invention stores a computer program that is not limited to the method operations described above, but can also execute related operations in a planning method for drone monitoring and operation of transmission lines provided by any embodiment of the present invention.

[0057] In the embodiments provided by the present invention, it should be understood that the disclosed structures and methods can be implemented in other ways. For example, the structural embodiments described above are merely illustrative. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interface, structure or unit, which can be electrical, mechanical or other forms.

[0058] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0059] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0060] The foregoing description is intended only to provide specific embodiments of the present invention, which will enable those skilled in the art to understand and implement the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not intended to be limited to the embodiments shown herein, but is intended to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A planning method for monitoring and maintaining power transmission lines by drones, characterized in that: include: Use raster maps to create digital scene models that describe the environment in which the transmission lines are located; Modeling of UAVs used for transmission line monitoring and maintenance from four aspects: function, operating status, energy consumption, and battery. A planner is constructed using a language model. Prompt words control the planner to generate conditional probabilities for controlling drones to move toward a specific transmission line area in the scenario model based on the current and previous drone states, the scenario model, the actions of each drone in the previous drone group, and the experience of a filtered experience pool. The control planner optimizes the local inspection routes of drones within each transmission line area with the goal of minimizing inspection flight costs. A reinforcement learning-based planning filter is used to filter out unreasonable task actions and inspection paths formed by the planner's hallucinations to avoid behaviors that are not safe or practical. The filtered experience content generated by the planning filter is provided to the language model, allowing the language model to refer to past filtering experience to continuously improve its decision-making.

2. The planning method for monitoring and maintaining power transmission lines by drones according to claim 1, characterized in that: The grid map of the scene model contains several grids, and each grid is configured with the following attributes: altitude attribute, no-fly zone, entity attribute, and entity status attribute. The grid and grid attributes are used to represent environmental information. Among them, the altitude attribute configured by the grid describes the altitude information of the environmental entity corresponding to the grid, and the grid position and its altitude attribute fully describe the structure of the environmental entity. The no-fly zone attribute configured by the grid describes whether the environmental area corresponding to the grid is no-fly zone. The entity attribute configured by the grid describes the type of environmental entity corresponding to the grid. The entity status attribute configured by the grid can describe the status of the entity corresponding to the grid.

3. The planning method for monitoring and maintaining power transmission lines by drones according to claim 1, characterized in that: The UAV used for transmission line monitoring and operation is modeled from four aspects: function, operating status, energy consumption and battery. The function description attribute of the UAV model describes the transmission line monitoring and operation function supported by the UAV. The UAV operating status attribute describes the UAV position and whether the UAV is charging, working or idle. Energy consumption modeling is performed for the two states of hovering and propulsion. The battery model includes: battery discharge power and SOC constraints. According to the law of conservation of energy and energy conversion efficiency, , x=1,2, where is the battery energy conversion efficiency, is the battery discharge power, is the non-driving power consumption, when x=1, is the power of the drone in hovering state, when x=2, The UAV propulsion power.

4. The planning method for monitoring and maintaining power transmission lines by drones according to claim 3, characterized in that: The power in hovering state is: ; in, Indicates the hovering drag power, which is caused by the friction resistance between the blades and the air when the rotor rotates, and is related to the blade drag coefficient , air density , number of blades N, blade chord length , speed Cubic and blade radius Proportional to the fourth power; It represents the hovering induced power, which is the energy required to offset the gravity by accelerating the air downward to generate a reaction force. mg is the gravity on the drone. is the induced power correction factor.

5. The planning method for monitoring and maintaining power transmission lines by drones according to claim 3, characterized in that: The power in propulsion state is: ； in, is the drag power correction factor for propulsion compared to hovering state, is the induced power correction factor for propulsion compared to hovering state, is the hovering induced speed; The drag power of level flight is the power required to overcome the aerodynamic resistance of the fuselage when the UAV is in level flight, and is related to the fuselage drag coefficient , air density , number of blades N, blade chord length Blade radius Proportional to the cube of the speed.

6. The planning method for monitoring and maintaining power transmission lines by drones according to claim 1, characterized in that: The prompt words include: The meaning description of the scenario model and the current scenario model; it is required to understand the information of the scenario in which the transmission line is located based on the provided scenario model and the meaning description of the scenario model; UAV functions and UAV status: Understand the UAV situation based on the UAV functions and UAV status; Filtering the experience pool; Responsible for drone planning in a scenario. The task is to determine the drone's mission action based on the experience of the filtered experience pool, the current scenario model, the drone's position and the drone's status; ensure that the selected mission action complies with the no-fly requirements in the scenario model and the drone's energy constraints; Minimize inspection flight costs and plan inspection routes based on the filtered experience pool, current transmission line area conditions, and drone status. Ensure the planned routes comply with the no-fly requirements and drone energy constraints in the scenario model. If the task action and path may violate any constraints, return a null decision instead of an unsafe decision; and interact with the planning filter to filter the task actions and paths.

7. The planning method for monitoring and maintaining power transmission lines by drones according to claim 1, characterized in that: Reinforcement learning-based planning filters include critic networks and actor networks; During training, the critic network and actor network are trained according to the drone planning process. The drone planning is formulated as a constrained Markov decision process, defined as follows: (S, A, r, P, γ, Ω), where S represents the state space, including: scene model, drone state, battery level, drones currently assigned to the power line area, and accumulated inspection distance; A represents the action space, including commands to assign drones to the power line area, return to the drone station, and switch between charging and idle states; r is the reward function based on state and action, P corresponds to the state transition probability, γ is the discount factor, and Ω is a set of constraints that the drone must meet; constraints are mandatory requirements for the drone to perform operations, and a cost function is constructed for each constraint. The planning filter maximizes the reward and adheres to the constraints through the following Lagrangian dual optimization: ； in, characterizes the decision made, λ ≥ 0 as the Lagrange multiplier, is the safety threshold; Actor Network Generating Actions , the critic network evaluates the expected return and associated constraint costs , the policy gradient is updated as follows: ； Where θ is the policy parameter; The actions formed by the rule filter interact with the scene model and the drone model to generate a new state space, and store (S, A, r, P, γ, Ω) in the filtered experience pool.

8. The planning method for monitoring and maintaining power transmission lines by drones according to claim 7, characterized in that: During training, if the cost of an action selected by the actor network exceeds a safety threshold, a safe alternative action replaces the original action.

9. The planning method for monitoring and maintaining power transmission lines by drones according to claim 7, characterized in that: During filtering, the actions and paths generated by the planner are directly evaluated by the trained critic network with respect to the constraint costs. , compared with the safety threshold to achieve action and path filtering generated by the planner.

10. The planning method for monitoring and maintaining power transmission lines by drones according to claim 9, characterized in that: The (S, A, r, P, γ, Ω) corresponding to the filtered action path is stored in the filtered experience pool, and the planner makes decision improvements based on the experience of the filtered experience pool.

Images