Unmanned aerial vehicle early warning platform and method based on distribution network tower

Abstract

The invention relates to an unmanned aerial vehicle early warning platform and method based on a distribution network tower. The platform comprises an unmanned aerial vehicle body, an adjusting mechanism, a first clamping arm and a second clamping arm. The adjusting mechanism drive is used for adjusting the included angle between the movable ends of the first clamping arm and the second clamping arm. The first clamping arm and the second clamping arm are symmetrically installed on the unmanned aerial vehicle body and are each of a pi-shaped structure, and the middle section of each pi-shaped structure is of an arc-shaped structure matched with the outer contour of a distribution network tower and used for clamping the distribution network tower. A binocular camera, a control module and an acousto-optic integrated alarm are arranged on the unmanned aerial vehicle body, the binocular camera obtains information of the agricultural robot and the surrounding environment and transmits the information to the control module, the control module evaluates the contact risk of the agricultural unmanned aerial vehicle and a wire, and when the contact risk exists, the acousto-optic integrated alarm is started to warn the agricultural unmanned aerial vehicle. And related information is stored. Compared with the prior art, the method has the advantages of long endurance time of the unmanned aerial vehicle, high adaptability, high detection accuracy and the like.

Description

Technical Field

[0001] This invention relates to the field of power grid protection technology, and in particular to a drone early warning platform and method based on distribution network towers. Background Technology

[0002] In recent years, with the rapid advancement of smart agriculture, agricultural drones, with their advantages of high efficiency and precision, have been widely used in all aspects of agricultural production, including sowing, fertilization, plant protection, and pest and disease monitoring. Currently, the agricultural drone market is experiencing an average annual growth rate of over 30%, covering more than 50% of the country's arable land. Especially in large-scale planting areas, drones can perform tens of thousands of sorties daily, making them one of the core pieces of equipment in modern agricultural production.

[0003] However, at the same time, the "machine-line conflict" between agricultural drone operations and rural power grid distribution lines is intensifying, gradually becoming a key hidden danger threatening the safe operation of the rural power grid. Rural power grid distribution lines are mostly distributed in fields, and some lines have problems such as low pole height and unclear line markings due to their long construction years; while the operating height of agricultural drones is mostly concentrated between 5-30 meters, which highly overlaps with the safe distance of distribution lines.

[0004] In actual operation, several factors can easily lead to safety accidents: First, most drone operators are farmers or grassroots workers, and some of them lack awareness of the safe distance of power distribution lines, which can easily cause the drone to approach the line at close range due to operational errors; Second, complex field environments such as strong electromagnetic interference and tree obstruction may cause the drone's GPS signal to become inaccurate or out of control, thus deviating from the preset flight path and colliding with the line; Third, different crop growth cycles, such as the wheat heading stage and the fruit tree peak production period, will change the required operating height, further compressing the safety buffer space between the drone and the line.

[0005] Currently, traditional monitoring methods for rural power grid lines are insufficient to address the new challenges posed by conflicts between drones and power lines. Existing monitoring mainly relies on manual inspections and fixed video surveillance: manual inspections are limited by terrain and manpower, resulting in narrow coverage, low inspection frequency, and slow emergency response, and cannot capture the dynamic risks of drones in real time; while fixed video surveillance can achieve real-time monitoring of local areas, it is limited by installation location and field of view, making it difficult to cover all distribution network lines in vast farmlands, and it lacks intelligent identification and early warning functions, failing to proactively identify drones and avoid risks in advance.

[0006] For example, the invention with publication number CN112214032A provides a drone inspection system and method. The drone inspection system automatically detects hot spot phenomena on solar modules. Utilizing a GPS system and thermal imaging devices, it solves the problems of time-consuming and labor-intensive traditional manual inspection, achieving efficient detection and maintenance. It identifies abnormal objects in target areas and captures corresponding images of these abnormalities. However, drones have limitations in terms of battery life and are not suitable for inspecting the basic wiring of agricultural drones.

[0007] For example, the invention with publication number CN118579286A involves a drone inspection device and a drone inspection system. By designing a protection module with a sliding plate and a linkage plate on the drone, combined with a hollow roller and a landing pole, it solves the problem that the camera cannot be protected from the side in the prior art. However, it still does not solve the problem of poor battery life of drone inspection.

[0008] Against this backdrop, building an intelligent early warning system capable of long-term, comprehensive monitoring of rural power grid lines has become an urgent priority for ensuring power grid safety and supporting the development of smart agriculture. This system needs to integrate multi-source sensing technologies such as drone inspections, infrared sensing, AI video recognition, wireless communication, and big data analytics to achieve comprehensive, real-time, and intelligent risk monitoring of rural power grid lines. It should be able to identify agricultural drones approaching the lines in advance, and prevent dangerous approaches through audible and visual warnings and real-time operator alerts, resolving conflicts between drones and power lines at the source and providing dual protection for both the safe operation of the rural power grid and the safety of agricultural production. Summary of the Invention

[0009] The purpose of this invention is to overcome the shortcomings of the existing technology, such as the poor endurance of inspection drones and their unsuitability for detecting contact between agricultural drones and power lines, and to provide a drone early warning platform and method based on power distribution towers.

[0010] The objective of this invention can be achieved through the following technical solutions: This solution provides a drone early warning platform based on power distribution network poles, including a drone body, an adjustment mechanism, a first clamping arm, and a second clamping arm. The adjustment mechanism drives and connects the first and second clamping arms to adjust the angle between their movable ends. The first and second clamping arms are symmetrically mounted on the drone body and are both π-shaped structures. The middle section of the π-shaped structure is an arc-shaped structure that matches the outer contour of the power distribution network pole for clamping the pole. The drone body is equipped with a binocular camera, a control module, and an integrated sound and light alarm. The binocular camera acquires information about the agricultural robot and its surrounding environment and transmits it to the control module. The control module assesses the contact risk between the agricultural drone and the power line. When a contact risk exists, the integrated sound and light alarm is activated, and the relevant information is stored.

[0011] Preferably, the adjusting mechanism includes an adjusting motor, a segmented worm gear, a first turbine, and a second turbine. The adjusting motor drives and connects to the segmented worm gear, which includes two segments with opposite directions of rotation. The first turbine and the second turbine mesh with the two segments of the segmented worm gear respectively, and their directions of rotation are opposite.

[0012] Preferably, the π-shaped structure includes a first vertical rod, a second vertical rod, and a first horizontal rod with identical structures, all of which are curved rods. The curved rods have an arc-shaped protrusion in the middle. The first vertical rod and the second vertical rod are vertically parallel and are both fixed to the first horizontal rod. The distance between the first vertical rod and the second vertical rod matches the diameter of the power distribution tower. The arc-shaped protrusions on the first vertical rod or the second vertical rod of the first clamping arm and the second clamping arm are arranged opposite to each other, and the arc-shaped protrusions on the first horizontal rod of the first clamping arm and the second clamping arm are arranged opposite to each other.

[0013] Preferably, the arc-shaped protrusion of the first crossbar is provided with an elastic rubber ring on the side near the power distribution tower, and a pressure sensor is installed on the elastic rubber ring. The pressure sensor is connected to the control module. When the pressure sensor detects a value greater than a threshold, the adjustment mechanism stops operating.

[0014] This solution also provides a drone early warning method for the aforementioned drone early warning platform based on power distribution towers. The method includes the following steps: The drone itself drives the early warning platform to move and captures power lines and the surrounding environment; When an agricultural drone is detected, the drone body approaches the power distribution tower and clamps the power distribution tower with its clamping arm. The system monitors the activity route, surrounding environment, and guide wire information of agricultural drones and transmits this information to the control module. Based on a causal inference model, it assesses the risk of agricultural drones colliding with guide wires. When the risk of an agricultural drone colliding with a guide wire exceeds a threshold, an integrated sound and light alarm is activated to warn the drone operator and the acquired core data is stored on-chain.

[0015] Furthermore, the causal inference model, based on the interpretability risk assessment framework of causal graphs, determines the causal relationship between environmental factors and the contact risk of drone wiring, specifically including: Define the core causal variable set W represents wind speed, Vis represents visibility, and Vr represents radial velocity. Where is the flight angle, D is the real-time distance, and R is the contact risk; Furthermore, a causal graph is constructed based on domain knowledge to clarify the parent-child dependency relationships between variables; Calculate the value of each variable Ci in the corresponding parent node set Pa(C) i Conditional probability under the given value: , This represents the number of samples in the data where Ci and its parent node both take a specific value. The total number of samples taking a specific value for the parent node is used to obtain the complete conditional probability table; Quantify the independent causal effects of a single environmental factor on exposure risk; For current high-risk scenarios, the risk changes after a single environmental factor change are simulated, the change in risk probability before and after the change is calculated, and the corresponding explanation of the risk change by the environmental factor change is output to support emergency obstacle avoidance strategies for agricultural drones. Based on new agricultural drone flight data, the conditional probability table and causal effect calculation results are updated to enable the model to continuously adapt to new scenarios.

[0016] Furthermore, the model dynamic threshold optimization of the causal inference model for the risk assessment of agricultural drone collisions with guide wires includes the following steps: Constructing a dynamic threshold self-evolution framework based on deep reinforcement learning, including: Define the state space as S={d,v} r Let ,θ,w,vis,Vline}, where d is the real-time distance between the UAV and the guide wire, v r θ is the radial velocity of the UAV relative to the conductor, θ is the UAV's flight angle, w is the ambient wind speed, Vis is the ambient visibility, and Vline is the conductor voltage level. And define the action space as A={Δk} w ,Δk vis}, where Δk w Δk is the amount of fine-tuning for the wind speed correction coefficient. vis This is the fine-tuning amount for the visibility correction factor, and Δk w and Δk vis The values ​​of all values ​​are in the range of [−0.1, 0.1]. Construct the reward function as F = α⋅(1−P) false )+β⋅T lead −γ⋅∣d−T dynamic |, where P false For the probability of misjudgment, T lead To provide advance warning time, T dynamic For dynamic time, These are the weighting coefficients; The dynamic threshold parameter is optimized using a deep deterministic gradient strategy, specifically including: Construct an Actor-Critic architecture, where the Actor network outputs continuous actions μ(s|θ). μ ), used to generate adjustments for wind speed correction and visibility correction factors, the Critic network evaluates the action value Q(s,a|θ).Q ), used to guide strategy optimization; The target network soft update mechanism is adopted, and the update formula is: θμ′←τθμ+(1−τ)θμ′ Where τ is the soft update coefficient, obtained through simulated training iterations.

[0017] Furthermore, the monitoring of traverse information is based on the traverse identification model of the early warning platform. During the training of the traverse identification model, a traverse feature sharing mechanism based on federated learning is adopted, specifically including: Each early warning platform node acts as an independent local training unit, initializing the MobileViT sub-model based on its own collected private farmland power grid data and performing local training. Each node uploads the locally trained parameters to the federated learning server. The server uses a voltage-level weighted federated averaging algorithm to aggregate the global model parameters and distributes them to each node to replace the old local parameters. When a new early warning platform node is added to the federated learning server, a parameter distillation mechanism is used to achieve rapid adaptation. Simultaneously, a cross-scene feature adapter is used to adapt to the differences in the appearance of conductors in different farmland areas. The expression of the cross-scene feature adapter is: In the formula, Indicates the local characteristics of the conductor. Indicates the global characteristics of the conductor. λ represents the average characteristics of the conductor, and λ represents the fit coefficient.

[0018] Furthermore, core data will be stored on-chain, specifically including: Calculate hash values ​​for the core collected data, including Hash (Radar) and Hash (Image); After combining Hash(Radar) and Hash(Image) with the risk assessment structure, calculate the SHA-256 hash as the on-chain digest, H=SHA256(Hash(Radar)||Hash(Image)||Result); And construct the corresponding block. Each block includes two parts: Header and Body. The Header is used to store the hash value, timestamp, and Merkle root of the previous block. The Body is used to store the on-chain summary, risk assessment result, and the node signature that generated the data. By using an optimized practical Byzantine fault-tolerant algorithm to achieve data consensus among nodes, on-chain data consistency is ensured, and blocks from nodes that meet the conditions are written into the blockchain.

[0019] Furthermore, an optimized practical Byzantine fault-tolerant algorithm is used to achieve data consensus among nodes, ensuring on-chain data consistency. Specifically, this includes: For each drone node or ground node participating in the consensus, a reputation value (Rep) is calculated based on its historical performance. i =α⋅Acc i −β⋅Err i Acc i To correctly determine the number of times, Err i The number of misjudgments is represented by α and β, which are weighting coefficients. Set a reputation threshold θrep, select nodes with reputation values ​​Repi>θrep to participate in the consensus process, and write the corresponding node's block into the blockchain.

[0020] Compared with the prior art, the present invention has the following advantages: (1) The key to achieving this goal is to utilize the distribution network poles to hold drones. Distribution network poles are widely distributed and cover the entire rural power grid, providing natural docking and resupply points for drones. This effectively solves the problem of limited drone endurance and ensures 24-hour uninterrupted monitoring of the distribution network. Corresponding clamping structures are installed on the drones to ensure that they are reliably and temporarily fixed to the distribution network poles, saving power and improving endurance while performing target drone detection.

[0021] (2) By deploying drones equipped with high-resolution cameras and infrared sensors on power distribution towers, the surrounding environment of the power lines and the activities of agricultural drones can be captured in real time, providing comprehensive data support for subsequent intelligent analysis. This stationary monitoring mode based on power distribution towers can not only meet the needs of long-term continuous monitoring, but also create conditions for the application of AI algorithms. The images and data captured by the drones can be transmitted in real time. With the help of deep learning algorithms, safety hazards such as the approach of agricultural drones can be accurately identified, and an early warning mechanism can be quickly activated. At the same time, relevant information is uploaded to the State Grid base in real time, providing decision-making basis for law enforcement personnel, which helps to strengthen farmers' safety awareness, fundamentally alleviate the conflict between drones and power lines, and ensure the stable operation of agricultural power lines.

[0022] (3) This scheme constructs a state space containing voltage level and environmental parameters, and uses a threshold self-evolution framework based on deep reinforcement learning to achieve intelligent optimization of judgment parameters through continuous interaction with the environment; and designs a reward function that integrates the misjudgment rate and warning time to achieve the self-evolution of dynamic threshold. Compared with the traditional dynamic threshold correction that relies on a fixed formula, it can effectively reduce the misjudgment rate of dynamic threshold.

[0023] (4) This scheme adopts a voltage level weighted parameter aggregation strategy and a cross-scene feature adaptation mechanism to improve recognition accuracy without sharing the original data. In response to the limited recognition accuracy caused by insufficient data samples from a single UAV, the wire feature sharing mechanism based on federated learning can achieve global model optimization and improve recognition accuracy while protecting data privacy. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the early warning platform provided by the present invention; Figure 2 This is a schematic diagram of the structure of the early warning platform provided by the present invention located near the power distribution tower; Figure 3 This is a schematic diagram of the structure of the early warning platform for landing on the power distribution tower provided by the present invention; Figure 4 A schematic diagram of the structure of the fixed power distribution tower for the early warning platform provided by the present invention; Figure 5 A front structural diagram of the fixed power distribution tower for the early warning platform provided by the present invention; Figure 6 A side view of the fixed distribution network tower for the early warning platform provided by the present invention; Figure 7 A flowchart of the early warning method for drone contact with a power line provided by the present invention; In the diagram: 1. UAV body, 2. First clamping arm, 3. Second clamping arm, 4. Power distribution tower. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0026] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0027] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0028] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship in which the product of this invention is usually placed during use. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0029] It should be noted that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0030] Furthermore, terms such as "horizontal" and "vertical" do not imply that components must be absolutely horizontal or suspended, but rather that they can be slightly tilted. For example, "horizontal" simply means that its direction is more horizontal than "vertical," and does not mean that the structure must be completely horizontal, but can be slightly tilted.

[0031] Example 1 like Figure 1 As shown, this embodiment provides a drone early warning platform based on power distribution towers, including a drone body 1, an adjustment mechanism, a first clamping arm 2, and a second clamping arm 3; the adjustment mechanism drives and connects the first clamping arm 2 and the second clamping arm 3 to adjust the angle between the movable ends of the first clamping arm 2 and the second clamping arm 3; the first clamping arm 2 and the second clamping arm 3 are symmetrically mounted on the drone body 1, and both are π-shaped structures. The middle section of the π-shaped structure is an arc-shaped structure that matches the outer contour of the power distribution tower 4, used to clamp the power distribution tower; the drone body 1 is equipped with a binocular camera, a control module, and an integrated sound and light alarm. The binocular camera acquires information about the agricultural robot and its surrounding environment and transmits it to the control module. The control module assesses the contact risk between the agricultural drone and the power line. When a contact risk exists, the integrated sound and light alarm is activated, and the relevant information is stored.

[0032] Using power distribution network poles to dock drones is a key means to achieve this goal. Power distribution network poles are widely distributed and cover the entire rural power grid, providing drones with natural docking and resupply points. This can effectively solve the problem of limited drone endurance and ensure 24-hour uninterrupted monitoring of power distribution network lines.

[0033] In a preferred embodiment, the adjusting mechanism includes an adjusting motor, a segmented worm gear, a first turbine, and a second turbine. The adjusting motor drives and connects to the segmented worm gear, which comprises two segments with opposite directions of rotation. The first and second turbines mesh with the two segments of the segmented worm gear, respectively, and with opposite directions of rotation. Based on the meshing characteristics of the worm gear, after the worm gear drives the two clamping arms to clamp the power distribution tower, the worm gear achieves self-locking of the turbine, thus preventing loosening between the clamping arms and the power distribution tower, ensuring the stability and reliability of the fixation.

[0034] Preferred implementation methods, such as Figure 5 and Figure 6 As shown, the π-shaped structure includes a first vertical bar, a second vertical bar, and a first horizontal bar with identical structures, all of which are curved bars. The curved bars have an arc-shaped protrusion in the middle. The first vertical bar and the second vertical bar are vertically parallel and are both fixed to the first horizontal bar. The distance between the first vertical bar and the second vertical bar matches the diameter of the power distribution tower. The arc-shaped protrusions on the first vertical bar or the second vertical bar of the first clamping arm 2 and the second clamping arm 3 are arranged opposite each other, and the arc-shaped protrusions on the first horizontal bar of the first clamping arm 2 and the second clamping arm 3 are arranged opposite each other.

[0035] like Figures 2 to 3 As shown, the first vertical bar on one side of the two clamping arms forms an H-shaped abutment structure, and the second vertical bar on the other side forms an H-shaped abutment structure, which helps to clamp the two sides of the distribution network tower that are not in contact with the crossbar. Moreover, the arc-shaped structure in the middle section of the crossbar can also fit the outer contour of the tower well, so as to reliably clamp the tower.

[0036] In this embodiment, an elastic rubber ring is provided on the side of the arc-shaped protrusion of the first crossbar near the power distribution tower, and a pressure sensor is installed on the elastic rubber ring. The pressure sensor is connected to the control module. When the pressure sensor detects a value greater than a threshold, the adjustment mechanism stops operating. The elastic rubber ring inside the arc-shaped protrusion of the crossbar ensures soft contact between the crossbar and the tower, protecting the drone alarm device, and also increases the friction between the crossbar and the tower. Furthermore, in conjunction with the pressure sensor inside the elastic rubber ring, the pressure between the clamping arm and the tower is controlled to ensure the accessibility of the clamping mechanism.

[0037] Specifically, the unmanned early warning platform uses a six-rotor drone as its large payload base, equipped with a binocular camera to capture the surrounding environment of the power line. When an agricultural drone is detected, it will approach a nearby pole and land on top of it. Subsequently, the variable support below will retract inward through an adjustable motor, clamping the pole. At this point, the stopped drone will be firmly fixed to the pole, saving energy. The platform also monitors the agricultural drone's movement route in real time through AI algorithms. When a risk of collision with the power line is detected, the unmanned platform will issue a warning to the surrounding agricultural drone operators through its integrated sound and light alarm, and simultaneously contact staff for manual monitoring through the platform.

[0038] Example 2 This embodiment is basically the same as Embodiment 1, such as... Figure 7 As shown, the difference lies in providing a drone early warning method, based on a drone early warning platform based on a power distribution tower in Embodiment 1. The method includes the following steps: The drone itself drives the early warning platform to move and captures power lines and the surrounding environment; When an agricultural drone is detected, the drone body approaches the power distribution pole and clamps the power distribution pole with its clamping arm. The system monitors the activity route, surrounding environment, and guide wire information of agricultural drones and transmits this information to the control module. Based on a causal inference model, it assesses the risk of agricultural drones colliding with guide wires. When the risk of an agricultural drone colliding with a guide wire exceeds a threshold, an integrated sound and light alarm is activated to warn the drone operator and the acquired core data is stored on-chain.

[0039] By deploying drones equipped with high-resolution cameras and infrared sensors on power distribution network towers, the surrounding environment and the activities of agricultural drones can be captured in real time, providing comprehensive data support for subsequent intelligent analysis. This stationary monitoring mode based on power distribution network towers not only meets the needs of long-term continuous monitoring but also creates conditions for the application of AI algorithms. Images and data captured by drones can be transmitted in real time, and deep learning algorithms can be used to accurately identify safety hazards such as the approach of agricultural drones and quickly activate the early warning mechanism. At the same time, relevant information is uploaded to the State Grid base in real time, providing decision-making basis for law enforcement personnel, helping to strengthen farmers' safety awareness, fundamentally alleviate the conflict between drones and power lines, and ensure the stable operation of rural power grid lines.

[0040] In this embodiment, the causal inference model, based on the interpretability risk assessment framework of causal graphs, determines the causal relationship between environmental factors and the contact risk of drone wiring, specifically including: Define the core causal variable set W represents wind speed, Vis represents visibility, and Vr represents radial velocity. Where is the flight angle, D is the real-time distance, and R is the contact risk; Furthermore, a causal graph is constructed based on domain knowledge to clarify the parent-child dependencies between variables; for example, increased wind speed may increase the risk of drone deviation, thereby increasing the probability of contact, W and Vis may be the parent nodes of R, and Vr and D may directly affect R.

[0041] Based on historical flight data, including observed values ​​of the aforementioned variables, conditional probability tables for each variable in the causal graph are learned through maximum likelihood estimation to quantify the strength of dependencies between variables. Calculate the value of each variable Ci in the corresponding parent node set Pa(C) iConditional probability under the given value: , This represents the number of samples in the data where Ci and its parent node both take a specific value. The total number of samples taking specific values ​​for the parent node is used to obtain a complete conditional probability table, which provides a parameter basis for subsequent causal effect calculations. Quantify the independent causal effects of a single environmental factor on exposure risk, eliminating confounding variables; Taking the causal effect of wind speed W on risk R as an example, the probability of risk when the intervention wind speed is w is calculated using Do-Calculus: The summation term covers all possible values ​​of other variables such as visibility and radial velocity. By marginalizing, the interference of non-target variables is eliminated, yielding the net impact of wind speed variation alone on risk. The independent causal risk probabilities of other factors can be obtained similarly.

[0042] For current high-risk scenarios, the risk changes after a single environmental factor change are simulated, the change in risk probability before and after the change is calculated, and the corresponding explanation of the risk change by the environmental factor change is output to support emergency obstacle avoidance strategies for agricultural drones. Taking the wind speed change scenario as an example, the current wind speed is w. Let's assume it's adjusted to w*. If the wind speed is reduced to a safe range, calculate the risk probability after the adjustment: CF(R∣W=w ∗ ,W=w)=P(R=1|W=w ∗ ,do(Vis=vis0,V r =v r0 )) Where Vis and Vr are the current observed values ​​of other factors, by intervening to fix these variables, the impact of wind speed adjustment is isolated, and the change in risk from the original risk probability P1 to the adjusted probability P2 is calculated. The output shows that reducing wind speed can decrease risk. The explainable conclusions support emergency obstacle avoidance strategies for drones, such as adjusting flight altitude and decelerating to fly around obstacles.

[0043] Finally, based on new agricultural drone flight data, the conditional probability table and causal effect calculation results were updated to ensure the model continuously adapts to new scenarios. Ultimately, a prediction accuracy of 92.3% and an AUC of 0.91 were achieved, ensuring the accuracy and interpretability of risk prediction.

[0044] This solution breaks through the limitations of traditional models that only know the risk is high but not why it is high. It not only achieves high-precision risk prediction but also identifies which factors are the main causes of risk and how to adjust them to reduce the risk, providing a transparent and reliable basis for safe flight decisions for drones.

[0045] In this embodiment, the model dynamic threshold optimization of the causal inference model for risk assessment of agricultural drone collisions with guide wires includes the following steps: Constructing a dynamic threshold self-evolution framework based on deep reinforcement learning, including: Define the state space as S={d,v} r Let ,θ,w,vis,Vline}, where d is the real-time distance between the UAV and the guide wire, v r θ is the radial velocity of the UAV relative to the conductor, θ is the UAV's flight angle, w is the ambient wind speed, Vis is the ambient visibility, and Vline is the conductor voltage level. And define the action space as A={Δk} w ,Δk vis}, where Δk w Δk is the amount of fine-tuning for the wind speed correction coefficient. vis This is the fine-tuning amount for the visibility correction factor, and Δk w and Δk vis The values ​​of all values ​​are in the range of [−0.1, 0.1]. Construct the reward function as F = α⋅(1−P) false )+β⋅T lead −γ⋅∣d−T dynamic |, where P false For the probability of misjudgment, T lead To provide advance warning time, T dynamic For dynamic time, The weighting coefficient is used to determine the reward. The system receives a positive reward when it provides a correct warning with an appropriate lead time, and a negative reward when it provides a false warning or an excessive warning.

[0046] The dynamic threshold parameter is optimized using a deep deterministic gradient strategy, specifically including: Construct an Actor-Critic architecture, where the Actor network outputs continuous actions μ(s|θ). μ ), used to generate adjustments for wind speed correction and visibility correction factors, the Critic network evaluates the action value Q(s,a|θ). Q ), used to guide strategy optimization; The target network soft update mechanism is adopted, and the update formula is: θμ′←τθμ+(1−τ)θμ′ Where τ is the soft update coefficient, iterated through simulation training. After 5000 simulation training cycles, the false positive rate of the dynamic threshold decreased by 42%, and the timeliness of early warning improved by 28%, significantly outperforming the traditional fixed formula correction method. This directly solves the problems of high false alarm rates and delayed early warnings in traditional models, making the determination of conductor contact risk more accurate and practical.

[0047] By constructing a state space that includes voltage level and environmental parameters, a threshold self-evolution framework based on deep reinforcement learning is used to achieve intelligent optimization of judgment parameters through continuous interaction with the environment. Furthermore, a reward function that integrates the false positive rate and the warning time is designed to achieve the self-evolution of dynamic thresholds. Compared with traditional dynamic threshold correction that relies on a fixed formula, this approach can effectively reduce the false positive rate of dynamic thresholds.

[0048] In actual farmland operations, this threshold allows drones to adapt to complex environments with sudden changes in wind speed and fluctuating visibility, reducing operational interruptions caused by misjudgments and preventing power line contact accidents due to delayed warnings. In the future, if new complex scenarios are encountered, such as different crop heights or new voltage levels of power lines, this optimization framework can be used for further iterative training to generate thresholds adapted to the new scenarios, ensuring the risk assessment model has continuous adaptability.

[0049] In this embodiment, the monitoring of guide wire information is based on the guide wire identification model of the early warning platform. During the training of the guide wire identification model, a guide wire feature sharing mechanism based on federated learning is adopted, specifically including: First, each early warning platform node acts as an independent local training unit, initializing the MobileViT sub-model based on its own collected private farmland power grid data, and then performing local training: Initial parameters are The parameters are updated using gradient descent, and the update formula is: Where η is the learning rate, Li is the local loss function used to measure the recognition error of the sub-model on local data, and Di is the private dataset of that node; only the updated model parameters are retained during training. We do not upload raw data to ensure data privacy. Secondly, each node will use the parameters trained locally. The parameters are uploaded to the federated learning server, which uses a voltage-level weighted federated averaging algorithm to aggregate global model parameters and distribute them to each node to replace the old local parameters. The aggregation formula is: Where n i V represents the amount of data in the i-th node. The larger the data volume, the higher the weight. i The node primarily monitors the conductor voltage level. Higher voltage conductors pose a greater risk, leading to increased weighting of corresponding node parameters. These parameters are then aggregated to generate a global optimization model. And distribute it to each node to replace the old local parameters; When a new early warning platform node joins the federated learning server, a parameter distillation mechanism is used to achieve rapid adaptation: Calculate the global model output probability distribution P global The local model output distribution P of the new node localThe KL divergence, as the distillation loss: L distill =KL(P global ||P local The new node quickly aligns with the global model by minimizing distillation loss, accelerating convergence and improving convergence speed by 3.7 times compared to traditional methods.

[0050] Each node continues to train with new local data based on the updated global model, uploads parameters and re-aggregates them until the wire feature recognition accuracy of the global model meets the risk assessment requirements and the optional recognition accuracy is improved to the preset threshold.

[0051] This core optimization addresses the pain point of low accuracy in conductor feature recognition due to insufficient data samples from individual drones. Its core objective is to improve the feature recognition capability of the global model through multi-node collaboration, without disclosing sensitive information such as farmland power grid distribution and conductor features, thereby providing more accurate foundational data support for subsequent risk assessments such as distance calculation and contact probability evaluation.

[0052] Meanwhile, a cross-scene feature adapter is used to adapt to the differences in the appearance of conductors in different farmland areas. The expression of the cross-scene feature adapter is: In the formula, Indicates the local characteristics of the conductor. Indicates the global characteristics of the conductor. Let represent the average feature of the conductor, and λ represent the fit coefficient. Experiments show that this training mechanism improves the conductor recognition mAP@.5 to 0.94, a 19% improvement compared to single-node training.

[0053] In this embodiment, when a risk of contact between the drone and the power line is detected, the core data will be stored on-chain, specifically including: Calculate hash values ​​for the core collected data, including Hash (Radar) and Hash (Image); After combining Hash(Radar) and Hash(Image) with the risk assessment structure, a SHA-256 hash is calculated as the on-chain digest, H=SHA256(Hash(Radar)||Hash(Image)||Result). Through this strategy, the amount of data stored on-chain is reduced by 99.7% compared to the original data, balancing storage efficiency and data integrity verification requirements.

[0054] The corresponding blocks are then constructed. Each block consists of two parts: a Header and a Body. The Header stores the hash value of the previous block—ensuring chain continuity—and a timestamp—recording the time the data was uploaded to the chain—generated by the Merkle root hash of all data in the Body for quick verification of block data integrity. The Body stores the on-chain digest (SHA-256 hash), risk assessment results, and the node signature that generated the data—ensuring data traceability. By using an optimized practical Byzantine fault-tolerant algorithm to achieve data consensus among nodes, on-chain data consistency is ensured, and blocks from nodes that meet the conditions are written into the blockchain.

[0055] The constructed block containing core data is written to the blockchain while controlling the consensus latency to within 200ms, meeting the time requirements for real-time risk assessment of drones. When subsequent data integrity verification is needed, the SHA-256 hash of the original data can be recalculated and compared with the digest H stored on the chain. If the data source needs to be traced, the data generation node can be located through the node signature in the Body, ensuring end-to-end data trustworthiness.

[0056] Specifically, an optimized practical Byzantine fault-tolerant algorithm is used to achieve data consensus among nodes, ensuring on-chain data consistency. This includes: For each drone node or ground node participating in the consensus, a reputation value (Rep) is calculated based on its historical performance. i =α⋅Acc i −β⋅Err i Acc i To correctly determine the number of times, Err i The number of misjudgments is represented by α and β, which are weighting coefficients. Set a reputation threshold θrep, select nodes with reputation values ​​Repi>θrep to participate in the consensus process, and write the corresponding node's block into the blockchain.

[0057] By employing a closed loop of hash compression to reduce storage, structured blocks to ensure integrity, and reputation consensus to prevent tampering, the reliability of critical data during transmission and storage is addressed. This ensures the immutability of sensor-collected data and risk assessment results, while also meeting real-time requirements through an efficient consensus mechanism. This provides underlying data security support for the reliability of the results of the drone-cable contact risk assessment model, and facilitates subsequent auditing and problem tracing.

[0058] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A drone early warning platform based on power distribution network poles, characterized in that, The system includes a drone body (1), an adjustment mechanism, a first clamping arm (2), and a second clamping arm (3). The adjustment mechanism drives and connects the first clamping arm (2) and the second clamping arm (3) to adjust the angle between the movable ends of the first clamping arm (2) and the second clamping arm (3). The first clamping arm (2) and the second clamping arm (3) are symmetrically mounted on the drone body (1) and are both π-shaped structures. The middle section of the π-shaped structure is an arc-shaped structure that matches the outer contour of the power distribution tower (4) and is used to clamp the power distribution tower. The drone body (1) is equipped with a binocular camera, a control module, and an integrated sound and light alarm. The binocular camera acquires information about the agricultural robot and its surrounding environment and transmits it to the control module. The control module assesses the contact risk between the agricultural drone and the power line. When there is a contact risk, the integrated sound and light alarm is activated and the relevant information is stored.

2. The UAV early warning platform based on power distribution network towers according to claim 1, characterized in that, The adjustment mechanism includes an adjustment motor, a segmented worm gear, a first turbine, and a second turbine. The adjustment motor drives and connects to the segmented worm gear, which includes two segments with opposite directions of rotation. The first turbine and the second turbine mesh with the two segments of the segmented worm gear respectively, and their directions of rotation are opposite.

3. The UAV early warning platform based on power distribution network towers according to claim 1, characterized in that, The π-shaped structure includes a first vertical bar, a second vertical bar, and a first horizontal bar with identical structures, all of which are curved bars. The curved bars have an arc-shaped protrusion in the middle. The first vertical bar and the second vertical bar are vertically parallel and are both fixed on the first horizontal bar. The distance between the first vertical bar and the second vertical bar matches the diameter of the power distribution tower. The arc-shaped protrusions on the first vertical bar or the second vertical bar of the first clamping arm (2) and the second clamping arm (3) are arranged opposite to each other. The arc-shaped protrusions on the first horizontal bar of the first clamping arm (2) and the second clamping arm (3) are arranged opposite to each other.

4. The UAV early warning platform based on power distribution network towers according to claim 3, characterized in that, The first crossbar has an elastic rubber ring on the side of the arc-shaped protrusion near the power distribution tower, and a pressure sensor is installed on the elastic rubber ring. The pressure sensor is connected to the control module. When the pressure sensor detects a value greater than a threshold, the adjustment mechanism stops operating.

5. A drone early warning method, based on a drone early warning platform based on a power distribution network tower as described in any one of claims 1-4, characterized in that, The method includes the following steps: The drone itself drives the early warning platform to move and captures power lines and the surrounding environment; When an agricultural drone is detected, the drone body approaches the power distribution tower and clamps the power distribution tower with its clamping arm. The system monitors the activity route, surrounding environment, and guide wire information of agricultural drones and transmits this information to the control module. Based on a causal inference model, it assesses the risk of agricultural drones colliding with guide wires. When the risk of an agricultural drone colliding with a guide wire exceeds a threshold, an integrated sound and light alarm is activated to warn the drone operator and the acquired core data is stored on-chain.

6. A method for early warning of unmanned aerial vehicles according to claim 5, characterized in that, The causal inference model, based on the interpretability risk assessment framework of causal graphs, determines the causal relationship between environmental factors and the contact risk of drone wiring, specifically including: Define the core causal variable set W represents wind speed, Vis represents visibility, and Vr represents radial velocity. Where is the flight angle, D is the real-time distance, and R is the contact risk; Furthermore, a causal graph is constructed based on domain knowledge to clarify the parent-child dependency relationships between variables; Calculate the value of each variable Ci in the corresponding parent node set Pa(C) i Conditional probability under the given value: , This represents the number of samples in the data where Ci and its parent node both take a specific value. The total number of samples taking a specific value for the parent node is used to obtain the complete conditional probability table; Quantify the independent causal effects of a single environmental factor on exposure risk; For current high-risk scenarios, the risk changes after a single environmental factor change are simulated, the change in risk probability before and after the change is calculated, and the corresponding explanation of the risk change by the environmental factor change is output to support emergency obstacle avoidance strategies for agricultural drones. Based on new agricultural drone flight data, the conditional probability table and causal effect calculation results are updated to enable the model to continuously adapt to new scenarios.

7. The unmanned aerial vehicle (UAV) early warning method according to claim 5, characterized in that, The dynamic threshold optimization of the causal inference model for risk assessment of agricultural drone collisions with guide wires includes the following steps: Constructing a dynamic threshold self-evolution framework based on deep reinforcement learning, including: Define the state space as S={d,v} r Let ,θ,w,vis,Vline}, where d is the real-time distance between the UAV and the guide wire, v r θ is the radial velocity of the UAV relative to the conductor, θ is the UAV's flight angle, w is the ambient wind speed, Vis is the ambient visibility, and Vline is the conductor voltage level. And define the action space as A={Δk} w ,Δk vis }, where Δk w Δk is the amount of fine-tuning for the wind speed correction coefficient. vis This is the fine-tuning amount for the visibility correction factor, and Δk w and Δk vis The values ​​of all values ​​are in the range of [−0.1, 0.1]. Construct the reward function as F = α⋅(1−P) false )+β⋅T lead −γ⋅∣d−T dynamic |, where P false For the probability of misjudgment, T lead To provide advance warning time, T dynamic For dynamic time, These are the weighting coefficients; The dynamic threshold parameter is optimized using a deep deterministic gradient strategy, specifically including: Construct an Actor-Critic architecture, where the Actor network outputs continuous actions μ(s|θ). μ ), used to generate adjustments for wind speed correction and visibility correction factors, the Critic network evaluates the action value Q(s,a|θ). Q ), used to guide strategy optimization; The target network soft update mechanism is adopted, and the update formula is: θμ′←τθμ+(1−τ)θμ′ Where τ is the soft update coefficient, obtained through simulated training iterations.

8. A method for early warning of unmanned aerial vehicles (UAVs) according to claim 5, characterized in that, The monitoring of traverse information is based on the traverse identification model of the early warning platform. During the training of the traverse identification model, a traverse feature sharing mechanism based on federated learning is adopted, specifically including: Each early warning platform node acts as an independent local training unit, initializing the MobileViT sub-model based on its own collected private farmland power grid data and performing local training. Each node uploads the locally trained parameters to the federated learning server. The server uses a voltage-level weighted federated averaging algorithm to aggregate the global model parameters and distributes them to each node to replace the old local parameters. When a new early warning platform node is added to the federated learning server, a parameter distillation mechanism is used to achieve rapid adaptation. Simultaneously, a cross-scene feature adapter is used to adapt to the differences in the appearance of conductors in different farmland areas. The expression of the cross-scene feature adapter is: In the formula, Indicates the local characteristics of the conductor. Indicates the global characteristics of the conductor. λ represents the average characteristics of the conductor, and λ represents the fit coefficient.

9. A method for early warning of unmanned aerial vehicles (UAVs) according to claim 5, characterized in that, Storing core data on-chain specifically includes: Calculate hash values ​​for the core collected data, including Hash (Radar) and Hash (Image); After combining Hash(Radar) and Hash(Image) with the risk assessment structure, calculate the SHA-256 hash as the on-chain digest, H=SHA256(Hash(Radar)||Hash(Image)||Result); And construct the corresponding block. Each block includes two parts: Header and Body. The Header is used to store the hash value, timestamp, and Merkle root of the previous block. The Body is used to store the on-chain summary, risk assessment result, and the node signature that generated the data. By using an optimized practical Byzantine fault-tolerant algorithm to achieve data consensus among nodes, on-chain data consistency is ensured, and blocks from nodes that meet the conditions are written into the blockchain.

10. A method for early warning of unmanned aerial vehicles (UAVs) according to claim 9, characterized in that, Achieving data consensus among nodes through an optimized practical Byzantine fault-tolerant algorithm ensures on-chain data consistency, specifically including: For each drone node or ground node participating in the consensus, a reputation value (Rep) is calculated based on its historical performance. i =α⋅Acc i −β⋅Err i Acc i To correctly determine the number of times, Err i The number of misjudgments is represented by α and β, which are weighting coefficients. Set a reputation threshold θrep, select nodes with reputation values ​​Repi>θrep to participate in the consensus process, and write the corresponding node's block into the blockchain.

Images