Unmanned aerial vehicle dynamic real-time risk early warning method and system based on hanging and unhooking point device

Abstract

A method and system for real-time dynamic risk warning of unmanned aerial vehicles (UAVs) based on a hook-and-unhook device. This method involves constructing a virtual risk receptor array to acquire fused data, performing risk precursor pattern recognition on the fused data, obtaining the risk precursor pattern recognition results, setting a confidence threshold, comparing the confidence threshold with the risk precursor pattern recognition results to obtain a judgment result, constructing an S-shaped nonlinear amplification function and combining it with an amplification upper limit to obtain an amplified risk index, constructing a dynamic decision graph, inputting the amplified risk index and the corresponding risk precursor pattern into the dynamic decision graph, outputting a parameter adjustment strategy matching the risk precursor pattern, adjusting the UAV's operating parameters based on the parameter adjustment strategy to obtain target operating parameters, acquiring adjusted fused data based on the target operating parameters, strengthening the verification of the amplified risk index, obtaining the strengthened verification results, and executing real-time risk warning. This invention improves the accuracy of UAV risk warning.

Description

Technical Field

[0001] This invention belongs to the field of unmanned aerial vehicle (UAV) control technology, specifically relating to a method and system for dynamic real-time risk warning of UAVs based on a hook-and-unhook device. Background Technology

[0002] In the construction and operation and maintenance of power lines, the installation and removal of attachment devices is a critical link. Traditional manual handling has problems such as low efficiency and high risk of high-altitude operation. Drones are widely used in this scenario due to their flexibility and safety.

[0003] However, the real-time risk warning technology for drone-mounted attachment points still has shortcomings in current operations. In practice, when drones are attaching and dismounting attachment points, the small size of these points necessitates greater precision. However, relying on single sensor data or simple threshold judgments makes it difficult to capture early risk precursors such as tension fluctuations at the attachment point, leading to missed or incorrect risk assessments. Furthermore, risk responses often employ fixed adjustment strategies, lacking adaptability to different safety levels and operational scenarios. Sudden parameter changes after risk reduction can cause operational fluctuations, affecting the accuracy and stability of the attachment point attachment process.

[0004] Therefore, there is an urgent need for a real-time risk warning method for drones that can identify minor risks and dynamically adapt and adjust them to meet the needs of dismantling and attaching operations at different safety levels. Summary of the Invention

[0005] To address the high false alarm rate caused by using a single model to identify multiple risk types in existing technologies, this invention provides a real-time risk warning method and system for drones based on attachment / removal point devices. This invention constructs a virtual risk receptor array, combining multi-sensor data fusion and deep learning algorithm units to achieve specific identification of various risk precursor patterns such as tension fluctuations. It obtains early, weak risk signals through small risk confidence levels, avoiding the missed and false judgments of traditional single-threshold judgments, thereby improving the accuracy of risk identification. It divides the work scenario into three safety levels and sets differentiated confidence thresholds, and uses an S-shaped nonlinear amplification function to achieve gradient amplification, generating an amplified risk index to adapt to attachment / removal operations with different levels of danger. Through a dynamic decision graph, it quickly outputs adjustment strategies that match the risk patterns and indices, and designs differentiated warning signals for different risk patterns to improve operator response efficiency. Through an adjustment-secondary verification-strategy iteration method, combined with a desensitization mechanism, it allows for a smooth transition of amplified risks, thereby improving the safety and operational efficiency of drones.

[0006] Specifically, in practice, the attachment point device is generally small, so the data obtained through the attachment point device has a small deviation. It is impossible to directly judge the risk based on these small risk deviation values. Therefore, such small risk deviations are amplified by nonlinear amplification to obtain an amplified risk index. The amplified risk index is then used to process such risks, thereby obtaining a real-time risk warning and alert method for UAVs.

[0007] The first aspect of this invention discloses a dynamic real-time risk warning method for unmanned aerial vehicles (UAVs) based on a hook-and-unhook device, employing the following technical solution: Define risk precursor patterns and collect corresponding historical feature data, assign algorithm units to each risk precursor pattern, train them based on historical feature data, integrate the trained algorithm units, and obtain a virtual risk receptor array. The system uses a virtual risk receptor array to identify risk precursor patterns in real-time hanging and dismantling data; based on the risk precursor pattern identification results, it determines the target risk of the current operation according to a preset confidence threshold, and performs nonlinear amplification based on the determination results; it outputs the amplified value and combines it with the amplification upper limit to obtain the amplified risk index; A dynamic decision graph is constructed, and the amplified risk index and the corresponding risk precursor mode are input into the dynamic decision graph. The parameter adjustment strategy matching the risk precursor mode is output. The working parameters of the UAV are adjusted based on the parameter adjustment strategy to obtain the target working parameters. Adjusted fusion data is obtained based on the target working parameters. Secondary micro-risk confidence is obtained through the adjusted fusion data. The secondary micro-risk confidence is compared with the micro-risk confidence to obtain the reinforcement verification results.

[0008] Furthermore, the risk precursor pattern identification of real-time hanging and unhanging data includes: Collect tension data at the ground wire attachment / removal point, drone fuselage tilt angle data, and data on obstacles around the ground wire to obtain real-time attachment / removal data; The real-time hanging and dismantling data is filtered and normalized, and the processed hanging and dismantling data is obtained by aligning with the timestamp. Feature extraction is performed on the processed hanging and dismantling data, and the extracted features are compared with the feature library in each algorithm unit to calculate the similarity and obtain the small risk confidence level corresponding to each risk precursor mode. The risk precursor pattern identification result is based on the risk precursor pattern and its corresponding small risk confidence level.

[0009] Furthermore, the determination result includes: Set confidence thresholds according to the safety level of the hanging and dismantling of the hanging point device operation; The confidence level of the tiny risk With confidence threshold If a comparison is made, If so, the result is that there is a target risk; if If the result is negative, the target risk is determined to be nonexistent, and data collection is performed again.

[0010] Furthermore, the acquisition of the amplified risk index includes: If the determination result indicates the existence of target risk, then an S-shaped nonlinear amplification function is constructed using the small risk confidence level, threshold offset, and amplification factor. The amplified value is obtained by using an S-shaped nonlinear amplification function, and then multiplied by the amplification upper limit to obtain the amplification risk index. The amplification factor can be dynamically adjusted according to the degree of danger of the risk precursor model.

[0011] Furthermore, the output parameter adjustment strategy includes: The historical risk case data and simulation data of the acquired wire hanging and dismantling operations are jointly trained to obtain a dynamic decision map; Input the risk precursor model and the amplified risk index into the dynamic decision graph, and output the parameter adjustment strategy that matches the risk precursor model. The decision graph includes risk level classification nodes and pattern matching nodes. The risk level classification node is used to classify risk levels based on risk precursor patterns and corresponding amplified risk indices. The pattern matching node is used to output the corresponding matching strategy based on the risk precursor pattern.

[0012] Furthermore, obtaining the target operating parameters includes: When the dynamic decision graph matches the corresponding risk precursor pattern, the corresponding parameter adjustment strategy is triggered to adjust the UAV's operating parameters in real time and obtain the target operating parameters. The operating parameters of the UAV include flight control parameters, warning parameters, and mechanism parameters.

[0013] Furthermore, the step of obtaining the enhanced verification result and performing corresponding measures includes: Based on the target working parameters, data on the tension at the attachment / removal point, the fuselage tilt angle, and the obstacles around the conductor and ground wire are collected, and adjustment and fusion data are obtained. The adjusted fusion data is input into a virtual risk receptor array for secondary identification to obtain secondary micro-risk confidence levels. The secondary minor risk confidence level is compared with the minor risk confidence level, and the reinforcement verification result is obtained according to the set first reinforcement threshold and second reinforcement threshold, and corresponding measures are implemented. If the enhanced verification result confirms the risk, then maintain the target operating parameters and continue inspection and monitoring. If the enhanced verification result indicates a reduced risk, then the desensitization mechanism will be activated; If the enhanced verification result shows that the risk is stable, then maintain the target operating parameters and shorten the data collection interval.

[0014] The second aspect of this invention discloses a dynamic real-time risk warning system for unmanned aerial vehicles (UAVs) based on a hook-and-unhook device, which implements the dynamic real-time risk warning method for UAVs as described in the first aspect of this invention. The system includes: Identification module: Define risk precursor patterns and collect corresponding historical feature data, assign algorithm units to each risk precursor pattern, train based on historical feature data, integrate the trained algorithm units, and obtain a virtual risk receptor array; Amplification Module: It performs risk precursor pattern recognition on real-time hanging and dismantling data through a virtual risk receptor array; it determines the target risk of the current operation based on the risk precursor pattern recognition results according to a preset confidence threshold, and performs nonlinear amplification based on the determination results; it outputs the amplified value and obtains the amplified risk index by combining the amplification upper limit. Adjustment module: Constructs a dynamic decision graph, inputs the amplified risk index and the corresponding risk precursor mode into the dynamic decision graph, outputs a parameter adjustment strategy that matches the risk precursor mode, and adjusts the working parameters of the UAV based on the parameter adjustment strategy to obtain the target working parameters; Early warning module: Based on the target operating parameters, it obtains the adjusted fusion data, obtains the secondary micro-risk confidence level through the adjusted fusion data, and compares the secondary micro-risk confidence level with the micro-risk confidence level to obtain the enhanced verification result.

[0015] The beneficial effects of this invention are that, compared with the prior art, 1. By constructing a virtual risk receptor array, risk precursor pattern recognition is performed on the drone attachment and dismantling device to obtain accurate risk precursor pattern recognition results.

[0016] 2. By using a nonlinear amplification method to amplify the risk precursor pattern recognition results, the minute deviations in the hanging and dismantling of the hanging point device are amplified to obtain an amplified risk index. The risk is then analyzed by amplifying the risk index, thereby improving the accuracy of data processing.

[0017] 3. Introduce a dynamic decision graph to adjust the amplified risk index, thereby improving the adaptability of the parameter adjustment of the corresponding attachment point device during the operation of the UAV. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the execution flow of the UAV dynamic real-time risk warning method in this embodiment.

[0019] Figure 2This is a structural block diagram of the UAV dynamic real-time risk warning system in this embodiment. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this invention. The embodiments described in this invention are merely some embodiments of this invention, and not all embodiments. Based on the spirit of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of this invention.

[0021] As an embodiment of the present invention, a specific implementation method for a dynamic real-time risk warning method for unmanned aerial vehicles based on a hook-and-unhook point device is disclosed. The execution flow of the method embodiment is as follows: Figure 1 .

[0022] S1: As one implementation method, when the drone attaches to or detaches from the attachment point device, a virtual risk receptor array is first constructed to acquire fused data of the drone attaching to or detaching from the attachment point device. Specifically, this includes: S1.1: Obtain feature data of a single risk precursor pattern in the historical scenario of attaching and removing ground wires of UAVs, assign each feature data to an algorithm unit, perform deep learning training, and construct a virtual risk receptor array from all algorithm units trained by deep learning.

[0023] Specifically, risk precursor modes are defined as risk types in conductor and ground wire operations, such as tension fluctuations at the attachment / removal point and fuselage tilt angle deviations. Feature data of various risk precursor modes from historical UAV conductor and ground wire attachment / removal operations are collected, such as spectral feature data of tension fluctuations at the attachment / removal point exceeding the rated value by 5%. Simultaneously, an independent algorithm unit is assigned to each risk precursor mode, and deep learning training is performed on the algorithm units using convolutional neural networks or long short-term memory networks. This ensures that each algorithm unit possesses the specific recognition capability for a single risk precursor mode. All trained algorithm units are then integrated to obtain a virtual risk receptor array.

[0024] It should be noted that by mapping risk precursor patterns one-to-one with algorithm units, cross-interference is avoided, preventing features such as aircraft tilting from being misjudged as obstacles being too close. At the same time, this distributed integration design allows all trained algorithm units to work in parallel, significantly shortening the recognition time. If new risk types emerge later, they can be directly added to the virtual risk receptor array through an algorithm unit without reconstructing the entire system.

[0025] In some possible embodiments, taking a power company as an example, the operation data of the power company in hanging and removing conductors and ground wires across highways, rural roads and open areas within 3 years are obtained. 100 sets of feature data where the distance between the conductor / ground wire and the tower is less than 1.2 times the safety threshold are extracted, and algorithm unit A is assigned to this risk precursor mode. In this embodiment, a convolutional neural network is used to train algorithm unit A so that it can identify the distance features between the conductor / ground wire and the tower. The same method is used to train algorithm unit B, which has a tension fluctuation of more than 5%, and algorithm unit C, which has a fuselage tilt angle deviation of more than 10°, etc. After training, all algorithm units A, B, C, etc. are integrated to form a virtual risk receptor array.

[0026] S1.2: Acquire data on the tension at the ground wire attachment / removal point, the fuselage tilt angle, and the obstacles around the ground wire from the UAV, and perform filtering and multi-data fusion operations to obtain fused data.

[0027] Specifically, the data collected by the sensor modules on the drone include the tension data at the ground wire attachment point, the fuselage tilt angle data, and the data of obstacles around the ground wire. Kalman filtering is used to remove noise interference from all the above data, and then weights are assigned to each data point to perform weighted fusion of multiple data to obtain fused data.

[0028] Understandably, acquiring the tension data at the attachment / removal point is used to directly reflect the connection strength and force balance. Excessive tension may damage the grounding wire or the drone. Acquiring the fuselage tilt angle data helps to control hovering accuracy and operational response. An excessive tilt angle may cause the drone to lose attitude control and even flip over, or it may cause the force direction at the attachment / removal point to shift, thus affecting the alignment accuracy of the grounding wire attachment / removal. Acquiring data on obstacles around the grounding wire is to identify the main risk sources that could cause collisions, obtaining their distance and orientation data to determine the safe distance between the drone and the grounding wire and surrounding objects. In actual use, users can make adaptive adjustments to the acquired data based on the actual situation.

[0029] It should be noted that before fusing the data on the tension at the ground wire attachment point, the fuselage tilt angle, and the data on obstacles around the ground wire, these data need to be normalized, such as by using min-max normalization to unify the dimensions of the data and ensure data synchronization.

[0030] In some possible embodiments, assuming that when the UAV is attaching and detaching ground wires across rural roads, the sensor module on the UAV collects multiple sets of attachment / detachment point tension data, multiple sets of fuselage tilt angle data, and multiple sets of data on obstacles around the ground wires. After Kalman filtering, abnormal data points are removed, and valid data is retained. Then, these data are aligned by timestamp, and the weights of the attachment / detachment point tension data (0.4), fuselage tilt angle data (0.3), and obstacles around the ground wires (0.3) are set to generate fused data D.

[0031] S1.3: Based on the feature extraction and cosine similarity calculation of each algorithm unit in the virtual risk receptor array, the matching quantification value of each risk precursor pattern is obtained as the small risk confidence. The risk precursor pattern recognition result is constructed based on the risk precursor pattern and the corresponding small risk confidence.

[0032] Specifically, the fused data is input into a virtual risk receptor array. Each algorithm unit extracts features from the risk precursor patterns corresponding to itself in the fused data. For example, algorithm unit B extracts frequency domain features such as fluctuation frequency and peak value of the tension data at the attachment / dismounting point, and algorithm unit C extracts time domain features of the rate of change of the fuselage tilt angle data. The extracted features are then compared with the feature library within each algorithm unit to calculate cosine similarity. The quantized value of the cosine similarity is the small risk confidence. Quantization maps the value range of the cosine similarity to the range of 0-1 and compresses it to 0-1%. That is, the value range of the small risk confidence is 0%-1%. The risk precursor pattern and the corresponding small risk confidence result are then output.

[0033] In some possible embodiments, continuing the example of step S1.2, the fused data D is input into the virtual risk receptor array, the fluctuation features of the tension data are extracted from algorithm unit B, and the cosine similarity is calculated with the features of tension fluctuation exceeding 5% in the feature library of algorithm unit B to obtain a small risk confidence of 0.35%; similarly, the tilt angle features are extracted from algorithm unit C, and the cosine similarity is calculated with the feature library of algorithm unit C to obtain a small risk confidence of 0.15%, etc.; all risk precursor patterns and corresponding small risk confidences are integrated to obtain the risk precursor pattern recognition result {tension fluctuation at the attachment / disassembly point 0.35%, fuselage tilt angle deviation 0.15%, ...}.

[0034] Risk precursor pattern recognition is performed on the fused data based on the virtual risk receptor array to obtain the risk precursor pattern recognition results. Through this step, the small deviations in the data when the UAV attaches and detaches the attachment point device are obtained as the basis for subsequent steps. S2: As one implementation of the embodiment, a confidence threshold is set, the risk precursor pattern recognition result is judged based on the confidence threshold, the judgment result is obtained, and the judgment result is nonlinearly amplified to obtain the amplified risk index. Through this step, the small deviation is nonlinearly amplified, so as to more clearly distinguish the deviation pattern of the UAV when attaching and detaching the attachment point device. When a drone attaches or detaches a mounting point on a ground wire, the mounting point is usually small in reality, resulting in a small deviation in the data obtained from it. It is not easy to directly judge the risk based on these small risk deviation values. Therefore, this method obtains an amplified risk index by nonlinear amplification of these small risk deviations, and then processes these risks through the amplified risk index to obtain a real-time risk warning and alert method for drones.

[0035] Preferably, step S2 includes: S2.1: Set a confidence threshold based on the safety level of the hanging and dismantling device operation.

[0036] The safety levels include Level 1, Level 2, and Level 3. Level 1 corresponds to high-risk operations across highways and railways, Level 2 corresponds to medium-risk operations across ordinary roads, and Level 3 corresponds to low-risk operations without any obstacles in the cross-regional area.

[0037] It should be noted that the safety levels are classified according to the environmental hazards of the operation of attaching and dismantling attachment points. For example, high-risk operation areas with dense personnel and vehicles and serious risks and consequences, such as crossing highways and railways, are classified as Level 1 safety; medium-risk operation areas with fewer personnel and vehicles, such as crossing ordinary roads, are classified as Level 2 safety; and low-risk operation areas with open areas and minor risks and consequences are classified as Level 3 safety.

[0038] Specifically, a first confidence threshold, a second confidence threshold, and a third confidence threshold are set based on the first-level security level, the second-level security level, and the third-level security level, respectively.

[0039] Furthermore, confidence thresholds are set based on historical accident statistics and safety regulations, and in combination with different safety levels. The confidence thresholds include a first confidence threshold of 0.2%, a second confidence threshold of 0.3%, and a third confidence threshold of 0.5%.

[0040] For example, if the work scenario is across a rural road, it corresponds to a level 2 safety level and a second confidence threshold of 0.3%; if the work scenario is across a highway, it corresponds to a level 1 safety level and a first confidence threshold of 0.2%; if the work scenario is an open area, it corresponds to a level 3 safety level and a third confidence threshold of 0.5%.

[0041] S2.2: Judge the risk precursor pattern recognition results based on the confidence threshold and obtain the judgment result.

[0042] Specifically, the confidence level of minor risks in the risk precursor pattern recognition results is compared with the confidence threshold corresponding to the current operational safety level. If the confidence level of a minor risk in the risk precursor pattern recognition result is greater than or equal to the corresponding confidence threshold, it is determined that a target risk exists. The confidence level of the minor risk is then nonlinearly amplified to obtain an amplified risk index. If the confidence levels of all minor risks in the risk precursor pattern recognition result are less than the corresponding confidence threshold, it is determined that no target risk exists, and the process returns to step S1 for data collection.

[0043] In some possible embodiments, assuming the operation is carried out on a rural road, if the confidence level of the tension fluctuation at the hook-and-unhook point in the identification result is 0.35%, which is greater than or equal to the set confidence level threshold of 0.3%, then it is determined that there is a target risk, and the confidence level of the small risk is nonlinearly amplified to obtain the amplified risk index; if the deviation of the fuselage tilt angle in the identification result is 0.15%, which is less than the set confidence level threshold of 0.3%, then it is determined that there is no target risk, and the process returns to step S1 to collect data.

[0044] S2.3: If the determination result is that there is a target risk, then construct an S-shaped nonlinear amplification function and obtain the amplification risk index.

[0045] Traditional exponential gain functions typically use a fixed gain factor, such as... Alternatively, using piecewise linear nonlinear functions may fail to match the risk levels in complex scenarios.

[0046] This invention employs a dynamically adjustable amplification factor to construct an S-shaped nonlinear amplification function, specifically expressed as follows: ,in, Represented as the magnified value, This is expressed as the magnification factor. This is expressed as a confidence level for minimal risk. Represented as threshold offset ( In this embodiment, 0.2% is used to compensate for recognition errors.

[0047] Furthermore, the confidence level of the small risk is nonlinearly amplified to obtain the amplified risk index. Specifically, the confidence level of the small risk is mapped to the range of 0-1, and then the standardized confidence level of the small risk is substituted into the S-shaped nonlinear amplification function to output the amplified value y. The amplified risk index is then obtained by combining the amplification upper limit. The formula for the amplified risk index is: ;in To increase the upper limit.

[0048] Understandably, the value of the amplification factor is dynamically adjusted according to the degree of danger of the risk precursor mode. Assuming the value of the amplification factor is [5,8], when the risk is a risk that directly threatens the survival of the UAV, such as the tension fluctuation at the attachment point exceeding the rated value by 6%, this risk may cause the clamping mechanism to fail, and the collision risk of this risk is extremely high; this type of risk develops quickly, has serious consequences, and is extremely dangerous, then the amplification factor will be dynamically adjusted to [7,8], thereby quickly amplifying the small confidence level and ensuring the triggering of the emergency adjustment strategy.

[0049] Furthermore, when the risk affects the accuracy of the action but does not involve safety in a short period of time, such as a deviation of 10°-15° in the fuselage tilt angle, this risk may cause misalignment of the mounting and dismounting, but it is not out of control; such risks need to be appropriately amplified, so the amplification factor is dynamically adjusted to [5,6] to balance the response speed and the stability of the adjustment.

[0050] It should be noted that setting an amplification limit in this step is to avoid excessive decision-making caused by unlimited amplification of the risk index. Without an upper limit, small changes in confidence levels may be amplified to an extreme value, causing the drone to execute adjustment strategies beyond actual needs, such as excessively tightening the clamping mechanism or suddenly reducing flight speed, which would damage operational stability and thus introduce new operational risks. Therefore, setting a reasonable amplification limit can control the risk index within a reasonable range and ensure that the adjustment strategy matches the actual level of risk.

[0051] Furthermore, in this embodiment, the amplification limit is set to 25%, which corresponds to the risk level set in step S3.2. The upper limit of the high risk level is 25%. Therefore, setting the amplification limit to 25% can highlight the high risk signal and prevent it from entering the invalid range beyond the high risk limit. Setting the amplification limit to 25% also takes into account historical operation data. The most serious controllable risks in the operation of attaching and dismantling attachment point devices, such as tension fluctuation exceeding the rated value by 8% or obstacle distance approaching the safety threshold by 1.1 times, after being amplified by the S-shaped nonlinear amplification function, have a maximum value of about 24%-25%. Therefore, setting the amplification limit to 25% can fully cover such high-risk scenarios.

[0052] It should be noted that during drone inspections, the amplification speed is slow when the confidence level of minor risks is low; when the confidence level of minor risks is close to the threshold offset, the amplification speed is extremely fast to ensure that no risks are missed; when the confidence level of minor risks is high, the speed is stabilized to prevent over-amplification.

[0053] In some possible embodiments, the small risk confidence level of the tension fluctuation at the hanging point is 0.35%, which is standardized to 0.0035. This is then substituted into the S-shaped nonlinear amplification function to obtain the value of y as 0.99005. Multiplying this by the amplification upper limit of 25%, we obtain the amplification risk index of 24.75%.

[0054] S2.4: If the determination result shows no target risk, return to step S1 to continue collecting fused data and performing risk precursor pattern recognition, without triggering any amplification or adjustment process, and the drone maintains its current working state.

[0055] In some possible embodiments, if the working scenario is across rural roads, and the small confidence level of all acquired risk precursor patterns is less than 0.3%, such as the small confidence level of the tension fluctuation at the attachment / removal point being 0.28% and the small confidence level of the obstacle distance being 0.12%, then the result is determined that there is no target risk, and the drone continues the attachment / removal operation, while the sensor module continues to collect data at the original frequency.

[0056] S3: As one implementation method of the embodiment, a dynamic decision graph is constructed, the amplified risk index and the corresponding risk precursor mode are input into the dynamic decision graph, and a parameter adjustment strategy matching the risk precursor mode is output. Based on the parameter adjustment strategy, the working parameters of the UAV are adjusted to obtain the target working parameters.

[0057] This step obtains the adjustment parameters corresponding to the amplified risk index through a dynamic decision graph. The adjustment parameters are displayed on the display screen operated by the drone pilot, who then adjusts the position of the attachment point device by adjusting the parameters.

[0058] S3.1: Obtain historical risk case data and simulation data for the installation and removal of conductors and ground wires. The historical risk case data includes risk precursor patterns, amplified risk indices, and optimal adjustment strategy data under different scenarios. The simulation data includes multiple sets of test data for wind speed, terrain, and conductor and ground wire type variables.

[0059] Specifically, historical risk case data for conductor and ground wire installation and removal operations are collected, including multiple sets of samples (approximately 500 sets). Each set of data includes the risk precursor pattern, the amplified risk index, and the corresponding optimal adjustment strategy. For example, in some cases, the tension fluctuation pattern has an amplified risk index of 20%, and the corresponding optimal adjustment strategy is to increase the clamping force redundancy by 10%. At the same time, simulation models are established based on the principle of multibody dynamics, such as MATLAB or Simulink simulation software, to simulate test data under different wind speeds (e.g., 3-10 m / s), terrain (e.g., plains, mountains), and conductor and ground wire models (e.g., JL / G1A-400 / 35). There are no fewer than 500 sets of test data in these cases.

[0060] In some possible embodiments, the historical case data includes 50 cases of tension fluctuations at the attachment and disassembly points, of which 20 cases correspond to an amplification risk index of 15%-20%. The optimal adjustment strategy is to increase the clamping force redundancy by 8% and adjust the hovering accuracy to ±0.08m. The simulation data includes 100 sets of test data simulating tension fluctuations at a wind speed of 5m / s, and the adjustment effects corresponding to different amplification risk indices are obtained simultaneously.

[0061] S3.2: Jointly train historical risk case data and simulation data to obtain a dynamic decision graph, which includes risk level classification nodes and pattern matching nodes.

[0062] Specifically, a decision tree algorithm is used to jointly train historical risk case data and simulation data to obtain a dynamic decision graph. The input to the dynamic decision graph is the risk precursor pattern (e.g., "tension fluctuation at the attachment / removal point") + amplified risk index (e.g., 24.75%), and the output is a parameter adjustment strategy matched to the risk (e.g., "increase dynamic clamping force redundancy by 8% and adjust hovering accuracy to ±0.08m"). The decision graph includes risk level classification nodes and pattern matching nodes. The risk level classification nodes are used to classify risk levels based on risk precursor patterns and corresponding amplified risk indices. The pattern matching nodes are used to output corresponding matching strategies based on risk precursor patterns.

[0063] In this embodiment, the risk level R is divided according to the following rules, with low risk level: Medium risk level: High-risk level: .

[0064] Furthermore, during training, a cost complexity pruning algorithm is used to remove redundant branches. This cost complexity pruning algorithm is a post-pruning algorithm for decision trees. It obtains a smaller tree by deleting sub-branches that do not contribute to the generalization accuracy, thereby matching the real-time computational load of the UAV.

[0065] In some possible embodiments, a dynamic decision map is obtained through historical risk case data and simulation data in step S3.1. According to the dynamic decision map, the risk level division node in the dynamic decision map classifies 24.75% in step S2.3 as a high-risk level. The pattern matching node in the dynamic decision map associates the risk precursor pattern of the tension fluctuation at the attachment point in step S3.2 with the adjustment strategies of increasing the clamping force redundancy by 12%, adjusting the hovering accuracy to ±0.05m, and the orange warning.

[0066] S3.3: Input the amplified risk index and the corresponding risk precursor mode into the dynamic decision graph, and output the parameter adjustment strategy that matches the risk precursor mode. The UAV adjusts the flight control parameters, warning parameters and mechanism parameters according to the parameter adjustment strategy to obtain the target working parameters. The parameter adjustment strategy includes the adjustment of flight control parameters, the adjustment of warning parameters and the adjustment of actuator parameters.

[0067] The adjustment of flight control parameters includes adjusting hovering accuracy and yaw angle response speed; the adjustment of warning parameters includes preloading risk type corresponding statements and light warning signals; and the adjustment of mechanism parameters includes adjusting the clamping force of the conductor ground wire clamping mechanism.

[0068] Furthermore, when the dynamic decision map matches the corresponding risk precursor mode, it triggers the adjustment of the corresponding warning parameters. For example, the drone's central control screen displays a text warning slogan corresponding to the risk precursor mode, and at the same time, it controls the warning light on the drone body corresponding to the risk precursor mode to flash, reminding the operator to take the corresponding adjustment strategy in time to avoid damage to the ground wire or the drone.

[0069] In a further implementation, after receiving the corresponding adjustment strategy, the UAV adjusts the flight control parameters, warning parameters, and mechanism parameters in real time. The parameters obtained after the adjustment are the target operating parameters.

[0070] For example, taking the adjustment of warning parameters as an example, when abnormal fluctuations in the tension at the attachment / removal point are detected, the drone's central control screen displays a warning message stating that the tension exceeds the safe range and urging immediate adjustment of the position, accompanied by a flashing orange light. Simultaneously, the dynamic decision map outputs a warning strategy for abnormal tension fluctuations at the attachment / removal point, and the central control screen displays corresponding text: "If the tension exceeds the safe range, please adjust the position immediately." The corresponding indicator light also begins to flash. If the orange light begins to flash, after seeing the warning, the operator will slightly adjust the drone to the left by 0.4m, reducing the tension fluctuation amplitude at the attachment / removal point from 6% to 3%. When an abnormally close obstacle is detected, the drone's central control screen displays a warning message stating that the distance to the pole is too close and urging immediate adjustment of the heading, accompanied by a flashing red light. Simultaneously, the dynamic decision map outputs a warning strategy for abnormally close obstacle distance, and the central control screen displays corresponding text: "If the distance to the pole is too close, please adjust the heading immediately." The corresponding indicator light also begins to flash. If the red light begins to flash, after seeing the warning, the operator will automatically shift the drone to the right by 1.2m, restoring the distance to 1.3 times the safe threshold.

[0071] For example, after receiving the above adjustment strategy, the UAV central control system will adjust the flight control parameters, warning parameters, and actuator parameters in real time. The system will automatically lock the current parameters, which are the target working parameters, such as hovering accuracy ±0.05m, yaw angle response speed 8° / s, orange light flashing at 1.5Hz, and ground wire clamping mechanism clamping force redundancy of 17%.

[0072] It should be noted that the above examples are just illustrations of how to adjust warning parameters and do not represent all situations. Adjustments may be made in other ways.

[0073] S4: As one implementation method of the embodiment, the adjustment and fusion data is obtained based on the target working parameters, the amplification risk index is strengthened and verified based on the adjustment and fusion data, the strengthened verification result is obtained, and real-time risk warning is executed based on the strengthened verification result.

[0074] S4.1: The UAV operates according to the target working parameters, collects and merges data on the sag of the ground wire, the tension at the attachment point, the tilt angle of the fuselage, and the obstacles around the ground wire, and obtains adjustment and fusion data.

[0075] Specifically, the UAV operates according to the target operating parameters. The sensor module collects data on the tension at the attachment point, the tilt angle of the fuselage, and the obstacles around the ground wire after operating based on the target operating parameters. These data are then used to obtain adjusted and fused data in the same way as in step S1.2.

[0076] In some possible embodiments, it is assumed that after the UAV operates according to the target operating parameters for a period of time, the sensor module collects new data on the tension of the ground wire attachment point, the fuselage tilt angle, and the obstacle data around the ground wire. After filtering and fusion, the data is adjusted and fused.

[0077] S4.2: Input the fused data into the virtual risk receptor array for secondary identification to obtain the secondary micro-risk confidence level.

[0078] Specifically, the fused data is input into the virtual risk receptor array constructed in step S1, and feature extraction and cosine similarity calculation are performed in the same way as in step S13 to perform secondary identification of the target risk precursor pattern and obtain secondary micro-risk confidence.

[0079] In some possible embodiments, assuming that the fused adjustment data is input into the virtual risk receptor array, the algorithm unit B performs secondary identification on the tension fluctuation at the hanging and dismantling point, and obtains a secondary small risk confidence level of 0.28%, while the small risk confidence level obtained in the first identification is 0.35%.

[0080] S4.3: Set a first reinforcement threshold and a second reinforcement threshold, and compare the secondary small risk confidence level with the small risk confidence level.

[0081] Specifically, the confidence level of the secondary small risk is compared with the confidence level of the small risk to obtain the comparison result. A first reinforcement threshold and a second reinforcement threshold are set. The risk is judged by comparing the comparison result with the first reinforcement threshold, and the risk is judged by comparing the comparison result with the second reinforcement threshold.

[0082] In some possible embodiments, the first reinforcement threshold is set to 1.3 times the confidence level of the minor risk itself; if the secondary minor risk confidence level exceeds the first reinforcement threshold, that is, the increase is more than 30% compared to the minor risk confidence level, then the risk is judged to have increased; the second reinforcement threshold is set to 0.8 times the confidence level of the minor risk itself; if the secondary minor risk confidence level is lower than the second reinforcement threshold, that is, the decrease is more than 20% compared to the minor risk confidence level, then the risk is judged to have decreased; if neither of these two situations applies, then the risk is judged to be stable.

[0083] S4.3.1: If the judgment result is "risk aggravated", it means that the risk is still aggravated after parameter adjustment, and the verification result is risk confirmation; at this time, the target working parameters are kept unchanged, and the continuous inspection and monitoring mode is started.

[0084] Furthermore, the inspection and monitoring mode specifically increases the sensor acquisition frequency to twice the original frequency, while the UAV sends a risk status report to the ground control center every 5 seconds.

[0085] In some possible embodiments, assuming that in the tension fluctuation risk, the confidence level of the secondary minor risk is 0.46% and the confidence level of the minor risk is 0.35%, and the calculated increase is greater than 30%, the risk is determined to be confirmed, the sampling frequency of the sensor module is increased from 100Hz to 200Hz, and the UAV sends a report on the increased tension fluctuation risk to the ground control center every 5 seconds.

[0086] S4.3.2: If the judgment result is "risk reduced", it means that the parameter adjustment is effective and the risk is reduced. The enhanced verification result is that the risk is reduced. At this time, the desensitization mechanism is activated.

[0087] Furthermore, this step uses a desensitization mechanism to prevent the UAV target operating parameters from suddenly reverting to the baseline operating parameters after the risk has been reduced. Directly switching parameters could lead to problems such as a rapid decrease in conductor clamping force and sudden changes in hovering accuracy. The desensitization mechanism gradually reduces the impact of risk by slowly reducing the weight, which can ensure that the UAV smoothly transitions from the risk adjustment state to the normal operation state, thereby ensuring the safety and continuity of the operation.

[0088] In some possible implementations, assuming that in the risk of tension fluctuation, the confidence level of the secondary minor risk is 0.26% and the confidence level of the minor risk is 0.35%, and the calculated decrease is greater than 20%, it is determined that the risk has been reduced, and at this time the desensitization mechanism is activated.

[0089] In a further implementation, the desensitization mechanism in this method achieves a gradual attenuation of risk impact by linearly reducing the weighting coefficient of the amplified risk index. Specifically, this includes: (1) Set an initial value for the weight coefficient of the amplified risk index, and dynamically adjust the rate of decrease of the weight coefficient based on the risk precursor mode. Reduce the weight coefficient based on the rate of decrease until the amplified risk index is less than or equal to the preset safety threshold.

[0090] Specifically, the initial value of the weighting coefficient of the amplified risk index is set to 1, meaning the index operates with 100% effectiveness. The reduction rate is dynamically adjusted according to the type of risk precursor mode. When the risk precursor mode is a high-risk mode such as tension fluctuation, the rate is reduced by 10%-12% every 0.5 seconds. When the risk precursor mode is a medium- or low-risk mode such as fuselage tilt angle deviation or obstacle distance approach, the rate is reduced by 8%-10% every 0.5 seconds. The weighting coefficient is continuously reduced, and the weighted amplified risk index is calculated until the index drops below the preset safety threshold of 3%.

[0091] It should be noted that the continuously decreasing weighting coefficient is calculated using the original amplified risk index multiplied by the weighting coefficient.

[0092] Understandably, the lower limit for low risk is 5%, so the safety threshold must be set to less than 5%. Setting the safety threshold to 3% retains a small risk buffer without causing excessively long desensitization time due to an excessively low threshold, thus balancing the safety and continuity of drone operations.

[0093] In some possible embodiments, taking the high-risk mode of tension fluctuation at the attachment / removal point as an example, the initial value of its weight coefficient is 1, and the reduction rate is set to decrease by 11% every 0.5 seconds. The weighted index at 0 seconds is 24.75%×1=24.75%. Through the set reduction rate, the weight at 0.5 seconds is calculated to be 0.89, and the weighted index is 24.75%×0.89=22.03%. The weight at 1 second is 0.78, and the weighted index is 17.18%, ..., and the weight at 4 seconds is 0.12, and the weighted index is 2.72%. At this time, the weighted index is less than the preset safety threshold.

[0094] (2) The working parameters of the UAV are restored from the target working parameters to the reference working parameters based on the low-pass filtering algorithm.

[0095] Specifically, when the weighted amplified risk index drops below the preset safety threshold, the drone's operating parameters are adjusted from the target operating parameters to the baseline parameters before adjustment, and the adjustment time is controlled within 1-2 seconds to avoid sudden parameter changes that could lead to operational instability.

[0096] In some possible embodiments, continuing the example of step S4.5.1, it is assumed that after the weighted index is reduced to 2.72%, the hovering accuracy is restored from ±0.05m to the reference ±0.1m within 1.5s by a low-pass filtering algorithm, and the clamping force redundancy is restored from 12% to the reference 5%.

[0097] S4.3.3: If the confidence level of the secondary minor risk does not increase by more than the first reinforcement threshold or decrease by more than the second reinforcement threshold compared to the confidence level of the minor risk, and the reinforcement verification result is that the risk is stable, then the target working parameters are maintained and the collection interval is shortened.

[0098] Specifically, if the confidence level of the secondary minor risk does not increase by more than the first reinforcement threshold or decrease by more than the second reinforcement threshold, i.e., within the range of -20% to 30%, it indicates that the risk is still stable after parameter adjustment, and the reinforcement verification result is that the risk is stable; at this time, the target working parameters are maintained and the collection interval is shortened.

[0099] In some possible implementations, assuming that in the risk of tensile fluctuations, the confidence level of the secondary minor risk is 0.30% and the confidence level of the minor risk is 0.35%, and the calculated change is within the range of -20% to 30%, it is determined that the risk is stable. In this case, the target operating parameters are maintained and the data collection interval is shortened by half.

[0100] refer to Figure 2 , Figure 2 This is a structural block diagram of a dynamic real-time risk warning system for unmanned aerial vehicles (UAVs) based on a hook-and-unhook device.

[0101] As an embodiment of the present invention, a dynamic real-time risk warning system for unmanned aerial vehicles (UAVs) based on a hook-and-unhook point device is disclosed. Employing the specific implementation method described above for dynamic real-time risk warning of UAVs, the system includes: Identification module: Define risk precursor patterns and collect corresponding historical feature data, assign algorithm units to each risk precursor pattern, train based on historical feature data, integrate the trained algorithm units, and obtain a virtual risk receptor array; Amplification Module: It performs risk precursor pattern recognition on real-time hanging and dismantling data through a virtual risk receptor array; it determines the target risk of the current operation based on the risk precursor pattern recognition results according to a preset confidence threshold, and performs nonlinear amplification based on the determination results; it outputs the amplified value and obtains the amplified risk index by combining the amplification upper limit. Adjustment module: Constructs a dynamic decision graph, inputs the amplified risk index and the corresponding risk precursor mode into the dynamic decision graph, outputs a parameter adjustment strategy that matches the risk precursor mode, and adjusts the working parameters of the UAV based on the parameter adjustment strategy to obtain the target working parameters; Early warning module: Based on the target operating parameters, it obtains the adjusted fusion data, obtains the secondary micro-risk confidence level through the adjusted fusion data, and compares the secondary micro-risk confidence level with the micro-risk confidence level to obtain the enhanced verification result.

[0102] As an embodiment of the present invention, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is loaded onto the processor, it employs the specific implementation method described above for the dynamic real-time risk warning method for unmanned aerial vehicles.

[0103] As an embodiment of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing a computer program, which, when executed by a processor, adopts the specific implementation method described above for the dynamic real-time risk warning method for unmanned aerial vehicles.

[0104] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A method for dynamic real-time risk early warning of unmanned aerial vehicles (UAVs) based on a hook-and-unhook device, characterized in that, include: Define risk precursor patterns and collect corresponding historical feature data, assign algorithm units to each risk precursor pattern, train them based on historical feature data, integrate the trained algorithm units, and obtain a virtual risk receptor array. Risk precursor pattern identification is performed on real-time attachment / removal data using a virtual risk receptor array. Based on the risk precursor pattern recognition results, the target risk of the current operation is determined according to the preset confidence threshold, and nonlinear amplification is performed based on the determination results; Output the amplified value and combine it with the amplification limit to obtain the amplification risk index; A dynamic decision graph is constructed, and the amplified risk index and the corresponding risk precursor mode are input into the dynamic decision graph. The parameter adjustment strategy matching the risk precursor mode is output. The working parameters of the UAV are adjusted based on the parameter adjustment strategy to obtain the target working parameters. Adjusted fusion data is obtained based on the target working parameters. Secondary micro-risk confidence is obtained through the adjusted fusion data. The secondary micro-risk confidence is compared with the micro-risk confidence to obtain the reinforcement verification results.

2. The method for dynamic real-time risk early warning of unmanned aerial vehicles based on a hook-and-unhook point device according to claim 1, characterized in that, The process of identifying risk precursor patterns in real-time attachment / removal data includes: Collect tension data at the ground wire attachment / removal point, drone fuselage tilt angle data, and data on obstacles around the ground wire to obtain real-time attachment / removal data; The real-time hanging and dismantling data is filtered and normalized, and the processed hanging and dismantling data is obtained by aligning with the timestamp. Feature extraction is performed on the processed hanging and dismantling data, and the extracted features are compared with the feature library in each algorithm unit to calculate the similarity and obtain the small risk confidence level corresponding to each risk precursor mode. The risk precursor pattern identification result is based on the risk precursor pattern and its corresponding small risk confidence level.

3. The method for dynamic real-time risk early warning of unmanned aerial vehicles based on a hook-and-unhook point device according to claim 1, characterized in that, The determination result includes: Set confidence thresholds according to the safety level of the hanging and dismantling of the hanging point device operation; The confidence level of the tiny risk With confidence threshold If a comparison is made, If so, the result is that there is a target risk; if If the result is negative, the target risk is determined to be nonexistent, and data collection is performed again.

4. The method for dynamic real-time risk early warning of unmanned aerial vehicles based on a hook-and-unhook point device according to claim 1, characterized in that, The acquisition of the amplified risk index includes: If the determination result indicates the existence of target risk, then an S-shaped nonlinear amplification function is constructed using the small risk confidence level, threshold offset, and amplification factor. The amplified value is obtained by using an S-shaped nonlinear amplification function, and then multiplied by the amplification upper limit to obtain the amplification risk index. The amplification factor can be dynamically adjusted according to the degree of danger of the risk precursor model.

5. The method for dynamic real-time risk early warning of unmanned aerial vehicles based on a hook-and-unhook point device according to claim 1, characterized in that, The output parameter adjustment strategy includes: The historical risk case data and simulation data of the acquired wire hanging and dismantling operations are jointly trained to obtain a dynamic decision map; Input the risk precursor model and the amplified risk index into the dynamic decision graph, and output the parameter adjustment strategy that matches the risk precursor model. The decision graph includes risk level classification nodes and pattern matching nodes. The risk level classification node is used to classify risk levels based on risk precursor patterns and corresponding amplified risk indices. The pattern matching node is used to output the corresponding matching strategy based on the risk precursor pattern.

6. The method for dynamic real-time risk early warning of unmanned aerial vehicles based on a hook-and-unhook point device according to claim 1, characterized in that, The acquisition of the target operating parameters includes: When the dynamic decision graph matches the corresponding risk precursor pattern, the corresponding parameter adjustment strategy is triggered to adjust the UAV's operating parameters in real time and obtain the target operating parameters. The operating parameters of the UAV include flight control parameters, warning parameters, and mechanism parameters.

7. The method for dynamic real-time risk early warning of unmanned aerial vehicles based on a hook-and-unhook point device according to claim 1, characterized in that, The step of obtaining the enhanced verification result and executing corresponding measures includes: Based on the target working parameters, data on the tension at the attachment / removal point, the fuselage tilt angle, and the obstacles around the conductor and ground wire are collected, and adjustment and fusion data are obtained. The adjusted fusion data is input into a virtual risk receptor array for secondary identification to obtain secondary micro-risk confidence levels. The secondary minor risk confidence level is compared with the minor risk confidence level, and the reinforcement verification result is obtained according to the set first reinforcement threshold and second reinforcement threshold, and corresponding measures are implemented. If the enhanced verification result confirms the risk, then maintain the target operating parameters and continue inspection and monitoring. If the enhanced verification result indicates a reduced risk, then the desensitization mechanism will be activated; If the enhanced verification result shows that the risk is stable, then maintain the target operating parameters and shorten the data collection interval.

8. A real-time dynamic risk warning system for unmanned aerial vehicles (UAVs) based on a hook-and-unhook device, comprising executing the real-time dynamic risk warning method for UAVs based on a hook-and-unhook device as described in any one of claims 1-7, characterized in that, The system includes: Identification Module: Defines risk precursor patterns and collects corresponding historical feature data, assigns algorithm units to each risk precursor pattern, trains the algorithm units based on historical feature data, integrates the trained algorithm units to obtain a virtual risk receptor array, and identifies risk precursor patterns in real-time attachment / removal data through the virtual risk receptor array. Amplification Module: It performs risk precursor pattern recognition on real-time hanging and dismantling data through a virtual risk receptor array; it determines the target risk of the current operation based on the risk precursor pattern recognition results according to a preset confidence threshold, and performs nonlinear amplification based on the determination results; it outputs the amplified value and obtains the amplified risk index by combining the amplification upper limit. Adjustment module: Constructs a dynamic decision graph, inputs the amplified risk index and the corresponding risk precursor mode into the dynamic decision graph, outputs a parameter adjustment strategy that matches the risk precursor mode, and adjusts the working parameters of the UAV based on the parameter adjustment strategy to obtain the target working parameters; Early warning module: Based on the target operating parameters, it obtains the adjusted fusion data, obtains the secondary micro-risk confidence level through the adjusted fusion data, and compares the secondary micro-risk confidence level with the micro-risk confidence level to obtain the enhanced verification result.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the computer program is loaded into the processor, it implements the dynamic real-time risk warning method for unmanned aerial vehicles according to any one of claims 1-7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the dynamic real-time risk warning method for unmanned aerial vehicles according to any one of claims 1-7.

Images