A high-voltage line in-situ multifunctional monitoring device and method

Abstract

The application discloses a kind of high-voltage line in-situ multifunctional monitoring device and method, belong to electric power system monitoring field, the device includes: mutually hinged upper shell and lower shell, setting on the guide rod of machine body component, guide rod is configured to connect flying vehicle to hang machine body component;Trigger driving mechanism is configured to, in response to high-voltage cable enters cable channel and touches the action of preset position;Further include: the automatic locking mechanism that is configured to automatically lock machine body component when upper shell is closed to position;And linkage disengaging mechanism, between automatic locking mechanism and installation guide mechanism, configured to convert the locking action of automatic locking mechanism into mechanical driving force, machine body component is completed locking while releasing the connection of guide rod and machine body component, so that guide rod is separated from machine body component.The device of the application can be installed on high-voltage cable by unmanned aerial vehicle, and has the characteristics of convenient recycling.

Description

Technical Field

[0001] This invention relates to the field of power system monitoring, and in particular to a multi-functional in-situ monitoring device and method for high-voltage lines. Background Technology

[0002] The operational status of high-voltage transmission lines directly affects the safety and stability of the power system. Because high-voltage cables often traverse complex terrains such as mountains and uninhabited areas, they are highly susceptible to environmental factors such as icing, wind deflection, high temperatures, and heavy loads. Currently, monitoring of high-voltage cables mainly relies on manual inspections, drone inspections, or single online monitoring devices. However, due to the high altitude of the high-voltage cables and the high voltage, in-situ installation is extremely difficult, and existing monitoring devices face installation challenges, particularly when installed under energized conditions. Because of the difficulty in in-situ installation, some technical solutions employ setting up a simulated cable section and installing monitoring devices on it to measure icing thickness. However, icing thickness is highly dependent on factors such as temperature, humidity, wind speed, cable sway, solar radiation, and cable temperature rise. The icing conditions on the simulated cable differ significantly from those on the actual cable, resulting in a large discrepancy between the measured values ​​and the actual values. The temperature, current, and sway acceleration of the actual cable cannot be measured on the simulated cable, and the existing monitoring methods have the following significant drawbacks: 1. Insufficient all-weather sensing capability: Traditional video surveillance relies on visible light and is basically ineffective at night or in dense fog, resulting in monitoring blind spots; 2. Ambiguous disaster type identification: Images alone cannot distinguish the physical properties of icing (such as dense rime versus loose frost), leading to frequent false alarms. Rime has a high density and is extremely harmful, while frost often does not require emergency treatment. The inability to distinguish between the two can lead to operational and maintenance decision errors; 3. Confusion about fault causes: Increased cable sag is a common safety hazard, but the cause of increased sag may be thermal expansion of materials due to high temperatures (thermal fault) or mechanical stretching due to icing (mechanical fault). Existing technology cannot decouple these two coupled factors, leading to incorrect decisions such as de-icing when load reduction is needed; 4. Predictive models lack physical constraints: Existing intelligent predictive models are mostly purely data-driven. In extreme weather conditions where historical fault samples are lacking, they are prone to outputting predictions that violate physical common sense (such as predicting increased ice weight but no change in sag).

[0003] Therefore, there is an urgent need for a remote monitoring technology for high-voltage cables that can integrate multi-dimensional sensing data, accurately distinguish disaster types, decouple fault causes, and have the ability to predict physical consistency. Summary of the Invention

[0004] One of the objectives of this invention is to provide a high-voltage line in-situ multifunctional monitoring device to solve the problems of poor all-weather sensing capability, inability to distinguish between rime and fog, difficulty in decoupling thermal / mechanical faults, and unreliable prediction results in the prior art.

[0005] This invention is achieved through the following technical solution: a high-voltage line in-situ multifunctional monitoring device, comprising a body assembly, an installation guide mechanism, and a trigger drive mechanism disposed within the body assembly. The body assembly includes an upper shell and a lower shell hinged to each other, the upper shell and the lower shell cooperating to form a cable channel for high-voltage cables to pass through. The installation guide mechanism includes a guide rod disposed on the body assembly, the guide rod being configured to connect to an aircraft to suspend the body assembly. The trigger drive mechanism is configured to, in response to the high-voltage cable entering the cable channel and touching a preset position, release a preset elastic potential energy to drive the upper shell to close relative to the lower shell. The high-voltage line in-situ multifunctional monitoring device further includes: an automatic locking mechanism configured to automatically lock the body assembly when the upper shell is closed; and a linkage release mechanism connected between the automatic locking mechanism and the installation guide mechanism, configured to convert the locking action of the automatic locking mechanism into a mechanical driving force, releasing the connection between the guide rod and the body assembly simultaneously with the body assembly completing locking, causing the guide rod to detach from the body assembly.

[0006] Furthermore, the triggering drive mechanism includes: a pressure triggering component disposed at the bottom of the cable channel and configured to be displaced by the pressure of the high-voltage cable; an energy storage component having a fixed end for storing the elastic potential energy and an actuating end for outputting power; and a transmission release component connected between the pressure triggering component and the energy storage component; wherein, the displacement of the pressure triggering component releases the limiting effect on the energy storage component through the transmission release component, and the actuating end of the energy storage component pushes the upper housing to close.

[0007] Furthermore, the energy storage component is a torsion spring, and a stop post is provided inside the body assembly; in the installation state, the fixed end of the torsion spring abuts against the stop post to maintain the energy storage state; the device also includes a remote recovery mechanism, which is configured to release the locking state of the automatic locking mechanism in response to a remote command, and simultaneously move the stop post or release the stop post from obstructing the fixed end of the torsion spring, so that the torsion spring unloads the elastic potential energy.

[0008] Furthermore, the remote recovery mechanism includes a winch and a pull rope; the winch is disposed within the body assembly, one end of the pull rope is wound around the winch, and the other end is connected to the hook rod of the automatic locking mechanism; the winch is configured to wind up the pull rope to pull the hook rod to rotate in the opposite direction to unlock, and the unlocking action of the hook rod triggers the displacement of the stop post.

[0009] Furthermore, the automatic locking mechanism includes: a latch, a top hook slider, and a hook connecting rod; the latch is disposed on the upper housing, and the top hook slider and the hook connecting rod are disposed on the lower housing; the hook connecting rod is configured to rotate under the action of a push rod spring to hook the latch after it is closed; the linkage release mechanism includes a push rod, a rotating component, and a locking rod tongue; one end of the push rod is connected to the hook connecting rod, and the other end is connected to the rotating component; the rotating component is kinetically connected to the locking rod tongue; the rotation of the hook connecting rod drives the rotating component to rotate through the push rod, thereby causing the locking rod tongue to retract to release the guide rod.

[0010] Furthermore, the device also includes: a multi-source sensing module and a main control unit integrated on the body assembly; a flexible inner wall liner is provided on the inner wall surface of the cable channel; the multi-source sensing module includes an ultrasonic detection unit, a cable temperature measurement unit, and a current sensing unit embedded in the inner wall liner; the probe surface of the ultrasonic detection unit is configured to be in close contact with the surface of the high-voltage cable when the device is closed, for emitting ultrasonic waves and receiving acoustic echoes characterizing the density of the icing medium; the cable temperature measurement unit uses a contact thermistor embedded in the inner wall liner to measure the temperature of the cable body; the current sensing unit includes: segments disposed on the upper and lower housings. The Rogowski coil or current transformer core at the joint, the current sensing unit is configured to dock and form a closed magnetic circuit in response to the closing and locking action of the upper housing, so as to measure the current of the high-voltage cable; the main control unit, electrically connected to the multi-source sensing module, is used to process the collected data and transmit it to the outside through the wireless communication module. The main control unit has a built-in attitude monitoring module and an edge computing module: the attitude monitoring module includes a six-axis inertial sensor, which is used to collect the tilt angle data and acceleration data of the device moving with the cable in real time; the edge computing module is used to receive the acoustic echo data of the ultrasonic detection unit and the tilt angle data of the attitude monitoring module, and run the freezing rain disaster monitoring logic.

[0011] Furthermore, the main control unit is also configured to perform a timestamp alignment operation, synchronously freeze the readings of all sensors in the multi-source sensing module at the same trigger time, and send the multi-dimensional sensing data packets containing timestamps to the cloud platform through the wireless communication module. The cloud platform stores a computer program, and when the computer program is executed by the processor, it implements the high-voltage cable remote multi-functional method as described in any one of claims 8 to 10.

[0012] Another aspect of the present invention provides a multi-functional in-situ monitoring method for high-voltage lines, comprising the following steps: S100, synchronously collecting multi-dimensional sensing data of high-voltage cables through a multi-source sensing module, and performing timestamp alignment on the multi-dimensional sensing data; S200, uploading the multi-dimensional sensing data to a back-end cloud platform via a wireless communication network, and using an intelligent visual recognition model to identify the visual recognition boundary of the cable; S300, executing freezing rain disaster monitoring logic based on the acoustic and mechanical characteristics in the multi-dimensional sensing data to distinguish the type of attachments on the cable surface; S400, constructing a load decoupling model containing the mechanical state equation of the cable, using the load decoupling model to decouple the icing state and mechanical state of the cable, calculating a fault discrimination operator to distinguish between high-temperature sag and icing sag of the cable; S500, constructing a cable state trend prediction model using a long short-term memory network of physical information fusion, and generating cable state trend prediction results based on the cable state trend prediction model.

[0013] Furthermore, the multidimensional sensing data includes at least: visible light images, temperature field distribution data, ground clearance data, medium echo data, and suspension point tangential tilt angle data; the timestamp alignment operation refers to simultaneously freezing the readings of multiple heterogeneous sensors at the same trigger moment to ensure that the parameters describing the cable status are consistent in physical time and space.

[0014] Furthermore, in S200, the process of uploading the multi-dimensional sensing data to the backend cloud platform is an adaptive edge-level transmission strategy: under normal operating conditions, only key feature vectors representing cable health are extracted and uploaded; when suspected fault features are identified, an abnormal interruption mode is triggered and the transmission channel is preempted, prioritizing the uploading of original data fragments containing evidence of the fault scene.

[0015] Furthermore, the upload process may also include: performing differentiated quality of service mapping based on data type, using a best-effort delivery protocol for regular heartbeat data, and a forced handshake confirmation protocol for alarm data; and performing semantic segmentation of the image using region of interest encoding logic, performing low compression rate encoding on the foreground region where cables and towers are located, and performing high compression rate encoding on the background region.

[0016] Furthermore, in S200, edge-sensitive geometric constraint logic is introduced into the training process of the intelligent visual recognition model: the geometric similarity index is used to measure the maximum mismatch between the contour point set predicted by the model and the real label contour point set; the maximum mismatch is introduced into the loss function as a geometric penalty term to constrain the cable edge curve predicted by the model to infinitely approach the real boundary in geometric space.

[0017] Furthermore, the intelligent visual recognition model adopts a dynamic gating fusion mechanism based on channel attention: it automatically adjusts the weight coefficients of the visible light branch and the infrared branch features according to the ambient light intensity; in low light environment, it automatically increases the feature response weight of the infrared branch and uses temperature difference features to identify the cable boundary.

[0018] Furthermore, the intelligent visual recognition model can be constructed through the following steps: 1) Construct a dual-stream feature fusion network based on a global gating mechanism. First, use two independent convolutional neural network branches to extract deep feature maps of visible light and infrared light respectively. Use global average pooling to compress spatial information and extract a global context vector representing the current environmental state. Then, use a learnable global gating weight matrix to calculate the response weight of each feature channel and automatically adjust the dependence on infrared or visible light features: automatically amplify the weight of infrared features at night and balance the two during the day.

[0019] 2) Introduce an edge-sensitive Hausdorff distance loss function. By incorporating Hausdorff distance as part of the loss function, even if the vast majority of pixels are predicted correctly, if the predicted edge deviates significantly from the real edge at some point, the Hausdorff distance will generate a huge penalty value. This forces the network to pay close attention to the continuity of the cable and the geometric accuracy of the edge during training, thereby achieving a keen perception of micron-level icing thickening.

[0020] Furthermore, the attention weighting mechanism for two-stream feature fusion can be calculated using the following formula:

[0021] in, This is the final fused multimodal feature map; The feature map is obtained by the visible light branch through a CNN feature extractor. Feature maps extracted from infrared thermal imaging branches; For channel-level splicing operations, heterogeneous features are merged into... tensor; This is a global average pooling operation, its function is to reduce the spatial dimension. Compress to Output a global context descriptor vector; This is the global gating weight matrix. The Sigmoid activation function is used to normalize the weight values ​​to... interval; The Hadamard product (element-wise multiplication) is used to weight and activate each channel of the original concatenated features using the calculated attention weight vector, thereby achieving feature recalibration.

[0022] Furthermore, the total loss function is a weighted average of the classification loss and the geometric loss, as shown in the following equation:

[0023] in, For conventional cross-entropy loss, This is the balance coefficient for the conventional cross-entropy loss; These are the balance coefficients for geometric loss. These two balance coefficients are used to adjust the proportion of geometric constraints in the overall optimization objective.

[0024] Furthermore, to address the issue of the standard Hausdorff distance calculation being non-differentiable, a weighted average Hausdorff distance is employed, with the specific calculation formula as follows:

[0025] in, This is the set of coordinates of the pixels at the edge of the cable icing predicted by the model; This is the set of coordinates of the cable edge pixels marked in the actual label (GroundTruth); and These represent the number of pixels in the prediction set and the true set, respectively; This represents the square of the Euclidean distance.

[0026] Furthermore, in S300, the freezing rain disaster monitoring logic includes: using the energy response of the sound wave pulse when reflected at the interface of a heterogeneous medium, combined with the distance parameters provided by spatial geometric detection methods, to compensate for the path loss of the sound wave; extracting acoustic impedance characteristics that characterize the physical density of the attachments on the cable surface based on the compensated energy response; and mapping the acoustic impedance characteristics with the macroscopic mechanical feedback of the cable under gravity load to determine the disaster type of the attachments.

[0027] Furthermore, the disaster type of the attached material is determined as follows: if the acoustic impedance characteristics indicate that the attached material is dense and the displacement deviation of the cable suspension point indicates that the mass growth rate exceeds the threshold, it is determined to be a frost disaster mode; if the acoustic impedance characteristics indicate that the attached material contains air gaps and the displacement deviation of the cable suspension point indicates that the mass growth is slow, it is determined to be a hoarfrost feature.

[0028] Furthermore, constructing an acoustic impedance-gravity coupling discrimination model based on acoustic and mechanical characteristics can include the following steps: 1) Constructing a medium density estimation algorithm based on ultrasonic reflection echo characteristics to transform the invisible internal physical structure into a quantifiable acoustic impedance index; first, using the precise distance obtained by lidar to correct the energy attenuation of ultrasonic waves during air transmission, thereby calculating the accurate interface reflection coefficient; then, based on the physical relationship between the reflection coefficient and acoustic impedance, the acoustic impedance value of the icing medium is inferred; finally, combined with empirical sound velocity, the volume density of the icing is calculated; 2) Constructing a gravity growth verification model based on tilt angle-tension transformation, and combining acoustic characteristics to achieve multi-dimensional coupling disaster type determination; by monitoring the tilt angle change rate and tension change rate of the cable, the real-time mass growth rate of the cable can be estimated; using trigonometric function relationships, the axial tension change of the cable is decoupled into the vertical gravity component change; only when the microscopic medium density and macroscopic gravity growth simultaneously exceed the high-risk threshold is the system determined to be a dense frost disaster.

[0029] Furthermore, the icing density can be calculated using the following formula: To calculate the reflection coefficient in the above formula This requires considering the ratio of transmitted to received signal energy and compensating for path loss. ,in, The desired bulk density of the icing is (unit: ); The known acoustic impedance of the ultrasonic probe surface; The speed of sound in ice; The reflection coefficient of sound waves at the icy interface; The amplitude of the ultrasonic echo received by the sensor; This is the initial amplitude of the ultrasonic wave during emission; The attenuation coefficient of sound waves in air; This is the precise distance from the sensor to the cable surface, measured by a coaxial lidar.

[0030] Furthermore, the gravity growth verification and final judgment logic can be expressed as follows: First, estimate the mass growth of the cable per unit time (gravity growth verification): Subsequently, a multidimensional coupling criterion is constructed: ,in, This represents the estimated cable quality growth rate. The rate of change of the axial tension of the cable over time; For tilt sensor readings; The geometric projection factor; For structural correction factors; This is a Boolean value. The system outputs a frost warning if and only if this value is true. This represents the critical threshold for density. The threshold for the rate of increase in gravity; For logical AND operation.

[0031] Furthermore, in S400, the process of calculating the fault discrimination operator includes: based on the cable's preset reference state parameters and combined with the real-time sensed temperature characteristics, constructing a thermodynamic simulation path under the assumption of no additional load constraints to obtain a theoretical reference value characterizing the cable's pure thermodynamic deformation; using spatial three-dimensional point cloud detection data to obtain the actual geometric shape of the cable to obtain a measured physical quantity characterizing the current comprehensive deformation; and extracting the offset of the measured physical quantity relative to the theoretical reference value as the mechanical load deformation residual, which serves as the fault discrimination operator.

[0032] Furthermore, the discrimination logic for distinguishing between high-temperature sag and icing sag of cables includes: if the fault discrimination operator approaches zero within a preset error range, the abnormal deviation of the cable is attributed to the thermal expansion of the material caused by high temperature; if the fault discrimination operator is a significantly positive value, the abnormal deviation of the cable is attributed to overload sag caused by external mechanical load.

[0033] Furthermore, the load decoupling model can be constructed through the following steps: 1) Using the improved cable state equation, a thermodynamic benchmark under theoretical ice-free conditions is constructed. The initial calibration state of the cable is used as the first state, and the current monitoring time is used as the second state. A nonlinear state equation is introduced to describe the cable under temperature changes. 1) Based on the stress evolution law under varying load, calculate the theoretical thermal stress at the current temperature; 2) Use lidar measurement data to invert the actual stress, calculate the difference between theoretical thermal sag and actual sag, construct a fault discrimination operator, and use the actual sag measured by lidar based on the parabolic approximation theory. To infer the current actual horizontal stress 3) Finally, define the fault detection operator. By calculating the deviation between the actual geometric shape and the theoretical thermodynamic shape, the load decoupling and attribution are achieved.

[0034] Furthermore, the state equation for the high-voltage cable changing from the initial state to the current monitoring state can be expressed as follows: ,in, The core unknown to be solved is the unknown quantity at the current temperature. Under the assumption of no icing, the theoretical thermal stress is calculated. The cable surface temperature is measured in real time by an infrared thermal imager or temperature sensor. The calibration temperature during system installation and initialization; To be at the calibration temperature The initial stress measured below; The specific load of the cable itself; The load ratio is the load under the initial conditions. The span of the cable (; The elastic modulus of a cable characterizes the material's ability to resist elastic deformation. The linear expansion coefficient of the cable.

[0035] Furthermore, the actual horizontal stress can be obtained by inversion using the following formula: Meanwhile, the theoretical thermal stress obtained in the aforementioned steps is used. Calculate the theoretical thermal arc sag under the ice-free assumption. : .

[0036] The further fault detection operator can be calculated using the following formula: ,in, This is the current actual stress inferred from measured data; The actual sag of the cable as measured in real time by the lidar; This is the theoretical thermal sag calculated considering only the effect of temperature; The mass per unit length of the cable; The mass of the icing layer is unknown. It is the acceleration due to gravity; For fault detection operators.

[0037] Furthermore, based on The judgment logic can be as follows: Case 1, thermal failure (high temperature sagging); if ,and A large value means that the actual observed sag is completely consistent with the theoretically calculated pure thermal sag; therefore, the conclusion is: the cable is not icy, and the increased sag is entirely caused by the thermal expansion of the material due to temperature rise; scenario two, mechanical failure; if This means the cable appears to sag lower than theoretically possible, because theoretical calculations have already eliminated the effects of temperature; this additional sag is... This can only be caused by the stretching effect of additional weight; therefore, the conclusion is that the cable is icing over. The size is proportional to the severity of the icing.

[0038] Furthermore, in S500, the construction process of the cable state trend prediction model includes: embedding a consistency check operator based on prior physical knowledge into the neural network architecture; using the consistency check operator to calculate the mechanical logic residuals between multiple prediction variables directly output by the model; and by minimizing the mechanical logic residuals, constraining the neural network to perform state evolution deduction within the physically feasible domain that conforms to the cable mechanical constitutive equation.

[0039] Furthermore, the neural network architecture performs multi-task coupled prediction: synchronously outputting the predicted icing amount reflecting the nature of the load and the predicted sag amount reflecting the geometric response; if the theoretical geometric value of the predicted icing amount derived from the physical equation does not match the predicted sag amount, the weight parameters of the neural network are forcibly corrected using the generated penalty signal.

[0040] Furthermore, after generating the cable status trend prediction results, a hierarchical early warning logic is executed: if the predicted trajectory shows that the mechanical tension or electrical clearance distance will exceed the physical safety limit at a future time, an early warning signal associated with the specific physical violation type is generated, and targeted de-icing or voltage regulation suggestions are given.

[0041] Furthermore, the cable condition trend prediction model can be constructed through the following steps: 1) Construct a time series input vector containing multidimensional meteorological and condition characteristics, and define the prediction target of coupled physical quantities. The input layer receives past data. The time window sequence includes surface temperature, ambient humidity, wind speed, tilt angle, and lidar ranging value. The output layer predicts the sag size and ice weight at the next moment. 2) Construct a physical constraint loss function and embed the cable state equation into the optimization objective of the neural network to force the model to learn physical laws. By treating the cable state equation as an inviolable physical operator, if the theoretical sag generated by the predicted ice weight is inconsistent with the predicted sag value under the predicted temperature conditions, it means that the network's prediction violates physical laws. This degree of violation is quantified as part of the loss function, so that when updating weights through backpropagation, the network will not only try to approximate historical data, but also try to avoid violating physical equations. Thus, even in extreme working conditions where data is scarce, the correct result can be deduced from physical logic.

[0042] Furthermore, the input-output mapping relationship of the cable condition trend prediction model can be defined as: The network's input vector and predicted output vector can be represented by the following equation: , ,in, for The input feature vector at time step; The surface temperature of the cable measured by infrared radiation; The relative humidity measured by the micro-weather station; Wind speed; It is the angle of inclination; Visibility distance (which reflects fog concentration); For model predictions The state vector at any given time; The predicted cable sag value; This is the predicted mass of icing per unit length; Represents the mapping function of the LSTM network; These are network parameters.

[0043] Furthermore, the loss function incorporating physical constraints can be represented by the following equation: Physical residuals It can be expressed by the following formula: ,in, Let this be the total loss function for network training; This is based on authentic historical observation data; This is the predicted output of the LSTM network; For traditional data-driven losses; These are the physical constraint weighting coefficients; The predicted sag is directly output by the network; The predicted temperature value for the next time step is input into the network. The predicted ice weight is directly output by the network; This is a mapping function for nonlinear state equations.

[0044] Compared with the prior art, the present invention has the following advantages and beneficial effects: 1. This invention, through precise mechanical design and multi-sensor fusion, ensures that the device's closing action relies entirely on the pre-stored elastic potential energy of the torsion spring. The closing is triggered by the gravity of the cable pressing a physical button, eliminating the need for a motor-driven mechanism. This significantly reduces power consumption during standby and installation, avoiding the risk of installation failure due to battery depletion or motor malfunction, and significantly improving operational reliability in harsh outdoor environments. Furthermore, the cleverly designed linkage transmission mechanism (first-stage push rod - second-stage turntable - third-stage locking tongue) automatically disconnects the guide rod from the main body upon locking. The guide rod is only used as an installation aid and is carried back by the drone, leaving no trace on the cable. This not only reduces the long-term load on the cable but also decreases the device's windward area, reducing the risk of wind deflection and swinging. Simultaneously, it features a unique torsion spring unloading and booster-opening linkage mechanism. During retrieval, the motor not only unlocks the device but also disables the powerful torsion spring (entering a free state), preventing the upper shell from being difficult to open or damaging the mechanism under elastic force. This achieves truly fully automatic detachment, eliminating the need for personnel to climb the tower for retrieval. Combined with parachute recovery, this protects the expensive internal sensors, allowing the device to be reused multiple times and significantly reducing operating costs.

[0045] 2. A dual-stream feature fusion network is adopted, which utilizes visible light texture features when there is sufficient light, and automatically switches to focus on infrared temperature difference features at night or in heavy fog, thus solving the failure problem of a single visual sensor in harsh environments. In addition, by introducing Hausdorff distance as a geometric constraint, the model is no longer limited to pixel classification, but focuses on the approximation of the overall geometric contour, thereby achieving accurate identification of micron-level early icing and blurred boundaries, realizing all-weather, high-precision visual perception capabilities.

[0046] 3. This invention constructs an acoustic impedance-gravity coupling discrimination logic, which uses ultrasonic waves to detect the internal density of the medium (microscopic physical property) and combines it with an tilt sensor to monitor changes in gravitational torque (macroscopic mechanical property), forming a rigorous AND gate criterion. This enables the system to accurately distinguish between dense frost (high density, high gravity) and loose hoarfrost (low density, low gravity) that have similar appearances but completely different hazards, effectively avoiding invalid de-icing operations caused by false alarms and greatly improving operation and maintenance efficiency.

[0047] 4. This invention proposes a fault discrimination operator based on state equations. By calculating the difference between the actual total sag and the theoretical thermal sag, it cleverly separates the mixed thermodynamic deformation (high-temperature thermal expansion) and mechanical deformation (ice-laden pressure), thus providing a decisive basis for power grid dispatch: if the operator is zero, it is a thermal fault and the current load should be reduced; if the operator is positive, it is a mechanical fault and de-icing measures should be initiated. This decoupling technology fills a gap in the industry.

[0048] 5. This invention addresses the problem of traditional intelligent models making haphazard predictions under extreme disasters by employing a Physical Information Fusion Long Short-Term Memory (PI-LSTM) network. By embedding the mechanical state equation of the cable into the loss function, the learning process of the neural network is forced to comply with physical laws (such as Hooke's Law and the principle of thermal expansion). This ensures that even under extreme conditions where historical data is lacking, the model's output prediction results (ice weight and sag) remain physically consistent, greatly improving the reliability of disaster early warning. Attached Figure Description

[0049] The accompanying drawings, which are included to provide a further understanding of embodiments of the invention and form part of this application, do not constitute a limitation thereof. In the drawings: Figure 1 This is an overall device diagram provided in Embodiment 1 of the present invention.

[0050] Figure 2 This is a front cross-sectional view of the device provided in Embodiment 1 of the present invention.

[0051] Figure 3 This is a cross-sectional view of the overall device provided in Embodiment 1 of the present invention.

[0052] Figure 4 This is a side sectional view of the linkage transmission mechanism of the lower housing provided in Embodiment 1 of the present invention.

[0053] Figure 5 This is an exploded view of the linkage transmission mechanism of the lower housing provided in Embodiment 1 of the present invention.

[0054] Figure 6 This is a flowchart of the method provided in Embodiment 2 of the present invention.

[0055] Figure 7 This is a schematic diagram of the timestamp alignment operation provided in Embodiment 2 of the present invention.

[0056] Figure 8 This is a schematic diagram of adaptive transmission bandwidth usage provided in Embodiment 2 of the present invention.

[0057] Figure 9 This is a schematic diagram comparing Hausdorff geometric distance and pixel loss as provided in Embodiment 2 of the present invention.

[0058] Figure 10 This is a schematic diagram of the weight distribution for adaptive dual-stream feature fusion provided in Embodiment 2 of the present invention.

[0059] Figure 11 This is a schematic diagram of the gradient advantage of the edge-sensitive loss function provided in Embodiment 2 of the present invention.

[0060] Figure 12 This is a schematic diagram of the disaster discrimination plane provided in Embodiment 2 of the present invention.

[0061] Figure 13 This is a schematic diagram of the response curve for the increase of icing load provided in Embodiment 2 of the present invention.

[0062] Figure 14 This is a schematic diagram illustrating the convergence of the Newton-Raphson method provided in Embodiment 2 of the present invention.

[0063] Figure 15 This is a schematic diagram of the fault cause analysis provided in Embodiment 2 of the present invention.

[0064] Figure 16 This is a schematic diagram of the convergence curve provided in Embodiment 2 of the present invention. Attached image description: 1-Upper housing, 2-Lower housing, 3-High voltage cable, 4-Guide rod, 11-Lock, 12-Torsion spring A end, 13-Torsion spring B end, 21-Touch sensor, 22-Hook connecting rod, 23-Top hook slider, 24-Pull rope, 25-Windlass, 26-Stop post, 27-Pressure guide connecting rod, 28-Fixing buckle, 29-Push rod spring, 221-Push rod, 222-Locking rod tongue, 223-Turntable. Detailed Implementation

[0066] 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.

[0067] Example 1 This embodiment discloses a high-voltage line in-situ multi-functional monitoring device. The multi-functional monitoring device in this embodiment has a clamp-type structure and can be attached to the surface of the high-voltage cable 3 in situ. The device integrates a multi-dimensional sensing module and an automatic locking / unlocking mechanism, and can be used with drones to achieve rapid installation and retrieval.

[0068] Figure 1 An overall device diagram of this embodiment is shown. Figure 2 A front cross-sectional view of this embodiment is shown; see below. Figure 1 As can be seen, in this embodiment, the main body of the remote multi-functional monitoring device for the high-voltage cable 3 is formed by hinged connection between the lower housing 2 and the upper housing 1. In the initial state (i.e., the state before installation), such as Figure 1 As shown, the upper housing 1 and the lower housing 2 are in the open position, and together they define an upward-opening U-shaped or semi-circular cable channel.

[0069] To ensure flexible clamping of the high-voltage cable 3 and effective coupling of the sensors after the device is closed, an inner wall liner is laid on the inner wall surface of the cable channel, and the detection end of a multi-source sensing module is embedded in the inner wall liner: The probe surface of the ultrasonic detection unit is exposed or flush with the surface of the inner wall liner; the cable temperature measurement unit (contact thermal element) is also embedded in the liner, with the probe part protruding from the liner surface to ensure physical contact.

[0070] See Figure 2 As shown, guide rods 4 are vertically arranged on both sides of the lower housing 2. The lower end of the guide rod 4 is inserted into a side hole of the lower housing 2 via a detachable mechanical connection (detailed in the subsequent linkage disengagement mechanism). The upper end of the guide rod 4 extends to the top of the device for connecting the suspension rope of the UAV. The two guide rods 4 form a funnel-shaped or door-frame-shaped guiding space. When the UAV hoisting device approaches the high-voltage cable 3 from bottom to top, the two guide rods 4 first restrict the lateral position of the high-voltage cable 3. As the device is raised, the high-voltage cable 3 smoothly slides into the cable channel along the inner side of the guide rods 4. See also Figure 1The diagram clearly shows the intricate mechanical triggering logic hidden inside the lower housing 2. The core of the automatic locking mechanism lies in the control and release of preset elastic potential energy. A pressure sensor 21 is installed at the lowest point of the cable channel in the lower housing 2 (i.e., at the center of the groove bottom). This pressure sensor 21 protrudes from the bottom of the channel in the form of a button or a push rod. This position typically mates with the opening of the aforementioned inner wall lining, ensuring that force can be directly applied to this component when the high-voltage cable 3 is seated. The lower end of the pressure sensor 21 is physically connected to an internal pressure guide rod 27, which is configured as a lever mechanism or linkage mechanism that can rotate around a fulcrum.

[0071] like Figure 3 As shown in the cross-sectional view, one end of the pressure-guided connecting rod 27 receives the downward pressure from the pressure sensor 21, and the other end extends to the fixing buckle 28. The fixing buckle 28 is a key locking component that maintains the device in the open state. In the initial state, the fixing buckle 28 hooks or locks the drive component of the upper housing 1 or the lever arm of the torsion spring, resisting the torsion force of the torsion spring. The closing force of the device comes from the torsion spring (or a similar torsion spring). Figure 3 As shown, the torsion spring is installed at the hinge shaft of the upper and lower housings 2. End A 12 (fixed end) of the torsion spring abuts tightly against a stop post inside the lower housing 2. The stop post provides a reference point for the reaction force, allowing the torsion spring to store elastic potential energy. End B 13 (actuating end) of the torsion spring abuts against or is fixedly connected to the upper housing 1. When the high-voltage cable 3 is fully submerged in the cable channel and presses against the bottom pressure sensor 21, the following continuous mechanical actions occur: Displacement transmission: The gravity of the high-voltage cable 3 overcomes the reset elastic force of the touch pressure sensor 21 component, pushing the touch pressure sensor 21 to move downward.

[0072] Link unlocking: The downward movement of the touch sensor 21 drives the pressure guide link 27 to rotate around the axis, causing the end of the pressure guide link 27 to trigger or actuate the fixing buckle 28.

[0073] Release potential energy: The fixed buckle 28 is displaced, instantly releasing the limiting restraint on the upper shell 1 or the torsion spring B end 13.

[0074] Driven closure: Since the A end 12 of the torsion spring is firmly held in place by the stop post and cannot move, the accumulated elastic potential energy forces the B end 13 of the torsion spring to rebound violently. The B end 13 of the torsion spring directly pushes the upper housing 1 to rotate inward around the hinge axis.

[0075] Cable locking: The upper housing 1 and the lower housing 2 quickly close together, and the inner wall pad (including the multi-source sensing module) is squeezed and deformed under the action of the torsion spring, tightly covering the surface of the high-voltage cable 3, completing the physical installation and sensor bonding.

[0076] When the upper housing 1 of the device rotates and closes (to the closed position) under the drive of the torsion spring, the following locking action occurs: Figure 3As shown, a protruding latch 11 is provided on the edge of the upper housing 1, and a sliding top hook slider 23 and a hook connecting rod 22 for locking are provided on the corresponding closed surface of the lower housing 2. When the upper housing 1 is closed in place, the inclined front end of the latch 11 first contacts and presses against the top hook slider 23. The downward pressure of the latch 11 forces the top hook slider 23 to move (usually downward), thereby making room for hooking. Under the continuous pushing force of the push rod 221 and the spring 29, the hook connecting rod 22 tends to rotate and reset. Once the top hook slider 23 moves out of place and the latch 11 enters the predetermined locking groove, the hook connecting rod 22 quickly rotates and resets under the action of the push rod 221 and the spring 29, and its hook head precisely engages in the groove of the latch 11.

[0077] At this point, the upper housing 1 and lower housing 2 are firmly locked and cannot be easily opened by external force, ensuring the stable suspension of the device on the high-voltage cable 3. Furthermore, with the locking buckle 11 engaged, a constant preload is applied to the segmented end faces of the current sensing unit (Rogowski coil or transformer core) located at the parting surface of the upper housing 1 and lower housing 2, thereby achieving a tight connection of the closed magnetic circuit, eliminating magnetic and air gaps, and ensuring the accuracy of high-load current measurements.

[0078] Figure 4 This figure shows a side sectional view of the linkage transmission mechanism of the lower housing 2 in this embodiment. Figure 5 An exploded view of the linkage transmission mechanism of the lower housing 2 in this embodiment is shown. As can be seen from the figure, the rotational locking action of the hook rod 22 can be converted into the unlocking action of the guide rod 4, such as... Figure 4 As shown, the linkage mechanism is mainly located in the bottom space of the lower housing 2: First-stage transmission (connecting rod rotation converted into linear thrust): A long strip-shaped push rod 221 is hinged to the tail (non-hook head end) of the hook connecting rod 22. When the hook connecting rod 22 is rotated and locked under the action of the spring 29 of the push rod 221, its rotational lever arm drives the push rod 221 to undergo longitudinal linear displacement.

[0079] Secondary transmission (converting linear thrust into rotational / disc force): The other end of push rod 221 is connected to a rotatable turntable 223 (or dial / cam). The linear movement of push rod 221 drives turntable 223 to rotate around its central axis by a certain angle.

[0080] Stage transmission (rotation converted to bidirectional retraction): A locking lever tongue 222 is connected to the turntable 223. It should be noted that, although Figure 4 Only one side view is shown, but based on the description of both guide rods 4 sliding out simultaneously, and Figure 5From this perspective, those skilled in the art will understand that the turntable 223 is typically connected to a symmetrical pair of locking lever tongues 222, or the locking lever tongues 222 on both sides are controlled simultaneously via a linkage mechanism. The rotational movement of the turntable 223 causes the locking lever tongues 222 to retract towards the center of the device (or away from the guide rod 4 slot). (Refer to...) Figure 4 and Figure 5 It can be seen that the automatic disengagement of the guide rod 4 in this embodiment includes: Initial locked state, i.e., before device installation (i.e.) Figure 1 (State), the locking tongue 222 is in the extended state, and its end is inserted into the guide rod 4, the pre-reserved locking hole or slot at the bottom, to rigidly fix the guide rod 4 to the lower housing 2 so that it can bear the weight of the device.

[0081] Instantaneous release state: Once the automatic locking mechanism completes its action, the locking lever tongue 222 retracts via the linkage of push rod 221 and turntable 223. Disengagement process: The locking hole at the bottom of the guide rod 4 loses the support of the locking lever tongue 222. At this moment, the connection between the guide rod 4 and the lower housing 2 is instantly released. Due to gravity (and the presence of a drone above), the guide rod 4 will slide out of the insertion hole in the lower housing 2. Operation completion: The drone flies upward, carrying away the two guide rods 4 for reuse, while the monitoring device remains independently on the high-voltage cable 3 to begin monitoring operations.

[0082] In addition, it will be combined Figure 4 The remote retrieval and reset mechanism of the device in this embodiment is described in detail. This mechanism allows operators to control the device to automatically detach and retrieve it from the ground, away from the high-voltage line environment, via wireless commands, without the need to send personnel to climb the tower for retrieval. When the monitoring task is completed or the device needs maintenance, the operator can issue a retrieval command to the device through the cloud platform. The main control unit inside the device (integrated on the circuit board) receives the command through the wireless communication module, and after security verification, outputs a drive signal.

[0083] like Figure 4 As shown, a winch 25 driven by a micro-motor is installed inside the device (usually located in the cavity of the lower housing 2). In response to a signal from the main control unit, the winch 25 begins to rotate, performing the rope winding action. A high-strength pull rope 24 is wound around the winch 25, and the other end of the pull rope 24 is connected to the lever arm end of the hook rod 22. As the winch 25 rotates, the pull rope 24 is tightened, applying a pulling force to the hook rod 22. This pulling force overcomes the resistance of the spring 29 of the push rod 221, forcing the hook rod 22 to rotate in the opposite direction about its axis (i.e., opposite to the direction of rotation when locked). As the hook rod 22 rotates, its hook head gradually disengages from the slot of the latch 11, thereby releasing the mechanical locking state between the upper housing 1 and the lower housing 2.

[0084] To ensure the device can be opened smoothly even under harsh conditions (such as icing and adhesion), this embodiment also includes a linkage between a booster opening and a spring-loaded unloading mechanism; such as Figure 4 As shown, when the hook rod 22 unlocks, the tension component of its mechanical structure or the pull rope 24 pushes the hook slider 23 upward. The upward-moving hook slider 23 directly acts on the opening edge of the upper housing 1, generating an active thrust that forces the upper housing 1 to open outward. To prevent the upper housing 1 from being hindered by the closing force of the torsion spring when it opens, this device is designed with a linkage between the stop post and the unlocking mechanism; in the closed state, the stop post blocks the A end 12 of the torsion spring, allowing it to accumulate closing torque; while in the retracted state, as the hook rod 22 is pulled and rotated by the pull rope 24, the stop post, which has a mechanical linkage with it (or is integrally formed), shifts position or rotates. Figure 3 As shown, after the stop post shifts, it no longer obstructs the torsion spring A end 12. At this time, the torsion spring A end 12 loses its support point, and the torsion spring instantly releases its remaining elastic potential energy and enters a free state, no longer applying a closing torque to the upper housing 1. This allows the upper housing 1 to open freely under the action of gravity or a very small external force. With the locking released, the torsion spring unloaded, and the push of the top hook slider 23, the upper housing 1 of the device is fully opened, losing its gripping force on the high-voltage cable 3. Under its own gravity, the device naturally slides off the high-voltage cable 3.

[0085] To prevent damage to the internal high-precision sensors (such as the multi-source sensing module) from a direct fall, the device is equipped with a parachute pack (not shown in detail in the figure) on its exterior or bottom. After the device detaches from the cable, the parachute automatically deploys, allowing the device to land at a safe speed for easy recovery by personnel. After recovery, personnel can manually rotate the reset stop and torsion spring A end 12 using a special tool, and reinsert the guide rod 4 into the locking hole at the bottom of the lower housing 2, restoring the device to its original position. Figure 1 The installation status shown is for use in the next task.

[0086] In addition, the device integrates the following sensors to perform the monitoring method described in Example 2, including an ultrasonic probe for monitoring icing thickness, a temperature sensor, a current transformer, and an inclination and accelerometer for monitoring cable condition; wherein, the temperature sensor is a contact-type thermistor probe embedded in the inner wall, directly measuring the temperature of the cable core, which is more accurate than infrared thermometry. The current transformer (CT) forms a closed magnetic circuit after the device is closed, and the built-in induction coil can measure the cable current in real time. The inclination and accelerometer integrates a MEMS six-axis inertial sensor (IMU) to monitor the cable's inclination angle (for calculating sag) and motion acceleration (for monitoring galloping and wind deflection frequency) in real time.

[0087] To accurately distinguish between the physical phases of frost and hoarfrost, an ultrasonic probe can be mounted on the lower housing 2, facing the icing-prone surface of the high-voltage cable 3. A miniature fisheye camera can also be integrated on the outer side of the upper housing 1 to capture the cable's external profile. The ultrasonic probe emits high-frequency pulse waves, and the main control unit collects the interface echo signal from the cable's upper surface, transmitting it back to the cloud platform to perform the acoustic impedance inversion step as described in Example 2.

[0088] In addition, in order to ensure that the device can achieve long-term maintenance-free operation in high-pressure environments in the field, the monitoring device in this embodiment can also be equipped with a dual-mode complementary power module of solar energy and magnetic field induction. The module includes a photovoltaic module disposed on the outer surface of the housing, an induction power collection coil integrated inside the housing, and a built-in energy storage battery unit.

[0089] When there is sufficient sunlight or the high-voltage line is under low load (low current, weak magnetic field), the solar unit is used as the main power source to charge the battery and maintain low power standby.

[0090] Simultaneously, utilizing the clamping and locking mechanical characteristics of this device, semi-annular high-permeability magnetic cores are embedded inside the mating surfaces of the upper housing 1 and the lower housing 2. When the device completes mechanical triggering and locking, the semi-annular magnetic cores inside the upper and lower housings align to form a complete closed magnetic ring, surrounding the high-voltage cable 3. The high-voltage cable 3 serves as the primary winding, and the power-taking coil wound on the closed magnetic ring serves as the secondary winding. According to the principle of electromagnetic induction, the power frequency load current in the cable induces an AC voltage in the power-taking coil. This induced voltage is processed by the rectification, filtering, and voltage regulation modules on the internal circuit board, and then converted into DC power and stored in the built-in battery. This method is suitable for nighttime, rainy days, or periods of high line load operation, ensuring the device's all-weather online monitoring capability under extreme weather conditions.

[0091] The built-in battery of this device can be a wide-temperature lithium battery or a supercapacitor pack, which can be installed in the sealed compartment of the lower housing 2 and connected to the power management unit (PMU). The PMU is configured with a two-way energy dispatch strategy: it prioritizes the use of inductive power, and automatically switches to solar power when the line current is insufficient (such as below the start-up threshold), and stores the excess energy in the battery to drive high-power actions (such as the rotation of the winch motor during recovery or the high-frequency emission of the ultrasonic probe).

[0092] The device described in this embodiment achieves touch-to-lock, eliminating the need for motor-driven closure and significantly reducing installation power consumption and complexity. Simultaneously, the in-situ clamp-type structure ensures zero-gap contact between the ultrasonic probe, temperature probe, and cable, making the data sources for icing density discrimination (rime / fog) and thermal fault decoupling analysis more accurate and reliable.

[0093] Example 2 This embodiment discloses a multi-functional in-situ monitoring method for high-voltage lines. Figure 6 The overall method flowchart of this embodiment is shown. As can be seen from the figure, this embodiment includes the following steps: Step 1: Synchronously collect multi-dimensional sensing data of high-voltage cables using a multi-source sensing module.

[0094] Among them, the multi-source sensing module refers to an integrated sensing device that can adapt to the complex electromagnetic environment and harsh weather conditions of high-voltage transmission lines in the field. This terminal can usually be deployed at the suspension point of the high-voltage cable tower, near the tension clamp, or at key monitoring points of the cable itself, aiming to achieve comprehensive physical quantitative capture of the cable's operating environment and its physical state, that is, the sensing module built into the device as shown in Example 1.

[0095] Multidimensional sensing data refers to a heterogeneous set of data used to comprehensively describe the current operating status of high-voltage cables from multiple dimensions, including optical characterization, thermodynamic distribution, spatial geometry, and acoustic medium properties. Specifically, this multidimensional sensing data includes visible light images of the cable surface acquired using high-definition cameras, temperature field distribution data of the cable surface acquired using infrared thermal imagers, cable clearance distance data to the ground acquired using lidar, medium echo data of the cable surface acquired using ultrasonic sensors, and tangential tilt angle data of the cable suspension point acquired using high-precision tilt sensors.

[0096] In this embodiment, to overcome the perception blind spots of a single sensor in specific environments (e.g., visible light fails at night, or infrared lacks texture details when illumination is uniform), the multi-source sensing module adopts a dual-channel synchronous acquisition strategy of visible light and infrared thermal imaging. Visible light images are mainly used to capture the texture features of the cable surface, the geometry of ice accumulation, and background environmental conditions (such as tree obstruction); while the temperature field distribution data of the cable surface reflects the temperature difference distribution between the cable body and surrounding attachments (such as ice layers and water droplets).

[0097] It is understandable that high-voltage cables are typically hotter than ambient temperatures due to the thermal effect of the current, while ice or snow accumulation is at or near ambient temperatures. Therefore, by collecting temperature field distribution data, thermal imaging principles can be used to logically segment the cable body and cold-state attachments on a thermal image, thus providing a thermal basis for solving the identification challenges of high-voltage cables at night or under low light conditions.

[0098] In this embodiment, the cable clearance data is obtained by vertically scanning downwards using a miniature lidar installed at the cable suspension point or midway between spans. This data characterizes the vertical distance from the lowest point of the cable to the ground or tree canopy. It is understood that, according to the geometry of catenary cables, the cable clearance to the ground is negatively correlated with the cable sag; when the cable sags due to high-temperature expansion or increased weight from icing, the clearance to the ground decreases accordingly. The logical significance of collecting this data is to provide a direct geometric observation value for subsequent verification of the accuracy of the theoretical sag derived from the mechanical model, forming the basis for constructing a dual geometric-mechanical verification system.

[0099] In this embodiment, the cable surface dielectric echo data is obtained by emitting pulse waves to the cable surface using a piezoelectric ultrasonic transducer and receiving the reflected echoes. This data includes not only the echo's flight time but, more importantly, its amplitude attenuation characteristics. It is understood that, based on the physical principle of acoustic impedance, when sound waves are reflected at interfaces of different densities (such as air-ice, air-water, and air-metal), their reflectivity is a function of the medium's density and sound velocity characteristics. The core logic of acquiring this data lies in utilizing the sensitivity of sound waves to medium density to distinguish between loose frost (low density, high gas content, low acoustic impedance) and dense glaze (high density, hardness, high acoustic impedance), which are difficult for visual sensors to differentiate. This compensates for the inability to perceive the internal physical texture of ice simply by relying on images.

[0100] In this embodiment, the tangential tilt angle data of the cable suspension point reflects the angle between the tangential direction of the cable at the suspension position and the horizontal plane. It is understood that, according to the derivative relationship of rigid body statics and the catenary equation, there is a strict monotonic mapping relationship between the tangential tilt angle of the suspension point and the total tension within the cable span and the total load per unit length (cable self-weight + icing weight + wind load). When the cable becomes heavier due to icing, its suspension angle will undergo a small but measurable change under tension. The purpose of collecting this data is to provide boundary condition input for calculating the equivalent weight of the cable from the perspective of mechanical equilibrium, enabling the system to detect early icing (such as transparent rime) that has not yet caused significant shape changes but has already resulted in significant weight changes.

[0101] In this embodiment, to ensure the spatiotemporal consistency of the aforementioned multi-dimensional sensing data in subsequent fusion calculations, the multi-source sensing module performs a strict timestamp alignment operation; that is, at the same trigger moment, the readings of the camera, radar, ultrasonic, and tilt sensor are simultaneously frozen to avoid data noise introduced by differences in acquisition time (such as wind-induced vibration causing inconsistencies in position at different times), thus ensuring that all data describe the cable status at the same physical moment. Figure 7 A schematic diagram of the timestamp alignment operation in this embodiment is shown.

[0102] Step 2: Upload the multi-dimensional sensing data to the multi-source data fusion sensing system of the backend cloud platform through the wireless communication network, and use the intelligent visual recognition model deployed in the cloud platform to identify the visual recognition boundary of the cable.

[0103] In the special operating conditions of high-voltage cable monitoring, the wireless communication network typically consists of industrial-grade wireless communication modules (such as multi-mode gateways supporting 4G / 5G / NB-IoT) deployed on poles and operator base stations. Given that high-voltage transmission lines often traverse uninhabited areas or regions with complex terrain, signal coverage is often unstable, and monitoring terminals typically rely on solar panels and batteries for power, making them extremely sensitive to power consumption. Therefore, this embodiment incorporates a state-aware adaptive hierarchical transmission logic.

[0104] To achieve efficient data transmission, this embodiment adopts an edge filtering-event triggered data flow control strategy; by changing the high bandwidth consumption mode of traditional full pass-through, it adopts a transmission mechanism that prioritizes feature uploading and supplements it with raw data.

[0105] Specifically, the edge computing unit built into the multi-source sensing module processes massive amounts of raw data (such as high frame rate video streams and high-frequency ultrasonic waveforms) in real time. Under normal operating conditions (i.e., no abnormal features are detected), the system only extracts and uploads lightweight feature vector data, such as the coordinates of the lowest point of the cable, average temperature value, maximum wind deflection angle, and average ultrasonic energy value every 15 minutes. This strategy is based on the logic of information entropy compression; under normal conditions, continuous raw waveform data contains a large amount of redundant information (low information entropy), while the feature data is sufficient to support health trend analysis in the cloud. When the edge side detects suspected fault features (such as a visually apparent white covering, sudden changes in ultrasonic impedance, or sharp fluctuations in tilt angle), an abnormal transmission interruption mode is immediately triggered.

[0106] By dynamically adjusting transmission priority and sampling density, the system can immediately identify original data segments (such as high-definition short videos and high-frequency vibration waveforms) before and after an anomaly occurs, mark them as high-priority data packets, and preempt the transmission channel. Simultaneously, the system automatically reduces the data update frequency of other non-critical sensors (such as ambient humidity) to ensure that limited bandwidth resources are entirely used to transmit critical on-site fault evidence. This mechanism ensures that even in bandwidth-constrained field environments, maintenance personnel can still obtain a complete chain of evidence for diagnosing faults.

[0107] Furthermore, to address the unique electromagnetic interference and weak signal scenarios inherent in high-voltage cables, a lightweight protocol based on a publish / subscribe model (such as MQTT) can be adopted for data transmission, with deep integration of application-layer QoS (Quality of Service) dynamic mapping logic. It is understandable that different data types have different reliability requirements. This embodiment maps different types of data packets to different QoS levels: for regular data, it is mapped to QoS 0 (at most once), allowing packet loss in extremely poor network conditions to avoid the retransmission mechanism draining battery power; for alarm and fault characteristic data, it is strictly mapped to QoS 2 (only once and must be delivered), ensuring that data is never lost or duplicated in unstable links through a two-way handshake confirmation mechanism, ensuring the integrity and timeliness of data transmission from the edge to the cloud-based fusion sensing center. Figure 8 A schematic diagram of adaptive transmission bandwidth usage in this embodiment is shown.

[0108] In addition, for the transmission of image data, a region of interest (ROI) encoding logic is adopted. The system identifies the cable and tower areas in the image as the foreground and encodes them with lossless or low compression ratio; while the background (sky, forest) is encoded with high compression ratio. Through this semantic-based compression logic, the transmission efficiency of images in narrowband networks can be further improved without sacrificing key details.

[0109] Additionally, this embodiment may include 'breakpoint resumption and local caching logic'. When the monitoring terminal detects a communication link interruption or a signal-to-noise ratio below the transmittable threshold via RSSI (Received Signal Strength Indicator), the system automatically writes the data to be sent into a circular buffer in local non-volatile memory. Once the signal is restored or a drone is used as a relay node, the system automatically uploads the cached content in reverse order (i.e., prioritizing the latest important data) based on the data's timestamp and priority, thus ensuring the spatiotemporal continuity of the monitoring data is not affected by communication blind spots.

[0110] The cloud platform serves as the core computing hub, deploying a heterogeneous parallel intelligent visual recognition model and a physical state calculation engine. These two components work together to achieve visual-physical dual verification of cable status. In this embodiment, the intelligent visual recognition model processes visible light images and infrared thermal imaging data, employing an edge-aware weighted dual-stream attention network architecture. The core design logic of this architecture lies in solving the challenge of recognizing high-voltage cables in complex backgrounds (such as mountains or forests) where the target is minute and the lighting conditions are variable.

[0111] Specifically, this embodiment includes two parallel feature extraction branches. The first branch processes visible light images to extract texture details, color features, and visual boundaries of icing on the cable surface. The second branch processes infrared thermal image data to extract the thermal radiation difference between the cable and its surrounding environment. It is understandable that simply superimposing the two types of data does not leverage the advantages of multimodal processing. Therefore, this embodiment introduces a dynamic gating fusion mechanism based on channel attention. The network automatically generates a set of targeted weight coefficients by learning the feature saliency in different scenarios. When ambient light is sufficient, the system assigns higher weights to the visible light branch to utilize its high-resolution texture advantage; while in nighttime or foggy environments where visible light information is lacking, the system automatically increases the weight of the infrared branch, using temperature difference features to identify the cable outline, ensuring all-weather perception capability.

[0112] In this embodiment, to accurately quantify the thickness of the icing, the training process of the visual recognition model abandons the traditional loss function that only focuses on pixel classification accuracy. Instead, it introduces Hausdorff distance constraint logic, which focuses on the degree of geometric shape matching. Mathematically, Hausdorff distance describes the maximum mismatch between two point sets (predicted contours and real contours). Translated into textual logic, this constraint mechanism means that during optimization, the model is no longer satisfied with merely finding the approximate location of the cable, but is forced to ensure that the predicted cable edge curves are geometrically infinitely close to the real cable edge. Even if the model correctly classifies the vast majority of background pixels, if there is a slight distance deviation between its predicted icing boundary and the real boundary, this constraint mechanism will generate a large penalty signal. This forces the model to pay extreme attention to changes in cable thickness, making it possible to infer the physical icing thickness from the image pixel width, significantly improving the detection accuracy of minute icing. Figure 9 A schematic diagram comparing the Hausdorff geometric distance with traditional pixel loss in this embodiment is shown, illustrating how this constraint mechanism captures the minute edge features of the cable.

[0113] Specifically, in this embodiment, to address the extremely fine line characteristics of high-voltage cables against a complex background, and the boundary blurring caused by the thin and highly transparent ice layer in the early stages of icing, an intelligent visual recognition model fusing infrared and visible light multimodal data is constructed. This model can be built through the following steps: 1) Construct a dual-stream feature fusion network based on a global gating mechanism to address the insufficient feature representation capability of a single sensor under different lighting and weather conditions. High-voltage cable monitoring scenarios cover all-weather environments. Visible light images (RGB) provide rich texture details during the day but fail at night or in foggy conditions; infrared thermal images (IR) provide contour information based on temperature differences but lack texture; simple channel stitching cannot reflect the difference in importance between the two in different scenarios. Therefore, it is necessary to construct a fusion module based on channel attention, modeling the feature fusion process as a dynamically weighted problem. Figure 10 A schematic diagram of the weight distribution for adaptive dual-stream feature fusion in this embodiment is shown.

[0114] Specifically, firstly, two independent convolutional neural network branches are used to extract deep feature maps for visible light and infrared light, respectively. To achieve adaptive fusion, global average pooling is used to compress spatial information and extract a global context vector representing the current environmental state (e.g., nighttime, bright light). Subsequently, a learnable global gating weight matrix is ​​used to calculate the response weight of each feature channel. This process essentially teaches the network how to automatically adjust its dependence on infrared or visible light features based on ambient light intensity: automatically amplifying the weight of infrared features at night and balancing the two during the day.

[0115] For example, in this embodiment, the attention weighting mechanism for dual-stream feature fusion can be calculated using the following formula:

[0116] in, The final fused multimodal feature map contains weighted infrared and visible light information; The feature map is obtained by the visible light branch through a CNN feature extractor. Feature maps extracted from infrared thermal imaging branches; For channel-level splicing operations, heterogeneous features are merged into... tensor; This is a global average pooling operation, its function is to reduce the spatial dimension. Compress to Output a global context descriptor vector, which does not contain spatial location information, but contains the statistical distribution characteristics of the entire screen; The global gate weight matrix is ​​a learnable parameter that is continuously optimized during training. It is a fully connected layer (DenseLayer) responsible for mapping the global context vector to the channel attention weight vector. The Sigmoid activation function is used to normalize the weight values ​​to... The interval represents the importance score of the feature channel; The Hadamard product (element-wise multiplication) is used to weight and activate each channel of the original concatenated features using the calculated attention weight vector, thereby achieving feature recalibration.

[0117] It should be noted that, through the above calculation process, when ambient light is sufficient, the cloud system automatically calculates high weighting coefficients for the visible light channel to leverage its high-resolution texture advantage; while in nighttime or foggy environments where visible light information is lacking, the system automatically increases the weight of the infrared branch, using temperature difference features to distinguish cable outlines, thereby ensuring the robustness of the perception model in all-weather environments. 2) An edge-sensitive Hausdorff distance loss function is introduced to address the challenge of conventional pixel-level loss functions failing to capture minute icing boundaries. After obtaining fused features, the network needs to segment the cable and its icing area. The conventional cross-entropy loss function is based on the assumption of pixel independence, focusing only on the correctness of individual pixel classification. However, cable icing often occupies only a very small proportion of pixels in an image (linear targets), and the icing edges are extremely faint. In cases of severe imbalance between positive and negative samples, cross-entropy loss is easily dominated by the background, leading to cable 'breaks' or blurred boundary predictions. Therefore, geometric constraints must be introduced to elevate the evaluation criterion from pixel accuracy to overall shape similarity.

[0118] Specifically, this embodiment introduces Hausdorff distance as part of the loss function; Hausdorff distance is a geometric metric that measures the similarity between two sets of points, focusing on the maximum mismatch between two contours. Even if the vast majority of pixels are predicted correctly, if the predicted edge deviates significantly from the true edge at some point, the Hausdorff distance will incur a large penalty value. This mechanism forces the network to pay close attention to the continuity of cables and the geometric accuracy of edges during training, thereby achieving a keen perception of micron-level icing thickening. Figure 11 The diagram illustrates the gradient advantage of the edge-sensitive loss function in this embodiment, showing that when the predicted edge deviates from the true edge, the Hausdorff loss can provide a sustained and large gradient penalty, while the cross-entropy is prone to saturation.

[0119] For example, in this embodiment, the total loss function is a weighted sum of the classification loss and the geometric loss, as shown in the following formula:

[0120] in, For conventional cross-entropy loss, This is the balance coefficient for the conventional cross-entropy loss; These are the balance coefficients for geometric loss. These two balance coefficients are used to adjust the proportion of geometric constraints in the overall optimization objective.

[0121] Furthermore, to address the issue of the standard Hausdorff distance calculation being non-differentiable, a weighted average Hausdorff distance is employed, with the specific calculation formula as follows:

[0122] in: This is the set of coordinates of the pixels at the edge of the cable icing predicted by the model; This is the set of coordinates of the cable edge pixels marked in the actual label (GroundTruth); and These represent the number of pixels in the prediction set and the true set, respectively; This represents the square of the Euclidean distance.

[0123] It should be noted that this formula consists of two parts, implementing a two-way distance constraint, wherein the first term... For the prediction set Each point in In the real set Find the point closest to it. And calculate the square of that minimum distance; this term penalizes false positives, i.e., predicted edge points that are far from the actual edge (e.g., predicted icing is thicker than it actually is). The second term... For the real set Each point in In the prediction set Find the point closest to it. This penalty applies to false negatives, i.e., cases where there are no predicted points near the true edge point (e.g., missing a section of icing). By minimizing this... The loss function allows the network to directly optimize the geometric fit between the predicted contour and the real contour, thereby outputting extremely high-precision cable diameter data, providing reliable geometric input for calculating icing thickness.

[0124] Step 3: Based on the characteristics of ultrasonic echoes and changes in gravitational torque, execute the special monitoring logic for freezing rain disasters to distinguish between rime ice and hoarfrost.

[0125] The special monitoring logic for freezing rain disasters refers to the processing flow for refined classification and early warning of the complex icing morphology unique to the cold and humid climate of southern China. Frost is usually formed by supercooled droplets freezing instantly upon contact with power lines. It has the physical characteristics of being dense, highly transparent, strongly adhesive, and dense, posing a significant threat to the mechanical load of transmission lines. In contrast, rime is usually formed by the sublimation of water vapor in the air. It has the characteristics of being loose, opaque, granular, and low in density, with a weight per unit volume much smaller than frost.

[0126] Specifically, in this embodiment, the special monitoring logic for freezing rain disasters is not the general icing monitoring, but a refined physical classification specifically for the complex phase transition processes unique to the cold and humid climate of southern China. Addressing the challenge of distinguishing between rime ice and hoarfrost, since both appear as white deposits in the visible light spectrum, it is difficult to determine their internal density using only optical images. Therefore, this step constructs an acoustic impedance-gravity coupling discrimination model, achieving physical identification of different disaster types through joint verification of acoustic medium detection and mechanical feedback.

[0127] Understandably, when sound waves propagate from one medium to another, their interface reflection characteristics depend on the difference in acoustic impedance between the two media. Since acoustic impedance is the product of medium density and sound velocity, there is a significant difference in acoustic response between dense frost and loose hoarfrost. When a pulse wave emitted by an ultrasonic sensor contacts the deposits on the cable surface, the system performs waveform analysis on the received echo signal. If the echo signal shows a high reflection coefficient or specific waveform attenuation characteristics, according to the acoustic physics model, this corresponds to a high acoustic impedance value, thus deducing that the deposit has a high physical density, logically pointing to frost characteristics; conversely, if the echo signal characteristics correspond to a low acoustic impedance, it is deduced that the deposit contains a large number of air gaps and has a low density, logically pointing to hoarfrost characteristics. This process transforms the abstract sound wave signal into a quantitative assessment of the microscopic density of the material.

[0128] In this embodiment, to further eliminate potential misjudgments from a single sensor (such as acoustic parameter drift caused by sensor surface contamination), the system executes mechanical verification logic based on changes in gravitational torque in parallel. It is understood that the high-voltage cable, as a flexible object suspended between two towers, experiences changes in gravitational load that are reflected in the tangential tilt angle at the suspension point in real time. With a constant span, an increase in the total weight of the cable inevitably leads to an increase in the vertical component of the force at the suspension point, causing a monotonic change in the tangential tilt angle. The system calculates the equivalent density increment of the cable by monitoring the rate of change of the tilt angle sensor readings over time and combining this with the current rate of increase in icing thickness (measured visually or ultrasonically as described above).

[0129] If monitoring data shows that the increase in ice thickness is not significant, but the tilt angle changes rapidly and significantly, it indicates that the deposit per unit thickness is extremely heavy and can only be high-density rime. Conversely, if the ice thickness increases rapidly but the tilt angle changes gradually, it indicates that the deposit is loose and light, and is most likely rime.

[0130] In this embodiment, the final disaster determination employs acoustic-mechanical dual AND gate logic; that is, the system only determines that it is currently in a high-risk freezing rain disaster mode when the medium density calculated from ultrasonic features exceeds a preset high-risk threshold, and the gravity increase inferred from the tilt angle change also simultaneously exceeds the corresponding gravity threshold. This design, based on cross-validation of multiple physical quantities, effectively prevents harmless light frost or brief bird stops from being falsely reported as severe line icing faults, thus ensuring the uniqueness and accuracy of monitoring conclusions under severe weather conditions.

[0131] Specifically, in this embodiment, the acoustic impedance-gravity coupling discrimination model can be constructed through the following steps: 1) Construct a medium density estimation algorithm based on ultrasonic reflection echo characteristics to transform the invisible internal physical structure into a quantifiable acoustic impedance index. Different types of icing (rime and hoarfrost) have drastically different microstructures. Rime has no air bubbles, a dense structure, and high acoustic impedance; hoarfrost is filled with air gaps, has a loose structure, and low acoustic impedance. According to acoustic wave propagation theory, when an ultrasonic beam is incident from a sensor (or air) onto the icy surface, the magnitude of its reflected energy directly depends on the acoustic impedance difference between the media on both sides of the interface.

[0132] Specifically, firstly, the precise distance obtained by lidar is used to correct for the energy attenuation of ultrasonic waves during air transmission, thereby calculating the accurate interface reflection coefficient. Then, based on the physical relationship between the reflection coefficient and acoustic impedance, the acoustic impedance value of the icing medium is inferred. Finally, combined with empirical sound velocity (which varies depending on the ice type), the volume density of the icing is calculated. This process achieves a physical inversion from surface echo to internal density. Figure 12 This embodiment shows a schematic diagram of a disaster discrimination plane based on both acoustic and mechanical features, defining the threshold areas for determining rime and hoarfrost.

[0133] For example, in this embodiment, the ice density can be calculated using the following formula:

[0134] To calculate the reflection coefficient in the above formula This requires considering the ratio of transmitted to received signal energy and compensating for path loss.

[0135] in, The desired bulk density of the icing is (unit: This is the most crucial physical quantity for distinguishing between freezing rain and rime ice; The known acoustic impedance of the ultrasonic probe surface is used as the reference value for calculation. This is the speed of sound in ice; this parameter is usually set to an empirical range value (e.g., the speed of sound in rime ice). Alternatively, it can be obtained through further measurement using dual-frequency ultrasound; The reflection coefficient of sound waves at the icy interface represents the proportion of energy reflected. The amplitude of the ultrasonic echo received by the sensor; This is the initial amplitude of the ultrasonic wave during emission; It is the sound wave attenuation coefficient in air, which is affected by ambient temperature and humidity; This is the precise distance from the sensor to the cable surface measured by the coaxial lidar, used to eliminate interference from distance variations on the echo intensity.

[0136] 2) Construct a gravity growth verification model based on tilt angle-tension transformation, and combine it with acoustic features to achieve multi-dimensional coupled disaster type determination. Considering the limitations of relying solely on acoustic density estimation due to single-point measurement (e.g., icing on sensor surfaces may lead to misjudgments), to ensure the reliability of early warnings, macroscopic mechanical features must be introduced for verification. The most significant hazard of rime ice lies in its high density, leading to a rapid increase in weight. Therefore, by monitoring the rate of change of cable tilt angle and tension, the real-time mass growth rate of the cable can be estimated.

[0137] Specifically, trigonometric relationships can be used to decouple the axial tension change of the cable into a vertical gravity component change. Only when both the microscopic medium density and the macroscopic gravity increase simultaneously exceed the high-risk threshold will the system determine that the current situation is a dense frost disaster. This dual verification mechanism of acoustics and mechanics constitutes a logically rigorous criterion.

[0138] For example, in this embodiment, the gravity growth verification and final discrimination logic can be represented by the following formula: First, estimate the mass gain of the cable per unit time (gravity growth verification):

[0139] Subsequently, a multidimensional coupling criterion is constructed:

[0140] in, This represents the estimated cable quality growth rate. The rate of change of axial tension in the cable over time is acquired in real time by a force sensor. The readings from the tilt sensor reflect the current sag of the cable. It is a geometric projection factor used to convert changes in axial tension into changes in gravitational load in the vertical direction; This is a structural correction factor, which is related to the cable span and specific installation method; This is a Boolean value. The system outputs a frost warning if and only if this value is true. The critical threshold for density (e.g., set to) A value greater than this indicates that the medium is dense; This is the threshold for the rate of increase in gravity, used to exclude instantaneous mechanical fluctuations caused by wind deflection or vibration. By using a logical AND operation, the system ensures that alarms are triggered only when the physical properties and mechanical performance are highly consistent, thus achieving high-precision, low-false-alarm discrimination. This cross-verification of microscopic physical properties and macroscopic mechanical performance ensures that the system only triggers alarms when a genuine "freezing rain and frost" disaster occurs, providing a low-false-alarm decision-making basis for power grid operation and maintenance departments to take de-icing measures.

[0141] Step 4: Construct a load decoupling algorithm and a physical state calculation engine that includes the mechanical state equation of the cable. Perform decoupling analysis on the icing state and mechanical state of the cable. By calculating the fault discrimination operator, distinguish between high-temperature sag and icing sag.

[0142] In this embodiment, the core task performed by the physical state calculation engine is state decoupling, that is, using the logic of state equations in physics to decompose the cable deformation into two independent components: thermodynamic deformation and mechanical deformation, thereby accurately identifying the cause of the fault. The basis for constructing the cable mechanical state equation is that, as a flexible catenary structure, the geometry of high-voltage cables (such as sag) is strictly controlled by their material properties (elastic modulus, coefficient of thermal expansion), ambient temperature, and the loads they bear (self-weight, ice weight, wind pressure). The specific process of decoupling analysis is as follows: First, the system uses cable point cloud data acquired by LiDAR to calculate the actual total sag of the cable at the current moment through orthogonal projection and curve fitting algorithms. This physical quantity represents the final equilibrium state of the cable under the combined action of all external factors (temperature, wind, icing, and current thermal effects).

[0143] Secondly, the system is based on the average surface temperature of the cable measured by an infrared sensor, combined with the inherent physical parameters of the cable at the time of manufacture (coefficient of linear expansion). Elastic modulus Weight per unit length The initial state parameters before icing are used to derive a theoretical thermal sag using the state equation. This theoretical thermal sag is a virtual reference value that represents the ideal physical state of the cable assuming that it is currently only affected by temperature changes and has no additional icing load.

[0144] Finally, by comparing the measured values ​​with the theoretical values, a fault discrimination operator is constructed. If the measured values ​​are similar to the theoretical values, it indicates that the cable sag is mainly caused by the current heating effect or the thermal expansion of the metal material due to the high ambient temperature. The system is judged to be in a high temperature and high load state, and the scheduling strategy should tend to reduce the current load. Conversely, if the measured values ​​are significantly greater than the theoretical values, according to the principle of force balance, the difference between the two (i.e., the deformation residual) can only be caused by an additional vertical downward mechanical force.

[0145] Understandably, by comparing the actual total sag with the theoretical thermal sag, different physical influencing factors can be separated. If the two values ​​are highly similar, it indicates that the cable sag is mainly caused by thermal expansion of the material due to high temperature, and the system should be judged as a high-temperature, high-load condition, rather than an icing fault. Conversely, if the actual total sag is significantly greater than the theoretical thermal sag, the difference between the two (i.e., the fault identification operator) can only be caused by additional mechanical load. Combined with the ambient temperature logic at this time (usually low temperature), this additional mechanical load is locked as the weight of icing. This decoupling logic based on first principles of physics effectively avoids misjudgments that may occur based solely on visual images (such as mistaking sag caused by thermal expansion for heavy icing pressure), providing a decisive basis for the formulation of operation and maintenance strategies (whether to reduce load or de-ic).

[0146] The fault identification operator is a computational index used to quantify and separate the sources of cable geometric deformation. During operation, the sag of high-voltage cables is typically driven by two main factors: one is the thermal expansion of materials due to current heating or environmental temperature rise (thermodynamic factors), and the other is the mechanical stretching caused by icing, snow accumulation, or strong wind loads (mechanical factors). By using a load decoupling algorithm, these two coupled deformation effects are decoupled, thereby accurately locating the physical root cause of abnormal sag in high-voltage cables. Figure 13 A schematic diagram of the response curve of the fault discrimination operator in this embodiment to the increase of icing load is shown.

[0147] In this embodiment, the basis for constructing the mechanical state equation of the cable is derived from the mechanics of materials and the catenary theory. This state equation essentially describes the nonlinear equilibrium relationship that a high-voltage cable, as an elastic extension body, must follow among its internal tension, ambient temperature, and external specific load (weight per unit length).

[0148] In this embodiment, the fault detection operator is calculated using differential comparison logic, which involves subtracting the actual total sag from the theoretical thermal sag. The physical meaning of this operator is to filter out the thermal effect component and extract the deformation margin purely caused by external mechanical weight gain. Based on this operator, the system can execute the following multi-functional detection logic: When monitoring data indicates a significant increase in cable sag (i.e., a safety hazard exists), but the calculated fault identification operator approaches zero (i.e., the actual sag closely matches the theoretical thermal sag), the underlying physical fact is that the cable sag is entirely consistent with the material expansion law under the current high temperature, and there is no additional weight stretching. At this time, the system will lock the fault logic as excessive sag caused by high temperature and heavy load, and the generated control strategy should be directed to the electrical side, such as scheduling current reduction or load shedding, rather than mechanical de-icing.

[0149] Conversely, when the calculated fault diagnosis operator is a significantly positive value, it means that the actual total sag is much greater than the theoretical thermal sag that should exist at that temperature. This indicates that although the current temperature (usually low) should cause the cable to contract and tighten, some external force is forcibly stretching and compressing it. Based on this physical contradiction, the system identifies the fault logic as excessive sag caused by icing overload. At this point, even if the visual image is blurred due to nighttime or dense fog, this mechanical evidence is sufficient to support the system in generating an emergency recommendation for mechanical de-icing or de-icing.

[0150] By employing this subtraction-based approach using physical models, this step effectively addresses the chaotic operational decision-making caused by the inability to distinguish between thermal expansion and heavy pressure in traditional monitoring. It ensures that the system can provide diagnostic results consistent with the physical nature of the problem in two drastically different meteorological scenarios: high temperatures in summer and cold waves in winter.

[0151] Specifically, in this embodiment, the load decoupling algorithm including the cable mechanical state equation can be constructed through the following steps: 1) Using an improved cable state equation, a thermodynamic benchmark for a theoretical ice-free state is constructed. It is understood that high-voltage cables are flexible cable structures, and their stress and sag changes follow specific physical state equations. To determine whether there is ice accumulation, a proof by contradiction assumption is made: assuming the cable is currently ice-free and only affected by its own weight and the current ambient temperature, what stress state should it be in at this moment?

[0152] Specifically, the initial calibration state of the cable (known temperature, initial stress, and specific load) is used as state 1, and the current monitoring moment is used as state 2; a nonlinear state equation is introduced to describe the cable's response to temperature changes. The goal of this step is to calculate the stress evolution under varying specific loads (assuming a constant specific load, only self-weight).

[0153] For example, in this embodiment, the state equation describing the change of the high-voltage cable from the initial state to the current monitoring state can be shown as follows:

[0154] in, The core unknown to be solved is the unknown quantity at the current temperature. The theoretical thermal stress is assumed to be without icing (only subject to its own weight); The cable surface temperature is measured in real time by an infrared thermal imager or temperature sensor. The calibration temperature during system installation and initialization; To be at the calibration temperature The initial stress measured below; The specific load of the cable itself (i.e., the gravity per unit length and unit cross-sectional area) is this value under the assumption of no ice. This is the load specific in the initial state (usually also the load specific due to the cable's own weight). The cable span (the horizontal distance between two towers); The elastic modulus of a cable characterizes the material's ability to resist elastic deformation. The linear expansion coefficient of the cable is a key thermodynamic parameter for distinguishing thermal faults, determining the extent to which the cable expands and contracts with temperature.

[0155] It should be noted that the above equation is a problem concerning... In engineering implementations, cubic equations are typically solved numerically using the Newton-Raphson method. Figure 14 A schematic diagram illustrating the convergence of the Newton-Raphson method in this embodiment is shown.

[0156] 2) The actual stress is inverted using lidar measurement data, and the difference between the theoretical thermal sag and the actual sag is calculated to construct a fault discrimination operator. After calculating the theoretical thermal stress, it needs to be converted into a geometric quantity, namely the theoretical thermal sag, so as to make a comparison with the actual sag directly measured by lidar in the same dimension.

[0157] First, based on the parabolic approximation theory (applicable to cables with high tension and small sag), the actual sag measured by lidar is used. To infer the current actual horizontal stress This step is to obtain the current true mechanical state of the cable (including all loads, i.e., its own weight plus any ice weight).

[0158] For example, in this embodiment, the actual horizontal stress can be obtained by inversion using the following formula:

[0159] At the same time, the theoretical thermal stress obtained in the aforementioned steps is used Calculate the theoretical thermal arc sag under the ice-free assumption. :

[0160] Finally, define the fault detection operator. By calculating the deviation between the actual geometric shape and the theoretical thermodynamic shape, the load decoupling and attribution are achieved.

[0161] For example, in this embodiment, the fault detection operator and detection logic can be calculated using the following formula:

[0162] in, This is the current actual stress inferred from measured data; The actual sag of the cable as measured in real time by the lidar; This is the theoretical thermal sag calculated considering only the effect of temperature; The mass per unit length of the cable (weight); For the unknown icing mass (implied in the actual sag in the inversion formula) (among the reasons for its formation) It is the acceleration due to gravity; For fault detection, its physical meaning is the sag increment caused by non-thermodynamic factors (i.e., mechanical icing). Figure 15 A schematic diagram of the fault cause deconstruction analysis in this embodiment is shown.

[0163] It should be noted that the method in this embodiment is based on The discrimination logic can be as follows: Scenario 1: Thermal failure (high temperature sagging); if (Within the allowable error range), and The value is relatively large, which means that the actual sag observed is completely consistent with the theoretically calculated pure thermal sag; therefore, the conclusion is that the cable is not covered with ice, and the increase in sag is entirely caused by the thermal expansion of the material due to the increase in temperature.

[0164] Scenario 2: Mechanical failure (ice accumulation and sagging); if (Significantly positive value) This means the cable appears to sag lower than theoretically possible. This additional sag is due to the fact that theoretical calculations have already eliminated the effects of temperature. This can only be caused by the stretching effect of additional weight (ice); therefore, the conclusion is that the cable is iced, and The size is proportional to the severity of the icing.

[0165] Step 5: Utilize a Physical Information Fusion Long Short-Term Memory (PI-LSTM) network to construct a cable status trend prediction model to generate cable status trend predictions.

[0166] Among them, the Physics-Informed Long Short-Term Memory (PI-LSTM) network refers to an improved recurrent neural network architecture. Its purpose is to break away from the black-box model of traditional deep learning, which relies solely on statistical data for regression prediction. By explicitly embedding the physical and mechanical constitutive equations of high-voltage cables at the network's objective function level, it achieves physical consistency constraints on the prediction results. This step aims to address the problem that, in extreme freezing rain or windstorm scenarios, traditional pure data-driven models are prone to overfitting or outputting predictions that violate physical principles due to a lack of sufficient historical fault samples (e.g., predicting a surge in icing weight while the sag remains unchanged). This allows for the generation of future state trends that both conform to historical meteorological evolution patterns and strictly adhere to physical and mechanical boundaries.

[0167] In this embodiment, the input layer logic for constructing the cable condition trend prediction model is based on the reconstruction of multi-dimensional time series. The system maps the environmental parameter sequence within the historical observation window (such as the temperature gradient, humidity change rate, and wind speed and direction vector over the past M hours) and the cable condition parameter sequence (such as historically calculated icing thickness and historically measured sag) into high-dimensional tensor inputs. The LSTM unit utilizes its internal cell state and gating mechanisms (forget gate, input gate, output gate) to capture the long-term temporal dependencies and nonlinear dynamic evolution characteristics between these meteorological factors and cable conditions. For example, the network can learn the implicit rule that sustained low temperature and high humidity usually predict an exponential increase in icing thickness.

[0168] In this embodiment, the PI-LSTM network output layer employs a multi-task coupled prediction strategy. At each prediction time step, the network does not simply output the future icing weight, but simultaneously outputs two key state variables: the predicted icing weight and the predicted cable sag. It is understood that these two output variables are not independent in physical reality, but are strongly coupled variables strictly constrained by the mechanical properties of the cable material. To ensure the network is aware of this coupling, this embodiment constructs a composite loss function containing physical residual terms during the model training and optimization phases. Specifically, the textual logic of this composite loss function consists of two parts: The first part is the data fitting error, which measures the statistical deviation between the network's predicted values ​​and the historical actual observations, driving the network to approximate the changing trends of historical data as closely as possible.

[0169] The second part is the core physical consistency residual. This residual substitutes the predicted icing weight output by the network and the predicted ambient temperature input into the high-voltage cable state equation, which describes the cable stress-sag relationship and was established in step 4. Through the rigid calculation of this physical equation, a physically correct theoretically derived sag is derived. Subsequently, the system calculates the numerical difference between the predicted cable sag directly output by the network and this theoretically derived sag. This difference is defined as the physical residual. If the network predicts a state combination that violates physical laws (e.g., predicting a huge ice weight but a small sag value, resulting in a large deviation from the theoretical sag calculated by the equation), this physical residual term will generate a large penalty value. During backpropagation, this penalty gradient forcibly adjusts the weight parameters of the neural network, forcing the network to converge to a physically feasible region that both conforms to statistical laws and strictly satisfies the physical state equation when searching for the optimal solution.

[0170] In this embodiment, to balance the relationship between experiential learning and physical constraints, an adaptive hyperparameter weight coefficient is introduced. In the early stages of training, by assigning greater weight to the physical residual term, the network is guided to quickly grasp the basic mechanical properties of the cable (i.e., learn the physical laws first). In the later stages of training, the weight of the data fitting error is gradually increased, enabling the network to fine-tune to adapt to the personalized characteristics of specific micro-meteorological regions (i.e., relearn environmental special cases).

[0171] In this embodiment, based on the trained PI-LSTM model, the system performs future disaster trend projection; the cloud system receives numerical weather forecast data for the next N hours issued by the meteorological department as input, and the model recursively generates the icing growth curve and sag descent curve for future times.

[0172] Because the prediction results have undergone rigorous physical and logical verification, they possess extremely high interpretability and reliability. When the model predicts that at some point in the future, the accumulated icing weight will cause the cable tension to exceed the tower's design safety factor, or that sag will cause the distance to ground to be less than the safe clearance for insulation breakdown, the cloud platform will immediately generate a tiered early warning signal. This warning not only includes the alarm time but also clearly indicates the predicted type of physical violation (whether it is mechanical overload or insufficient electrical clearance), thus providing the power grid dispatching department with precise decision-making basis. For example, it allows for advance scheduling of DC de-icing operations or line load transfer, achieving a technological leap from reactive post-event repair to proactive pre-event defense. Figure 16 A schematic diagram of the convergence curve of the loss function during the PI-LSTM training process in this embodiment is shown.

[0173] Specifically, in this embodiment, the cable status trend prediction model can be constructed through the following steps: 1) Construct a time-series input vector containing multi-dimensional meteorological and state characteristics, and define the prediction target of coupled physical quantities. To predict the future evolution trend of cables, it is first necessary to time-align and normalize heterogeneous data from different sensors to construct a feature vector describing the complete environmental state at the current moment. Simultaneously, it is clarified that the network output is not just a single sag value, but must also include the intrinsic physical factor causing sag changes—ice weight. This is because sag is a phenomenon, while ice weight is the essence, and the two are physically strongly coupled. By predicting these two variables simultaneously, a computational foundation is provided for subsequent physical constraints.

[0174] Specifically, the input layer receives the past The time window sequence includes surface temperature, ambient humidity, wind speed, tilt angle, and lidar ranging values. The output layer predicts the sag size and icing weight for the next time step.

[0175] For example, in this embodiment, the input-output mapping relationship of PI-LSTM can be defined as:

[0176] The network's input vector and predicted output vector can be represented by the following equation:

[0177]

[0178] in, for The input feature vector at time step; The surface temperature of the cable measured by infrared radiation; The relative humidity measured by the micro-weather station; Wind speed; It is the angle of inclination; Visibility distance (which reflects fog concentration); For model predictions The state vector at any given time; The predicted cable sag value; This is the predicted mass of icing per unit length; Represents the mapping function of the LSTM network; These are network parameters.

[0179] 2) Construct a physical constraint loss function, embedding the cable state equation into the optimization objective of the neural network, forcing the model to learn physical laws. Traditional deep learning only focuses on the error (MSE) between the predicted value and the true label; as long as the value is close, it is considered correct. However, in this embodiment, the concept of physical residual is introduced: that is, the aforementioned cable state equation is used to verify whether the sag and ice weight predicted by the network are contradictory.

[0180] Specifically, this is achieved by treating the cable state equation as an inviolable physical operator. If, under the predicted temperature conditions, the theoretical sag of the predicted ice weight is inconsistent with the predicted sag value, it indicates that the network's prediction violates physical laws. This degree of violation is quantified as part of the loss function. Thus, during backpropagation to update weights, the network not only strives to approximate historical data but also actively avoids violating physical equations, enabling it to deduce correct results from physical logic even in extreme conditions lacking data.

[0181] For example, in this embodiment, the physical constraint loss function can be represented by the following formula:

[0182] Physical residuals It can be expressed by the following formula:

[0183] in, Let be the total loss function for network training, and the optimization objective is to minimize it; This is based on authentic historical observation data; This is the predicted output of the LSTM network; The traditional data-driven loss (MSE) is used to ensure that the prediction results conform to the historical data distribution; These are the physical constraint weights (hyperparameters) used to balance the importance of data fit and physical consistency. They should be increased when training data is scarce. The value is more dependent on physical laws; The predicted sag is directly output by the network; The next time-period temperature prediction (or known weather forecast value) is input into the network. The predicted ice weight is directly output by the network; This is a nonlinear state equation mapping function. The function encapsulates the aforementioned cable state equation solution process. It takes the predicted ice weight and predicted temperature as inputs and calculates the theoretical sag that should exist based on the principles of rigid body physics and thermodynamics.

[0184] It should be noted that in the formula Essentially, it is a physical consistency checker. If the neural network predicts: the temperature remains constant, the ice weight increases, but the sag remains constant, then: 1. 1. A larger theoretical sag is calculated based on the increase in ice weight; 2. Direct prediction by the network. However, it remains unchanged; 3. The difference between the two will lead to 4. The huge gradient will be fed back to the network, forcing it to correct its parameters until it predicts the result that the increase in ice weight leads to an increase in sag, which is consistent with physical logic.

[0185] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A multi-functional in-situ monitoring device for high-voltage lines, comprising: The airframe assembly, the mounting guide mechanism, and the trigger drive mechanism disposed within the airframe assembly, wherein the airframe assembly includes: an upper housing and a lower housing hinged to each other, the upper housing and the lower housing cooperating to form a cable channel for high-voltage cables to pass through; characterized in that the mounting guide mechanism includes: a guide rod disposed on the airframe assembly, the guide rod being configured to connect to an aircraft to suspend the airframe assembly; The trigger drive mechanism is configured to release a preset elastic potential energy in response to the action of the high-voltage cable entering the cable channel and touching a preset position, so as to drive the upper housing to close relative to the lower housing. The high-voltage line in-situ multi-functional monitoring device further includes: an automatic locking mechanism configured to automatically lock the body components when the upper housing is closed in place; and The linkage release mechanism is connected between the automatic locking mechanism and the installation guide mechanism. It is configured to convert the locking action of the automatic locking mechanism into a mechanical driving force, and release the connection between the guide rod and the body assembly at the same time as the body assembly completes locking, so that the guide rod is released from the body assembly.

2. The high-voltage line in-situ multifunctional monitoring device according to claim 1, characterized in that, The triggering drive mechanism includes: A pressure-triggered component is located at the bottom of the cable channel and configured to displace under the pressure of the high-voltage cable; An energy storage component has a fixed end for storing the elastic potential energy and an actuating end for outputting power; and A transmission release assembly is connected between the pressure trigger assembly and the energy storage component; The displacement of the pressure-triggered component releases the limiting effect on the energy storage component through the transmission release component, and the actuating end of the energy storage component pushes the upper housing to close.

3. The high-voltage line in-situ multifunctional monitoring device according to claim 2, characterized in that, The energy storage component is a torsion spring, and the body assembly is equipped with a baffle. In the state of waiting to be installed, the fixed end of the torsion spring abuts against the stop post to maintain the energy storage state; The device further includes a remote recovery mechanism configured to release the locking state of the automatic locking mechanism in response to a remote command, and simultaneously move the stop post or release the stop post from obstructing the fixed end of the torsion spring, so that the torsion spring unloads its elastic potential energy.

4. The high-voltage line in-situ multifunctional monitoring device according to claim 3, characterized in that, The remote recovery mechanism includes a winch and a pull rope; The winch is located inside the machine body assembly, one end of the pull rope is wound around the winch, and the other end is connected to the hook rod of the automatic locking mechanism; The winch is configured to wind up the pull rope to pull the hook rod to rotate in the opposite direction to unlock it, and the unlocking action of the hook rod triggers the displacement of the stop post.

5. The high-voltage line in-situ multifunctional monitoring device according to claim 1, characterized in that, The automatic locking mechanism includes: a latch, a top hook slider, and a hook connecting rod; The latch is disposed on the upper housing, and the top hook slider and the hook connecting rod are disposed on the lower housing; The hook rod is configured to rotate under the action of the push rod spring to hook the latch after it is closed; The linkage disengagement mechanism includes a push rod, a rotating component, and a locking lever tongue; One end of the push rod is connected to the hook connecting rod, and the other end is connected to the rotating component; The rotating component is connected to the locking lever tongue via a transmission connection; The rotation of the hook rod drives the rotating component to rotate via the push rod, which in turn causes the locking tongue to retract and release the guide rod.

6. The high-voltage line in-situ multifunctional monitoring device according to claim 1, characterized in that, The high-voltage line in-situ multi-functional monitoring device also includes: a multi-source sensing module and a main control unit integrated on the body components; The inner wall of the cable channel is provided with an inner wall liner, and the multi-source sensing module includes an ultrasonic detection unit, a cable temperature measurement unit, and a current sensing unit embedded in the inner wall liner. The probe surface of the ultrasonic detection unit is configured to emit ultrasonic waves and receive acoustic echoes characterizing the density of the icing medium. The cable temperature measuring unit uses a contact-type thermistor element embedded in the inner wall liner to measure the temperature of the cable body. The current sensing unit includes: a Rogowski coil or a current transformer core split at the junction of the upper and lower housings. The current sensing unit is configured to form a closed magnetic circuit in response to the closing and locking action of the upper housing, so as to measure the current of the high-voltage cable. The main control unit, electrically connected to the multi-source sensing module, is used to process the collected data and transmit it externally via a wireless communication module. The main control unit has a built-in attitude monitoring module and an edge computing module. The attitude monitoring module includes a six-axis inertial sensor for real-time acquisition of tilt and acceleration data as the device moves with the cable; the edge computing module receives acoustic echo data from the ultrasonic detection unit and tilt data from the attitude monitoring module, and runs freezing rain disaster monitoring logic.

7. The high-voltage line in-situ multifunctional monitoring device according to claim 6, characterized in that, The main control unit is also configured to perform a timestamp alignment operation, synchronously freeze the readings of all sensors in the multi-source sensing module at the same trigger time, and send the multi-dimensional sensing data packets containing timestamps to the cloud platform through the wireless communication module. The cloud platform stores a computer program, and when the computer program is executed by the processor, it implements the high-voltage cable remote multi-functional method as described in any one of claims 8 to 10.

8. A multi-functional in-situ monitoring method for high-voltage lines, characterized in that, The remote multi-functional monitoring method includes: S100. The high-voltage line in-situ multi-functional monitoring device as described in any one of claims 1 to 7 synchronously collects multi-dimensional sensing data of the high-voltage line, and performs a timestamp alignment operation on the multi-dimensional sensing data. S200: Upload the multi-dimensional sensing data to the backend cloud platform through a wireless communication network, and use an intelligent visual recognition model to identify the visual recognition boundary of the cable. S300. Based on the acoustic and mechanical characteristics in the multi-dimensional sensing data, execute the freezing rain disaster monitoring logic to distinguish the type of attachments on the cable surface; S400. Construct a load decoupling model that includes the mechanical state equation of the cable. Use the load decoupling model to decouple the icing state and mechanical state of the cable and calculate the fault discrimination operator to distinguish between high-temperature sag and icing sag of the cable. S500. Construct a cable state trend prediction model using a long short-term memory network that integrates physical information, and generate cable state trend prediction results based on the cable state trend prediction model. In step S200, edge-sensitive geometric constraint logic is introduced during the training process of the intelligent visual recognition model: The geometric similarity index is used to measure the maximum mismatch between the contour point set predicted by the model and the true label contour point set; The maximum mismatch is introduced as a geometric penalty term into the loss function to constrain the cable edge curve predicted by the model to infinitely approximate the real boundary in geometric space; The intelligent visual recognition model adopts a dynamic gating fusion mechanism based on channel attention: it automatically adjusts the weight coefficients of the visible light branch and the infrared branch features according to the ambient light intensity; in low light environment, it automatically increases the feature response weight of the infrared branch and uses temperature difference features to identify the cable boundary. In the S300, the freezing rain disaster monitoring logic includes: using the energy response of the sound wave pulse when it is reflected at the interface of a heterogeneous medium, and combining the distance parameters provided by spatial geometric detection methods, to perform gain compensation for the path loss of the sound wave; extracting acoustic impedance characteristics that characterize the physical density of the attachments on the cable surface based on the compensated energy response; and mapping the acoustic impedance characteristics with the macroscopic mechanical feedback of the cable under gravity load to determine the disaster type of the attachments. The method for determining the type of disaster caused by the attached material includes: if the acoustic impedance characteristics indicate that the attached material is dense and the displacement deviation of the cable suspension point indicates that the mass growth rate exceeds a threshold, then it is determined to be a frost disaster mode; if the acoustic impedance characteristics indicate that the attached material contains air gaps and the displacement deviation of the cable suspension point indicates that the mass growth is gradual, then it is determined to be a hoarfrost feature.

9. The high-voltage line in-situ multifunctional monitoring method according to claim 8, characterized in that, The high-voltage line in-situ multi-functional monitoring method according to claim 1 is characterized in that, in step S400, the process of calculating the fault discrimination operator includes: Based on the preset baseline state parameters of the cable and combined with the real-time temperature characteristics, a thermodynamic simulation path is constructed under the assumption of no additional load constraints in order to obtain the theoretical baseline value characterizing the pure thermodynamic deformation of the cable. The actual geometric shape of the cable is obtained by using spatial three-dimensional point cloud detection data, so as to obtain the measured physical quantity characterizing the current comprehensive deformation; The offset of the measured physical quantity relative to the theoretical benchmark value is extracted as the mechanical load deformation residual, which is used as the fault discrimination operator.

10. The high-voltage line in-situ multifunctional monitoring method according to claim 8, characterized in that, The logic for distinguishing between high-temperature sag and icing sag of cables includes: If the fault detection operator approaches zero within the preset error range, the abnormal deviation of the cable is attributed to the thermal expansion of the material caused by high temperature. If the fault identification operator is significantly positive, the abnormal cable offset is attributed to overload sagging caused by external mechanical load. In the S500, the process of constructing the cable status trend prediction model includes: Embed consistency verification operators based on prior physical knowledge in neural network architecture; The mechanical logic residuals between multiple predictive variables directly output by the model are calculated using the consistency check operator. By minimizing the mechanical logic residual, the neural network is constrained to perform state evolution deduction within the physically feasible domain that conforms to the mechanical constitutive equation of the cable. The neural network architecture performs multi-task coupled prediction: The system synchronously outputs the predicted icing amount, which reflects the nature of the load, and the predicted sag amount, which reflects the geometric response. If the theoretical geometric value of the predicted icing amount derived from the physical equation does not match the predicted sag amount, the weight parameters of the neural network are forcibly corrected using the generated penalty signal.

Images