Power transmission line icing thickness dynamic estimation method and system

Abstract

The invention discloses a method and a system for dynamically estimating the icing thickness of a power transmission line, which are used for realizing real-time high-precision online detection on the icing thickness of the surface of an overhead power transmission conductor. According to the method, an ultrasonic excitation and receiving module is arranged on the side edge of a power transmission line, emits ultrasonic waves to the surface of a wire and receives echo signals of the wire; and transmitting the received echo signal to a far-end processing terminal, performing short-time Fourier transform to obtain a time-frequency spectrogram, extracting a Mel frequency spectrum to construct a multi-dimensional feature vector, inputting the multi-dimensional feature vector into the trained deep neural network model for reasoning calculation, and outputting an icing thickness estimated value of a corresponding position. The system has the functions of remote ultrasonic acquisition, autonomous alignment, signal enhancement, multi-dimensional feature extraction, deep learning prediction and the like, can adapt to the icing state detection requirements of the power transmission line under complex environmental conditions, has the advantages of high real-time performance, high precision, flexible deployment and the like, and is suitable for scenes such as power grid icing risk early warning, intelligent operation and maintenance, remote inspection and the like.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of transmission line state detection, and particularly relates to a transmission line icing thickness dynamic estimation method and system. BACKGROUND

[0002] In cold and high-humidity areas, transmission line icing phenomena occur frequently, and in severe cases, can lead to conductor breakage, tower collapse and other accidents, bringing great hidden dangers to the safe operation of the power grid. With the increase in extreme weather events, the power system has higher requirements for real-time monitoring and early warning of the icing state of the transmission line. Especially in complex climate conditions such as mountainous areas, plateaus and wind-affected areas, icing occurs frequently and develops rapidly. The traditional manual inspection method has been difficult to meet the demand for dynamic control of icing thickness changes and timely response due to information lag and limited human resources.

[0003] The current mainstream transmission line icing detection methods mainly include video image recognition, conductor tension monitoring, laser ranging and meteorological model calculation, etc. Among them, the image recognition method relies on visible light acquisition equipment and good environmental conditions, and is easily affected by factors such as fog, light, etc., with poor stability and adaptability; the inference method based on tension changes, capacitance, resistance and other electrical parameters can indirectly perceive the conductor load, but it is extremely sensitive to external disturbance factors such as wind and temperature changes, and can only determine whether icing occurs, but cannot quantitatively reflect the change in ice thickness; the laser ranging method has high precision, but is high in cost, complex in structure and sensitive to installation angle, making it difficult to be widely deployed; the prediction path based on the meteorological model relies on historical meteorological data such as temperature, humidity, wind speed and physical modeling, and its prediction has regional and trend characteristics, lacking the ability to feedback on the specific line segment, making it difficult to support fine and dynamic response of the icing thickness control strategy.

[0004] In order to improve the monitoring accuracy and reliability, some research attempts to introduce multi-physical quantity fusion technology, combining temperature, humidity, current, capacitance, resistance and other parameters to infer the icing state. However, such methods often require multiple sensors to work together, and the system integration is difficult, and still mainly focuses on state judgment, making it difficult to realize quantitative dynamic estimation of ice thickness. In some academic research, new wave detection technologies such as microwave radar and ground wave radar have also been explored for their potential application in icing perception, but these methods often have high cost, poor anti-interference ability, strong dependence on installation location and angle, and other engineering application obstacles. In addition, these technical paths generally focus on the judgment of the presence or absence of ice layer and the auxiliary analysis of visual features, and lack a dynamic detection mechanism that can directly associate the change in ice thickness with the physical propagation mechanism and has strong universality and does not depend on specific image scenes. SUMMARY

[0005] To solve the above technical problems, the application provides a power transmission line icing thickness dynamic estimation method and system, which uses high-frequency non-interference active ultrasonic excitation signals to obtain reflected echo data under non-contact sensing conditions, and realizes high-precision estimation of ice layer thickness through deep learning modeling.

[0006] Specifically, the technical scheme provided by the application is as follows: A power transmission line icing thickness dynamic estimation method, comprising the steps of: S1, emitting an ultrasonic excitation signal to the power transmission line and collecting a corresponding ultrasonic echo signal; S2, pre-processing and short-time Fourier transform of the echo signal to obtain a frequency domain power spectrum; S3, Mel spectrum mapping of the frequency domain power spectrum to construct a two-dimensional time-frequency graph; S4, inputting the two-dimensional time-frequency graph into a neural network to obtain an icing thickness prediction value; S5, dynamic correction of the prediction value to output an icing thickness estimation value.

[0007] Further, the pre-processing in step S2 includes: setting the frequency band range of the band-pass filter according to the excitation frequency of the ultrasonic excitation signal, band-pass filtering the collected echo signal, and performing normalization processing and amplitude standardization operation on the filtered signal; then performing echo envelope extraction and main reflection segment cropping: constructing a signal envelope line through Hilbert transform, extracting a main reflection energy segment, and intercepting the signal segment before and after the first significant peak value.

[0008] Further, the intercepted signal segment is framed and windowed: the whole signal is cut into several short-time signals according to the set frame length and frame shift, and each short-time signal is multiplied by a window function; then through fast Fourier transform, the frequency domain power spectrum of each frame signal is obtained.

[0009] Further, in step S3: In the desired frequency range, M frequency bands are uniformly divided according to the Mel scale, and a triangular band-pass filter is designed for each frequency band, each filter reaches a maximum value of 1 at the center frequency and gradually decays to 0 towards adjacent frequency bands, constructing an overlapping filter bank; Multiply the frequency domain power spectrum of each frame signal by the Mel filter bank and accumulate to get the power in each Mel frequency band:

[0010] Wherein, Pm represents the power in the mth Mel frequency band, Pm(k) is the power of the kth FFT frequency point, k ​is the response of the mth Mel filter at frequency point k , are the FFT indices corresponding to the m-1th and m+1th Mel frequencies, respectively; The power values on each Mel band are spliced to obtain a Mel power spectrum vector M with a length of , which is used to represent the energy distribution of the current frame on the Mel scale; The Mel power spectrum vectors corresponding to the frames of signals are spliced into a two-dimensional time-frequency graph in time sequence.

[0011] Further, the neural network in step S4 includes a time-frequency convolution module, a BiLSTM unit and a full connection regression module connected in sequence; The time-frequency convolution module regards the two-dimensional time-frequency graph as an image and extracts local deep features of the two-dimensional time-frequency graph by using multiple convolution layers; the BiLSTM unit is used to input the feature vector output by the time-frequency convolution module into the BiLSTM network to extract time sequence evolution features; and the full connection regression module is used to input the feature vector output by the BiLSTM unit into the full connection neural network to obtain the predicted value of the ice thickness through multiple full connection layers.

[0012] Further, the dynamic correction in step S5 includes: introducing a sliding window filter in the time dimension to smooth the predicted values for multiple times in succession; through calculation of the deviation between the predicted value and the measured value, using a sliding mean offset or weight scaling correction strategy to make the predicted value approach the true ice thickness; and establishing a constraint rule library to eliminate abnormal predicted values that violate physical common sense.

[0013] A power line ice thickness dynamic estimation system based on the above method includes an ultrasonic excitation and reception module, a data acquisition and caching module, a communication transmission module, a remote processing terminal, a result feedback and display module, and a remote control terminal for issuing control instructions to the ultrasonic excitation and reception module, and data links are established between the modules through wireless or optical fiber networks; The ultrasonic excitation and reception module is used to convert the excitation control instructions sent by the remote control terminal into physical ultrasonic pulse signals transmitted to the power line, and to receive ultrasonic echo signals in real time; The data acquisition and caching module is used to sample the echo signals at a high speed and convert them into digital signals, and to package the digital signals and return them to the remote processing terminal through the communication module after sampling is completed; The remote processing terminal is provided with a complete feature extraction module and a neural network model, and is used to output corresponding ice thickness estimation values according to input signals; ​The result feedback and display module is used to push the estimated result to the power grid dispatch center interface, mobile terminal APP or ground inspection workstation in real time, and is also used to link the early warning module, and when the estimated thickness exceeds the set threshold, an alarm is triggered.

[0014] Preferably, the system is also configured with a dynamic correction and output module, which is used to make boundary judgment and dynamic calibration on the model output result in real time; at the same time, the module also supports interface interaction with external unmanned aerial vehicle systems, infrared imaging devices or laser range finders, fuses external sensing information, compares and verifies the estimated result and makes micro deviation correction.

[0015] Further, the ultrasonic excitation and receiving module is installed on the side of the power transmission line, including an ultrasonic exciter, an ultrasonic receiver, a fixed support, a power supply and a control module; the ultrasonic exciter selects a piezoelectric ceramic transducer or a MEMS ultrasonic transducer, and the excitation signal waveform adopts a frequency-modulated pulse or a Gaussian window envelope sine pulse wave.

[0016] Further, the ultrasonic excitation and receiving module is also configured with an angle adjusting component and a visual positioning component; the visual positioning component includes a camera and an image processing unit, which is used to identify the spatial orientation information of the power transmission line in real time through image processing technology, and calculate the relative deviation angle between the module and the power transmission line; the angle adjusting component automatically drives the integrated micro servo motor or piezoelectric driving mechanism according to the deviation angle, drives the ultrasonic excitation and receiver to pitch or horizontally rotate around the center axis of the installation support, until the transmission axis of the module is aligned with the power transmission line, realizing directional excitation and echo collection.

[0017] Traditional ice coating detection mainly relies on image recognition, infrared imaging or laser ranging, etc. These methods are easily affected by light, haze, rain and snow, etc. It is difficult to stably obtain the real ice coating state of the conductor surface, and most of them can only realize rough judgment of the presence or absence of ice coating or the shape of the ice layer, and it is difficult to realize accurate thickness quantitative estimation. The present application uses ultrasonic technology to emit excitation signals from the side of the power transmission conductor and receive echoes, analyzes the propagation characteristics of the echo signals in the conductor-ice-air multi-medium structure, and can reflect the change of ice thickness with high resolution, has stronger physical penetration and environmental adaptability. Further, the present application proposes a scheme combining short-time Fourier transform and Mel spectrum extraction acoustic features, which not only retains the time-frequency information of the echo signal, but also considers the compactness and discriminability of the perceptual features. After inputting into the deep neural network for modeling, it can realize accurate mapping and prediction of different thickness ice states. The overall system of the present application adopts a remote deployment architecture, and after the original echo signal collection is completed, it is transmitted to the remote processing terminal for algorithm reasoning calculation through wireless or wired mode, avoiding the influence of the on-site environment on the model running, and improving the integration and expansibility of the system.

[0018] In summary, the application can be widely applied to the ice prevention and disaster reduction, intelligent inspection and ice risk management of power lines, and has good engineering adaptability and popularization prospect. BRIEF DESCRIPTION OF DRAWINGS

[0019] The accompanying drawings are included to provide a further understanding of the application and are incorporated in and constitute a part of this specification, illustrate embodiments of the application and are used to explain the application, but do not constitute a limitation on the application.

[0020] Figure 1 is a flowchart of the ice thickness dynamic estimation method provided by an embodiment of the application; Figure 2 is a schematic diagram of the erection of the ultrasonic excitation and reception module provided by an embodiment of the application. DETAILED DESCRIPTION

[0021] In order to make the purpose, technical scheme and advantages of the embodiments of the application clearer, the technical scheme in the embodiments of the application will be described clearly and completely below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the application.

[0022] The propagation speed and attenuation characteristics of ultrasonic waves in a medium are closely related to the physical parameters of the medium, such as density, elastic modulus and acoustic impedance. When an ice layer is formed on the surface of a power transmission line, the external medium of the power transmission line changes from air to a double-layer structure of air and ice. This layered structure will significantly change the reflection mode and energy loss characteristics of ultrasonic waves on the surface of the power transmission line. Specifically, there is a significant difference in acoustic impedance between the ice layer and the power transmission line, so that part of the ultrasonic waves is reflected at the interface between the ice and the power transmission line, and another part is transmitted in the ice layer and propagates along the power transmission line. Due to the different thicknesses of the ice layer, the propagation distance and round-trip time of ultrasonic waves in the ice layer will change accordingly, and this change will leave measurable characteristics in the time delay, phase, amplitude attenuation and spectral distribution of the received echo. In addition, the existence of the ice layer will introduce additional multiple reflection and scattering effects, resulting in specific interference fringes and harmonic changes in the spectral structure of the echo signal.

[0023] Embodiment one The embodiment provides a power transmission line ice thickness dynamic estimation method, as shown in Figure 1 The method mainly includes the following steps: I. Ultrasonic excitation and echo acquisition In the actual operating environment of the power transmission line, in order to realize real-time perception and thickness estimation of the ice layer on the conductor surface, a set of ultrasonic sensing device with excitation and receiving functions needs to be set on the side of the conductor. The device mainly includes an ultrasonic exciter, an ultrasonic receiver, a fixed support, a power supply and a control module, and is packaged in a shell with strong weather resistance and insulation performance in an integrated structure, so as to be suitable for stable operation in complex environments such as high cold, high humidity, wind and snow.

[0024] The ultrasonic exciter selects a piezoelectric ceramic transducer (such as PZT-based material) or a MEMS ultrasonic transducer, and the frequency range should meet the sufficient resolution capability for the change of millimeter-level or even sub-millimeter-level ice thickness. Usually, a transducer with a center frequency in the range of 300 kHz to 800 kHz is selected, and the excitation signal waveform adopts a frequency-modulated pulse or a Gaussian window envelope sine pulse wave to improve the time domain resolution and noise immunity in the echo. The excitation period can be set to emit once every 10 to 60 minutes, and the specific period can be adjusted according to the environmental temperature and the development trend of the ice layer.

[0025] After the ultrasonic signal is emitted from the exciter, under normal circumstances, it propagates to the conductor through the air medium, and the reflected signal mainly comes from the outer surface of the conductor. When there is an ice layer on the surface of the conductor, the ultrasonic signal propagation path becomes transducer -> air -> ice layer -> conductor, and reflection caused by acoustic impedance mismatch occurs at the air-ice layer interface and the ice layer-conductor interface. Due to the significant difference in acoustic properties (such as acoustic impedance, density, and wave speed) between the ice layer and the conductor, the two interfaces will form two reflection peaks with distinguishable time delays in the echo signal.

[0026] The ultrasonic receiver should have high sensitivity and high time resolution, and the sampling rate should not be less than 5 MHz. The receiver is usually set on the same side as the exciter to form a pulse-echo mode, which can effectively avoid the wiring complexity and phase error accumulation problems caused by setting a second receiving node remotely.

[0027] II. Signal preprocessing and feature extraction After completing ultrasonic excitation and echo signal collection, the original signal obtained usually has the following problems: first, the signal is mixed with multiple source environmental noise (such as wind vibration, electromagnetic interference, scattering clutter of ice and snow attachments, etc.); second, amplitude imbalance and phase drift caused by conductor geometric structure differences, excitation power fluctuations and other factors; third, signal overlap caused by multi-path reflection and interface ambiguity. Therefore, before entering the feature extraction and modeling link, the original echo signal must be preprocessed to ensure that the key features related to the ice thickness can be accurately and efficiently identified subsequently.

[0028] Firstly, the original signal is band-pass filtered. The upper and lower frequency bands of the filter are set according to the center frequency of the excitation frequency to filter out stray signals far below or above the excitation frequency. For example, when the excitation signal frequency is set to 500 kHz, the band-pass filter frequency band can be set to 300 kHz to 700 kHz to cover the main energy band and appropriately retain the high harmonic information. The filtering process is implemented using FIR or IIR filter structures, and the filter design needs to consider the passband flatness, phase response, and real-time processing efficiency.

[0029] Next, to eliminate systematic deviations caused by transducer performance fluctuations, contact pressure changes, and other factors, the filtered signal is normalized and amplitude standardized to map it to the [-1, 1] range. Or based on energy normalization standards, the signal amplitudes at different sampling times or device states are uniformly adjusted to a consistent scale.

[0030] Then, echo envelope extraction and main reflection segment cropping are performed. The signal envelope is constructed by Hilbert transform, and the main reflection energy segment is extracted. The signal segment before and after the first significant peak is intercepted to avoid introducing invalid tail waves or early interference. At this time, a clear structure echo time-domain waveform segment is obtained, which has good signal-to-noise ratio and time delay characteristics.

[0031] In the feature extraction stage, the time-domain signal obtained above is first framed and windowed. The entire signal is divided into small segments (i.e. short-time signals) according to a certain frame length and frame shift. After framing, each frame of signal is multiplied by a window function (such as Hamming window, Hannah window or Kaiser window) to reduce boundary effects.

[0032] Each frame is subjected to a fast Fourier transform (FFT) to obtain the frequency domain power spectrum of each frame: , wherein, is the complex spectrum of each frame of signal, is its amplitude spectrum, and the square is the power spectrum.

[0033] After completing the short-time Fourier transform (STFT) and obtaining the power spectrum of each frame, the next step is to map this spectral information to the Mel scale that simulates human auditory characteristics. The non-linear characteristics of the Mel scale can more sensitively reflect subtle changes in the frequency spectrum, which is particularly important in determining whether the power line is iced and the thickness of the ice layer, because icing changes the propagation, reflection, and attenuation characteristics of ultrasonic waves by the conductor, which is then reflected in the energy changes of the frequency spectrum distribution.

[0034] Mel spectrum mapping mainly includes the following steps: In the desired frequency range, it is uniformly divided into M frequency bands (usually 20~40) according to the Mel scale, and a triangular band-pass filter is designed for each frequency band, each filter reaches a maximum value of 1 at the center frequency and gradually decays to 0 towards adjacent frequency bands, constructing an overlapping filter bank.

[0035] Multiply the power spectrum of each frame with the Mel filter bank and accumulate to get the power in each Mel frequency band:

[0036] where, Pm represents the power in the mth Mel frequency band, Pm(f) is the power of the mth FFT frequency point, k Hm(f) is the response of the mth Mel filter at frequency point f, and k are the FFT indices corresponding to the m-1th and m+1th Mel frequencies, respectively. Spitch(m) is the power value of each Mel frequency band, and a vector of length M is obtained , which represents the energy distribution of the current frame on the Mel scale, i.e., the Mel power spectrum.

[0037] In order to compress the dynamic range and enhance the expression ability of small energy changes, the power value of each Mel frequency band is taken as the logarithm: where

[0038] is a small positive number to prevent meaningless operations when taking the logarithm. In some specific environments, in order to enhance the response ability to icing state changes, high-order spectral features such as cepstrum delay spectrum (CDP), wavelet packet entropy (WPE), spectral centroid (Spectral Centroid) and spectral skewness (Spectral Skewness) can be further introduced. These features can describe the changes in frequency distribution patterns of signals and have significant sensitivity to early icing formation.

[0039] So far, this step has completed the preprocessing and feature extraction of the echo signal, and through short-time Fourier transform (STFT), a number of Mel power spectra representing the echo signal are obtained, as well as other high-order spectral features. These features implicitly contain the physical mapping rules of ice layer existence and thickness changes in sound wave propagation.

[0040]

[0041] III. Deep feature learning and thickness modeling ​​After the Mel power spectrum and other features are obtained, the ultrasonic echo signal of each frame is compressed into a set of feature vectors with high correlation but low dimension. These features reflect the complex reflection, attenuation, and propagation path changes between the ultrasonic wave and the iced conductor, and have high distinguishability. The next core task is to use a deep neural network structure to learn these features, thereby establishing a mapping relationship model between the echo features and the actual ice thickness.

[0042] Before building the model, the extracted Mel spectrum and other features need to be further processed to adapt to the input of the deep neural network. In this embodiment, the continuous multi-frame Mel power spectrum vectors are directly spliced to form a two-dimensional time-frequency graph , T represents the number of frames, M is the number of Mel bands (i.e., the number of Mel filters), and such a matrix can be regarded as an image or a time series input.

[0043] To simultaneously learn the time-frequency spatial structure and the time series evolution characteristics, the neural network model designed in this embodiment includes: Time-frequency convolution: a two-dimensional convolution network is used to extract local features from the input time-frequency image; BiLSTM time series modeling: the convolution output is flattened into a time series input BiLSTM network; Fully connected regression: finally, one or more fully connected layers are used to output the ice thickness prediction value.

[0044] Specifically, in this embodiment, the two-dimensional time-frequency graph obtained by splicing is regarded as an image, and a two-layer convolutional neural network (CNN) is used to extract local features. The first layer of CNN uses 32 3×3 convolution kernels, and the second layer uses 64 3×3 convolution kernels. After each layer of convolution, a ReLU activation and a 2×2 maximum pooling operation are performed, and a local representation with frequency band coupling information is extracted.

[0045] Subsequently, the convolution output is flattened in the time dimension into a time series vector and input into a BiLSTM network containing 128 hidden units. This structure can model the forward and reverse trends of the time series in parallel, and better capture the feature patterns of the ultrasonic echo evolving over time under the ice state, such as energy drift, time delay increase, and spectral broadening of the feature frequency band over time.

[0046] After the BiLSTM output is flattened, it enters a fully connected regression network and finally outputs a continuous numerical value representing the predicted ice thickness (in millimeters) of the current conductor segment. The entire network is trained with the mean squared error (MSE) as the objective function, and iteratively optimized in the data set to approximate the true ice thickness label. At the same time, by setting Dropout = 0.3 to suppress overfitting, and using early stopping mechanism between the training set and the validation set to avoid overfitting noise.

[0047] In order to construct a representative and high-precision training data set to support the modeling of the mapping relationship between ultrasonic echo signals and transmission line ice thickness, the present application mainly relies on laboratory measurement and field engineering deployment to obtain data. In the laboratory stage, first, a test platform containing standard transmission conductor segments is built, and different thickness of ice is simulated on the conductor surface by a temperature and humidity control system or a low-temperature spray ice device. To ensure the thickness is controllable, high-precision measurement equipment such as laser displacement sensor and contact thickness gauge is used to measure the ice thickness point by point to form high-credibility label data. The ultrasonic exciter and echo receiver are fixed on the side wall or clamping bracket of the conductor, respectively, the excitation signal is triggered by a function signal generator, and a high-speed signal acquisition card is used to record the echo time-domain signal synchronously, ensuring that the acquisition process has high time resolution and high signal-to-noise ratio. Repeat the collection of sufficient signal samples under different ice thickness conditions, and recalibrate the thickness after each change of ice state, and finally form a structured data set, each sample of which contains the original echo signal and its corresponding true ice thickness.

[0048] In the field engineering deployment stage, a section of actual running high-voltage transmission line is selected for system deployment, and the ultrasonic sensing components are installed on the conductor side wall or insulator bracket to realize automatic acquisition of echo signals in long-time continuous running state. To obtain the corresponding thickness label information, an unmanned aerial vehicle equipped with an infrared thermal imaging device can be used to scan the conductor ice profile, and an image recognition algorithm can be used to estimate the ice layer size, or a ground laser rangefinder can be used to measure the conductor diameter change data to calculate the ice layer thickness. In some accessible areas, manual tower climbing combined with thickness measuring tools can also be used for on-site calibration to further improve the data labeling accuracy. The echo signals collected in the field not only cover various weather conditions, terrain environments and running voltage backgrounds, but also introduce complex factors such as electromagnetic interference, wind vibration effect and uneven ice coverage in real working conditions, thereby enhancing the generalization ability and anti-interference robustness of the neural network model in actual application. These high-quality data sets obtained through measurement provide a solid foundation for model training and optimization.

[0049] Four, dynamic estimation and result correction In the practical application stage, first, the ultrasonic echo signal is acquired, and short-time Fourier transform (STFT) and Mel feature extraction are completed based on preset parameters (such as frame length, frame shift, window function, etc.), so that the ultrasonic echo time-frequency feature at the current moment is converted into a two-dimensional feature matrix consistent with the format of the training sample. The matrix is then input into a deep learning model, such as the CNN-BiLSTM structure network designed as described above. After feature extraction and time series modeling by multiple convolutional layers and bidirectional long short-term memory layers, the model outputs a real value representing the estimated ice thickness value of the current conductor segment.

[0050] However, since the neural network may produce unstable predictions when facing extreme noise interference or boundary thickness (such as 0 mm or > 30 mm), in some embodiments, a multiple result correction mechanism is also provided. First, a sliding window filter is introduced in the time dimension to smooth the prediction values for multiple consecutive times, avoiding prediction jumps caused by transient noise; second, a set of physical constraint rule library is established, such as the logic rule that the ice thickness should not continuously increase when the temperature is positive, and these constraint conditions are embedded in the post-processing process of the prediction result to eliminate or correct abnormal predictions that obviously violate physical common sense.

[0051] In addition, to improve adaptability in different regions and different weather conditions, in some embodiments, a dynamic drift correction function is also supported. Some manually measured correction points (such as ice layer thickness obtained by laser ranging or unmanned aerial vehicle image recognition) can be collected on site, compared with the current model prediction result, the deviation is calculated and the model output is dynamically adjusted, for example, using a sliding mean offset or weight scaling correction strategy, so that the prediction value gradually approaches the true ice layer thickness level. For long-term deployment scenarios, an online fine-tuning mechanism can also be introduced to continuously optimize model parameters using new labeled samples to ensure the accuracy stability in continuous operation.

[0052] Embodiment Two Based on the above method, the embodiment provides a power transmission line ice thickness dynamic estimation system, which aims to monitor the change of the ice thickness on the surface of the overhead power transmission conductor in real time. The system has compact structure, moderate algorithm complexity, high robustness, high sensitivity and strong engineering adaptability, and can effectively support online ice risk warning and power grid operation management of large-scale power transmission lines. The system as a whole includes the following functional modules: remote control terminal, ultrasonic excitation and reception module, data acquisition and caching module, communication transmission module, remote processing terminal, result feedback and display module. The modules are connected through wireless or optical fiber network to form a clear structure, efficient distributed collaborative system.

[0053] The ultrasonic excitation and receiving module is installed at the side of the power transmission line, including a piezoelectric ultrasonic excitation device and a receiving sensor, which is used to convert the excitation control command from the remote control terminal into a physical ultrasonic pulse signal and receive the ultrasonic echo signal in real time. The module is provided with a microcontroller unit (MCU) and a driving circuit, which controls the frequency, pulse width and repetition interval of the ultrasonic excitation through a high-precision timer, so as to ensure the controllability and stability of the signal.

[0054] As shown in Figure 2 , the ultrasonic excitation and receiving module can be arranged on adjacent power transmission lines, and can also be installed in various structures with electrical safety. For example, the module can be integrated and installed below the insulator string or arranged on the non-metallic arm of the hardware such as the tension clamp and the suspension clamp, and the signal excitation and receiving can be performed close to the side of the conductor through the suspension mode, which not only ensures electrical isolation but also facilitates maintenance. In addition, the module can be miniaturized and embedded in the existing shock absorber or metal damper, and the signal coupling can be completed by using the original structure, so as to realize function reuse, reduce structural complexity and additional mass. At the head of the tension tower or the cross arm, an elastic coupling arm or a waveguide coupling plate can be arranged, and the ultrasonic excitation and receiving device can be arranged at the end thereof, so as to direct the ultrasonic wave into the conductor surface in a non-contact manner, avoid direct contact with the conductor in the high potential area, and enhance the adaptability in multi-circuit or high-voltage systems. These installation modes can be flexibly selected according to the actual application scene, so as to effectively improve the flexibility of system deployment and the stability of signal detection.

[0055] The data acquisition and caching module includes a high sampling rate ADC, an anti-aliasing filter, a low-noise amplifier and a caching chip, which is responsible for high-speed sampling and converting the analog echo signal into a digital signal. The sampling frequency is usually set to 200-500 kHz to ensure the time detail resolution of weak echoes. After sampling is completed, the module packages each signal in a structured format (including timestamp, echo intensity, environmental parameters, etc.), temporarily caches and prepares to send.

[0056] The communication transmission module is a data bridge between the remote acquisition end and the back-end processing platform. According to the deployment environment, the module can adopt 4G / 5G cellular communication, LoRa long-distance communication, satellite relay or power optical fiber channel, and has a multi-channel redundant backup mechanism to ensure real-time and safe transmission of data in long-distance and high-interference environments. The communication module supports MQTT or custom TCP / IP protocol, has a certain degree of data compression and encryption capability, and can realize remote upgrading and device state monitoring.

[0057] The remote processing terminal can be a centralized cloud server, a dispatch center AI platform, or a local embedded inference server, which internally deploys a complete feature extraction module and a deep neural network model. The processing terminal first performs STFT processing and Mel feature extraction on the received original signal to construct a two-dimensional time-frequency graph, and then inputs it into the CNN-BiLSTM neural network model to complete the dynamic estimation of ice thickness.

[0058] The result feedback and display module is used for multi-channel output of the estimation results. On the one hand, the system can push the results to the power grid dispatch center interface, mobile terminal APP, or ground patrol workstation in real time, directly presenting the ice thickness distribution of each section of the conductor; on the other hand, the system can also be linked with the early warning module, and when the estimated thickness exceeds the set threshold, it triggers various means such as SMS, email, sound and light alarm to prompt the risk.

[0059] The remote control terminal is responsible for issuing control instructions to the on-site devices (ultrasonic excitation and reception module), such as excitation parameter adjustment, communication frequency change, fault diagnosis command, etc., to realize remote controllable and adaptive management of the system.

[0060] The workflow of the entire system is as follows: the remote control terminal remotely sends excitation control instructions to the edge node according to the set strategy or manual intervention, triggering one or more ultrasonic excitation-echo acquisition-data caching processes; then the acquisition module returns the digitized signal to the remote processing terminal through the communication module; the processing terminal completes the whole process of echo feature extraction, model inference, etc.; finally, the estimation results are output and presented through the graphical interface or trigger the alarm linkage. The modules are asynchronously coordinated by timestamp marking and queue scheduling mechanism, ensuring that the system has strong real-time performance and large-scale deployment capability.

[0061] To avoid false judgments of the model under extreme interference, the system is also configured with a dynamic correction and output module, which introduces physical rules such as moving average filtering, change rate constraint, and environmental temperature and humidity coordination restriction to perform boundary judgment and dynamic calibration on the model output results in real time. At the same time, this module also supports interface interaction with external unmanned aerial systems, infrared imaging devices, or laser range finders, fuses information from other sensing sources, compares and verifies the estimation results, and performs micro-deviation correction to ensure the absolute accuracy of thickness estimation.

[0062] In some embodiments, in order to ensure that the ultrasonic excitation and receiving module can accurately align the power transmission line for signal excitation and echo acquisition, an angle adjustment module and a visual positioning component are introduced in the structural design thereof. The visual positioning component includes an industrial camera, an image processing unit and a feature recognition algorithm, which is used to obtain image information of the space in front of the module in real time, and accurately identify the spatial orientation information of the power transmission line through image segmentation, edge detection and target recognition algorithm. Based on the identified center line or surface contour line of the power transmission line, the relative deviation angle between the line and the module is calculated. Subsequently, the angle adjustment module automatically drives the integrated micro servo motor or piezoelectric driving mechanism according to the deviation angle input, drives the ultrasonic excitation and receiving head to pitch or horizontally rotate around the center axis of the installation support, until the transmission axis of the module is aligned with the direction of the power transmission line, realizing efficient directional excitation and echo acquisition. The whole angle adjustment process is executed in real time through closed-loop control logic, and has the ability of self-adaptive line searching, self-alignment and fault tolerance.

[0063] The design not only improves the installation flexibility and detection accuracy of the module under complex field conditions, but also significantly enhances the automatic correction capability of the system when the direction deviates due to line swing, wind disturbance or component displacement during long-time operation, ensuring stable echo signal quality and strong repeatability, and providing a reliable signal basis for subsequent echo feature extraction and ice thickness prediction.

[0064] The above system can realize the power transmission line ice thickness dynamic estimation method described in embodiment one, has the corresponding function modules and beneficial effects of the method, and the technical details not described in detail in this embodiment can be referred to the power transmission line ice thickness dynamic estimation method provided in embodiment one of the application.

[0065] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be realized by means of software plus a general hardware platform, and of course, it can also be realized by hardware. Based on such understanding, the above technical solutions or the part that contributes to the related art can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes a plurality of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the method described in each embodiment or some part of the embodiment.

[0066] Finally, it should be noted that: the above examples are only used to illustrate the technical solutions of the present application, but not limited to them; under the idea of the present application, the technical features of the above examples or different examples can also be combined, the steps can be implemented in any order, and there are many other changes of different aspects of the present application as described above, which are not provided in details for simplicity; although the present application has been described in detail with reference to the foregoing examples, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing examples, or make equivalent replacement to part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for dynamically estimating the ice thickness on a power transmission line, characterized in that, The method comprises the steps of: S1, transmitting an ultrasonic excitation signal to the power transmission line and collecting a corresponding ultrasonic echo signal; S2, preprocessing and short-time Fourier transform of the echo signal to obtain a frequency domain power spectrum; S3, Mel spectrum mapping of the frequency domain power spectrum to construct a two-dimensional time-frequency graph; S4, inputting the two-dimensional time-frequency graph into a neural network to obtain an ice thickness prediction value; S5, dynamic correction of the prediction value to output an ice thickness estimation value.

2. The method for dynamically estimating the ice thickness on a power transmission line according to claim 1, wherein, The preprocessing in step S2 includes: setting the frequency band range of the band-pass filter according to the excitation frequency of the ultrasonic excitation signal, band-pass filtering the collected echo signal, and normalizing and amplitude standardizing the filtered signal; then performing echo envelope extraction and main reflection segment cropping: constructing a signal envelope line through Hilbert transform, extracting the main reflection energy segment, and intercepting the signal segment before and after the first significant peak appears.

3. The method for dynamic estimation of ice thickness on power transmission lines according to claim 2, characterized in that, The intercepted signal segment is framed and windowed: the whole signal is divided into several short-time signals according to the set frame length and frame shift, and each short-time signal is multiplied by a window function; then through fast Fourier transform, the frequency domain power spectrum of each frame signal is obtained.

4. The method for dynamically estimating the ice thickness on a power transmission line according to claim 3, wherein, In step S3: In the desired frequency range, M frequency bands are uniformly divided according to the Mel scale, and a triangular band-pass filter is designed for each frequency band, each filter reaches a maximum value of 1 at the center frequency and gradually decays to 0 towards adjacent frequency bands, constructing an overlapping filter bank; Multiply the frequency domain power spectrum of each frame signal by the Mel filter bank and accumulate to get the power in each Mel band: wherein, Pmrepresents the power on the mthMel band, is the power of the mthFFT bin, k is the power of the mthFFT bin, is the response of the mthMel filter at the bin k is the response of the mthMel filter at the bin and are the FFT indices corresponding to the m-1thand m+1thMel frequencies, respectively. The power values on each Mel band are concatenated to obtain a Mel power spectrum vector M of length , which is used to represent the energy distribution of the current frame on the Mel scale. The Mel power spectrum vectors corresponding to each frame signal are spliced into a two-dimensional time-frequency graph in time sequence.

5. The method of dynamic estimation of ice thickness on power transmission lines of claim 1, wherein, The neural network in step S4 includes a time-frequency convolution module, a BiLSTM unit and a fully connected regression module connected in turn; The time-frequency convolution module regards the two-dimensional time-frequency graph as an image and uses multiple convolution layers to extract the local deep features of the two-dimensional time-frequency graph; the BiLSTM unit is used to input the feature vector output by the time-frequency convolution module into the BiLSTM network to extract the time sequence evolution features; the fully connected regression module is used to input the feature vector output by the BiLSTM unit into the fully connected neural network to obtain the prediction value of the ice thickness through multiple fully connected layers.

6. The method of dynamic estimation of ice thickness on power transmission lines of claim 1, wherein, The dynamic correction in step S5 includes: introducing a sliding window filter in the time dimension to smooth the prediction values of multiple continuous times; by calculating the deviation between the prediction value and the measured value, using sliding mean offset or weight scaling correction strategy to make the prediction value approach the true ice thickness; a constraint rule library is established to eliminate abnormal prediction values that violate physical common sense.

7. A system for dynamic estimation of ice thickness on power lines based on the method according to any one of claims 1 to 6, characterized in that, It comprises an ultrasonic excitation and reception module, a data acquisition and cache module, a communication transmission module, a remote processing terminal, a result feedback and display module, and a remote control terminal for issuing control instructions to the ultrasonic excitation and reception module, and data links are established between the modules through wireless or optical fiber networks; The ultrasonic excitation and reception module is used to convert the excitation control instructions sent by the remote control terminal into physical ultrasonic pulse signals transmitted to the power transmission line, and to receive ultrasonic echo signals in real time; The data acquisition and caching module is configured to sample the echo signal at a high speed and convert it into a digital signal, and to package the digital signal and return it to a remote processing terminal through a communication module after sampling is completed; The remote processing terminal is provided with a complete feature extraction module and a neural network model, and is configured to output a corresponding ice thickness estimation value according to an input signal; The result feedback and display module is configured to push the estimation result to a power grid dispatching center interface, a mobile terminal APP or a ground patrol workstation in real time, and is also configured to link an early warning module, and trigger an alarm prompt when the estimated thickness exceeds a set threshold.

8. The transmission line ice thickness dynamic estimation system of claim 7, wherein, The system is also provided with a dynamic correction and output module, which is configured to perform boundary judgment and dynamic calibration on the model output result in real time; at the same time, the module also supports interface interaction with an external unmanned aerial vehicle system, an infrared imaging device or a laser range finder, fuses external sensing information, compares and verifies the estimation result, and performs micro-deviation correction.

9. The transmission line ice thickness dynamic estimation system of claim 7, wherein, The ultrasonic excitation and receiving module is installed on the side of the power transmission line, and includes an ultrasonic exciter, an ultrasonic receiver, a fixing support, a power supply and a control module; the ultrasonic exciter selects a piezoelectric ceramic transducer or a MEMS ultrasonic transducer, and the excitation signal waveform adopts a frequency-modulated pulse or a sine pulse wave with a Gaussian window envelope.

10. The transmission line ice accretion thickness dynamic estimation system of claim 7, wherein, The ultrasonic excitation and receiving module is also provided with an angle adjusting assembly and a visual positioning assembly; the visual positioning assembly includes a camera and an image processing unit, and is configured to identify the spatial orientation information of the power transmission line in real time through image processing technology, and calculate the relative deviation angle between the module and the power transmission line; the angle adjusting assembly automatically drives an integrated micro servo motor or a piezoelectric driving mechanism according to the deviation angle, drives the ultrasonic excitation and receiving module to perform pitch or horizontal rotation adjustment around the center axis of the installation support, until the transmission axis of the module is aligned with the power transmission line, so as to realize directional excitation and echo collection.

Images