High-voltage line interference signal feature analysis method and system for sag monitoring

Abstract

This application discloses a method and system for analyzing interference signal characteristics of high-voltage lines for sag monitoring, which relates to the field of transmission line sag monitoring. The method includes: deploying a polarized antenna assembly containing horizontally and vertically polarized antennas below the conductor to collect electromagnetic interference signals in real time, extracting IMF components through bandpass filtering and EMD decomposition, and performing sparse coding based on a physical dictionary constructed based on sag and temperature to extract structural state characteristics; simultaneously using RTK positioning information to differentially calculate geometric sag, and combining ambient temperature and wind speed information to construct EMI and GNSS dual-branch features; finally, adaptively determining fusion weights based on signal-to-noise ratio and positioning reliability, generating multi-source fusion features, and inputting them into a sag prediction model. This method achieves accurate monitoring of the entire span and continuous dynamics of high-voltage lines, capable of outputting high-precision sag estimates.

Description

Technical Field

[0001] The present application relates to the technical field of transmission line sag monitoring, and in particular to a method and system for analyzing characteristics of high-voltage line interference signals for sag monitoring. Background Art

[0002] The sag of a high-voltage transmission line refers to the static downward curve formed by the conductor's weight, tension, and environmental factors. Its magnitude directly affects the safe spanning distance and operational stability of the line. Accurately understanding changes in conductor sag helps promptly identify spanning risks caused by loose conductors, abnormal tension, or sudden environmental changes, thereby preventing safety incidents such as collisions and oscillation instability.

[0003] Currently, optical scanning and laser ranging can directly measure the spatial sag height of the conductor with high accuracy. However, they are limited by severe weather conditions such as rain, fog, and snow, as well as interference from multipath reflections, resulting in poor measurement stability. At the same time, the equipment usually needs to be deployed on tracks, high towers, or aerial platforms, which makes deployment complex, construction difficult, and maintenance costly, making it unsuitable for large-scale deployment.

[0004] In the inverse method based on tensiometers and accelerometers, the sag is indirectly estimated using the tension-deformation model by measuring the conductor tension or vibration information. However, there are problems with complex modeling and strong coupling relationships, and the vibration signal is easily affected by external disturbances such as wind and bird strikes, resulting in poor accuracy and robustness of the estimated results, and limited practical application effects.

[0005] In sag monitoring based on the GNSS (Global Navigation Satellite System), high-precision positioning systems using RTK (Real-Time Kinematic) differential positioning technology can achieve centimeter-level accuracy, making them suitable for sag monitoring at conductor installation points. However, because this method can only locate discrete points where antenna equipment is installed, it is difficult to continuously describe the sag shape of the entire conductor. Furthermore, due to the limited number of equipment deployed, it is difficult to meet the requirements for comprehensive sensing of multi-point, high-frequency, and dynamic changes. Summary of the Invention

[0006] The present application provides a method, system, storage medium, computer program product and electronic device for analyzing the characteristics of high-voltage line interference signals for sag monitoring, which are used to at least solve the problems of large environmental interference, high deployment cost, incomplete data coverage or unstable accuracy in the current related sag monitoring technology.

[0007] In the first aspect, the embodiment of the present application provides a method for analyzing the characteristics of high-voltage line interference signals for sag monitoring, comprising: equidistantly arranged polarized antenna assemblies between two adjacent towers along the bottom of the high-voltage conductor to collect electromagnetic interference signals; the polarized antenna assemblies include horizontally polarized antennas and vertically polarized antennas; band-pass filtering is performed on the collected electromagnetic interference signals to obtain preprocessed signals, and empirical mode decomposition is performed on the preprocessed signals to extract M IMF components; M Is an integer greater than 1; based on the physical dictionary constructed by combining the state of the preset sag range and the preset temperature range, the M The invention relates to a method for obtaining a sparse coefficient vector by sparsely encoding the IMF components; calculating the three-dimensional coordinates by differential solution output by an RTK receiver suspended on a high-voltage conductor; calculating the difference between the lowest point of conductor sag and the height of the tower; collecting conductor temperature information and ambient wind speed information, encoding the conductor temperature information and the sparse coefficient vector to generate a corresponding EMI branch feature coding vector, encoding the ambient wind speed information and the geometric sag observation value to generate a corresponding GNSS branch feature coding vector, and performing weighted fusion on the EMI branch feature coding vector and the GNSS branch feature coding vector according to a fusion weight to obtain a multi-source fusion feature; the fusion weight is determined according to the EMI signal-to-noise ratio and the GNSS positioning reliability; and inputting the multi-source fusion feature into a sag prediction model to output a corresponding sag estimation value.

[0008] In the second aspect, the embodiment of the present application provides a high-voltage line interference signal feature analysis system for sag monitoring, including: a polarized EMI signal acquisition unit, which is used to arrange polarized antenna components equidistantly between two adjacent towers along the bottom of the high-voltage conductor to collect electromagnetic interference signals; the polarized antenna components include horizontally polarized antennas and vertically polarized antennas; an IMF component extraction unit, which is used to perform bandpass filtering on the collected electromagnetic interference signals to obtain preprocessed signals, and perform empirical mode decomposition on the preprocessed signals to extract M IMF components; M is an integer greater than 1; a sparse coding unit for encoding the physical dictionary constructed based on a combination of a preset sag range and a preset temperature range. MThe IMF components are sparsely coded to obtain a sparse coefficient vector; a geometric sag measurement unit is used to use the differential decomposition output by the RTK receiver suspended on the high-voltage wire to calculate the three-dimensional coordinates, calculate the difference between the lowest point of the wire sag and the tower height, to obtain a geometric sag observation value; a multi-source coding fusion unit is used to collect wire temperature information and ambient wind speed information, encode the wire temperature information and the sparse coefficient vector to generate a corresponding EMI branch feature coding vector, and encode the ambient wind speed information and the geometric sag observation value to generate a corresponding GNSS branch feature coding vector, and perform weighted fusion on the EMI branch feature coding vector and the GNSS branch feature coding vector according to a fusion weight to obtain a multi-source fusion feature; the fusion weight is determined according to the EMI signal noise ratio and the GNSS positioning reliability; a sag prediction unit is used to input the multi-source fusion feature into a sag prediction model to output a corresponding sag estimation value.

[0009] In a third aspect, an electronic device is provided, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can perform the steps of the high-voltage line interference signal characteristic analysis method for sag monitoring according to any embodiment of the present application.

[0010] In a fourth aspect, an embodiment of the present application provides a storage medium on which a computer program is stored, characterized in that when the program is executed by a processor, the steps of the high-voltage line interference signal characteristic analysis method for sag monitoring of any embodiment of the present application are implemented.

[0011] In a fifth aspect, an embodiment of the present application provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the high-voltage line interference signal characteristic analysis method for sag monitoring of any embodiment of the present application.

[0012] The present application provides a method and system for analyzing high-voltage line interference signal characteristics for sag monitoring, which can produce at least the following technical effects:

[0013] (1) Using polarized antenna components equidistantly arranged below the high-voltage conductor, the electromagnetic interference signal caused by the conductor's operating status is actively collected and decomposed into signal components in the horizontal and vertical polarization directions. The equidistant antenna arrangement combines the physical characteristics of the continuous distribution of electromagnetic signals around the conductor to achieve a continuous perception of the conductor's sag shape. Compared with point-based monitoring methods, it has higher spatial coverage capabilities, provides a complete signal input source for subsequent signal modeling and sag estimation, and improves the system's spatial resolution of the conductor's status.

[0014] (2) By performing bandpass filtering and empirical mode decomposition (EMD) on the collected electromagnetic interference signal, the complex mixed non-stationary signal is converted into several physically meaningful intrinsic mode functions (IMFs), which constitute the typical disturbance characteristics of the conductor under the operating state and, to a certain extent, contain the response relationship between the conductor sag and external environmental factors. By sparsely encoding the IMF components based on a physical dictionary constructed based on a combination of preset sag and temperature states, a sparse mapping of complex signals to structural parameters is achieved, representative state feature codes are effectively extracted, and the ability to identify subtle changes in conductor sag is enhanced.

[0015] (3) In view of the respective advantages and limitations of the EMI branch and the GNSS branch, the conductor temperature and ambient wind speed are collected simultaneously, and the impact of environmental changes on electromagnetic characteristics and geometric observations is made explicit through encoding. The fusion weight is dynamically adjusted based on the EMI signal noise ratio and RTK positioning reliability, achieving a complementary synergy between EMI continuity and GNSS absolute accuracy. In addition, the adaptive fusion strategy can balance the contributions of the two types of information in real time according to scene changes. When one signal source is unstable, the fusion strategy can automatically increase the weight of the other branch, thereby significantly improving the estimation accuracy and anti-interference capability under complex meteorological and external interference conditions.

[0016] (4) The fused multi-source features are input into the pre-trained sag prediction model, which can achieve fine-grained response across sections with the help of dense EMI driving data while retaining the RTK observation benchmark. The sag prediction model can capture the slight deformation of the conductor caused by temperature expansion and contraction, tension relaxation or wind load changes, and output high-precision sag estimation values, realizing accurate monitoring of the entire span, multiple points and continuous dynamics of the high-voltage line.

[0017] Through this technical solution, multi-scale feature extraction and physical dictionary sparse coding based on polarized EMI networks are used to achieve full-span, high-frequency wire vibration deformation perception; the multi-source adaptive fusion strategy driven by environmental perception parameters enhances the accuracy and robustness of the estimation. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0019] Figure 1 A flowchart illustrating an example of a method for analyzing characteristics of high-voltage line interference signals for sag monitoring according to an embodiment of the present application is shown;

[0020] Figure 2 A schematic diagram of the system architecture of an example of a sag monitoring platform according to an embodiment of the present application is shown;

[0021] Figure 3 An operational flow chart illustrating an example of constructing a physical dictionary according to an embodiment of the present application is shown;

[0022] Figure 4 An operational flow chart of an example of sparse coding solution according to an embodiment of the present application is shown;

[0023] Figure 5 An operational flow chart illustrating an example of multi-source feature weighted fusion processing according to an embodiment of the present application is shown;

[0024] Figure 6 A schematic diagram showing an example of a comparison effect of a sag monitoring result based on EMI and GNSS fusion according to an embodiment of the present application and a sag monitoring result based on GNSS;

[0025] Figure 7 A structural block diagram of an example of a high-voltage line interference signal characteristic analysis system for sag monitoring according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0026] To make the purpose, technical solutions, and advantages of the embodiments of this application more clear, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0027] It should be noted that the current mainstream sag monitoring technologies mainly include traditional geometric measurement methods and sag monitoring methods based on RTK positioning technology of GNSS signals, but they have many bottlenecks in engineering applications. As the scale of transmission lines and operating voltage levels continue to increase, the requirements for the accuracy and frequency of online monitoring of conductor sag are becoming increasingly stringent. Although existing geometric measurement and RTK positioning methods can provide high-precision sag height, they are limited by the complexity of equipment layout, discrete observation points and meteorological obstructions, making it difficult to conduct high-frequency, full-time continuous monitoring of the entire line; indirect inference methods based on tensiometers or accelerometers are difficult to achieve both accuracy and robustness due to the complex conductor tension-deformation coupling relationship and significant external disturbances. Therefore, there is an urgent need for a new monitoring approach that can take into account large-scale deployment, online high frequency, and strong environmental adaptability.

[0028] On the one hand, RTK positioning technology based on GNSS (such as Beidou) can provide centimeter-level absolute geometric height references. However, due to the discrete measurement of the equipment mounting point, it is difficult to reflect the continuous deformation of the conductor at the span between the two towers. In addition, the RTK mounting point is generally set at the center of the conductor. By default, the center of the conductor is the lowest point of the conductor and can be used to monitor sag. However, due to the influence of various factors such as wind load, temperature gradient, uneven tension, and the absolute height deviation between the two tower tops, the actual lowest point of the conductor often deviates from the midpoint of the span, resulting in systematic deviations when the midpoint height is used as the sag value.

[0029] During power system operation, conductor deformation and vibration not only alter the mechanical stress state but also cause subtle modulation of the current distribution and electromagnetic field radiation along the line. This variation in conductor sag leaves a recognizable "deformation fingerprint" in the time-frequency characteristics of EMI (electromagnetic interference) signals. The tension changes caused by conductor sag alter the local current distribution density and radiation impedance, resulting in a slight amplitude or phase shift in the EMI waveform received by the antenna that is positively correlated with the deformation amplitude. Although EMI signals are superimposed with multiple noise sources such as switching surges and lightning pulses, the inherent deformation modulation components can be separated and extracted in the high-resolution time-frequency domain. Therefore, EMI signals can be viewed as "implicit" deformation observables, providing a lightweight and easily deployable approach for sag monitoring.

[0030] Based on this, the inventors of the present application discovered in the process of practicing the present application that by combining the high-frequency dynamic characteristics of the EMI signal with the absolute position calibration of RTK, it is possible to use EMI to capture the transient information of the wire vibration and deformation, and to perform error correction and parameter calibration through the uninterrupted geometric reference provided by RTK, thereby achieving high-precision, full-time, online dynamic monitoring of the wire sag.

[0031] Further, Figure 1 A flowchart of an example of a method for analyzing high-voltage line interference signal characteristics for sag monitoring according to an embodiment of the present application is shown.

[0032] The execution entity of the method of the embodiment of the present application can be any controller or processor with computing or processing capabilities. Specifically, it can be implemented by a sag monitoring platform. By integrating electromagnetic interference signal feature extraction, GNSS geometric observation, environmental perception, and multi-source data fusion modeling, a high-voltage conductor sag monitoring system with continuous perception capabilities, high robustness, and dynamic response performance is constructed.

[0033] In some examples, it can be integrated into an electronic device or terminal through software, hardware, or a combination of software and hardware, and the type of terminal or electronic device can be diverse, such as a mobile phone, tablet computer, or desktop computer, etc.

[0034] like Figure 1 As shown, in step S110, polarized antenna assemblies are equidistantly arranged between two adjacent towers along the bottom of the high-voltage wire to collect electromagnetic interference signals, wherein the polarized antenna assemblies include horizontally polarized antennas and vertically polarized antennas.

[0035] It should be noted that during the transmission process, high-voltage transmission lines generate perceptible electromagnetic radiation fields due to the passage of power-frequency current, the effects of high-voltage electric fields, and nonlinear changes caused by environmental disturbances (such as wind-induced vibration and corona discharge). These radiation signals carry physical information such as the conductor's geometric deformation, tension changes, and operating status. Because the spatial distribution of electromagnetic signals is closely related to conductor sag, their spatial intensity and characteristic morphology can reflect the conductor's sag state.

[0036] In order to fully collect this electromagnetic information, the solution uses a polarized antenna assembly, which includes two parts: a horizontally polarized antenna and a vertically polarized antenna, which correspond to the sensitive directions of the horizontal and vertical electric fields in the electromagnetic field of the conductor respectively. Compared with the single polarization acquisition method, the dual-polarization acquisition method can significantly enhance the ability to capture the components of the signal in all directions, which is conducive to obtaining a more complete electromagnetic interference feature. In addition, by arranging the antenna components at equal distances, the spatial sampling uniformity can be effectively guaranteed, while having good scalability and spatial coverage capabilities. The antenna spacing can be based on the span of the conductor. L and the required spatial resolution Δ x OK, generally take L / N ,in N The signals in each polarization direction record the changes in the electromagnetic field caused by the conductor's vibration in the horizontal and vertical planes. This acquisition network does not require an elevated foundation and can be directly fixed to a ground support or the lower edge of a tower.

[0037] Figure 2 A schematic diagram of the system architecture of an example of a sag monitoring platform according to an embodiment of the present application is shown.

[0038] like Figure 2 As shown, the system platform consists of three main components: distributed monitoring stations, a base station, and a centralized data processing center. At each monitoring station, horizontally / vertically polarized EMI antennas, RTK receivers, and temperature and wind speed sensors share synchronized trigger signals to collect pre-processed EMI waveforms, GNSS differential positioning data, and environmental parameters (including conductor temperature and ambient wind speed) in real time. These sensors are powered by solar energy. The base station, equipped with a GNSS base station antenna, continuously outputs a differential correction stream via network equipment to ensure the monitoring station's positioning accuracy. The monitoring and base stations are interconnected via cellular / microwave links, transmitting multi-source raw data in real time to the data processing center. Upon receiving the data, the data processing center first performs EMI-GNSS feature fusion and sag prediction model inference to generate spatially resolved sag curves and lowest point estimates. The real-time monitoring results are then displayed on a visual interface and alarms are triggered based on thresholds, completing the entire online, high-precision sag monitoring process for high-voltage lines.

[0039] In some examples of the present application, the main plane of a horizontally polarized antenna is parallel to the high-voltage conductor, making it sensitive to the horizontal component of radiation caused by the asymmetric distribution of the conductor current. The main plane of a vertically polarized antenna is perpendicular to the high-voltage conductor, making it sensitive to the vertical component generated by the conductor oscillation. To decouple the two interference components of conductor current distribution and mechanical vibration, an orthogonal polarization layout is adopted: the horizontally polarized antenna is most sensitive to the main radiation field caused by the current, while the vertically polarized antenna is most sensitive to the field momentum caused by the conductor oscillation.

[0040] Specifically, the antennas are arranged equidistantly between two adjacent towers and fixed with non-metallic brackets and insulating bolts. A 1-50 MHz broadband logarithmic periodic dipole antenna with a feed impedance of 50Ω can be used, which can significantly improve the separation of horizontal and vertical polarization signals and reduce cross-coupling between different interference sources.

[0041] Specifically, in the acquisition of EMI signals, the time synchronization reference signal output by the RTK receiver is first obtained. Then, using the time synchronization reference signal as a trigger signal, the analog-to-digital converter is triggered to synchronously digitally sample the outputs of the horizontally polarized antenna and the vertically polarized antenna to obtain an electromagnetic interference signal that is time-aligned with the RTK positioning signal.

[0042] It should be understood that if the EMI signal and GNSS positioning data are not strictly aligned, phase misalignment and time drift will occur during feature fusion. Therefore, using the time synchronization reference signal (e.g., a 1 PPS pulse) output by the RTK receiver as the sampling trigger reference can achieve refined data time synchronization. For example, the 1 PPS pulse output of the RTK receiver is directly connected to the trigger input of the antenna assembly's analog-to-digital converter (ADC). After the ADC detects a rising edge, both antenna signals are continuously sampled simultaneously and a timestamp aligned with the PPS is added, thus ensuring strict timing synchronization between the raw EMI sampled data and the RTK positioning stream.

[0043] In step S120, the collected electromagnetic interference signal is band-pass filtered to obtain a pre-processed signal, and the pre-processed signal is subjected to empirical mode decomposition to extract M IMF components, M is an integer greater than 1.

[0044] It should be noted that the raw electromagnetic signals collected are non-stationary, nonlinear, and multi-component. Direct use of these signals is susceptible to noise, leading to inaccurate feature extraction. Therefore, bandpass filtering is required to limit the spectral range and remove frequency interference unrelated to sag changes, such as power frequency background noise and high-frequency random disturbances.

[0045] Then, the empirical mode decomposition algorithm is used to adaptively decompose the preprocessed signal and extract the M IMF components, each representing a local oscillation mode, correspond to a frequency component in the signal, and its physical meaning may be related to conductor vibration, tension fluctuation, or geometric deformation. This effectively decomposes the multi-scale physical characteristics contained in the complex original signal into several eigenmodes, facilitating the identification of signal components closely related to conductor sag.

[0046] In some examples of the embodiments of the present application, the electromagnetic interference signal is band-pass filtered from 1 MHz to 10 MHz using a linear phase FIR (Finite Impulse Response) band-pass filter to obtain a preprocessed signal that does not contain power frequency interference and high-frequency radio frequency interference.

[0047] It should be noted that in high-voltage transmission line operating environments, EMI signals contain valuable 1-10 MHz frequency modulation components caused by conductor vibration and tension variations, as well as significant 50 Hz power frequency interference and its harmonics, and RF background noise above 10 MHz. Therefore, based on grid operation and conductor radiation characteristics, the 1-10 MHz target passband was selected. This frequency band precisely covers the electromagnetic modulation components generated by conductor vibration and tension variations, while significantly attenuating the 50 Hz power frequency and its harmonics, as well as RF clutter above 10 MHz. This significantly improves the separation between the target signal and other noise components, thereby enhancing the signal-to-noise ratio of the EMI signal.

[0048] In some implementations, the passband boundaries are set at 1 MHz and 10 MHz, and the stopband starts at 0.9 MHz and 10.5 MHz, leaving ample transition band. The bandpass filter operates on the digital signal output by the ADC and maintains a consistent sampling rate, ensuring that the Nyquist frequency is above the upper passband limit. By employing a linear-phase FIR structure, the transient characteristics of the EMI signal are preserved, ensuring the integrity of the time-domain pulse morphology and preventing distortion of the signal's time-domain characteristics due to group delay variations.

[0049] In this embodiment, by selecting a linear phase FIR bandpass filter with a constant group delay characteristic, it is possible to filter out industrial frequency and high-frequency radio frequency interference without introducing time domain distortion, thereby retaining the transient pulse shape and phase information in the EMI signal as is, avoiding signal distortion.

[0050] Then, adaptive empirical mode decomposition is performed on the preprocessed signal to sequentially remove the energy accumulation ratio exceeding 95%. M IMF components.

[0051] It should be noted that EMI signals exhibit significant nonstationary and nonlinear characteristics as they vary with operating conditions and the environment, making them difficult to adaptively capture using classical time-frequency methods (such as STFT and wavelets). EMD decomposes the signal into a set of IMF components without pre-determining basis functions. Each IMF component corresponds to a component of the signal at a specific time-frequency scale, facilitating precise matching with sag-temperature response atoms in the physical dictionary.

[0052] More specifically, EMD processing includes the following operations:

[0053] 1) Extreme point detection: Scan all local maxima and minima within the preprocessed signal frame;

[0054] 2) Envelope construction: Construct the upper envelope through cubic spline interpolation With lower envelope , calculate the envelope average ;

[0055] 3) Atomic stripping: Let the candidate IMF ,in Is the original input signal, which represents the original electromagnetic interference signal input to the EMD processing at a certain moment; if The difference between the number of zero crossings and the number of extreme values ​​is ≤1 and the average amplitude of the envelope is less than 10 -3 ×std( x ),but as the first IMF; otherwise Continue to perform the IMF sifting iteration step;

[0056] 4) Residual processing: the remaining residual Repeat steps 1) to 3) as the new signal until the residual energy accounts for ≤ 5% or the residual is monotonic. For the The signal of the empirical mode component, means that by putting all The reconstructed signal is obtained by adding.

[0057] Mode number selection: Count the energy proportions of all IMF components and keep the first M IMF components with cumulative energy ≥ 95%, and M >1. Thus, the original EMI signal is successfully separated into multi-scale IMFs, each of which corresponds to a different physical cause, and only the front-end that contributes most of the signal energy is retained. M An IMF can not only cover the main components caused by sag changes and environmental jitter, but also discard the noise components with extremely low energy, achieving the effects of noise reduction, dimensionality reduction and focusing on key modes.

[0058] In step S130, based on the physical dictionary constructed based on the state combination of the preset sag range and the preset temperature range, M The IMF components are sparsely coded to obtain a sparse coefficient vector.

[0059] It should be noted that sparse coding is a representation learning method that aims to decompose signals into linear combinations of a few typical patterns. The core of the physical dictionary is to construct the characteristic patterns of EMI signals under different sag conditions and temperature levels through simulation or testing in advance, forming a set of typical primitives. Based on this, the real-time IMF components can be expressed as a sparse combination of these primitives, and the coding coefficients reflect the mapping position of the current conductor state in the entire state space.

[0060] Here, a physical dictionary based on the combination of sag and temperature state is constructed, and the obtained IMF components are sparsely coded to extract a sparse coefficient vector representing the conductor state. For example, the output sparse coefficient vector Corresponding to the state combination in the physical dictionary —That is The discrete sag value and discrete temperature values. The non-zero or large magnitude components in the vector identify This is the sag-temperature condition that the current EMI signal is most likely to match.

[0061] In some implementations, a physical dictionary is constructed by analyzing electromagnetic radiation test data obtained under different conductor sag curvature and temperature field conditions, and then discretizing and combining them to form typical signal samples. Furthermore, sparse coding employs optimization algorithms such as OMP (Orthogonal Matching Pursuit), LASSO (Least Absolute Shrinkage and Selection Operator), or ADMM (Alternating Direction Method of Multipliers) to obtain the minimum set of non-zero coefficients under error control, which should not be restricted here. Thus, sparse coding compresses high-dimensional, complex IMF features into low-dimensional, physically meaningful sparse coefficients, significantly improving the efficiency and identifiability of feature representation.

[0062] In step S140, the three-dimensional coordinates output by the RTK receiver suspended on the high-voltage wire are calculated by differential analysis to calculate the difference between the lowest point of the wire sag and the tower height to obtain a geometric sag observation value.

[0063] For example, the three-dimensional coordinates provided by a high-precision RTK receiver based on Beidou satellites can be used to extract geometric observations of the actual sag shape, which can be used as a geometric reference for the conductor state. For example, by setting the apex positions of two towers as reference points and selecting the center of the conductor as the lowest point, the conductor sag is calculated based on the coordinates of the lowest point, that is, the difference between the "lowest point height" and the "tower height."

[0064] Specifically, an RTK base station is deployed at a fixed point (such as a tower base) near the monitoring line. Using a dual-frequency (L1 / L2) GNSS receiver, the precise geographic coordinates (WGS-84) of this point are measured. A rover equipped with the same dual-frequency receiver is then mounted midway along the line. The antenna is secured to the lowest point of the line using an insulating bracket to minimize horizontal and vertical offset from the line's center. The base station uses real-time differential correction data (in RTCM format) to perform real-time dual-frequency carrier phase differential analysis, achieving centimeter-level real-time positioning.

[0065] For example, the known elevation of the tower base (In the same coordinate system value) and the satellite elevation measured by the rover antenna Subtracting them, we can get the height difference between the lowest point of the wire sag and the tower top: .

[0066] For details on geometric sag measurement based on GNSS signals, please refer to previous research and will not be repeated here. However, as previously described, GNSS-based measurements can only support discrete single-point measurements and cannot fully reproduce the sag shape. Furthermore, it requires assuming an ideal minimum sag point for the conductor (i.e., the midpoint between the two towers). However, the actual minimum sag point may deviate, resulting in poor measurement accuracy.

[0067] It should be noted that in the embodiments of this application, EMI does not refer to all "noise interference" that is harmful to the system. Rather, it refers to the electromagnetic field signals or electromagnetic radiation signals radiated by high-voltage conductors when carrying current, vibrating, or experiencing temperature changes. Although it is often viewed as an "interference source" in power systems, these signals contain information about conductor sag, vibration, and tension changes. Therefore, it is not considered noise to be removed, but rather a valuable passive sensor, namely, extracting components highly correlated with conductor deformation to compensate and correct GNSS geometric measurements.

[0068] In step S150, conductor temperature information and ambient wind speed information are collected, the conductor temperature information and the sparse coefficient vector are encoded to generate a corresponding EMI branch feature coding vector, and the ambient wind speed information and the geometric sag observation value are encoded to generate a corresponding GNSS branch feature coding vector. The EMI branch feature coding vector and the GNSS branch feature coding vector are weightedly fused according to the fusion weight to obtain a multi-source fusion feature. The fusion weight is determined based on the EMI signal-to-noise ratio and the GNSS positioning reliability.

[0069] Here, the EMI branch jointly encodes the conductor temperature and sparse coefficient vector to construct a temperature-signal coupling feature, reflecting the relationship between physical properties and structural state. The GNSS branch jointly models wind speed and geometric sag, reflecting the impact of external disturbances on sag geometric deformation. Furthermore, the weight ratio of the two branches is dynamically determined based on the EMI signal-to-noise ratio (SNR) and GNSS positioning accuracy, improving the system's adaptability.

[0070] In some implementations, temperature and wind speed are collected using temperature sensors and wind speed sensors, respectively. For example, using the time synchronization reference signal output by an RTK receiver as a trigger source, conductor temperature information and ambient wind speed information are read from a platinum resistance temperature sensor on the conductor surface and a wind speed sensor in an open area, respectively. These environmental sensing parameters are then concatenated and transformed with the EMI branch sparsity coefficients and GNSS branch geometry observations.

[0071] Specifically, in the EMI branch feature coding, the temperature value With sparse coefficient vector First, normalize them separately:

[0072] , formula (1)

[0073] Where, and represent the minimum and maximum values ​​of the temperature range considered, represents the normalized temperature value, represents the normalized sparse coefficient vector;

[0074] Constructing joint input , Represents a vector transpose operation.

[0075] Will Enter a two-layer fully connected network:

[0076] , formula (2)

[0077] Where, Represents the ReLU activation function; and are the weight matrix and bias term of the first fully connected network, and are the weight matrix and bias term of the second fully connected network respectively.

[0078] Output features That is the EMI branch feature coding vector.

[0079] Through GNSS branch feature coding, wind speed (Sliding mean) and geometric sag observations Normalization:

[0080] , formula (3)

[0081] Where, represents the normalized wind speed value, represents the maximum wind speed of the standardized reference, represents the normalized GNSS positioning accuracy, Indicates the maximum value in the GNSS accuracy metric, Indicates the minimum value in the GNSS accuracy metric.

[0082] Get the concatenation vector: ;

[0083] Input a two-layer fully connected network with the same structure:

[0084] , formula (4)

[0085] Where, and are the weight matrices of the first and second fully connected layers respectively, and are the bias terms of the first fully connected layer and the second fully connected layer respectively.

[0086] Output That is the GNSS branch feature coding vector.

[0087] In the EMI branch, temperature encoding enables automatic adjustment of the sparsity coefficient to different conductor thermal expansion and contraction conditions, thereby improving the EMI signature's ability to distinguish sag conditions. Wind speed is the primary external disturbance causing conductor oscillation and transient displacement. In the GNSS branch, wind speed encoding can compensate for short-period oscillations in geometric measurements caused by wind loads, improving the stability of geometric observations.

[0088] After the EMI and GNSS branches each encode feature vectors, fusion weights need to be dynamically assigned based on the real-time reliability of the two observation types to fully leverage their complementary strengths. Specifically, when the EMI signal-to-noise ratio is high and GNSS positioning stability is poor, the system should increase the trustworthiness of the EMI branch; conversely, when the signal-to-noise ratio is high, the GNSS branch should be prioritized. This effectively suppresses error amplification caused by the unreliability of a single sensor, and adaptively corrects observations under different operating conditions, improving the stability and accuracy of multi-source fusion features.

[0089] In step S160 , the multi-source fusion features are input into the sag prediction model to output the corresponding sag estimation value.

[0090] Here, by inputting multi-source fusion features into a pre-trained sag prediction model, an estimated sag value for the current conductor state is output for use in alarming, monitoring, or trend forecasting. The sag estimation problem is essentially a nonlinear regression problem. Considering the influence of time series and physical factors, various time series analysis models or regression analysis models, such as LSTM, Transformer, or lightweight neural networks, can be used. In some implementations, historical samples and actual observation data are fused during model training, with the input features being fused vectors and the output being continuous sag values, to improve prediction accuracy and generalization capabilities. Furthermore, during the online inference process, the prediction output is regularly refreshed using a sliding window to capture the nonlinear response of the conductor due to temperature, wind speed, and vibration, achieving high-precision sag estimation.

[0091] In some examples of the embodiments of the present application, the sag prediction model adopts an end-to-end multilayer perceptron (MLP) regression network.

[0092] It should be noted that the end-to-end multi-layer perceptron (MLP) can learn nonlinear mappings directly from high-dimensional fused feature vectors without the need to manually design intermediate features or sub-models. Its multi-layer structure and ReLU activation function can approximate any continuous function, meeting the fitting requirements of the complex coupling relationship between sag and features. At the same time, the MLP inference process only involves matrix multiplication and addition and simple activation, supporting parallel acceleration to achieve efficient inference.

[0093] Here, based on an end-to-end multilayer perceptron regression network, combined with data-driven and physical priors, the loss function simultaneously minimizes the network prediction error and the consistency difference between the predicted value and the sag value of the theoretical model. Specifically, during the training phase of the sag prediction model, the network parameters are optimized by using the following loss function:

[0094] , formula (5)

[0095] , formula (6)

[0096] Where, represents the training loss function, Indicates the number of samples in a batch during training, represents the sample index, Indicates the The predicted sag value of samples output by the regression network is: Indicates the The true sag value of the samples, Indicates the The measured temperature value of the wire corresponding to the sample; The theoretical droop function based on the catenary physical model of the conductor is input as temperature The droop value of the physical model obtained by post-calculation; represents the weight coefficient of the physical consistency term; is the weight per unit length of the wire, which is equal to the product of the wire density and the acceleration due to gravity; is the horizontal span of the conductor between two adjacent towers, is the reference temperature Initial tension of the wire under It is the conductor temperature-tension attenuation coefficient, which indicates the relative change in conductor tension when the temperature rises by 1°C.

[0097] On the Theoretical Droop Function As explained in the literature, a conductor forms a catenary under the influence of its own weight and tension. The amount of sag is proportional to the inverse of the tension, which in turn decays predictably with temperature. A theoretical sag function, using temperature as the sole independent variable, can effectively combine real-time temperature measurement with line geometry, providing a physical mapping for estimating static sag without the need for additional tension sensing or comprehensive distance measurement. In some cases, the theoretical sag function can be calibrated through temperature-tension experiments and subjected to multiple field tests to ensure its reliability in predicting sag over a range of common temperature variations.

[0098] In the embodiment of the present application, the sag prediction model adopts a hybrid loss function including a physical consistency term. On the basis of the traditional method of minimizing the mean squared error (MSE), the physical consistency term is introduced. As a regularization measure, coupling the predicted values ​​with theoretical droop values ​​based on a catenary model and real-time temperature calculations significantly reduces the deviation between the model output and physical laws, preventing the network from making physically unreasonable predictions when noisy or with low-label data. This reduces the error between the network output and the theoretical catenary droop while maintaining high fitting accuracy to real calibration data, ensuring prediction stability and physical interpretability.

[0099] Figure 3 An operational flow chart of an example of building a physical dictionary according to an embodiment of the present application is shown.

[0100] like Figure 3 As shown, in step S310, within the preset sag range Internal equidistant dispersion J Status , and the wire working temperature is within the preset temperature range Internal equidistant dispersion K Status .

[0101] Here, in order to make the physical dictionary cover the different sags and ambient temperatures that may occur in high-voltage conductors during operation, it is necessary to pre-grid the two-dimensional state space. For example, the minimum sag allowed for the conductor is determined based on engineering design specifications or historical monitoring data. and maximum sag , and are equally spaced J Status ; Determine the operating temperature range based on extreme weather records , equidistantly scattered Status Therefore, through discretization processing, it is ensured that each dictionary atom can fully represent all working conditions from the most relaxed to the most tense, from low temperature to high temperature, providing a complete state basis for sparse coding.

[0102] In step S320, for each state combination , based on the positions of the horizontally polarized antenna and the vertically polarized antenna, electromagnetic interference signals are collected and extracted from the electromagnetic interference signals. M IMF components, the extracted M IMF components are spliced ​​into corresponding state combinations Dictionary atoms , each IMF component contains corresponding horizontal polarization sub-component and vertical polarization sub-component.

[0103] Here, the change in conductor sag affects the tension and geometric deformation, and the temperature changes the material impedance and resonance characteristics. The two jointly modulate the time-frequency characteristics of the EMI signal. Therefore, the corresponding signal is obtained under each state combination in order to construct a dictionary atom with physical meaning.

[0104] More specifically, for each state pair Do the following:

[0105] On the test bench or on site, use the pre-installed horizontal / vertical polarization antenna to adjust the wire tension (or directly adjust the hanging weight) to the corresponding Sag amount, and heat / cool to ;

[0106] Then, the collected EMI raw data is first subjected to 1-10 MHz bandpass filtering, and then adaptive EMD is performed to extract the EMI with cumulative energy ≥ 95%. IMF components, each IMF contains both horizontal and vertical polarization components;

[0107] To ensure repeatability, each state was sampled at least 10 times, and the average response of the IMF component was calculated as the final result.

[0108] Therefore, the IMF components obtained under precisely controlled sag and temperature conditions can better reflect the resonance and coupling characteristics of the EMI signal under this working condition, laying the foundation for the physical interpretability and matching accuracy of the dictionary atoms.

[0109] In step S330, a physical dictionary matrix for sparse coding is constructed based on the dictionary atoms corresponding to all state combinations. .

[0110] In some embodiments, the extracted The IMF components are linearly concatenated to form a "template vector" that can uniquely identify the state.

[0111] Specifically, for each state combination :

[0112] Take out the extracted IMF components, each containing horizontal and vertical subcomponents, are divided into The order of splicing into a vector ; Indicates the M The empirical mode decomposition components of the horizontal components, Indicates the M The empirical mode decomposition components of the vertical components.

[0113] Then, yes Perform L2 normalization to eliminate the impact of overall amplitude differences on subsequent matching. Cache to the local dictionary storage unit. Thus, through splicing and normalization, it ensures that each dictionary atom only reflects the state difference and is not disturbed by signal strength fluctuations, improving the discrimination ability and numerical stability during dictionary matching.

[0114] Furthermore, all dictionary atoms are arranged in a fixed index order to construct a matrix format that is convenient for fast retrieval by the sparse coding algorithm.

[0115] For example, the sag index is and temperature index Double loop, each Splice by column to form a physical dictionary matrix , corresponding to the matrix metadata record pairing, so that it can be directly used in dictionary sparse coding algorithms.

[0116] In the embodiment of the present application, the two-dimensional state space of the sag and temperature of the conductor is pre-dispersed with equal distances, and the EMI signal is collected, filtered, decomposed and IMF is extracted under each state combination, and then the EMI signal under the same state is obtained. MThe IMFs are spliced ​​in sequence and normalized into dictionary atoms, and finally assembled into a physical dictionary in matrix form according to the dual order of sag index and temperature index, which realizes the precise mapping between physical state and EMI characteristics, enables sparse coding to directly find the most matching EMI pattern from the prior state space, and establishes a mapping from physical quantity to signal characteristics.

[0117] Therefore, the constructed physical dictionary not only covers the typical EMI responses of conductors under different sag and temperature conditions, but also ensures numerical stability and efficient retrieval through normalization and ordered matrix structure, greatly improving the matching accuracy and computational efficiency of sparse coding.

[0118] Figure 4 An operational flowchart of an example of sparse coding solution according to an embodiment of the present application is shown.

[0119] like Figure 4 As shown, in step S410, according to M IMF components, forming the observation vector .

[0120] Here, the adaptive EMD extraction M The IMF components are uniformly flattened into a single vector to integrate the dual-polarization features into the same space.

[0121] For example, the IMF components under horizontal and vertical polarization are read in sequence. ( ), and concatenate them sequentially into observation vectors:

[0122] , formula (7)

[0123] Furthermore, we can also subtract The mean of the value is divided by the standard deviation to eliminate the amplitude offset by zero-mean processing, and the processed cache.

[0124] In step S420, based on the constructed physical dictionary matrix and the observation vector , establish an L1 regularized least squares model.

[0125] Here, the physical dictionary As the basis, L1 constraints are used to promote the sparsity of the solution, so that only a few dictionary atoms that are most relevant to the current observation are activated, corresponding to the most likely sag-temperature state.

[0126] Specifically, by loading the physical dictionary matrix , determine the regularization parameter by cross-validation offline or online (usually in the range of 0.01-0.1) to balance residual error with sparsity.

[0127] Construct the L1 regularized least squares model as:

[0128] , formula (8)

[0129] Where, is the L1 regularization weight, is a sparse coefficient vector.

[0130] Will and After normalization, input the solver at the same time and initialize , ADMM Lagrange multiplier And set the penalty factor Therefore, the model form is compatible with L2 residual minimization and L1 sparsity constraint, which not only ensures fitting accuracy but also compresses redundant atoms.

[0131] It should be noted that the mapping between the EMI signal and the physical dictionary matrix needs to achieve accurate matching of the sag-temperature state under high-dimensional and noise superposition conditions. By adopting the L1 regularized least squares model to construct the objective function, the fitting residual (L2 term) and the sparsity constraint (L1 term) are organically combined: on the one hand, the minimum error between the observed signal and the dictionary reconstruction is approached; on the other hand, only a small number of the most relevant dictionary atoms are forced to be activated, thereby retaining the most physically relevant features in an environment with complex interference signals and changeable working conditions, and generating a highly sparse and physically interpretable coefficient vector.

[0132] In step S430, the L1 regularized least squares model is iteratively solved by the ADMM method until the original residual is less than the first positive convergence threshold and the dual residual is less than the second positive convergence threshold, and the converged sparse coefficient vector is output. .

[0133] Here, the original problem is decomposed into two sub-problems: one is to minimize the residual L2 term, and the other is to perform the soft threshold processing L1 term. Alternating updates can converge quickly and are easy to parallelize.

[0134] Specifically, for the subproblem Update: Fixed , solved by the conjugate gradient method:

[0135] , formula (9)

[0136] Where, Indicates the The sparse coefficient vector updated in the iteration, and Respectively represent Auxiliary and dual variables in iterations;

[0137] Subproblems z renew:

[0138] right Perform soft thresholding element-wise:

[0139] , formula (10)

[0140] Where, Indicates the updated Auxiliary variables for elements In the The value in the iteration; The intermediate value for the soft threshold operation No. The value of an element.

[0141] Multiplier u renew:

[0142] , formula (11)

[0143] When the original residual in ADMM and the dual residual are all less than the preset threshold (e.g. ), the iteration ends and the converged sparse coefficient vector is output . Further, the final sparse coefficient It is considered as the activation weight of each state atom, directly reflecting the sag-temperature state most likely corresponding to the current EMI signal, providing physically interpretable input features for the regression model. Therefore, using sparse coefficients as features has high mapping efficiency and can ensure the speed of subsequent regression model training and inference.

[0144] Here, by using the ADMM method to iteratively solve the above composite objective function, the L2+L1 composite problem that is difficult to optimize directly is split into two tractable subproblems: one subproblem focuses on minimizing the residual (L2 objective), and the other subproblem implements the sparse constraint (L1 objective) through a soft threshold operation. The two are coordinated in the iteration through Lagrange multipliers, so that a closed form or efficiently parallelized update step can be obtained in each iteration.

[0145] This embodiment utilizes the alternating direction method of multipliers (ADMM) to separate residual minimization and soft-threshold sparse processing into two efficient subproblems, enabling rapid convergence to the optimal solution. This accurately maps the observation vector to a small number of the most relevant dictionary atoms, establishing an intuitive connection between the EMI signal and the physical state. This activates only the few dictionary atoms that represent the actual sag-temperature condition, significantly suppressing noise and redundant information and achieving state recognition with a high signal-to-noise ratio.

[0146] Figure 5 An operational flowchart of an example of multi-source feature weighted fusion processing according to an embodiment of the present application is shown.

[0147] like Figure 5 As shown, in step S510, a sparse reconstructed signal is generated based on the sparse coefficient vector and the physical dictionary, and the EMI signal-to-noise ratio is calculated according to the sparse reconstructed signal and the preprocessed signal.

[0148] , formula (12)

[0149] , formula (13)

[0150] Where, Indicates the signal-to-noise ratio of the EMI signal, reflecting the comparison between the reconstruction error and the original signal strength; Indicates that the preprocessed signal The amplitude of each sampling point; Represented by the physical dictionary matrix and the converged sparse coefficient vector The reconstructed signal is The amplitude of each sampling point; It means taking the expectation of all time series samples, which is used to estimate the average energy of the signal or error.

[0151] Here, the EMI branch measures the dictionary matching accuracy by reconstructing the error. Specifically, by preprocessing the real-time signal With physical dictionary and sparse coefficients The reconstructed signal Energy comparison quantifies the ratio of the residual energy from dictionary matching to the original signal energy, yielding a signal-to-noise ratio (SNR) metric that directly reflects the quality of EMI feature extraction. Specifically, if the dictionary and coefficients accurately capture the deformation and vibration components of the current EMI signal, the reconstruction error will be minimal and the SNR will significantly increase. Conversely, in the presence of noise or model mismatch, the SNR will decrease.

[0152] In step S520, the GNSS positioning reliability index is calculated based on the vertical precision factor output by the RTK differential calculation.

[0153] , formula (14)

[0154] Where, Represents the GNSS positioning reliability index; This is the vertical precision factor output by the RTK differential solution. The smaller the value, the more accurate the altitude solution.

[0155] It should be understood that Vertical Dilution of Precision (VDOP) is a well-established positioning quality metric in GNSS / RTK systems. Specifically, during differential analysis, an RTK receiver outputs a set of DOP values, including PDOP (position), HDOP (horizontal), and VDOP (vertical). Therefore, VDOP can be directly derived. GNSS vertical positioning accuracy reflects altitude measurement reliability.

[0156] In step S530, the EMI signal SNR and the GNSS positioning reliability index are respectively normalized by min-max to obtain the corresponding normalized EMI signal SNR. and normalized GNSS positioning reliability index .

[0157] Specifically, the fusion weight is calculated by the following formula:

[0158] , formula (15)

[0159] , formula (16)

[0160] Where, It is the fusion scaling coefficient, which is used to adjust the sensitivity of different normalized inputs to weights; Represents the original weight score of the EMI branch, as exponential growth; represents the original weight score of the GNSS branch, as exponential growth; is the fusion weight.

[0161] In step S540, the EMI branch feature encoding vector is coded according to the fusion weight. and GNSS branch feature coding vector Perform weighted fusion to obtain multi-source fusion features :

[0162] , formula (17)

[0163] Here, the normalized SNR and Rel are respectively index-mapped and normalized to and , dynamically assigning weights between the EMI and GNSS branches in feature fusion so that the fusion weights are proportional to the real-time quality of the corresponding observations, enabling the system to automatically switch its reliance on high-confidence sources. Consequently, when EMI noise levels are low and GNSS accuracy is high, the system naturally favors the GNSS contribution; when EMI signals are clear and GNSS is obstructed, the system shifts to strengthening the EMI branch.

[0164] Through the embodiments of this application, fusion weights are dynamically calculated based on real-time signal-to-noise ratio and vertical dilution of precision, and the feature vectors of the EMI branch and the GNSS branch are linearly superimposed according to adaptive weights, achieving accurate measurement and switching of the dynamic credibility of the two types of observations under different operating conditions. This allows for both capturing high-frequency information about conductor vibration and deformation using fast-responding EMI signals and relying on the absolute height reference provided by GNSS positioning. Consequently, adaptive weighting greatly improves the accuracy and stability of the fused features under various complex electromagnetic and meteorological conditions, significantly optimizing the real-time accuracy and reliability of sag monitoring.

[0165] In addition to using EMI signals to correct transient offsets in GNSS geometric measurements, when GNSS positioning experiences weak signals or temporary failures due to signal obstruction, ionospheric disturbances, or severe multipath, the EMI branch can still continuously output sag estimates based on the electromagnetic characteristics of the conductor's own vibration and temperature modulation. By dynamically weighting EMI with real-time GNSS data, the system can automatically increase confidence in EMI measurements when GNSS quality degrades, ensuring continuous monitoring under all operating conditions.

[0166] It should be emphasized that in this embodiment, the "global implicit perception" of the EMI branch on the sag shape of the entire span at different positions is adopted to compensate and correct the discrete geometric measurement of GNSS.

[0167] Specifically, the EMI branch uses dictionary-driven sparse coding to identify the sag state corresponding to the entire span (i.e., the offset and sag of the true lowest point relative to the span), rather than the deformation at a specific location (such as the midpoint between two towers). The GNSS branch provides the absolute height reference at that point. Furthermore, the EMI branch's sparse coefficients, obtained at different equidistant locations, describe the sag state that best matches the current span and guide the network to automatically adjust the most likely lowest point location and sag value corresponding to the GNSS observation, thus breaking the inherent assumption that "midpoint = lowest point." When the actual lowest point deviates from the midpoint, the GNSS observation is automatically corrected based on the EMI signature to accurately locate and quantify the sag after the offset, truly enabling fine-grained and continuous deformation monitoring of the span.

[0168] Figure 6 A schematic diagram showing an example of a comparison effect of a sag monitoring result based on EMI and GNSS fusion according to an embodiment of the present application and a sag monitoring result based on GNSS is shown.

[0169] like Figure 6 , which shows the comparison of the sag estimation error over time based on the fusion of EMI and GNSS and based only on GNSS according to the embodiment of the present application. The yellow curve represents the single-point sag estimation error based only on GNSS positioning, with an average error in the range of 0.04-0.06 m and large fluctuations due to wind load and multipath interference. The orange curve represents the estimation error after EMI-GNSS fusion according to the embodiment of the present application, with an average error stable in the range of 0.015-0.03 m and a significantly reduced fluctuation amplitude. From this, it can be intuitively seen that through EMI signal compensation and multi-source dynamic weighted fusion, the sag estimation accuracy can be improved by about one-fold, and higher robustness can be maintained in complex electromagnetic environments.

[0170] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity of description, they are all expressed as a series of combined actions, but those skilled in the art should be aware that this application is not limited by the order of the actions described, because according to this application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by this application. In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0171] Figure 7 A structural block diagram of an example of a high-voltage line interference signal characteristic analysis system for sag monitoring according to an embodiment of the present application is shown.

[0172] like Figure 7 As shown, the high-voltage line interference signal feature analysis system 700 for sag monitoring includes a polarization EMI signal acquisition unit 710, an IMF component extraction unit 720, a sparse coding unit 730, a geometric sag measurement unit 740, a multi-source coding fusion unit 750 and a sag prediction unit 760.

[0173] The polarized EMI signal acquisition unit 710 is used to arrange polarized antenna components equidistantly between two adjacent towers along the bottom of the high-voltage wire to collect electromagnetic interference signals; the polarized antenna components include horizontally polarized antennas and vertically polarized antennas.

[0174] The IMF component extraction unit 720 is used to perform bandpass filtering on the collected electromagnetic interference signal to obtain a preprocessed signal, and perform empirical mode decomposition on the preprocessed signal to extract M IMF components; M is an integer greater than 1.

[0175] The sparse coding unit 730 is used to generate the sparse coding unit 730 based on the physical dictionary constructed based on the state combination of the preset sag range and the preset temperature range. M The IMF components are sparsely coded to obtain a sparse coefficient vector.

[0176] The geometric sag measurement unit 740 is used to calculate the three-dimensional coordinates using the differential calculation output by the RTK receiver suspended on the high-voltage wire, and calculate the difference between the lowest point of the wire sag and the tower height to obtain the geometric sag observation value.

[0177] The multi-source coding fusion unit 750 is used to collect conductor temperature information and ambient wind speed information, encode the conductor temperature information and the sparse coefficient vector to generate a corresponding EMI branch feature coding vector, and encode the ambient wind speed information and the geometric sag observation value to generate a corresponding GNSS branch feature coding vector. The EMI branch feature coding vector and the GNSS branch feature coding vector are weightedly fused according to a fusion weight to obtain a multi-source fusion feature; the fusion weight is determined based on the EMI signal-to-noise ratio and the GNSS positioning reliability.

[0178] The sag prediction unit 760 is configured to input the multi-source fusion features into a sag prediction model to output a corresponding sag estimation value.

[0179] In some embodiments, an embodiment of the present application provides a non-volatile computer-readable storage medium, in which one or more programs including execution instructions are stored. The execution instructions can be read and executed by an electronic device (including but not limited to a computer, a server, or a network device, etc.) to execute the steps of any of the above-mentioned high-voltage line interference signal characteristic analysis methods for sag monitoring in the present application.

[0180] In some embodiments, the embodiments of the present application also provide a computer program product, which includes a computer program stored on a non-volatile computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer executes any step of the above-mentioned high-voltage line interference signal characteristic analysis method for sag monitoring.

[0181] In some embodiments, an embodiment of the present application also provides an electronic device, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the steps of a high-voltage line interference signal characteristic analysis method for sag monitoring.

[0182] The above-mentioned product can execute the method provided in the embodiment of this application, and has the functional modules and beneficial effects corresponding to the execution method. For technical details not fully described in this embodiment, please refer to the method provided in the embodiment of this application.

[0183] The electronic devices of the embodiments of the present application exist in various forms, including but not limited to:

[0184] (1) Mobile communication devices: These devices are characterized by their mobile communication capabilities and their primary purpose is to provide voice and data communications. These terminals include smartphones, multimedia phones, feature phones, and low-end phones.

[0185] (2) Ultra-mobile personal computer devices: These devices fall under the category of personal computers and have computing and processing capabilities, and generally also have mobile Internet access. These terminals include PDAs, MIDs, and UMPCs.

[0186] (3) Portable entertainment devices: These devices can display and play multimedia content. They include audio and video players, handheld game consoles, e-books, smart toys, and portable car navigation devices.

[0187] (4) Other onboard electronic devices with data interaction functions, such as onboard computer devices installed in vehicles.

[0188] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of this embodiment.

[0189] Through the description of the above embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a general hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the relevant technology, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, a server, or a network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0190] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for analyzing the characteristics of high-voltage line interference signals for sag monitoring, characterized in that: The method comprises: Polarized antenna assemblies are arranged equidistantly between two adjacent towers and below the high-voltage conductors to collect electromagnetic interference signals; the polarized antenna assemblies include horizontally polarized antennas and vertically polarized antennas; The collected electromagnetic interference signal is band-pass filtered to obtain a pre-processed signal, and the pre-processed signal is subjected to empirical mode decomposition to extract M IMF components; M is an integer greater than 1; On the physical dictionary constructed based on the combination of the state of the preset sag range and the preset temperature range, the M The IMF components are sparsely coded to obtain a sparse coefficient vector; The 3D coordinates are calculated using the differential output of the RTK receiver suspended on the high-voltage conductor. The difference between the lowest point of the conductor sag and the tower height is calculated to obtain the geometric sag observation value. collecting conductor temperature information and ambient wind speed information, encoding the conductor temperature information and the sparse coefficient vector to generate a corresponding EMI branch feature encoding vector, encoding the ambient wind speed information and the geometric sag observation value to generate a corresponding GNSS branch feature encoding vector, and performing weighted fusion on the EMI branch feature encoding vector and the GNSS branch feature encoding vector according to a fusion weight to obtain a multi-source fusion feature; the fusion weight is determined based on the EMI signal-to-noise ratio and the GNSS positioning reliability; The multi-source fusion features are input into a sag prediction model to output corresponding sag estimation values.

2. The method according to claim 1, characterized in that The main plane of the horizontally polarized antenna is parallel to the high-voltage conductor and is sensitive to the horizontal component radiation caused by the asymmetric distribution of the conductor current; the main plane of the vertically polarized antenna is perpendicular to the high-voltage conductor and is sensitive to the vertical component generated by the conductor oscillation; The collecting of electromagnetic interference signals includes: Acquire a time synchronization reference signal output by the RTK receiver; The time synchronization reference signal is used as a trigger signal to trigger an analog-to-digital converter to synchronously digitally sample the outputs of the horizontal polarization antenna and the vertical polarization antenna to obtain an electromagnetic interference signal that is time-aligned with the RTK positioning signal.

3. The method according to claim 1, characterized in that The collected electromagnetic interference signal is band-pass filtered to obtain a pre-processed signal, and the pre-processed signal is subjected to empirical mode decomposition to extract M IMF components, including: The electromagnetic interference signal is subjected to a 1 MHz to 10 MHz bandpass filtering by a linear phase FIR bandpass filter to obtain a preprocessed signal free of power frequency interference and high frequency radio frequency interference; Perform adaptive empirical mode decomposition on the pre-processed signal and sequentially remove the signals with a cumulative energy share of more than 95%. M IMF components.

4. The method according to claim 1, wherein The construction of the physical dictionary includes: In the preset sag range Internal equidistant dispersion J Status , and the wire working temperature is within the preset temperature range Internal equidistant dispersion K Status ; For each state combination , based on the positions of the horizontally polarized antenna and the vertically polarized antenna, electromagnetic interference signals are collected and extracted from the electromagnetic interference signals. M IMF components, the extracted M IMF components are spliced ​​into corresponding state combinations Dictionary atoms ; Each IMF component contains corresponding horizontal polarization sub-component and vertical polarization sub-component; Based on the dictionary atoms corresponding to all state combinations, a physical dictionary matrix for sparse coding is constructed .

5. The method according to claim 4, characterized in that The physical dictionary constructed based on the state combination of the preset sag range and the preset temperature range is used to M The IMF components are sparsely coded to obtain a sparse coefficient vector, including: According to the M IMF components, forming the observation vector ; Based on the constructed physical dictionary matrix and the observation vector , establish an L1 regularized least squares model: ， Where, is the L1 regularization weight, is a sparse coefficient vector; The L1 regularized least squares model is iteratively solved by the ADMM method until the original residual is less than the first positive convergence threshold and the dual residual is less than the second positive convergence threshold, and the converged sparse coefficient vector is output. .

6. The method according to claim 5, characterized in that The step of performing weighted fusion on the EMI branch feature coding vector and the GNSS branch feature coding vector according to the fusion weight to obtain the multi-source fusion feature includes: A sparse reconstructed signal is generated based on the sparse coefficient vector and the physical dictionary, and an EMI signal-to-noise ratio is calculated based on the sparse reconstructed signal and the preprocessed signal: ， ， Where, Indicates the signal-to-noise ratio of the EMI signal, reflecting the comparison between the reconstruction error and the original signal strength; Indicates that the preprocessed signal The amplitude of each sampling point; Represented by the physical dictionary matrix and the converged sparse coefficient vector The reconstructed signal is The amplitude of each sampling point; It means taking the expectation of all time series samples to estimate the average energy of the signal or error; Calculate the GNSS positioning reliability index based on the vertical precision factor output by RTK differential analysis: ， Where, Represents the GNSS positioning reliability index; This is the vertical precision factor output by the RTK differential solution. The smaller the value, the more accurate the altitude solution. The EMI signal SNR and GNSS positioning reliability index are normalized by min-max respectively to obtain the corresponding normalized EMI signal SNR. and normalized GNSS positioning reliability index , and the fusion weight is calculated by the following formula: ， ， Where, It is the fusion scaling coefficient, which is used to adjust the sensitivity of different normalized inputs to weights; Represents the original weight score of the EMI branch, as exponential growth; represents the original weight score of the GNSS branch, as exponential growth; is the fusion weight; The EMI branch feature encoding vector is coded according to the fusion weight and GNSS branch feature coding vector Perform weighted fusion to obtain multi-source fusion features : 。 7. The method according to claim 1, characterized in that The sag prediction model adopts an end-to-end multi-layer perceptron regression network; During the training phase of the sag prediction model, the network parameters are optimized by adopting the following loss function: ， ， Where, represents the training loss function, Indicates the number of samples in a batch during training, represents the sample index, Indicates the The predicted sag value of samples output by the regression network is: Indicates the The true sag value of the samples, Indicates the The measured temperature value of the wire corresponding to the sample; The theoretical droop function based on the catenary physical model of the conductor is input as temperature The droop value of the physical model obtained by post-calculation; represents the weight coefficient of the physical consistency term; is the weight per unit length of the wire, which is equal to the product of the wire density and the acceleration due to gravity; is the horizontal span of the conductor between two adjacent towers, is the reference temperature Initial tension of the wire under It is the conductor temperature-tension attenuation coefficient, which indicates the relative change in conductor tension when the temperature rises by 1°C.

8. A high-voltage line interference signal characteristic analysis system for sag monitoring, characterized in that: The system comprises: A polarized EMI signal acquisition unit is used to collect electromagnetic interference signals by using polarized antenna assemblies arranged equidistantly between two adjacent towers along the bottom of the high-voltage conductor; the polarized antenna assemblies include horizontally polarized antennas and vertically polarized antennas; The IMF component extraction unit is used to perform bandpass filtering on the collected electromagnetic interference signal to obtain a preprocessed signal, and perform empirical mode decomposition on the preprocessed signal to extract M IMF components; M is an integer greater than 1; The sparse coding unit is used to generate a sparse coding matrix based on a physical dictionary constructed based on a state combination of a preset sag range and a preset temperature range. M The IMF components are sparsely coded to obtain a sparse coefficient vector; The geometric sag measurement unit is used to calculate the three-dimensional coordinates using the differential output of the RTK receiver suspended on the high-voltage conductor, and calculate the difference between the lowest point of the conductor sag and the tower height to obtain the geometric sag observation value; a multi-source coding fusion unit, configured to collect conductor temperature information and ambient wind speed information, encode the conductor temperature information and the sparse coefficient vector to generate a corresponding EMI branch feature coding vector, encode the ambient wind speed information and the geometric sag observation value to generate a corresponding GNSS branch feature coding vector, and perform weighted fusion of the EMI branch feature coding vector and the GNSS branch feature coding vector according to a fusion weight to obtain a multi-source fusion feature; the fusion weight is determined based on the EMI signal-to-noise ratio and the GNSS positioning reliability; The sag prediction unit is used to input the multi-source fusion features into the sag prediction model to output a corresponding sag estimation value.

Images