Line icing early warning method and system based on millimeter wave radar and multi-source fusion

Abstract

The application provides a line icing early warning method and system based on millimeter wave radar and multi-source fusion, relates to the technical field of power system safety monitoring, and comprises the following steps: collecting multi-dimensional data of a power transmission line operation environment and millimeter wave radar echo signals, processing the echo signals to extract range-Doppler images, analyzing icing shape characteristic parameters from the range-Doppler images and inputting the icing shape characteristic parameters into a support vector machine classifier to identify the icing shape, establishing a Bayesian network model to calculate the prior probability of the icing shape, updating the posterior probability of the Bayesian network based on the identification result, determining a risk level, and remotely transmitting the determination result and position information. The application improves the identification accuracy by extracting the icing shape characteristic, improves the robustness of risk determination by fusing multi-source information through the Bayesian network, and solves the data transmission problem in remote areas by using satellite communication.

Description

Technical Field

[0001] This invention relates to the field of power system safety monitoring technology, specifically to a method and system for early warning of line icing based on millimeter-wave radar and multi-source fusion. Background Technology

[0002] Currently, transmission line icing monitoring technologies are mainly divided into three categories: manual inspection, indirect measurement, and direct detection. Manual inspection is inefficient, lacks real-time performance, and is difficult to adapt to complex terrain and severe weather. Indirect measurement methods estimate icing thickness indirectly through parameters such as conductor tension and sag, but are easily affected by environmental factors, resulting in insufficient measurement accuracy. Among direct detection methods, optical imaging is greatly affected by lighting and weather, while ultrasonic and capacitive sensor methods have limited monitoring ranges. In recent years, millimeter-wave radar technology has been applied to icing detection, but existing technologies mostly focus on the quantitative measurement of icing thickness. This method is highly dependent on geometric models, making it difficult to guarantee accuracy in complex environments, and it ignores the crucial impact of icing morphology on risk assessment. Furthermore, a single sensor cannot comprehensively reflect the icing formation mechanism, leading to a high misjudgment rate under complex meteorological conditions. Therefore, there is an urgent need for an intelligent early warning technology that can effectively identify icing morphology, accurately assess risk levels, and possess reliable long-distance communication capabilities. Summary of the Invention

[0003] To address the problems of existing technologies, such as the inability to effectively identify icing patterns, difficulty in accurately assessing icing risk levels, and poor reliability under complex weather conditions, this application proposes a line icing early warning method and system based on millimeter-wave radar and multi-source fusion, which enables intelligent perception, accurate classification, and reliable early warning of icing risks on transmission lines.

[0004] To achieve the above objectives, the present invention adopts the following technical solution: A method for early warning of line icing based on millimeter-wave radar and multi-source fusion includes the following steps: Step 1: Collect multi-dimensional data on the operating environment of the target transmission line, and collect electromagnetic echo parameters of the conductors using millimeter-wave radar; Step 2: Perform signal processing on the electromagnetic echo parameters to extract the range-Doppler image; Step 3: Extract the multi-dimensional feature parameter set representing the icing morphology of the conductor from the distance-Doppler image, input the multi-dimensional feature parameter set into the support vector machine classifier, and output the icing morphology recognition result. Step 4: Construct a Bayesian network model, using multidimensional data from the operating environment as input, to calculate the prior probability distribution of various icing morphologies; Step 5: Use the ice frost pattern recognition results as observation evidence to perform a posterior probability update on the Bayesian network model; Step 6: Calculate the posterior probability of different icing risk levels based on the posterior probability distribution of various icing patterns, and take the risk level corresponding to the highest posterior probability as the judgment result; Step 7: Transmit the judgment result and the location parameters of the target transmission line to the remote terminal.

[0005] Preferably, in step 1, the multidimensional data includes ambient temperature, ambient relative humidity, and ambient wind speed data collected by temperature sensors, humidity sensors, and wind speed sensors; the method for collecting electromagnetic echo parameters of the conductor by millimeter-wave radar is as follows: the millimeter-wave radar transmits frequency-modulated continuous wave signals toward the target transmission line and receives reflected echo signals from the conductor and the ice layer.

[0006] Preferably, in step 2, signal processing is performed on the electromagnetic echo parameters, specifically including the sequential processing of signal decoupling, dimension transformation, and target detection.

[0007] Preferably, in step 3, the multi-dimensional feature parameter set includes the relative change rate of the millimeter-wave radar cross section, the distortion value of the echo waveform, the statistical features of the Doppler spectrum, and the amplitude and phase feature parameters of the dual-polarized echo. The relative change rate of the radar cross section is the ratio of the measured value of the current time-domain radar cross section to the reference value of the radar cross section under the non-icing state. The distortion value of the echo waveform is the difference between 1 and the cross-correlation coefficient between the current echo waveform and the non-icing reference waveform. The statistical features of the Doppler spectrum are the centroid frequency and spectral width of the target peak group in the velocity dimension. The amplitude and phase feature parameters of the dual-polarized echo are the amplitude ratio and phase difference between the vertically polarized and horizontally polarized echoes.

[0008] Preferably, in step 3, the inputs of the support vector machine classifier are the relative change rate of the radar cross section, the distortion value of the echo waveform, the statistical features of the Doppler spectrum, and the amplitude and phase feature parameters of the dual-polarized echo, and the output is the morphological classification result of no icing, frost, hoarfrost, mixed frost, and rain frost.

[0009] Preferably, in step 6, the risk level is divided into three levels according to the ratio of the conductor load increment to the design value. The first level is a low-risk level where the load increment does not exceed 20% of the design value; the second level is a medium-risk level where the load increment is between 20% and 50% of the design value; and the third level is a high-risk level where the load increment exceeds 50% of the design value.

[0010] Furthermore, this invention also provides a line icing early warning system based on millimeter-wave radar and multi-source fusion, including a multi-source sensing module, a data processing and fusion module, a communication module, and a remote terminal. The multi-source sensing module includes an environmental parameter sensing unit and an electromagnetic echo sensing unit. The environmental parameter sensing unit is used to collect temperature, relative humidity, and wind speed parameters around the transmission line, while the electromagnetic echo sensing unit is used to transmit electromagnetic signals to the transmission line conductors and receive reflected echo signals. The data processing and fusion module uses an edge computing terminal as the core control unit, integrating a multi-channel data acquisition unit and a signal intelligent processing unit to realize parameter acquisition, signal preprocessing, feature analysis, morphological discrimination, and probability fusion calculation of the multi-source sensing module, and output the icing risk level judgment result. The communication module is used to realize the remote transmission of the icing risk level judgment result and location parameters. The remote terminal includes a data receiving server and a monitoring terminal to realize the reception, storage, analysis, visualization, and alarm triggering of transmitted data.

[0011] Preferably, the communication module is a satellite communication module, which includes a satellite short message transmission unit and a satellite positioning unit. The satellite positioning unit is used to determine the location of the icing transmission line, and the satellite short message transmission unit is used to send the icing risk level determination result and location parameter information to a remote terminal.

[0012] Preferably, the satellite short message transmission unit is based on the BeiDou RDSS system to realize data transmission, supports message length of 100-1000 Chinese characters, and has a transmission delay of less than 5 seconds.

[0013] Preferably, the edge computing terminal communicates with the millimeter-wave radar sensor via a serial peripheral interface, with the temperature and humidity sensors of the environmental parameter sensing unit via an integrated circuit bus interface, with the wind speed sensor of the environmental parameter sensing unit via a universal asynchronous transceiver interface, and with the communication module via a universal serial bus interface.

[0014] Beneficial effects: This invention extracts multi-dimensional morphological features of icing morphology, such as radar cross-section change rate, waveform distortion, Doppler characteristics, and polarization characteristics, from millimeter-wave radar, rather than simply measuring thickness. This more fundamentally reflects the physical differences between different icing morphologies (such as rime, hoarfrost, and mixed rime), significantly improving the accuracy of icing morphology identification. By constructing a Bayesian network model, prior information such as environmental temperature, humidity, and wind speed is probabilistically fused with radar morphology identification results, effectively reducing the risk of misjudgment by a single sensor under complex meteorological conditions and improving the robustness and reliability of risk level determination. Employing satellite communication methods such as BeiDou short message service overcomes the communication blind spots at power transmission line monitoring points in remote areas, enabling real-time remote transmission of early warning information. The system adopts an edge computing architecture, completing data acquisition, processing, and decision-making locally, reducing dependence on cloud resources and improving the system's response speed and autonomous operation capabilities. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the overall architecture of the line icing early warning system according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the overall process of the line icing early warning method according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the millimeter-wave radar signal processing flow according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the Bayesian network model structure according to an embodiment of the present invention; Figure 5 This is a schematic diagram of the BeiDou short message communication data frame format according to an embodiment of the present invention; Figure 6 is a simulation result of millimeter-wave radar signal processing according to an embodiment of the present invention, where (a) is the time-domain waveform of the FMCW transmitted signal, (b) and (c) are the intermediate frequency signals under the conditions of no icing and rime ice, respectively, (d) and (e) are the corresponding range-dimensional FFT spectra, (f) is the comparison of range spectra before and after icing, (g) is the range-Doppler image, and (h) is the CFAR constant false alarm rate detection result. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0017] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention.

[0018] Example 1: like Figure 2 As shown, this embodiment provides a method for early warning of power line icing based on millimeter-wave radar and multi-source fusion. This method achieves accurate early warning of transmission line icing risk by fusing environmental meteorological data with radar electromagnetic echo characteristics, specifically including the following steps: Step S100: Collect multi-dimensional data on the operating environment of the target transmission line and collect electromagnetic echo parameters of the conductor using millimeter-wave radar.

[0019] Specifically, this step establishes a data foundation based on dual environmental and electromagnetic sensing. The multidimensional data includes ambient temperature, relative humidity, and wind speed data collected by temperature, humidity, and wind speed sensors. In practical deployment, the temperature sensor can be installed near the lightning rod at the top of the monitoring tower, approximately 2 meters high, to avoid interference from ground heat radiation. Its measurement range covers -40℃ to +85℃, with an accuracy of ±0.3℃. The humidity sensor is installed in the upper middle part of the tower, ensuring good ventilation, and its measurement range is 0% to 100%RH. The wind speed sensor is installed at the top of the tower, using ultrasonic principles to measure wind speed and direction. Through this sensor array, the system can acquire real-time data on the meteorological factors inducing icing formation.

[0020] Simultaneously, this step involves acquiring electromagnetic echo parameters of the power line using millimeter-wave radar. Specifically, the millimeter-wave radar transmits a frequency-modulated continuous wave (FMCW) signal towards the target power line and receives reflected echo signals from the power line and the icing layer. In this embodiment, the millimeter-wave radar preferably operates in the 77GHz to 81GHz frequency band, employing a FMCW system with a bandwidth of 4GHz and a theoretical range resolution of 3.75cm. The transmitted signal is a linear frequency-modulated pulse, which is reflected upon encountering the power line and the icing layer, and the radar receiver captures the echo signal. Due to differences in icing morphology (such as rime and hoarfrost), their dielectric constant and surface roughness vary, leading to changes in the amplitude, phase, and scattering characteristics of the reflected echo. This provides a physical basis for subsequent morphological identification based on electromagnetic features. It should be understood that although this embodiment lists specific sensor types and radar frequency bands, in other embodiments, sensors such as air pressure and rainfall can be added, or the radar operating frequency band can be adjusted, as long as synchronous acquisition of environmental parameters and electromagnetic echoes is achieved.

[0021] Step S200: Perform signal processing on the electromagnetic echo parameters to extract the range-Doppler image.

[0022] This step aims to filter out noise from the raw echo and extract the target's range and velocity information. By decoupling and transforming the echo signal, the one-dimensional time-domain signal is converted into a two-dimensional range-Doppler image, thereby separating the target conductor from background clutter in the range and velocity dimensions.

[0023] Step S300: Extract the multi-dimensional feature parameter set representing the icing morphology of the conductor from the distance-Doppler image, input the multi-dimensional feature parameter set into the support vector machine classifier, and output the icing morphology recognition result.

[0024] This step utilizes machine learning algorithms to replace traditional threshold judgment. The multi-dimensional feature parameter set covers physical quantities such as radar cross-section change, waveform distortion, Doppler features, and polarization features, which can comprehensively characterize the electromagnetic scattering properties of icing. The support vector machine classifier maps the input feature vector to specific icing morphology categories, such as no icing, frost, hoarfrost, mixed frost, or rain frost, using a pre-trained model.

[0025] Step S400: Construct a Bayesian network model, using multidimensional data from the operating environment as input, to calculate the prior probability distribution of various icing morphologies.

[0026] This step introduces a probabilistic reasoning mechanism, utilizing meteorological data such as ambient temperature, humidity, and wind speed, combined with historical icing patterns, to calculate the prior probability of various icing patterns occurring under current meteorological conditions. This process simulates the physical mechanism of icing formation, providing environmental background support for the final risk assessment.

[0027] Step S500: Use the ice frost pattern recognition result as observation evidence to perform a posterior probability update on the Bayesian network model.

[0028] This step achieves deep fusion of multi-source information. The morphological results identified by radar in step S300 are used as strong observational evidence and input into the Bayesian network to correct the prior probability obtained in step S400. This fusion method effectively reduces the risk of misjudgment by a single sensor under complex weather conditions and improves the robustness of the identification results.

[0029] Step S600: Based on the posterior probability distribution of various icing morphologies, calculate the posterior probability of different icing risk levels, and take the risk level corresponding to the highest posterior probability as the judgment result.

[0030] This step makes a decision based on the maximum a posteriori probability criterion, outputting the final risk level. This judgment comprehensively considers environmental factors and electromagnetic characteristics, and has higher reliability.

[0031] Step S700: The determination result and the location parameters of the target transmission line are transmitted to the remote terminal.

[0032] This step enables remote delivery of early warning information. By packaging and sending the risk level and location information, remote maintenance personnel can promptly grasp the line status and take appropriate measures.

[0033] Through steps S100 to S700 described above, this embodiment constructs a complete method for early warning of icing. This method not only utilizes millimeter-wave radar to achieve precise perception of the microscopic morphological characteristics of icing, but also integrates macroscopic environmental information through Bayesian networks, solving the problem of poor reliability of single-sensor monitoring and providing strong protection for the safe operation of transmission lines.

[0034] Example 2: This embodiment is a further refinement of step S200 in Embodiment 1. For example... Figure 3 As shown, in step S200, signal processing is performed on the electromagnetic echo parameters, specifically including the serial processing of signal decoupling, dimension transformation, and target detection.

[0035] Specifically, the signal processing aims to extract the target's range and velocity information from complex raw echo data, and includes the following sub-steps: Step S201, signal decoupling. The millimeter-wave radar transmits a frequency-modulated continuous wave signal, typically a linear frequency-modulated pulse, towards the target power line. When the electromagnetic wave encounters the conductor and icing, it is reflected, and the radar receiver captures the echo signal. To obtain the target's range information, the system first performs de-chirping on the echo signal. This process mixes the received echo signal with a copy of the transmitted signal, converting the high-frequency signal into an intermediate-frequency (IF) signal. In this IF signal, the frequency components are proportional to the target's range; that is, the farther the distance, the higher the IF frequency. Through this decoupling process, the system transforms time delay information in the time domain into frequency distribution information in the frequency domain, laying the foundation for subsequent range dimension analysis.

[0036] Step S202, Dimension Transformation. After signal decoupling, the system performs a dimension transformation on the intermediate frequency (IF) signal to construct a range-Doppler image. First, a one-dimensional Fast Fourier Transform (FFT) is performed on the IF signal, i.e., range-dimensional FFT. This transform converts the time-domain IF signal into a frequency-domain signal, where the peak position in the spectrum corresponds to the target's distance, thus enabling the differentiation of targets at different distances. Subsequently, the system performs a two-dimensional Fast Fourier Transform (FFT), i.e., Doppler-dimensional FFT, between multiple consecutive pulses at the same range gate. This transform utilizes the Doppler frequency shift effect caused by the target's micro-motion (such as the vibration of a conductor under wind excitation) to extract the target's velocity information. Through the above two FFT transformations, the original one-dimensional time-domain echo data is converted into a two-dimensional range-Doppler image containing distance and velocity information. It should be understood that although this embodiment uses the FFT algorithm for dimension transformation, in other embodiments, time-frequency analysis methods such as Discrete Fourier Transform (DFT) or wavelet transform can also be used, as long as the distance and velocity features can be extracted from the echo signal.

[0037] Step S203, Target Detection. Due to interference from ground clutter and atmospheric noise in the actual monitoring environment, range-Doppler images often contain a large number of non-target signals. To accurately extract target information of transmission line conductors, the system uses a constant false alarm rate (CFAR) detection algorithm to process the image. Specifically, the CFAR algorithm calculates the local noise power level within the protection and reference cells and adaptively adjusts the detection threshold according to the set false alarm probability. Only when the signal strength is higher than this adaptive threshold is it determined to be a valid target. In this way, the system can effectively filter out background clutter and accurately locate the position of the conductor target in the range-Doppler image, thus providing clean target area data for subsequent extraction of icing morphology features.

[0038] The above-described serial processing logic for signal decoupling, dimension transformation, and target detection transforms the original electromagnetic echo parameters into an intuitive range-Doppler image. This not only enables accurate measurement of target distance and velocity, but more importantly, the processing link effectively suppresses noise and highlights the weak scattering characteristics of the icy conductor, providing high-quality input data for the subsequent step S300 to analyze the multi-dimensional feature parameter set.

[0039] To verify the effectiveness of the above signal processing procedure and the accuracy of feature extraction, this embodiment presents the entire process of millimeter-wave radar signal processing based on simulation data. As shown in Figure 6, Figure 6(a) shows the time-domain waveform of the FMCW transmitted signal, with a sweep period of 50μs, a bandwidth of 4GHz, and a center frequency of 79GHz. The comparison between Figure 6(b) and Figure 6(c) clearly shows that due to the high density of rime ice (approximately 0.9g / cm³),... 3Furthermore, the smooth surface of the ice-covered surface produces a strong specular reflection effect on electromagnetic waves. The amplitude of the mid-frequency signal under the icing condition is significantly higher than that under the uniced condition, demonstrating the sensitivity of millimeter-wave radar to changes in the dielectric constant of icing. A comparison between Figure 6(d) and Figure 6(e) shows that the peak value of the range-dimensional FFT spectrum under the icing condition is significantly larger than that under the uniced condition, intuitively verifying the increase in the relative rate of change of the radar cross section. Figure 6(f) further compares the range spectra before and after icing in the same coordinate system, clearly showing the RCS increase effect caused by icing. Figure 6(g) shows the range-Doppler image obtained after range-dimensional FFT and Doppler-dimensional FFT processing. The target conductor forms a clear target peak group in the range and velocity dimensions, indicating that the dimension transformation step successfully converts the time-domain signal into an analytical two-dimensional feature space. Figure 6(h) shows the results processed using the CFAR constant false alarm rate detection algorithm. In complex background noise, the adaptive threshold effectively filters out ground clutter, retaining only the clear peak value of the target conductor. This demonstrates that the target detection step described in this application can accurately lock onto the target under adverse weather conditions, providing a high signal-to-noise ratio data foundation for subsequent feature extraction. A four-dimensional feature vector (relative rate of change of radar cross section, distorted echo waveform value, Doppler spectrum statistical features, and dual-polarized echo amplitude and phase feature parameters) can be extracted from the detection results. The output after classification by the SVM classifier is rime ice. The above results indicate that the signal decoupling, dimension transformation, and target detection serial processing link described in this invention can effectively analyze the electromagnetic scattering characteristics of iced conductors and accurately extract multi-dimensional feature parameters characterizing the icing morphology, providing reliable feature input for subsequent icing morphology identification.

[0040] Example 3: This embodiment is a further refinement of step S300 in Embodiment 1. In step S300, a multi-dimensional feature parameter set characterizing the icing morphology of the conductor is parsed from the range-Doppler image. This multi-dimensional feature parameter set is then input into a support vector machine classifier, which outputs the icing morphology recognition result. Specifically, this step aims to extract key indicators that characterize the physical morphology of the icing from electromagnetic scattering features and to achieve automatic morphology discrimination through machine learning algorithms.

[0041] Step S301: Analyze the multi-dimensional feature parameter set. In this embodiment, the multi-dimensional feature parameter set includes the relative change rate of the millimeter-wave radar cross section, the distortion value of the echo waveform, the statistical characteristics of the Doppler spectrum, and the amplitude and phase characteristic parameters of the dual-polarized echo. These four feature parameters characterize the icing morphology from four dimensions: energy scattering, waveform structure, micro-motion characteristics, and polarization properties. The specific calculation process and physical meaning are as follows: It is important to note that there is a clear physical mapping relationship between the above four characteristic parameters and the physical morphology of ice accumulation; they are not simply numerical additions. For example, in distinguishing between rime ice and hoarfrost: rime ice has a higher density (approximately 0.9 g / cm³).3 Furthermore, the smooth surface of rime ice produces strong specular reflection of millimeter waves, significantly increasing the relative rate of change of radar cross section (typically +60% to +100%), while frost ice has low density and a loose structure, with an RCS change rate typically only between +15% and +30%. The smooth surface of rime ice results in relatively small distortion values ​​in the echo waveform, while mixed rime ice or frost ice, due to its inhomogeneous internal structure, produces volume scattering, leading to more severe waveform distortion. Rime ice has strong adhesion, resulting in low and stable conductor vibration frequencies and a narrower Doppler spectrum, while frost ice has weak adhesion, easily producing high-frequency micro-motions and a wider spectrum. The surface of rime ice is sensitive to polarization direction, exhibiting specific patterns in its dual-polarization amplitude and phase characteristics, while the rough surface of frost ice results in a significant depolarization effect, with an amplitude ratio approaching 1. Through the joint calculation of the above multi-dimensional features, this embodiment can accurately distinguish between icing morphologies that are similar in appearance but have drastically different physical properties, thereby verifying the effectiveness of the feature extraction method described in claim 3.

[0042] First, the relative rate of change of the millimeter-wave radar cross section is calculated. This characteristic parameter is defined as the ratio of the measured value of the radar cross section in the current time domain to the reference value of the radar cross section under icing-free conditions. In actual operation, the system establishes a reference value of the radar cross section through long-term statistics during the initial commissioning of the transmission line or during periods when icing is confirmed. When icing occurs, the physical dimensions, dielectric constant, and surface roughness of the conductor surface change, leading to a change in the radar cross section (RCS). For example, frost has a high density and a smooth surface, producing a strong specular reflection effect on electromagnetic waves, resulting in a significant increase in RCS; while rime has a low density and a loose structure, producing a volume scattering effect on electromagnetic waves, resulting in a relatively smaller increase in RCS. By calculating the ratio of the measured value to the reference value, the influence of differences in the conductor's own dimensions can be effectively eliminated, quantifying the change in scattered energy caused by icing. It should be understood that although this embodiment uses a ratio as the relative rate of change, in other embodiments, a difference or logarithmic difference can also be used, as long as it reflects the trend of RCS change.

[0043] In this embodiment, the radar cross-section change rate is calculated as follows: First, a baseline value for the radar cross-section is established during the non-icing period. Then calculate the radar cross-section at the current moment. Finally, calculate the rate of change. To further eliminate the influence of environmental factors, a dual-target ratio method is adopted: a calibrated corner reflector is set up outside the monitoring area as a reference target. By comparing the change in the ratio of the conductor echo to the corner reflector echo, the influence of atmospheric attenuation on the common path is eliminated, ensuring that the RCS change rate only reflects the change in icing status. The correspondence between the icing morphology and the icing morphology is shown in Table 1 below: The RCS change rate data mentioned above were obtained based on the following experimental and simulation conditions: the millimeter-wave radar operated at a center frequency of 79 GHz, a bandwidth of 4 GHz, a sweep period of 50 μs, a target distance of 2.5 m, and a sampling rate of 60 MHz. During the non-icing benchmark calibration phase, the system continuously collected radar echo data for 72 hours under clear weather conditions, and the median was used as the benchmark RCS. The RCS data under icing conditions was obtained through two methods: one was based on simulation calculations using an electromagnetic scattering physical model, with the dielectric constant and density parameters for each icing morphology being: frost / ice ε≈1.5, ρ≈100 kg / m³. 3 The rime ice has an ε≈2.1 and a ρ≈300kg / m³. 3 Mixed frost ε≈2.6, ρ≈600kg / m 3 Ice frost ε≈3.2, ρ≈900kg / m 3 Secondly, in an artificial climate laboratory, temperature, humidity, and wind speed conditions were controlled to conduct field measurements and calibrations on conductors with icing thicknesses. The deviations between the simulation and measured data were weighted and fused to obtain the RCS change rate range and typical values ​​shown in Table 1. The typical values ​​correspond to the statistical median under icing thickness conditions of 15mm~25mm.

[0044] Secondly, the distortion variability value of the echo waveform is calculated. This characteristic parameter is defined as the difference between 1 and the cross-correlation coefficient of the current echo waveform and the un-iced reference waveform. Icing not only changes the echo intensity but also alters the phase distribution and time-domain waveform structure. Without icing, the conductor echo waveform exhibits a standard frequency-modulated continuous wave compression waveform; after icing, due to multiple reflections and refractions within the ice layer and the discontinuity of the dielectric constant, the echo waveform shows broadening, tailing, or multi-peak phenomena. The cross-correlation coefficient measures the similarity between two waveforms; the closer the coefficient is to 1, the more similar the waveforms. Therefore, a larger distortion variability value indicates more severe waveform distortion. For example, mixed rime, due to its non-uniform internal structure, often leads to severe waveform distortion, and its distortion variability value is usually higher than that of uniformly structured rime.

[0045] In this embodiment, the method for calculating the echo waveform distortion is as follows: perform cross-correlation analysis between the current echo waveform and the non-icing reference waveform, and calculate the correlation coefficient. Then calculate the distortion degree. Based on the analysis of experimental data, The relationship between the value and the icing morphology is as follows: Frost Ice rime Mixed frost , rime .

[0046] Next, the statistical characteristics of the Doppler spectrum are calculated. These characteristics are defined as the centroid frequency and spectral width of the target peak group in the velocity dimension. Under wind excitation, iced conductors will exhibit minute vibrations. Different icing morphologies have different mass, stiffness, and damping characteristics, leading to variations in vibration characteristics. By analyzing the distribution of the target peak group in the velocity dimension of the range-Doppler image, vibration information can be extracted. The centroid frequency reflects the average velocity of the conductor's vibration, while the spectral width reflects the degree of velocity dispersion. For example, frost has weak adhesion and easily generates high-frequency vibrations in light winds, resulting in a wider Doppler spectral width; while rime has strong adhesion, low and stable vibration frequencies, and a narrower Doppler spectral width. This characteristic provides a dynamic criterion for identifying icing morphologies.

[0047] Finally, the amplitude and phase characteristic parameters of the dual-polarized echo are calculated. These parameters are defined as the amplitude ratio and phase difference between the vertically polarized and horizontally polarized echoes. This embodiment uses a dual-polarized antenna to simultaneously transmit and receive vertically and horizontally polarized electromagnetic waves. Different icing morphologies exhibit varying degrees of sensitivity to polarization. For example, rime ice, with its smooth, near-mirror surface, is sensitive to polarization direction, and the amplitude ratio and phase difference between the vertically and horizontally polarized echoes show specific patterns; while hoarfrost, with its rough surface exhibiting diffuse reflection characteristics, shows a significant depolarization effect, with an amplitude ratio approaching 1 and a random phase difference distribution. By extracting the amplitude and phase characteristics, icing morphologies can be further distinguished from the polarization domain.

[0048] Step S302, Icing morphology identification. The radar cross-section relative change rate, echo waveform distortion value, Doppler spectrum statistical features, and amplitude and phase feature parameters of the dual-polarized echo extracted in step S301 are combined into a feature vector, which is input into a support vector machine classifier. The output is the morphology classification result of no icing, frost, hoarfrost, mixed frost, and rain frost.

[0049] Specifically, Support Vector Machine (SVM) is a classification model based on statistical learning theory. Its core idea is to find an optimal hyperplane in a high-dimensional feature space that maximizes the margin between samples of different classes. In this embodiment, a large amount of sample data with different icing morphologies is collected in advance. The four feature parameters mentioned above are calculated as inputs, and the observed icing morphologies are used as labels to train the SVM classifier. During training, the Radial Basis Function (RBF) is used to map low-dimensional features to a high-dimensional space to solve the nonlinear separability problem, and the penalty coefficient and kernel parameters are optimized through grid search.

[0050] In actual operation, the feature vector calculated in real time is input into the trained SVM model. The model calculates the probability or decision value of the feature vector belonging to each morphological category, and outputs the category with the highest probability as the discrimination result. For example, when the input feature vector shows a large RCS change rate, small waveform distortion, narrow Doppler spectral width, and obvious polarization features, the model tends to classify it as "rime"; when the input feature vector shows a small RCS change rate, large waveform distortion, wide Doppler spectral width, and indistinct polarization features, the model tends to classify it as "fog".

[0051] Through feature extraction and SVM classification across the four dimensions described above, this embodiment achieves a precise mapping from electromagnetic echo signals to specific icing morphologies. This multi-dimensional feature fusion-based identification method, compared to single thickness measurement or simple threshold judgment, can more comprehensively reflect the physical nature of icing, significantly improving identification accuracy and robustness in complex environments, and providing reliable data support for subsequent risk level determination.

[0052] Specifically, the input features of the classifier are the four morphological feature parameters mentioned above: radar cross-section change rate. echo waveform distortion Doppler spectral width and polarization characteristics ( , To improve classification accuracy, the two parameters of the polarization feature are combined into one feature, resulting in a 5-dimensional feature vector.

[0053] The training process of the classifier is as follows: First, radar feature data of various types of icing under different meteorological conditions are collected to establish a training sample set. The training sample set contains at least 500 sets of valid icing event data, with no less than 100 sets of samples for each type. Then, the feature data is standardized to eliminate dimensional differences. Finally, grid search and 5-fold cross-validation are used to optimize the kernel function parameters (radial basis function, parameters) of the SVM. and Cross-validation showed that the overall recognition accuracy of the classifier was no less than 90%. After training, the model parameters were stored in the Raspberry Pi for online recognition.

[0054] Example 4: This embodiment is a further refinement of steps S400 to S600 in Embodiment 1. For example... Figure 4 As shown, this embodiment describes in detail the risk level reasoning process based on Bayesian networks, focusing on the engineering significance and probabilistic fusion logic of the risk level classification standard.

[0055] Step S400 involves constructing a Bayesian network model, using multidimensional data of the operating environment as input, to calculate the prior probability distribution of various icing morphologies. Specifically, the Bayesian network model includes environment nodes, morphology nodes, and risk nodes. Environment nodes include temperature, humidity, and wind speed, serving as the root node and inputting the current meteorological data. Morphology nodes, influenced by environment nodes, output the prior probability of various icing morphologies (such as rime ice and hoarfrost). This model is constructed based on historical icing data statistics and physical mechanisms. For example, when the ambient temperature is close to 0°C and the humidity is extremely high, the model calculates that the prior probability of rime ice is significantly higher than that of other morphologies. This step utilizes the causal logic of the meteorological environment to provide background probability support for subsequent judgments.

[0056] Step S500 involves updating the posterior probability of the Bayesian network model using the icing morphology recognition results as observational evidence. In this step, the morphology recognition results output by the SVM classifier in Example 3 are used as strong observational evidence and input into the Bayesian network. The posterior probability of each icing morphology is updated using the Bayesian formula, combining the prior probability with the likelihood of the observational evidence. This fusion mechanism effectively corrects the bias of a single sensor. For example, when environmental conditions favor rime ice (high prior probability), but radar features are closer to hoarfrost (observational evidence), the posterior probability will combine the information from both to provide a compromise probability distribution, avoiding extreme errors from a single judgment.

[0057] Step S600: Based on the posterior probability distribution of various icing morphologies, calculate the posterior probability of different icing risk levels, and take the risk level corresponding to the highest posterior probability as the judgment result. The core of this step lies in the quantitative classification of risk levels. In step 6, the risk levels are divided into three levels according to the ratio of the conductor load increment to the design value: the first level is a low-risk level where the load increment does not exceed 20% of the design value; the second level is a medium-risk level where the load increment is between 20% and 50% of the design value; and the third level is a high-risk level where the load increment exceeds 50% of the design value.

[0058] Specifically, the critical values ​​of 20% and 50% have clear engineering and physical significance. For transmission line conductors, a 20% load increase typically corresponds to the starting point where conductor stress reaches its elastic limit. At this point, icing is relatively light, conductor sag changes are within a safe range, and no immediate emergency measures are required; this is considered a low-risk level. A 50% load increase, however, corresponds to a critical threshold where conductor stress approaches its yield point or the risk of galloping increases sharply. Once this value is exceeded, catastrophic accidents such as line breakage and tower collapse are highly likely, necessitating immediate activation of de-icing or power outage plans; this is considered a high-risk level. The medium-risk level, falling between these two values, indicates that maintenance personnel need to strengthen monitoring and prepare for de-icing.

[0059] The determination of the above risk level probability is based on the following engineering mechanics analysis: according to the design parameters of the LGJ-240 / 30 type conductor (nominal cross-section 275.96mm²). 2 The calculated breaking force is 83.39 kN, with a design safety factor of 2.5. The allowable design stress for the conductor is 121 MPa. (Frost / ice, density 100 kg / m³) 3 The resulting increase in load per unit length is approximately 8% to 12% of the conductor's self-weight, corresponding to a stress increase of approximately 1015 MPa (approximately 8% to 12% of the design value); frost (density 300 kg / m³) 3 This results in a load increase of approximately 25% to 40% of its own weight, corresponding to a stress increase of 3048 MPa (approximately 25% to 40% of the design value); mixed frost (density 600 kg / m³) 3 This results in a load increase of approximately 50% to 80% of its own weight, corresponding to a stress increase of 6097 MPa (approximately 50% to 80% of the design value); rime ice (density 900 kg / m³) 3 This results in a load increase of approximately 80% to 120% of the self-weight, corresponding to a stress increase of 97,145 MPa (approximately 80% to 120% of the design value). The probability distribution of P(R|S) shown in Table 3 is determined based on the above load-stress correspondence, combined with the statistical results of icing morphology. For example, under the rime ice morphology, 80% of historical accidents correspond to the high-risk level, therefore P(R=high|rime ice)=0.80.

[0060] It should be understood that the above threshold settings are scientific values ​​verified through extensive engineering practice. If the threshold is set too low, for example, setting the high-risk level threshold to 10%, although it can greatly reduce the false alarm rate, it will cause the system to frequently trigger high-risk alarms even with slight icing (such as frost), requiring maintenance personnel to frequently dispatch personnel for investigation, resulting in serious waste of human and material resources, creating a "crying wolf" effect, and reducing the reliability of the early warning system. Conversely, if the threshold is set too high, for example, setting the high-risk level threshold to 60%, although it reduces false alarms, it will cause the system to still judge medium or even low risk when icing is already quite severe (such as load increase reaching 55%), missing the best time for de-icing and causing safety accidents. Therefore, this embodiment selects 20% and 50% as the classification boundaries, achieving the best balance between false alarm rate and false alarm rate, ensuring the timeliness and accuracy of the early warning.

[0061] After calculating the posterior probability of each risk level, the system uses the maximum a posteriori probability criterion for final determination. That is, it compares the posterior probability values ​​of low-risk, medium-risk, and high-risk levels, selecting the level with the highest probability as the final risk status of the current line. For example, if the calculated probability of low risk is 0.1, medium-risk probability is 0.3, and high-risk probability is 0.6, then the determination result is high-risk. This determination result combines the dual verification of environmental factors and electromagnetic characteristics, possessing extremely high reliability.

[0062] Specifically, in this embodiment, the Bayesian network structure used includes five nodes: temperature T, humidity H, wind speed V, icing morphology S, and risk level R. The value range of node T is {-15℃, -10℃, -5℃, 0℃}, corresponding to four discrete states. The value range of node H is {low, medium, high}, corresponding to relative humidity of <85%, 85%-95%, and >95%, respectively. The value range of node V is {low, medium, high}, corresponding to wind speeds of <5m / s, 5-10m / s, and >10m / s, respectively. The value range of node S is {no, frost, hoarfrost, mixed frost, rain frost}, corresponding to five morphology recognition results from millimeter-wave radar. The value range of node R is {low, medium, high}, corresponding to three risk levels.

[0063] The conditional dependencies between nodes are as follows: T, H, and V are root nodes with no parent nodes; S's parent nodes are T, H, and V, indicating that the icing pattern is determined by meteorological conditions. S also has an autoregressive structure, meaning that the current icing pattern is affected by the previous period. The system sets a status persistence counter, which only updates the status when the same icing pattern is identified for three consecutive monitoring cycles, thus avoiding misjudgments caused by instantaneous noise; R's parent node is S, indicating that the risk level is determined by the icing pattern.

[0064] The conditional probability table was established based on historical icing event data and expert knowledge. Data collection covered the winter icing period (November to March of the following year) of a provincial power grid from 2015 to 2024, totaling nearly 10 years, and collected 512 valid icing event records. The criteria for determining icing morphology were as follows: Frost ice – white crystals on the conductor surface, loose structure, easily detached; Hoarfrost – translucent or milky white, feather-like or granular, with moderate adhesion; Mixed rime – dense, translucent ice body, elliptical or round, with strong adhesion; Rain rime – transparent or translucent smooth ice shell, streamlined, with extremely strong adhesion. Each record included ambient temperature (accuracy ±0.3℃), relative humidity (accuracy ±2%RH), wind speed (accuracy ±0.1m / s), and a label indicating the icing morphology confirmed by manual inspection. Temperature was divided into four discrete ranges: -15℃, -10℃, -5℃, and 0℃; humidity was divided into three ranges: <85%, 85%-95%, and >95%. The frequency of each icing pattern within each interval was statistically analyzed, smoothed using Laplace (plus-1 smoothing), and then normalized to obtain the conditional probabilities shown in Tables 2 & 3. For intervals with a sample size of less than 5, empirical values ​​from power system icing prevention experts were used for correction, with a correction margin not exceeding ±5%. For example: The conditional probability of P(S=frost|T,H,V) (Table 2): The conditional probability of P(R|S) (Table 3): When determining the risk level, the current temperature, humidity, and wind speed data are first discretized and then input into a Bayesian network to calculate the prior probabilities of various icing patterns. Then, the millimeter-wave radar morphology recognition results... As observational evidence, the posterior probability is updated using Bayes' theorem: in, The likelihood of radar identification is determined by the confusion matrix of millimeter-wave radar morphology identification, reflecting the identification accuracy. S′ represents the probability of identifying all possible icing morphologies. This is the normalization term for the total probability.

[0065] Finally, based on the posterior probability distribution of various icing patterns, the posterior probability of the risk level is calculated: The risk level with the highest posterior probability is taken as the judgment result.

[0066] Example 5: This embodiment provides a line icing early warning system based on millimeter-wave radar and multi-source fusion. This system is the hardware carrier for implementing the methods described in embodiments 1 to 4 above. Figure 1 As shown, the system includes a multi-source sensing module, a data processing and fusion module, a communication module, and a remote terminal.

[0067] Specifically, the multi-source sensing module acts as the system's "tentacles," responsible for sensing changes in the external physical world. This module includes an environmental parameter sensing unit and an electromagnetic echo sensing unit. The environmental parameter sensing unit collects temperature, relative humidity, and wind speed parameters around the transmission line. It is typically deployed at different heights on the transmission line towers to obtain representative micro-meteorological data. The electromagnetic echo sensing unit transmits electromagnetic signals to the transmission line conductors and receives reflected echo signals. Its core component is a millimeter-wave radar sensor, which captures the electromagnetic scattering characteristics of the conductors and icing layers by transmitting frequency-modulated continuous waves and receiving echoes. It should be understood that although this embodiment lists specific sensor types, in actual engineering applications, the multi-source sensing module can also be expanded to include barometric pressure sensors, rainfall sensors, or conductor tilt sensors, as long as they can provide parameters reflecting the environment or state of icing formation. In this embodiment, the millimeter-wave radar operates in two modes: continuous monitoring mode and triggered monitoring mode. In continuous monitoring mode, the radar continuously transmits and receives signals according to a set period (default 1 second). The transmitted signal is a linear frequency modulated continuous wave with a frequency sweep period of 50μs, which is suitable for close monitoring during periods of high icing incidence. In triggered monitoring mode, the radar only starts collecting data when environmental parameters change beyond a threshold. During sleep mode, the power consumption drops to below 1W, which is suitable for remote sites powered by batteries.

[0068] The data processing and fusion module is the "brain" of the system, with an edge computing terminal as the core control unit, integrating a multi-channel data acquisition unit and a signal intelligent processing unit. This module is used to realize parameter acquisition, signal preprocessing, feature analysis, morphological discrimination, and probabilistic fusion calculation of the multi-source sensing module, and output the icing risk level judgment result. The edge computing terminal has powerful local computing capabilities, enabling it to complete complex radar signal processing (such as FFT transformation and CFAR detection) and Bayesian network inference in real time in the field, without uploading raw data to the cloud, thus greatly reducing the dependence on network bandwidth and improving the system's response speed. In actual deployment, the edge computing terminal can use an industrial-grade Raspberry Pi, NVIDIA Jetson series development board, or ARM architecture embedded industrial control computer, and be packaged in a chassis with IP67 protection rating to ensure long-term stable operation in harsh field environments. In this embodiment, the data processing and fusion module is installed in a protective box at the bottom of the tower, including a Raspberry Pi 4B development board, a multi-channel data acquisition card, a power management module, and a communication interface. The Raspberry Pi 4B uses a Broadcom BCM2711 quad-core Cortex-A72 processor with a clock speed of 1.5GHz, equipped with 4GB of LPDDR4 memory and 32GB of SD card storage, and runs the Raspberry Pi iOS operating system. The data acquisition card provides 8 analog inputs, 4 digital inputs, and 2 SPI interfaces, allowing simultaneous connection to various types of sensors. The power management module converts the DC power supplied by the solar panels or wind turbines on the pole into the various voltage levels required by the system, including 5V power to the Raspberry Pi and 12V power to the millimeter-wave radar and BeiDou module.

[0069] The communication module is the system's "mouthpiece," used to remotely transmit icing risk level assessment results and location parameters. Considering that power transmission lines often traverse remote areas such as mountains, forests, and uninhabited regions, public mobile communication signals (such as 4G / 5G) often have limited coverage or are extremely unstable. If public network communication is used, and the line happens to be in a communication blind spot when icing occurs, the warning information cannot be sent, leading to serious consequences. Therefore, in this embodiment, the communication module is preferably a satellite communication module. The satellite communication module includes a satellite short message transmission unit and a satellite positioning unit.

[0070] The satellite positioning unit is used to determine the location of iced transmission lines. Since transmission lines typically stretch for hundreds of kilometers, accurate location information is crucial for maintenance personnel to quickly locate fault points. The satellite positioning unit receives satellite navigation signals (such as BeiDou and GPS), calculates the latitude and longitude coordinates of the current monitoring device, and updates it in real time. The satellite short message transmission unit is used to send the icing risk level assessment results and location parameter information to a remote terminal. Unlike traditional satellite phones or satellite broadband, short message communication has advantages such as small terminal size, low power consumption, and low communication costs, making it very suitable for transmitting early warning information with small data volumes but high real-time requirements. In this embodiment, after receiving a high-risk alarm signal from the edge computing terminal, the satellite short message transmission unit immediately packages a data packet containing the risk level, icing pattern, latitude and longitude coordinates, and timestamp into a short message format, sends it to the ground station via a satellite link, and then forwards it to the remote terminal. This combination of satellite positioning and short message communication completely solves the communication blind spot problem in remote transmission line monitoring, ensuring the flawless transmission of early warning information.

[0071] The remote terminal serves as the system's "command center," comprising a data receiving server and a monitoring terminal. The data receiving server receives data packets transmitted by the communication module, and then decrypts, parses, and stores them. The monitoring terminal receives, stores, parses, visualizes, and triggers alarms on the transmitted data. Maintenance personnel can visually view the real-time location and risk status of each monitoring point through the monitoring terminal's GIS map interface. Once a high-risk alarm is issued at a monitoring point, the monitoring terminal will automatically pop up an alarm window and notify on-duty personnel via SMS, audio-visual alerts, etc., thus achieving a closed-loop process from data collection to alarm triggering.

[0072] Example 6: This embodiment is a further refinement of the specific communication protocol of the communication module and the hardware interface connection relationship of the data processing and fusion module in Embodiment 5.

[0073] In terms of satellite communication, the satellite short message transmission unit uses the BeiDou RDSS system for data transmission, supporting message lengths of 100-1000 Chinese characters and a transmission latency of less than 5 seconds. The BeiDou RDSS system possesses unique short message communication capabilities, independent of terrestrial mobile communication networks, making it particularly suitable for power transmission line monitoring points in areas with poor public network signal coverage, such as high mountains and forests. The message length is set at 100-1000 Chinese characters, a range that meets the service capability limitations of BeiDou civilian cards while also being sufficient to accommodate complete monitoring data frames. Figure 5As shown, the data frame format defined in this embodiment includes a frame header, frame sequence number, device ID, timestamp, latitude and longitude coordinates, risk level, ambient temperature, humidity, wind speed, icing morphology, checksum, and frame trailer. The risk level is encoded using 3 bits of binary code, the icing morphology using 1 byte of code, and the location information retains 6 decimal places of precision. After encoding and compression, the entire data frame is typically between tens and hundreds of bytes in length, well within the 100-1000 Chinese character message carrying range. The technical specification of a transmission latency of less than 5 seconds ensures that when the system determines a high-risk icing state, the alarm information can be quickly sent to the remote terminal via satellite link, meeting the timeliness requirements of power system emergency response. If the transmission latency is too long, for example, exceeding 1 minute, maintenance personnel may miss the optimal time for de-icing, potentially leading to tower collapse and line breakage accidents.

[0074] In terms of hardware interface connectivity, the edge computing terminal communicates with the millimeter-wave radar sensor via a serial peripheral interface, with the temperature and humidity sensors of the environmental parameter sensing unit via an integrated circuit bus interface, with the wind speed sensor of the environmental parameter sensing unit via a universal asynchronous transceiver interface, and with the communication module via a universal serial bus interface. This interface allocation scheme is an optimized configuration based on the data transmission characteristics and real-time requirements of each sensor. Specifically, the millimeter-wave radar sensor generates a large amount of raw echo data during operation, including range and Doppler information. It has a high data throughput and requires high-speed real-time transmission. The serial peripheral interface, as a high-speed full-duplex synchronous communication bus, can achieve a transmission rate of several Mbps to tens of Mbps, meeting the high-speed transmission requirements of radar data and avoiding data backlog. The temperature and humidity sensors typically have low sampling rates, such as once per second to once per minute, and the data volume is extremely small; only temperature and humidity values ​​need to be transmitted. The integrated circuit bus interface only requires two signal lines to connect multiple slave devices, effectively saving pin resources of the edge computing terminal and simplifying circuit design. Wind speed sensors typically output standard serial data streams. Universal Asynchronous Receiver / Transmitter (UART), as a general asynchronous communication protocol, requires no clock line, has simple wiring, and strong compatibility, making it suitable for low-speed point-to-point data transmission like that of wind speed sensors. Beidou communication modules require complex protocol encapsulation and command interaction before sending short messages, and the amount of data transmitted is relatively large. Universal Serial Bus (USB) interfaces provide plug-and-play convenience and high transmission bandwidth, stably supporting configuration writing and data reading for Beidou modules. It should be understood that although this embodiment lists specific interface types, in actual engineering applications, technicians can make adaptive adjustments based on the specific chip selection. For example, the UART can be converted to an RS485 interface to extend the transmission distance, or an analog serial peripheral interface can be used to connect to the radar, as long as reliable data transmission can be achieved.

[0075] Example 7: This embodiment uses an icing monitoring scenario on a 110kV transmission line in a mountainous area as an example to illustrate the specific application of the methods and systems described in Embodiments 1 to 6 above. This application scenario aims to verify the practical effectiveness of the invention under complex meteorological environments and communication blind spots.

[0076] In this mountainous environment, the transmission lines cross high-altitude areas, constantly exposed to low temperatures and high humidity, and suffer from extremely poor public mobile signal coverage. During system deployment, multi-source sensing modules were installed on key towers of the line. Environmental parameter sensing units were installed at different heights on the towers, while electromagnetic echo sensing units were installed on the central platform of the towers, with the beam pointing towards the target conductor. The data processing and fusion module was encapsulated in a protective box, and the communication module adopted the BeiDou satellite communication solution.

[0077] In the early morning of a winter day, the area experienced freezing rain, with the ambient temperature plummeting to -5°C, relative humidity reaching 95%, and wind speed at 2 m / s. The system initiated real-time monitoring procedures: Step S701, Data Acquisition. The environmental parameter sensing unit acquires current ambient temperature, relative humidity, and wind speed data. Simultaneously, the millimeter-wave radar transmits frequency-modulated continuous wave signals towards the target power transmission line and receives reflected echo signals from the conductor and icing layer.

[0078] Step S702, Signal Processing and Feature Extraction. The data processing and fusion module performs sequential processing of signal decoupling, dimension transformation, and target detection on the electromagnetic echo parameters, extracting the range-Doppler image. Multi-dimensional feature parameter sets are analyzed from the image. The calculation results show that: the relative change rate of the radar cross-section reaches +85%, the distortion value of the echo waveform is 0.48, the centroid frequency of the Doppler spectrum is low and the spectral width is narrow, and the amplitude ratio and phase difference of the dual-polarized echo exhibit obvious specular reflection characteristics.

[0079] Step S703, morphological recognition. The above feature parameters are input into a support vector machine classifier. The classifier outputs "rime". This recognition result conforms to the physical laws of rime formation under freezing rain weather, namely, high density, smooth surface, and strong adhesion.

[0080] Step S704, Risk Fusion and Determination. A Bayesian network model is constructed, using current ambient temperature, humidity, and wind speed data as input, to calculate the prior probability distribution of various icing patterns. Since the environmental conditions meet the typical meteorological characteristics of rime ice formation, the model calculates a high prior probability for the rime ice pattern. Subsequently, the "rime ice" identification result obtained in step S703 is used as observational evidence to update the posterior probability of the Bayesian network model. The updated posterior probability further confirms the high confidence level of the rime ice pattern. According to the risk level classification criteria, the conductor load increment corresponding to the rime ice pattern exceeds 50% of the design value, and the system ultimately determines it to be a high-risk level.

[0081] Step S705, Remote Early Warning. The system triggers the alarm mechanism, and the satellite positioning unit obtains the precise latitude and longitude coordinates of the current tower. The satellite short message transmission unit sends a message containing "high risk level," "rime ice morphology," location coordinates, and environmental data to the remote terminal via the BeiDou RDSS system. Since there is no public network signal in this area, BeiDou short message communication plays a crucial role, ensuring that the alarm information is successfully sent within 5 seconds, unaffected by the lack of ground communication facilities.

[0082] Step S706, Visualization. After receiving the alarm message, the data receiving server of the remote terminal parses the data and automatically displays a red warning icon on the GIS map of the monitoring terminal. Maintenance personnel see the flashing red alarm for that line segment on the large screen in the monitoring center and view the specific icing pattern and estimated load increment. They then activate the de-icing plan and dispatch maintenance personnel to the site for verification.

[0083] The implementation of this application scenario verifies that the present invention can effectively integrate the electromagnetic characteristics of millimeter-wave radar with environmental meteorological data to accurately identify icing patterns and determine risk levels. Particularly in communication blind spots, reliable remote early warning was achieved using BeiDou short message communication, providing strong technical support for the safe operation of power transmission lines and demonstrating extremely high practical application value.

[0084] In summary, the technical solution of this embodiment has the following beneficial effects: First, an innovative technical solution is proposed that utilizes millimeter-wave radar to estimate icing morphology characteristics rather than thickness. By analyzing the radar cross-section change rate, echo waveform distortion, Doppler spectrum characteristics, and polarization characteristics of the radar echo signal, different icing morphologies such as rime, hoarfrost, mixed rime, and frost can be effectively identified, providing accurate morphological information for icing risk assessment. This technical solution avoids the dependence of traditional thickness measurement methods on precise geometric models, improving its applicability and reliability in complex electromagnetic environments.

[0085] Second, a multi-sensor data fusion method based on Bayesian networks is proposed, which deeply integrates environmental parameters such as temperature, humidity, and wind speed with millimeter-wave radar morphological features. By establishing a probabilistic relationship model between meteorological conditions and icing morphology, and comprehensively considering the complementarity and uncertainty of multi-source information, intelligent determination of icing risk level is achieved. Compared with single-sensor determination methods, this invention significantly improves the accuracy and stability of risk level determination.

[0086] Third, the use of BeiDou short message communication to remotely transmit monitoring data solves the problem of the lack of conventional communication methods for power transmission lines in remote areas. BeiDou short message communication has advantages such as wide coverage, reliable transmission, and no impact from ground communication infrastructure, making it particularly suitable for monitoring power transmission lines in complex terrains such as mountainous areas, forest areas, and lake areas.

[0087] Fourth, by using Raspberry Pi as an edge computing terminal, the localization of sensor data acquisition, preprocessing, fusion computing, and risk level determination is achieved, reducing the dependence on cloud computing resources and the demand for network bandwidth, and improving the system's autonomy and response speed.

[0088] Fifth, the proposed technical solution has advantages such as low cost, easy deployment, and strong scalability. The sensors are all mature commercial products with controllable costs; the Raspberry Pi and BeiDou short message module are small in size and low in power consumption, facilitating on-site installation and deployment; the system architecture is open, allowing for the addition of other types of sensors as needed, providing excellent scalability.

[0089] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A line icing early warning method based on millimeter-wave radar and multi-source fusion, characterized in that, Includes the following steps: Step 1: Collect multi-dimensional data on the operating environment of the target transmission line, and collect electromagnetic echo parameters of the conductors using millimeter-wave radar; Step 2: Perform signal processing on the electromagnetic echo parameters to extract the range-Doppler image; Step 3: Extract the multi-dimensional feature parameter set representing the icing morphology of the conductor from the distance-Doppler image, input the multi-dimensional feature parameter set into the support vector machine classifier, and output the icing morphology recognition result. Step 4: Construct a Bayesian network model, using multidimensional data from the operating environment as input, to calculate the prior probability distribution of various icing morphologies; Step 5: Use the ice frost pattern recognition results as observation evidence to perform a posterior probability update on the Bayesian network model; Step 6: Based on the posterior probability distribution of various icing patterns, calculate the posterior probability of different icing risk levels, and take the risk level corresponding to the highest posterior probability as the judgment result. Step 7: Transmit the judgment result and the location parameters of the target transmission line to the remote terminal.

2. The line icing early warning method based on millimeter-wave radar and multi-source fusion according to claim 1, characterized in that, In step 1, the multidimensional data includes ambient temperature, ambient relative humidity, and ambient wind speed data collected by temperature sensors, humidity sensors, and wind speed sensors; the method for collecting electromagnetic echo parameters of the conductor by millimeter-wave radar is as follows: the millimeter-wave radar transmits frequency-modulated continuous wave signals toward the target transmission line and receives reflected echo signals from the conductor and the ice layer.

3. The line icing early warning method based on millimeter-wave radar and multi-source fusion according to claim 1, characterized in that, In step 2, signal processing is performed on the electromagnetic echo parameters, specifically including the sequential processing of signal decoupling, dimension transformation, and target detection.

4. The line icing early warning method based on millimeter-wave radar and multi-source fusion according to claim 1, characterized in that, In step 3, the multi-dimensional feature parameter set includes the relative change rate of the millimeter-wave radar cross section, the distortion value of the echo waveform, the statistical features of the Doppler spectrum, and the amplitude and phase feature parameters of the dual-polarized echo. The relative change rate of the radar cross section is the ratio of the measured value of the current time-domain radar cross section to the reference value of the radar cross section under the non-icing state. The distortion value of the echo waveform is the difference between 1 and the cross-correlation coefficient between the current echo waveform and the non-icing reference waveform. The statistical features of the Doppler spectrum are the centroid frequency and spectral width of the target peak group in the velocity dimension. The amplitude and phase feature parameters of the dual-polarized echo are the amplitude ratio and phase difference between the vertically polarized and horizontally polarized echoes.

5. The line icing early warning method based on millimeter-wave radar and multi-source fusion according to claim 4, characterized in that, In step 3, the inputs of the support vector machine classifier are the relative change rate of radar cross section, the distortion value of echo waveform, the statistical features of Doppler spectrum and the amplitude and phase feature parameters of dual-polarized echo, and the output is the morphological classification result of no icing, frost, hoarfrost, mixed frost and rain frost.

6. The line icing early warning method based on millimeter-wave radar and multi-source fusion according to claim 1, characterized in that, In step 6, the risk level is divided into three levels according to the ratio of the conductor load increment to the design value. The first level is a low-risk level where the load increment does not exceed 20% of the design value; the second level is a medium-risk level where the load increment is between 20% and 50% of the design value; and the third level is a high-risk level where the load increment exceeds 50% of the design value.

7. A line icing early warning system based on millimeter-wave radar and multi-source fusion, applicable to the line icing early warning method based on millimeter-wave radar and multi-source fusion as described in any one of claims 1-6, characterized in that, It includes a multi-source sensing module, a data processing and fusion module, a communication module, and a remote terminal; The multi-source sensing module includes an environmental parameter sensing unit and an electromagnetic echo sensing unit. The environmental parameter sensing unit is used to collect temperature, relative humidity, and wind speed parameters around the transmission line. The electromagnetic echo sensing unit is used to transmit electromagnetic signals to the transmission line conductors and receive reflected echo signals. The data processing and fusion module uses an edge computing terminal as the core control unit, integrating a multi-channel data acquisition unit and a signal intelligent processing unit. It is used to realize the parameter acquisition, signal preprocessing, feature analysis, morphology discrimination and probability fusion calculation of the multi-source sensing module, and output the icing risk level judgment result. The communication module is used to remotely transmit the icing risk level determination results and location parameters. The remote terminal includes a data receiving server and a monitoring terminal, which realizes the reception, storage, parsing, visualization, and alarm triggering of transmitted data.

8. The line icing early warning system based on millimeter-wave radar and multi-source fusion according to claim 7, characterized in that, The communication module is a satellite communication module, which includes a satellite short message transmission unit and a satellite positioning unit. The satellite positioning unit is used to determine the location of the icing transmission line, and the satellite short message transmission unit is used to send the icing risk level determination result and location parameter information to the remote terminal.

9. The line icing early warning system based on millimeter-wave radar and multi-source fusion according to claim 8, characterized in that, The satellite short message transmission unit is based on the BeiDou RDSS system to realize data transmission, supports message length of 100-1000 Chinese characters, and has a transmission delay of less than 5 seconds.

10. The line icing early warning system based on millimeter-wave radar and multi-source fusion according to claim 7, characterized in that, The edge computing terminal communicates with the millimeter-wave radar sensor through a serial peripheral interface, with the temperature and humidity sensors of the environmental parameter sensing unit through an integrated circuit bus interface, with the wind speed sensor of the environmental parameter sensing unit through a universal asynchronous transceiver interface, and with the communication module through a universal serial bus interface.

Images