Power transmission tower attitude monitoring and vibration intrusion identification method based on multi-source data fusion

Abstract

The invention discloses a power transmission tower attitude monitoring and vibration intrusion identification method based on multi-source data fusion, and the method comprises the steps: carrying out the deep fusion of GPS displacement data and accelerometer inclination angle data, employing a dynamic weight fusion mechanism: adjusting a fusion coefficient alpha in real time according to the GPS signal quality, guaranteeing the accuracy and stability of data fusion, and at the same time, carrying out the real-time adjustment of the fusion coefficient alpha, and carrying out the recognition of the vibration intrusion of a power transmission tower. And continuously optimizing the error propagation model by taking the final inclination angle theta final of the power transmission tower as a state variable, updating the attitude estimation of the power transmission tower, and realizing + / -0.01-degree high-precision attitude monitoring. According to the invention, the system can output a high-precision three-dimensional attitude angle in real time by combining the rapid dynamic response of the accelerometer and the long-term stability of the GPS. Besides, an LCD-MPE (feature extraction technology) is introduced, lightweight neural network classification is combined, vibration signal sampling, noise reduction and abnormal signal segment identification, analysis and extraction are carried out on the transmission tower through an MEMS accelerometer, and second-level detection and behavior type discrimination of invasion behaviors around the transmission tower are realized. The core technical barrier from'inability perception 'to'accurate identification' is overcome, and subversive upgrading of power facility safety protection from'passive disposal 'to'active defense' is promoted.

Description

Technical Field

[0001] This invention belongs to the field of power transmission tower safety monitoring technology, and more specifically, it relates to a method for power transmission tower attitude monitoring and vibration intrusion identification based on multi-source data fusion. Background Technology

[0002] With the rapid expansion of power grid construction and the continuous increase in the scale of transmission towers, the demand for real-time, high-precision monitoring of the safety of high-voltage transmission towers is becoming increasingly urgent. Currently, traditional transmission tower safety monitoring mainly relies on a combination of high-precision GPS positioning and regular manual inspections, but both methods have significant shortcomings:

[0003] First, while existing high-precision GPS positioning can detect the horizontal displacement of power transmission towers, it can only sense large-scale positional changes, such as overall tilting or movement, and cannot accurately assess subtle changes in the tower's attitude. Furthermore, limited by sampling frequency and data transmission delays, high-precision GPS positioning has almost no ability to monitor the instantaneous dynamic behavior of power transmission towers, such as structural vibrations caused by strong winds or instantaneous deformations caused by mechanical impacts, making it difficult to provide timely warnings of sudden risks.

[0004] Secondly, manual inspections rely on visual observation and simple instrument measurements, which are inefficient and costly, especially in remote mountainous areas. Furthermore, they are difficult to implement during dangerous weather conditions such as typhoons and thunderstorms, posing safety risks and preventing 24 / 7 monitoring. In addition, inspection cycles are typically monthly, leading to significant delays in the discovery of safety hazards.

[0005] Third, traditional monitoring methods are almost ineffective in monitoring intrusion around power transmission towers. Mechanical collisions, such as those caused by construction vehicles or the instantaneous vibration signals generated by illegal blasting, cannot be captured. Illegal climbing, such as the theft of tower materials or human-caused damage, usually occurs at night or in areas without monitoring, and manual inspections can usually only find traces afterward.

[0006] Chinese invention patent application CN119104057A, published on December 10, 2024, discloses a method, device, equipment, and storage medium for monitoring the attitude of a power transmission tower. This method overcomes the significant errors inherent in directly combining BeiDou positioning results with inertial navigation positioning results for attitude determination, and the susceptibility of inertial sensors to accumulated errors over long periods of operation, which reduces the accuracy of tower attitude monitoring. The method includes determining the antenna pose information of each positioning antenna based on the baseline vector of at least one positioning antenna relative to a reference antenna. The antenna pose information includes the antenna coordinates of the positioning antenna, and the heading and pitch angles determined based on these coordinates. Based on the antenna pose information, the method determines the target antenna attitude information and attitude correction information. It then acquires the inertial navigation positioning information and corrects it using the attitude correction information to obtain corrected inertial navigation positioning information. Finally, based on the target antenna attitude information and the corrected inertial navigation positioning information, the method determines the tower attitude information of the target power transmission tower. The technical solution of this invention reduces the impact of observation conditions on the observation by placing multiple positioning antennas for observation and solving for the target antenna attitude information. It also reduces the amplification of carrier phase double-difference errors and utilizes attitude correction information to reduce the cumulative errors of inertial sensor devices over long-term operation, thereby improving the accuracy of transmission tower attitude monitoring. However, this invention requires setting up multiple antennas and performing extensive calculations in hardware, and it only monitors the attitude of transmission towers, lacking vibration intrusion detection functionality. Summary of the Invention

[0007] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a method for monitoring the attitude of transmission towers and identifying vibration intrusion based on multi-source data fusion. By combining the rapid dynamic response of accelerometers and the long-term stability of GPS positioning, it can output high-precision three-dimensional attitude angles (pitch angle, roll angle, and yaw angle) in real time, while simultaneously identifying vibration intrusion.

[0008] To achieve the above-mentioned objectives, the present invention provides a method for monitoring the attitude of transmission towers and identifying vibration intrusion based on multi-source data fusion, characterized by comprising the following steps:

[0009] (1) GPS positioning + accelerometer data fusion outputs continuous and reliable transmission tower attitude information.

[0010] 1.1) Data Synchronization

[0011] GPS second pulse calibration: The 1PPS (second pulse) signal output by the GPS module is used to synchronize the accelerometer acquisition clock to ensure that the accelerometer timestamp is consistent with the GPS system time.

[0012] TCXO Backup Clock: If the GPS signal is briefly interrupted, the accelerometer acquisition clock switches to a highly stable temperature-compensated crystal oscillator (TCXO) clock to maintain the stability of the time base in the short term and avoid data drift.

[0013] Synchronization guarantee mechanism: Regardless of changes in GPS signal status, accelerometer data always relies on stable time base sampling to ensure that subsequent data fusion computing can process time-aligned data;

[0014] 1.2) Posture fusion

[0015] Weighted fusion of GPS measurement data and accelerometer data:

[0016] Prediction phase: Short-term attitude calculation of the transmission tower is performed based on accelerometer data. By removing the gravity component and integrating the acceleration, the velocity and displacement changes are obtained, and the tower tilt angle θ_acc, pitch angle and roll angle are estimated in real time.

[0017] Correction Phase: Periodically introduce tower displacement and direction information provided by the GPS module to correct accumulated errors during acceleration calculation.

[0018] First, calculate the tilt angle θ_GPS based on the GPS displacement data of the transmission tower:

[0019]

[0020] Where Δx and Δy represent the horizontal displacement increments in the x and y directions obtained by the GPS module from the transmission tower, respectively, and H is the height of the transmission tower;

[0021] Then, calculate the final transmission tower tilt angle θ_final:

[0022] θ_final=α×θ_GPS+(1-α)×θ_acc

[0023] Where α is a dynamically adjusted fusion coefficient, which is adjusted in real time according to the quality of the GPS signal;

[0024] Continuous iterative prediction and correction, continuously outputting the final transmission tower tilt angle θ_final;

[0025] 1.3) Attitude angle output

[0026] First, the yaw angle is estimated by combining the GPS travel direction. Then, based on the attitude calculation characteristics, the final tower tilt angle θ_final is used as the state variable to continuously optimize the error propagation model, update the tower attitude estimation, and output the corrected three-dimensional attitude angles of the tower in real time: pitch angle, roll angle and yaw angle.

[0027] (2) Intrusion behavior detection and type identification

[0028] 2.1) High-sensitivity vibration sensing

[0029] The MEMS accelerometer samples the vibration signal of the transmission tower at a sampling rate of 125Hz to obtain the detection signal. The MEMS accelerometer adopts triaxial synchronous measurement and adaptive range switching, and has a sensitivity of 0.001g.

[0030] 2.2) Signal Denoising

[0031] First, time-frequency analysis is performed on the acquired detection signal, the background noise is sampled and its energy spectrum is analyzed. Then, in the frequency domain, the background noise in the detection signal is removed by spectral decomposition based on the background noise energy spectrum. Finally, the background noise-removed detection signal is converted to the time domain to obtain the noise-reduced detection signal.

[0032] 2.3) Abnormal signal segment identification

[0033] First, in the time domain, a time window is opened for the noise-reduced detection signal, and the short-term average and long-term average of each time window are calculated. Appropriate thresholds are set according to the field experimental data for discrimination. If all thresholds are exceeded, it is considered an abnormal signal segment.

[0034] Secondly, in the frequency domain, a Hanning window is opened for the noise-reduced detection signal, the power spectrum of each Hanning window data segment is calculated, and an appropriate threshold is set according to the field experimental data for discrimination. If it exceeds the threshold, it is considered an abnormal signal segment.

[0035] Finally, segments that are considered abnormal signals in both the time and frequency domains are considered abnormal signal segments in the detection signal.

[0036] 2.4) Feature Analysis and Extraction of Abnormal Signal Segments

[0037] First, each identified anomalous signal segment is extracted separately, and local feature scale decomposition is performed on each extracted anomalous signal segment: the anomalous signal segment x(t) is split into two parts: the feature mode term and the residual term.

[0038]

[0039] Among them, the characteristic modality term includes n modes c i (t), r n (t) represents the residual term;

[0040] Then, each mode is encoded: using the multi-scale permutation entropy (MPE) analysis technique in machine learning, the dimensionality of each feature mode is reduced to highlight the vibration features, so as to facilitate automatic category analysis and obtain an m-dimensional feature vector. After mode encoding, the m-dimensional feature vector is arranged in rows to obtain an m×n MPE feature matrix.

[0041] 2.5) Classification of Intrusion Types

[0042] A lightweight neural network model was built and optimized based on the ResNet-18 architecture. By inputting the MPE feature matrix, the type of intrusive behavior can be obtained.

[0043] The objective of this invention is achieved as follows:

[0044] Traditional GPS monitoring can capture the macroscopic horizontal displacement of transmission towers, such as overall settlement, but it cannot accurately detect subtle changes in three-dimensional attitude, such as tower torsion or crossarm deflection. This invention, a method for transmission tower attitude monitoring and vibration intrusion identification based on multi-source data fusion, deeply fuses GPS displacement data and accelerometer tilt data, employing a dynamic weighted fusion mechanism: the fusion coefficient α is adjusted in real time according to GPS signal quality to ensure the accuracy and stability of data fusion. Simultaneously, using the final transmission tower tilt angle θ_final as the state variable, the error propagation model is continuously optimized, and the transmission tower attitude estimation is updated, achieving high-precision attitude monitoring of ±0.01°, comprehensively covering the displacement and attitude changes of transmission towers. This invention combines the rapid dynamic response of accelerometers with the long-term stability of GPS, enabling the system to output high-precision three-dimensional attitude angles in real time.

[0045] Furthermore, traditional methods for monitoring transmission towers (such as GPS positioning and manual inspection) completely lack the ability to detect intrusions in the surrounding area in real time, and their technological limitations lead to serious safety hazards. This invention breaks through the traditional technical framework by introducing LCD-MPE (Local Feature Decomposition-Multi-Scale Permutation Entropy) feature extraction technology, combined with lightweight neural network classification, and using MEMS accelerometers to sample, denoise, identify, analyze, and extract abnormal signal segments from transmission towers. This achieves second-level detection and behavior type discrimination of intrusions around transmission towers, overcoming the core technical barrier from "no ability to perceive" to "accurate identification," and promoting a disruptive upgrade of power facility security protection from "passive handling" to "active defense." Attached Figure Description

[0046] Figure 1 This is a flowchart of a specific implementation of the method for monitoring the attitude of transmission towers and identifying vibration intrusion based on multi-source data fusion according to the present invention;

[0047] Figure 2 This is a flowchart illustrating the continuous and reliable output of transmission tower attitude information through the fusion of GPS positioning and accelerometer data.

[0048] Figure 3 This is a flowchart of intrusion behavior detection and type identification;

[0049] Figure 4These are time-frequency analysis (TFA) comparison charts of the detected signal before and after frequency domain noise reduction, where (a) is before noise reduction and (b) is after noise reduction:

[0050] Figure 5 These are waveforms of the detected signal before and after time-domain noise reduction;

[0051] Figure 6 This is a schematic diagram of time-domain STA / LTA detection signal identification;

[0052] Figure 7 This is a schematic diagram of signal identification based on power spectrum detection.

[0053] Figure 8 It is a waveform diagram that identifies an abnormal signal segment;

[0054] Figure 9 These are the six modal diagrams separated from the abnormal signal segment;

[0055] Figure 10 This is a waveform diagram of the detection signal for mechanical impact on the tower base;

[0056] Figure 11 This is a waveform diagram of the detection signal for illegal climbing. Detailed Implementation

[0057] The specific embodiments of the present invention will now be described with reference to the accompanying drawings to enable those skilled in the art to better understand the invention. It should be particularly noted that in the following description, detailed descriptions of known functions and designs that might obscure the main content of the invention will be omitted here.

[0058] To overcome technical bottlenecks and solve the aforementioned problems, this invention proposes a method for monitoring the attitude and identifying vibration intrusion of transmission towers based on multi-source data fusion. Through deep fusion of GPS displacement data and triaxial accelerometer attitude data, it achieves comprehensive monitoring of the tower's "macroscopic displacement, microscopic attitude, and external intrusion." Simultaneously, through intelligent algorithms, the monitoring frequency is increased from "manual monthly" to "millisecond," completely resolving the perception blind spots and response lag issues of traditional methods. Specifically, in this embodiment, as... Figure 1 As shown, the transmission tower attitude monitoring and vibration intrusion identification method based on multi-source data fusion of this invention includes two parts: intrusion behavior detection and type identification technology, specifically:

[0059] Step S1: GPS positioning + accelerometer data fusion outputs continuous and reliable transmission tower attitude information.

[0060] To overcome the limitations of a single sensor, this invention employs data fusion technology combining GPS displacement data and accelerometer data. By combining the rapid dynamic response of the accelerometer with the long-term stability of GPS, the system can output high-precision three-dimensional attitude angles (pitch, roll, and yaw angles) in real time. This includes data synchronization, attitude fusion, dynamic weight adjustment, and attitude angle output mechanisms. Figure 2 As shown, it specifically includes:

[0061] Step S1.1: Data Synchronization

[0062] Data fusion requires a high degree of time synchronization. This invention achieves precise alignment between GPS displacement data and accelerometer data through the following method:

[0063] GPS second pulse calibration: The 1PPS (second pulse) signal output by the GPS module is used to synchronize the accelerometer acquisition clock to ensure that the accelerometer timestamp is consistent with the GPS system time.

[0064] TCXO Backup Clock: If the GPS signal is briefly interrupted, the accelerometer acquisition clock switches to a highly stable temperature-compensated crystal oscillator (TCXO) clock to maintain the stability of the time base in the short term and avoid data drift.

[0065] Synchronization guarantee mechanism: Regardless of changes in GPS signal status, accelerometer data always relies on stable time base sampling to ensure that subsequent data fusion computing can process time-aligned data.

[0066] Step S1.2: Pose Fusion

[0067] Weighted fusion of GPS measurement data and accelerometer data:

[0068] Prediction phase: Short-term attitude estimation of transmission towers is performed based on accelerometer data. By removing the gravity component and integrating the acceleration, the velocity and displacement changes are obtained. The tower tilt angle θ_acc, pitch angle and roll angle are estimated in real time to achieve high-frequency attitude updates.

[0069] Correction Phase: Periodically introduce tower displacement and direction information provided by the GPS module to correct accumulated errors during acceleration calculation.

[0070] First, calculate the tilt angle θ_GPS based on the GPS displacement data of the transmission tower:

[0071]

[0072] Where Δx and Δy represent the horizontal displacement increments in the x and y directions obtained by the GPS module from the transmission tower, respectively, and H is the height of the transmission tower.

[0073] This method calculates the attitude angle change by the ratio of the horizontal displacement change in the x and y directions to the height, and is suitable for approximate scenarios with small angle changes.

[0074] Then, the final transmission tower tilt angle θ_final is calculated.

[0075] θ_final=α×θ_GPS+(1-α)×θ_acc

[0076] Wherein, α is a dynamically adjusted fusion coefficient (typically between 0.6 and 0.8), adjusted in real time according to the GPS signal quality. Since the reliability of sensor data varies under different GPS signal quality conditions, this embodiment designs an adaptive weight adjustment mechanism to dynamically adjust the fusion coefficient α, i.e.:

[0077] When GPS signal is good: When a strong GPS signal and high positioning accuracy are detected, the weight of GPS in the fusion process is increased, that is, the fusion coefficient α is increased, to accelerate attitude correction and improve overall accuracy.

[0078] When GPS signals are limited: In scenarios where GPS signals are weak or lost, the GPS weight is reduced, i.e., the fusion coefficient α is decreased. This relies primarily on the calculation results from the accelerometer, avoiding interference introduced by GPS displacement data with large errors. This dynamic adjustment of the fusion coefficient α allows the system to maintain continuous and smooth attitude estimation in tunnels and obstructed environments, and enables rapid calibration after GPS signals are restored.

[0079] Continuous iterative prediction and correction continuously outputs the final transmission tower tilt angle θ_final, thus maintaining both dynamic response and long-term accuracy.

[0080] Step S1.3: Attitude Angle Output

[0081] First, the yaw angle is estimated by combining the GPS travel direction. Then, based on the attitude calculation characteristics, the final tower tilt angle θ_final is used as the state variable to continuously optimize the error propagation model, update the tower attitude estimation, and output the corrected three-dimensional attitude angles of the tower in real time: pitch angle, roll angle, and yaw angle.

[0082] The fusion algorithm outputs three-dimensional attitude angles in real time, including:

[0083] Pitch: Describes the device's rotation (tilt forward or backward) around its horizontal axis;

[0084] Roll angle: describes the rotation (left or right tilt) of the equipment about its longitudinal axis;

[0085] Yaw: Describes the rotation (heading angle) of the equipment about its vertical axis.

[0086] Attitude angles are output in degrees, and a reference zero point can be defined as needed (such as initial stationary state or true north). Pitch and roll angles are mainly calculated using accelerometers, while yaw angles are estimated and calibrated in conjunction with the GPS direction of travel.

[0087] Step S2: Intrusion Detection and Type Recognition

[0088] Through breakthroughs in three levels of technology—vibration sensing, intelligent feature extraction, and machine learning classification—accurate detection and type identification of intrusive behaviors are achieved, such as... Figure 3 As shown, it specifically includes:

[0089] Step S2.1: High-sensitivity vibration sensing (hardware layer)

[0090] The MEMS accelerometer samples the vibration signal of the transmission tower at a sampling rate of 125Hz (supporting the 0.1-500Hz frequency band, covering the full frequency band characteristics of intrusion behavior) to obtain the detection signal. The MEMS accelerometer adopts triaxial synchronous measurement and adaptive range switching, and has a sensitivity of 0.001g.

[0091] Step S2.2: Signal noise reduction

[0092] When large vibration sources occur around transmission towers, such as impact drills, rotary drilling rigs, or heavy vehicles passing by, the received vibration signal (detection signal) will contain both abnormal signals and background noise signals. Using spectral decomposition technology, the background noise component of the detection signal is removed, leaving only the abnormal signal. This effectively improves the signal-to-noise ratio and the accuracy of intrusion detection and type identification. Specifically:

[0093] First, time-frequency analysis is performed on the acquired detection signal, the background noise is sampled and its energy spectrum is analyzed. Then, in the frequency domain, the background noise in the detection signal is removed using spectral decomposition technology based on the background noise energy spectrum. Finally, the background noise-removed detection signal is converted to the time domain to obtain the noise-reduced detection signal.

[0094] The time-frequency analysis comparison diagram of the detected signal before and after frequency domain noise reduction is shown below. Figure 4 As shown, from Figure 4 It can be seen that the noise frequency is significantly reduced after the detection signal is denoised.

[0095] The comparison graph of the detected signal after noise reduction (converted from the frequency domain back to the time domain) and the detected signal before noise reduction is shown below. Figure 5 As shown, from Figure 5 It can be seen that the noise points are significantly reduced after the detection signal is denoised.

[0096] Step S2.3: Abnormal signal segment identification

[0097] Detecting intrusion requires identifying abnormal signal segments within continuously monitored signals. This invention performs this identification in both the time and frequency domains, with the two analysis methods mutually verifying each other to ensure accuracy. In the time domain, we select the STA / LTA value as the discrimination criterion. The specific steps are as follows:

[0098] First, in the time domain, a time window is opened for the denoised detection signal, and the short-time average and long-time average of each time window are calculated. Appropriate thresholds are set based on field experimental data for discrimination; if all thresholds are exceeded, it is considered an abnormal signal segment. In this embodiment, the detection signal identification based on time-domain STA / LTA is as follows: Figure 6 As shown, when an abnormal signal segment occurs, the STA / LTA value increases significantly, and the abnormal signal segment can be identified based on the threshold.

[0099] Secondly, in the frequency domain, a Hanning window is opened for the denoised detection signal, and the power spectrum of each Hanning window data segment is calculated. An appropriate threshold is set based on field experimental data for discrimination; if the threshold is exceeded, it is considered an abnormal signal segment. In this embodiment, power spectrum-based detection signal identification is used for... Figure 7 As shown, the power spectrum increases significantly when an abnormal signal segment occurs, and the abnormal signal segment can be identified based on the threshold.

[0100] Finally, segments that are considered abnormal in both the time and frequency domains are identified as abnormal signal segments in the detection signal.

[0101] Step S2.4: Feature Analysis and Extraction of Abnormal Signal Segments

[0102] The above processing and analysis steps are sufficient to detect intrusion, but this is not enough. Besides intrusion, many other factors can cause signal anomalies, such as construction, collisions, and heavy vehicle passage. If these are not differentiated and treated as a single event, a large number of false alarms will inevitably occur. Therefore, it is necessary to perform feature analysis on all detected abnormal signal segments, extracting the characteristic parameters of each abnormal signal to determine its source type. The specific steps are as follows:

[0103] First, each identified abnormal signal segment is extracted individually. In this embodiment, the extracted abnormal signal segments are as follows: Figure 8 As shown.

[0104] For each extracted anomalous signal segment, a Local Characteristics-Scale Decomposition (LCD) is performed: the anomalous signal segment x(t) is split into two parts: the characteristic mode term and the residual term.

[0105]

[0106] Among them, the characteristic modality term includes n modes c i (t), r n (t) represents the residual term.

[0107] In this embodiment, as Figure 9 As shown, the abnormal signal segment is divided into 6 modal diagrams, namely Model1 to Model6.

[0108] Then, each mode is encoded: using the multi-scale permutation entropy (MPE) analysis technique in machine learning, the dimensionality of each feature mode is reduced to highlight the vibration features, so as to facilitate automatic category analysis and obtain an m-dimensional feature vector. After mode encoding, the m-dimensional feature vector is arranged in rows to obtain an m×n MPE feature matrix.

[0109] Step S2.5: Intrusion Type Classification

[0110] A lightweight neural network model was built and optimized based on the ResNet-18 architecture. By inputting the MPE feature matrix, the types of intrusion behaviors were obtained. Table 1 shows the seven types of intrusion behaviors.

[0111]

[0112]

[0113] Table 1

[0114] This invention can also be extended to similar tall structures, such as communication towers and wind turbines, for attitude monitoring and intrusion detection. Furthermore, mode decomposition can be replaced with other mode decomposition methods (such as EMD) or localization algorithms (such as beamforming).

[0115] Specific examples

[0116] I. Transmission tower attitude monitoring test:

[0117] Test 1: Static monitoring under clear weather conditions

[0118] Input data:

[0119] GPS displacement data: Δx = 1mm (θ_GPS = 0.0011°), SNR = 45dB, HDOP = 1.2, number of satellites = 10;

[0120] Accelerometer: Transmission tower tilt angle θ_acc=0.0015°, zero drift error 0.0008°.

[0121] Parameter calculation:

[0122] α = 0.8 (high confidence level for GPS).

[0123] θ_final=0.8×0.0011°+0.2×0.0015°=0.00118°.

[0124] Results: The tilt angle error after fusion is <±0.0003°, which is better than that of a single sensor.

[0125] Test 2: Dynamic monitoring under heavy rain conditions

[0126] Input data:

[0127] GPS: Signal lost (α = 0.6).

[0128] Accelerometer: detected instantaneous impact vibration (amplitude 0.005g, frequency 50Hz), and the tower tilt angle θ_acc = 0.003°.

[0129] Output: θ_final = 0.6 × 0° (GPS invalid) + 0.4 × 0.003° = 0.0012° (smoothed based on historical data).

[0130] II. Vibration Intrusion Test:

[0131] Test 1: Mechanical impact on the tower base

[0132] Simulating external damage, the main impact frequency is approximately 2Hz, and the average amplitude is approximately 0.4g. The impact signal captured by the system is the detection signal. Figure 10 As shown, after noise reduction and abnormal signal segment identification, the detected signal undergoes feature analysis and extraction to obtain the MPE feature matrix. The lightweight neural network model identifies it as "mechanical impact" and sends alarm information to the background, while simultaneously pushing location information.

[0133] Test 2: Illegal Climbing

[0134] Testers simulated a thief climbing a steel tower, with a vibration frequency of 1.2Hz and an amplitude of 0.03g. The climbing signal captured by the system is the detection signal. Figure 11 As shown, after noise reduction and abnormal signal segment identification, the detected signal undergoes feature analysis and extraction to obtain the MPE feature matrix. The lightweight neural network model identifies it as "illegal climbing," sends a warning message to the background, and simultaneously pushes location information.

[0135] Although the illustrative specific embodiments of the present invention have been described above to enable those skilled in the art to understand the invention, it should be understood that the invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the invention as defined and determined by the appended claims, and all inventions utilizing the concept of the present invention are protected.

Claims

1. A method for monitoring the attitude of transmission towers and identifying vibration intrusion based on multi-source data fusion, characterized by comprising the following steps: (1) GPS positioning + accelerometer data fusion outputs continuous and reliable transmission tower attitude information. 1.1) Data Synchronization GPS second pulse calibration: The 1PPS (second pulse) signal output by the GPS module is used to synchronize the accelerometer acquisition clock to ensure that the accelerometer timestamp is consistent with the GPS system time. TCXO Backup Clock: If the GPS signal is briefly interrupted, the accelerometer acquisition clock switches to a highly stable temperature-compensated crystal oscillator (TCXO) clock to maintain the stability of the time base in the short term and avoid data drift. Synchronization guarantee mechanism: Regardless of changes in GPS signal status, accelerometer data always relies on stable time base sampling to ensure that subsequent data fusion computing can process time-aligned data; 1.2) Posture fusion Weighted fusion of GPS measurement data and accelerometer data: Prediction phase: Short-term attitude calculation of the transmission tower is performed based on accelerometer data. By removing the gravity component and integrating the acceleration, the velocity and displacement changes are obtained, and the tower tilt angle θ_acc, pitch angle and roll angle are estimated in real time. Correction Phase: Periodically introduce tower displacement and direction information provided by the GPS module to correct accumulated errors during acceleration calculation. First, calculate the tilt angle θ_GPS based on the GPS displacement data of the transmission tower: in, Δx and Δy represent the horizontal displacement increments in the x and y directions obtained by the GPS module from the transmission tower, respectively, and H is the height of the transmission tower; Then, calculate the final transmission tower tilt angle θ_final: θ_final=α×θ_GPS+(1-α)×θ_acc Where α is a dynamically adjusted fusion coefficient, which is adjusted in real time according to the quality of the GPS signal; Continuous iterative prediction and correction, continuously outputting the final transmission tower tilt angle θ_final; 1.3) Attitude angle output First, the yaw angle is estimated by combining the GPS travel direction. Then, based on the attitude calculation characteristics, the final tower tilt angle θ_final is used as the state variable to continuously optimize the error propagation model, update the tower attitude estimation, and output the corrected three-dimensional attitude angles of the tower in real time: pitch angle, roll angle and yaw angle. (2) Intrusion behavior detection and type identification 2.1) High-sensitivity vibration sensing The MEMS accelerometer samples the vibration signal of the transmission tower at a sampling rate of 125Hz to obtain the detection signal. The MEMS accelerometer adopts triaxial synchronous measurement and adaptive range switching, and has a sensitivity of 0.001g. 2.2) Signal Denoising First, time-frequency analysis is performed on the acquired detection signal, the background noise is sampled and its energy spectrum is analyzed. Then, in the frequency domain, the background noise in the detection signal is removed by spectral decomposition based on the background noise energy spectrum. Finally, the background noise-removed detection signal is converted to the time domain to obtain the noise-reduced detection signal. 2.3) Abnormal signal segment identification First, in the time domain, a time window is opened for the noise-reduced detection signal, and the short-term average and long-term average of each time window are calculated. Appropriate thresholds are set according to the field experimental data for discrimination. If all thresholds are exceeded, it is considered an abnormal signal segment. Secondly, in the frequency domain, a Hanning window is opened for the noise-reduced detection signal, the power spectrum of each Hanning window data segment is calculated, and an appropriate threshold is set according to the field experimental data for discrimination. If it exceeds the threshold, it is considered an abnormal signal segment. Finally, segments that are considered abnormal signals in both the time and frequency domains are considered abnormal signal segments in the detection signal. 2.4) Feature Analysis and Extraction of Abnormal Signal Segments First, each identified anomalous signal segment is extracted separately, and local feature scale decomposition is performed on each extracted anomalous signal segment: the anomalous signal segment x(t) is split into two parts: the feature mode term and the residual term. Among them, the characteristic modality term includes n modes c i (t), r n (t) represents the residual term; Then, each mode is encoded: using the multi-scale permutation entropy (MPE) analysis technique in machine learning, the dimensionality of each feature mode is reduced to highlight the vibration features, so as to facilitate automatic category analysis and obtain an m-dimensional feature vector. After mode encoding, the m-dimensional feature vector is arranged in rows to obtain an m×n MPE feature matrix. 2.5) Classification of Intrusion Types A lightweight neural network model was built and optimized based on the ResNet-18 architecture. By inputting the MPE feature matrix, the type of intrusive behavior can be obtained.

2. The method for transmission tower attitude monitoring and vibration intrusion identification based on multi-source data fusion according to claim 1, characterized in that the dynamically adjusted fusion coefficient α is adjusted in real time according to the GPS signal quality as follows: When a strong GPS signal and high positioning accuracy are detected, the weight of GPS in the fusion process is increased, i.e., the fusion coefficient α is increased, to accelerate attitude correction and improve overall accuracy. In scenarios where GPS signals are weak or lost, reducing the GPS weight means reducing the fusion coefficient α.

3. The method for transmission tower attitude monitoring and vibration intrusion identification based on multi-source data fusion according to claim 2, characterized in that, The dynamically adjusted fusion coefficient α is between 0.6 and 0.8.

Images