Multi-source fusion perception power transmission tower pole structure abnormity early warning method and device

Abstract

The invention discloses a power transmission tower pole structure abnormity early warning method based on multi-source fusion perception, and belongs to the technical field of power transmission tower pole structure abnormity early warning. The method comprises the following steps: 1, adaptively activating a multi-source sensor to execute data acquisition, and establishing a multi-source data set; 2, establishing a multi-modal physical structure feature vector; 3, respectively establishing a local abnormal index vector and a global abnormal index vector through the local sensing layer and the global sensing layer; 4, executing self-correction and mutual verification of the multi-source sensor, establishing a correction factor, and executing dynamic fusion of a local abnormal index vector and a global abnormal index vector by using the correction factor; and 5, carrying out space-time adaptive anomaly identification on a dynamic fusion result, and outputting a structure anomaly early warning. According to the invention, the technical problems of low anomaly recognition accuracy, high false alarm rate and poor early warning reliability caused by incapability of distinguishing environmental interference and structural anomaly are solved, and the anomaly recognition accuracy and the early warning reliability are improved.

Description

Technical Field

[0001] This invention relates to the field of early warning technology for abnormal structures of power transmission towers, and specifically to a method and device for early warning of abnormal structures of power transmission towers using multi-source fusion sensing. Background Technology

[0002] As a core component of power grid infrastructure, the structural health of transmission towers directly impacts the safe and stable operation of the grid. Due to long-term exposure to the natural environment, transmission towers are susceptible to wind, icing, geological disasters, and material aging, leading to structural performance degradation or even failure, and consequently, power outages. Therefore, real-time and accurate monitoring and early warning of anomalies in transmission tower structures are crucial. Traditional transmission tower structural monitoring primarily relies on vibration sensors, strain gauges, or image sensors for data acquisition. These methods are easily affected by environmental interference, resulting in low data reliability and difficulty in comprehensively capturing the multi-dimensional characteristics of the structure. Furthermore, they lack adaptive capabilities, failing to distinguish between abnormal signals and environmental noise, leading to high false alarm rates and poor early warning timeliness. In addition, existing multi-source sensor collaborative acquisition and data processing lack adaptive mechanisms, resulting in resource waste and insufficient real-time performance, making it difficult to distinguish between structural anomalies and environmental interference, thus compromising the accuracy of transmission tower structural anomaly identification and the reliability of early warnings.

[0003] Therefore, current related technologies suffer from the inability to distinguish between environmental interference and structural anomalies, resulting in low accuracy in anomaly identification, high false alarm rates, and poor reliability of early warning. To address these issues, this invention provides a method and device for early warning of structural anomalies in transmission towers based on multi-source fusion sensing. Summary of the Invention

[0004] To address the aforementioned technical shortcomings, the present invention aims to provide a method for early warning of structural anomalies in transmission towers based on multi-source fusion sensing. This method adaptively activates multi-source sensors to perform data acquisition, establishing a multi-source dataset and a multimodal physical structure feature vector. Local and global anomaly indicator vectors are established through a local and a global sensing layer, respectively. The correction factor is used to dynamically fuse the local and global anomaly indicator vectors, and the fusion result is then used for space-time adaptive anomaly identification. This method solves the technical problems in existing technologies, such as the inability to distinguish between environmental interference and structural anomalies, resulting in low anomaly identification accuracy, high false alarm rate, and poor early warning reliability. It achieves the technical effect of improving the accuracy of anomaly identification and the reliability of early warning.

[0005] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: The present invention provides a method for early warning of structural anomalies in transmission towers based on multi-source fusion sensing, comprising the following steps: Step 1: Put the multi-source sensor into a semi-dormant state; if the multi-source sensor acquires an abnormal local signal or generates an abnormal event, the multi-source sensor will be adaptively activated to perform data acquisition and establish a multi-source dataset; otherwise, the semi-dormant state will be maintained. Step 2: Map the multi-source dataset to a unified structural feature space to establish a multimodal physical structure feature vector; Step 3: After collecting environmental data vectors, use the structure-environment causal graph to perform causal reasoning fusion of environmental data vectors and physical structure feature vectors, and establish local anomaly index vectors and global anomaly index vectors through the local perception layer and the global perception layer, respectively. Step 4: Determine whether the local anomaly index vector and the global anomaly index vector need to be calibrated by the multi-source sensor; if yes, perform self-calibration and mutual verification of the multi-source sensor, establish a calibration factor, and use the calibration factor to perform dynamic fusion of the local anomaly index vector and the global anomaly index vector; if no, proceed to step 5. Step 5: Perform spatial-temporal adaptive anomaly identification on the dynamic fusion results and output a structural anomaly warning.

[0006] Preferably, in step one, the adaptive activation of the multi-source sensors to perform data acquisition is as follows: Based on the predicted range of impact or event development, the system intelligently wakes up the corresponding sensors and switches the sensors in a semi-dormant state to full-power operation mode according to the type and location of the trigger signal. This enables collaborative high-frequency data acquisition, simultaneously collecting various data such as vibration, tilt, stress, and images, and integrating them to establish a multi-source dataset.

[0007] Preferably, the abnormal local signals collected by the multi-source sensors in step one are as follows: If any multi-source sensor collects an abnormal local signal, then an adaptive influence range prediction is performed based on the abnormal value and abnormal node of the abnormal local signal to establish a first prediction result; historical abnormal data of the first prediction result is obtained, and the first prediction result is compensated based on the historical abnormal data to establish a second prediction result; within the range of the second prediction result, the corresponding multi-source sensor is activated to collect data.

[0008] Preferably, the abnormal event data collected by the multi-source sensors in step one is as follows: If an abnormal event occurs, a multi-source sensor wake-up evaluation is performed based on the event association, and a wake-up evaluation result is established. The abnormal event is predicted, a tolerance factor is configured using the event prediction result, and the wake-up evaluation result is compensated based on the tolerance factor. Then, the multi-source sensors are adaptively activated to perform data acquisition, and a multi-source dataset is established. If an abnormal event is determined, a multi-source sensor wake-up evaluation is performed based on the event association. The abnormal event's development is predicted, including using a time series prediction model to predict its future intensity and evolution trend based on current abnormal event characteristics and historical abnormal event data, thus determining the prediction result. A tolerance factor is configured using the prediction result to control the aggressiveness or conservatism of the wake-up strategy. If the abnormal event is predicted to worsen, a high tolerance factor is configured; if the event is predicted to weaken, a low tolerance factor is configured. The wake-up evaluation result is dynamically compensated based on the tolerance factor to obtain a sensor wake-up scheme and adaptively activate multi-source sensors to perform data acquisition. Specifically, a high tolerance factor expands the wake-up range, while a low tolerance factor shrinks the wake-up range.

[0009] Preferably, step two includes the following: Feature indicators were extracted from the raw data of each mode, including the characteristic frequency, amplitude, root mean square value, and kurtosis related to the vibration signal; the average strain, strain change amplitude, and number of cycles of the strain signal; the displacement of feature points, width of structural gaps, and area ratio of rusted regions of the image data; and the temperature gradient characteristics of the environmental data. All feature values ​​are processed by Min-Max scaling or Z-Score standardization to eliminate differences in dimensions and orders of magnitude, obtain multi-dimensional structural features, establish a unified structural feature space, and establish multimodal physical structure feature vectors.

[0010] Preferably, in step three, the structure-environment causal graph is used to perform causal reasoning fusion of environmental data vectors and physical structure feature vectors, as detailed below: Environmental factors are collected by meteorological sensors and quantified into external environmental data vectors affecting the transmission tower at the current moment. Then, a structure-environment causal graph is used to perform causal reasoning fusion of the environmental data vectors and physical structure feature vectors. The structure-environment causal graph is a predefined graph model based on historical anomaly data. The nodes in the structure-environment causal graph represent variables, and the directed edges represent the causal influence directions. The causal reasoning fusion refers to substituting the environmental data vectors and multimodal physical structure feature vectors into the structure-environment causal graph for reasoning, including causal effect prediction, residual calculation, and causal interpretation. In step three, local anomaly indicator vectors and global anomaly indicator vectors are established through the local perception layer and the global perception layer, respectively. The specific details are as follows: Reasoning is performed on the nodes representing local components in the structure-environment causal graph and their directly connected environmental nodes, and anomaly scores for each local component are output. Then, a local anomaly index vector containing the anomaly scores of multiple local components is constructed. Finally, a global anomaly index vector is established through a global perception layer.

[0011] Preferably, the determination in step four of whether the local anomaly index vector and the global anomaly index vector require multi-source sensor correction is as follows: Real-time noise spectra of multi-source sensors are extracted, and template comparison is performed between the real-time noise spectra and the mapped sensor fingerprint template to establish a log score. Environmental impact and environmental sensitivity analysis of multi-source sensors are performed based on the environmental data vector to establish an adaptation score. A self-verification factor is established based on the log score and the adaptation score. Adjacent sensors and equivalent physical quantity sensors of multi-source sensors are identified to establish an identification cluster. Cooperative residual calculation is performed using the identification cluster to establish a mutual verification factor.

[0012] Preferably, step four involves performing self-calibration of the multi-source sensors, the specific details of which are as follows: Real-time noise spectra of multi-source sensors are extracted, and template comparison is performed between the real-time noise spectra and the mapped sensor fingerprint template. The differences are calculated, and log scores are output based on the magnitude of the differences. High scores indicate that the noise spectrum is highly consistent with the template, while low scores indicate that the noise spectrum has drifted or that new noise peaks have appeared. Environmental impact and environmental sensitivity analyses of multi-source sensors are performed based on environmental data vectors. Environmental impact analysis refers to determining the sensor reading deviation under extreme environments based on a preset environmental database, while environmental sensitivity analysis refers to analyzing whether the sensitivity of the current sensor readings to environmental changes is within the expected range, and then outputting an adaptation score. High scores indicate that the sensor's behavior in the current environment is in line with expectations, while low scores indicate that the sensor exhibits abnormal environmental sensitivity or a decrease in anti-interference ability. The log scores and adaptation scores are weighted and calculated, and a self-verification factor is established by fusing them. A value close to 1 indicates that the sensor itself has high health and the readings are reliable, while a value close to 0 indicates that the sensor itself may have a fault or severe performance degradation, and its readings have low reliability. Step four involves performing mutual verification of the multi-source sensors, the details of which are as follows: Based on the topology and sensor configuration table, the system performs neighbor and equivalent physical quantity sensor identification for multi-source sensors. This includes grouping physically adjacent sensors into neighbor sensor identification clusters and sensors measuring the same physical quantity but with different types or principles into equivalent physical quantity sensor identification clusters. The system then uses these identification clusters to perform collaborative residual calculation. It normalizes and weights the self-verification factor and mutual verification factor, outputting a correction factor. Finally, it uses the correction factor as a weight to perform dynamic fusion of local and global anomaly index vectors.

[0013] Preferably, step five includes the following: The dynamic fusion result is decomposed into multi-scale components to establish long-term trend components and short-term residual components. Based on the short-term residual components, abrupt change detection is performed to establish a short-term anomaly score. Based on the long-term trend components, change point detection and trend slope evaluation are performed to establish a long-term anomaly score. The short-term anomaly score and the long-term anomaly score are then verified and weighted to fuse, and a structural anomaly warning is output.

[0014] A multi-source fusion sensing early warning device for transmission tower structural anomalies includes: The adaptive acquisition module is used to put the multi-source sensors into a semi-dormant state. When any multi-source sensor acquires an abnormal local signal or generates an abnormal event, it adaptively activates the multi-source sensor to perform data acquisition and establish a multi-source dataset. The data processing module is used to map the multi-source dataset to a unified structural feature space and establish a multimodal physical structure feature vector. The reasoning and recognition module is used to perform causal reasoning fusion of environmental data vectors and physical structure feature vectors after collecting environmental data vectors, using a structure-environment causal graph. Local anomaly index vectors and global anomaly index vectors are established through a local perception layer and a global perception layer, respectively. The first calibration module is used to perform self-calibration and mutual verification of multi-source sensors, establish calibration factors, and use the calibration factors to perform dynamic fusion of local anomaly index vectors and global anomaly index vectors. The second correction module is used to perform spatial-temporal adaptive anomaly identification on the dynamic fusion results and output structural anomaly warning; the device executes the above-mentioned multi-source fusion sensing method for early warning of transmission tower structural anomalies.

[0015] The beneficial effects of this invention are as follows: This invention adaptively activates multi-source sensors to perform data acquisition, establishes a multi-source dataset, and creates a multimodal physical structure feature vector. It then establishes local and global anomaly indicator vectors through a local and a global perception layer, respectively. The invention utilizes a correction factor to dynamically fuse the local and global anomaly indicator vectors, and performs space-time adaptive anomaly identification using the fusion result. This solves the technical problems in existing technologies, such as the inability to distinguish between environmental interference and structural anomalies, leading to low anomaly identification accuracy, high false alarm rate, and poor early warning reliability. The invention achieves the technical effect of improving anomaly identification accuracy and early warning reliability. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 A flowchart of a method for early warning of structural anomalies in transmission towers based on multi-source fusion sensing provided in an embodiment of the present invention.

[0018] Figure 2 This is a schematic diagram of the structure of the multi-source fusion sensing transmission tower structural anomaly early warning device provided in an embodiment of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Example: Figure 1 As shown, this invention provides a method for early warning of structural anomalies in transmission towers based on multi-source fusion sensing, comprising the following steps: Step 1: Put the multi-source sensor into a semi-dormant state; if the multi-source sensor acquires an abnormal local signal or generates an abnormal event, the multi-source sensor will be adaptively activated to perform data acquisition and establish a multi-source dataset; otherwise, the semi-dormant state will remain.

[0021] Step 2: Map the multi-source dataset to a unified structural feature space and establish a multimodal physical structure feature vector.

[0022] Step 3: After collecting environmental data vectors, use the structure-environment causal graph to perform causal reasoning fusion of environmental data vectors and physical structure feature vectors, and establish local anomaly indicator vectors and global anomaly indicator vectors through the local perception layer and the global perception layer, respectively.

[0023] Step 4: Determine whether the local anomaly index vector and the global anomaly index vector need to be calibrated by the multi-source sensor; if yes, perform self-calibration and mutual verification of the multi-source sensor, establish a calibration factor, and use the calibration factor to perform dynamic fusion of the local anomaly index vector and the global anomaly index vector; if no, proceed to Step 5.

[0024] Step 5: Perform spatial-temporal adaptive anomaly identification on the dynamic fusion results and output a structural anomaly warning.

[0025] In this embodiment, data acquisition is performed by adaptively activating multi-source sensors to establish a multi-source dataset and a multimodal physical structure feature vector. Local anomaly index vectors and global anomaly index vectors are established through a local perception layer and a global perception layer, respectively. The local anomaly index vectors and global anomaly index vectors are dynamically fused using the correction factor, and the dynamic fusion result is used for space-time adaptive anomaly identification. This solves the technical problems in the prior art that cannot distinguish between environmental interference and structural anomalies, resulting in low anomaly identification accuracy, high false alarm rate, and poor early warning reliability, and achieves the technical effect of improving the accuracy of anomaly identification and the reliability of early warning.

[0026] Furthermore, the adaptive activation of multi-source sensors to perform data acquisition in step one is as follows: Specifically, the multi-source sensors are placed in a semi-dormant state, that is, in a low-power standby mode, maintaining only basic anomaly monitoring functions. When any multi-source sensor acquires an abnormal local signal or generates an abnormal event, such as when a strain gauge reading detects a signal exceeding a preset threshold within its monitoring range or receives an external anomaly alarm, the multi-source sensors are adaptively activated to perform data acquisition. This includes intelligently waking up the corresponding sensor based on the anomaly's impact range or event development prediction, and switching the semi-dormant sensor to full-power operating mode based on the type and location of the trigger signal to perform collaborative high-frequency data acquisition. Simultaneously, multiple data such as vibration, tilt, stress, and images are acquired and integrated to establish a multi-source dataset, thereby ensuring that no key anomaly information is missed and minimizing normal power consumption.

[0027] Furthermore, the abnormal local signals collected by the multi-source sensors in step one are as follows: If any multi-source sensor collects an abnormal local signal, an adaptive impact range prediction is performed based on the abnormal value and abnormal node of the abnormal local signal to establish a first prediction result. Historical abnormal data of the first prediction result is acquired, and the first prediction result is compensated based on the historical abnormal data to establish a second prediction result. Within the range of the second prediction result, corresponding multi-source sensor activation is performed for data acquisition. Furthermore, if any multi-source sensor collects an abnormal local signal, i.e., after a single sensor detects an anomaly, the impact range is intelligently predicted and relevant sensors are accurately activated. Specifically, the abnormal value of the abnormal local signal refers to the specific abnormal data value of the sensor, and the abnormal node is the sensor location identifier that triggers the abnormal signal. Based on the transmission tower structural mechanics model and fault propagation knowledge base, the most likely propagation path and impact range of stress, vibration, etc., when the abnormal value occurs at the abnormal node are calculated in real time, and a first prediction result is output, including potentially affected areas. For example, the first prediction result may be the sensor coverage area of ​​all sections 2 to 4 on the left wing of tower A. Then, the impact range is... Take the historical anomaly data of the first prediction result, that is, the actual records when anomalies occurred at the same or similar locations in the past, including the historical anomaly impact range. Then compare and analyze the first prediction result with the historical anomaly data, and adjust and compensate the first prediction result of the model according to the deviation to determine the second prediction result, that is, the precise impact range verified and compensated by the historical anomaly data. For example, it is necessary to activate all sensors in sections 1 to 5 of the left wing of tower A. Finally, according to the precise impact range determined by the second prediction result, send activation commands to the sensors within the range, switch the working state of the multi-source sensors, start synchronous high-frequency data acquisition, and form a multi-source dataset.

[0028] Furthermore, the abnormal events collected by the multi-source sensors in step one are as follows: If an abnormal event occurs, a multi-source sensor wake-up evaluation based on event association is performed, and a wake-up evaluation result is established. The abnormal event's development is predicted, and a tolerance factor is configured using the prediction result. After compensating the wake-up evaluation result based on the tolerance factor, multi-source sensors are adaptively activated to perform data acquisition, establishing a multi-source dataset. If an abnormal event is determined, such as a strong typhoon warning, a seismic wave warning issued by earthquake monitoring, a tower foundation landslide risk identified by drone inspection, or a suspected insulator icing event determined by the system itself, a multi-source sensor wake-up evaluation based on event association is performed. Specifically, a pre-set event-sensor association rule base is used to define the most critical sensors for different event types. For example, for a typhoon warning, anemometers, tower tilt sensors, and micro-wind vibration sensors are evaluated with high priority; for suspected icing, temperature sensors, humidity sensors, cameras, and tension sensors are evaluated with high priority. All sensors are then scored and ranked based on their importance, and a sensor wake-up scheme list is output as the wake-up evaluation result, including sensors recommended for immediate activation and their wake-up priorities. Anomaly events are predicted by a time-series forecasting model based on current event characteristics and historical event data. This prediction determines the event's future intensity and evolution trend. The prediction results are then used to configure a tolerance factor, controlling the aggressiveness or conservatism of the wake-up strategy. A high tolerance factor is configured if the anomaly is predicted to worsen, indicating a willingness to accept higher energy costs for more comprehensive data monitoring and to prevent missed detections. Conversely, a low tolerance factor is configured if the event is predicted to weaken, indicating that only the most critical sensors are activated for more energy-efficient strategies to avoid overreaction. Finally, the wake-up evaluation results are dynamically compensated based on the tolerance factor. A high tolerance factor expands the wake-up range, while a low tolerance factor shrinks it. This results in a sensor wake-up scheme that adaptively activates multiple sensors to perform data acquisition, ultimately generating a highly accurate multi-source dataset for the specific anomaly event. This effectively balances the reliability of anomaly monitoring and early warning with overall power consumption.

[0029] Furthermore, the specific details of step two are as follows: Feature indicators are extracted from the raw data of each mode, including characteristic frequencies, amplitudes, root mean square values, and kurtosis related to vibration signals; average strain, strain change amplitude, and cycle number of strain signals; feature point displacements, structural gap widths, and corrosion area proportions from image data; and temperature gradient features from environmental data. All feature values ​​are processed using Min-Max scaling or Z-Score normalization to eliminate differences in dimensions and orders of magnitude, obtaining multi-dimensional structural features, establishing a unified structural feature space, and creating a multi-modal physical structure feature vector. Each element corresponds to a normalized physical feature of a sensor. For example, the multi-modal physical structure feature vector includes the dominant frequency offset corresponding to the vibration sensor, the strain amplitude of key parts corresponding to the strain gauge, the tower tilt corresponding to the tilt sensor, the maximum crack width corresponding to the image sensor, and the current wind speed influence coefficient corresponding to the anemometer.

[0030] Furthermore, in step three, the structure-environment causal graph is used to fuse causal reasoning between environmental data vectors and physical structure feature vectors. The specific details are as follows: Environmental factors are collected by meteorological sensors and quantified into external environmental data vectors affecting the transmission tower at the current moment. A structure-environment causal graph is then used to fuse the environmental data vectors and physical structure feature vectors through causal reasoning. The structure-environment causal graph is a predefined graphical model based on historical anomaly data, used to describe the causal relationship between environmental factors, structural component states, and the overall tower state. Nodes in the structure-environment causal graph represent variables, such as wind speed, strain of the third member, and tower top displacement, while directed edges represent the direction of causal influence. The causal reasoning fusion refers to substituting the environmental data vectors and multimodal physical structure feature vectors into the structure-environment causal graph for reasoning, including causal effect prediction, residual calculation, and causal interpretation. Specifically, based on the current environmental data vectors and the structure-environment causal graph, the theoretical expected value of the physical structure feature vector of the transmission tower structure response under the current environmental conditions is calculated. The theoretical expected value is then compared with the actual measured physical structure feature vector value to determine the residual. If the residual cannot be explained by the known environmental factors through the causal graph, it may indicate a potential structural anomaly or damage.

[0031] In step three, local anomaly indicator vectors and global anomaly indicator vectors are established through the local perception layer and the global perception layer, respectively. The specific details are as follows: The system infers the anomaly scores of each local component and its directly connected environmental nodes in the structure-environment causal graph, thus constructing a local anomaly index vector containing the anomaly scores of multiple local components. A global anomaly index vector is then established through a global perception layer. The global perception layer infers the macroscopic behavior, stability, and modal characteristics of the entire transmission tower structure, such as overall modal frequency, overall tower top displacement, and torsion. This involves inferring the anomaly scores of the nodes representing the overall tower performance in the structure-environment causal graph and all environmental nodes, considering neighboring tower information to eliminate the influence of common environmental factors such as regional strong winds, and outputting anomaly scores for overall performance indicators. Finally, a global anomaly index vector is constructed, allowing for precise localization of anomalies to specific local components or overall structural performance, thereby reducing false alarm rates and ensuring the reliability and accuracy of anomaly identification and early warning.

[0032] Furthermore, in step four, it is determined whether the local anomaly index vector and the global anomaly index vector require multi-source sensor correction. The specific details are as follows: Real-time noise spectra of multi-source sensors are extracted, and template comparison is performed between the real-time noise spectra and the mapped sensor fingerprint template to establish a log score. Environmental impact and sensitivity analysis of the multi-source sensors are performed based on the environmental data vector to establish an adaptation score. A self-verification factor is established based on the log score and the adaptation score. Adjacent sensors and equivalent physical quantity sensors of the multi-source sensors are identified to establish a recognition cluster. Cooperative residual calculation is performed using the recognition cluster to establish a mutual verification factor. After normalizing the self-verification factor and the mutual verification factor, a correction factor is established.

[0033] Furthermore, step four involves performing self-calibration of the multi-source sensors, the details of which are as follows: The real-time noise spectrum of the multi-source sensors is extracted, which is the background noise output when the sensor has no signal input or the signal is stable. The sensor fingerprint template is the background noise spectrum of each sensor after factory calibration or in a healthy state. The real-time noise spectrum is compared with the mapped sensor fingerprint template, and the difference is calculated, such as the cosine similarity of the spectrum or the mean square error. The log score is output according to the size of the difference. A high score indicates that the noise spectrum is highly consistent with the template, indicating that the sensor performance is stable. A low score indicates that the noise spectrum has drifted or new noise peaks have appeared, indicating that the sensor may be aging, damp, or has an internal fault. Environmental impact and sensitivity analyses are performed on multi-source sensors based on environmental data vectors. Environmental impact analysis determines the sensor reading deviation under extreme conditions based on a pre-set environmental database. Environmental sensitivity analysis analyzes whether the current sensor readings are within the expected range to respond to environmental changes, and outputs an adaptation score. A high score indicates that the sensor's behavior in the current environment meets expectations, while a low score indicates that the sensor exhibits abnormal environmental sensitivity or decreased anti-interference ability. The log score and adaptation score are weighted and calculated, and a self-verification factor is established. A value close to 1 indicates that the sensor itself has high health and the readings are reliable, while a value close to 0 indicates that the sensor itself may have a fault or severe performance degradation, and its readings have low reliability.

[0034] Step four involves performing mutual verification of the multi-source sensors, the details of which are as follows: Based on the topology and sensor configuration table, the system performs neighbor and equivalent physical quantity sensor identification for multi-source sensors. This includes grouping physically adjacent sensors into neighbor sensor identification clusters and sensors measuring the same physical quantity but with different types or principles into equivalent physical quantity sensor identification clusters. The identification clusters are then used to perform collaborative residual calculation. Specifically, the most reliable sensor reading within an identification cluster is used as a reference benchmark. The reference benchmark is subtracted from the readings of other sensors within the cluster to obtain multiple differences as residuals. Statistical analysis is then performed on these residuals, such as calculating the standard deviation. A value close to 1 indicates that the sensor reading is highly consistent with its neighbor readings, mutually corroborating each other and indicating high reliability. A value close to 0 indicates that the sensor reading differs from its neighbor readings, suggesting potential abnormal readings. For example, if a vibration sensor reading spikes while other sensor readings within the same cluster remain stable, the sensor's mutual verification factor becomes extremely low, suggesting a possible fault rather than a genuine structural anomaly. The self-verification factor and mutual verification factor are normalized and weighted and fused to output a correction factor, which comprehensively reflects the overall reliability of the sensor. The correction factor is used as a weight to perform dynamic fusion of local anomaly index vector and global anomaly index vector. Sensors with high correction factors have higher weights, while sensors with low correction factors have reduced or even ignored weights, thereby preventing erroneous data from faulty sensors from affecting early warning and improving the accuracy and reliability of anomaly identification and early warning of transmission tower structures.

[0035] Furthermore, the specific details of step five are as follows: The dynamic fusion results are decomposed into multi-scale components to establish long-term trend components and short-term residual components. Abrupt change detection is performed based on the short-term residual components to establish a short-term anomaly score. Change point detection and trend slope evaluation are performed based on the long-term trend components to establish a long-term anomaly score. The short-term and long-term anomaly scores are then validated and weighted to generate a structural anomaly warning. Based on time series decomposition, wavelet transform, or empirical mode decomposition, the dynamic fusion results are decomposed into long-term trend components and short-term residual components. The long-term trend component represents slow, gradual changes in anomaly indicators, reflecting chronic degradation such as material fatigue, corrosion, and foundation settlement. The short-term residual component contains rapid changes, sudden fluctuations, and noise, reflecting acute damage such as bolt breakage, member buckling, and external impact. Z-score mutation detection is used to detect mutations in the short-term residual components, identifying sudden, large-amplitude pulses or step changes in the residual sequence. If a mutation is detected, the short-term anomaly score at that point increases sharply, with the score magnitude based on the severity of the mutation and the event probability, thus establishing the short-term anomaly score. Bayesian change point detection is used to detect change points and evaluate the trend slope in the long-term trend components. Specifically, it identifies the moment when the slope of the long-term trend component changes and performs linear fitting on the trend component to determine the trend slope, thereby comprehensively determining the long-term anomaly score. The larger the slope and / or the more recent the occurrence of multiple change points, the higher the score. Then, weights are assigned to the short-term and long-term anomaly scores based on the current environmental context and historical warning information. The short-term and long-term anomaly scores are then validated and weighted to fuse them, outputting a structural anomaly warning. The warning information includes the type and level of the transmission tower structural anomaly, thereby improving the accuracy of transmission tower structural anomaly identification, reducing the false alarm rate, and improving the reliability of anomaly warnings.

[0036] like Figure 2 As shown, the multi-source fusion sensing transmission tower structural anomaly early warning device, which implements the above-mentioned multi-source fusion sensing method for early warning of transmission tower structural anomalies, includes: The adaptive acquisition module is used to put the multi-source sensors into a semi-dormant state. When any multi-source sensor acquires an abnormal local signal or generates an abnormal event, it adaptively activates the multi-source sensor to perform data acquisition and establish a multi-source dataset. The data processing module is used to map the multi-source dataset to a unified structural feature space and establish a multimodal physical structure feature vector. The reasoning and recognition module is used to perform causal reasoning fusion of environmental data vectors and physical structure feature vectors after collecting environmental data vectors, using a structure-environment causal graph. Local anomaly index vectors and global anomaly index vectors are established through a local perception layer and a global perception layer, respectively. The first calibration module is used to perform self-calibration and mutual verification of multi-source sensors, establish calibration factors, and use the calibration factors to perform dynamic fusion of local anomaly index vectors and global anomaly index vectors. The second correction module is used to perform spatial-temporal adaptive anomaly identification on the dynamic fusion results and output structural anomaly warning; the device executes the above-mentioned multi-source fusion sensing method for early warning of transmission tower structural anomalies.

[0037] Specifically, the configuration of the reasoning and recognition module also includes a construction unit, used to establish a hierarchical diagram of pole-connection point-to-overall tower using the topology of the transmission tower as prior data, apply environmental nodes to the hierarchical diagram, perform data-driven causal edge inference, and establish a structure-environment causal graph using the causal edge inference results; a local recognition unit, used to input the environmental data vector and physical structure feature vector into the structure-environment causal graph, and use a local perception layer to perform local anomaly perception under environment-structure causal inference, and establish a local anomaly index vector; and a global recognition unit, used to input the environmental data vector and physical structure feature vector into the structure-environment causal graph, and use a global perception layer to perform global anomaly perception under environment-structure causal inference, and establish a global anomaly index vector.

[0038] More specifically, based on the design drawings, BIM models, or 3D point cloud data of the transmission tower, the topology of the transmission tower is constructed, clarifying the members, connection points, and physical connections of the transmission tower. The topology of the transmission tower is used as prior data to establish a hierarchical graph of members-connection points-the entire tower, that is, the topology is transformed into a hierarchical graph structure. In this graph, the bottom layer nodes represent individual members and connection points, the middle layer nodes represent substructures composed of multiple members and connection points, and the top layer nodes represent the performance indicators of the entire transmission tower. The edges between nodes represent physical connections. Then, environmental nodes are added to the hierarchical graph, and data-driven causal edge inference is performed. That is, using historical monitoring data of environmental and structural responses, statistical analysis is performed through causal discovery algorithms to identify the causal relationships and causal directions between environmental nodes and various structural nodes in the hierarchical graph. For example, there is a strong causal relationship between wind speed and tower top displacement, with the direction being wind speed → tower top displacement. This results in the output of a structure-environment causal graph. Furthermore, environmental data vectors and physical structural feature vectors are input into the structure-environment causal graph. Local anomaly perception is then performed through a local identification unit under environment-structure causal reasoning. This involves performing causal reasoning on each bottom-level node in the structure-environment causal graph to predict the theoretical strain / vibration values ​​of the members under the current environmental conditions. The predicted values ​​are then compared with actual measurements to determine the anomaly score for each local component, ultimately outputting a local anomaly index vector to achieve preliminary localization of structural anomalies in the transmission tower. Simultaneously, a global perception layer is used for global anomaly perception under environment-structure causal reasoning. This involves performing causal reasoning on the top-level and middle-level nodes in the structure-environment causal graph to predict the theoretical values ​​of the first-order frequency, damping ratio, and overall stiffness of the entire transmission tower under the current environmental conditions. Again, the predicted values ​​are compared with actual measurements to determine the anomaly score for each overall performance index, ultimately outputting a global anomaly index vector to determine whether and how local anomalies affect the overall safety of the transmission tower structure. This clearly distinguishes between environmental interference and actual structural damage, reducing the false alarm rate of structural anomaly warnings.

[0039] Furthermore, the specific configuration of the global identification unit also includes a global linkage layer, used to send linkage sensing signals to neighboring transmission towers, receive linkage sensing feedback, perform sensing compensation for the global anomaly sensing using the linkage sensing feedback, and update the global anomaly index vector. When the global identification unit of this transmission tower outputs a high global anomaly index vector, for example, detecting increased overall swaying of the transmission tower structure, the global linkage layer is activated and a linkage sensing signal is sent to neighboring transmission towers. This signal may be a standardized query request including the anomaly type code detected by this tower, current environmental data, the time period of the anomaly, and the corresponding sensing data request information sent to the neighbor. After receiving the linkage sensing signal, the neighboring towers determine whether they have detected similar or related anomalies in the same time period, and determine whether their own local and global anomaly indicators are normal, and then generate linkage sensing feedback and send it to this transmission tower. The linkage sensing feedback may include consistent feedback or inconsistent feedback. Then, the perception compensation for global anomaly detection is performed using the linkage perception feedback. If the feedback from neighboring towers is consistent, indicating that they have detected similar or identical anomalies, it may be a wide-area environmental event, such as a gust of wind, a small-scale earthquake, or line galloping. In this case, the global anomaly index of this transmission tower is lowered, a corresponding compensation coefficient is generated, and the global anomaly index vector is updated to reduce the warning level, thus preventing high-level transmission tower structural damage warnings. If the feedback from neighboring towers is inconsistent, meaning that the surrounding towers are all normal, it indicates that this transmission tower has structural damage. In this case, the global anomaly index of this transmission tower is increased, thus providing more accurate structural anomaly warnings and ensuring the accuracy and reliability of the structural anomaly warnings. If the feedback from neighboring towers is partially consistent, for example, if the upstream tower in the wind direction shows an anomaly first, followed by the downstream tower, it is determined to be an environmental event. At the same time, it can be verified whether the event propagation path matches the prediction of the structure-environment causal graph, thereby optimizing the structure-environment causal graph itself.

[0040] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for early warning of structural anomalies in transmission towers based on multi-source fusion sensing, characterized in that, Includes the following steps: Step 1: Put the multi-source sensor into a semi-dormant state; if the multi-source sensor acquires an abnormal local signal or generates an abnormal event, the multi-source sensor will be adaptively activated to perform data acquisition and establish a multi-source dataset; otherwise, the semi-dormant state will be maintained. Step 2: Map the multi-source dataset to a unified structural feature space to establish a multimodal physical structure feature vector; Step 3: After collecting environmental data vectors, use the structure-environment causal graph to perform causal reasoning fusion of environmental data vectors and physical structure feature vectors, and establish local anomaly index vectors and global anomaly index vectors through the local perception layer and the global perception layer, respectively. Step 4: Determine whether the local anomaly index vector and the global anomaly index vector need to be calibrated by the multi-source sensor; if yes, perform self-calibration and mutual verification of the multi-source sensor, establish a calibration factor, and use the calibration factor to perform dynamic fusion of the local anomaly index vector and the global anomaly index vector; if no, proceed to step 5. Step 5: Perform spatial-temporal adaptive anomaly identification on the dynamic fusion results and output a structural anomaly warning.

2. The method for early warning of structural anomalies in transmission towers based on multi-source fusion sensing as described in claim 1, characterized in that, In step one, the adaptive activation of the multi-source sensors to perform data acquisition is as follows: Based on the predicted range of impact or event development, the system intelligently wakes up the corresponding sensors and switches the sensors in a semi-dormant state to full-power operation mode according to the type and location of the trigger signal. This enables collaborative high-frequency data acquisition, simultaneously collecting various data such as vibration, tilt, stress, and images, and integrating them to establish a multi-source dataset.

3. The method for early warning of structural anomalies in transmission towers based on multi-source fusion sensing as described in claim 1, characterized in that, The abnormal local signals collected by the multi-source sensors in step one are as follows: If any multi-source sensor collects an abnormal local signal, then an adaptive influence range prediction is performed based on the abnormal value and abnormal node of the abnormal local signal to establish a first prediction result; historical abnormal data of the first prediction result is obtained, and the first prediction result is compensated based on the historical abnormal data to establish a second prediction result; within the range of the second prediction result, the corresponding multi-source sensor is activated to collect data.

4. The method for early warning of structural anomalies in transmission towers based on multi-source fusion sensing as described in claim 1, characterized in that, The abnormal events collected by the multi-source sensors in step one are as follows: If an abnormal event occurs, a multi-source sensor wake-up evaluation is performed based on the event association, and a wake-up evaluation result is established. The abnormal event is predicted, a tolerance factor is configured using the event prediction result, and the wake-up evaluation result is compensated based on the tolerance factor. Then, the multi-source sensors are adaptively activated to perform data acquisition, and a multi-source dataset is established. If an abnormal event is determined to have occurred, then the multi-source sensor wake-up evaluation under the event association is performed according to the abnormal event. Predicting the development of anomalous events involves using time series prediction models to forecast the future intensity and evolution trend of the event based on current anomalous event characteristics and historical anomalous event data, and determining the event development prediction results. The event development prediction results are then used to configure tolerance factors to control the aggressiveness or conservatism of the wake-up strategy. If the predicted abnormal event intensifies, a high tolerance factor is configured; if the predicted event will weaken, a low tolerance factor is configured. The wake-up evaluation results are dynamically compensated based on the tolerance factor to obtain a sensor wake-up scheme and adaptively activate multi-source sensors to perform data acquisition. Among them, the high tolerance factor expands the wake-up range, and the low tolerance factor shrinks the wake-up range.

5. The method for early warning of structural anomalies in transmission towers based on multi-source fusion sensing as described in claim 1, characterized in that, Step two details are as follows: Feature indicators were extracted from the raw data of each mode, including the characteristic frequency, amplitude, root mean square value, and kurtosis related to the vibration signal; the average strain, strain change amplitude, and number of cycles of the strain signal; the displacement of feature points, width of structural gaps, and area ratio of rusted regions of the image data; and the temperature gradient characteristics of the environmental data. All feature values ​​are processed by Min-Max scaling or Z-Score standardization to eliminate differences in dimensions and orders of magnitude, obtain multi-dimensional structural features, establish a unified structural feature space, and establish multimodal physical structure feature vectors.

6. The method for early warning of structural anomalies in transmission towers based on multi-source fusion sensing as described in claim 1, characterized in that, Step three utilizes a structure-environment causal graph to fuse environmental data vectors and physical structure feature vectors for causal reasoning. The specific details are as follows: Environmental factors are collected by meteorological sensors and quantified into external environmental data vectors affecting the transmission tower at the current moment. Then, a structure-environment causal graph is used to perform causal reasoning fusion of the environmental data vectors and physical structure feature vectors. The structure-environment causal graph is a predefined graph model based on historical anomaly data. The nodes in the structure-environment causal graph represent variables, and the directed edges represent the causal influence directions. The causal reasoning fusion refers to substituting the environmental data vectors and multimodal physical structure feature vectors into the structure-environment causal graph for reasoning, including causal effect prediction, residual calculation, and causal interpretation. In step three, local anomaly indicator vectors and global anomaly indicator vectors are established through the local perception layer and the global perception layer, respectively. The specific details are as follows: Reasoning is performed on the nodes representing local components in the structure-environment causal graph and their directly connected environmental nodes, and anomaly scores for each local component are output. Then, a local anomaly index vector containing the anomaly scores of multiple local components is constructed. Finally, a global anomaly index vector is established through a global perception layer.

7. The method for early warning of structural anomalies in transmission towers based on multi-source fusion sensing as described in claim 1, characterized in that, Step four involves determining whether the local anomaly index vector and the global anomaly index vector require multi-source sensor correction. The specific details are as follows: Real-time noise spectra of multi-source sensors are extracted, and template comparison is performed between the real-time noise spectra and the mapped sensor fingerprint template to establish a log score. Environmental impact and environmental sensitivity analysis of multi-source sensors are performed based on the environmental data vector to establish an adaptation score. A self-verification factor is established based on the log score and the adaptation score. Adjacent sensors and equivalent physical quantity sensors of multi-source sensors are identified to establish an identification cluster. Cooperative residual calculation is performed using the identification cluster to establish a mutual verification factor.

8. The method for early warning of structural anomalies in transmission towers based on multi-source fusion sensing as described in claim 1, characterized in that, Step four involves performing self-calibration of the multi-source sensors, the details of which are as follows: Real-time noise spectra of multi-source sensors are extracted, and template comparison is performed between the real-time noise spectra and the mapped sensor fingerprint template. The differences are calculated, and log scores are output based on the magnitude of the differences. High scores indicate that the noise spectrum is highly consistent with the template, while low scores indicate that the noise spectrum has drifted or new noise peaks have appeared. Environmental impact and environmental sensitivity analysis of multi-source sensors are performed based on environmental data vectors. Step four involves performing mutual verification of the multi-source sensors, the details of which are as follows: Based on the topology and sensor configuration table, the system performs neighbor and equivalent physical quantity sensor identification for multi-source sensors. This includes grouping physically adjacent sensors into neighbor sensor identification clusters and sensors measuring the same physical quantity but with different types or principles into equivalent physical quantity sensor identification clusters. The system then uses these identification clusters to perform collaborative residual calculation. It normalizes and weights the self-verification factor and mutual verification factor, outputting a correction factor. Finally, it uses the correction factor as a weight to perform dynamic fusion of local and global anomaly index vectors.

9. The method for early warning of structural anomalies in transmission towers based on multi-source fusion sensing as described in claim 1, characterized in that, Step five details are as follows: The dynamic fusion result is decomposed into a multi-scale component to establish a long-term trend component and a short-term residual component; a mutation detection is performed based on the short-term residual component to establish a short-term anomaly score. Based on the long-term trend components, change points are detected and trend slopes are evaluated to establish long-term anomaly scores. The short-term and long-term abnormal scores are verified and weighted and fused to output a structural anomaly warning.

10. A multi-source fusion sensing early warning device for power transmission tower structural anomalies, characterized in that, include: The adaptive acquisition module is used to put the multi-source sensors into a semi-dormant state. When any multi-source sensor acquires an abnormal local signal or generates an abnormal event, it adaptively activates the multi-source sensor to perform data acquisition and establish a multi-source dataset. The data processing module is used to map the multi-source dataset to a unified structural feature space and establish a multimodal physical structure feature vector. The reasoning and recognition module is used to perform causal reasoning fusion of environmental data vectors and physical structure feature vectors after collecting environmental data vectors, using a structure-environment causal graph. Local anomaly index vectors and global anomaly index vectors are established through a local perception layer and a global perception layer, respectively. The first calibration module is used to perform self-calibration and mutual verification of multi-source sensors, establish calibration factors, and use the calibration factors to perform dynamic fusion of local anomaly index vectors and global anomaly index vectors. The second correction module is used to perform spatial-temporal adaptive anomaly identification on the dynamic fusion results and output structural anomaly warnings. The device performs the multi-source fusion sensing method for early warning of structural anomalies in transmission towers as described in any one of claims 1-9.

Images