Power line earthquake prevention and disaster reduction multidimensional monitoring and early warning system

Abstract

This invention relates to the field of power facility safety monitoring technology, specifically a multi-dimensional monitoring and early warning system for earthquake disaster prevention and mitigation of power lines. The system includes: collecting data through sensor arrays deployed on poles and conductors; utilizing edge computing nodes to perform spatiotemporal registration and fusion of heterogeneous data to construct a multi-dimensional monitoring data cube indexed by geographic coordinates and timestamps, achieving spatiotemporal alignment of multi-source information. Based on this cube, multi-dimensional joint features are extracted and input into a pre-trained multi-scale risk assessment network. This network integrates historical earthquake data to generate risk scenario evolution paths, thereby outputting a structural risk probability sequence. Finally, combined with real-time meteorological and geological data, structured early warning information including warning level, location, and impact range is generated. This system improves the efficiency of fusion processing of massive heterogeneous monitoring data and the dynamic prediction capability of risk assessment, achieving accurate early warning of earthquake disaster risks to power lines.

Description

Technical Field

[0001] This invention relates to the field of power facility safety monitoring technology, and in particular to a multi-dimensional monitoring and early warning system for earthquake prevention and disaster reduction of power lines. Background Technology

[0002] As a critical infrastructure, the safe and stable operation of power lines during natural disasters such as earthquakes is of paramount importance. Currently, deploying various sensors on towers and conductors for condition monitoring has become a common technical approach. However, the monitoring data such as tilt angle, tension, and geological data generated by these sensors are often collected, transmitted, and stored independently within traditional systems, forming multiple parallel data streams. In this mode, data from different sources lack precise temporal and spatial correlation and alignment, making it difficult to accurately characterize the overall state of a specific location at a specific time. Existing technologies struggle to achieve real-time, accurate spatiotemporal fusion of massive heterogeneous monitoring data at the edge, limiting the ability to subsequently perform refined joint feature extraction and analysis.

[0003] At the risk assessment level, existing methods mainly rely on comparing current monitoring data with preset thresholds or performing calculations based on mechanical models with fixed parameters. These methods are essentially static or quasi-static, and their assessment results reflect the current instantaneous risk level of the power line. However, the occurrence and development of disasters such as earthquakes are dynamic processes, and their impact on the power line structure evolves continuously with the intensity and duration of the disaster, as well as the dynamic response of the power line itself. Current technologies lack the ability to deeply integrate historical disaster event data with real-time monitoring, and cannot simulate or predict how the risk of the power line dynamically evolves over time and space during the disaster's impact. This limits early warning information to the level of status alerts, failing to provide predictive insights into the course of the disaster's impact. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by proposing a multi-dimensional monitoring and early warning system for earthquake prevention and disaster reduction of power lines.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a multi-dimensional monitoring and early warning system for earthquake prevention and disaster reduction of power lines, comprising: The data acquisition module deploys various types of sensor arrays on the poles and conductors along the power lines. The data fusion module collects heterogeneous monitoring data streams generated by multiple types of sensor arrays in real time through edge computing nodes, performs spatiotemporal registration and fusion processing on the heterogeneous monitoring data streams, and constructs a multidimensional monitoring data cube indexed by geographic coordinates and timestamps. The feature extraction module extracts multi-dimensional joint features reflecting the structural and environmental conditions of the power line based on the multi-dimensional monitoring data cube. The multi-dimensional joint features include tower attitude tilt features, conductor tension features, and geological condition features around the tower foundation. The risk assessment module inputs the multi-dimensional joint features into a pre-trained multi-scale risk assessment network. The multi-scale risk assessment network outputs a structural risk probability sequence for different power line sections and integrates historical earthquake event data to generate a risk scenario evolution path during the internal calculation process of the multi-scale risk assessment network. The early warning generation module generates structured early warning information, including the early warning level, risk location, and estimated impact range, based on the structural risk probability sequence and risk scenario evolution path, combined with real-time meteorological forecast data and regional geological activity monitoring data.

[0006] As a further aspect of the present invention, the step of acquiring heterogeneous monitoring data streams generated by multi-type sensor arrays in real time through edge computing nodes includes: The multi-type sensor array includes dynamic sensors for monitoring displacement and acceleration, strain sensors for monitoring stress, and environmental sensors for monitoring environmental parameters. The heterogeneous monitoring data stream includes time-domain vibration waveforms generated by dynamic sensors, micro-strain sequences generated by strain sensors, and environmental parameter sequences generated by environmental sensors. The edge computing nodes deployed on the poles synchronously receive raw monitoring signals from all dynamic sensors, strain sensors, and environmental sensors within the same power line section. The original monitoring signals are preprocessed, including timestamp synchronization, unit unification, and data packet parsing, to generate time-domain vibration waveforms, micro-strain sequences, and environmental parameter sequences with a unified time reference and standardized data format, which together constitute a heterogeneous monitoring data stream. The heterogeneous monitoring data stream is segmented and cached according to a preset data slice length, and then the corresponding power line section identifier and tower location code are attached before being transmitted to the upper-level data processing node.

[0007] As a further aspect of the present invention, the step of performing spatiotemporal registration and fusion processing on the heterogeneous monitoring data stream to construct a multidimensional monitoring data cube indexed by geographic coordinates and timestamps includes: Based on the timing signals of the Global Navigation Satellite System and the geographical coordinates of the towers, each data point in the heterogeneous monitoring data stream is assigned a unified Coordinated Universal Time timestamp and latitude and longitude coordinates; Using the tower location as a spatial reference point, the Kriging spatial interpolation method is used to fuse discrete sensor data into continuous spatial field data covering the power line corridor. The continuous spatial field data are aligned according to the time dimension to form a multidimensional monitoring data cube that is continuous in both the spatial and time dimensions and includes displacement field, strain field and environmental parameter field.

[0008] As a further aspect of the present invention, the extraction of multi-dimensional joint features reflecting the structural and environmental conditions of power lines based on the multi-dimensional monitoring data cube includes: From the displacement field data of the multidimensional monitoring data cube, the standard deviation and peak factor of the displacement time history curve of each tower position in the three-dimensional direction are calculated as the tower attitude tilt angle characteristics. From the strain field data of the multidimensional monitoring data cube, the equivalent stress time history of the conductor suspension point is calculated, and the dynamic tension variation range of the conductor is inferred from the stress-strain relationship, which serves as the characteristic of the conductor tension. By combining surface acceleration data and soil moisture data near the tower foundation in the multidimensional monitoring data cube, the shear wave velocity and liquefaction potential of the soil around the tower foundation are assessed as characteristics of the geological state around the tower foundation.

[0009] As a further aspect of the present invention, the step of inputting the multi-dimensional joint features into a pre-trained multi-scale risk assessment network includes: The multi-scale risk assessment network includes a local tower scale assessment branch and a line corridor area scale assessment branch. In the local tower scale assessment branch, the tower attitude tilt angle characteristics, conductor tension characteristics and geological state characteristics around the tower foundation are spliced ​​together, and the local spatial correlation pattern is extracted through convolutional layer to output the structural risk probability of a single tower. In the regional scale assessment branch of the line corridor, the multi-dimensional joint features of all towers in the same line section are arranged in spatial order, and the risk propagation pattern along the line direction is extracted by recurrent neural network to output the overall risk probability of the line section. The outputs of the local tower scale assessment branch and the line corridor area scale assessment branch are combined to form a structural risk probability sequence.

[0010] As a further aspect of the present invention, the step of integrating historical earthquake event data to generate a risk scenario evolution path during the internal calculation process of the multi-scale risk assessment network includes: A scene memory is integrated into the multi-scale risk assessment network. The scene memory stores power line response data and corresponding damage modes recorded in historical earthquake events. The extracted multi-dimensional joint features are matched with historical patterns in the scene memory database to retrieve the K most similar historical risk scenes. Using retrieved historical risk scenarios as samples, a probabilistic graphical model is used to deduce the path and intensity changes of the risk propagating from the initial position along the route corridor under the current environmental parameters and route conditions, generating multiple risk scenario evolution paths with occurrence probability weights.

[0011] As a further aspect of the present invention, the generation of structured early warning information, including early warning level, risk location, and estimated impact range, based on the structural risk probability sequence and risk scenario evolution path, combined with real-time meteorological forecast data and regional geological activity monitoring data, includes: Set multi-level risk probability thresholds and mark towers or sections in the structural risk probability sequence that exceed the corresponding thresholds as warning points of concern; Based on the evolution path of the risk scenario, determine the risk propagation direction of the warning focus and the adjacent towers or sections affected, and delineate the estimated impact range; By integrating wind speed and precipitation data from real-time weather forecasts with microseismic activity frequency data from regional geological activity monitoring, the risk level of the warning focus is dynamically corrected, ultimately generating structured warning information that includes a polygonal coordinate set containing the warning level, risk location coordinates, and estimated impact range.

[0012] As a further aspect of the present invention, it also includes: The closed-loop feedback module pushes structured early warning information to the line operation and maintenance decision support interface, and at the same time starts the early warning response plan matching process in the background, calling the pre-stored emergency response plan that matches the early warning level and risk location. Based on the structured early warning information pushed by the system, the monitoring parameters of multiple types of sensor arrays are dynamically adjusted to implement intensified and key monitoring of high-risk areas, forming a closed-loop feedback between monitoring and early warning. The process of pushing structured early warning information to the line operation and maintenance decision support interface, and simultaneously initiating an early warning response plan matching process in the background, includes: The structured early warning information is visualized and rendered on the digital map of power lines in the decision support interface by overlaying geographic information system layers, highlighting the towers and estimated impact ranges of different early warning levels. Based on the warning level and risk location in the structured early warning information, the system automatically retrieves the preset emergency response plan from the contingency plan database. The emergency response plan includes personnel dispatch plan, material allocation list and on-site handling procedure.

[0013] As a further aspect of the present invention, the dynamic adjustment of monitoring parameters of multiple types of sensor arrays based on the pushed structured early warning information includes: When the structured early warning information indicates that the early warning level of a certain section is raised, an instruction is sent to the edge computing node corresponding to the section to increase the sampling frequency of the dynamic sensor and the strain sensor. At the same time, the environmental sensors are instructed to increase the monitoring density of specific meteorological parameters such as precipitation, wind direction and wind speed, and to activate the video monitoring devices deployed in the area for coordinated observation.

[0014] As a further aspect of the present invention, the implementation of intensified and focused monitoring of high-risk areas to form a closed-loop feedback mechanism for monitoring and early warning includes: During the encrypted monitoring process, edge computing nodes upload higher-density heterogeneous monitoring data streams in real time; The system uses the data stream obtained from encrypted monitoring to update the multidimensional monitoring data cube of the corresponding section, and re-executes multidimensional joint feature extraction and multi-scale risk assessment; By comparing the structured early warning information generated in the new round of assessment with the early warning information in the previous round, the risk evolution is verified and the early warning level and response plan are dynamically adjusted, thereby realizing a closed-loop feedback control in which monitoring data drives early warning and early warning guides the monitoring focus.

[0015] Compared with the prior art, the advantages and positive effects of the present invention are as follows: By performing spatiotemporal registration and fusion of heterogeneous monitoring data streams at edge computing nodes, a multidimensional monitoring data cube indexed by geographic coordinates and timestamps is constructed. This technical solution achieves unified spatiotemporal benchmark alignment and structured organization of multi-source, asynchronous, and heterogeneous sensor data. It solves the fragmentation problem of monitoring data in both time and space, enabling all state parameters of any tower location at any time to be accessed and analyzed as a complete, accurate, and traceable data entity. This provides a unique and reliable data source for extracting multidimensional joint features with physical consistency and spatiotemporal correlation, ensuring that subsequent features accurately reflect the comprehensive state of the line at that spatiotemporal point.

[0016] This approach integrates historical earthquake event data into a pre-trained multi-scale risk assessment network to generate risk scenario evolution paths. It embeds structural response patterns and damage development sequences from historical disaster cases as prior knowledge into the network's inference and computation process, enabling dynamic extrapolation capabilities for risk assessment. Based on real-time characteristics, the network can simulate the changing trends of risk probabilities of various components of a railway line over time and spatially under specific disaster disturbances. The output not only includes the current risk level but also reveals the potential dynamic development process and spatial transmission relationship of the risk, achieving a transformation from static state assessment to dynamic process prediction. Attached Figure Description

[0017] Figure 1 This is a timing diagram of the multi-dimensional monitoring and early warning system for earthquake prevention and disaster reduction of power lines as described in this invention; Figure 2A flowchart for the acquisition and preprocessing of heterogeneous monitoring data streams; Figure 3 A flowchart for the construction of a multidimensional monitoring data cube; Figure 4 This is a probability distribution diagram of structural risks of towers in the line section; Figure 5 This is a comparison chart of the multi-dimensional joint features of the tower. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0019] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0020] See Figure 1 The overall implementation scheme of the multi-dimensional monitoring and early warning system for earthquake prevention and disaster reduction of power lines includes: a data acquisition module, a data fusion module, a feature extraction module, a risk assessment module, and an early warning generation module connected in sequence. The data acquisition module deploys multi-type sensor arrays on the towers and conductors along the power line. The data fusion module collects heterogeneous monitoring data streams generated by the multi-type sensor arrays in real time through edge computing nodes, performs spatiotemporal registration and fusion processing on the heterogeneous monitoring data streams, and constructs a multi-dimensional monitoring data cube indexed by geographic coordinates and timestamps. The feature extraction module extracts features reflecting the structural and environmental conditions of the power line based on the multi-dimensional monitoring data cube. The risk assessment module inputs the multi-dimensional joint features, including tower attitude and tilt characteristics, conductor and ground wire tension characteristics, and geological conditions around the tower foundation, into a pre-trained multi-scale risk assessment network. The multi-scale risk assessment network outputs structural risk probability sequences for different power line sections and integrates historical earthquake event data to generate risk scenario evolution paths during its internal calculation process. The early warning generation module generates structured early warning information, including early warning level, risk location, and estimated impact range, based on the structural risk probability sequence and risk scenario evolution path, combined with real-time meteorological forecast data and regional geological activity monitoring data.

[0021] In one embodiment of the present invention, see [reference] Figure 2 In practice, various types of sensor arrays deployed along the power line towers and conductors generate raw monitoring signals. For example, in a mountainous high-voltage line section containing five towers, dynamic sensors monitoring displacement and acceleration are deployed on the tower body, crossarms, and conductor suspension points of each tower; strain sensors monitoring stress are deployed on the key stress-bearing parts of the tower material and on the conductors of each tower; and environmental sensors monitoring environmental parameters are deployed near the foundation of each tower. The dynamic sensors generate time-domain vibration waveforms at a rate of 200 samples per second, the strain sensors generate micro-strain sequences at a rate of 100 samples per second, and the environmental sensors generate environmental parameter sequences at a rate of 1 sample per second. These signals together constitute the raw, unprocessed monitoring data.

[0022] In some embodiments, edge computing nodes deployed on each tower synchronously receive raw monitoring signals from all dynamic sensors, strain sensors, and environmental sensors on that tower and adjacent towers within the same line segment via wired or wireless communication protocols. The edge computing nodes have a built-in global navigation satellite system timing module, which adds a unified timestamp based on Coordinated Universal Time (UTC) to each received data packet. The edge computing nodes preprocess the raw monitoring signals, including timestamp synchronization, dimensional unification, and data packet parsing. Timestamp synchronization aligns the data time base of all sensors to the UTC clock of the edge computing node. Dimensional unification converts the voltage values ​​output by dynamic sensors into acceleration values ​​in meters per second squared (m²) or displacement values ​​in meters, and converts the resistance change values ​​output by strain sensors into micro-strain values. Data packet parsing decapsulates data according to a predefined communication protocol to extract valid waveform or sequence data.

[0023] In practical implementation, after preprocessing, time-domain vibration waveforms, micro-strain sequences, and environmental parameter sequences with a unified time reference and standardized data format are generated. These sequences together constitute a heterogeneous monitoring data stream. The time-domain vibration waveforms contain three-dimensional acceleration or displacement data, the micro-strain sequences contain strain time histories from multiple measurement points, and the environmental parameter sequences contain time-series readings of parameters such as temperature, humidity, wind speed, and precipitation. The heterogeneous monitoring data stream is segmented and buffered within edge computing nodes according to a preset data slice length. The data slice length is set considering data transmission efficiency and subsequent processing requirements; for example, it can be set to 10 seconds or 30 seconds per data slice. Each buffered data slice is appended with a corresponding power line segment identifier and tower location code. The segment identifier is used to identify the segment to which the line belongs, and the tower location code is used to accurately locate the tower from which the data originates. After the appended information is completed, the data slice is transmitted to the upper-level data processing node via a communication network.

[0024] Optionally, when generating standardized data with a unified time base, edge computing nodes employ an interpolation synchronization algorithm to handle minor time deviations caused by network latency or differences in sensor response times. Assume the original data timestamp sequence of a certain sensor is... The corresponding observation sequence is The standard timestamp sequence for edge computing nodes is For each time point in the standard timestamp sequence Find the matching data in the original timestamp sequence of the sensor. The interval is used to calculate the observation value at the standard time using linear interpolation. : in: and These are two adjacent timestamps in the sensor's raw data. and These are the corresponding observed values. It is a unified standard timestamp required by edge computing nodes. It is obtained after interpolation and synchronization. Standardized observations at specific times. In this way, data from different sensors with different native sampling times are aligned to the same high-precision time base, thereby generating time-domain vibration waveforms, micro-strain sequences, and environmental parameter sequences with strictly synchronized timestamps. It can be understood that transmitting heterogeneous monitoring data stream slices with power line section identifiers and tower location codes to upper-level data processing nodes completes the transformation from field physical signals to standardized, structured data streams. The upper-level data processing nodes, based on the power line section identifiers and tower location codes, classify and summarize the data slices from different edge computing nodes, providing input for subsequent spatiotemporal fusion processing.

[0025] In one embodiment of the present invention, see [reference] Figure 3In practice, the data processing node receives heterogeneous monitoring data stream slices from multiple edge computing nodes. Each data point is pre-stamped with a preliminary timestamp and tower location code. Based on the GPS timing signal and the precise geographic coordinates of the tower, each data point in the heterogeneous monitoring data stream is assigned a unified UTC timestamp and latitude / longitude coordinates. For example, for tower T23 within the line section, its pre-determined latitude / longitude coordinates are (E118.764532°, N32.051247°). The acceleration data packet collected by the dynamic sensor on this tower at a certain moment is calibrated by the data processing node in conjunction with its reception time and the satellite timing signal, and finally assigned a UTC timestamp of "2025-07-15 08:30:15.123456UTC". The geographic coordinates of tower T23 are then permanently associated with this data point.

[0026] In some embodiments, the location of each tower is used as a spatial reference point, and the Kriging spatial interpolation method is employed to fuse discrete sensor data into continuous spatial field data covering the power line corridor. The Kriging spatial interpolation method is a spatial estimation method based on variogram theory. It assumes that a series of known points exist within the plane of the power line corridor. Its observed value is (For example, the horizontal displacement value at tower T23), it is necessary to estimate the unknown points. value The Kriging estimator is expressed as a linear combination of observations at known points: in: It is the number of known points used for estimation. It is to assign a known point The weighting coefficients, It is a known point Observations at that location It is an unknown point The estimated value at that location. Weighting coefficients. By solving the Kriging equations, it is determined that the estimation error must be minimized and the estimation itself must be unbiased. Using this method, the observed displacement, strain, and environmental parameters at tower locations can be extrapolated to various points on the conductor sag between towers and any point on the ground along the line corridor, thus generating a continuous displacement field, continuous strain field, and continuous environmental parameter field covering the entire line space. The continuous spatial field data are aligned along the time dimension, and the continuous spatial field data in each time slice are arranged with a fixed time interval (e.g., 1 second) to form a multidimensional monitoring data cube that is continuous in both spatial and temporal dimensions, containing displacement, strain, and environmental parameter fields.

[0027] In practical implementation, multi-dimensional joint features reflecting the structural and environmental conditions of power lines are extracted based on a multi-dimensional monitoring data cube. From the displacement field data of the multi-dimensional monitoring data cube, displacement time history curves in three dimensions (along the line, across the line, and vertical) are extracted for each tower location. The standard deviation and peak factor of each displacement time history curve are calculated. The standard deviation reflects the average sway amplitude of the tower over a period of time, and the peak factor reflects the ratio of extreme displacement to average sway amplitude. The standard deviation and peak factor in these three directions are used together as the tower attitude tilt angle feature. From the strain field data of the multi-dimensional monitoring data cube, the strain monitoring point at the conductor suspension point is located, and the equivalent stress time history at this point is calculated. The equivalent stress time history comprehensively considers the combined effect of multi-directional stress. Through the stress-strain relationship of the conductor material, the dynamic tension range of the conductor is inferred from the equivalent stress time history. The dynamic tension range includes the minimum, maximum, and average rate of change of the tension force, and is used as the tension force feature of the conductor.

[0028] Optionally, by combining surface acceleration data and soil moisture data near the tower foundation in the multi-dimensional monitoring data cube, the shear wave velocity and liquefaction potential of the soil surrounding the tower foundation can be assessed. Surface acceleration data comes from dynamic sensors deployed on the surface near the tower foundation, and soil moisture data comes from environmental sensors. Using the time-history data of surface acceleration, the average shear wave velocity of the soil is estimated through spectral analysis. Shear wave velocity is an important parameter characterizing soil density and stiffness. Simultaneously, combining soil moisture, peak surface acceleration, and prior information on soil type, a standardized liquefaction discrimination method is used to calculate the liquefaction potential index at this location under the current state. The liquefaction potential index is a dimensionless value representing the probability of soil liquefaction. The estimated shear wave velocity and the calculated liquefaction potential index are used together as characteristics of the geological state around the tower foundation. It is understandable that, through the above process, the characteristics of tower attitude tilt angle, conductor tension, and geological condition around the tower foundation are systematically extracted from the multidimensional monitoring data cube. These multidimensional joint features quantitatively describe the comprehensive state of the power line from three aspects: structural response, component stress, and foundation environment, providing structured input for subsequent risk assessment.

[0029] In one embodiment of the present invention, in a specific implementation, the pre-trained multi-scale risk assessment network receives multi-dimensional joint feature data from the feature extraction module. The multi-scale risk assessment network includes two parallel data processing paths, namely a local tower scale assessment branch and a line corridor area scale assessment branch. The local tower scale assessment branch focuses on the status assessment of a single tower and its directly related components, while the line corridor area scale assessment branch focuses on the overall and cascading risk assessment of a line segment composed of continuous multi-tower towers. In some embodiments, in the local tower scale assessment branch, the tower attitude tilt angle characteristics, conductor tension characteristics, and geological state characteristics around the tower foundation of a single tower are spliced ​​together. The tower attitude tilt angle characteristics may contain six numerical components (standard deviation in three directions and peak factor in three directions), the conductor tension characteristics may contain three numerical components (minimum, maximum, and average rate of change of dynamic tension), and the geological state characteristics around the tower foundation contain two numerical components (shear wave velocity and liquefaction potential index). After splicing, a fixed-length feature vector is formed. This feature vector is processed by one or more one-dimensional convolutional layers. The filter of the convolutional layer slides along the dimension of the feature vector to extract the local spatial correlation patterns and nonlinear combination relationships between different feature components. The feature map processed by the convolutional layer is then comprehensively judged by a fully connected layer, and finally outputs a scalar value between 0 and 1, which represents the structural risk probability of a single tower.

[0030] In practical implementation, in the regional scale assessment branch of the line corridor, the multi-dimensional joint features of all towers within the same line segment are arranged in spatial order. This spatial order is typically determined by the geographical location of the towers along the line, from front to back or from back to front. This arrangement forms a feature matrix, where rows correspond to the tower order and columns correspond to the multi-dimensional joint feature components of each tower. This feature matrix is ​​input as a time series to a recurrent neural network (RNN). The RNN treats the spatial sequence along the line corridor as a special type of time series. The hidden state of the RNN is updated when traversing each row of the feature matrix (i.e., each tower), thereby capturing and memorizing the spatial patterns and dependencies of risk features propagating along the line corridor. The RNN ultimately outputs a vector representing the overall risk state of the entire line segment. This vector is then mapped to the overall risk probability of the line segment via a fully connected layer. Optionally, when updating the hidden state to extract risk propagation patterns along the line corridor, the RNN employs a gated recurrent unit structure. Its state update process can be described by the following formula: Where: subscript Represents the index in the spatial sequence, corresponding to the first... The position of the base tower in the sequence. It is the input to the recurrent neural network, representing the first The multi-dimensional joint feature vector of the base tower Before it was finished The hidden state vector obtained after the base tower. It is input The weight matrix, It is a preceding hidden state The weight matrix, It is a bias vector. It is a non-linear activation function. The calculated candidate hidden states are further controlled by a gating mechanism (update gate and reset gate) to obtain the hidden state after the current tower position is updated. This hidden state Encoded up to the number Information on the propagation of spatial risks along the railway line up to the base tower.

[0031] In some embodiments, the outputs of the local tower scale assessment branch and the line corridor area scale assessment branch are fused to form a structural risk probability sequence. The fusion operation adopts a weighted splicing and fully connected integration method. The individual structural risk probabilities of all towers output by the local tower scale assessment branch are arranged in spatial order to form a local risk vector. At the same time, the overall risk probability of the line section output by the line corridor area scale assessment branch is taken as a global scalar. This global scalar is copied and extended into a global influence vector with the same dimension as the local risk vector. The local risk vector and the global influence vector are spliced. The spliced ​​fusion vector is integrated and adjusted through a fully connected layer. The fully connected layer outputs the final structural risk probability sequence. Each probability value in this sequence corresponds to a tower. This probability value considers both the local state of the tower itself and its relative status under the overall risk background of the line section. It is understandable that, through the aforementioned dual-branch assessment structure of local and regional levels, the multi-scale risk assessment network can simultaneously output a refined risk probability for each specific tower and a structural risk probability sequence reflecting the overall risk status of the line section. The output of the local tower-scale assessment branch provides the accuracy of risk location, while the output of the line corridor regional-scale assessment branch provides a perspective on risk chaining and propagation. The fusion of the two makes the final structural risk probability sequence both accurate and systematic.

[0032] In one embodiment of the present invention, in a specific implementation, the multi-scale risk assessment network integrates a scene memory library, which stores power line response data and corresponding damage modes recorded in historical earthquake events. The historical earthquake event data comes from post-earthquake investigation records of power lines damaged in previous earthquakes, retrospective historical monitoring data of the same line, and laboratory simulation test data. Each record includes basic earthquake parameters, multi-dimensional joint feature data sequence of each tower in the line section, and a description of the finally observed damage mode. The damage mode may include tower buckling, foundation hollowing, conductor and ground wire breakage, or insulator detachment, etc.

[0033] In some embodiments, the scene memory database organizes historical data in a structured tabular format for efficient retrieval. The tables include fields such as event number, magnitude, epicenter distance from the line, tower attitude and tilt characteristic vectors for each tower, conductor tension characteristic vectors, geological condition characteristic vectors around the tower foundations, and recorded main failure modes. See Table 1 for a simplified scene memory database table.

[0034] Table 1: Historical Earthquake Event Scene Memory Database In practice, the extracted multi-dimensional joint features are matched with historical patterns in the scene memory database for similarity. For the line segment to be evaluated, the system calculates the similarity between the current feature vector of each tower within the segment and the feature vector corresponding to each historical record in the scene memory database. The similarity is measured using the reciprocal of the weighted Euclidean distance, with weights allocated based on the contribution of each feature to the damage pattern. Assume the feature vector of the current i-th tower is... The feature vector of the j-th historical record in the scene memory bank is Then its distance The calculation is as follows: in: It is the total dimension of the feature vector. It is the k-th component of the current eigenvector. It is the k-th component of the historical eigenvector. This is the weighting coefficient assigned to the k-th feature component, which is predetermined through statistical analysis of historical data. Distance The smaller the value, the higher the similarity. The system retrieves the K most similar historical risk scenarios with the smallest distance for the current tower. Optionally, using the K most similar historical risk scenarios as samples, the system uses a probabilistic graphical model to deduce the evolution path of the risk scenario. The probabilistic graphical model constructs a directed graph with towers as nodes and risk propagation probability as edges. Each historical risk scenario provides prior information about the risk initiation node (i.e., the tower where the damage started) and the path of risk propagation along the line corridor. The model integrates the environmental parameters of the current line section and the state of all towers to calculate the conditional probability of risk propagating from the current warning concern point to adjacent towers, generating multiple possible paths starting from the initiation point with different branches. Each path is attached with an occurrence probability weight calculated based on the historical scenario matching degree and the current state, thus forming a set of risk scenario evolution paths.

[0035] In some embodiments, structured early warning information is generated based on the structural risk probability sequence and the risk scenario evolution path. Multi-level risk probability thresholds are set, for example, a low-risk threshold of 0.3, a medium-risk threshold of 0.6, and a high-risk threshold of 0.8. Towers or sections that exceed the corresponding thresholds in the structural risk probability sequence are marked as early warning concern points. Based on the risk scenario evolution path, the risk propagation direction of the early warning concern points and the adjacent towers or sections that may be affected are determined, thereby delineating the estimated impact range. The estimated impact range is geographically represented as a polygonal area that includes the early warning concern points and their downstream towers that may be affected.

[0036] It is understandable that the warning level is dynamically corrected by combining real-time weather forecast data and regional geological activity monitoring data. Real-time weather forecast data provides wind speed and precipitation predictions for a period of time in the future, while regional geological activity monitoring data provides the frequency of monitored microseismic activity. The system adjusts the risk level of the warning focus point according to predefined rules. For example, when the liquefaction potential index in the geological state characteristics of the area surrounding the warning focus point is high and the real-time precipitation forecast shows that there will be heavy rain, the risk level of the point may be increased by one level. The final structured warning information includes the determined warning level, risk location coordinates, and a set of polygon coordinates of the estimated impact range.

[0037] See Figure 4This is a probability distribution map of structural risks for 20 power line towers, showing the changing trends of structural risk probabilities. It is a core visual output of the multi-scale risk assessment stage in the multi-dimensional monitoring and early warning system for earthquake prevention and disaster reduction of power lines. The risk probability gradually increases from 0.12 at tower 01, peaking at 0.85 at tower 09, then gradually decreases to 0.15 at tower 16, before slightly rising again. This clearly identifies the highest-risk section of the line, providing a basis for scheduling maintenance resources. The risk attenuates from tower 09 to both sides, which can be used to deduce the risk diffusion path. Towers exceeding the threshold can trigger corresponding levels of early warning, guiding intensified monitoring and emergency response. Intensified monitoring is implemented for towers 08–10, increasing the sensor sampling frequency and focusing on conductor and ground wire tension and foundation geological conditions.

[0038] In one embodiment of the present invention, in a specific implementation, the closed-loop feedback module pushes the structured early warning information output by the early warning generation module to the line operation and maintenance decision support interface. The structured early warning information includes the early warning level, risk location coordinates, and a set of polygon coordinates of the estimated impact range. The line operation and maintenance decision support interface is a software platform that integrates a geographic information system, line asset management, and operation and maintenance processes. At the same time, the system automatically starts the early warning response plan matching process in the background. The plan matching process calls the pre-stored emergency response plan that matches the early warning level and risk location based on the key fields in the structured early warning information. In some embodiments, the structured early warning information is pushed to the line operation and maintenance decision support interface through a geographic information system (GIS) layer overlay. The structured early warning information is converted into GIS-recognizable vector layer data. The early warning level is visualized on the digital map of the power line in the decision support interface using different colors and highlighting effects. For example, high risk corresponds to a flashing red icon, medium risk to a solid orange icon, and low risk to a yellow icon. The risk location coordinates are marked as the center point of the layer, and the estimated impact range polygon coordinate set is rendered as a semi-transparent overlay area, allowing operation and maintenance personnel to intuitively grasp the geographical distribution and spatial range of the risk. Based on the early warning level and risk location in the structured early warning information, the system automatically performs a retrieval operation in the contingency plan database. The contingency plan database is a relational database that stores various preset emergency response plans. The retrieval conditions include exact matching of the early warning level field and spatial correlation matching between the risk location and the section under the responsibility of the contingency plan. The automatically retrieved emergency response plans include detailed personnel dispatch plans, material allocation lists, and on-site handling procedures. These plans are pushed to the decision support interface for operation and maintenance personnel to call and execute.

[0039] In practical implementation, based on the pushed structured early warning information, the closed-loop feedback module dynamically adjusts the monitoring parameters of multiple types of sensor arrays. When the structured early warning information indicates that the warning level of a certain section is raised, for example, from "yellow" low risk to "orange" medium risk, the system generates a control command and sends it to the edge computing node corresponding to that section. The control command increases the sampling frequency of all dynamic sensors and strain sensors in that section. The sampling frequency of dynamic sensors may be increased from 200 samples per second to 500 samples per second, and the sampling frequency of strain sensors may be increased from 100 samples per second to 200 samples per second. At the same time, the control command requires environmental sensors to increase the monitoring density of specific meteorological parameters such as precipitation, wind direction, and wind speed. For example, the sampling and reporting interval of precipitation and wind speed may be shortened from once per minute to once every ten seconds. The video monitoring device deployed in that section is also activated for linked observation. The video monitoring device continuously photographs or records high-risk towers and their conductors at preset positions and zoom levels.

[0040] Optionally, when the system adjusts the warning level by integrating real-time weather forecast data and regional geological activity monitoring data, a quantified correction coefficient may be used. This coefficient affects the determination of the final warning level; correction coefficient. The calculation comprehensively considers external dynamic factors: in: It is a forecast precipitation intensity index for a specified future period. It is the forecast average wind speed. This is a wind speed reference value. This refers to the recent frequency of microseismic activity in the region. This is a reference value for microseismic frequency. These are weighting factors corresponding to precipitation, wind speed, and microseismic activity, respectively. These weighting factors are pre-set based on historical disaster correlation analysis, and the resulting correction coefficients are calculated accordingly. The original structural risk probability output by the multi-scale risk assessment network is multiplied to obtain a comprehensive risk probability value adjusted for environmental factors. This comprehensive risk probability value is used to finally determine the warning level.

[0041] In some embodiments, high-risk sections are subject to encrypted and focused monitoring. During encrypted monitoring, edge computing nodes operate at a higher sampling frequency and monitoring density, generating heterogeneous monitoring data streams with higher spatiotemporal resolution and uploading them to data processing nodes in real time. The system utilizes the data streams obtained from encrypted monitoring to quickly update the multidimensional monitoring data cube for the corresponding section. It then re-executes the multidimensional joint feature extraction and multi-scale risk assessment process using the updated multidimensional monitoring data cube to generate a new round of structured early warning information. This new round of structured early warning information is compared with the previous round, including changes in warning levels, increases or decreases in risk locations, and the evolution of the estimated impact range. The system verifies whether the actual evolution of the risk matches the predicted risk scenario evolution path and dynamically adjusts the warning level and response plan based on the comparison results. This includes maintaining, upgrading, downgrading the warning, or changing the focus of the emergency response plan, thereby achieving closed-loop feedback control where monitoring data drives early warning and early warning guides monitoring priorities.

[0042] See Figure 5 This is a multi-dimensional joint feature comparison chart of 10 power poles. The chart displays three core feature values: tilt angle, tension, and geological condition. It is a key visual output of the feature extraction stage in the power line earthquake prevention and disaster reduction system, used to assess the structural health of the poles. Pole 8 has the highest tension in the entire section, while its tilt angle is at a moderate level. Combined with stable geological conditions, this suggests that abnormal conductor-to-ground wire tension is the main risk source, requiring close inspection of the hardware and suspension points. Pole 4 and 9 have significantly higher tilt angles than other poles. Although their tension and geological conditions are not obviously abnormal, the risk of deformation of the pole foundation or tower structure should be noted. The overall geological condition feature values ​​are low and stable, indicating that the geological conditions of the line corridor are good, and the main risks come from the pole structure itself and conductor-to-ground wire tension.

[0043] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A multi-dimensional monitoring and early warning system for earthquake prevention and disaster reduction of power lines, characterized in that, include: The data acquisition module deploys various types of sensor arrays on the poles and conductors along the power lines. The data fusion module collects heterogeneous monitoring data streams generated by multiple types of sensor arrays in real time through edge computing nodes, performs spatiotemporal registration and fusion processing on the heterogeneous monitoring data streams, and constructs a multidimensional monitoring data cube indexed by geographic coordinates and timestamps. The feature extraction module extracts multi-dimensional joint features reflecting the structural and environmental conditions of the power line based on the multi-dimensional monitoring data cube. The multi-dimensional joint features include tower attitude tilt features, conductor tension features, and geological condition features around the tower foundation. The risk assessment module inputs the multi-dimensional joint features into a pre-trained multi-scale risk assessment network. The multi-scale risk assessment network outputs a structural risk probability sequence for different power line sections and integrates historical earthquake event data to generate a risk scenario evolution path during the internal calculation process of the multi-scale risk assessment network. The early warning generation module generates structured early warning information, including the early warning level, risk location, and estimated impact range, based on the structural risk probability sequence and risk scenario evolution path, combined with real-time meteorological forecast data and regional geological activity monitoring data.

2. The multi-dimensional monitoring and early warning system for earthquake prevention and disaster reduction of power lines according to claim 1, characterized in that, The heterogeneous monitoring data stream generated by real-time acquisition of multiple types of sensor arrays through edge computing nodes includes: The multi-type sensor array includes dynamic sensors for monitoring displacement and acceleration, strain sensors for monitoring stress, and environmental sensors for monitoring environmental parameters. The heterogeneous monitoring data stream includes time-domain vibration waveforms generated by dynamic sensors, micro-strain sequences generated by strain sensors, and environmental parameter sequences generated by environmental sensors. The edge computing nodes deployed on the poles synchronously receive raw monitoring signals from all dynamic sensors, strain sensors, and environmental sensors within the same power line section. The original monitoring signals are preprocessed, including timestamp synchronization, unit unification, and data packet parsing, to generate time-domain vibration waveforms, micro-strain sequences, and environmental parameter sequences with a unified time reference and standardized data format, which together constitute a heterogeneous monitoring data stream. The heterogeneous monitoring data stream is segmented and cached according to a preset data slice length, and then the corresponding power line section identifier and tower location code are attached before being transmitted to the upper-level data processing node.

3. The multi-dimensional monitoring and early warning system for earthquake prevention and disaster reduction of power lines according to claim 1, characterized in that, The process of performing spatiotemporal registration and fusion on the heterogeneous monitoring data streams to construct a multidimensional monitoring data cube indexed by geographic coordinates and timestamps includes: Based on the timing signals of the Global Navigation Satellite System and the geographical coordinates of the towers, each data point in the heterogeneous monitoring data stream is assigned a unified Coordinated Universal Time timestamp and latitude and longitude coordinates; Using the tower location as a spatial reference point, the Kriging spatial interpolation method is used to fuse discrete sensor data into continuous spatial field data covering the power line corridor. The continuous spatial field data are aligned according to the time dimension to form a multidimensional monitoring data cube that is continuous in both the spatial and time dimensions and includes displacement field, strain field and environmental parameter field.

4. The multi-dimensional monitoring and early warning system for earthquake prevention and disaster reduction of power lines according to claim 3, characterized in that, Based on the multidimensional monitoring data cube, multidimensional joint features reflecting the structural and environmental conditions of power lines are extracted, including: From the displacement field data of the multidimensional monitoring data cube, the standard deviation and peak factor of the displacement time history curve of each tower position in the three-dimensional direction are calculated as the tower attitude tilt angle characteristics. From the strain field data of the multidimensional monitoring data cube, the equivalent stress time history of the conductor suspension point is calculated, and the dynamic tension variation range of the conductor is inferred from the stress-strain relationship, which serves as the characteristic of the conductor tension. By combining surface acceleration data and soil moisture data near the tower foundation in the multidimensional monitoring data cube, the shear wave velocity and liquefaction potential of the soil around the tower foundation are assessed as characteristics of the geological state around the tower foundation.

5. A multi-dimensional monitoring and early warning system for earthquake prevention and disaster reduction of power lines according to claim 1, characterized in that, The step of inputting the multi-dimensional joint features into a pre-trained multi-scale risk assessment network includes: The multi-scale risk assessment network includes a local tower scale assessment branch and a line corridor area scale assessment branch. In the local tower scale assessment branch, the tower attitude tilt angle characteristics, conductor tension characteristics and geological state characteristics around the tower foundation are spliced ​​together, and the local spatial correlation pattern is extracted through convolutional layer to output the structural risk probability of a single tower. In the regional scale assessment branch of the line corridor, the multi-dimensional joint features of all towers in the same line section are arranged in spatial order, and the risk propagation pattern along the line direction is extracted by recurrent neural network to output the overall risk probability of the line section. The outputs of the local tower scale assessment branch and the line corridor area scale assessment branch are combined to form a structural risk probability sequence.

6. A multi-dimensional monitoring and early warning system for earthquake prevention and disaster reduction of power lines according to claim 5, characterized in that, The process of generating risk scenario evolution paths by integrating historical earthquake event data during the internal calculation of the multi-scale risk assessment network includes: A scene memory is integrated into the multi-scale risk assessment network. The scene memory stores power line response data and corresponding damage modes recorded in historical earthquake events. The extracted multi-dimensional joint features are matched with historical patterns in the scene memory database to retrieve the K most similar historical risk scenes. Using retrieved historical risk scenarios as samples, a probabilistic graphical model is used to deduce the path and intensity changes of the risk propagating from the initial position along the route corridor under the current environmental parameters and route conditions, generating multiple risk scenario evolution paths with occurrence probability weights.

7. A multi-dimensional monitoring and early warning system for earthquake prevention and disaster reduction of power lines according to claim 1, characterized in that, Based on the structural risk probability sequence and risk scenario evolution path, combined with real-time meteorological forecast data and regional geological activity monitoring data, structured early warning information is generated, including warning level, risk location, and estimated impact range, including: Set multi-level risk probability thresholds and mark towers or sections in the structural risk probability sequence that exceed the corresponding thresholds as warning points of concern; Based on the evolution path of the risk scenario, determine the risk propagation direction of the warning focus and the adjacent towers or sections affected, and delineate the estimated impact range; By integrating wind speed and precipitation data from real-time weather forecasts with microseismic activity frequency data from regional geological activity monitoring, the risk level of the warning focus is dynamically corrected, ultimately generating structured warning information that includes a polygonal coordinate set containing the warning level, risk location coordinates, and estimated impact range.

8. A multi-dimensional monitoring and early warning system for earthquake prevention and disaster reduction of power lines according to claim 7, characterized in that, Also includes: The closed-loop feedback module pushes structured early warning information to the line operation and maintenance decision support interface, and at the same time starts the early warning response plan matching process in the background, calling the pre-stored emergency response plan that matches the early warning level and risk location. Based on the structured early warning information pushed by the system, the monitoring parameters of multiple types of sensor arrays are dynamically adjusted to implement intensified and key monitoring of high-risk areas, forming a closed-loop feedback between monitoring and early warning. The process of pushing structured early warning information to the line operation and maintenance decision support interface, and simultaneously initiating an early warning response plan matching process in the background, includes: The structured early warning information is visualized and rendered on the digital map of power lines in the decision support interface by overlaying geographic information system layers, highlighting the towers and estimated impact ranges of different early warning levels. Based on the warning level and risk location in the structured early warning information, the system automatically retrieves the preset emergency response plan from the contingency plan database. The emergency response plan includes personnel dispatch plan, material allocation list and on-site handling procedure.

9. A multi-dimensional monitoring and early warning system for earthquake prevention and disaster reduction of power lines according to claim 1, characterized in that, The structured early warning information based on push notifications dynamically adjusts the monitoring parameters of multiple types of sensor arrays, including: When the structured early warning information indicates that the early warning level of a certain section is raised, an instruction is sent to the edge computing node corresponding to the section to increase the sampling frequency of the dynamic sensor and the strain sensor. At the same time, the environmental sensors are instructed to increase the monitoring density of specific meteorological parameters such as precipitation, wind direction and wind speed, and to activate the video monitoring devices deployed in the area for coordinated observation.

10. A multi-dimensional monitoring and early warning system for earthquake prevention and disaster reduction of power lines according to claim 1, characterized in that, The aforementioned implementation of intensified and focused monitoring of high-risk areas to form a closed-loop feedback mechanism for monitoring and early warning includes: During the encrypted monitoring process, edge computing nodes upload higher-density heterogeneous monitoring data streams in real time; The system uses the data stream obtained from encrypted monitoring to update the multidimensional monitoring data cube of the corresponding section, and re-executes multidimensional joint feature extraction and multi-scale risk assessment; By comparing the structured early warning information generated in the new round of assessment with the early warning information in the previous round, the risk evolution is verified and the early warning level and response plan are dynamically adjusted, thereby realizing a closed-loop feedback control in which monitoring data drives early warning and early warning guides the monitoring focus.

Images