Communication tower structure state early warning method and device, storage medium and electronic equipment

Abstract

The application discloses a communication tower structure state early warning method and device, a storage medium and an electronic equipment. It relates to the field of artificial intelligence, and the method comprises the following steps: acquiring a multi-modal data set of a monitored communication tower in a current monitoring period; acquiring a structure dynamics generative causal diagram corresponding to the current monitoring period, wherein the structure dynamics generative causal diagram comprises a plurality of physical nodes and directed edges connecting the plurality of physical nodes, and the directed edges correspond to a vibration transmission model, which is obtained by training a generative adversarial network through a sample data set; performing inference calculation on the multi-modal data set of the current monitoring period and the structure dynamics generative causal diagram corresponding to the current monitoring period to obtain an overall structure health index corresponding to the monitored communication tower; and determining early warning information corresponding to the monitored communication tower according to the overall structure health index. The method solves the problem of low early warning accuracy in the related art that the structure state early warning of the communication tower is realized according to the sensor threshold alarm.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence, and more specifically, to a method and apparatus for early warning of the structural status of a communication tower, a storage medium, and an electronic device. Background Technology

[0002] Structural health monitoring of communication towers primarily employs sensor threshold alarm methods, which suffer from low accuracy and significant delays. Alarms are only triggered when structural damage becomes macroscopically visible and sensor readings exceed preset thresholds, by which time the structure is often already on the verge of failure, failing to provide sufficient maintenance windows. Other supervised learning models rely on training with a large number of fault samples involving structural damage or even collapse. However, communication tower structural failures are low-probability events, resulting in a limited number of valid fault samples, poor model generalization ability, high false alarm and false false alarm rates, and low overall accuracy.

[0003] The existing technologies that rely on sensor threshold alarms to provide early warnings of the structural status of communication towers suffer from low accuracy, and no effective solution has yet been proposed. Summary of the Invention

[0004] The main objective of this application is to provide a method, device, storage medium, and electronic device for early warning of the structural status of communication towers, in order to solve the problem that the accuracy of early warning is low when relying on sensor threshold alarms to achieve early warning of the structural status of communication towers.

[0005] To achieve the above objectives, according to one aspect of this application, a method for early warning of the structural status of a communication tower is provided. The method includes: acquiring a multimodal dataset of the monitored communication tower for the current monitoring period; acquiring a structural dynamics generative causal graph corresponding to the current monitoring period, wherein the structural dynamics generative causal graph includes multiple physical nodes and directed edges connecting the multiple physical nodes, the directed edges representing the causal relationship of vibration transmission between nodes, each directed edge corresponding to a vibration transmission model, the vibration transmission model being obtained by training a generative adversarial network using a sample dataset; performing inference calculations based on the multimodal dataset of the current monitoring period and the structural dynamics generative causal graph corresponding to the current monitoring period to obtain an overall structural health index corresponding to the monitored communication tower; and determining early warning information corresponding to the monitored communication tower based on the overall structural health index.

[0006] Furthermore, based on the multimodal dataset of the current monitoring period and the structural dynamics generative causal graph corresponding to the current monitoring period, inference calculations are performed to obtain the overall structural health index of the monitored communication tower. This includes: inputting the multimodal dataset of the current monitoring period into the structural dynamics generative causal graph corresponding to the current monitoring period; for each directed edge, using the vibration transmission model corresponding to the directed edge to predict the vibration response based on the multimodal dataset of the current monitoring period, and obtaining the predicted value; calculating and smoothing the deviation between the predicted value and the actual observed value corresponding to the directed edge to obtain the target deviation value corresponding to the directed edge; calculating the weighted average based on the target deviation values ​​corresponding to all directed edges and the weight corresponding to each directed edge to obtain the overall deviation value corresponding to the monitored communication tower, and determining the overall structural health index corresponding to the monitored communication tower based on the overall deviation value.

[0007] Further, the structural dynamics generative causal graph is obtained through the following steps: obtaining the target finite element model of the monitored communication tower, wherein the target finite element model is obtained by calibrating the initial finite element model of the monitored communication tower using the historical multimodal dataset of the monitored communication tower; determining multiple physical nodes of the monitored communication tower based on the target finite element model, and constructing an initial causal graph topology based on the multiple physical nodes; and determining the structural dynamics generative causal graph based on the historical multimodal dataset and the initial causal graph topology.

[0008] Furthermore, based on the historical multimodal dataset and the initial causal graph topology, the determination of the structural dynamics generative causal graph includes: configuring an initial vibration transmission model for each directed edge in the initial causal graph topology to obtain the initial structural dynamics generative causal graph; constructing a generative adversarial network based on the initial structural dynamics generative causal graph and a multilayer fully connected neural network; extracting sample datasets from the historical multimodal dataset and training the generative adversarial network based on the sample datasets until the preset conditions are met to obtain the structural dynamics generative causal graph.

[0009] Further, the structural dynamics generative causal graph corresponding to the current monitoring period is obtained through the following steps: acquiring the target multimodal dataset, wherein the target multimodal dataset is the dataset extracted from the multimodal dataset of the previous calibration period of the current monitoring period, excluding data within the time period during which an early warning was triggered, and data where the wind speed is greater than a preset threshold or the temperature exceeds a preset range; calibrating and updating the structural dynamics generative causal graph corresponding to the previous calibration period of the current monitoring period based on the target multimodal dataset, thereby obtaining the structural dynamics generative causal graph corresponding to the current monitoring period.

[0010] Furthermore, determining the early warning information corresponding to the monitored communication tower based on the overall structural health index includes: determining the early warning level corresponding to the monitored communication tower based on the overall structural health index; determining the anomaly diagnosis strategy corresponding to the monitored communication tower based on the early warning level, and executing the anomaly diagnosis strategy to obtain anomaly diagnosis results, wherein the anomaly diagnosis results include at least the anomaly node identifier, anomaly type, or anomaly severity; and generating early warning information based on the early warning level and the anomaly diagnosis results.

[0011] Furthermore, before acquiring the multimodal dataset of the monitored communication tower in the current monitoring period, the method further includes: acquiring the initial multimodal dataset of the monitored communication tower in the current monitoring period, wherein the initial multimodal dataset includes at least wind speed, wind direction, temperature, and structural response data; using the initial multimodal dataset as the multimodal dataset for the current monitoring period, or analyzing based on wind speed, wind direction, temperature, and the target finite element model to obtain predicted structural response data, and calculating based on the structural response data and the predicted structural response data to obtain the target response dataset, and using the target response dataset as the multimodal dataset for the current monitoring period.

[0012] To achieve the above objectives, according to another aspect of this application, a structural status early warning device for a communication tower is provided. The device includes: a first acquisition unit for acquiring a multimodal dataset of the monitored communication tower during the current monitoring period; a second acquisition unit for acquiring a structural dynamics generative causal graph corresponding to the current monitoring period, wherein the structural dynamics generative causal graph includes multiple physical nodes and directed edges connecting the multiple physical nodes, the directed edges representing the causal relationship of vibration transmission between nodes, each directed edge corresponding to a vibration transmission model, and the vibration transmission model being obtained by training a generative adversarial network using a sample dataset; a first processing unit for performing inference calculations based on the multimodal dataset of the current monitoring period and the structural dynamics generative causal graph corresponding to the current monitoring period to obtain an overall structural health index corresponding to the monitored communication tower; and a first determination unit for determining the early warning information corresponding to the monitored communication tower based on the overall structural health index.

[0013] Further, the first processing unit includes: a first processing subunit, used to input the multimodal dataset of the current monitoring period into the structural dynamics generative causal graph corresponding to the current monitoring period; for each directed edge, to predict the vibration response based on the multimodal dataset of the current monitoring period using the vibration transmission model corresponding to the directed edge, and obtain a predicted value; a second processing subunit, used to perform deviation calculation and smoothing processing based on the predicted value and the actual observed value corresponding to the directed edge, and obtain the target deviation value corresponding to the directed edge; and a third processing subunit, used to perform weighted average calculation based on the target deviation values ​​corresponding to all directed edges and the weight corresponding to each directed edge, to obtain the overall deviation value corresponding to the monitored communication tower, and to determine the overall structural health index corresponding to the monitored communication tower based on the overall deviation value.

[0014] Furthermore, the device also includes the following units for obtaining a structural dynamics generative causal graph through the following steps: a third acquisition unit for acquiring the target finite element model of the monitored communication tower, wherein the target finite element model is obtained by calibrating the initial finite element model of the monitored communication tower using the historical multimodal dataset of the monitored communication tower; a second determination unit for determining multiple physical nodes of the monitored communication tower based on the target finite element model, and constructing an initial causal graph topology based on the multiple physical nodes; and a third determination unit for determining the structural dynamics generative causal graph based on the historical multimodal dataset and the initial causal graph topology.

[0015] Furthermore, the third determining unit includes: a fourth processing subunit, used to configure an initial vibration transmission model for each directed edge in the initial causal graph topology to obtain an initial structural dynamics generative causal graph; a fifth processing subunit, used to construct a generative adversarial network based on the initial structural dynamics generative causal graph and a multilayer fully connected neural network; and a sixth processing subunit, used to extract sample datasets from historical multimodal datasets and train the generative adversarial network based on the sample datasets until preset conditions are met to obtain a structural dynamics generative causal graph.

[0016] Furthermore, the device also includes the following units for obtaining the structural dynamics generative causal graph corresponding to the current monitoring period through the following steps: a fourth acquisition unit for acquiring a target multimodal dataset, wherein the target multimodal dataset is a dataset extracted from the multimodal dataset of the previous calibration period of the current monitoring period, excluding data within the time period during which an early warning was triggered, and data where the wind speed is greater than a preset threshold or the temperature exceeds a preset range; and a second processing unit for calibrating and updating the structural dynamics generative causal graph corresponding to the previous calibration period of the current monitoring period based on the target multimodal dataset, thereby obtaining the structural dynamics generative causal graph corresponding to the current monitoring period.

[0017] Furthermore, the first determining unit includes: a first determining subunit, used to determine the warning level corresponding to the monitored communication tower based on the overall structural health index; a second determining subunit, used to determine the anomaly diagnosis strategy corresponding to the monitored communication tower based on the warning level, and execute the anomaly diagnosis strategy to obtain anomaly diagnosis results, wherein the anomaly diagnosis results include at least anomaly node identifier, anomaly type, or anomaly severity; and a third determining subunit, used to generate warning information based on the warning level and the anomaly diagnosis results.

[0018] Furthermore, the device also includes: a fifth acquisition unit, used to acquire an initial multimodal dataset of the monitored communication tower in the current monitoring period before acquiring the multimodal dataset of the monitored communication tower in the current monitoring period, wherein the initial multimodal dataset includes at least wind speed, wind direction, temperature and structural response data; and a third processing unit, used to use the initial multimodal dataset as the multimodal dataset of the current monitoring period, or to analyze based on wind speed, wind direction, temperature and target finite element model to obtain predicted structural response data, and to calculate based on the structural response data and predicted structural response data to obtain a target response dataset, and to use the target response dataset as the multimodal dataset of the current monitoring period.

[0019] According to another aspect of the present invention, an electronic device is also provided, comprising: a memory storing an executable program; and a processor for running the program, wherein the program executes the communication tower structure status early warning method described above during runtime.

[0020] According to another aspect of the present invention, a computer-readable storage medium is also provided, the storage medium storing a program, wherein, when the program is running, the device where the storage medium is located executes the communication tower structure status early warning method described above.

[0021] In this embodiment, the following steps are employed: First, a multimodal dataset of the monitored communication tower for the current monitoring period is obtained. Second, a generative causal graph of structural dynamics corresponding to the current monitoring period is obtained. This graph includes multiple physical nodes and directed edges connecting these nodes. The directed edges characterize the causal relationship of vibration transmission between nodes, and each directed edge corresponds to a vibration transmission model. These vibration transmission models are obtained by training a generative adversarial network using a sample dataset. Third, inference calculations are performed based on the multimodal dataset and the generative causal graph of structural dynamics corresponding to the current monitoring period to obtain the overall structural health index of the monitored communication tower. Finally, the early warning information corresponding to the monitored communication tower is determined based on the overall structural health index. This solves the technical problem in related technologies where structural status early warning of communication towers based on sensor threshold alarms has low accuracy.

[0022] This solution constructs a generative causal graph of structural dynamics based on physical causal relationships and utilizes generative adversarial networks to learn the vibration transmission patterns under normal operating conditions. This achieves deep fusion of multimodal data and analysis of physical mechanisms, overcoming the problems of poor generalization ability and fragmented multimodal data fusion caused by the reliance on fault samples in traditional methods. It does not require any fault samples of structural damage or collapse, thus solving the problem of supervised learning models failing in small-sample scenarios. Simultaneously, by calculating the overall structural health index based on causal bias inference, it can capture subtle abnormal changes in the structure, thereby achieving accurate and advanced graded early warnings in the early stages of damage. This improves the accuracy, real-time performance, and early warning lead time of communication tower structural health monitoring. Attached Figure Description

[0023] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0024] Figure 1 A hardware block diagram of a computer terminal for implementing a method for early warning of the structural status of communication towers is shown.

[0025] Figure 2 This is a flowchart of a communication tower structural status early warning method provided according to an embodiment of this application;

[0026] Figure 3 This is a schematic diagram of a communication tower structural status early warning device provided according to an embodiment of this application;

[0027] Figure 4 This is a structural block diagram of an electronic device according to an embodiment of this application. Detailed Implementation

[0028] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0029] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0030] It should be noted that the information collected in this application (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for display, data used for analysis, etc.) are information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of this data all comply with relevant laws, regulations, and standards, and necessary measures have been taken to ensure compliance with public order and good morals. Corresponding access points are provided for users to choose whether to authorize or refuse. For example, interfaces are established between this system and relevant users or organizations, providing users with corresponding access points to choose whether to agree to or refuse automated decision-making results; if the user chooses to refuse, the process proceeds to the expert decision-making stage.

[0031] Example 1

[0032] According to an embodiment of this application, a method embodiment for early warning of the structural status of a communication tower is also provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0033] The method embodiment provided in Embodiment 1 of this application can be executed on a mobile terminal, computer terminal, or similar computing device. Figure 1 A hardware block diagram of a computer terminal (or mobile device) for implementing a structural status early warning method for communication towers is shown. Figure 1As shown, the computer terminal 10 (or mobile device) may include one or more processors 102 (shown as 102a, 102b, ..., 102n in the figure) 102 (processor 102 may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.), a memory 104 for storing data, and a transmission device 106 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of a BUS bus), a network interface, a power supply, and / or a camera. Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the aforementioned electronic device. For example, computer terminal 10 may also include... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.

[0034] It should be noted that the aforementioned one or more processors 102 and / or other data processing circuits are generally referred to herein as "data processing circuits". These data processing circuits may be embodied, in whole or in part, in software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuits may be a single, independent processing module, or may be integrated, in whole or in part, into any other element within the computer terminal 10 (or mobile device). As involved in the embodiments of this application, the data processing circuits serve as a processor control mechanism (e.g., selection of a variable resistor termination path connected to an interface).

[0035] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the communication tower structural status early warning method in this embodiment of the application. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, thereby realizing the aforementioned communication tower structural status early warning method. The memory 104 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to the computer terminal 10 via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0036] The transmission device 106 is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of the computer terminal 10. In one example, the transmission device 106 includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 106 may be a Radio Frequency (RF) module, used for wireless communication with the Internet.

[0037] The display may be a touchscreen liquid crystal display (LCD) that allows the user to interact with the user interface of the computer terminal 10 (or mobile device).

[0038] Under the aforementioned operating environment, this application provides the following: Figure 2 The method for early warning of the structural status of the communication tower is shown. Figure 2 This is a flowchart of a structural condition early warning method for a communication tower according to Embodiment 1 of this application. The structural condition early warning method for the communication tower includes:

[0039] Step S201: Obtain the multimodal dataset of the monitored communication tower in the current monitoring period;

[0040] Step S202: Obtain the structural dynamics generative causal graph corresponding to the current monitoring period. The structural dynamics generative causal graph includes multiple physical nodes and directed edges connecting multiple physical nodes. The directed edges are used to characterize the causal relationship of vibration transmission between nodes. Each directed edge corresponds to a vibration transmission model. The vibration transmission model is obtained by training a generative adversarial network using a sample dataset.

[0041] Step S203: Based on the multimodal dataset of the current monitoring period and the structural dynamics generative causal graph corresponding to the current monitoring period, inference calculation is performed to obtain the overall structural health index corresponding to the monitored communication tower.

[0042] Step S204: Determine the early warning information corresponding to the monitored communication tower based on the overall structural health index.

[0043] Optionally, the structural status early warning system of the communication tower is used as the main execution body. The system performs three working phases: initialization phase (executed once when the system runs for the first time), real-time monitoring phase (executed continuously in a preset monitoring cycle), and calibration phase. The three phases run in parallel and do not interfere with each other.

[0044] Optionally, during the real-time monitoring phase, the system acquires a multimodal dataset of the monitored communication tower for the current monitoring period. This dataset includes acceleration, strain, tilt angle, wind speed, temperature, and image data. Then, it acquires the structural dynamics generative causality graph corresponding to the current monitoring period. This causality graph is obtained by calibrating and updating the structural dynamics generative causality graph corresponding to the previous calibration period. Then, based on the multimodal dataset and the corresponding structural dynamics generative causality graph for the current monitoring period, inference calculations are performed to obtain the global structural health index (i.e., the overall structural health index) for the monitored communication tower. For example, the multimodal dataset of the current monitoring period is input into the corresponding structural dynamics generative causality graph, the local causal deviation value of each causal edge is calculated and smoothed, and the global structural health index is calculated based on the smoothed local causal deviation values ​​of all causal edges.

[0045] Optionally, in the structural status early warning method for communication towers provided in this application embodiment, the overall structural health index of the monitored communication tower is obtained by reasoning and calculating based on the multimodal dataset of the current monitoring period and the structural dynamics generative cause-effect graph corresponding to the current monitoring period. This includes: inputting the multimodal dataset of the current monitoring period into the structural dynamics generative cause-effect graph corresponding to the current monitoring period; for each directed edge, predicting the vibration response based on the vibration transmission model corresponding to the directed edge and the multimodal dataset of the current monitoring period to obtain a predicted value; calculating and smoothing the deviation between the predicted value and the actual observed value corresponding to the directed edge to obtain the target deviation value corresponding to the directed edge; calculating a weighted average based on the target deviation values ​​corresponding to all directed edges and the weight corresponding to each directed edge to obtain the overall deviation value corresponding to the monitored communication tower, and determining the overall structural health index corresponding to the monitored communication tower based on the overall deviation value.

[0046] In an optional embodiment, the system inputs the multimodal dataset of the current monitoring period into the structural dynamics generative causal graph corresponding to the current monitoring period. For each directed edge, the vibration response is predicted based on the multimodal dataset of the current monitoring period using the vibration transmission model corresponding to that directed edge, thus obtaining the predicted value. For example, a sliding time window is first constructed, and the real-time data is divided into blocks according to the sliding time window. For example, the time window length is 1000 sampling points (corresponding to 10 seconds), and the sliding step size is 10 sampling points (corresponding to 0.1 seconds). Each time window contains a cause node. Historical data from the 10 seconds prior to the current moment is used to predict the outcome node. The value at the next time step. Then perform causal edge prediction calculation: for each causal edge ( (representing the current set of causal edges), and the cause node. Historical data within the current time window The input is fed into the corresponding transfer function (i.e., the vibration transmission model) to obtain the result node. Predicted value at time t+1 , means as follows:

[0047]

[0048] Where p and q are the denominator and numerator orders of the transfer function, respectively. For example, this application uses a 10th-order modal superposition transfer function, which is discretized... . and These are the optimal parameters for the calibrated transfer function. Indicates in Time, Cause Node The actual observed vibration response values ​​(such as measured acceleration). This is a lag order index.

[0049] Optionally, for strain and tilt data, since their sampling frequency is lower than that of acceleration data, a linear interpolation method is used to interpolate them to 100Hz before prediction calculation.

[0050] In an optional embodiment, bias calculation and smoothing are performed based on the predicted value and the actual observed value corresponding to the directed edge to obtain the smoothed local bias value (i.e., the target bias value) corresponding to the directed edge. For example, the original local bias value of each causal edge is first calculated. , i.e., result node The absolute difference between the actual observed value and the predicted value is expressed as follows:

[0051]

[0052] in, For the result node The actual observed value at time t+1.

[0053] To eliminate the influence of random noise on the deviation value, the exponential moving average method is used to smooth the original local deviation value, as shown below:

[0054]

[0055] in, Let be the smoothed local deviation value at time t. This is the smoothing coefficient. The value is set to 0.1 because it effectively smooths out random fluctuations and quickly responds to changes in deviation caused by structural damage. During initial system startup... .

[0056] In an optional embodiment, the above calculation process can yield a set of smoothed local deviation values ​​for all causal edges. Then, based on the target deviation values ​​corresponding to all directed edges and the weights corresponding to each directed edge, a weighted average is calculated to obtain the overall deviation value for the monitored communication tower. Based on this overall deviation value, the overall structural health index for the monitored communication tower is determined. For example, first, the causal edge weights are calculated, and the weight of each causal edge... It consists of two parts: modal contribution and node importance, as shown below:

[0057]

[0058] in, The modal contribution weighting coefficient. Causal edge The average modal contribution is the arithmetic mean of the contributions of each modality. The importance weights of nodes are determined by their relative importance. For example, the weight of nodes at the base and top of the tower is 1.5, and the weight of nodes in the middle of the tower body is 1.0.

[0059] Furthermore, the weights of all causal edges are normalized so that the sum of the weights is 1. Then, the weighted global deviation value is calculated. (i.e., the overall deviation value), which is the weighted average of the smoothed local deviation values ​​of all causal edges. This weighted global deviation value characterizes the overall degree of deviation in the physical causal relationship of the entire monitored communication tower structure. The larger the value, the worse the structural health. The formula is as follows:

[0060]

[0061] Furthermore, the weighted global bias value is mapped to the Global Structure Health Index (GSHI) in the range of 0 to 100, as follows:

[0062]

[0063] Among them, the proportionality coefficient The weighted global deviation is determined by statistical analysis of global deviation values ​​under normal operating conditions. For example, during the first 7 days of initial system operation, the average weighted global deviation value under normal operating conditions is calculated. ,make This results in an average GSHI of approximately 95 under normal operating conditions. The GSHI value ranges from 0 to 100, with higher values ​​indicating better structural health. A GSHI of 100 indicates a completely healthy structure, while a GSHI of 0 indicates a completely failed structure.

[0064] The real-time global structural health index is obtained through the above calculation process. and weighted global deviation time series .

[0065] Optionally, the system continues to determine the corresponding early warning information for the monitored communication tower based on the overall structural health index. For example, the system calculates the preset time-based moving average and rate of decline of the global structural health index, and generates early warning signals of corresponding levels according to preset standards. When the early warning level reaches the attention level or above, the system sorts the smoothed local causal deviation values ​​of all causal edges, extracts the preset number of causal edges with the largest deviation values, and determines the physical nodes connected by these causal edges as suspected damage nodes. The system judges the damage type based on the distribution characteristics of the local causal deviation values. For example, if the deviations of multiple causal edges connecting the same node increase simultaneously, the corresponding node connection becomes loose; if the deviation of a single causal edge significantly increases, the corresponding member becomes damaged; if the deviation of the causal edge related to the tower base increases, the corresponding foundation settlement becomes damaged. When the early warning level reaches the alarm level or above, the system calls the calibrated finite element model of the communication tower, sets different levels of damage parameters at the suspected damage location for simulation calculation, and matches the simulated local deviation value distribution with the actual observed distribution to obtain an accurate assessment result of the damage severity.

[0066] Optionally, in the structural status early warning method for communication towers provided in this application embodiment, determining the early warning information corresponding to the monitored communication tower based on the overall structural health index includes: determining the early warning level corresponding to the monitored communication tower based on the overall structural health index; determining the anomaly diagnosis strategy corresponding to the monitored communication tower based on the early warning level, and executing the anomaly diagnosis strategy to obtain anomaly diagnosis results, wherein the anomaly diagnosis results include at least anomaly node identifier, anomaly type, or anomaly severity; and generating early warning information based on the early warning level and the anomaly diagnosis results.

[0067] In an optional embodiment, the system determines the warning level corresponding to the monitored communication tower based on the overall structural health index. For example, the system first calculates the 7-day moving average of GSHI. , means as follows:

[0068]

[0069] Where N=604800 is the number of sampling points within 7 days (sampling frequency 1Hz). The second is the sampling interval.

[0070] Then calculate the daily rate of descent of GSHI. , means as follows:

[0071]

[0072] Where 86400 represents the number of seconds in a day. A negative daily rate of decline indicates a decrease in GSHI, and the larger the absolute value, the faster the decline.

[0073] Optionally, the four-level warning level can be determined according to the following criteria, and the corresponding level will be triggered if any one of the conditions is met:

[0074] Normal state: ;

[0075] Attention status: ;

[0076] Alarm status: ;

[0077] Dangerous situation: .

[0078] Optionally, the warning level determination is performed once per hour to avoid misjudgments caused by instantaneous fluctuations.

[0079] Furthermore, the system determines the corresponding anomaly diagnosis strategy for the monitored communication tower based on the warning level, executes the anomaly diagnosis strategy, and obtains the anomaly diagnosis result. This result includes at least the anomaly node identifier, anomaly type, or anomaly severity. Then, based on the warning level and the anomaly diagnosis result, a warning message is generated. For example, when the warning level remains at "Attention" or above for 24 consecutive hours, a warning notification is automatically triggered. The warning notification is simultaneously sent to relevant management personnel via SMS, email, and platform push notifications. The warning notification content includes: warning level, trigger time, current GSHI value, 7-day trend chart, and recommended measures. Different warning levels are sent to different levels of management personnel; for example, "Attention" level notifications are sent to maintenance personnel, "Alarm" level notifications are sent to department heads, and "Danger" level notifications are sent to company leaders.

[0080] In an optional embodiment, the system performs damage localization and type determination based on local causal bias. First, it extracts abnormal causal edges: when the warning level reaches "attention" or higher, the smoothed local bias values ​​of all causal edges are sorted from largest to smallest. The top 5 causal edges with the largest bias values ​​are extracted and denoted as the set of abnormal causal edges. ; Calculate the average deviation of abnormal causal edges and the average deviation value under normal operating conditions Comparison, requirements Otherwise, the increased deviation is considered to be caused by random fluctuations. Then, suspected damaged nodes are located: Each physical node connected to each edge in the abnormal causal edge set is counted, and the number of times each physical node is connected by the abnormal causal edge is calculated. For example, the physical node with the most connections is designated as a first-level suspected damaged node, and the physical node with the second most connections is designated as a second-level suspected damaged node. For example, if the abnormal causal edge set is... Node 1 is connected 3 times, and nodes 2 and 3 are each connected 2 times. Therefore, node 1 is a first-level suspected damaged node, and nodes 2 and 3 are second-level suspected damaged nodes. Then, the damage type is determined based on the distribution characteristics of the abnormal causal edges, as shown in the following example:

[0081] Node connection loosening damage: If the deviation of multiple (≥3) causal edges connecting the same node increases simultaneously, and all of these causal edges have the node as the result node, then the node is judged to have connection loosening damage, such as loose bolts or cracked welds.

[0082] Member damage: If the deviation of a single causal edge increases significantly (the deviation value is more than twice the average deviation value of other abnormal causal edges), and the deviations of other causal edges connecting the two nodes of this edge do not change significantly, then the member corresponding to this causal edge is judged to be damaged, such as bending, corrosion, or cracks.

[0083] Foundation settlement damage: If the deviation of all causal edges related to the tower base (all causal edges pointing to the tower base nodes) generally increases (the deviation value is greater than twice the normal average value), and the deviation value gradually increases from the top of the tower to the base of the tower, then it is determined that there is foundation settlement damage.

[0084] In an optional embodiment, the system performs false alarm suppression verification: calculating the duration and spatial spread of the local deviation value of the abnormal causal edge over the past 24 hours. If the local deviation value suddenly increases (more than 10 times in 1 hour) and recovers spontaneously within 24 hours, while the deviation values ​​of all causal edges exhibit globally synchronized fluctuation characteristics, it is determined to be a false alarm caused by environmental interference, the warning is automatically lifted, and the interference event is recorded. If the local deviation value shows a continuous increasing trend (more than 5% increase per day for 3 consecutive days), and the deviation value is concentrated only in the local structural regions corresponding to a few adjacent causal edges, it is determined to be real structural damage, and the warning level is maintained and upgraded.

[0085] Through the above process, a list of suspected damaged node locations can be obtained. Damage type inference results Damage confidence assessment value (0-100%).

[0086] In an optional embodiment, the system performs a quantitative assessment of damage severity based on a finite element model. For example, it first performs parametric damage modeling: according to the damage type, corresponding damage parameters are set at the suspected damage locations in the finite element model, as shown in the following example:

[0087] Node connection loosening damage: The stiffness of all bolted spring units corresponding to the node is reduced proportionally, ranging from 10% to 90%, with increments of 10%. For example, a 30% reduction in stiffness indicates a 30% loss of bolt preload.

[0088] Member damage: The cross-sectional area of ​​the member is reduced proportionally, ranging from 10% to 90%, with increments of 10%. The reduction ratio corresponds to the degree of corrosion or damage to the member.

[0089] Foundation settlement damage: The vertical displacement of the four corner nodes of the tower base is set with different values, ranging from 1mm to 10mm, with a step size of 1mm.

[0090] Then, finite element simulation calculations are performed: for each damage parameter setting, finite element modal analysis is executed to obtain the simulated modal parameter set. Based on the simulated modal parameter set, the transfer function parameters of the structural dynamics generative causality graph are recalculated. Using the same input data as in the real-time monitoring phase, the simulated local causal deviation value distribution is calculated. , where k is the index of the damage parameter.

[0091] Then, damage severity matching is performed: the Pearson correlation coefficient between each simulated deviation value distribution and the actual observed deviation value distribution is calculated, as shown in the following formula:

[0092]

[0093] in, The average value of the distribution of the k-th simulated deviation value. This is the average value of the distribution of actual observed deviation values. Correlation coefficient. The closer the value is to 1, the better the simulation results match the actual observation results. Optionally, the damage parameter with the highest correlation coefficient can be selected as the assessment result of the actual damage severity, and the output can be a quantitative assessment result of the damage severity (e.g., bolt stiffness reduction of 40%, bar section damage of 20%, foundation settlement of 3mm), the prediction result of the remaining service life, and a maintenance recommendation report.

[0094] Through the above process, the quantified structural health index is transformed into actionable early warning information and maintenance guidance. Based on the changing trend of the global structural health index, hierarchical early warning is achieved. By analyzing the distribution characteristics of local causal deviation values, accurate damage location and type judgment are achieved. Finally, the calibrated finite element model is used to quantitatively assess the severity of damage, thereby improving the accuracy of structural condition early warning.

[0095] In summary, by constructing a generative causal graph of structural dynamics based on physical causal relationships and utilizing generative adversarial networks to learn the vibration transmission laws under normal operating conditions, deep fusion and physical mechanism analysis of multimodal data are achieved. This overcomes the problems of poor generalization ability and fragmented multimodal data fusion caused by the reliance on fault samples in traditional methods. It does not require any fault samples of structural damage or collapse, thus solving the problem of supervised learning models failing in small sample scenarios. Furthermore, by calculating the overall structural health index based on causal bias inference, subtle abnormal changes in the structure can be captured, enabling accurate and advanced hierarchical early warnings in the early stages of damage. This improves the accuracy, real-time performance, and early warning lead time of communication tower structural health monitoring.

[0096] Optionally, in the structural state early warning method for communication towers provided in this application embodiment, a structural dynamics generative causal graph is obtained through the following steps: obtaining a target finite element model of the monitored communication tower, wherein the target finite element model is obtained by calibrating the initial finite element model of the monitored communication tower using the historical multimodal dataset of the monitored communication tower; determining multiple physical nodes of the monitored communication tower based on the target finite element model, and constructing an initial causal graph topology based on the multiple physical nodes; and determining the structural dynamics generative causal graph based on the historical multimodal dataset and the initial causal graph topology.

[0097] Optionally, in the structural state early warning method for communication towers provided in this application embodiment, determining the structural dynamics generative causal graph based on the historical multimodal dataset and the initial causal graph topology includes: configuring an initial vibration transmission model for each directed edge in the initial causal graph topology to obtain the initial structural dynamics generative causal graph; constructing a generative adversarial network based on the initial structural dynamics generative causal graph and a multilayer fully connected neural network; extracting sample datasets from the historical multimodal dataset and training the generative adversarial network based on the sample datasets until preset conditions are met to obtain the structural dynamics generative causal graph.

[0098] In an optional embodiment, in obtaining the generative causal graph of structural dynamics, a target finite element model of the monitored communication tower is first acquired. This target finite element model is obtained by calibrating the initial finite element model of the monitored communication tower using a historical multimodal dataset. For example, a historical multimodal dataset is acquired, a digital twin initial finite element model of the monitored communication tower is constructed, and the initial finite element model is calibrated using the historical multimodal dataset to obtain a calibrated finite element model (i.e., the target finite element model). The modal parameter set of the calibrated finite element model is then extracted. For example, the modal parameter set of the calibrated finite element model includes the first 10 natural frequencies, mode shape vectors, and damping ratios. Then, multiple physical nodes of the monitored communication tower are determined based on the target finite element model, and an initial causal graph topology is constructed based on these multiple physical nodes. Based on the historical multimodal dataset and the initial causal graph topology, the generative causal graph of structural dynamics is determined. For example, the calibrated finite element model is discretized into a set of key physical nodes containing all sensor installation locations and structural connection nodes. This set of key physical nodes serves as the node set for the causal graph. For any two physical nodes, if the vibration amplitude of the node at the higher position is greater than that of the node at the lower position in any first-order vibration mode, a directed causal edge is generated from the higher node to the lower node. All indirect causal edges with a transmission path length exceeding 3 are removed, resulting in the initial causal graph topology. A parameterized transfer function based on the modal superposition principle is assigned to each initial directed causal edge. The parameterized transfer function is discretized into a difference equation form, forming a parameterized initial structural dynamics generative causal graph. Using a historical multimodal dataset, a generative adversarial learning approach is employed to calibrate the parameterized initial structural dynamics generative causal graph. The generator is the parameterized initial structural dynamics generative causal graph, with historical data of the cause nodes as input and future prediction data of the result nodes as output.

[0099] Optionally, before the initialization phase begins, various sensors are deployed at key nodes of the communication tower. For example, at least one triaxial accelerometer is deployed at the tower base, one-third of the tower height, two-thirds of the tower height, and the top of the tower; at least one fiber optic strain gauge is deployed at each connection node of each section of the tower; at least one dual-axis tilt sensor is deployed at one-third of the tower height, two-thirds of the tower height, and the top of the tower; at least one weather station is deployed at the top of the tower to measure wind speed, wind direction, and temperature; and at least one high-definition pan-tilt camera is deployed at the base of the tower to capture a panoramic view of the tower. All sensors are uniformly connected to the edge computing gateway, and the Global Positioning System (GPS) timing module adds an absolute timestamp to each data frame.

[0100] Optionally, data is acquired according to a preset sampling frequency. For example, the triaxial accelerometer samples at 100 Hz, the fiber optic strain gauge at 10 Hz, the biaxial tilt sensor and weather station at 1 Hz, and the high-definition pan-tilt camera captures a panoramic view of the tower once per hour. Then, the data from different sampling frequencies are time-axis aligned. Using the accelerometer's sampling time as a reference, the data from other sensors are aligned to the same time axis using linear interpolation, generating a synchronized data frame sequence. The synchronized data frame sequence is then preprocessed, as shown in the example below.

[0101] Detrend term removal: The low-frequency linear trend term in the acceleration and strain signals is fitted using the least squares method, and then the trend term is subtracted from the original signal to eliminate the effects of temperature drift and sensor zero-point drift.

[0102] Wavelet denoising: The acceleration signal is decomposed into 5 levels using the db4 wavelet. The coefficients with absolute values ​​less than the threshold in the high-frequency coefficients obtained from the decomposition are set to zero. Then, the signal is reconstructed to remove high-frequency noise components. The strain and tilt angle signals are denoised using a soft threshold method, with the threshold selected as 3 times the standard deviation of the signal.

[0103] Outlier removal: The 3σ criterion is used to identify isolated outliers in the data. A data point is considered an outlier if its difference from the mean of the data series is greater than three times the standard deviation. Outliers are replaced by linear interpolation of the two preceding and following normal data points.

[0104] Standardization: Perform Z-score standardization on all numerical data, that is, subtract the mean of the data series from each data point and then divide by the standard deviation of the data series, so that the mean of the processed data is 0 and the standard deviation is 1.

[0105] Through the above process, a preprocessed clean multimodal dataset is obtained, which serves as a historical multimodal dataset. This dataset contains time-synchronized acceleration, strain, tilt angle, wind speed, temperature, and image data.

[0106] Optionally, during the initialization phase, an initial finite element model of the communication tower is constructed. First, a geometric model is established: the communication tower design drawings and the mechanical parameters of the steel (elastic modulus) are obtained. (Pa, Poisson's ratio μ=0.3, density ρ=7850kg / m³). A 1:1 scale 3D geometric model of the communication tower was established in a multiphysics simulation software platform to accurately reproduce the cross-sectional dimensions, length, and spatial position of all members. For angle steel members, the actual cross-sectional shape was used for modeling; for connecting plates and reinforcing plates, the actual thickness and dimensions were used for modeling. A geometric model of the reinforced concrete foundation of the tower was established, with the foundation dimensions precisely set according to the foundation construction drawings. The angle steel members of the tower body were simulated using BEAM188 elements, which are 3D linear finite strain beam elements with 6 degrees of freedom, capable of accurately simulating the axial deformation, bending deformation, and torsional deformation of the beam. Bolted connections were simulated using COMBIN14 spring elements, with each bolt connection corresponding to 3 COMBIN14 elements, simulating axial stiffness, shear stiffness, and rotational stiffness respectively. The tower foundation was simulated using SOLID185 solid elements, which are 3D 8-node structural solid elements, capable of accurately simulating the stress characteristics of concrete. The geometric model was then meshed, with the following requirements: tower body member element length no greater than 0.5m; connecting plate and reinforcing plate element length no greater than 0.2m; and tower foundation element length no greater than 1m. Material properties were then assigned to the meshed model: all angle steel members and connecting plates of the tower body were assigned the material properties of Q235 steel. The tower foundation was assigned the material properties of C30 concrete: elastic modulus. Pa, Poisson's ratio μ=0.2, density ρ=2400kg / m³. Assign initial stiffness values ​​to the COMBIN14 spring element; the initial axial stiffness is taken as... N / m, initial shear stiffness is taken as N / m, initial rotational stiffness taken N m / rad. Then, boundary conditions and loads are set: the bottom surface of the tower foundation is set as a fixed constraint, restricting displacement and rotation in all six degrees of freedom. The connection between the tower and the foundation is simulated using anchor bolts, which are modeled using BEAM188 elements, with the lower end coupled to the top node of the foundation and the upper end coupled to the bottom node of the tower. The structural self-weight load is applied by setting gravitational acceleration. Equipment loads are applied, based on the weight and location of the antennas, feeders, and other equipment actually installed on the communication tower, applying the loads to the corresponding nodes, ultimately obtaining the initial finite element model of the communication tower.

[0107] Optionally, measured modal parameters are extracted, i.e., the current vibration characteristics (natural frequency, damping, and mode shape) of the communication tower are extracted from sensor data. First, data preprocessing is performed: Normal operating condition data from the previous three days are obtained from a clean multimodal dataset. Bandpass filtering is applied to the acceleration data, with a filtering frequency range of 0.1Hz to 20Hz, to remove DC components and high-frequency noise. The data is divided into multiple time windows with a length of 2048 sampling points, with adjacent time windows overlapping by 50%. Then, modal identification is performed: a covariance-driven random subspace method is used for modal parameter identification. A Hankel matrix is ​​constructed, with twice the number of rows and half the length of the time window. Singular value decomposition is performed on the Hankel matrix to obtain the system's state matrix and output matrix. The natural frequency and damping ratio are calculated from the eigenvalues ​​of the state matrix, and the mode shape vector is calculated from the output matrix. The modal order is determined: a stability diagram is plotted, with the x-axis representing frequency and the y-axis representing the model order. As the model order increases, the points where the frequency, damping ratio, and mode shape vector remain stable are the true modes. Extract the parameters of the first 10 stable modes, including their natural frequencies. Damping ratio And mode vector , Modal verification: Calculate the Modal Confidence Criterion (MAC) values ​​for each mode to verify the orthogonality of the mode shape vectors. The MAC value between any two modes must be less than 0.1; otherwise, the modal identification result is considered unreliable and requires re-identification. Finally, the measured modal parameter set is obtained. .

[0108] Optionally, finite element model calibration is performed: modal analysis is performed on the initial finite element model, and the first 10 modal parameters are extracted using the block Lanzos method to obtain the initial computational modal parameter set. Select calibration parameters in the following priority order:

[0109] Boundary condition parameters: equivalent stiffness of the tower foundation and stiffness of the anchor bolts;

[0110] Material parameters: elastic modulus and density of steel;

[0111] Connection parameters: axial stiffness, shear stiffness, and rotational stiffness of bolted connections.

[0112] Optionally, the objective function can be defined as calculating the weighted sum of squared errors between the modal parameters and the measured modal parameters, as shown in the following formula:

[0113]

[0114] in, =0.6 is the frequency weight. =0.1 is the damping ratio weight. =0.3 is the mode shape weight; To calculate the modal confidence criterion value between the mode shape and the measured mode shape.

[0115] Optionally, optimization calibration is performed: a genetic algorithm is used for parameter optimization, with a population size of 50, a crossover probability of 0.8, and a mutation probability of 0.1. The objective function value is calculated in each iteration, and optimization stops when the objective function value is less than 0.05 or the number of iterations reaches 100. During optimization, it is ensured that all calibration parameters are within a reasonable physical range to obtain the optimized calibration parameters.

[0116] Optionally, the calibration results can be verified: the finite element model is updated with the optimized parameters, and modal analysis is performed again to obtain the calibrated set of calculated modal parameters. The error between the calculated and measured modal parameters after calibration is verified, requiring the relative error of each natural frequency to be less than 5% and the MAC value of the mode shape to be greater than 0.9. If these requirements are met, the calibrated finite element model of the communication tower is obtained. calibrated set of computational modal parameters .

[0117] Optionally, node mapping is performed: A set of key physical nodes is determined. For example, the set V of key physical nodes consists of four types of nodes: tower base nodes, tower body connection nodes, sensor installation nodes, and tower top nodes. The selection criteria for each type of node are as follows: Tower base nodes: 4 nodes in total, located at the center of the top surface of the four corners of the tower base. Alternatively, the center point of the tower base can be used. Tower body connection nodes: One node is set at each end of each segment of the tower body. For a 10-segment, 50-meter-high communication tower, a total of 20 tower body connection nodes are set. Sensor installation nodes: One node corresponds to the installation position of each accelerometer, strain gauge, and tilt sensor. In this embodiment, 4 accelerometer nodes, 9 strain gauge nodes, and 3 tilt sensor nodes are included, for a total of 16 nodes. Tower top nodes: 2 nodes in total, located at the center point of the tower top platform and the highest point of the antenna installation position, respectively. Node mapping and numbering: In the calibrated finite element model of the communication tower, the finite element node closest to the location of the key physical node is found, and a one-to-one mapping relationship is established to obtain a mapping table between key physical nodes and finite element nodes. Each critical physical node is assigned a unique integer identifier, for example, the numbering rule is: tower base nodes: 1-4; tower body connection nodes: 5-24; sensor installation nodes: 25-40; tower top nodes: 41-42, resulting in the set of critical physical nodes. Node attribute definition: Define attributes for each critical physical node, including node number, spatial coordinates, corresponding finite element node number, node type, and installed sensor type (if any). Create a node attribute table to record detailed information for all critical physical nodes, resulting in the critical physical node attribute table.

[0118] Optionally, a structural dynamics generative causal graph is constructed: based on the fundamental principles of structural dynamics, the topology and transfer function of the causal graph are built. Parameters are calibrated through generative adversarial learning to obtain a mathematical model that accurately describes the physical causal relationships under normal operating conditions of the communication tower, i.e., the calibrated structural dynamics generative causal graph. First, an initial causal graph topology is generated, as shown in the example below:

[0119] Node spatial location sorting: based on the spatial coordinates of each key physical node According to height coordinates Sort all nodes from largest to smallest to obtain the node height sequence. For nodes of the same height, sort them in ascending order of their horizontal distance from the center of the tower.

[0120] Generate initial causal edges: for any two distinct physical nodes Perform the following judgment: if the node The height is greater than the node height Then calculate the ratio of the vibration amplitudes of the two nodes in all 10 modes. If there exists at least one mode n such that... Then a path is generated from the node. Pointing to node Directed causal edge The physical meaning of this generation rule is as follows: the vibration energy of a tall steel structure is mainly transmitted from the top to the bottom, and the vibration amplitude usually decreases as the height decreases. If, in a certain vibration mode, the vibration amplitude of the upper node is greater than that of the lower node, it indicates that the vibration of the upper node will causally cause the vibration of the lower node.

[0121] Causal edge pruning: A breadth-first search algorithm is used to calculate the shortest path length between any two nodes. Path length is defined as the number of edges traversed from the cause node to the result node. All indirect causal edges with a shortest path length greater than 3 are removed, retaining only direct causal relationships. All self-loop edges (i.e., edges pointing from a node to itself) are removed. The pruned result is the initial set of causal edges. Based on the set of key physical nodes V and the pruned set of causal edges The initial causal graph topology is obtained. .

[0122] Furthermore, a parameterized initial structural dynamics generative causal graph is generated: First, the modal contribution of each causal edge in the pruned causal edge set is calculated to obtain the modal contribution set. For example, for each retained causal edge Calculate its modal contribution in each vibration mode. :

[0123]

[0124] in, Cause node The vibration amplitude in the nth mode shape, For the result node The vibration amplitude in the nth mode shape. The modal contribution reflects the degree of influence of the nth mode on the causal side transmission characteristics.

[0125] Then, the causal edge transfer function is parameterized: based on the modal superposition principle of linear systems, the vibration transfer relationship between any two nodes can be expressed as a weighted sum of the transfer functions of each order of modes. For a single-degree-of-freedom system, its displacement transfer function (output displacement / input force) is:

[0126]

[0127] in, For the nth modal mass, The damping ratio of the nth mode is... = Let be the nth natural angular frequency. s is the complex frequency variable in the Laplace transform.

[0128] For the acceleration transfer function (output acceleration / input acceleration), taking the second derivative with respect to the displacement transfer function yields:

[0129]

[0130] Considering the influence of modal contribution, nodes in a multi-degree-of-freedom system To the node The acceleration transfer function is:

[0131]

[0132] Since this application uses an acceleration signal as the main input, the above-described acceleration transfer function form is adopted.

[0133] Transfer function correction and normalization: To make the gain of the transfer function zero at zero frequency (which conforms to the low-frequency characteristics of the actual structure), the above transfer function is corrected to obtain the final parameterized transfer function:

[0134]

[0135] This transfer function accurately describes the cause node. How are acceleration signals transmitted through the structure to the result node? .

[0136] Transfer function discretization: The Laplace domain transfer function is discretized using the bilinear transform method. Discretized into Z-domain transfer function The sampling frequency is consistent with the sampling frequency of the accelerometer. The bilinear transformation formula is:

[0137]

[0138] Substituting the bilinear transformation formula into the Laplace domain transfer function, we obtain the general form of the Z-domain transfer function:

[0139]

[0140] Where p and q are the denominator and numerator orders of the transfer function, respectively. For the 10th-order modal superposition transfer function used in this application, after discretization... .

[0141] Converting to difference equation form: The Z-domain transfer function is converted into difference equation form for easier online real-time calculation, as shown below:

[0142]

[0143] in, Cause node The input acceleration signal at time t, For the result node The predicted output acceleration signal at time t, and These are the parameters to be calibrated for the transfer function.

[0144] Finally, a parameterized initial structural dynamics generative causal graph is obtained. ,in, This represents the set of causal edges that have been assigned parameters to the corresponding transfer functions, and each causal edge... Corresponding to a set of initial transfer function parameters .

[0145] Optionally, the parameterized initial structural dynamics generative causal graph is calibrated based on generative adversarial learning, as follows:

[0146] Constructing the training dataset: Divide the normal operating condition acceleration data from the first 7 days into time windows of length L=1000 sampling points, corresponding to a duration of 10 seconds. Adjacent time windows overlap by 50% to increase the number of training samples. For each time window, use the first 900 sampling points (first 9 seconds) as the historical input data for the cause node, and the last 100 sampling points (last 1 second) as the future output data for the result node, thus obtaining the training dataset. ,in, For the cause node in the k-th window Historical input data, For the result node in the k-th window The future output data, where K is the total number of training samples.

[0147] Constructing Generative Adversarial Networks: Generating Causal Graphs from Parameterized Initial Structural Dynamics As a generator, the generator's input is the historical data of the cause node. The output is the future prediction data for the result node. The trainable parameters of the generator are the transfer function parameters of all causal edges. The discriminator employs a 3-layer fully connected neural network, with the following structure: Input layer: 100 neurons, corresponding to the future output data of the result node 1 second in advance; Hidden layer 1: 64 neurons, activation function is ReLU; Hidden layer 2: 32 neurons, activation function is ReLU; Output layer: 1 neuron, activation function is Sigmoid, outputting the probability that the data comes from the true distribution; The trainable parameters of the discriminator are the weights and biases of the neural network.

[0148] Loss function definition: A loss function with gradient penalty is used to avoid the mode collapse and gradient vanishing problems in traditional generative adversarial networks (GANs). The generator's loss function is:

[0149]

[0150] The loss function of the discriminator is:

[0151]

[0152]

[0153] in, For the true data distribution, To generate a data distribution, The interpolation distribution between real data and generated data. =10 is the gradient penalty coefficient.

[0154] Alternating training: An adaptive moment estimation optimizer is used, with a generator learning rate of [missing information]. The discriminator learning rate is For example, train the discriminator 5 times, then train the generator once. Set the batch size to 32. Monitor the loss function in real time during training. When the generator loss function decreases by less than [a certain value] for 10 consecutive rounds... Training should be stopped when the total number of training rounds reaches 100.

[0155] Calibration result validation: The calibrated causal graph is validated using a validation set (10% of the data randomly selected from the training dataset). The mean squared error (MSE) between the predicted and actual data is calculated, and the average MSE is required to be less than [value missing]. If the validation fails, adjust the training parameters and retrain.

[0156] Finally, the calibrated structural dynamics generative causal graph is obtained. , where each causal edge Corresponding to a set of optimal transfer function parameters .

[0157] Optionally, in the structural status early warning method for communication towers provided in this application embodiment, the structural dynamics generative causal graph corresponding to the current monitoring period is obtained through the following steps: obtaining a target multimodal dataset, wherein the target multimodal dataset is a dataset extracted from the multimodal dataset of the previous calibration period of the current monitoring period, excluding data within the time period during which an early warning was triggered, and data where the wind speed is greater than a preset threshold or the temperature exceeds a preset range; calibrating and updating the structural dynamics generative causal graph corresponding to the previous calibration period of the current monitoring period based on the target multimodal dataset, thereby obtaining the structural dynamics generative causal graph corresponding to the current monitoring period.

[0158] In an optional embodiment, in the process of obtaining the structural dynamics generative causal graph corresponding to the current monitoring period, a target multimodal dataset is first obtained. This target multimodal dataset is a dataset extracted from the multimodal dataset of the previous calibration period of the current monitoring period, excluding data within the time period during which an early warning was triggered, and data where the wind speed is greater than a preset threshold or the temperature exceeds a preset range. Then, the structural dynamics generative causal graph corresponding to the previous calibration period of the current monitoring period is calibrated and updated based on the target multimodal dataset to obtain the structural dynamics generative causal graph corresponding to the current monitoring period. For example, the multimodal dataset of the previous calibration period of the current monitoring period is obtained, and the normal operating condition multimodal dataset that did not trigger any warnings in the previous calibration period of the current monitoring period is extracted. The transfer function parameters of the structural dynamics generative causal graph corresponding to the previous calibration period of the current monitoring period are used as initial values, and incremental generative adversarial learning is used to fine-tune the parameters. In this process, the generator keeps the topology of the causal graph unchanged and only updates the transfer function parameters of each causal edge. The discriminator distinguishes between the generated result node data and the actual collected normal operating condition data. The update stops when the loss function of parameter fine-tuning converges to a preset threshold or reaches the maximum number of iterations, thus obtaining the structural dynamics generative causal graph corresponding to the current monitoring period.

[0159] Optionally, during the calibration phase, after obtaining the multimodal dataset from the previous calibration period for the current monitoring period, data from warning periods are removed: All time periods during which warnings were triggered within the previous calibration period are extracted, including the warning start and end times. All data from these warning periods are removed from the clean multimodal dataset, retaining only the normal operating condition data after the warning is lifted. If the warning duration in the previous calibration period exceeds 12 hours, the data acquisition time is extended until at least 12 hours of normal operating condition data are accumulated before calibration. Extreme environment data is removed: Wind speed and temperature data are extracted from the clean multimodal dataset, and data from extreme environmental periods with wind speeds greater than 15 m / s or temperatures below -20℃ or above 40℃ are removed. This avoids abnormal structural responses under extreme environments affecting the calibration results. Data consistency is verified: Statistical characteristics of the remaining normal operating condition data are calculated, including the mean, standard deviation, and maximum value. These are compared with the statistical characteristics of historical normal operating condition data to ensure consistent data distribution. If the standard deviation of the data exceeds twice the historical average, the data is considered abnormal, and data filtering is performed again. Extracting Pure Structural Damage Response: Using the same external load decoupling method as in the real-time monitoring phase, pure structural damage response data is extracted based on the filtered normal operating condition data and the calibrated finite element model. Only the pure structural damage response data is used for subsequent causal graph calibration to eliminate the influence of environmental loads. The final result is the pure structural damage response dataset under normal operating conditions from the previous calibration cycle. The data length should be no less than 12 hours.

[0160] Optionally, an incremental training dataset is constructed: the pure structural damage response dataset under normal operating conditions is divided into time windows of length L=1000 sampling points, corresponding to a duration of 10 seconds. Adjacent time windows overlap by 50% to increase the number of training samples. For each time window, the first 900 sampling points (the first 9 seconds) are used as the historical input data for the cause node, and the last 100 sampling points (the last 1 second) are used as the future output data for the result node. Sample quality is evaluated: for each time window, the effective causal graph from the previous period is used. Make predictions and calculate the mean squared error (MSE) between the predicted and actual data. Remove data with an MSE greater than [value missing]. The low-quality samples typically contain undetected anomalies or noise. This results in the final incremental training dataset. Where K is the total number of valid training samples. Dataset partitioning: The incremental training dataset is randomly divided into a training set and a validation set, with the training set accounting for 90% and the validation set accounting for 10%. The training set is used for updating model parameters, and the validation set is used to monitor the training process and prevent overfitting.

[0161] Optionally, incremental generative adversarial learning calibration involves first initializing the network structure and parameters. The generator maintains the exact same structure as in the initial calibration phase, meaning the topology of the causal graph remains unchanged; only the transfer function parameters of each causal edge are updated. The generator's initial parameters are set to the causal graph valid in the previous cycle. Optimal parameters The discriminator retains the exact same 3-layer fully connected neural network structure as in the initial calibration phase, with initial parameters set to the discriminator parameters from the previous calibration cycle.

[0162] Define an incremental loss function: To prevent the model from forgetting previously learned causal relationships, a regularization term is added to the traditional loss function with gradient penalty to limit the magnitude of parameter updates. The incremental loss function for the generator is:

[0163]

[0164] in, These are the parameters of the current generator. These are the parameters of the generator from the previous cycle. This is the regularization coefficient, used to control the magnitude of parameter updates. The loss function of the discriminator is the same as that in the initial calibration phase.

[0165] Incremental training process: An adaptive moment estimation optimizer is used, and the generator learning rate is... The discriminator learning rate is Train the discriminator 3 times, then train the generator 1 time (to reduce the number of generator training iterations and prevent overfitting). Set the batch size to 32. Set the maximum number of iterations to 50. During training, calculate the validation error every 5 iterations using the validation set. If the validation error does not decrease for 3 consecutive iterations, stop training early to prevent overfitting. Validate the trained generator using the validation set and calculate the average MSE between the predicted and actual data. The average MSE should be less than [value missing]. If the validation fails, adjust the training parameters and retrain.

[0166] Optionally, in the structural status early warning method for communication towers provided in this application embodiment, before obtaining the multimodal dataset of the monitored communication tower in the current monitoring period, the method further includes: obtaining an initial multimodal dataset of the monitored communication tower in the current monitoring period, wherein the initial multimodal dataset includes at least wind speed, wind direction, temperature, and structural response data; using the initial multimodal dataset as the multimodal dataset for the current monitoring period, or, analyzing based on wind speed, wind direction, temperature, and a target finite element model to obtain predicted structural response data, and calculating based on the structural response data and the predicted structural response data to obtain a target response dataset, and using the target response dataset as the multimodal dataset for the current monitoring period.

[0167] In an optional embodiment, an initial multimodal dataset of the monitored communication tower in the current monitoring period is obtained. This dataset includes at least wind speed, wind direction, temperature, and structural response data. The initial multimodal dataset is then used as the multimodal dataset for the current monitoring period. Alternatively, based on the wind speed, wind direction, temperature, and target finite element model, predicted structural response data is obtained. The target response dataset is then calculated based on the structural response data and the predicted structural response data and used as the multimodal dataset for the current monitoring period.

[0168] Optionally, based on the real-time wind speed and temperature meteorological parameters in the initial multimodal dataset of the current monitoring period, the calibrated finite element model of the communication tower is used to generate predicted structural response values ​​caused only by wind and temperature loads. Then, the actual multimodal data of the current monitoring period is input into the external load response prediction model to obtain predicted structural response values ​​caused only by wind and temperature loads. The corresponding predicted external load response values ​​are then subtracted from the actual observed multimodal data of the current monitoring period to obtain pure structural damage response data that has eliminated the influence of environmental loads. For example, external load decoupling preprocessing is first performed: real-time wind speed v(t) and wind direction are extracted from the clean multimodal dataset of the current monitoring period. Environmental load parameters were obtained by sampling temperature T(t) data at a frequency of 1 Hz. The environmental load data were then interpolated to 100 Hz using a linear interpolation method, consistent with the sampling frequency of acceleration, strain, and tilt data. External load response prediction was then performed: real-time wind speed, wind direction, and temperature data were input into the calibrated finite element model for transient dynamic analysis. For wind loads, the wind load time histories acting on each node of the tower were calculated based on wind speed and direction; for temperature loads, the temperature deformation of the structure was calculated based on temperature data. Predicted structural responses caused solely by wind and temperature loads, including acceleration response predictions, were generated. Strain response prediction value and tilt response prediction value Extracting pure structural damage response: This involves extracting the multimodal data actually observed during the current monitoring period (…). Subtracting the corresponding predicted external load response from the original data yields the pure structural damage response data. ), which is represented as follows:

[0169]

[0170]

[0171]

[0172] Pure structural damage response data only reflects the vibration and deformation caused by changes in the mechanical properties of the structure itself, and does not include the influence of environmental loads, thus obtaining a pure structural damage response dataset. .

[0173] Furthermore, the real-time calculation of local causal bias is performed, which is exactly the same as the steps in the previous embodiment, except that the input real-time preprocessed multimodal clean dataset is replaced with a pure structural damage response dataset. The upcoming pure structural damage response dataset Input the structural dynamics generative causal graph corresponding to the current monitoring period, and output the set of smoothed local deviation values ​​for all causal edges. The calculation of the global structural health index follows the same steps as in the previous embodiments, and will not be repeated here.

[0174] The structural status early warning method for communication towers provided in this application includes the following steps: acquiring a multimodal dataset of the monitored communication tower in the current monitoring period; acquiring a structural dynamics generative causal graph corresponding to the current monitoring period, wherein the structural dynamics generative causal graph includes multiple physical nodes and directed edges connecting the multiple physical nodes, the directed edges are used to characterize the causal relationship of vibration transmission between nodes, each directed edge corresponds to a vibration transmission model, and the vibration transmission model is obtained by training a generative adversarial network using a sample dataset; performing inference calculations based on the multimodal dataset of the current monitoring period and the structural dynamics generative causal graph corresponding to the current monitoring period to obtain the overall structural health index corresponding to the monitored communication tower; and determining the early warning information corresponding to the monitored communication tower based on the overall structural health index. This solves the technical problem of low early warning accuracy in related technologies that rely on sensor threshold alarms to achieve structural status early warning for communication towers.

[0175] This solution constructs a generative causal graph of structural dynamics based on physical causal relationships and utilizes generative adversarial networks to learn the vibration transmission patterns under normal operating conditions. This achieves deep fusion of multimodal data and analysis of physical mechanisms, overcoming the problems of poor generalization ability and fragmented multimodal data fusion caused by the reliance on fault samples in traditional methods. It does not require any fault samples of structural damage or collapse, thus solving the problem of supervised learning models failing in small-sample scenarios. Simultaneously, by calculating the overall structural health index based on causal bias inference, it can capture subtle abnormal changes in the structure, thereby achieving accurate and advanced graded early warnings in the early stages of damage. This improves the accuracy, real-time performance, and early warning lead time of communication tower structural health monitoring.

[0176] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0177] Example 2

[0178] This application also provides a structural status early warning device for communication towers. It should be noted that the structural status early warning device for communication towers in this application can be used to execute the structural status early warning method for communication towers provided in this application. The structural status early warning device for communication towers provided in this application will be described below.

[0179] According to an embodiment of this application, a structural condition early warning device for a communication tower used to implement the above-described structural condition early warning method for communication towers is also provided, such as... Figure 3 As shown, the device includes: a first acquisition unit 301, a second acquisition unit 302, a first processing unit 303, and a first determination unit 304.

[0180] The first acquisition unit 301 is used to acquire the multimodal dataset of the monitored communication tower in the current monitoring period;

[0181] The second acquisition unit 302 is used to acquire the structural dynamics generative causal graph corresponding to the current monitoring period. The structural dynamics generative causal graph includes multiple physical nodes and directed edges connecting multiple physical nodes. The directed edges are used to characterize the causal relationship of vibration transmission between nodes. Each directed edge corresponds to a vibration transmission model. The vibration transmission model is obtained by training a generative adversarial network through a sample dataset.

[0182] The first processing unit 303 is used to perform reasoning calculations based on the multimodal dataset of the current monitoring period and the structural dynamics generative causal graph corresponding to the current monitoring period to obtain the overall structural health index of the monitored communication tower.

[0183] The first determining unit 304 is used to determine the early warning information corresponding to the monitored communication tower based on the overall structural health index.

[0184] Optionally, in the structural status early warning device for communication towers provided in this application embodiment, the first processing unit includes: a first processing subunit, used to input the multimodal dataset of the current monitoring period into the structural dynamics generative cause-effect graph corresponding to the current monitoring period, and for each directed edge, to perform vibration response prediction based on the multimodal dataset of the current monitoring period through the vibration transmission model corresponding to the directed edge to obtain a predicted value; a second processing subunit, used to perform deviation calculation and smoothing processing based on the predicted value and the actual observed value corresponding to the directed edge to obtain the target deviation value corresponding to the directed edge; and a third processing subunit, used to perform weighted average calculation based on the target deviation values ​​corresponding to all directed edges and the weight corresponding to each directed edge to obtain the overall deviation value corresponding to the monitored communication tower, and to determine the overall structural health index corresponding to the monitored communication tower based on the overall deviation value.

[0185] Optionally, in the structural status early warning device for communication towers provided in this application embodiment, the device further includes the following units for obtaining a structural dynamics generative causal graph through the following steps: a third acquisition unit for acquiring a target finite element model of the monitored communication tower, wherein the target finite element model is obtained by calibrating the initial finite element model of the monitored communication tower using the historical multimodal dataset of the monitored communication tower; a second determination unit for determining multiple physical nodes of the monitored communication tower based on the target finite element model, and constructing an initial causal graph topology based on the multiple physical nodes; and a third determination unit for determining the structural dynamics generative causal graph based on the historical multimodal dataset and the initial causal graph topology.

[0186] Optionally, in the structural state early warning device for communication towers provided in this application embodiment, the third determining unit includes: a fourth processing subunit, used to configure an initial vibration transmission model for each directed edge in the initial causal graph topology to obtain an initial structural dynamics generative causal graph; a fifth processing subunit, used to construct a generative adversarial network based on the initial structural dynamics generative causal graph and a multilayer fully connected neural network; and a sixth processing subunit, used to extract sample datasets from historical multimodal datasets and train the generative adversarial network based on the sample datasets until preset conditions are met to obtain a structural dynamics generative causal graph.

[0187] Optionally, in the structural status early warning device for communication towers provided in this application embodiment, the device further includes the following unit, used to obtain the structural dynamics generative causal graph corresponding to the current monitoring period through the following steps: a fourth acquisition unit, used to acquire a target multimodal dataset, wherein the target multimodal dataset is a dataset extracted from the multimodal dataset of the previous calibration period of the current monitoring period, excluding data within the time period during which an early warning was triggered, and data where the wind speed is greater than a preset threshold or the temperature exceeds a preset range; a second processing unit, used to calibrate and update the structural dynamics generative causal graph corresponding to the previous calibration period of the current monitoring period based on the target multimodal dataset, to obtain the structural dynamics generative causal graph corresponding to the current monitoring period.

[0188] Optionally, in the structural status early warning device for communication towers provided in this application embodiment, the first determining unit includes: a first determining subunit, used to determine the early warning level corresponding to the monitored communication tower based on the overall structural health index; a second determining subunit, used to determine the abnormal diagnosis strategy corresponding to the monitored communication tower based on the early warning level, and execute the abnormal diagnosis strategy to obtain an abnormal diagnosis result, wherein the abnormal diagnosis result includes at least an abnormal node identifier, an abnormal type, or an abnormal severity; and a third determining subunit, used to generate early warning information based on the early warning level and the abnormal diagnosis result.

[0189] Optionally, in the structural status early warning device for communication towers provided in this application embodiment, the device further includes: a fifth acquisition unit, used to acquire an initial multimodal dataset of the monitored communication tower in the current monitoring period before acquiring the multimodal dataset of the monitored communication tower in the current monitoring period, wherein the initial multimodal dataset includes at least wind speed, wind direction, temperature and structural response data; and a third processing unit, used to use the initial multimodal dataset as the multimodal dataset of the current monitoring period, or to analyze based on wind speed, wind direction, temperature and target finite element model to obtain predicted structural response data, and to calculate based on the structural response data and predicted structural response data to obtain a target response dataset, and to use the target response dataset as the multimodal dataset of the current monitoring period.

[0190] It should be noted that the first acquisition unit 301, the second acquisition unit 302, the first processing unit 303, and the first determination unit 304 mentioned above correspond to steps S201 to S204 in Embodiment 1. The four units and the corresponding steps implement the same instances and application scenarios, but are not limited to the content disclosed in Embodiment 1. It should be noted that the above modules or units can be hardware or software components stored in memory (e.g., memory 104) and processed by one or more processors (e.g., processors 102a, 102b, ..., 102n). The above units can also be part of a device and run in the computer terminal 10 provided in Embodiment 1.

[0191] Example 3

[0192] Embodiments of this application may provide an electronic device. Figure 4 This is a structural block diagram of an electronic device according to an embodiment of this application. Figure 4 As shown, the electronic device may include: one or more ( Figure 4 (Only one is shown) Processor 402, memory 404, memory controller, and peripheral interface, wherein the peripheral interface is connected to the radio frequency module, audio module and display.

[0193] The memory can be used to store software programs and modules, such as the program instructions / modules corresponding to the methods and apparatus in the embodiments of this application. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, thereby implementing the above-described methods. The memory may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include memory remotely located relative to the processor, and these remote memories can be connected to the terminal via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0194] Those skilled in the art will understand that Figure 4 The structure shown is for illustrative purposes only. Electronic devices can also be smartphones, tablets, handheld computers, mobile internet devices (MIDs), PADs, and other terminal devices. Figure 4 This does not limit the structure of the aforementioned electronic device. For example, electronic devices may also include components that are more... Figure 4 The more or fewer components shown (such as network interfaces, display devices, etc.), or having the same Figure 4 The different configurations shown.

[0195] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing the hardware related to the terminal device. The program can be stored in a computer-readable storage medium, which may include: flash drive, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0196] Example 4

[0197] Embodiments of this application also provide a computer-readable storage medium. Optionally, in this embodiment, the storage medium can be used to store the program code executed by the communication tower structural status early warning method provided in Embodiment 1.

[0198] Optionally, in this embodiment, the storage medium may be located in any computer terminal in a group of computer terminals in a computer network, or in any mobile terminal in a group of mobile terminals.

[0199] This application also provides a computer program product, which, when executed on a data processing device, is suitable for performing the steps of a method for early warning of the structural status of a communication tower.

[0200] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0201] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0202] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0203] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0204] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0205] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard drive, magnetic disk, or optical disk.

[0206] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for early warning of the structural status of a communication tower, characterized in that, include: Obtain the multimodal dataset of the monitored communication tower during the current monitoring period; Obtain the structural dynamics generative causal graph corresponding to the current monitoring period, wherein the structural dynamics generative causal graph includes multiple physical nodes and directed edges connecting the multiple physical nodes, the directed edges are used to characterize the vibration transmission causal relationship between nodes, each directed edge corresponds to a vibration transmission model, and the vibration transmission model is obtained by training a generative adversarial network through a sample dataset; Based on the multimodal dataset of the current monitoring period and the structural dynamics generative causal graph corresponding to the current monitoring period, inference calculations are performed to obtain the overall structural health index corresponding to the monitored communication tower; The early warning information corresponding to the monitored communication tower is determined based on the overall structural health index.

2. The method according to claim 1, characterized in that, Based on the multimodal dataset of the current monitoring period and the structural dynamics generative causal graph corresponding to the current monitoring period, inference calculations are performed to obtain the overall structural health index corresponding to the monitored communication tower, including: The multimodal dataset of the current monitoring period is input into the structural dynamics generative cause-effect graph corresponding to the current monitoring period. For each directed edge, the vibration response is predicted based on the multimodal dataset of the current monitoring period using the vibration transmission model corresponding to the directed edge, and the predicted value is obtained. Based on the predicted value and the actual observed value corresponding to the directed edge, the deviation is calculated and smoothed to obtain the target deviation value corresponding to the directed edge; The overall deviation value of the monitored communication tower is obtained by performing a weighted average calculation based on the target deviation value corresponding to all the directed edges and the weight corresponding to each directed edge, and the overall structural health index of the monitored communication tower is determined based on the overall deviation value.

3. The method according to claim 1, characterized in that, The structural dynamics generative causal graph is obtained through the following steps: A target finite element model of the monitored communication tower is obtained, wherein the target finite element model is obtained by calibrating the initial finite element model of the monitored communication tower using the historical multimodal dataset of the monitored communication tower; Based on the target finite element model, multiple physical nodes of the monitored communication tower are determined, and an initial causal graph topology is constructed based on the multiple physical nodes; Based on the historical multimodal dataset and the initial causal graph topology, the structural dynamics generative causal graph is determined.

4. The method according to claim 3, characterized in that, Based on the historical multimodal dataset and the initial causal graph topology, the structural dynamics generative causal graph is determined as follows: An initial vibration transmission model is configured for each directed edge in the initial causal graph topology to obtain an initial structural dynamics generative causal graph; Based on the initial structural dynamics generative causal graph and the multilayer fully connected neural network, the generative adversarial network is constructed. The sample dataset is extracted from the historical multimodal dataset, and the generative adversarial network is trained based on the sample dataset until the preset conditions are met, thereby obtaining the structural dynamics generative causal graph.

5. The method according to claim 1, characterized in that, The structural dynamics generative cause-effect graph corresponding to the current monitoring period is obtained through the following steps: Obtain the target multimodal dataset, wherein the target multimodal dataset is the dataset extracted from the multimodal dataset of the previous calibration period of the current monitoring period, excluding data within the time period during which an early warning was triggered, and data where the wind speed is greater than a preset threshold or the temperature exceeds a preset range; Based on the target multimodal dataset, the structural dynamics generative causal graph corresponding to the previous calibration cycle of the current monitoring cycle is calibrated and updated to obtain the structural dynamics generative causal graph corresponding to the current monitoring cycle.

6. The method according to claim 1, characterized in that, The early warning information corresponding to the monitored communication tower determined based on the overall structural health index includes: The warning level corresponding to the monitored communication tower is determined based on the overall structural health index. Based on the warning level, an anomaly diagnosis strategy corresponding to the monitored communication tower is determined, and the anomaly diagnosis strategy is executed to obtain an anomaly diagnosis result, wherein the anomaly diagnosis result includes at least the anomaly node identifier, anomaly type, or anomaly severity. The warning information is generated based on the warning level and the anomaly diagnosis results.

7. The method according to claim 1, characterized in that, Before acquiring the multimodal dataset of the monitored communication tower in the current monitoring period, the method further includes: The initial multimodal dataset of the monitored communication tower in the current monitoring period is obtained, wherein the initial multimodal dataset includes at least wind speed, wind direction, temperature and structural response data; The initial multimodal dataset is used as the multimodal dataset for the current monitoring period. Alternatively, the predicted structural response data is obtained by analyzing the wind speed, wind direction, temperature, and target finite element model. The target response dataset is then calculated based on the structural response data and the predicted structural response data, and used as the multimodal dataset for the current monitoring period.

8. A structural status early warning device for a communication tower, characterized in that, include: The first acquisition unit is used to acquire the multimodal dataset of the monitored communication tower in the current monitoring period; The second acquisition unit is used to acquire the structural dynamics generative causal graph corresponding to the current monitoring period. The structural dynamics generative causal graph includes multiple physical nodes and directed edges connecting the multiple physical nodes. The directed edges are used to characterize the vibration transmission causal relationship between nodes. Each directed edge corresponds to a vibration transmission model. The vibration transmission model is obtained by training a generative adversarial network using a sample dataset. The first processing unit is used to perform inference calculations based on the multimodal dataset of the current monitoring period and the structural dynamics generative causal graph corresponding to the current monitoring period to obtain the overall structural health index corresponding to the monitored communication tower. The first determining unit is used to determine the early warning information corresponding to the monitored communication tower based on the overall structural health index.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored executable program, wherein, when the executable program is executed, it controls the device containing the computer-readable storage medium to perform the structural status early warning method for a communication tower as described in any one of claims 1 to 7.

10. An electronic device, characterized in that, include: Memory, which stores executable programs; A processor for running the program, wherein the program executes the structural status early warning method for communication towers according to any one of claims 1 to 7.

Images