A line structure health early warning method and system

Abstract

This invention provides a method and system for early warning of line structure health. By collecting multi-source heterogeneous monitoring data of the line structure, and based on the reliability differences of different types of monitoring data, an adaptive weight fusion algorithm is designed to fuse the multi-source heterogeneous monitoring data to obtain comprehensive feature values. Historical health data, historical minor hazard data, and historical serious hazard data are selected as training samples to train an improved attention ELM model, completing model training. The comprehensive feature values ​​collected and fused in real time are used to construct feature vectors according to time windows, which are then input into the trained improved attention ELM model to output the probability distribution of the line health status level. The probability of the line health status level and the comprehensive feature values ​​are compared with the corresponding probability thresholds and warning thresholds to perform graded warnings, achieving accurate identification and graded warning of early hazards in the line structure, improving the reliability and real-time performance of the warnings.

Description

Technical Field

[0001] This invention belongs to the field of line structure health early warning technology, and specifically relates to a line structure health early warning method and system. Background Technology

[0002] As a core component of infrastructure, the health of power lines is directly related to public safety, energy transmission, and smooth communication.

[0003] Existing line structure health early warning methods suffer from the following shortcomings: First, the monitoring indicators are singular, mostly focusing on monitoring single physical quantities such as current and vibration, ignoring the coupling influence of environmental factors and structural parameters, resulting in strong one-sidedness and high false alarm rate in early warnings; second, the data fusion methods are simple, mostly using fixed weight fusion or direct splicing, unable to dynamically adjust the fusion strategy according to data reliability, and difficult to adapt to the fluctuating characteristics of sensor data in complex environments; third, the anomaly early warning algorithms have low accuracy and strong lag, traditional threshold methods can only identify obvious anomalies that have already occurred, and cannot capture early gradual hidden dangers, and machine learning algorithms often suffer from overfitting and insufficient sensitivity to key features, making it difficult to meet the requirements of high-precision early warning. Summary of the Invention

[0004] Based on this, the present invention provides a method and system for early warning of line structure health, which aims to achieve accurate identification and graded early warning of early hidden dangers in line structure, and improve the reliability and real-time performance of early warning.

[0005] A first aspect of this invention provides a method for early warning of line structure health, the method comprising: Collect multi-source heterogeneous monitoring data of the line structure, including strain data, vibration data, traveling wave current data, and environmental data; Based on the reliability differences of different types of monitoring data, an adaptive weighted fusion algorithm is designed to fuse multi-source heterogeneous monitoring data and obtain comprehensive feature values ​​of the line structure health status. Historical health data, historical minor hidden danger data, and historical serious hidden danger data were selected as training samples. The comprehensive feature values ​​were divided according to time windows, the feature vector of each time window was determined, and the corresponding line health status level was constructed to build training set and test set. An improved attention ELM model is constructed by inputting the training set into the improved attention ELM model and completing the model training by solving the output layer weights. The comprehensive feature values ​​collected and fused in real time are used to construct feature vectors according to time windows, which are then input into the trained improved attention ELM model to output the probability distribution of the line health status level. The probability and comprehensive characteristic value of the line health status level are compared with the corresponding probability threshold and warning threshold to conduct graded warnings.

[0006] Furthermore, based on the reliability differences of different types of monitoring data, an adaptive weighted fusion algorithm is designed to fuse multi-source heterogeneous monitoring data to obtain the comprehensive feature value of the line structure health status. The formula for calculating the comprehensive feature value is as follows: ; ; ; ; ; in, Let i be the normalized value of the monitoring data of the i-th type. For the adaptive fusion weights of the i-th type of monitoring data, Let i be the reliability index of the i-th type of monitoring data. Let be the signal-to-noise ratio of the i-th type of monitoring data. For the i-th type of monitoring raw data, Let N be the denoised monitoring data of the i-th class, where N is the data length. Let be the coefficient of variation of the i-th type of monitoring data. Let be the standard deviation of the i-th type of monitoring data. Let α be the mean of the i-th type of monitoring data, α be the weight coefficient, and j be the global traversal index.

[0007] Furthermore, the improved attention ELM model includes an input layer, an attention layer, a hidden layer, and an output layer. The input of the input layer is a feature vector, which corresponds to the comprehensive feature value of the time window. The attention layer employs a self-attention mechanism, assigning different weights to each time point of the input feature vector to highlight key anomaly features. The formula for calculating the attention weights is as follows: ; x is the weight coefficient of the k-th element. k Let k be the input feature vector of the kth element. Let b be the attention weight matrix. a As the bias term, the softmax function is used to normalize the attention weights, δ is the feedback coefficient, and the output of the attention layer is: ; The hidden layer has L hidden neurons, using sigmoid as the activation function. The formula for calculating the hidden layer output H is: ; This is the hidden layer weight matrix. For hidden layer bias terms, For the set of real numbers, It is an L-row, 1-column column vector of real numbers; The output layer uses the Softmax function as its activation function, and the output layer weights are based on the probability distribution of the output circuit's health status level. The least squares method is used to solve this problem, and an L2 regularization term is introduced to avoid overfitting. The solution formula is as follows: ; Y is the training sample label matrix, λ is the regularization coefficient, and I is the identity matrix. This is the transpose of the hidden layer output matrix.

[0008] Further, the process involves collecting and fusing comprehensive feature values ​​in real time, constructing feature vectors according to time windows, inputting them into a trained improved attention ELM model, and outputting the probability distribution of the line health status level. Following this step, the process includes: Based on the probability distribution of the line health status level, it is divided into three categories: stable state, gradual abnormal state, and sudden abnormal state. These are used as trigger conditions for adjusting the warning threshold. Among them, the stable state is when the probability of a single health status level is greater than or equal to the first preset value, and the prediction results of a consecutive preset number of time windows do not fluctuate. In this case, it is marked as no need to adjust the warning threshold. If the probability of a gradually changing abnormal state being a minor or serious hazard continues to rise, but has not reached the threshold, it is marked as a minor adjustment of the warning threshold. If the probability of a sudden abnormal state being a minor or serious hazard increases instantly, or if the probability of both minor and serious hazard states is greater than or equal to the second preset value, then it is marked as an emergency adjustment warning threshold. The warning threshold is dynamically updated based on the stable state, the gradually changing abnormal state, and the sudden abnormal state. The calculation formula is as follows: ; This is the real-time early warning threshold for health state type l. The historical mean of the comprehensive characteristic values ​​of the l-th health state. To adjust the coefficient, Let be the real-time standard deviation of the comprehensive characteristic value of the l-th health state, and k be the linkage coefficient. Different health states correspond to different linkage coefficients. The formula for calculating the real-time standard deviation of the comprehensive characteristic value of the l-th health state is: ; Where S(tk) is the (k+1)th element in the feature vector.

[0009] Furthermore, the step of dynamically updating the early warning threshold based on stable state, gradual abnormal state, and sudden abnormal state includes: At each preset number of time windows, the current warning threshold is calibrated by combining real-time prediction results with the comprehensive feature values ​​at the corresponding time. This includes: If the prediction result is a healthy state, but the comprehensive feature value does not reach the health threshold, and this situation occurs twice in a row, it is determined that the health threshold is too high. The adjustment coefficient of the next warning threshold update will be lowered, and the corresponding feature vector will be recorded as a new negative sample for improving the attention ELM model. If the prediction result is a minor hazard state, but the comprehensive feature value reaches the serious hazard threshold, it is determined that the minor hazard threshold is too high. The adjustment coefficient will be increased, and the corresponding feature vector will be recorded to optimize and improve the attention ELM model's accuracy in distinguishing between minor and serious hazard states. If the prediction result indicates a serious hidden danger state, but the comprehensive feature value does not reach the serious hidden danger threshold, it is determined that the serious hidden danger threshold is too low. The adjustment coefficient will be lowered, and the attention weight of the corresponding feature vector will be strengthened. Abnormal matching samples recorded during the threshold calibration process (samples where the predicted result does not match the threshold determination) are fed back in real time to the training step of the improved attention ELM model. These samples serve as incremental samples for online fine-tuning of the improved attention ELM model. The fine-tuning formula is as follows: ; in, To fine-tune the output layer weights, For the original weights, For learning rate, Labels for abnormal matching samples. This is the hidden layer output for the abnormal matching samples.

[0010] Furthermore, if the input to the attention layer is a high-sensitivity feature identified in the threshold calibration, a preset value is assigned to the feedback coefficient to increase the weight coefficient of the corresponding feature vector.

[0011] Furthermore, in the step of comparing the probability and comprehensive characteristic value of the line health status level with the corresponding probability threshold and warning threshold to conduct graded warnings, When the probability of the line health status level is greater than or equal to the first probability threshold, and the comprehensive feature value is greater than or equal to the first warning threshold, a normal operation signal is output. When the probability of the line health status level is greater than or equal to the second probability threshold, and the comprehensive feature value is between the first warning threshold and the second warning threshold, a hidden danger warning signal is output. An emergency warning signal is output when the probability of the line health status level is greater than or equal to the second probability threshold and the comprehensive feature value is less than the second warning threshold.

[0012] A second aspect of this invention provides a line structure health early warning system for implementing the line structure health early warning method of the first aspect. The system includes: The acquisition module is used to collect multi-source heterogeneous monitoring data of the line structure, including strain data, vibration data, traveling wave current data, and environmental data. The fusion module is used to design an adaptive weighted fusion algorithm based on the reliability differences of different types of monitoring data, to fuse multi-source heterogeneous monitoring data and obtain a comprehensive feature value of the health status of the line structure. The module is used to select historical health data, historical minor hidden danger data, and historical serious hidden danger data as training samples, divide the comprehensive feature values ​​according to time windows, determine the feature vector of each time window and the corresponding line health status level, and build training set and test set. The training module is used to build an improved attention ELM model. The training set is input into the improved attention ELM model, and the model training is completed by solving the output layer weights. The input module is used to construct a feature vector by dividing the real-time collected and fused comprehensive feature values ​​into time windows, input the trained improved attention ELM model, and output the probability distribution of the line health status level. The comparison module is used to compare the probability and comprehensive feature value of the line health status level with the corresponding probability threshold and warning threshold to perform graded warnings.

[0013] A third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the circuit structure health early warning method provided in the first aspect.

[0014] A fourth aspect of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the circuit structure health early warning method provided in the first aspect.

[0015] This invention provides a method and system for early warning of line structure health. By collecting multi-source heterogeneous monitoring data of the line structure, and considering the reliability differences of different types of monitoring data, an adaptive weight fusion algorithm is designed to fuse the multi-source heterogeneous monitoring data to obtain comprehensive feature values. Historical health data, historical minor hazard data, and historical serious hazard data are selected as training samples. The comprehensive feature values ​​are divided according to time windows to determine the feature vector of each time window and the corresponding line health status level, thus constructing a training set and a test set. An improved attention ELM model is constructed, and the training set is input into the improved attention ELM model. The model is trained by solving the output layer weights. The comprehensive feature values ​​collected and fused in real time are used to construct feature vectors according to time windows and input into the trained improved attention ELM model to output the probability distribution of the line health status level. The probability of the line health status level and the comprehensive feature value are compared with the corresponding probability threshold and warning threshold for graded warning. This achieves accurate identification and graded warning of early hazards in the line structure, improving the reliability and real-time performance of the warning. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the implementation of a line structure health early warning method according to Embodiment 1 of the present invention. Figure 2 This is a structural block diagram of a line structure health early warning system provided in Embodiment 2 of the present invention; Figure 3 This is a structural block diagram of an electronic device provided in Embodiment 3 of the present invention. Detailed Implementation

[0017] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Several embodiments of the invention are illustrated in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.

[0018] It should be noted that when a component is said to be "fixed to" another component, it can be directly on the other component or there may be an intervening component. When a component is said to be "connected to" another component, it can be directly connected to the other component or there may be an intervening component. The terms "vertical," "horizontal," "left," "right," and similar expressions used in this document are for illustrative purposes only.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0020] Example 1 According to an embodiment of the present invention, a method for early warning of line structure health is 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.

[0021] This first embodiment provides a method for early warning of circuit structure health, which can be used in electronic devices, such as computers. Please refer to... Figure 1 , Figure 1 The flowchart of a line structure health early warning method provided in Embodiment 1 of the present invention is shown, specifically including steps S01 to S06.

[0022] Step S01: Collect multi-source heterogeneous monitoring data of the line structure, including strain data, vibration data, traveling wave current data, and environmental data.

[0023] Specifically, the multi-source heterogeneous monitoring data actually includes structural parameter data, electrical parameter data, and environmental parameter data. It can be understood that the structural parameter data is collected by fiber optic strain sensors to collect strain data of key sections of the line, and by triaxial vibration sensors to collect vibration acceleration data (i.e., vibration data) of the line; the electrical parameter data is collected by traveling wave current sensors to collect traveling wave current signals (i.e., traveling wave current data); and the environmental parameter data (i.e., environmental data) is collected by temperature and humidity sensors.

[0024] Furthermore, a timestamp synchronization mechanism is adopted to uniformly synchronize the data collected by all sensors to the edge computing node, with the synchronization error controlled within ±5ms, to avoid data fusion distortion caused by timing deviations.

[0025] Step S02: Based on the reliability differences of different types of monitoring data, an adaptive weight fusion algorithm is designed to fuse multi-source heterogeneous monitoring data to obtain a comprehensive feature value of the line structure health status.

[0026] It should be noted that prior to this, the monitoring data underwent preprocessing operations including outlier removal, signal denoising, and normalization. Specifically, the formula for calculating the comprehensive feature value is: ; ; ; ; ; Where i = 1, 2, 3, 4, corresponding to strain, vibration, traveling wave current, and environmental data, respectively. Let i be the normalized value of the monitoring data of the i-th type. For the adaptive fusion weights of the i-th type of monitoring data, Let i be the reliability index of the i-th type of monitoring data. Let be the signal-to-noise ratio of the i-th type of monitoring data. For the i-th type of monitoring raw data, Let N be the denoised monitoring data of the i-th class, where N is the data length. Let be the coefficient of variation of the i-th type of monitoring data. Let be the standard deviation of the i-th type of monitoring data. Let S(t) be the mean of the i-th type of monitoring data, α be the weight coefficient, and j be the global traversal index. It can be understood that S(t)∈[0,1]. The closer S(t) is to 1, the better the health status of the line structure. The closer it is to 0, the higher the probability of hidden dangers in the line structure.

[0027] Step S03: Select historical health data, historical minor hidden danger data, and historical serious hidden danger data as training samples. Divide the comprehensive feature values ​​according to time windows, determine the feature vector of each time window, and the corresponding line health status level, and construct the training set and test set.

[0028] In this embodiment of the invention, historical health data, minor hazard data, and serious hazard data are selected as training samples. The comprehensive feature value S(t) is divided into time windows with a window length of 50 (corresponding to 5 seconds of data). The feature vector of each window is X = [S(t-49), S(t-48), ..., S(t)], and the corresponding label is the line health status level (0: healthy, 1: minor hazard, 2: serious hazard). The training set X_train and the test set X_test are constructed.

[0029] The improved attention ELM model consists of an input layer, an attention layer, hidden layers, and an output layer, where the input layer takes a feature vector as its input. This corresponds to the comprehensive feature value of 50 time windows. 1 represents the feature vector of a single sample, and 50 represents that the vector contains 50 consecutive time-series feature values, corresponding to the time window length set above. The attention layer employs a self-attention mechanism, assigning different weights to each time point of the input feature vector to highlight key anomaly features. The formula for calculating the attention weights is as follows: ; x is the weight coefficient of the k-th element. k Let k be the input feature vector of the kth element. Let b be the attention weight matrix. a As a bias term, the softmax function is used to normalize the attention weights, and δ is the feedback coefficient. If x k For high-sensitivity features identified during threshold calibration, δ=0.02 (increase the attention weight of this feature); otherwise, δ=0, to achieve collaborative extraction of anomalous features. The output of the attention layer is: ; The hidden layer has L hidden neurons, using sigmoid as the activation function. The formula for calculating the hidden layer output H is: ; This is the hidden layer weight matrix. For hidden layer bias terms, For the set of real numbers, It is an L-row, 1-column column vector of real numbers; The output layer uses the Softmax function as the activation function to output the probability distribution of the health status level of the output line.

[0030] Step S04: Construct an improved attention ELM model by inputting the training set into the improved attention ELM model and solving for the output layer weights to complete the model training.

[0031] Specifically, output layer weights The least squares method is used to solve this problem, and an L2 regularization term is introduced to avoid overfitting. The solution formula is as follows: ; Y is the training sample label matrix ( (where n is the number of training samples), Y is an n-row, 3-column real matrix, where 3 represents the number of health status categories, corresponding to three levels: healthy, minor risk, and serious risk. λ is the regularization coefficient, and I is the identity matrix. This is the transpose of the hidden layer output matrix.

[0032] Step S05: The comprehensive feature values ​​collected and fused in real time are used to construct feature vectors according to time windows, which are then input into the trained improved attention ELM model to output the probability distribution of the line health status level.

[0033] It should be noted that, based on the probability distribution of the line health status level, it is divided into three categories: stable state, gradual abnormal state, and sudden abnormal state, which serve as the triggering conditions for adjusting the warning threshold. Among them, the stable state is when the probability of a single health status level is greater than or equal to the first preset value, and the prediction results of a consecutive preset number of time windows are without fluctuation, and it is marked as no need to adjust the warning threshold. For example, if the probability of a single health status level is ≥90%, and the prediction results are without fluctuation for 5 consecutive time windows (25s), it is marked as "no need to adjust the threshold". If the probability of a gradually changing abnormal state being a minor or serious hazard continues to rise but does not reach the threshold, it is marked as a minor adjustment to the warning threshold. For example, if the probability of a certain abnormal level (minor / serious hazard) continues to rise (the increase is ≥5% for 3 consecutive time windows) but does not reach the warning threshold, it is marked as a "minor adjustment threshold". If the probability of a sudden abnormal state is a minor or serious hazard that rises instantly, or if the probability of both minor and serious hazard states is greater than or equal to the second preset value, it is marked as an emergency adjustment warning threshold. For example, if the probability of a certain abnormal state rises instantly (the increase is ≥20% in a single time window), or if the probability of two abnormal states occurs simultaneously with ≥50%, it is marked as an "emergency adjustment threshold". The warning threshold is dynamically updated based on the stable state, the gradually changing abnormal state, and the sudden abnormal state. The calculation formula is as follows: ; The real-time warning threshold for the l-th type of health status is defined as l=0 or 1, where l=0 corresponds to a healthy state and l=1 corresponds to a serious hazard state. T0(t) is the health status threshold used to determine whether the line is in a healthy state, and T1(t) is the critical threshold for serious hazard, used to distinguish between minor and serious hazards. The historical mean of the comprehensive characteristic values ​​of the l-th health state. To adjust the coefficient, Let be the real-time standard deviation of the comprehensive characteristic value of the l-th health state, and k be the linkage coefficient. Different health states correspond to different linkage coefficients. In the embodiments of this invention, k=0.8 in the stable state (reducing the threshold adjustment range to avoid over-adjustment); k=1.2 in the gradually changing abnormal state (maintaining the original adjustment range to adapt to the gradual trend); and k=1.5 in the sudden abnormal state (increasing the adjustment range to quickly adapt to sudden hidden dangers). In addition, in the stable state, the threshold update frequency is changed from real-time update to update once every 3 time windows (15s) to reduce computational overhead; in the gradually changing abnormal state, real-time update is maintained (updated once every 1 time window); in the sudden abnormal state, the threshold is triggered to be updated urgently (updated once every 0.5 time windows, i.e., updated once every 2.5s) to ensure rapid response to sudden hidden dangers. The formula for calculating the real-time standard deviation of the comprehensive characteristic values ​​of health state type l is: ; Where S(tk) is the (k+1)th element in the feature vector.

[0034] Furthermore, at preset time windows, the current warning threshold is calibrated by combining the real-time prediction results with the comprehensive feature values ​​at the corresponding times. It should be noted that data points with inconsistent prediction results and comprehensive feature values ​​S(t) within the corresponding time windows are uniformly marked as high-sensitivity features, specifically including: If the predicted result is a healthy state, but the comprehensive feature value does not reach the health threshold (S(t) < T0(t)), and this situation occurs twice in a row, it is determined that the health threshold is too high. The adjustment coefficient of the next warning threshold update is lowered (for example, by 0.1, and at the lowest to 0.9). At the same time, the corresponding feature vector is recorded as a new negative sample for improving the attention ELM model (to avoid misjudging the health state). If the prediction result is a minor hazard state, but the comprehensive feature value reaches the serious hazard threshold (S(t) < T1(t)), it is determined that the minor hazard threshold is too high, and the adjustment coefficient is increased (for example, increased by 0.1, up to a maximum of 1.5). At the same time, the corresponding feature vector is recorded to optimize and improve the attention ELM model's accuracy in distinguishing between minor and serious hazard states. If the prediction result is a serious hidden danger state, but the comprehensive feature value does not reach the serious hidden danger threshold (S(t)≥T1(t)), it is determined that the serious hidden danger threshold is too low, and the adjustment coefficient is lowered (for example, lowered by 0.1), while strengthening the attention weight of the corresponding feature vector (increasing the training weight of the feature vector in the attention layer of the improved attention ELM model). The abnormal matching samples recorded during the threshold calibration process, where the abnormal matching samples are those with prediction results not matching the determination of the comprehensive eigenvalue, are fed back in real time to the training step of the improved attention ELM model and used as incremental samples to perform online fine-tuning on the improved attention ELM model. The fine-tuning formula is: ; where, is the weight of the output layer after fine-tuning, is the original weight, is the learning rate, is the label of the abnormal matching sample, is the output of the hidden layer of the abnormal matching sample.

[0035] Step S06: Compare the probability of the line health status level and the comprehensive eigenvalue with the corresponding probability threshold and warning threshold to perform hierarchical warning.

[0036] Specifically, when the probability of the line health status level is greater than or equal to the first probability threshold (exemplarily, the first probability threshold is 90%), and the comprehensive eigenvalue is greater than or equal to the first warning threshold (i.e., S(t) ≥ T0(t)), a normal operation signal is output; when the probability of the line health status level is greater than or equal to the second probability threshold (exemplarily, the second probability threshold is 60%), and the comprehensive eigenvalue is between the first warning threshold and the second warning threshold (i.e., T1(t) ≤ S(t) < T0(t)), a potential hazard reminder signal is output; when the probability of the line health status level is greater than or equal to the second probability threshold, and the comprehensive eigenvalue is less than the second warning threshold (i.e., S(t) < T1(t)), an emergency warning signal is output.

[0037] In summary, the line structure health early warning method in the above embodiments of the present invention collects multi-source heterogeneous monitoring data of the line structure, designs an adaptive weight fusion algorithm based on the reliability differences of different types of monitoring data, fuses the multi-source heterogeneous monitoring data to obtain comprehensive feature values; selects historical health data, historical minor hidden danger data, and historical serious hidden danger data as training samples, divides the comprehensive feature values ​​according to time windows, determines the feature vector of each time window and the corresponding line health status level, and constructs training and testing sets; constructs an improved attention ELM model, inputs the training set into the improved attention ELM model, and completes model training by solving the output layer weights; constructs feature vectors according to time windows using the comprehensive feature values ​​collected and fused in real time, inputs them into the trained improved attention ELM model, and outputs the probability distribution of the line health status level; compares the probability of the line health status level and the comprehensive feature value with the corresponding probability threshold and early warning threshold to perform graded early warning, realizing accurate identification and graded early warning of early hidden dangers in line structures, and improving the reliability and real-time performance of early warning.

[0038] Example 2 Please see Figure 2 , Figure 2 This is a structural block diagram of a line structure health early warning system provided in Embodiment 2 of the present invention. This line structure health early warning system 200 is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the apparatus described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0039] Specifically, the line structure health early warning system 200 includes: a data acquisition module 21, a fusion module 22, a construction module 23, a training module 24, an input module 25, and a comparison module 26, wherein: The acquisition module 21 is used to acquire multi-source heterogeneous monitoring data of the line structure, including strain data, vibration data, traveling wave current data and environmental data. Fusion module 22 is used to design an adaptive weighted fusion algorithm based on the reliability differences of different types of monitoring data to fuse multi-source heterogeneous monitoring data and obtain a comprehensive feature value of the line structure health status. The formula for calculating the comprehensive feature value is as follows: ; ; ; ; ; in, Let i be the normalized value of the monitoring data of the i-th type. For the adaptive fusion weights of the i-th type of monitoring data, Let i be the reliability index of the i-th type of monitoring data. Let be the signal-to-noise ratio of the i-th type of monitoring data. For the i-th type of monitoring raw data, Let N be the denoised monitoring data of the i-th class, where N is the data length. Let be the coefficient of variation of the i-th type of monitoring data. Let be the standard deviation of the i-th type of monitoring data. Let α be the mean of the i-th type of monitoring data, α be the weight coefficient, and j be the global traversal index. Module 23 is used to select historical health data, historical minor hidden danger data, and historical serious hidden danger data as training samples, divide the comprehensive feature values ​​according to time windows, determine the feature vector of each time window and the corresponding line health status level, and construct training set and test set; Training module 24 is used to construct an improved attention ELM model. The training set is input into the improved attention ELM model, and the model training is completed by solving the output layer weights. The improved attention ELM model includes an input layer, an attention layer, a hidden layer and an output layer. The input of the input layer is a feature vector, which corresponds to the comprehensive feature value of the time window. The attention layer employs a self-attention mechanism, assigning different weights to each time point of the input feature vector to highlight key anomaly features. The formula for calculating the attention weights is as follows: ; x is the weight coefficient of the k-th element. k Let k be the input feature vector of the kth element. Let b be the attention weight matrix. a As the bias term, the softmax function is used to normalize the attention weights, δ is the feedback coefficient, and the output of the attention layer is: ; The hidden layer has L hidden neurons, using sigmoid as the activation function. The formula for calculating the hidden layer output H is: ; This is the hidden layer weight matrix. For hidden layer bias terms, For the set of real numbers, It is an L-row, 1-column column vector of real numbers; The output layer uses the Softmax function as its activation function, and the output layer weights are based on the probability distribution of the output circuit's health status level. The least squares method is used to solve this problem, and an L2 regularization term is introduced to avoid overfitting. The solution formula is as follows: ; Y is the training sample label matrix, λ is the regularization coefficient, and I is the identity matrix. This is the transpose of the hidden layer output matrix; If the input to the attention layer is a high-sensitivity feature identified in the threshold calibration, then a preset value is assigned to the feedback coefficient to increase the weight coefficient of the corresponding feature vector. Input module 25 is used to construct feature vectors by time window based on the comprehensive feature values ​​collected and fused in real time, input them into the trained improved attention ELM model, and output the probability distribution of the line health status level. The comparison module 26 is used to compare the probability and comprehensive feature value of the line health status level with the corresponding probability threshold and warning threshold, and to perform graded warning. When the probability of the line health status level is greater than or equal to the first probability threshold and the comprehensive feature value is greater than or equal to the first warning threshold, a normal operation signal is output. When the probability of the line health status level is greater than or equal to the second probability threshold, and the comprehensive feature value is between the first warning threshold and the second warning threshold, a hidden danger warning signal is output. An emergency warning signal is output when the probability of the line health status level is greater than or equal to the second probability threshold and the comprehensive feature value is less than the second warning threshold.

[0040] Furthermore, in some optional embodiments of the present invention, the line structure health early warning system 200 further includes: The classification module is used to classify the line health status level into three categories: stable state, gradually abnormal state, and sudden abnormal state, based on the probability distribution of the line health status level. These categories serve as trigger conditions for adjusting the warning threshold. The stable state is defined as a single health status level with a probability greater than or equal to the first preset value, and the prediction results for a consecutive preset number of time windows show no fluctuations. In this case, the warning threshold does not need to be adjusted. If the probability of a gradually changing abnormal state being a minor or serious hazard continues to rise, but has not reached the threshold, it is marked as a minor adjustment of the warning threshold. If the probability of a sudden abnormal state being a minor or serious hazard increases instantly, or if the probability of both minor and serious hazard states is greater than or equal to the second preset value, then it is marked as an emergency adjustment warning threshold. The update module is used to dynamically update the warning threshold based on stable state, gradual abnormal state, and sudden abnormal state. The calculation formula is as follows: ; This is the real-time early warning threshold for health state type l. The historical mean of the comprehensive characteristic values ​​of the l-th health state. To adjust the coefficient, Let be the real-time standard deviation of the comprehensive characteristic value of the l-th health state, and k be the linkage coefficient. Different health states correspond to different linkage coefficients. The formula for calculating the real-time standard deviation of the comprehensive characteristic value of the l-th health state is: ; Where S(tk) is the (k+1)th element in the feature vector.

[0041] Furthermore, in some optional embodiments of the present invention, the line structure health early warning system 200 further includes: The calibration module, used at preset intervals within a certain number of time windows, combines real-time prediction results with the comprehensive feature values ​​at the corresponding time points to calibrate the current warning threshold. Specifically, it includes: If the prediction result is a healthy state, but the comprehensive feature value does not reach the health threshold, and this situation occurs twice in a row, it is determined that the health threshold is too high. The adjustment coefficient of the next warning threshold update will be lowered, and the corresponding feature vector will be recorded as a new negative sample for improving the attention ELM model. If the prediction result is a minor hazard state, but the comprehensive feature value reaches the serious hazard threshold, it is determined that the minor hazard threshold is too high. The adjustment coefficient will be increased, and the corresponding feature vector will be recorded to optimize and improve the attention ELM model's accuracy in distinguishing between minor and serious hazard states. If the prediction result indicates a serious hidden danger state, but the comprehensive feature value does not reach the serious hidden danger threshold, it is determined that the serious hidden danger threshold is too low. The adjustment coefficient will be lowered, and the attention weight of the corresponding feature vector will be strengthened. Abnormal matching samples recorded during the threshold calibration process (samples where the predicted result does not match the threshold determination) are fed back in real time to the training step of the improved attention ELM model. These samples serve as incremental samples for online fine-tuning of the improved attention ELM model. The fine-tuning formula is as follows: ; in, To fine-tune the output layer weights, For the original weights, For learning rate, Labels for abnormal matching samples. This is the hidden layer output for the abnormal matching samples.

[0042] Example 3 In another aspect, the present invention also proposes an electronic device, please refer to [link to relevant documentation]. Figure 3The electronic device shown is an embodiment of the present invention, including a memory 20, a processor 10, and a computer program 30 stored in the memory and executable on the processor. When the processor 10 executes the computer program 30, it implements the circuit structure health warning method as described above.

[0043] In some embodiments, the processor 10 may be a central processing unit (CPU), controller, microcontroller, microprocessor or other data processing chip, used to run program code stored in memory 20 or process data, such as executing access restriction programs.

[0044] The memory 20 includes at least one type of readable storage medium, including flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 20 can be an internal storage unit of an electronic device, such as the hard disk of the electronic device. In other embodiments, the memory 20 can also be an external storage device of the electronic device, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc. Furthermore, the memory 20 can include both internal and external storage units of the electronic device. The memory 20 can be used not only to store application software and various types of data of the electronic device, but also to temporarily store data that has been output or will be output.

[0045] It should be pointed out that, Figure 3 The structure shown does not constitute a limitation on the electronic device. In other embodiments, the electronic device may include fewer or more components than shown, or combine certain components, or have different component arrangements.

[0046] This invention also proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the circuit structure health early warning method described above.

[0047] Those skilled in the art will understand that the logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can mean any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0048] More specific examples (a non-exhaustive list) of computer-readable media include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which programs can be printed, because programs can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0049] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0050] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0051] The above embodiments merely illustrate several implementation methods of the present invention, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this patent should be determined by the appended claims.

Claims

1. A method for early warning of line structure health, characterized in that, The methods include: Collect multi-source heterogeneous monitoring data of the line structure, including strain data, vibration data, traveling wave current data, and environmental data; Based on the reliability differences of different types of monitoring data, an adaptive weighted fusion algorithm is designed to fuse multi-source heterogeneous monitoring data and obtain comprehensive feature values ​​of the line structure health status. Historical health data, historical minor hidden danger data, and historical serious hidden danger data were selected as training samples. The comprehensive feature values ​​were divided according to time windows, the feature vector of each time window was determined, and the corresponding line health status level was constructed to build training set and test set. An improved attention ELM model is constructed by inputting the training set into the improved attention ELM model and completing the model training by solving the output layer weights. The comprehensive feature values ​​collected and fused in real time are used to construct feature vectors according to time windows, which are then input into the trained improved attention ELM model to output the probability distribution of the line health status level. The probability and comprehensive characteristic value of the line health status level are compared with the corresponding probability threshold and warning threshold to conduct graded warnings.

2. The line structure health early warning method according to claim 1, characterized in that, Based on the reliability differences of different types of monitoring data, an adaptive weighted fusion algorithm is designed to fuse multi-source heterogeneous monitoring data to obtain the comprehensive feature value of the line structure health status. The formula for calculating the comprehensive feature value is as follows: ； ； ； ； ； in, Let i be the normalized value of the monitoring data of the i-th type. For the adaptive fusion weights of the i-th type of monitoring data, Let i be the reliability index of the i-th type of monitoring data. Let be the signal-to-noise ratio of the i-th type of monitoring data. For the i-th type of monitoring raw data, Let N be the denoised monitoring data of the i-th class, where N is the data length. Let be the coefficient of variation of the i-th type of monitoring data. Let be the standard deviation of the i-th type of monitoring data. Let α be the mean of the i-th type of monitoring data, α be the weight coefficient, and j be the global traversal index.

3. The line structure health early warning method according to claim 2, characterized in that, The improved attention ELM model consists of an input layer, an attention layer, a hidden layer, and an output layer. The input of the input layer is a feature vector, which corresponds to the comprehensive feature value of the time window. The attention layer employs a self-attention mechanism, assigning different weights to each time point of the input feature vector to highlight key anomaly features. The formula for calculating the attention weights is as follows: ； x is the weight coefficient of the k-th element. k Let k be the input feature vector of the kth element. Let b be the attention weight matrix. a As the bias term, the softmax function is used to normalize the attention weights, δ is the feedback coefficient, and the output of the attention layer is: ； The hidden layer has L hidden neurons, using sigmoid as the activation function. The formula for calculating the hidden layer output H is: ； This is the hidden layer weight matrix. For hidden layer bias terms, For the set of real numbers, It is an L-row, 1-column column vector of real numbers; The output layer uses the Softmax function as its activation function, and the output layer weights are based on the probability distribution of the output circuit's health status level. The least squares method is used to solve this problem, and an L2 regularization term is introduced to avoid overfitting. The solution formula is as follows: ； Y is the training sample label matrix, λ is the regularization coefficient, and I is the identity matrix. This is the transpose of the hidden layer output matrix.

4. The line structure health early warning method according to claim 3, characterized in that, The steps following the real-time collection and fusion of comprehensive feature values, constructing feature vectors according to time windows, inputting them into the trained improved attention ELM model, and outputting the probability distribution of the line health status level include: Based on the probability distribution of the line health status level, it is divided into three categories: stable state, gradual abnormal state, and sudden abnormal state. These are used as trigger conditions for adjusting the warning threshold. Among them, the stable state is when the probability of a single health status level is greater than or equal to the first preset value, and the prediction results of a consecutive preset number of time windows do not fluctuate. In this case, it is marked as no need to adjust the warning threshold. If the probability of a gradually changing abnormal state being a minor or serious hazard continues to rise, but has not reached the threshold, it is marked as a minor adjustment of the warning threshold. If the probability of a sudden abnormal state being a minor or serious hazard increases instantly, or if the probability of both minor and serious hazard states is greater than or equal to the second preset value, then it is marked as an emergency adjustment warning threshold. The warning threshold is dynamically updated based on the stable state, the gradually changing abnormal state, and the sudden abnormal state. The calculation formula is as follows: ； This is the real-time early warning threshold for health state type l. The historical mean of the comprehensive characteristic values ​​of the l-th health state. To adjust the coefficient, Let be the real-time standard deviation of the comprehensive characteristic value of the l-th health state, and k be the linkage coefficient. Different health states correspond to different linkage coefficients. The formula for calculating the real-time standard deviation of the comprehensive characteristic value of the l-th health state is: ； Where S(tk) is the (k+1)th element in the feature vector.

5. The line structure health early warning method according to claim 4, characterized in that, The step of dynamically updating the warning threshold based on stable state, gradual abnormal state, and sudden abnormal state includes: At each preset number of time windows, the current warning threshold is calibrated by combining real-time prediction results with the comprehensive feature values ​​at the corresponding time. This includes: If the prediction result is a healthy state, but the comprehensive feature value does not reach the health threshold, and this situation occurs twice in a row, it is determined that the health threshold is too high. The adjustment coefficient of the next warning threshold update will be lowered, and the corresponding feature vector will be recorded as a new negative sample for improving the attention ELM model. If the prediction result is a minor hazard state, but the comprehensive feature value reaches the serious hazard threshold, it is determined that the minor hazard threshold is too high. The adjustment coefficient will be increased, and the corresponding feature vector will be recorded to optimize and improve the attention ELM model's accuracy in distinguishing between minor and serious hazard states. If the prediction result indicates a serious hidden danger state, but the comprehensive feature value does not reach the serious hidden danger threshold, it is determined that the serious hidden danger threshold is too low. The adjustment coefficient will be lowered, and the attention weight of the corresponding feature vector will be strengthened. Abnormal matching samples recorded during the threshold calibration process (samples where the predicted result does not match the threshold determination) are fed back in real time to the training step of the improved attention ELM model. These samples serve as incremental samples for online fine-tuning of the improved attention ELM model. The fine-tuning formula is as follows: ； in, To fine-tune the output layer weights, For the original weights, For learning rate, Labels for abnormal matching samples. This is the hidden layer output for the abnormal matching samples.

6. The line structure health early warning method according to claim 5, characterized in that, If the input to the attention layer is a high-sensitivity feature identified in the threshold calibration, then a preset value is assigned to the feedback coefficient to increase the weight coefficient of the corresponding feature vector.

7. The line structure health early warning method according to claim 6, characterized in that, In the step of comparing the probability and comprehensive characteristic value of the line health status level with the corresponding probability threshold and warning threshold to conduct graded warnings, When the probability of the line health status level is greater than or equal to the first probability threshold, and the comprehensive feature value is greater than or equal to the first warning threshold, a normal operation signal is output. When the probability of the line health status level is greater than or equal to the second probability threshold, and the comprehensive feature value is between the first warning threshold and the second warning threshold, a hidden danger warning signal is output. An emergency warning signal is output when the probability of the line health status level is greater than or equal to the second probability threshold and the comprehensive feature value is less than the second warning threshold.

8. A line structure health early warning system, characterized in that, For implementing the line structure health early warning method as described in any one of claims 1-7, the system comprises: The acquisition module is used to collect multi-source heterogeneous monitoring data of the line structure, including strain data, vibration data, traveling wave current data, and environmental data. The fusion module is used to design an adaptive weighted fusion algorithm based on the reliability differences of different types of monitoring data, to fuse multi-source heterogeneous monitoring data and obtain a comprehensive feature value of the health status of the line structure. The module is used to select historical health data, historical minor hidden danger data, and historical serious hidden danger data as training samples, divide the comprehensive feature values ​​according to time windows, determine the feature vector of each time window and the corresponding line health status level, and build training set and test set. The training module is used to build an improved attention ELM model. The training set is input into the improved attention ELM model, and the model training is completed by solving the output layer weights. The input module is used to construct a feature vector by dividing the real-time collected and fused comprehensive feature values ​​into time windows, input the trained improved attention ELM model, and output the probability distribution of the line health status level. The comparison module is used to compare the probability and comprehensive feature value of the line health status level with the corresponding probability threshold and warning threshold to perform graded warnings.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the line structure health early warning method as described in any one of claims 1-7.

10. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the circuit structure health early warning method as described in any one of claims 1-7.

Images