Cable strand breakage early warning method, device and equipment and storage medium
Patent Information
- Application Number
- CN202411122054.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-15
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2044-08-15
AI Technical Summary
Existing cable strand detection methods are inefficient and fail to detect potential breakage risks in a timely manner. Traditional single-sensor methods cannot fully reflect complex working conditions. Multimodal sensor data fusion and feature extraction are insufficient, and the challenges of dynamic characteristics and nonlinear feature estimation have not been effectively addressed.
By employing multimodal sensor data acquisition, combined with multi-band distributed analysis, adaptive feature selection, covariance matching adaptive factor update mechanism, and dynamic window adjustment technology, an adaptive Kalman filter combined early warning algorithm is constructed. This algorithm is then combined with a multi-layer machine learning model for line break risk prediction, including a multi-channel temporal convolutional network, an autoencoder, a hierarchical attention mechanism network, and a recurrent neural network tree, to achieve data fusion and intelligent early warning.
It enables comprehensive perception and accurate description of cable status, improves the reliability and accuracy of status monitoring, enhances the sensitivity of anomaly detection and the robustness of status estimation, realizes intelligent line breakage early warning, and improves the pertinence and effectiveness of early warning.
Smart Images

Figure CN118940008A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of cable strands, and in particular to a cable strand disconnection warning method, device, equipment and storage medium. Background Art
[0002] Cable stranding systems play a key role in power transmission, communication transmission and other fields. Their safe and reliable operation is directly related to social and economic development and people's quality of life. However, due to the influence of long-term mechanical stress, environmental corrosion and other factors, cable stranding is prone to fatigue damage, structural degradation and other problems. In severe cases, it may lead to wire breakage accidents, causing huge economic losses and safety hazards. Traditional cable stranding detection methods mainly rely on manual inspections and regular inspections, which have defects such as low efficiency and poor real-time performance, and it is difficult to detect potential wire breakage risks in a timely manner.
[0003] In recent years, with the rapid development of sensing technology, signal processing and artificial intelligence, cable stranding condition monitoring and early warning methods based on multimodal sensing and intelligent algorithms have gradually become a research hotspot. However, there are still some problems in the current research: a single sensing method is difficult to fully reflect the complex working conditions of cable stranding, and the fusion and feature extraction of multimodal sensing data still need to be further studied; secondly, the dynamic characteristics and nonlinear characteristics of the cable stranding system bring challenges to state estimation, and it is necessary to develop more robust and adaptive algorithms; how to effectively use historical data and expert knowledge to build an accurate and reliable disconnection risk prediction model is still a problem that needs to be solved urgently. Summary of the invention
[0004] The present invention provides a cable wire break warning method, device, equipment and storage medium. The present invention realizes intelligent break warning, can take corresponding preventive measures according to the risk level, and improves the pertinence and effectiveness of the warning.
[0005] In a first aspect, the present invention provides a cable strand disconnection warning method, the cable strand disconnection warning method comprising: Multi-modal sensor data acquisition is performed on the cable stranding system to obtain vibration acceleration data and stress-strain data; Performing multi-band distributed analysis on the vibration acceleration data to obtain a global vibration state diagram, and performing time-frequency domain feature extraction on the stress-strain data to obtain a stress-strain feature diagram; Performing data fusion and adaptive feature selection on the global vibration state diagram and the stress-strain characteristic diagram to obtain a target fusion feature set; Based on the adaptive factor updating mechanism of covariance matching, an adaptive Kalman filter combined warning algorithm is constructed in combination with the dynamic window adjustment technology, and the cable state estimation is performed on the target fusion feature set to obtain the cable state estimation result; A multi-layer machine learning model analysis is performed on the cable status estimation result to obtain a disconnection risk prediction value, and a multi-level warning strategy decision is performed on the disconnection risk prediction value to output intelligent disconnection warning information.
[0006] In a second aspect, the present invention provides a cable strand disconnection warning device, the cable strand disconnection warning device comprising: The acquisition module is used to collect multi-modal sensor data of the cable stranding system to obtain vibration acceleration data and stress strain data; An extraction module, used to perform multi-band distributed analysis on the vibration acceleration data to obtain a global vibration state diagram, and to perform time-frequency domain feature extraction on the stress-strain data to obtain a stress-strain feature diagram; A fusion module, used for performing data fusion and adaptive feature selection on the global vibration state diagram and the stress-strain characteristic diagram to obtain a target fusion feature set; The state estimation module is used to construct an adaptive Kalman filter combined early warning algorithm based on an adaptive factor update mechanism of covariance matching and a dynamic window adjustment technology, and to perform cable state estimation on the target fusion feature set to obtain a cable state estimation result; The output module is used to perform a multi-layer machine learning model analysis on the cable status estimation result to obtain a disconnection risk prediction value, and to make a multi-level warning strategy decision on the disconnection risk prediction value to output intelligent disconnection warning information.
[0007] The third aspect of the present invention provides a computer device, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the computer device executes the above-mentioned cable wire break warning method.
[0008] A fourth aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned cable strand break warning method.
[0009] In the technical solution provided by the present invention, by collecting multimodal sensor data of the cable stranding system, obtaining vibration acceleration and stress-strain data, and performing data fusion and adaptive feature selection, the cable state is fully perceived and accurately described, and the reliability and accuracy of state monitoring are improved. The vibration acceleration data is subjected to multi-band distributed analysis to obtain a global vibration state diagram, which can effectively capture the vibration characteristics of the cable stranding in different frequency bands and improve the sensitivity and accuracy of abnormal detection. Based on the adaptive factor update mechanism of covariance matching and the dynamic window adjustment technology, an adaptive Kalman filter combined warning algorithm is constructed to improve the robustness and adaptability of state estimation and adapt to the dynamic characteristics of the cable stranding system. A multi-layer machine learning model is used to predict the risk of line breakage, including a multi-channel time convolutional network, an autoencoder, a hierarchical attention mechanism network, and a recursive neural network tree, etc., which fully mines the temporal characteristics and semantic information of the cable state data and improves the accuracy of risk prediction. A multi-level warning strategy is executed according to the predicted value of the line breakage risk, and an intelligent line breakage warning is realized. Corresponding preventive measures can be taken according to the degree of risk, which improves the pertinence and effectiveness of the warning. BRIEF DESCRIPTION OF THE DRAWINGS
[0010] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.
[0011] Figure 1 Schematic diagram of the steps of the cable strand break warning method according to an embodiment of the present invention; Figure 2 Schematic diagram of the structure of a cable wire break warning device in an embodiment of the present invention. DETAILED DESCRIPTION
[0012] Embodiments of the present invention provide a cable strand break warning method, device, equipment and storage medium. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe the order or precedence of the targets. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments described here can be implemented in an order other than that illustrated or described here. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.
[0013] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 , an embodiment of the cable strand disconnection warning method in the embodiment of the present invention includes: Step S1, performing multi-modal sensor data acquisition on the cable stranding system to obtain vibration acceleration data and stress strain data; It is understandable that the execution subject of the present invention may be a cable wire break warning device, or a terminal or a server, which is not limited here. The embodiment of the present invention is described by taking a server as the execution subject as an example.
[0014] Specifically, the cable stranding system is segmented to arrange a multimodal sensor array, establish a sensor network topology, and clarify the relative position relationship of each sensor. According to the sensor network topology, the sampling frequency of the sensor is dynamically adjusted through an adaptive algorithm to obtain the optimal sampling frequency parameters to ensure the efficiency and accuracy of data acquisition. Using the optimal sampling frequency parameters, multi-channel synchronous data acquisition is performed to capture the original sensor signal of the system. The signal is denoised using wavelet denoising technology, and the signal separation algorithm is applied to separate different signal components to obtain vibration acceleration signals and stress-strain signals respectively. The instantaneous frequency spectrum of the vibration acceleration signal is extracted, the frequency domain characteristics of the signal are analyzed, and the effective frequency band range is identified. These frequency bands are then filtered to obtain vibration acceleration data for analysis. At the same time, the stress-strain signal is subjected to principal component analysis to extract the main stress direction and strain direction, reduce the dimension of the data and retain the most important feature information. According to the principal stress direction and principal strain direction, the stress-strain signal is tensor decomposed to decompose the complex signal into multiple manageable tensor components. The decomposed stress-strain tensor is processed, and the characteristics of the material or structure under stress are characterized by extracting the stress-strain eigenvalues, and the stress-strain eigenvalues are obtained. The stress-strain eigenvalues are normalized to eliminate the influence of different dimensions and obtain stress-strain data.
[0015] Step S2, performing multi-band distributed analysis on the vibration acceleration data to obtain a global vibration state diagram, and performing time-frequency domain feature extraction on the stress-strain data to obtain a stress-strain feature diagram; Specifically, the vibration acceleration data is decomposed by wavelet packets, and the original vibration signal is decomposed into multiple frequency band sub-signals, which represent the vibration characteristics in different frequency ranges. By calculating the energy of the frequency band sub-signals, the energy distribution of each frequency band is obtained, and the frequency band energy spectrum is constructed. The frequency band energy spectrum shows the energy distribution of the vibration signal at different frequencies. The frequency band energy spectrum is processed by adaptive threshold segmentation to identify the frequency band where the energy is abnormally concentrated, and the abnormal frequency band identification is generated. The abnormal frequency band identification can indicate the frequency interval where there may be problems in the vibration signal. Based on the abnormal frequency band identification, each monitoring section of the cable strand is analyzed, the specific location of the vibration abnormality is located, and a local abnormal area map is obtained. The local abnormal area map is subjected to spatial correlation analysis to reveal the propagation path of the vibration abnormality and reflect the way and degree of the abnormality propagating from one area to other areas. Combining the local abnormal area map and the abnormal propagation path, a global vibration state map is constructed to show the vibration state of the entire cable strand system. At the same time, the stress-strain data is subjected to Hilbert-Huang transform to decompose the time-frequency characteristics of the nonlinear and non-stationary signals and obtain the time-frequency energy distribution map of the stress-strain data. According to the time-frequency energy distribution diagram, the instantaneous frequency and instantaneous amplitude are extracted to form a time-frequency feature vector, which reflects the change law of stress-strain signals in time and frequency. The time-frequency feature vector is subjected to feature dimensionality reduction, and the most representative feature information is retained to form a feature set after dimensionality reduction. According to the feature set after dimensionality reduction, a stress-strain feature map is constructed to show the characteristic changes of stress-strain in different regions and time.
[0016] Step S3, performing data fusion and adaptive feature selection on the global vibration state diagram and the stress-strain characteristic diagram to obtain a target fusion feature set; Specifically, the global vibration state diagram is feature segmented and divided into multiple local vibration regions, representing the vibration characteristics in different positions or time periods. The vibration intensity, frequency and duration of each local vibration region are extracted to form a vibration feature vector. At the same time, the stress-strain feature diagram is decomposed into a trend term, a period term and a residual term. The trend term reflects the long-term change trend of the stress-strain signal, the period term reveals the periodic fluctuations in the signal, and the residual term represents the part of the signal that cannot be explained by the trend and period. By analyzing these components, the change rate and accumulation of stress-strain are calculated to obtain the stress-strain feature vector. The vibration feature vector and the stress-strain feature vector are fused by Dempster-Shafer evidence theory. By combining evidence from different sources, uncertainty can be effectively handled to obtain an initial fused feature set. The initial fused feature set is subjected to feature importance analysis. By calculating the feature importance score of each feature, the contribution of each feature to the overall prediction is understood, and the features are ranked based on this. According to the ranking result of the importance score, threshold screening is performed to select the most critical candidate feature subset. The feature contribution in the candidate feature subset is calculated. The feature contribution measures the actual impact of each feature in the disconnection warning. According to the calculation results, the contribution of each feature is dynamically weighted to obtain the adaptive feature weight. The weight allocation is dynamically adjusted based on the performance of the feature under different conditions, so that the final feature combination can better adapt to various situations. According to the adaptive feature weight, the candidate feature subset is weighted and combined to obtain the target fusion feature set.
[0017] Step S4, based on the adaptive factor updating mechanism of covariance matching, an adaptive Kalman filter combined early warning algorithm is constructed in combination with the dynamic window adjustment technology, and the cable state estimation is performed on the target fusion feature set to obtain the cable state estimation result; Specifically, the state space model of the target fusion feature set is constructed to obtain the system state equation and observation equation, which jointly describe the evolution law of the cable system state over time and the relationship between the observation data and the system state. Through these two equations, the parameters of the Kalman filter are initialized to obtain the initial state estimate and covariance matrix. The observation data sequence is divided into sliding windows to generate a dynamic observation window. The dynamic observation window is used to capture the changing characteristics of the cable system state over time, and the adaptive observation noise covariance matrix is constructed by calculating the observation noise covariance within the dynamic observation window. This matrix can dynamically adjust the noise level in the filtering process, thereby improving the accuracy and adaptability of the filtering. The adaptive observation noise covariance matrix is subjected to singular value decomposition to obtain the covariance eigenvector, which reflects the key information of the noise covariance matrix, and the adaptive factor is calculated based on it. The calculation of the adaptive factor is an important part of the filtering process, which can adjust the covariance of the state prediction according to the real-time data, so that the filter can better adapt to the changing system state. The updated state prediction covariance is decomposed to extract the lower triangular matrix. The lower triangular matrix is used to generate an untraceable transformation sample set, representing multiple possibilities of the system state. By performing state prediction and measurement prediction on the unscented transformation sample set, the mean and covariance of the predicted state are obtained to describe the state distribution of the system at the next moment. The Kalman gain is calculated based on the predicted state mean and covariance. The Kalman gain is a dynamic adjustment coefficient that combines the observed data and the state prediction results to update the state estimation of the system. Through the Kalman gain, the new information in the observed data is incorporated into the state estimation, making the final state estimation more accurate, and finally obtaining the cable state estimation result.
[0018] Step S5: perform a multi-layer machine learning model analysis on the cable status estimation result to obtain a disconnection risk prediction value, perform a multi-level warning strategy decision on the disconnection risk prediction value, and output intelligent disconnection warning information.
[0019] Specifically, the cable state estimation result is subjected to empirical mode decomposition. Empirical mode decomposition is an analysis method suitable for nonlinear and non-stationary signals. By this method, the complex cable state estimation result is decomposed into multiple intrinsic mode functions, which respectively characterize the characteristic information of different frequency components. The intrinsic mode function is input into the multi-channel time convolution network in the multi-layer machine learning model for processing, and the time series feature representation of the data is extracted. The time convolution network has the ability to capture time series patterns, can analyze data at multiple time scales, and obtain rich time series feature information. The time series feature representation is input into the autoencoder in the multi-layer machine learning model for noise reduction. The autoencoder maps the data to a low-dimensional space and then reconstructs the data from the low-dimensional space to achieve effective filtering of noise and obtain a robust feature vector. The robust feature vector is input into the hierarchical attention mechanism network in the multi-layer machine learning model for context weighted analysis. The hierarchical attention mechanism network can weight the features by calculating the importance of different features in different contexts to obtain weighted context information. This information is processed by dynamic time warping to adjust the feature alignment between different time periods to obtain the aligned time series pattern. This processing can effectively reduce the error caused by different time scales and ensure the consistency of features in the time dimension. The aligned time series pattern is input into the recursive neural network tree in the multi-layer machine learning model for processing. The recursive neural network tree can extract the semantic information of the data at multiple levels and obtain hierarchical semantic features. By performing multi-task learning on these features, shared representation and task target representation are obtained at the same time. The shared representation contains feature information that is meaningful to all tasks, while the task target representation focuses on the feature information of a specific task. The shared representation and task target representation are input into the integrated gradient boosting tree in the multi-layer machine learning model for disconnection risk prediction. The gradient boosting tree can build a strong classifier based on multiple weak classifiers to obtain a more accurate disconnection risk prediction value. The disconnection risk prediction value is integrated and fused, and a more robust comprehensive risk assessment result is obtained by weighted combination of the prediction results of different models. According to the assessment result, a multi-level warning strategy is executed, and finally the intelligent disconnection warning information is output.
[0020] The aligned time series patterns are segmented by sliding windows, and the entire time series data is divided into multiple continuous time segments, capturing the feature information of different time intervals in the time series data. The time segments are sequentially input into the leaf nodes of the recursive neural network tree for processing to obtain the initial feature representation, which covers the basic information in the time series segments. The initial feature representation is subjected to local time series dependency analysis, and the pattern features that change in a short time are captured by identifying the local dependencies between time segments. The local time series dependency features reveal the dynamic changes of the data at the micro level, and these features are input into the middle nodes of the recursive neural network tree for processing to form a hierarchical feature combination. The hierarchical feature combination gradually builds a more complex feature representation by fusing and refining features at different levels, reflecting the multi-level characteristics of the data. The hierarchical feature combination is subjected to long-term dependency analysis to capture the dependency of the data in a longer time range, and obtain the long-term dependency features, which reflect the state change trend of the cable stranding system in a longer time. The long-term dependency features are input into the root node of the recursive neural network tree for processing to obtain the global semantic features. The global semantic features integrate all hierarchical information from the leaf nodes to the root nodes, providing an overall understanding of the time series data. Perform attention mechanism analysis on global semantic features, identify the most important parts of global semantic features, and perform weighted processing on these parts to generate weighted semantic features. Input weighted semantic features into the multi-task learning network to construct a task sharing layer. The task sharing layer is the core part of multi-task learning, which provides a shared feature basis for different tasks and ensures that each task can be learned and predicted in a unified feature space. Perform task branch network analysis on the task sharing layer to process the features of specific tasks, and obtain the task target representation by processing the information of the sharing layer. The task target representation focuses on the prediction requirements of each specific task and can generate the feature representation that best suits the task. The task sharing layer and the task target representation are feature fused to obtain the final shared representation and task target representation. These two representations contain the shared information in multi-task learning and task-specific information, respectively.
[0021] In the embodiment of the present invention, by collecting multimodal sensor data of the cable stranding system, obtaining vibration acceleration and stress-strain data, and performing data fusion and adaptive feature selection, the cable state is fully perceived and accurately described, and the reliability and accuracy of state monitoring are improved. The vibration acceleration data is subjected to multi-band distributed analysis to obtain a global vibration state diagram, which can effectively capture the vibration characteristics of the cable stranding in different frequency bands and improve the sensitivity and accuracy of abnormal detection. Based on the adaptive factor update mechanism of covariance matching and the dynamic window adjustment technology, an adaptive Kalman filter combined warning algorithm is constructed to improve the robustness and adaptability of state estimation and adapt to the dynamic characteristics of the cable stranding system. A multi-layer machine learning model is used to predict the risk of line breakage, including a multi-channel time convolutional network, an autoencoder, a hierarchical attention mechanism network, and a recursive neural network tree, etc., which fully mines the temporal characteristics and semantic information of the cable state data and improves the accuracy of risk prediction. A multi-level warning strategy is executed according to the predicted value of the line breakage risk, and an intelligent line breakage warning is realized. Corresponding preventive measures can be taken according to the degree of risk, which improves the pertinence and effectiveness of the warning.
[0022] In a specific embodiment, the process of executing step S1 may specifically include the following steps: Arrange a multi-modal sensor array in segments of the cable strand system to obtain a sensor network topology, and adaptively adjust the sensor sampling frequency according to the sensor network topology to obtain the optimal sampling frequency parameters; Perform multi-channel synchronous data acquisition on the optimal sampling frequency parameters to obtain the original sensor signal, and perform wavelet denoising and signal separation on the original sensor signal to obtain the vibration acceleration signal and stress strain signal; Extracting the instantaneous frequency spectrum of the vibration acceleration signal, and filtering the vibration acceleration signal according to the instantaneous frequency spectrum to obtain vibration acceleration data; Performing principal component analysis on the stress-strain signal to obtain the principal stress direction and the principal strain direction, and performing tensor decomposition on the stress-strain signal according to the principal stress direction and the principal strain direction to obtain the stress-strain tensor; The stress-strain tensor is decomposed to obtain the stress-strain eigenvalues, and the stress-strain eigenvalues are normalized to obtain the stress-strain data.
[0023] Specifically, a multimodal sensor array is arranged in segments on the cable strand system, and a sensor network topology is constructed through these sensor arrays. In order to fully cover all possible stress and vibration parts of the cable strand, the sensors are usually arranged in segments along the length of the cable. Multimodal sensors may include accelerometers and strain sensors, etc., which can respectively collect vibration and stress and strain data of the cable. The sensor network topology refers to the relative layout of these sensors in space and their interconnection relationship. This structure determines the path and method of data transmission. The sampling frequency of the sensor is adaptively adjusted according to the sensor network topology to obtain the optimal sampling frequency parameters. The sampling frequency is an important parameter in the signal acquisition process. Too high a sampling frequency will lead to data redundancy and increase the computational burden, while too low a sampling frequency may miss important signal features. The method of adaptively adjusting the sampling frequency can determine the main frequency components of the signal by analyzing the initial data collected by the sensor and using frequency domain analysis techniques, such as Fourier transform, and then select an optimal sampling frequency that can cover these main frequencies. Assume is the sampling frequency, is the maximum frequency of the signal. According to the Nyquist sampling theorem, the sampling frequency needs to satisfy In practical applications, in order to ensure signal integrity, the sampling frequency is usually set to Or higher. After determining the optimal sampling frequency parameters, perform multi-channel synchronous data acquisition on the cable stranding system. Process the original signal. Use wavelet denoising technology to denoise the signal. Wavelet denoising is a signal processing method with localized characteristics in both time domain and frequency domain. It can effectively remove noise from the signal without losing important signal components. Then perform signal separation on the denoised signal, and extract the vibration acceleration signal and stress strain signal respectively. Extract the instantaneous frequency spectrum of the vibration acceleration signal. The instantaneous frequency spectrum can describe the frequency changes of the signal at different times, reflecting the dynamic characteristics of the vibration signal. Assume that the vibration signal The instantaneous frequency is , then the instantaneous frequency can be obtained by Hilbert transform. Specifically, the signal The Hilbert transform is defined as ; in, yes The Hilbert transform of Indicates the principal value integral. Instantaneous frequency Then we can solve the envelope signal of the signal The derivative of is obtained. After the instantaneous frequency spectrum is calculated, the vibration acceleration signal is filtered according to the result of the frequency spectrum to remove unnecessary frequency components and obtain the vibration acceleration data for analysis. The stress-strain signal is subjected to principal component analysis. Principal component analysis is a commonly used dimensionality reduction method. It maps the data from the original high-dimensional space to the low-dimensional space by performing a linear transformation on the data, mainly retaining the direction with the largest variance in the data. Assume that the data matrix of the stress-strain signal is , where each row represents a sample and each column represents a variable. The core of principal component analysis is to solve the covariance matrix The eigenvalues and eigenvectors of are: ; in is the data matrix The transpose of is the number of samples. By performing eigenvalue decomposition on the covariance matrix, the principal stress directions and principal strain directions are found, representing the components with the largest changes in the stress-strain data. According to the principal component directions, the stress-strain signal is tensor decomposed to obtain the stress-strain tensor. Tensor decomposition is a high-order data analysis method that can decompose high-dimensional data into multiple low-dimensional tensor components, thereby simplifying the data structure and extracting the most important characteristic information. The obtained stress-strain tensor is decomposed to extract the stress-strain eigenvalues. The stress-strain eigenvalues represent the stress-strain intensity in different directions and are important indicators for evaluating the state of the material. The stress-strain eigenvalues are normalized and mapped to a unified scale. They are usually scaled to the range of [0, 1] to eliminate the influence between different dimensions and obtain stress-strain data.
[0024] In a specific embodiment, the process of executing step S2 may specifically include the following steps: Decomposing the vibration acceleration data by wavelet packet to obtain a plurality of frequency band sub-signals, and calculating the energy distribution of each frequency band according to the plurality of frequency band sub-signals to obtain a frequency band energy spectrum; Adaptively segment the frequency band energy spectrum using thresholds to obtain abnormal frequency band identifiers, and locate the abnormalities in each monitoring section based on the abnormal frequency band identifiers to obtain a local abnormal area map; Perform spatial correlation analysis on the local abnormal area map to obtain the abnormal propagation path, and construct a global vibration state map based on the abnormal propagation path and the local abnormal area map; Perform Hilbert-Huang transform on stress-strain data to obtain time-frequency energy distribution diagram, and extract instantaneous frequency and instantaneous amplitude according to the time-frequency energy distribution diagram to obtain time-frequency feature vector; The time-frequency feature vector is subjected to feature dimensionality reduction to obtain a feature set after dimensionality reduction, and a stress-strain feature map is constructed based on the feature set after dimensionality reduction.
[0025] Specifically, the vibration acceleration data is decomposed by wavelet packet. Wavelet packet decomposition is an effective signal processing method that can decompose the signal into sub-signals of different frequency bands, thereby analyzing the characteristics of the signal in each frequency band. For the vibration acceleration data, multiple frequency band sub-signals are obtained by wavelet packet decomposition. Assume that the original vibration signal is , the process of wavelet packet decomposition will be Decompose into sub-signals of different scales and frequencies ,in represents the number of decomposition levels, Indicates the number of the frequency band. The energy distribution of each frequency band is calculated based on multiple frequency band sub-signals to construct a frequency band energy spectrum. The frequency band energy spectrum is a graph that describes the distribution of signal energy in different frequency segments. For each frequency band sub-signal , its energy can be calculated by the following formula: ; in, Indicates Tier The energy of the frequency band, is the signal strength of the frequency band. By calculating the energy distribution of all frequency bands, the frequency band energy spectrum of the entire signal is obtained, which reflects the energy distribution characteristics of the signal at different frequencies. The frequency band energy spectrum is segmented by adaptive threshold to identify the frequency bands that may have abnormalities. The adaptive threshold segmentation method determines one or more thresholds based on the statistical characteristics of the frequency band energy spectrum to distinguish normal frequency bands from abnormal frequency bands. For example, by setting a threshold based on the standard deviation, when the energy of a frequency band exceeds the threshold, it is identified as an abnormal frequency band. According to the abnormal frequency band identification information, the abnormal location of each monitoring section is performed, and the identified abnormal frequency band is matched with the actual monitoring position. A local abnormal area map is drawn to show the areas where vibration abnormalities are concentrated in the cable stranding system. These areas may be structural weak points or parts impacted by external forces. The local abnormal area map is subjected to spatial correlation analysis to reveal the propagation path of vibration abnormalities in space. The spatial correlation analysis determines how abnormal vibration propagates between different areas by calculating the correlation between different areas. According to the abnormal propagation path and the local abnormal area map, a global vibration state map is constructed to describe the vibration state of the cable stranding system at different times and spaces. Perform Hilbert-Huang transform on stress-strain data. Hilbert-Huang transform is a time-frequency analysis method suitable for nonlinear and non-stationary signals. Through Hilbert-Huang transform, stress-strain signals are decomposed into a series of intrinsic mode functions (IMFs), and the instantaneous frequency and instantaneous amplitude of the signals are extracted from them. Assume that the stress-strain signal is , the instantaneous frequency at each moment is obtained through the Hilbert-Huang transform and the instantaneous amplitude . Time-frequency energy distribution diagram It can be expressed by the following formula: ; in, is the Dirac function, is the instantaneous frequency, is the instantaneous amplitude. The time-frequency energy distribution diagram shows the distribution of signal energy in time and frequency. According to the time-frequency energy distribution diagram, the instantaneous frequency and instantaneous amplitude are extracted to obtain the time-frequency feature vector. The time-frequency feature vector includes the frequency and amplitude information of the signal at different times, reflecting the dynamic change characteristics of the stress-strain signal. The time-frequency feature vector is subjected to feature dimensionality reduction processing, and the dimension of the feature vector is reduced while retaining the most important information, thereby improving the computational efficiency. According to the reduced feature set, a stress-strain feature diagram is constructed to show the characteristic distribution of stress-strain signals in different time and space, and to reveal the change law of materials under different stress states. For example, the signal collected by a certain cable stranding system through distributed stress sensors shows that the instantaneous frequency of the signal changes significantly in certain time periods, indicating possible structural stress concentration. Through the Hilbert-Huang transform and feature dimensionality reduction processing, the constructed stress-strain feature diagram clearly shows the stress concentration area and time point, thereby indicating that there is a potential risk of disconnection in this area.
[0026] In a specific embodiment, the process of executing step S3 may specifically include the following steps: Performing feature segmentation on the global vibration state diagram to obtain multiple local vibration regions, and extracting vibration intensity, frequency and duration features based on the multiple local vibration regions to obtain a vibration feature vector; The stress-strain characteristic diagram is decomposed into a time series to obtain trend terms, periodic terms and residual terms, and the stress-strain change rate and cumulative amount are calculated based on the trend terms, periodic terms and residual terms to obtain the stress-strain characteristic vector; The vibration eigenvector and stress-strain eigenvector are fused by Dempster-Shafer evidence theory to obtain the initial fusion feature set. Perform feature importance analysis on the initial fusion feature set to obtain feature importance scores, and sort and threshold screen the feature importance scores to obtain candidate feature subsets; The feature contribution is calculated according to the candidate feature subsets, and the feature contribution is dynamically weighted to obtain the adaptive feature weights. The candidate feature subsets are weighted combined according to the adaptive feature weights to obtain the target fusion feature set.
[0027] Specifically, feature segmentation is performed on the global vibration state diagram, and the global vibration state diagram is divided into multiple local vibration regions, which may represent different vibration modes or abnormal regions in the system. Each segmented local region corresponds to a period of time and a specific spatial position. Vibration intensity, frequency and duration features are extracted based on multiple local vibration regions. Vibration intensity is usually measured by the amplitude or energy of the vibration signal, frequency represents the periodic change of the vibration signal, and duration reflects the time span of a specific vibration event. By extracting these features, a vibration feature vector is formed. Assume that the vibration signal of a local vibration region is , its vibration intensity It can be defined as the signal energy: ; in, and is the start and end time of the vibration event. Frequency The duration can be obtained by spectrum analysis of the signal. for . Vibration eigenvector The characteristics of the vibration event can be fully described. The stress-strain characteristic diagram is decomposed into time series to extract trend terms, periodic terms and residual terms. The stress-strain characteristic diagram shows the stress and strain changes of each part of the cable system at different times. The purpose of time series decomposition is to decompose the original signal into three parts: trend terms , Periodic Item and the residual The trend term reflects the long-term change trend of the signal, the periodic term reveals the periodic fluctuations in the signal, and the residual term represents the irregular fluctuations in the signal. The stress-strain change rate and cumulative amount are calculated based on the trend term, periodic term, and residual term. Stress-strain change rate It can be expressed as: ; in, is the stress or strain signal. The cumulative amount is the cumulative effect of the signal over time, which can usually be calculated by integration. Through calculation, the stress-strain characteristic vector is formed ,in is the cumulative stress-strain variable. The vibration eigenvector and stress-strain eigenvector are fused by Dempster-Shafer evidence theory to obtain the initial fusion feature set. Dempster-Shafer theory is a method for dealing with uncertainty and multi-source information fusion. For each eigenvector, a trust function is defined , which represents the confidence that a feature belongs to a specific state. By applying the Dempster-Shafer rule to the vibration eigenvector and the stress-strain eigenvector, these confidences are combined to generate the initial fusion feature set. Assume that the two eigenvectors and The trust functions are and , then the fused trust function It can be expressed as: ; in, is a subset of the feature space, and is the support set of the feature vector. Perform feature importance analysis on the initial fusion feature set. By calculating the contribution of each feature to the final prediction result, an importance score is assigned to each feature. The score calculation can be achieved by a variety of methods, such as the feature importance measurement method based on the gradient boosting decision tree. The obtained feature importance score reflects the contribution of each feature in the disconnection warning. According to the feature importance score, the features are sorted and thresholds are set to filter out the candidate feature subsets. Based on the candidate feature subsets, the feature contribution of each feature is calculated. The feature contribution can be evaluated by the performance of the feature under different conditions, taking into account the performance of the feature in various samples. According to the feature contribution, each feature is dynamically weighted to obtain an adaptive feature weight. The adaptive feature weight enables the influence of each feature to be dynamically adjusted according to actual needs at different times and under different circumstances. The candidate feature subsets are weighted combined according to the adaptive feature weights to obtain the target fusion feature set. The target fusion feature set is the result of weighted combination of multiple features, representing the overall state of the cable stranding system, and integrating the key information of vibration features and stress-strain features.
[0028] In a specific embodiment, the process of executing step S4 may specifically include the following steps: Construct a state space model for the target fusion feature set to obtain the system state equation and observation equation, and initialize the Kalman filter parameters according to the system state equation and observation equation to obtain the initial state estimate and covariance matrix; The observation data sequence is divided into sliding windows to obtain a dynamic observation window, and the observation noise covariance is calculated according to the dynamic observation window to obtain an adaptive observation noise covariance matrix; Perform singular value decomposition on the adaptive observation noise covariance matrix to obtain the covariance eigenvector, and calculate the adaptive factor based on the covariance eigenvector to obtain the updated state prediction covariance; Decompose the updated state prediction covariance to obtain a lower triangular matrix, and generate an untraceable transformation sample set based on the lower triangular matrix; The state prediction and measurement prediction are performed on the unscented transformation sample set to obtain the predicted state mean and covariance, and the Kalman gain is calculated based on the predicted state mean and covariance. The state estimation is updated in combination with the observed data to obtain the cable state estimation result.
[0029] Specifically, a state space model is constructed for the target fusion feature set to describe the dynamic behavior of the system. The state space model consists of the system state equation and the observation equation. The former is used to describe the evolution of the system state over time, and the latter is used to associate the system state with the observation data. Assume that the target fusion feature set is , the state variables of the system are , then the state equation can be expressed as: ; in, is the state transfer matrix, which describes the transfer relationship between states; is the control input matrix, is the control input vector; is process noise, which is usually assumed to have a mean of zero and a covariance of Gaussian white noise. The observation equation can be expressed as: ; in, is the observation matrix, which maps the state variables to the observation space; is the observation noise, assuming that its mean is zero and its covariance is Gaussian white noise. Initialize the Kalman filter parameters according to the system state equation and observation equation. Initial state estimation Usually obtained by averaging historical data or directly measuring, the initial covariance matrix It reflects the uncertainty of the initial state estimate and can usually be set to a small positive definite matrix. The observation data sequence is divided into sliding windows so that dynamic changes in time can be captured in data processing. The sliding window technology can analyze the local characteristics of the data by moving a fixed-size window on the time series. The data in each window is called a dynamic observation window. By processing these data, the observation noise covariance matrix is calculated. , which reflects the uncertainty in the observed data. Its calculation method is usually to perform statistical analysis on the residuals of the observed data. Assuming that the residual is , then the covariance matrix can be expressed as: ; in, is the mean of the residuals, is the number of data points in the window. After calculating the adaptive observation noise covariance matrix, perform singular value decomposition on it to obtain the covariance eigenvector. Singular value decomposition can decompose the covariance matrix into eigenvectors and eigenvalues. These eigenvectors represent the main change direction in the observation data. Assume that the covariance matrix The singular value decomposition of is: ; in, is the eigenvector matrix, is a diagonal matrix containing the singular values of the covariance matrix. Based on these eigenvectors, the adaptive factors are calculated , this factor is used to update the covariance matrix of state prediction, the formula is: ; Updated state prediction covariance It is adjusted by multiplying the adaptive factor to better adapt to the characteristics of the current observation data. The updated state prediction covariance is decomposed to extract the lower triangular matrix , through Cholesky decomposition, the covariance matrix It can be expressed as: ; in, is a lower triangular matrix. Using this matrix, unscented transformation sample sets are generated. These sample sets capture various possibilities of system states by introducing nonlinear transformations. The generation of sample sets usually adopts the method of unscented Kalman filtering, which reflects the state distribution through sampling points in the center of the distribution. State prediction and measurement prediction are performed on the unscented transformation sample set. State prediction is performed based on the state equation of the system, and measurement prediction is completed through the observation equation. These predictions will generate the predicted state mean respectively and the predicted covariance . Use these predictions to calculate the Kalman gain , Kalman gain is a balancing coefficient used to combine observation data and prediction results to update the state estimate. The calculation formula is: ; Using the Kalman gain, update the state estimate and the covariance matrix ; ; in, is the identity matrix, The actual observation data is combined with the updated state estimation to obtain the estimated result of the cable state, which reflects the current operating state of the system, and ensures the accuracy of the state estimation through the above dynamic adjustment and prediction process.
[0030] In a specific embodiment, the process of executing step S5 may specifically include the following steps: Performing empirical mode decomposition on the cable state estimation results to obtain multiple intrinsic mode functions, and inputting the multiple intrinsic mode functions into a multi-channel time convolutional network in a multi-layer machine learning model to obtain a time series feature representation; The time series feature representation is input into the autoencoder in the multi-layer machine learning model for noise reduction to obtain a robust feature vector; The robust feature vector is input into the hierarchical attention mechanism network in the multi-layer machine learning model for context weighted analysis to obtain weighted context information, and the weighted context information is dynamically time-warped to obtain the aligned time series pattern; The aligned temporal patterns are input into a recursive neural network tree in a multi-layer machine learning model for processing to obtain hierarchical semantic features, and the hierarchical semantic features are processed by multi-task learning to obtain shared representations and task target representations; The shared representation and the task target representation are input into the integrated gradient boosting tree in the multi-layer machine learning model to predict the disconnection risk, and the disconnection risk prediction value is obtained; The predicted values of line disconnection risk are integrated and fused to obtain comprehensive risk assessment results. Based on the comprehensive risk assessment results, a multi-level warning strategy is implemented to output intelligent line disconnection warning information.
[0031] Specifically, the cable state estimation result is subjected to empirical mode decomposition to extract different frequency components in the signal. Empirical mode decomposition is an adaptive signal processing method that decomposes complex nonlinear and non-stationary signals into several intrinsic mode functions (IMFs) by layer-by-layer decomposition. IMF represents the fluctuation characteristics of the signal on different time scales, and each IMF captures the signal components within a specific frequency range. Assume that the cable state estimation result is a signal , after empirical mode decomposition, can be expressed as: ; in, For the The intrinsic mode functions, is the mode number, is the residual term. The intrinsic mode function is input into the multi-channel time convolution network in the multi-layer machine learning model for processing. The multi-channel time convolution network is a deep learning model suitable for processing time series data, which can extract the time series features of the signal at different time scales. By performing convolution operations on IMF in parallel on multiple channels, the multi-channel time convolution network can capture the complex time dependency of the signal and obtain the time series feature representation, which reflects the changing law of the signal at different times. The time series feature representation is input into the autoencoder in the multi-layer machine learning model for denoising. The autoencoder is a neural network used to learn compact representation of data. It learns the main features of the input data and removes noise by encoding the input data into a low-dimensional space and then decoding it from the low-dimensional space to the original data space. In this process, the time series feature representation is compressed and reconstructed by the autoencoder, and the obtained robust feature vector not only retains the main features of the signal, but also reduces the interference of noise and redundant information. The robust feature vector is input into the hierarchical attention mechanism network in the multi-layer machine learning model for context weighted analysis. The hierarchical attention mechanism network performs weighted processing on important features by calculating the importance of features in different contexts. Through the attention mechanism, the model can automatically focus on those time segments or features that contribute more to the prediction and obtain weighted contextual information. These weighted feature information is then input into the dynamic time warping module for processing. Dynamic time warping is a method used to align time series data. It can adjust the inconsistency of feature vectors in the time dimension, ensure that data from different time periods can be compared and analyzed on the same time scale, and obtain the aligned time series pattern. The aligned time series pattern is input into the recursive neural network tree in the multi-layer machine learning model for processing. The recursive neural network tree is a model that combines the advantages of recursive neural networks and decision trees. It can capture the long-term dependencies of signals and extract hierarchical semantic features of data through the hierarchical structure of the tree. When processing time series data, the recursive neural network can retain the information connection between the previous and next time points, while the tree structure helps to extract higher-level semantic features from the data. After being processed by the recursive neural network tree, the hierarchical semantic features obtained by the model represent the semantic information of the signal at different times and frequencies. The hierarchical semantic features are input into the multi-task learning network for processing. The multi-task learning network processes multiple related tasks simultaneously through shared representation and task branching. In this network, hierarchical semantic features are first used to construct a task sharing layer, which contains common features that are meaningful to all tasks. The shared representation is assigned to the task branch network to extract the target representation of a specific task. The shared representation captures the common characteristics of the cable status, while the task target representation is optimized for the specific disconnection risk prediction task. The shared representation and the task target representation are input into the integrated gradient boosting tree in the multi-layer machine learning model for disconnection risk prediction.The ensemble gradient boosting tree is an ensemble learning method that combines multiple decision tree models to obtain more accurate prediction results. In the ensemble gradient boosting tree model, each tree modifies the prediction results of the previous tree. The final prediction value of the line break risk combines the prediction information of all models and is an accurate estimate of the line break risk of the cable system. The line break risk prediction value is integrated and fused, and multiple prediction results are weighted averaged. The stability and accuracy of the prediction are improved by adjusting the weights. By integrating the prediction results of different time points and different models, a comprehensive risk assessment result is obtained.
[0032] In a specific embodiment, the execution step inputs the aligned time series pattern into the recursive neural network tree in the multi-layer machine learning model for processing to obtain hierarchical semantic features, and performs multi-task learning processing on the hierarchical semantic features to obtain the shared representation and the task target representation. The process may specifically include the following steps: Perform sliding window segmentation on the aligned time series pattern to obtain multiple time segments, and input the multiple time segments into the leaf nodes of the recursive neural network tree for processing to obtain the initial feature representation; Perform local temporal dependency analysis on the initial feature representation to obtain local temporal dependency features, and input the local temporal dependency features into the intermediate nodes of the recursive neural network tree for processing to obtain a hierarchical feature combination; Perform long-term dependency analysis on hierarchical feature combinations to obtain long-term dependency features, and construct the root nodes of a recursive neural network tree with the long-term dependency features for processing to obtain global semantic features; Perform attention mechanism analysis on global semantic features to obtain weighted semantic features, and construct a multi-task learning network with the weighted semantic features to obtain a task sharing layer; The task branch network analysis is performed on the task sharing layer to obtain the task target representation, and feature fusion is performed based on the task sharing layer and the task target representation to obtain the shared representation and the task target representation.
[0033] Specifically, the aligned time series pattern is segmented by sliding windows to obtain multiple time segments. Sliding window segmentation technology is a time series data processing method that can capture the local characteristics of the signal in different time periods by moving a fixed-size window on the time axis. Assume that the aligned time series pattern is ,in Represents time, split by sliding window to get a series of time segments ,in is the index of the window, is the time point in the window. The input is processed into the leaf nodes of the recursive neural network tree to obtain the initial feature representation. The recursive neural network tree combines the advantages of recursive neural networks and decision trees. At the leaf nodes, the recursive neural network can capture the short-term dependencies within the time segment. Through the loop structure of the recursive neural network, the leaf nodes can learn the change pattern of the signal in a short period of time and obtain the initial feature representation of each time segment. , these initial feature representations reflect the basic temporal characteristics of the time segment. Perform local temporal dependency analysis to obtain more detailed feature information. Local temporal dependency analysis focuses on the relationship between features in a short period of time and can reveal the dynamic change pattern of the signal in a small-scale time window. Through the intermediate nodes of the recursive neural network tree, the local temporal dependency features are further processed and combined to form a hierarchical feature combination. . For hierarchical feature combinations The memory mechanism of the neural network for long-term dependency analysis can retain the dependency relationship of the signal over a longer time span and capture a wider range of time features. The long-term dependency features obtained by processing the root node of the recursive neural network tree It is a summary of the entire time series pattern, reflecting the behavior and pattern of the signal on a large time scale. Input to the attention mechanism for analysis. The attention mechanism can highlight those features that are more critical to the overall prediction or classification task by giving different weights to different time segments or features. is the feature corresponding to The attention weights, the attention mechanism will calculate the weighted semantic features ; in, It is the weight value calculated by the attention mechanism, reflecting the importance of each long-term dependency feature in the final semantic representation. , the model can better focus on those key time series features and improve the accuracy and robustness of prediction. The input is processed in a multi-task learning network. The multi-task learning network can solve multiple related prediction tasks at the same time by combining shared features and task-specific features. In the multi-task learning network, the weighted semantic features are used to Building a task sharing layer , which contains common feature representations that are meaningful to all tasks. The construction of the shared layer can be achieved through a set of shared neural network layers, which are responsible for extracting the most critical shared features from the weighted semantic features of the input. Perform task branch network analysis. Each task branch network is responsible for extracting features related to a specific task from the shared layer to form a task target representation. Assumptions It is with A task-related target representation can be obtained by extracting these features from the shared layer through a set of specific neural network layers: ; in, Yes and task Related feature extraction functions. In this way, different tasks can share some features while also retaining their own task-specific feature representations. and task objectives Perform feature fusion to obtain the final shared representation and task objectives Feature fusion can be achieved through simple weighting or concatenation operations. Assume that the feature fusion operation is , then the shared representation and task target representation can be expressed as: ; Through the fusion operation, the final model can simultaneously consider shared global features and task-specific local features, providing more accurate prediction or classification results for each task.
[0034] The above describes the cable strand disconnection warning method in the embodiment of the present invention. The following describes the cable strand disconnection warning device in the embodiment of the present invention. Figure 2 , an embodiment of the cable strand disconnection warning device in the embodiment of the present invention includes: The acquisition module is used to collect multi-modal sensor data of the cable stranding system to obtain vibration acceleration data and stress strain data; An extraction module is used to perform multi-band distributed analysis on vibration acceleration data to obtain a global vibration state diagram, and to perform time-frequency domain feature extraction on stress-strain data to obtain a stress-strain feature diagram; A fusion module is used to perform data fusion and adaptive feature selection on the global vibration state diagram and the stress-strain characteristic diagram to obtain a target fusion feature set; The state estimation module is used to construct an adaptive Kalman filter combined warning algorithm based on the adaptive factor update mechanism of covariance matching and dynamic window adjustment technology, and to estimate the cable state of the target fusion feature set to obtain the cable state estimation result; The output module is used to perform multi-layer machine learning model analysis on the cable status estimation results to obtain the disconnection risk prediction value, make multi-level warning strategy decisions on the disconnection risk prediction value, and output intelligent disconnection warning information.
[0035] Through the cooperation of the above components, the cable stranding system is subjected to multimodal sensor data collection, vibration acceleration and stress-strain data are obtained, and data fusion and adaptive feature selection are performed to achieve comprehensive perception and accurate description of the cable state, thereby improving the reliability and accuracy of state monitoring. The vibration acceleration data is subjected to multi-band distributed analysis to obtain a global vibration state diagram, which can effectively capture the vibration characteristics of the cable stranding in different frequency bands and improve the sensitivity and accuracy of anomaly detection. Based on the adaptive factor update mechanism of covariance matching and the dynamic window adjustment technology, an adaptive Kalman filter combined warning algorithm is constructed to improve the robustness and adaptability of state estimation and adapt to the dynamic characteristics of the cable stranding system. A multi-layer machine learning model is used for line break risk prediction, including multi-channel time convolutional network, autoencoder, hierarchical attention mechanism network and recursive neural network tree, etc., which fully mines the temporal characteristics and semantic information of the cable state data and improves the accuracy of risk prediction. According to the predicted value of line break risk, a multi-level warning strategy is implemented to realize intelligent line break warning, and corresponding preventive measures can be taken according to the degree of risk, thereby improving the pertinence and effectiveness of the warning.
[0036] The present invention also provides a computer device, which includes a memory and a processor. The memory stores computer-readable instructions. When the computer-readable instructions are executed by the processor, the processor executes the steps of the cable wire break warning method in the above-mentioned embodiments.
[0037] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the cable wire break warning method.
[0038] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.
[0039] If the integrated unit is implemented in the form of 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 the present invention is essentially or the part that contributes to the prior art or the whole or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.
[0040] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A cable strand break warning method, characterized in that: The cable strand disconnection warning method comprises: Multi-modal sensor data acquisition is performed on the cable stranding system to obtain vibration acceleration data and stress-strain data; Performing multi-band distributed analysis on the vibration acceleration data to obtain a global vibration state diagram, and performing time-frequency domain feature extraction on the stress-strain data to obtain a stress-strain feature diagram; Performing data fusion and adaptive feature selection on the global vibration state diagram and the stress-strain characteristic diagram to obtain a target fusion feature set; Based on the adaptive factor updating mechanism of covariance matching, an adaptive Kalman filter combined warning algorithm is constructed in combination with the dynamic window adjustment technology, and the cable state estimation is performed on the target fusion feature set to obtain the cable state estimation result; A multi-layer machine learning model analysis is performed on the cable status estimation result to obtain a disconnection risk prediction value, and a multi-level warning strategy decision is performed on the disconnection risk prediction value to output intelligent disconnection warning information.
2. The cable strand break warning method according to claim 1, characterized in that: The multi-modal sensor data acquisition is performed on the cable stranding system to obtain vibration acceleration data and stress strain data, including: Arranging a multi-modal sensor array in sections of the cable strand system to obtain a sensor network topology, and adaptively adjusting the sensor sampling frequency according to the sensor network topology to obtain an optimal sampling frequency parameter; Performing multi-channel synchronous data acquisition on the optimal sampling frequency parameters to obtain original sensor signals, and performing wavelet denoising and signal separation on the original sensor signals to obtain vibration acceleration signals and stress strain signals; Extracting the instantaneous frequency spectrum of the vibration acceleration signal, and filtering the vibration acceleration signal according to the instantaneous frequency spectrum to obtain vibration acceleration data; Performing principal component analysis on the stress-strain signal to obtain a principal stress direction and a principal strain direction, and performing tensor decomposition on the stress-strain signal according to the principal stress direction and the principal strain direction to obtain a stress-strain tensor; The stress-strain tensor is decomposed to obtain stress-strain eigenvalues, and the stress-strain eigenvalues are normalized to obtain stress-strain data.
3. The cable strand break warning method according to claim 1, characterized in that: The multi-band distributed analysis of the vibration acceleration data is performed to obtain a global vibration state diagram, and the time-frequency domain feature extraction of the stress-strain data is performed to obtain a stress-strain feature diagram, including: Performing wavelet packet decomposition on the vibration acceleration data to obtain a plurality of frequency band sub-signals, and calculating the energy distribution of each frequency band according to the plurality of frequency band sub-signals to obtain a frequency band energy spectrum; Adaptively segment the frequency band energy spectrum using a threshold value to obtain abnormal frequency band identifiers, and perform abnormal location on each monitoring section according to the abnormal frequency band identifiers to obtain a local abnormal area map; Performing spatial correlation analysis on the local abnormal region map to obtain an abnormal propagation path, and constructing a global vibration state map according to the abnormal propagation path and the local abnormal region map; Performing Hilbert-Huang transform on the stress-strain data to obtain a time-frequency energy distribution diagram, and extracting instantaneous frequency and instantaneous amplitude according to the time-frequency energy distribution diagram to obtain a time-frequency feature vector; The time-frequency feature vector is subjected to feature dimensionality reduction to obtain a feature set after dimensionality reduction, and a stress-strain feature map is constructed according to the feature set after dimensionality reduction.
4. The cable strand break warning method according to claim 1, characterized in that: The data fusion and adaptive feature selection are performed on the global vibration state diagram and the stress-strain characteristic diagram to obtain a target fusion feature set, including: Performing feature segmentation on the global vibration state diagram to obtain a plurality of local vibration regions, and extracting vibration intensity, frequency and duration features according to the plurality of local vibration regions to obtain a vibration feature vector; Decomposing the stress-strain characteristic diagram in time series to obtain trend terms, periodic terms and residual terms, and calculating the stress-strain change rate and cumulative amount according to the trend terms, periodic terms and residual terms to obtain a stress-strain characteristic vector; Performing Dempster-Shafer evidence theory fusion on the vibration feature vector and the stress-strain feature vector to obtain an initial fusion feature set; Performing feature importance analysis on the initial fusion feature set to obtain feature importance scores, and sorting and threshold screening the feature importance scores to obtain a candidate feature subset; The feature contribution is calculated according to the candidate feature subsets, the feature contribution is dynamically weighted to obtain an adaptive feature weight, and the candidate feature subsets are weighted combined according to the adaptive feature weight to obtain a target fusion feature set.
5. The cable strand break warning method according to claim 1, characterized in that: The adaptive factor update mechanism based on covariance matching is combined with the dynamic window adjustment technology to construct an adaptive Kalman filter combined warning algorithm, and the cable state estimation is performed on the target fusion feature set to obtain the cable state estimation result, including: Constructing a state space model for the target fusion feature set to obtain a system state equation and an observation equation, and initializing Kalman filter parameters according to the system state equation and the observation equation to obtain an initial state estimate and a covariance matrix; Performing sliding window division on the observation data sequence to obtain a dynamic observation window, and calculating the observation noise covariance according to the dynamic observation window to obtain an adaptive observation noise covariance matrix; Performing singular value decomposition on the adaptive observation noise covariance matrix to obtain a covariance eigenvector, and calculating an adaptive factor based on the covariance eigenvector to obtain an updated state prediction covariance; Decomposing the updated state prediction covariance to obtain a lower triangular matrix, and generating an untraceable transformation sample set according to the lower triangular matrix; The state prediction and measurement prediction are performed on the unscented transformation sample set to obtain the predicted state mean and covariance, and the Kalman gain is calculated according to the predicted state mean and covariance, and the state estimation is updated in combination with the observed data to obtain the cable state estimation result.
6. The cable strand break warning method according to claim 1, characterized in that: The cable state estimation result is subjected to a multi-layer machine learning model analysis to obtain a disconnection risk prediction value, and a multi-level warning strategy decision is made on the disconnection risk prediction value to output intelligent disconnection warning information, including: Performing empirical mode decomposition on the cable state estimation result to obtain a plurality of intrinsic mode functions, and inputting the plurality of intrinsic mode functions into a multi-channel time convolution network in a multi-layer machine learning model to obtain a time series feature representation; Inputting the time series feature representation into the autoencoder in the multi-layer machine learning model for noise reduction to obtain a robust feature vector; Inputting the robust feature vector into a hierarchical attention mechanism network in a multi-layer machine learning model for context weighted analysis to obtain weighted context information, and performing dynamic time warping on the weighted context information to obtain an aligned timing pattern; Inputting the aligned temporal patterns into a recursive neural network tree in a multi-layer machine learning model for processing to obtain hierarchical semantic features, and performing multi-task learning processing on the hierarchical semantic features to obtain a shared representation and a task target representation; Inputting the shared representation and the task target representation into an integrated gradient boosting tree in a multi-layer machine learning model to perform disconnection risk prediction, thereby obtaining a disconnection risk prediction value; The disconnection risk prediction values are integrated and fused to obtain a comprehensive risk assessment result, and a multi-level warning strategy is executed according to the comprehensive risk assessment result to output intelligent disconnection warning information.
7. The cable strand break warning method according to claim 6, characterized in that: The aligned time series pattern is input into a recursive neural network tree in a multi-layer machine learning model for processing to obtain hierarchical semantic features, and the hierarchical semantic features are processed by multi-task learning to obtain a shared representation and a task target representation, including: Performing sliding window segmentation on the aligned time series pattern to obtain multiple time segments, and inputting the multiple time segments into leaf nodes of a recursive neural network tree for processing to obtain initial feature representation; Performing local temporal dependency analysis on the initial feature representation to obtain local temporal dependency features, and inputting the local temporal dependency features into the intermediate nodes of the recursive neural network tree for processing to obtain a hierarchical feature combination; Performing long-term dependency analysis on the hierarchical feature combination to obtain long-term dependency features, and constructing the root node of a recursive neural network tree with the long-term dependency features for processing to obtain global semantic features; Performing attention mechanism analysis on the global semantic features to obtain weighted semantic features, and constructing a multi-task learning network with the weighted semantic features to obtain a task sharing layer; A task branch network analysis is performed on the task sharing layer to obtain a task target representation, and feature fusion is performed based on the task sharing layer and the task target representation to obtain a sharing representation and a task target representation.
8. A cable strand break warning device, characterized in that: Used to execute the cable strand disconnection warning method according to any one of claims 1 to 7, the cable strand disconnection warning device comprising: The acquisition module is used to collect multi-modal sensor data of the cable stranding system to obtain vibration acceleration data and stress strain data; An extraction module, used to perform multi-band distributed analysis on the vibration acceleration data to obtain a global vibration state diagram, and to perform time-frequency domain feature extraction on the stress-strain data to obtain a stress-strain feature diagram; A fusion module, used for performing data fusion and adaptive feature selection on the global vibration state diagram and the stress-strain characteristic diagram to obtain a target fusion feature set; The state estimation module is used to construct an adaptive Kalman filter combined early warning algorithm based on an adaptive factor update mechanism of covariance matching and a dynamic window adjustment technology, and to perform cable state estimation on the target fusion feature set to obtain a cable state estimation result; The output module is used to perform a multi-layer machine learning model analysis on the cable status estimation result to obtain a disconnection risk prediction value, and to make a multi-level warning strategy decision on the disconnection risk prediction value to output intelligent disconnection warning information.
9. A computer device, characterized in that: It comprises a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and is characterized in that when the processor executes the computer program, the cable wire breakage warning method described in any one of claims 1 to 7 is implemented.
10. A computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the processor is enabled to execute the cable strand break warning method according to any one of claims 1 to 7.
Citation Information
Patent Citations
Optical fiber state detection method based on multi-source information fusion
CN115905810A
Cable external damage prevention monitoring and early warning method based on multi-modal data fusion
CN118395377A
Intelligent integrated wiring board with network cable quality monitoring
KR102648377B1
Systems and methods for utilizing machine learning to minimize a potential of damage to fiber optic cables
US20220136891A1
Cited By
Method and system for monitoring laying state of pipeline optical cable based on artificial intelligence
CN120178669A
Artificial intelligence-based pipeline optical cable laying status monitoring method and system
CN120178669B
Dry hot rock reservoir horizontal well fracturing fiber breakage early warning method based on DAS monitoring
CN120291864A
Vehicle suspension control method and device, computer equipment and readable storage medium
CN120422604A
Device, method and equipment for detecting loose connection of automobile cable
CN120993277A
