Cable vibration signal feature extraction and mode recognition method and cable maintenance method
By optimizing compression and feature mapping technology, identifying abnormal characteristics of cable vibration signals, the problems of low data processing efficiency and information omission in the prior art are solved, and accurate fault identification and prediction of the cable system is achieved, and the failure risk is reduced.
Patent Information
- Application Number
- CN202411084887.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-08
- Publication Date
- 2024-12-03
AI Technical Summary
The existing cable vibration signal feature extraction and pattern recognition methods have low data processing efficiency, information omission, and limited prediction accuracy, making it difficult to identify complex failure modes, which increases the risk of unexpected downtime and failure.
By collecting vibration data along the cable, dividing data levels according to signal strength, and optimizing and compression; segmenting according to signal change frequency, identifying signal mutations and frequency changes, and establishing a time-frequency characteristic map; generating an abnormal signal characteristic library for matching and analysis, identifying Potential abnormalities and targeted maintenance measures are formulated.
It improves data storage efficiency and processing speed, retains important information, enhances the identification accuracy and prediction ability of abnormal signals, reduces the risk of unexpected downtime, and improves the stability and safety of the cable system.
Smart Images

Figure CN119066568A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of signal processing, and in particular to a cable vibration signal feature extraction and pattern recognition method and a cable maintenance method. Background Art
[0002] The field of signal processing technology is a technical science involving the analysis, processing and interpretation of signals. In the context of cable vibration signal feature extraction and pattern recognition, it focuses on understanding the intrinsic characteristics and patterns of vibration signals generated by cables under differentiated working conditions. The technical field includes a wide range of methods and techniques, from basic filtering and noise elimination to complex feature extraction, classification algorithms and pattern recognition. The goal of signal processing technology is to extract useful information, reduce data complexity, and improve data interpretability to support decision making and prediction.
[0003] Among them, the cable vibration signal feature extraction and pattern recognition method is a technology designed to understand and classify cable vibration signals. The purpose of this method is to accurately analyze the cable vibration signal, extract key features, and accurately identify and classify the cable in differentiated working conditions or fault situations. Its goal is to achieve real-time monitoring of the health status of the cable system, predict potential faults, and take maintenance or repair measures in advance to ensure the reliable operation of the system and extend the service life of the cable, which can greatly reduce the risk of unexpected downtime and improve the stability and safety of the system.
[0004] Traditional cable vibration signal feature extraction and pattern recognition methods focus on basic filtering and noise elimination, or rely on relatively complex and inflexible feature extraction and classification algorithms. This leads to low data processing efficiency, information omissions, limited prediction accuracy and other deficiencies in actual operations. They fail to effectively utilize signal intensity and frequency information for data hierarchical management, resulting in the loss of important vibration information during compression or analysis. They lack effective abnormal signal feature identification and matching mechanisms, making it difficult for the method to accurately predict and identify anomalies when faced with unknown or complex failure modes, increasing the risk of unexpected downtime and failure.
[0005] Therefore, those skilled in the art are committed to developing a cable vibration signal feature extraction and pattern recognition method and a cable maintenance method with high efficiency, comprehensive information and good accuracy. Summary of the invention
[0006] In view of the above-mentioned defects in the prior art, the technical problem to be solved by the present invention is to provide a cable vibration signal feature extraction and pattern recognition method and a cable maintenance method with high efficiency, comprehensive information and good accuracy.
[0007] To achieve the above object, the present invention provides a method for extracting cable vibration signal features, comprising the following steps:
[0008] S1: Collect vibration data along the cable under differentiated conditions, divide the data levels according to signal strength, and obtain an optimized compressed data set;
[0009] S2: segment the optimized compressed data set according to the signal change frequency, divide the signals in each segment into multiple categories according to the energy level, and generate a segmented compressed signal set;
[0010] S3: Based on the segmented compressed signal set, key time points are extracted, including the identification of signal mutations and frequency changes, and feature mapping is established to obtain time-frequency feature mapping;
[0011] S4: Through time-frequency feature mapping, the feature points of abnormal signals are identified, the time and frequency band of the abnormal features are recorded, and an abnormal signal feature library is generated.
[0012] In the present invention, the optimized compressed data set includes a signal strength classification identifier, a corresponding compression ratio and data size information; the segmented compressed signal set includes signal energy level classification categories, a compression ratio for each type of signal and overlapping window configuration parameters; the time-frequency feature map includes a signal mutation point time mark, a frequency change point time mark and a signal strength change feature; the abnormal signal feature library includes the timestamp of the abnormal signal, the frequency band information and a feature description of the abnormal signal.
[0013] Preferably, the step S1 comprises the following steps:
[0014] S101: collecting vibration data based on vibration detection equipment installed along the cable, classifying the data according to vibration intensity thresholds, and generating vibration data classification results;
[0015] S102: Based on the vibration data classification result, a high compression rate is applied to low-intensity signals, and a low compression rate is applied to high-intensity signals, and the data intensity is used as a parameter to optimize the compression process and construct hierarchical compression data;
[0016] S103: Adjust compression algorithm parameters for the hierarchically compressed data, balance storage efficiency and data access speed, and generate an optimized compressed data set.
[0017] Preferably, the step S2 comprises the following steps:
[0018] S201: performing frequency analysis on the optimized compressed data set, segmenting the data according to the changing frequency of the signal, including data with similar frequency characteristics, and generating a frequency segmented data set;
[0019] S202: Based on the frequency segmented data set, using overlapping window technology to identify and classify signals with differentiated energy levels, reduce information omission, and obtain energy classification signal data;
[0020] S203: Designing a corresponding compression strategy for the energy level data according to the energy classification signal data, refining compression parameters to optimize data access efficiency, and obtaining a segmented compression signal set.
[0021] Preferably, the step S3 comprises the following steps:
[0022] S301: traverse the data segments of the segmented compressed signal set, calculate the changes in amplitude and frequency, record the amplitude change exceeding the threshold as the signal mutation point, record the frequency change exceeding the threshold as the frequency change point, and obtain a set of time points of signal mutation and frequency change;
[0023] S302: Setting a time window for the points in the set of time points of signal mutation and frequency change, using a moving average method and a frequency-weighted average method to count the average strength and dominant frequency of the signal in the window, dividing the time points into differentiated intervals according to feature differences, and generating a set of feature change intervals;
[0024] S303: assigning identifiers to the intervals in the characteristic change interval set, recording the start and end time points, average signal strength and dominant frequency, establishing a mapping relationship from time points to signal characteristics, and constructing a time-frequency characteristic mapping.
[0025] Preferably, the step S4 comprises the following steps:
[0026] S401: After collecting the cable vibration signal based on the time-frequency feature mapping, perform time-frequency transformation on it to highlight the time and frequency characteristics of the abnormal part, reveal the potential abnormal characteristics in the signal, and generate a time-frequency mapping diagram;
[0027] S402: Based on the time-frequency map, screen abnormal signal feature points, including points where the signal strength suddenly changes and the frequency deviates from the normal range, record the time and frequency of the points, and form an abnormal point record table;
[0028] S403: According to the abnormal point record table, the identified abnormal feature points are sorted and summarized, including timestamps, frequency values and corresponding intensities, identifiers are assigned to the abnormal points, and an abnormal signal feature library is established.
[0029] The present invention also provides a cable pattern recognition method, comprising the cable vibration signal feature extraction method as described above, and further comprising the following steps:
[0030] S5: matching the newly collected signal with the abnormal signal feature library, marking the signal with good matching degree as a potential abnormality, recording the characteristics of the signal, and obtaining a potential abnormal signal record;
[0031] S6: Analyze the potential abnormal signal records, compare historical data, identify the abnormal type and cause, generate abnormal signal analysis results, and complete cable vibration pattern recognition.
[0032] The potential abnormal signal record includes signal feature matching information, signal timestamp and signal frequency range, and the abnormal signal analysis result includes abnormal type identification, cause analysis description and historical data comparison result.
[0033] Preferably, the step S5 further comprises the following steps:
[0034] S501: Based on the abnormal signal feature library, perform time-frequency transformation on the collected cable vibration signal, make a preliminary comparison with the abnormal signal feature library, screen the feature-related signals in the abnormal signal feature library, and generate a matching signal set;
[0035] S502: Analyze the signals in the matching signal set, compare their timestamps, frequencies and intensity details, determine that they are potential abnormal signals, record the characteristics of the signals, and obtain a potential abnormal signal detail table;
[0036] S503: Based on the information in the potential abnormal signal detail table, the signal matching degree, characteristic parameters and time frequency information are summarized, and iterative analysis and processing are performed to provide a decision basis, thereby generating a potential abnormal signal record.
[0037] Preferably, the step S6 further comprises the following steps:
[0038] S601: traverse the potential abnormal signal records, compare the time series of the signal with the time series of the historical abnormal pattern, mark the signal matching the historical pattern, classify the signal into an abnormal type according to the matching result, and obtain an abnormal type identification set;
[0039] S602: Analyze characteristic parameters of the signals in the abnormal type identification set, identify potential causes of signal abnormalities by comparing the similarities between the parameters and historical fault cases, and generate an abnormal cause analysis set;
[0040] S603: Based on the abnormal cause analysis set, a decision tree algorithm is used to formulate targeted processing measures for each cause, evaluate the feasibility and potential effect of the measures, and construct abnormal signal analysis results.
[0041] The present invention also provides a cable maintenance method, which performs cable maintenance based on the results obtained by the cable pattern recognition method as described above. According to the abnormal signal analysis results, targeted measures are formulated, working parameters are adjusted and monitoring measures are strengthened to obtain a maintenance preventive measures plan. The maintenance preventive measures plan includes a list of adjusted working parameters, details of strengthened monitoring measures and an equipment maintenance management plan.
[0042] Preferably, the method comprises the following steps:
[0043] S701: Based on the abnormal signal analysis result, analyze the occurrence frequency and distribution of the abnormal signal, determine the impact on the equipment components, monitor the strength and frequency of the abnormal signal, identify the affected part of the key equipment, and generate a key impact analysis result;
[0044] S702: Based on the key impact analysis results, plan adjustment measures for corresponding components and modification plans for working parameters, and obtain an adjustment execution plan by referring to the feasibility of the adjustment and the expected improvement effect;
[0045] S703: Based on the adjustment execution plan, a multi-dimensional maintenance and management strategy is formulated to optimize the stability and performance of the equipment through regular maintenance, key component monitoring and parameter optimization, and to build a maintenance preventive measures plan.
[0046] The beneficial effects of the present invention are as follows: the present invention collects the vibration number of the cable in a differentiated state by installing vibration detection equipment along the cable, and uses a differentiated data compression rate to preliminarily process the vibration data, so that data storage is more efficient, and the method of dividing the levels based on the signal strength and applying the differentiated compression rate is used to optimize the size and processing speed of the data set while retaining important information. By segmenting the signal according to the frequency of signal change and dividing the signal into multiple categories according to the energy level in each segment, the management efficiency and analysis accuracy of the data are iteratively improved. The strategy uses overlapping window technology to reduce information omission and ensure data integrity. The extraction of key time points and the construction of time-frequency feature mapping based on this provide an accurate analysis basis for vibration data analysis, making the identification of abnormal signals more accurate, the establishment of abnormal signal feature libraries and the matching of newly collected signals enhance the prediction ability of potential abnormalities. By analyzing the potential abnormal signal records and comparing them with historical data, the abnormal type and cause can be accurately identified, and more targeted maintenance and preventive measures can be formulated. The present invention can make the identification of cable abnormal information faster and more accurate, enhance the prediction ability of potential abnormalities, and help formulate more targeted maintenance and preventive measures. BRIEF DESCRIPTION OF THE DRAWINGS
[0047] Figure 1 It is a schematic diagram of a specific implementation method of the present invention.
[0048] Figure 2 It is a flow chart of step S1 of a specific implementation mode of the present invention.
[0049] Figure 3 It is a flow chart of step S2 of a specific implementation mode of the present invention.
[0050] Figure 4 It is a flow chart of step S3 of a specific implementation mode of the present invention.
[0051] Figure 5 It is a flow chart of step S4 of a specific implementation mode of the present invention.
[0052] Figure 6 It is a flow chart of step S5 of a specific implementation mode of the present invention.
[0053] Figure 7 It is a flow chart of step S6 of a specific implementation mode of the present invention.
[0054] Figure 8 It is a flow chart of step S7 of a specific implementation mode of the present invention. DETAILED DESCRIPTION
[0055] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.
[0056] A cable vibration signal feature extraction method, such as Figure 1 As shown, the following steps are included:
[0057] S1: Collect vibration data of cables in differentiated states, classify data according to signal strength, and obtain optimized compressed data sets. Install vibration detection equipment along the cables, collect vibration data of cables in differentiated states, and classify data according to signal strength.
[0058] This step can be performed by the following method:
[0059] S101: Based on the vibration detection equipment installed along the cable, vibration data is collected, data is classified according to the vibration intensity threshold, and vibration data classification results are generated. In this step, vibration detection equipment is installed along the cable so that regular vibration data collection can be performed.
[0060] Based on the installation of vibration detection equipment along the cable, the Fourier transform algorithm is used to perform frequency domain analysis of the vibration signal. The NumPy library in Python is used to execute the fft function. The input parameter is the vibration signal data. The vibration signal in the time domain is converted into a frequency domain signal. The threshold judgment logic is applied to the converted frequency domain signal. The specific threshold parameter is set as the median value of the vibration intensity. The frequency domain signal is classified into high intensity and low intensity. The classification logic is implemented through the if-else statement, and the classification results are stored in the database. The SQL statement INSERT INTO is used to insert data to generate the vibration data classification results.
[0061] S102: Based on the vibration data classification result, a high compression rate is applied to low-intensity signals, and a low compression rate is applied to high-intensity signals. The data intensity is used as a parameter to optimize the compression process and construct hierarchical compressed data.
[0062] Based on the classification results of vibration data, the Lempel-Ziv-Welch (LZW) compression algorithm is used to perform hierarchical compression on the data. For low-intensity signals, the compression rate parameter of the LZW compression algorithm is set to a high compression rate, and the specific value is set to 9:1. For high-intensity signals, the compression rate parameter is set to a low compression rate, and the specific value is set to 4:1. The compress function is called using Python's zlib library, and the input parameters are vibration data and compression rate parameters. Compression operations with different compression rates are performed on vibration data of different intensities. The compressed data is stored in the data set to generate hierarchical compressed data.
[0063] S103: Adjust compression algorithm parameters for the hierarchically compressed data, balance storage efficiency and data access speed, and generate an optimized compressed data set.
[0064] For hierarchical compressed data, a dynamic adjustment strategy of compression parameters is adopted, and a Python script is used to calculate the current values of storage efficiency and data access speed. The calculation method is to access the compressed data records in the database, count the access time and data size, and use the information to dynamically adjust the compression rate parameters in the LZW compression algorithm. The adjustment logic is: if the data access speed is lower than the preset standard speed threshold, reduce the compression rate to reduce the decompression time; if the storage efficiency is lower than the preset storage space utilization threshold, increase the compression rate to save storage space. The adjustment process is cyclically adjusted until the balance point of storage efficiency and data access speed is reached to generate an optimized compressed data set.
[0065] The optimized compressed data set includes signal strength classification identification, corresponding compression ratio and data size information.
[0066] S2: Segment the optimized compressed data set according to the signal change frequency, and divide the signals in each segment into multiple categories according to the energy level to generate a segmented compressed signal set. Figure 3 As shown, this step can be specifically performed by the following steps:
[0067] S201: Perform frequency analysis on the optimized compressed data set, segment the data according to the changing frequency of the signal, including data with similar frequency characteristics, and generate a frequency segmented data set.
[0068] Based on the optimized compressed data set, the short-time Fourier transform (STFT) algorithm is used for frequency analysis. The stft function in Python's SciPy library is used. The input parameters include vibration signal data and window function type. The specific window function is the Hanning window. The frequency change analysis of the signal is performed in different time periods. According to the frequency characteristics of the signal in different time windows, the data is segmented. Each segment contains data with similar frequency characteristics. All signal segments are traversed using a for loop and classified according to the frequency characteristics. The results are stored in a dataset named after the frequency characteristics to generate a frequency segmented dataset.
[0069] S202: Based on the frequency segmented data set, overlapping window technology is applied to identify and classify signals with differentiated energy levels, reduce information omission, and obtain energy classification signal data.
[0070] In this embodiment, based on the frequency segmented data set, the overlapping window technology is combined with the power spectral density (PSD) analysis, and the specgram function of matplotlib.mlab in Python is used. The input parameters include signal data, window size and overlapping window ratio. The window size is specifically set to 256 data points and the overlapping window ratio is 50%. This method can increase the frequency resolution while maintaining the time resolution, identify and classify signals of different energy levels, and divide the signals into high energy and low energy by setting a specific energy threshold. If-else logic is used for judgment and classification, energy classification signal data is obtained.
[0071] S203: Designing a corresponding compression strategy for the energy level data according to the energy classification signal data, refining compression parameters to optimize data access efficiency, and obtaining a segmented compression signal set.
[0072] Signal data is classified according to energy. In this embodiment, adaptive differential pulse code modulation (ADPCM) compression technology is used to apply different compression strategies to data of different energy levels. The wave library in Python is used to call the adpcm function. The input parameters are signal data and initial step size. For high-energy signals, a smaller initial step size value is set to maintain signal quality. For low-energy signals, a larger initial step size value is set to improve the compression ratio. All signal data are looped through, and corresponding compression parameters are applied according to the energy level to optimize access efficiency and obtain a segmented compressed signal set.
[0073] The segmented compression signal set includes signal energy level classification categories, compression ratio of each type of signal and overlapping window configuration parameters.
[0074] S3: Based on the segmented compressed signal set, extract key time points, including the identification of signal mutations and frequency changes, establish feature mapping, and obtain time-frequency feature mapping. The specific steps include:
[0075] S301: traverse the data segments of the segmented compressed signal set, calculate the changes in amplitude and frequency, record the amplitude changes exceeding the threshold as signal mutation points, record the frequency changes exceeding the threshold as frequency change points, and obtain a set of time points for signal mutation and frequency change.
[0076] In this embodiment, signal processing technology is used to perform spectrum analysis on data segments in the traversed segmented compressed signal set using a short-time Fourier transform (STFT) algorithm. The window function is set to a Hamming window, the window length is set to 256 sample points, and the step length is set to 128 sample points. The spectrum of each data segment is calculated using the STFT algorithm to obtain the amplitude and frequency values of each frequency component. A threshold judgment method is used to set the amplitude change threshold to 0.5 and the frequency change threshold to 5 Hz. The STFT results are traversed, and data points exceeding the amplitude change threshold are recorded as signal mutation points, and data points exceeding the frequency change threshold are recorded as frequency change points, to generate a set of time points for signal mutations and frequency changes.
[0077] S302: Setting a time window for the points in the set of time points of the signal mutation and frequency change, using the moving average method and the frequency-weighted average method to count the average strength and dominant frequency of the signal in the window, dividing the time points into differentiated intervals according to feature differences, and generating a set of feature change intervals.
[0078] Based on time series analysis, the moving average method is used to process the time point set of signal mutations and frequency changes obtained in S301. The time window length is set to 1 second and the step length is set to 0.5 seconds. The average intensity and dominant frequency of the signal data in each time window are calculated. The dominant frequency is calculated using the frequency weighted average method, that is, the amplitude of each frequency component is used as the weight, and the frequency value obtained after weighted average calculation is used as the dominant frequency. According to the average intensity and dominant frequency in the time window, the k-means clustering algorithm is used, and the preset number of clusters is 3. The time points are divided into differentiated intervals according to the feature differences to generate a set of feature change intervals.
[0079] In this embodiment, the moving average method is performed according to the following formula:
[0080]
[0081] Where MA′ is the moving average, N is the time window length, indicating the number of data points considered, and X i is the signal strength value within the time window, indicating the signal strength collected within the specified time window, W i is the weight coefficient of each point in the time window, which is used to adjust the influence of each point in the average value calculation, and C is the constant offset, which is used to adjust the baseline of the entire signal intensity.
[0082] The frequency-weighted average method is performed according to the following formula:
[0083]
[0084] Where FWA′ is the frequency weighted average, f i is the frequency component, indicating the value of each frequency in the signal, A i is the corresponding amplitude, indicating the intensity of each frequency component, W i ′ is the weight coefficient of the frequency component, which is used to strengthen or weaken the role of a specific frequency component in the calculation, and D is the adjustment constant, which is used to fine-tune the calculated dominant frequency value.
[0085] The execution process is as follows;
[0086] New weight coefficients and constant adjustment items are introduced to improve the flexibility and accuracy of calculation. i and W i ′ can be determined by adopting a strategy based on the amplitude of signal changes within the time window, for example, a larger amplitude of change is given a higher weight. This requires setting a reasonable weight distribution strategy through analysis of historical data. The setting of C and D can be based on prior knowledge of the overall level of the signal, or obtained through statistical analysis in the data preprocessing stage to ensure that the algorithm has good adaptability and accuracy to data in different scenarios. Such improvements make the original calculation method more flexible and can better adapt to the complex changes of signal strength and frequency in practical applications.
[0087] S303: assigning identifiers to the intervals in the characteristic change interval set, recording the start and end time points, average signal strength and dominant frequency, establishing a mapping relationship from time points to signal characteristics, and constructing a time-frequency characteristic mapping.
[0088] The time-frequency feature map includes the time mark of the signal mutation point, the time mark of the frequency change point and the signal strength change feature.
[0089] In this embodiment, the execution process is as follows:
[0090] An identifier generation algorithm is used to assign a unique identifier to each interval in the feature change interval set in S302, and the start and end time points, average signal strength and dominant frequency of each interval are recorded. A dictionary data structure is constructed, in which the key is the identifier of the time point and the value is a sub-dictionary containing information such as the start and end time points, average signal strength and dominant frequency. This data structure is used to establish a mapping relationship from time points to signal features, and a time-frequency feature mapping is constructed.
[0091] S4: Through time-frequency feature mapping, identify the feature points of abnormal signals, record the time and frequency band of the abnormal features, and generate an abnormal signal feature library. Figure 5 As shown,
[0092] S401: Based on the time-frequency feature mapping, after collecting the cable vibration signal, perform time-frequency transformation on it to highlight the time and frequency characteristics of the abnormal part, reveal the potential abnormal characteristics in the signal, and generate a time-frequency mapping diagram.
[0093] Based on time-frequency feature mapping, the wavelet transform method is used to analyze the cable vibration signal. The cwt function is executed using Python's PyWavelets library. The input parameters include vibration signal data and wavelet function. The specific wavelet function selects Morlet wavelet to highlight the time and frequency characteristics of the abnormal part. Wavelet transform decomposes the signal into different frequency bands, which significantly improves the identifiability of the abnormal signal. For each decomposed frequency band, the numpy library is used to calculate its energy distribution. By comparing the changes in energy distribution, the potential abnormal characteristics in the signal are revealed. The matplotlib library is used to draw the time-frequency mapping diagram, and the time and frequency characteristics of the vibration signal are displayed in a graphical form to generate a time-frequency mapping diagram.
[0094] S402: Based on the time-frequency map, filter out abnormal signal feature points, including points where the signal strength suddenly changes and the frequency deviates from the normal range, record the time and frequency of the points, and form an abnormal point record table.
[0095] In this embodiment, based on the time-frequency mapping diagram, the threshold analysis technology is applied to screen abnormal signal feature points, and the data points in the mapping diagram are traversed using a Python script. The threshold is set to the upper quartile of the signal strength and the frequency deviation from the normal range of ±20%. For data points that exceed the signal strength threshold or the frequency deviates from the set range, their time and frequency are recorded, and the pandas library is used to create a DataFrame with the column names set to time and frequency. For each data point that meets the conditions, the append method of DataFrame is used to add records to form an abnormal point record table.
[0096] S403: According to the abnormal point record table, the identified abnormal feature points are sorted and summarized, including timestamps, frequency values and corresponding intensities, identifiers are assigned to the abnormal points, and an abnormal signal feature library is established.
[0097] According to the abnormal point record table, the abnormal feature points are sorted and summarized using a clustering algorithm. The K-means algorithm is used, and the KMeans function is executed using Python's scikit-learn library. The input parameters include the timestamp, frequency value and corresponding intensity of the abnormal point. The number of clusters is set to be automatically determined according to the distribution of the abnormal points. The algorithm calculates the Euclidean distance between data points, classifies the abnormal points with similar timestamps, frequency values and intensities into one category, assigns a unique identifier to each cluster, uses the UUID library to generate identifiers, stores the clustering results and their identifiers in the abnormal signal feature library, and uses the SQLite database to execute the INSERT INTO statement for data storage to establish the abnormal signal feature library.
[0098] The abnormal signal feature library includes the timestamp, frequency band information and feature description of the abnormal signal.
[0099] The present invention also provides a cable pattern recognition method, comprising the cable vibration signal feature extraction method as described above, and further comprising the following steps:
[0100] S5: Match the newly collected signal with the abnormal signal feature library, mark the signal with a good matching degree as a potential abnormality, record the characteristics of the signal, and obtain a potential abnormal signal record.
[0101] This step specifically includes:
[0102] S501: Based on the abnormal signal feature library, perform time-frequency transformation on the collected cable vibration signal, make a preliminary comparison with the abnormal signal feature library, screen the feature-related signals in the abnormal signal feature library, and generate a matching signal set;
[0103] Based on the abnormal signal feature library, the fast Fourier transform (FFT) algorithm is used to perform time-frequency transformation on the collected cable vibration signal. The number of FFT transformation points is set to 1024, and the window function is Hanning window to reduce the edge effect. The frequency domain representation of the signal is obtained by FFT algorithm conversion. The cosine similarity calculation method is used to compare the spectrum of each transformed signal with the spectrum in the abnormal signal feature library. The similarity threshold is set to 0.8. For each signal, its similarity with all signals in the feature library is calculated, and the signals with similarity greater than or equal to the threshold are screened to generate a matching signal set.
[0104] S502: Analyze the signals in the matching signal set, compare their timestamps, frequencies and intensity details, determine that they are potential abnormal signals, record the characteristics of the signals, and obtain a potential abnormal signal detail table.
[0105] The specific implementation method of this step is as follows: for each signal in the matching signal set, a detail comparison algorithm is used, which analyzes the timestamp, frequency and intensity details of the signal. The timestamp comparison is calculated using absolute time difference, and the frequency and intensity comparison are based on standard deviation analysis. The frequency comparison standard deviation threshold is set to 0.2Hz, and the intensity comparison standard deviation threshold is set to 5dB. The timestamp, frequency and intensity details of each signal are compared by the algorithm, and the signal that meets all threshold conditions is determined as a potential abnormal signal. The timestamp, frequency and intensity information of the signal is recorded to obtain a potential abnormal signal detail table.
[0106] S503: Based on the information in the potential abnormal signal detail table, the signal matching degree, characteristic parameters and time frequency information are summarized, and iterative analysis and processing are performed to provide a decision basis, thereby generating a potential abnormal signal record.
[0107] During specific execution, based on the information in the potential abnormal signal detail table, an iterative analysis method is adopted. This method uses a linear regression model to analyze the relationship between signal matching, characteristic parameters and time-frequency information. The number of iterations is set to 10. The coefficients of the regression model are adjusted in each iteration to optimize the model's explanatory power for signal characteristic parameters. During the iterative analysis process, a decision tree algorithm is used with a tree depth of 5 as the basis for determining whether a signal is abnormal. The data after iterative analysis and processing is used to generate potential abnormal signal records.
[0108] The potential abnormal signal records include signal feature matching information, signal timestamp and signal frequency band range.
[0109] S6: Analyze the potential abnormal signal records, compare historical data, identify the abnormal type and cause, generate abnormal signal analysis results, and complete cable vibration pattern recognition.
[0110] The specific implementation includes the following steps:
[0111] S601: traverse the potential abnormal signal records, compare the time series of the signal with the time series of the historical abnormal pattern, mark the signal matching the historical pattern, classify the signal into an abnormal type according to the matching result, and obtain an abnormal type identification set.
[0112] Traverse the potential abnormal signal records and adopt the time series similarity analysis method. Specifically, the dynamic time warping (DTW) algorithm is applied to compare the signal time series with the time series of historical abnormal patterns. The dtw function is executed using Python's tslearn library. The input parameters are the current signal time series and the historical abnormal time series. The DTW algorithm calculates the minimum cumulative distance to find the best alignment between time series, thereby identifying similar patterns. For each signal, based on its DTW distance with the historical pattern, the if-else logic is used to determine whether the distance is lower than the preset threshold. The preset threshold is determined according to the average distance of historical data. Signals below the threshold are marked as matching the historical pattern. The matching results are stored in the database. The SQL statement INSERT INTO is used to insert data to obtain the abnormal type recognition set.
[0113] S602: Analyze characteristic parameters of the signals in the abnormal type identification set, identify potential causes of signal abnormalities by comparing the similarities between the parameters and historical fault cases, and generate an abnormal cause analysis set.
[0114] During execution, the characteristic parameter analysis method is applied to the signals in the abnormal type identification set, including the principal component analysis (PCA) method to analyze the characteristic parameters of the signal. The PCA function is executed using Python's sklearn library. The input parameters include the signal data set. PCA extracts the most important characteristic parameters through dimensionality reduction processing, and compares the parameters with the characteristic parameters in the historical fault case database. The comparison method uses Euclidean distance calculation. For each signal characteristic parameter set, the Euclidean distance between it and the characteristic parameter set of the historical fault case is calculated. According to the size of the distance, the if-else logic is used to determine which signals have similar characteristic parameters to the historical fault cases. Signals with high similarity are classified as corresponding fault types to generate an abnormal cause analysis set.
[0115] S603: Based on the abnormal cause analysis set, a decision tree algorithm is used to formulate targeted processing measures for each cause, evaluate the feasibility and potential effect of the measures, and construct abnormal signal analysis results.
[0116] Based on the abnormal cause analysis set, the decision tree method is used to formulate targeted treatment measures. The DecisionTreeClassifier function is executed using Python's sklearn.tree library. The input parameters include the characteristic parameters of the abnormal cause analysis set and the corresponding fault type. The decision tree learns through training data and formulates specific treatment measures for each cause. The feasibility and potential effect of the treatment measures are evaluated through the evaluation indicators of the decision tree model, such as accuracy and recall. A set of treatment measures is formulated for each abnormal signal, including necessary maintenance or adjustment suggestions. The treatment measures are recorded in detail in the report. The pandas library is used to save the report in a CSV file to construct the abnormal signal analysis results.
[0117] The decision tree algorithm follows the formula:
[0118]
[0119] Among them, IG′(D, f) is the information gain measure, D is the entire data set, f is the feature selected for segmentation, and D v is the subset of the data set when the value of feature f is v, Entropy(D) is the entropy of data set D, w1 and w2 are weight coefficients used to adjust the contribution ratio of entropy and subset entropy, α and β are newly added weight coefficients, SD(D) is the standard deviation of data set D, which is used to measure the dispersion of data, Var(f) is the variance of feature f, which is used to measure the variation of feature values, and Values(f) is the value of feature f in data set D.
[0120] The execution process is as follows;
[0121] First, calculate the entropy of the dataset D and each subset D v The entropy is consistent with the original formula. Next, two new parameters are introduced: the standard deviation SD(D) of the data set D and the variance Var(f) of the feature f. The standard deviation is used to measure the degree of deviation of the data points in the data set from the mean value, and the variance is used to measure the degree of variation of the feature values. By adding these two parameters, the accuracy of the model in processing complex data can be improved. The weight coefficients w1 and w2 are used to adjust the influence of the original entropy value and the subset entropy value, while the newly added weight coefficients α and β are used to adjust the contribution of the standard deviation and variance to the information gain. The weight coefficient is determined by optimizing the training data through cross-validation or model selection technology to find the optimal model performance.
[0122] The abnormal signal analysis results include abnormal type identification, cause analysis description and historical data comparison results.
[0123] The present invention also provides a cable maintenance method, which performs cable maintenance based on the results obtained by the cable vibration signal feature extraction method as described above. According to the abnormal signal analysis results, targeted measures are formulated, working parameters are adjusted and monitoring measures are strengthened to obtain a maintenance preventive measures plan. The specific steps are as follows:
[0124] S701: Based on the abnormal signal analysis result, analyze the occurrence frequency and distribution of the abnormal signal, determine the impact on the equipment components, monitor the strength and frequency of the abnormal signal, identify the impact part of the key equipment, and generate the key impact analysis result.
[0125] Specifically, based on the abnormal signal analysis results, the statistical analysis method can be used to analyze the frequency and distribution of abnormal signals. The Kurtosis and Skewness calculations can be performed through Python's SciPy library to determine the skewness and kurtosis of the signal distribution. The abnormal signal strength monitoring threshold is set to -20dB to 0dB, and the frequency monitoring range is set to 20Hz to 2000Hz. Signal processing technology is used, specifically a bandpass filter is used, and the lower limit frequency of the filter is set to 20Hz and the upper limit frequency is set to 2000Hz to eliminate signals that are not within the frequency range of interest. The strength and frequency of abnormal signals are monitored in real time. Through the frequency distribution analysis of abnormal signals, the key components associated with the degradation of equipment performance are identified, and key impact analysis results are generated.
[0126] S702: Based on the key impact analysis results, plan the adjustment measures of the corresponding components and the modification plan of the working parameters, and obtain the adjustment execution plan with reference to the feasibility of the adjustment and the expected improvement effect.
[0127] Based on the results of key impact analysis, the decision tree algorithm is used, specifically the DecisionTreeClassifier in Python's Scikit-learn library, and the tree depth is set to 4. It is used to plan the adjustment measures of the corresponding components and the modification plans of the working parameters. The feasibility of the adjustment measures is analyzed through cost-benefit analysis, specifically using Net Present Value (NPV) and Internal Rate of Return (IRR) calculation. The expected improvement effect is predicted by simulating the impact of the adjusted working parameters on the equipment performance. The Monte Carlo simulation method is used, and the number of simulations is set to 1000 to estimate the potential improvement brought about by the adjustment execution plan, and the adjustment execution plan is obtained.
[0128] S703: Based on the adjustment execution plan, a multi-dimensional maintenance and management strategy is formulated to optimize the stability and performance of the equipment through regular maintenance, key component monitoring and parameter optimization, and to build a maintenance preventive measures plan.
[0129] The maintenance preventive action plan includes a list of adjusted operating parameters, details of enhanced monitoring measures and an equipment maintenance management plan.
[0130] In this embodiment, the execution plan is adjusted comprehensively, and the project management algorithm is adopted, specifically using the critical path method (CPM) and the project evaluation and review technique (PERT). The regular maintenance cycle is set to once a quarter, and the key components are monitored using vibration analysis and temperature monitoring technology. The parameter optimization uses a genetic algorithm, and the population size is set to 50 and the number of iterations is set to 20. This is used to optimize the equipment operating parameters. Through multi-dimensional maintenance and management strategies, the stability and performance of the equipment are systematically optimized, and a maintenance preventive measures plan is constructed.
[0131] The preferred specific embodiments of the present invention are described in detail above. It should be understood that a person skilled in the art can make many modifications and changes based on the concept of the present invention without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art through logical analysis, reasoning or limited experiments based on the concept of the present invention on the basis of the prior art should be within the scope of protection determined by the claims.
Claims
1. A cable vibration signal feature extraction method, Its characteristics are: The following steps are involved: S1: Collect vibration data along the cable under differentiated conditions, divide the data levels according to signal strength, and obtain an optimized compressed data set; S2: segment the optimized compressed data set according to the signal change frequency, divide the signals in each segment into multiple categories according to the energy level, and generate a segmented compressed signal set; S3: Based on the segmented compressed signal set, key time points are extracted, including the identification of signal mutations and frequency changes, and feature mapping is established to obtain time-frequency feature mapping; S4: Through time-frequency feature mapping, the feature points of abnormal signals are identified, the time and frequency band of the abnormal features are recorded, and an abnormal signal feature library is generated.
2. The cable vibration signal feature extraction method according to claim 1, Its characteristics are: The step S1 comprises the following steps: S101: collecting vibration data based on vibration detection equipment installed along the cable, classifying the data according to vibration intensity thresholds, and generating vibration data classification results; S102: Based on the vibration data classification result, a high compression rate is applied to low-intensity signals, and a low compression rate is applied to high-intensity signals, and the data intensity is used as a parameter to optimize the compression process and construct hierarchical compression data; S103: Adjust compression algorithm parameters for the hierarchically compressed data, balance storage efficiency and data access speed, and generate an optimized compressed data set.
3. The cable vibration signal feature extraction method according to claim 1, Its characteristics are: The step S2 comprises the following steps: S201: performing frequency analysis on the optimized compressed data set, segmenting the data according to the changing frequency of the signal, including data with similar frequency characteristics, and generating a frequency segmented data set; S202: Based on the frequency segmented data set, using overlapping window technology to identify and classify signals with differentiated energy levels, reduce information omission, and obtain energy classification signal data; S203: Designing a corresponding compression strategy for the energy level data according to the energy classification signal data, refining compression parameters to optimize data access efficiency, and obtaining a segmented compression signal set.
4. The cable vibration signal feature extraction method according to claim 1, Its characteristics are: The step S3 comprises the following steps: S301: traverse the data segments of the segmented compressed signal set, calculate the changes in amplitude and frequency, record the amplitude change exceeding the threshold as the signal mutation point, record the frequency change exceeding the threshold as the frequency change point, and obtain a set of time points of signal mutation and frequency change; S302: Setting a time window for the points in the set of time points of signal mutation and frequency change, using a moving average method and a frequency-weighted average method to count the average strength and dominant frequency of the signal in the window, dividing the time points into differentiated intervals according to feature differences, and generating a feature change interval set; S303: assigning identifiers to the intervals in the feature change interval set, recording the start and end time points, average signal strength and dominant frequency, establishing a mapping relationship from time points to signal features, and constructing a time-frequency feature mapping.
5. The cable vibration signal feature extraction method according to claim 1, Its characteristics are: The step S4 comprises the following steps: S401: After collecting the cable vibration signal based on the time-frequency feature mapping, perform time-frequency transformation on it to highlight the time and frequency characteristics of the abnormal part, reveal the potential abnormal characteristics in the signal, and generate a time-frequency mapping diagram; S402: Based on the time-frequency map, filter out abnormal signal feature points, including points where the signal strength suddenly changes and the frequency deviates from the normal range, record the time and frequency of the points, and form an abnormal point record table; S403: According to the abnormal point record table, the identified abnormal feature points are sorted and summarized, including timestamps, frequency values and corresponding intensities, identifiers are assigned to the abnormal points, and an abnormal signal feature library is established.
6. A cable pattern recognition method, Its characteristics are: The method for extracting cable vibration signal features as claimed in any one of claims 1 to 5 further comprises the following steps: S5: matching the newly collected signal with the abnormal signal feature library, marking the signal with good matching degree as a potential abnormality, recording the characteristics of the signal, and obtaining a potential abnormal signal record; S6: Analyze the potential abnormal signal records, compare historical data, identify the abnormal type and cause, generate abnormal signal analysis results, and complete cable pattern recognition.
7. The cable pattern recognition method according to claim 6, Its characteristics are: The step S5 further comprises the following steps: S501: Based on the abnormal signal feature library, perform time-frequency transformation on the collected cable vibration signal, make a preliminary comparison with the abnormal signal feature library, screen the feature-related signals in the abnormal signal feature library, and generate a matching signal set; S502: Analyze the signals in the matching signal set, compare their timestamps, frequencies and intensity details, determine that they are potential abnormal signals, record the characteristics of the signals, and obtain a potential abnormal signal detail table; S503: Based on the information in the potential abnormal signal detail table, the signal matching degree, characteristic parameters and time frequency information are summarized, and iterative analysis and processing are performed to provide a decision basis, thereby generating a potential abnormal signal record.
8. The cable pattern recognition method according to claim 6, Its characteristics are: The step S6 further comprises the following steps: S601: traverse the potential abnormal signal records, compare the time series of the signal with the time series of the historical abnormal pattern, mark the signal matching the historical pattern, classify the signal into an abnormal type according to the matching result, and obtain an abnormal type identification set; S602: Analyze characteristic parameters of the signals in the abnormal type identification set, identify potential causes of signal abnormalities by comparing the similarities between the parameters and historical fault cases, and generate an abnormal cause analysis set; S603: Based on the abnormal cause analysis set, a decision tree algorithm is used to formulate targeted processing measures for each cause, evaluate the feasibility and potential effect of the measures, and construct abnormal signal analysis results.
9. A cable maintenance method, Its characteristics are: Cable maintenance is performed based on the results obtained by the cable pattern recognition method as described in any one of claims 6 to 8.
10. The cable maintenance method according to claim 9, Its characteristics are: The following steps are involved: S701: Based on the abnormal signal analysis result, analyze the occurrence frequency and distribution of the abnormal signal, determine the impact on the equipment components, monitor the strength and frequency of the abnormal signal, identify the affected part of the key equipment, and generate a key impact analysis result; S702: Based on the key impact analysis results, plan adjustment measures for corresponding components and modification plans for working parameters, and obtain an adjustment execution plan by referring to the feasibility of the adjustment and the expected improvement effect; S703: Based on the adjustment execution plan, a multi-dimensional maintenance and management strategy is formulated to optimize the stability and performance of the equipment through regular maintenance, key component monitoring and parameter optimization, and to build a maintenance preventive measures plan.
Citation Information
Patent Citations
Submarine cable fault feature extraction and signal processing method and system
CN116127283A
Hydroelectric equipment fault diagnosis method based on fault matching algorithm
CN117009828A
Cable vibration signal time frequency analysis system
CN118168652A
High-voltage cable external damage prevention monitoring system based on distributed optical fiber vibration sensing technology
CN118258450A
Systems and methods for utilizing machine learning to minimize a potential of damage to fiber optic cables
US20220136891A1
Cited By
High-performance buried cable for airport navigation aid lamp
CN119291364A
Abrupt change signal positioning method, device and equipment based on PCA algorithm
CN119669990A
Method, device and equipment for locating mutation signals based on PCA algorithm
CN119669990B
Lightweight power cable material characteristic database and analysis platform
CN119892109A
Lightweight Power Cable Material Property Database and Analysis Platform
CN119892109B
