Multi-scattered wave identification method for broadband environment vibration
By employing a sliding fixed-length window segmentation and factor number grouping method, combined with an attention-fully connected network, the problem of feature extraction and classification for multiple scattering wave identification in broadband environments was solved, achieving high-precision vibration event identification and risk warning.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-03
- Publication Date
- 2026-03-10
AI Technical Summary
Existing vibration event recognition technologies suffer from limited feature extraction, low signal-to-noise ratio, and high false alarm rate in broadband environments due to the varying lengths of multiple scattered wave sequences and complex temporal coupling relationships. Furthermore, the lack of a data richness pre-screening mechanism leads to information loss and wasted computing power.
Multiple scattering wave data is segmented using a sliding fixed-length window to ensure that each segment has a consistent length. The data is grouped by the number of factors and significant features are extracted. An attention-fully connected network is then used for classification to suppress data interference and ensure the accuracy of feature extraction and classification.
It improves the accuracy and robustness of vibration event identification, reduces the false identification rate, and achieves high-precision real-time identification and risk warning of events such as micro-vibrations and blasting.
Smart Images
Figure CN121637032A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of monitoring and signal processing technology, and specifically to a method for identifying multiple scattered waves from broadband environmental vibrations. Background Technology
[0002] Existing vibration event recognition technologies are mostly based on extracting features directly from single-segment waveforms or fixed-duration windows, without considering the "path-period" coupling characteristics of multiple scattered waves propagating in complex media. This results in a single feature dimension, low signal-to-noise ratio, and insufficient ability to distinguish short-period events such as micro-vibrations and blasting.
[0003] Traditional sliding window segmentation methods generally use empirical window lengths, which are mismatched with the actual scattering sequence lengths. This results in the same event being split or different events being mixed, leading to inter-class overlap and intra-class discrepancies, and the classification model's false alarm rate remains high.
[0004] Existing technologies lack constraints on the consistency of the number of factors, resulting in waveform segments of different lengths being directly fed into the same feature extractor. This leads to feature drift due to differences in sampling density, inconsistent distribution between the training and test sets, and limited generalization performance of the model.
[0005] Existing methods typically process all raw waveforms uniformly without pre-screening for data richness of the vibration events to be processed. When the total amount of available historical descriptions is too small, forced segmentation can amplify noise, waste computing power, and reduce real-time performance.
[0006] Existing systems often employ a strategy of first globally stitching and then extracting data, which fails to preserve subtle differences between different short-term modes, resulting in the loss of broadband information and consequently affecting the accurate location and early warning of subsequent abnormal vibration events.
[0007] Therefore, in broadband environmental vibration monitoring, the varying lengths of multiple scattered wave sequences, complex temporal coupling relationships, and the mismatch between traditional fixed-window empirical segmentation methods and actual scattering paths lead to the fragmentation of the same vibration event or the mixing of different events. This results in inter-class crossover and intra-class discrepancies during the feature extraction stage, leading to high false alarm rates and poor generalization performance in subsequent classification models. Furthermore, existing technologies lack a pre-screening mechanism for historical data richness, often forcibly segmenting data even when there is insufficient available descriptive content, amplifying noise and wasting computational power. In addition, the traditional "global stitching followed by extraction" strategy cannot retain subtle differences between different short-term patterns, resulting in the loss of broadband information and making it difficult to achieve high-precision, low-latency real-time identification and risk warning for multiple events such as micro-seismic events, blasting events, and mining tremors simultaneously. Summary of the Invention
[0008] The technical problem this invention aims to solve is that traditional strategies fail to retain subtle differences between different short-term modes, resulting in the loss of broadband information. The goal is to provide a method for identifying multiple scattering waves from broadband environmental vibrations. Based on the number of vibration waveform description data extracted, segmentation is only initiated when the total number of historical descriptions exceeds a threshold, effectively avoiding over-segmentation under low signal-to-noise ratio conditions and improving computational resource utilization. A sliding fixed-length window is used to sequentially segment along the extraction dimension, ensuring that each segment of multiple scattering wave data elements is seamless in time and of consistent length, eliminating boundary misalignment and event fragmentation caused by traditional empirical windows. The number of factors is used as the sole criterion for identification. Based on the grouping criteria, elements of equal length are grouped into the same queue to ensure that the scattering path sampling density is the same within the group, eliminating feature drift caused by length differences and ensuring that the distribution of the training set and the test set are consistent. Significant vibration waveform features are extracted and concatenated for each queue group to form a high-dimensional feature vector that combines multiple short-term modes and broadband information, enabling different types of events such as micro-seismic events, blasting events, and mining tremors to have separable boundaries in low-dimensional space. Finally, an attention-fully connected network is used to quickly classify the fused features, effectively suppressing data interference caused by group control batch operations or malicious injection, ensuring the accuracy, robustness, and traceability of abnormal vibration identification results.
[0009] This invention is achieved through the following technical solution:
[0010] The first aspect of this invention provides a method for identifying multiple scattered waves from broadband environmental vibrations, comprising the following specific steps:
[0011] Obtain a set of historical vibration data for the vibration event to be processed; the set of historical vibration data includes several historical vibration descriptions sorted by vibration period, and each historical vibration description is a segment of multiple scattered wave data;
[0012] Determine the number of vibration waveform description data to be extracted that are associated with the historical vibration data set;
[0013] Based on the number of vibration waveform description data extracted, several consecutive historical vibration description contents are extracted from the historical vibration data set to form corresponding multiple scattering wave data elements;
[0014] Multiple scattering wave data with the same number of factors are divided into the same queue group, and a data queue is set up for each queue group.
[0015] Feature extraction is performed on the data queue to obtain the vibration waveform features of the vibration event to be processed;
[0016] Based on the vibration waveform characteristics of the vibration events to be processed, the data of the vibration events to be processed are divided into data segments, and the data segmentation results are obtained.
[0017] Furthermore, the step of determining the number of vibration waveform description data extractions includes:
[0018] The vibration period of each historical vibration description in the historical vibration data set is statistically analyzed to obtain the period amplitude;
[0019] Obtain the vibration queue of annotated abnormal vibration events within the stated periodic amplitude;
[0020] Statistical analysis was performed on the number of vibration waveform description data covered by each vibration queue to determine the lower limit and upper limit of the number of data to be extracted.
[0021] At least two values located between the lower and upper limits are determined as the number of vibration waveform description data extracted.
[0022] Furthermore, statistical analysis was performed on the number of vibration waveform description data covered by each vibration queue to derive the lower and upper limits of the number of data to be extracted, including:
[0023] Obtain the vibration queue of several annotated abnormal vibration events within the same period amplitude;
[0024] The number of vibration waveform description data covered by each vibration queue is counted to obtain the number set;
[0025] Statistical analysis is performed on the set of numbers to calculate its quantiles or cluster centers, in order to determine the lower limit and upper limit of the number of numbers to be extracted.
[0026] At least two values between the lower and upper limits are used as the number of vibration waveform description data extracted from the historical vibration data set.
[0027] Furthermore, the statistical analysis of the number of vibration waveform description data covered by each vibration queue also includes:
[0028] Obtain the historical vibration data set of the vibration event to be processed, and count the total number N of vibration waveform description data covered by the historical vibration data set;
[0029] The total number N is compared with the preset number of vibration waveform description data extractions M;
[0030] If N≤M, then the probability that the vibration event to be processed is abnormal is lower than the set threshold, and the sub-queue extraction and subsequent abnormal identification steps of the historical vibration data set are skipped.
[0031] If N>M, then the probability that the vibration event to be processed is abnormal is determined to be higher than the set threshold, and the sub-queue extraction, feature extraction and abnormal classification steps continue to be executed.
[0032] Furthermore, the step of extracting the multiple scattering wave data elements includes:
[0033] The first starting point of the historical vibration data set is used as the starting position for extraction, and the extraction dimension associated with the first starting point is determined.
[0034] Based on the extraction dimensions, starting from the extraction start position, historical vibration descriptions are extracted sequentially from the historical vibration data set.
[0035] When the number of extracted vibration waveform description data reaches the number of vibration waveform description data extracted, the extracted historical vibration description content is used as a component factor to construct the multiple scattered wave data corresponding to the extraction start position.
[0036] Based on the extraction dimension, a new extraction starting position is determined from the historical seismic data set, and the steps of extracting historical seismic descriptions sequentially from the historical seismic data set starting from the extraction starting position are returned until the end factor of the multiple scattering wave data is at the second starting point in the historical seismic data set.
[0037] Furthermore, according to the extraction dimension, determining the new extraction starting position from the historical vibration data set includes:
[0038] Obtain the positional difference parameter representing the two consecutive extraction start positions;
[0039] If the terminal factor of the multiple scattering wave data component is in the middle position in the historical vibration data set, then according to the extraction dimension and position difference parameters, the next position of the historical vibration description content with a set interval from the extraction start position of the previous round will be used as the extraction start position of this round.
[0040] Furthermore, the process of assembling the data queue includes:
[0041] Acquire several multiple scattering wave data and determine the number of constituent factors covered by each multiple scattering wave data;
[0042] Multiple scattering data with the same number of factors are grouped into the same queue group;
[0043] For each queue group, based on the composition factors of each multiple scattering wave data in the group, similar queues are selected within the group, and the composition factors of the similar queues are rearranged to form a data queue corresponding to the queue group.
[0044] Furthermore, the step of selecting similar queues within the group and rearranging the constituent factors of the similar queues includes:
[0045] Calculate the similarity of the component factor sequences of any two multiple scattering wave data, and group the multiple scattering wave data with similarity higher than a preset threshold into the same similarity queue;
[0046] The constituent factors within similar queues are reordered according to time sequence or energy weight.
[0047] Furthermore, feature extraction is performed on the data queue to obtain the vibration waveform features of the vibration event to be processed, including:
[0048] Feature extraction was performed on each data queue to obtain significant vibration waveform features corresponding to each data queue.
[0049] The significant vibration waveform features are spliced together in a preset order or with preset weights to form the final vibration waveform features of the vibration issue to be processed.
[0050] Furthermore, based on the vibration waveform characteristics of the vibration events to be processed, the data of the vibration events to be processed is segmented to obtain the data segmentation results, including:
[0051] Based on the attention network, the vibration waveform features of the vibration event to be processed are reconstructed to obtain the reconstructed features of the vibration event to be processed.
[0052] By using the reconstructed features as input to a fully connected network, vibration event identifiers for the vibration events to be processed are obtained.
[0053] Based on the vibration event identifier, the historical vibration description content of the vibration event to be processed is searched to obtain a historical vibration data set covering several historical vibration description contents;
[0054] When the vibration event to be processed is an abnormal vibration event, a risk is identified, an early warning is issued, and risk control operations are performed.
[0055] If the vibration event to be processed is a normal vibration event, and it is determined that there is no risk, no risk control operation will be performed.
[0056] Compared with the prior art, the present invention has the following advantages and beneficial effects:
[0057] By transforming the originally scattered multiple scattering wave records into a set of elements of uniform length and seamless temporal sequence at one time, the classification accuracy is improved. Based on the number of factors, elements of equal length are grouped into the same queue group. Data within the group maintains the same scattering path sampling density, eliminating feature drift caused by length differences and ensuring that the vibration waveform characteristics have a monotonic mapping relationship with the source-propagation path, thus reducing the false identification rate.
[0058] Significant vibration waveform features are extracted from the grouped data queues and then spliced together to form a high-dimensional feature vector that integrates wideband and multiple short-term modes. This enables different types of events, such as micro-seismic events, blasting events, and mining seismic events, to have separable boundaries in low-dimensional space, providing highly discriminative input for subsequent attention-fully connected networks.
[0059] It effectively suppresses data interference caused by group control batch operations, and can still maintain the robustness and traceability of the division results even if abnormal vibration events are maliciously injected or repeatedly uploaded. Attached Figure Description
[0060] To more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be considered as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort. In the drawings:
[0061] Figure 1 This is a flowchart of the method for identifying multiple scattered waves of broadband environmental vibration in an embodiment of the present invention. Detailed Implementation
[0062] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments and accompanying drawings. The illustrative embodiments and descriptions of the present invention are only used to explain the present invention and are not intended to limit the present invention.
[0063] As one possible implementation method, such as Figure 1 As shown, this embodiment provides a method for identifying multiple scattered waves from broadband environmental vibrations, including the following specific steps:
[0064] Obtain a set of historical vibration data for the vibration event to be processed; the set of historical vibration data includes several historical vibration descriptions sorted by vibration period, and each historical vibration description is a segment of multiple scattered wave data;
[0065] Determine the number of vibration waveform description data to be extracted that are associated with the historical vibration data set;
[0066] Based on the number of vibration waveform description data extracted, several consecutive historical vibration description contents are extracted from the historical vibration data set to form corresponding multiple scattering wave data elements;
[0067] Multiple scattering wave data with the same number of factors are divided into the same queue group, and a data queue is set up for each queue group.
[0068] Feature extraction is performed on the data queue to obtain the vibration waveform features of the vibration event to be processed;
[0069] Based on the vibration waveform characteristics of the vibration events to be processed, the data of the vibration events to be processed are divided into data segments, and the data segmentation results are obtained.
[0070] This embodiment acquires a historical seismic data set and extracts historical seismic elements containing multiple scattered wave data from it. This fully utilizes the rich structural information carried by seismic waves propagating in complex media. Several multiple scattered wave data points are extracted from the historical seismic data set. Since each historical seismic event in the set is described by one or more multiple scattered wave data points, it is equivalent to describing the sequence of events from a higher dimension, thus obtaining multiple scattered wave data. A data queue is constructed based on the distribution of the multiple scattered wave data, effectively preserving the spatiotemporal evolution characteristics of the seismic signal under different propagation paths, improving the characterization ability of complex seismic events. By extracting features from the data queue, the seismic waveform characteristics of the seismic event to be processed are obtained, accurately characterizing the key dynamic properties of the seismic signal and enhancing the ability to identify and distinguish the essential characteristics of the seismic event. Finally, the data of the seismic event to be processed is divided based on the extracted seismic waveform features. In practical applications, abnormal vibration events are often controlled in batches, and these data often exhibit similarities, potentially leading to data interference. This embodiment improves the accuracy and reliability of data extraction, achieving high-precision classification and identification of vibration events. It significantly enhances the accuracy and reliability of vibration data processing in areas such as source identification, type discrimination, and event classification. It overcomes the limitations of traditional methods in identifying complex vibration events and their limited feature extraction capabilities. It possesses technical advantages in handling complex seismic environments and improving vibration event identification performance, making it suitable for fields such as earthquake monitoring, structural health monitoring, and underground resource exploration.
[0071] The specific implementation process of this embodiment is as follows:
[0072] Step S1: Obtain a set of historical vibration data for the vibration event to be addressed through vibration monitoring equipment.
[0073] The seismic events can be geological phenomena or man-made events. The historical seismic dataset includes descriptions of several historical seismic events ordered by their seismic cycles. These descriptions are represented by seismic events, and the historical seismic dataset refers to the queue of seismic events that have occurred since the seismic events to be processed.
[0074] In some possible implementations, the server can obtain several historical vibration descriptions of the vibration event to be processed, and construct a historical vibration data set for the event. The server can obtain the historical vibration data set for the event within a set time period, such as the most recent week, the most recent month, or the most recent three days. The server can obtain the historical vibration data set actively or passively.
[0075] Step S2: Generate multiple scattering wave data elements.
[0076] Each historical seismic description in a historical seismic dataset consists of one or more multiplescattering data points. In other words, the union of all multiplescattering data points constitutes the historical seismic dataset, and each multiplescattering data point is a subset of the historical seismic dataset. The queue continuity of the multiplescattering data refers to the sequential continuity of the historical seismic descriptions encompassed within the multiplescattering data. The multiplescattering data elements of historical seismic data are ordered subsets or supersets of historical seismic descriptions that satisfy the queue continuity constraint.
[0077] In one possible implementation, the server can randomly select any two selected vibration waveform descriptions from a set of historical vibration descriptions. Using these two selected vibration waveform descriptions, along with the consecutive vibration waveform descriptions between them, as constituent factors, the server obtains the multiple scattered wave data elements of the historical vibration data. Similarly, by selecting different selected vibration waveform descriptions each time, several different multiple scattered wave data can be obtained.
[0078] In one possible implementation, the server can first determine the number of vibration waveform description data extractions associated with the historical vibration data set, and then extract multiple scattering wave data elements of several historical vibration data from the historical vibration data set according to the number of vibration waveform description data extractions.
[0079] It is understandable that the number of constituent factors in each multiple scattering wave data can be the same or different.
[0080] In one possible implementation, extracting multiple scattering wave data elements from a set of historical vibration data includes: determining the number of vibration waveform description data extractions associated with the set of historical vibration data; and extracting multiple scattering wave data elements from the set of historical vibration data according to the number of vibration waveform description data extractions.
[0081] The number of vibration waveform description data extracted refers to the number of historical vibration description contents required to construct a multiple scattering wave dataset. It is understood that in this application, the number of vibration waveform description data extracted is greater than 1, ensuring that at least two consecutive historical vibration description contents are extracted from the queue. For example, the server can determine the number of vibration waveform description data extracted associated with the historical vibration dataset. Optionally, the server can determine the number of vibration waveform description data extracted based on the total number of historical vibration description contents covered in the historical vibration dataset, wherein the number of vibration waveform description data extracted is less than the total number. Optionally, the server can also count the vibration period of each historical vibration description content in the historical vibration dataset, determine the length of the historical vibration dataset's period, obtain several annotated abnormal vibration events within their respective vibration queues within the specified period length, and finally perform statistical analysis on the number of vibration waveform description data covered by each vibration queue to determine the number of vibration waveform description data extracted associated with the historical vibration dataset.
[0082] After determining the number of vibration waveform description data to be extracted, the server then extracts multiple scattering wave data elements from the historical vibration data set according to this number. Optionally, the server can randomly extract multiple scattering wave data elements from the historical vibration data set according to the number of vibration waveform description data to be extracted, so that the number of component factors covered in each multiple scattering wave data is equal to the number of vibration waveform description data to be extracted. Optionally, the server can also sequentially extract multiple scattering wave data from the historical vibration data set according to the queue and information extraction number, resulting in several multiple scattering wave data arranged in a queue.
[0083] Understandably, the server can determine either a single number of vibration waveform description data extractions or several different numbers. Furthermore, the information extraction methods corresponding to each extraction number can be consistent or inconsistent. In other words, the server can extract multiple scattering wave data elements from the historical vibration data set based on each extraction number of vibration waveform description data, thereby obtaining several multiple scattering wave data points corresponding to each extraction number of vibration waveform description data.
[0084] In the above embodiments, the number of vibration waveform description data extracted associated with the historical vibration data set is first determined, and then, according to the number of vibration waveform description data extracted, several multiple scattering wave data elements of historical vibration data are extracted from the historical vibration data set. This can target multiple scattering wave data of different lengths of historical vibration data sets, which is beneficial to improving the flexibility of the multiple scattering wave identification method based on broadband environmental vibration.
[0085] In one possible implementation, based on the number of vibration waveform description data extracted, multiple scattering wave data elements of several historical vibration data are extracted from the historical vibration data set. This includes: taking a first starting point of the historical vibration data set as the extraction start position and determining the extraction dimension associated with the first starting point; extracting historical vibration description content sequentially from the historical vibration data set according to the extraction dimension, starting from the extraction start position; when the number of extracted vibration waveform description data reaches the number of vibration waveform description data extractions, constructing multiple scattering wave data corresponding to the extraction start position using each extracted historical vibration description content as a component factor; determining a new extraction start position from the historical vibration data set according to the extraction dimension, and returning to the step of extracting historical vibration description content sequentially from the historical vibration data set starting from the extraction start position, until the end factor of the multiple scattering wave data is at a second starting point in the historical vibration data set.
[0086] In the above embodiments, starting from the first starting point of the historical vibration data set, the multiple scattering wave data is extracted sequentially according to the extraction dimension. This allows the sorting of the extracted multiple scattering wave data to correspond to the short-term queues of each multiple scattering wave data, which facilitates the subsequent construction of data queues and improves work efficiency.
[0087] In one possible implementation, determining a new extraction starting position from the historical seismic data set according to the extraction dimension includes: obtaining a position difference parameter representing the difference between two consecutive extraction starting positions; if the terminal factor of the multiple scattering wave data is in the middle position in the historical seismic data set, then determining a new extraction starting position from the historical seismic data set according to the extraction dimension and the difference parameter.
[0088] The difference parameter represents the positional interval between two consecutive extraction start positions; in other words, in this embodiment, the positional interval between two consecutive extraction start positions is consistent. Specifically, the server can obtain the positional difference parameter representing the positional difference between two consecutive extraction start positions. During the extraction of multiple scattering wave data, if the terminal factor of the multiple scattering wave data assembled in the previous round is in the middle position in the historical seismic data set, a new extraction start position is determined from the historical seismic data set according to the extraction dimension and the difference parameter, and the next round of multiple scattering wave data assembly is performed.
[0089] Optionally, when the difference parameter represents a set number of historical vibration descriptions between two consecutive extraction start positions, the server can, according to the extraction dimension, take the next position of the historical vibration descriptions that is a set number of times apart from the extraction start position of the previous round as the extraction start position of the current round.
[0090] In one possible implementation, determining the number of vibration waveform description data extracted associated with the historical vibration data set includes: counting the vibration period of each historical vibration description content in the historical vibration data set to determine the period length of the historical vibration data set; obtaining several annotated abnormal vibration events within their respective period lengths; and performing statistical analysis on the number of vibration waveform description data covered by each vibration queue to determine the number of vibration waveform description data extracted associated with the historical vibration data set.
[0091] The vibration period refers to the historical occurrence time recorded in the historical vibration description. The period length refers to the duration of each vibration period.
[0092] In practical applications, abnormal vibration events exhibit certain patterns in their number. For example, in charitable donation scenarios, donations related to abnormal vibration events tend to be more frequent and involve larger sums within the same timeframe. Based on this, the server can determine the timeframe of the historical vibration dataset by statistically analyzing the vibration cycles of each historical vibration description. Then, it obtains vibration queues for several annotated abnormal vibration events within that timeframe, determines the number of vibration waveform description data points covered by each queue, and finally performs statistical analysis on the number of vibration waveform description data points to determine the number of vibration waveform description data points associated with the historical vibration dataset.
[0093] Furthermore, the server can compare the number of vibration waveform description data points covered by the historical vibration data set with the number of extracted vibration waveform description data points. If the number of vibration waveform description data points is less than or equal to the number of extracted vibration waveform description data points, the probability of the vibration event to be processed being abnormal is considered low, and no further sub-queue extraction is performed. Conversely, if the number of vibration waveform description data points is greater than the number of extracted vibration waveform description data points, the probability of the vibration event to be processed being abnormal is considered high, and further sub-queue extraction is performed to further determine whether the vibration event to be processed is abnormal. Treating vibration events with a number of vibration waveform description data points greater than the number of extracted vibration waveform description data points as vibration events requiring further determination of abnormality can reduce the workload of the abnormal vibration event identification process and improve efficiency.
[0094] In the above embodiments, by statistically analyzing the number of vibration waveform description data covered by the vibration queue within the same period of time for annotated abnormal vibration events, the number of vibration waveform description data extracted that are associated with the historical vibration data set can be determined. This can associate the characteristics of abnormal vibration events in different scenarios, thereby improving the accuracy and reliability of the data segmentation results.
[0095] In one possible implementation, statistical analysis is performed on the number of vibration waveform description data covered by each vibration queue to determine the number of vibration waveform description data to be extracted associated with the historical vibration data set. This includes: performing statistical analysis on the number of vibration waveform description data covered by each vibration queue to determine a lower limit and an upper limit for the number of extractions; and taking at least two extraction numbers that fall within the range represented by the lower limit and the upper limit as the number of vibration waveform description data to be extracted associated with the historical vibration data set.
[0096] The extraction method for each number of vibration waveform description data extractions is consistent.
[0097] After determining the lower limit and upper limit of the number of extractions, the server can use at least two extraction numbers that fall within the range represented by the lower limit and upper limit of the number of extractions as the number of vibration waveform description data to be extracted associated with the historical vibration data set. Then, in the same information extraction method, information is extracted separately according to the number of vibration waveform description data to be extracted.
[0098] In the above embodiments, statistical analysis is performed on the number of vibration waveform description data covered by each vibration queue to determine the lower limit and upper limit of the extraction number. At least two extraction numbers within the range represented by the lower limit and upper limit of the extraction number are used as the vibration waveform description data extraction numbers associated with the historical vibration data set. This can determine several vibration waveform description data extraction numbers based on the numerical characteristics of abnormal vibration events, which is beneficial to improving the accuracy of the content description of each multiple scattering wave data extracted based on each vibration waveform description data extraction number, thereby improving the accuracy and reliability of the vibration event data segmentation results.
[0099] Step S3: According to the data queue corresponding to each multiple scattering wave data.
[0100] Multiple scattering data is used to represent the short-term nature of the vibration event to be processed. Based on this, after the server extracts several multiple scattering data points from the historical vibration data set, it can organize them according to the data queues corresponding to each multiple scattering data point.
[0101] In one possible implementation, according to the data queue corresponding to each multiple scattering wave data, the process includes: selecting similar queues from each multiple scattering wave data based on the constituent factors of each multiple scattering wave data; rearranging the factors of each similar queue to obtain optimized multiple scattering wave data; and following the data queue corresponding to each optimized multiple scattering wave data.
[0102] In the above embodiments, based on the constituent factors of each multiple scattering wave data, similar queues are selected from each multiple scattering wave data, and the factors of each similar queue are rearranged to obtain optimized multiple scattering wave data. By following the data queues corresponding to the optimized multiple scattering wave data, redundant information can be removed as much as possible while ensuring more detailed information differentiation. In this way, the accuracy and reliability of the vibration event data classification results are ensured, while improving the efficiency of data extraction.
[0103] In practical applications, the server can construct a data queue based on each multiple scattering data point. Alternatively, the server can construct separate data queues based on the number of factors covered by each multiple scattering data point, resulting in data queues corresponding to each number of factors. In one possible implementation, constructing data queues based on the number of factors for each multiple scattering data point includes: grouping multiple scattering data points with the same number of factors into the same queue group; and for each queue group, constructing a corresponding data queue based on the multiple scattering data points within that queue group.
[0104] Specifically, the server can determine the number of factors covered by each multiple scattering data point and group multiple scattering data points with the same number of factors into the same queue group. Then, for each queue group, a corresponding data queue is created based on the multiple scattering data points in that queue group.
[0105] In this embodiment, data queues are constructed according to the number of factors covered in the multiple scattering wave data, which can obtain data queues that can represent different short-term modes respectively. This improves the feature representation ability of significant vibration waveform features obtained by feature extraction based on the data queues, and helps to further improve the accuracy and reliability of data segmentation results.
[0106] Step S4: Extract features from the data queue to obtain the vibration waveform features of the vibration event to be processed.
[0107] Understandably, when there are several data queues, the server can extract features from each data queue separately to obtain the significant vibration waveform features of each queue, and then further concatenate them to obtain the vibration waveform features of the vibration event to be processed; alternatively, it can first concatenate the data queues, and then extract features from the concatenated results to obtain the vibration waveform features of the vibration event to be processed. Specific concatenation methods may include summation, etc.
[0108] In one possible implementation, feature extraction is performed on the data queues to obtain the vibration waveform features of the vibration event to be processed, including: performing feature extraction on each data queue separately to obtain the salient vibration waveform features of each data queue; and splicing the salient vibration waveform features to obtain the vibration waveform features of the vibration event to be processed.
[0109] Specifically, the server can extract features from each data queue separately to obtain the salient vibration waveform features for each data queue. Then, the salient vibration waveform features are concatenated to obtain the vibration waveform features of the vibration event to be processed.
[0110] In this embodiment, feature extraction is performed on each data queue to obtain the salient vibration waveform features of each data queue. Then, the salient vibration waveform features are spliced together to obtain the vibration waveform features of the vibration event to be processed. This can ensure that the obtained vibration waveform features can represent multiple short-term patterns, which is beneficial to improving the accuracy and reliability of the data segmentation results determined based on the vibration waveform features.
[0111] Step S5: Based on the vibration waveform characteristics of the vibration event to be processed, the data of the vibration event to be processed is divided to obtain the data division result.
[0112] The data segmentation results can be represented by vibration event categories. These categories can include normal vibration events and abnormal vibration events. For example, the server can input the vibration waveform characteristics of the vibration event to be processed into a vibration event data segmentation network used for data segmentation, and obtain the data segmentation result of the vibration event to be processed based on the output of the vibration event data segmentation network.
[0113] In one possible implementation, the server can obtain a vibration event identifier for the vibration event to be processed, and search for historical vibration descriptions of the vibration event based on the vibration event identifier to obtain a historical vibration data set covering several historical vibration descriptions. If the vibration event to be processed is an abnormal vibration event, a risk is identified, an early warning is issued, and risk control operations are performed; if the vibration event to be processed is a normal vibration event, no risk is identified, and no risk control operations are performed. The process of identification based on each multiple scattering wave data includes: according to the data queue corresponding to each multiple scattering wave data; performing feature extraction on the data queue to obtain the vibration waveform features of the vibration event to be processed; and dividing the vibration event to be processed into data based on the vibration waveform features to obtain the data division results.
[0114] In one possible implementation, the vibration event to be processed is divided into data based on the vibration waveform features of the vibration event to be processed, and the data division result is obtained. This includes: reconstructing the vibration waveform features of the vibration event to be processed based on an attention network to obtain the reconstructed features of the vibration event to be processed; using the reconstructed features as the input of a fully connected network, and determining the data division result of the vibration event to be processed based on the output of the fully connected network.
[0115] In the above embodiments, combining attention networks and fully connected networks to segment the vibration events to be processed can further ensure the accuracy and reliability of the data segmentation results.
[0116] The above specific embodiments further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A method for identifying multiple scattered waves of broadband ambient vibrations, characterized by, The method comprises the following specific steps: obtaining a historical seismic data set of a to-be-processed seismic event; the historical seismic data set comprises a plurality of historical seismic description contents sorted by seismic period, and each historical seismic description content is a piece of multiple scattering wave data; determining a seismic waveform description data extraction number associated with the historical seismic data set; extracting a plurality of continuous historical seismic description contents from the historical seismic data set according to the seismic waveform description data extraction number to form corresponding multiple scattering wave data elements; dividing the multiple scattering wave data elements with the same factor number into the same queue group, and respectively establishing data queues for each queue group; extracting features of the data queues to obtain seismic waveform features of the to-be-processed seismic event; dividing data of the to-be-processed seismic event based on the seismic waveform features of the to-be-processed seismic event to obtain a data division result.
2. The method of claim 1, wherein the broadband ambient vibration is a multiple scattered wave. The step of determining the seismic waveform description data extraction number comprises: counting the seismic periods of each historical seismic description content in the historical seismic data set to obtain a period amplitude; obtaining seismic queues of an abnormal seismic event within the period amplitude; statistically analyzing the number of seismic waveform description data covered by each seismic queue to obtain a lower limit of the extraction number and an upper limit of the extraction number; determining at least two values between the lower limit and the upper limit as the seismic waveform description data extraction number.
3. The method of claim 2, wherein the step of identifying the multiple scattered wave of the broadband ambient vibration is characterized by, The step of statistically analyzing the number of seismic waveform description data covered by each seismic queue to obtain a lower limit of the extraction number and an upper limit of the extraction number comprises: obtaining seismic queues of a plurality of abnormal seismic events within the same period amplitude; counting the number of seismic waveform description data covered by each seismic queue to obtain a number set; statistically analyzing the number set to calculate a quantile value or a clustering center to determine the lower limit of the extraction number and the upper limit of the extraction number; determining at least two values between the lower limit and the upper limit as the seismic waveform description data extraction number associated with the historical seismic data set.
4. The method of claim 2, wherein the broadband ambient vibration is a multiple scattered wave. The step of statistically analyzing the number of seismic waveform description data covered by each seismic queue further comprises: obtaining a historical seismic data set of a to-be-processed seismic event, and counting a total number N of seismic waveform description data covered by the historical seismic data set; comparing the total number N with a preset seismic waveform description data extraction number M; if N≤M, determining that the probability of the to-be-processed seismic event being abnormal is lower than a set threshold, and skipping the steps of subqueue extraction and subsequent abnormality identification of the historical seismic data set; if N>M, determining that the probability of the to-be-processed seismic event being abnormal is higher than the set threshold, and continuing to perform the steps of subqueue extraction, feature extraction and abnormality classification.
5. The method of claim 1, wherein the broadband ambient vibration is a multiple scattered wave. The step of extracting the multiple scattering wave data elements comprises: taking a first starting point of the historical seismic data set as an extraction starting position, and determining an extraction dimension associated with the first starting point; extracting historical seismic description contents from the historical seismic data set successively from the extraction starting position according to the extraction dimension; In a case where the number of extracted seismic waveform description data reaches the seismic waveform description data extraction number, a plurality of scattered wave data corresponding to the extraction start position is constructed by taking each historical seismic description content as a constituent factor; According to the extraction dimension, a new extraction start position is determined from the historical seismic data set, and the step of extracting historical seismic description content from the historical seismic data set sequentially starting from the extraction start position is returned until the end factor of the plurality of scattered wave data is at the second start point in the historical seismic data set.
6. The method of claim 5, wherein the step of identifying the multiple scattered wave of the broadband ambient vibration is characterized by, According to the extraction dimension, the new extraction start position is determined from the historical seismic data set, including: Obtain a position difference parameter representing the difference between the two consecutive extraction start positions; If the end factor of the plurality of scattered wave data constituent factors is at the intermediate position in the historical seismic data set, the next position of the historical seismic description content spaced by a certain number of extraction start positions from the last round is taken as the extraction start position of this round according to the extraction dimension and the position difference parameter.
7. The method of claim 1, wherein the broadband ambient seismic wavefield is characterized by, The data queue is constructed, including: Obtain a plurality of scattered wave data and determine the number of constituent factors covered by each scattered wave data; The scattered wave data with the same number of factors are divided into the same queue group; For each queue group, based on the constituent factors of each scattered wave data in the group, similar queues are selected in the group, and the constituent factors of the similar queues are rearranged to construct a data queue corresponding to the queue group.
8. The method of claim 7, wherein the broadband ambient vibration is a multiple scattered wave. The similar queues are selected in the group, and the constituent factors of the similar queues are rearranged, including: Calculate the similarity of the constituent factor sequences of any two scattered wave data, and classify the scattered wave data with a similarity higher than a preset threshold into the same similar queue; The constituent factors in the similar queue are reordered in time sequence or energy weight.
9. The method of claim 1, wherein the broadband ambient seismic wavefield is characterized by, Feature extraction is performed on the data queue to obtain the seismic waveform features of the to-be-processed seismic event, including: Respectively performing feature extraction on each data queue to obtain a significant seismic waveform feature corresponding to each data queue; The significant seismic waveform features are spliced according to a preset order or weight to form the final seismic waveform feature of the to-be-processed seismic event.
10. The method of claim 1, wherein the broadband ambient seismic wavefield is characterized by, Based on the seismic waveform feature of the to-be-processed seismic event, the to-be-processed seismic event is divided into data to obtain a data division result, including: Based on the attention network, the seismic waveform feature of the to-be-processed seismic event is reconstructed to obtain the reconstruction feature of the to-be-processed seismic event; Taking the reconstruction feature as the input of the full connection network, the seismic event identifier of the to-be-processed seismic event is obtained; Based on the seismic event identifier, the historical seismic description content of the to-be-processed seismic event is searched to obtain a historical seismic data set covering a plurality of historical seismic description contents; In a case where the to-be-processed seismic event is an abnormal seismic event, it is determined that there is a risk, a warning is given and a risk control operation is performed; In a case where the to-be-processed seismic event is a normal seismic event, it is determined that there is no risk, and no risk control operation is performed.