Ocean seismic signal processing method based on distributed optical fiber vibration sensing

Abstract

The invention provides an ocean seismic signal processing method based on distributed optical fiber vibration sensing, and belongs to the technical field of seismic signal processing. The method comprises the following steps: acquiring a marine seismic signal through a distributed optical fiber vibration sensor; performing data preprocessing on the collected seismic signals to construct a seismic oscillogram data set; according to the preprocessed data, feature extraction is carried out on the preprocessed seismic signals through a U-KAN network to obtain feature data; identifying and classifying the feature data to obtain a P wave and an S wave; and calculating the time difference between the P wave and the S wave according to the identified P wave and S wave, and estimating the seismic source distance and magnitude according to the time difference and the seismic wave propagation speed. The earthquake signal processing automation degree is improved, the earthquake wave arrival time is accurately determined, the marine earthquake monitoring response speed and accuracy are improved, and a scientific basis is provided for prevention and reduction of marine earthquake disasters.

Description

Technical Field

[0001] The present invention relates to the technical field of seismic signal processing, and particularly to a method for processing marine seismic signals based on distributed fiber optic vibration sensing technology. Background Art

[0002] The technical background related to the present invention mainly focuses on the field of marine seismic signal processing, especially the application of distributed fiber optic vibration sensing technology (DAS). As a new means of seismic data acquisition, DAS technology uses existing communication optical fibers as sensors to achieve continuous monitoring of seismic waves, thereby providing high-resolution seismic data over a large area. With its high sensitivity, low-cost maintenance, and strong environmental adaptability, this technology shows great potential in marine seismic exploration.

[0003] Recently, Kolmogorov - Arnold networks (KANs) have been proposed as a new type of neural network structure, which allows learning customized activation values at the edges of the network, thus providing a more transparent view of the network decision-making process. The introduction of KANs provides a new paradigm for neural networks as an alternative to multi-layer perceptron models (MLPs). In fields such as image segmentation, the combination of the U-Net architecture and KAN layers (U-KAN) has shown excellent accuracy and efficiency.

[0004] For the specific application of marine seismic signal processing, this study proposes a method for processing marine seismic signals based on DAS data. The technical background of the present invention covers the application of DAS technology, traditional and modern methods of seismic signal processing, the progress of deep learning in seismic signal processing, and the introduction and optimization of the U-KAN network structure. By combining these technical backgrounds, the present invention aims to provide a new method for processing marine seismic signals to improve the efficiency and accuracy of data processing, and further optimize the performance of the earthquake early warning system. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a method for processing seismic signals based on distributed fiber optic vibration sensing (DAS) technology. The present invention improves the degree of automation of seismic signal processing, accurately determines the arrival time of seismic waves, enhances the response speed and accuracy of marine seismic monitoring, and provides a scientific basis for the prevention and mitigation of marine seismic disasters.

[0006] To solve the above technical problems, the technical solution of the present invention is as follows: In the first aspect, a method for processing seismic signals based on distributed fiber optic vibration sensing technology, the method includes: Collecting marine seismic signals through a distributed fiber optic vibration sensor; Preprocess the collected seismic signals to construct a seismic waveform dataset; Extract features from the preprocessed seismic signals through the U-KAN network based on the preprocessed data to obtain feature data; Identify and classify the feature data to obtain P-waves and S-waves; Calculate the time difference between the P-wave and the S-wave based on the identified P-wave and S-wave, and estimate the earthquake source distance and magnitude based on the time difference and the seismic wave propagation speed; Set the threshold for earthquake early warning, and compare the earthquake source distance and magnitude with the warning threshold to obtain a comparison result; Send an earthquake early warning signal based on the comparison result to achieve earthquake early warning.

[0007] Furthermore, collect ocean seismic signals through a distributed fiber optic vibration sensor, including: Determine the optimization objective and the key parameters of the genetic algorithm; Based on the key parameters of the genetic algorithm, take the positions of the sensor arrangements as individuals of the genetic algorithm, and randomly generate an initial population according to the coding scheme, where each individual represents a sensor arrangement scheme; Based on the initial population, set the fitness function, and select individuals to enter the next generation through roulette wheel selection according to the fitness function; Exchange part of the genes of two excellent individuals at a certain crossover rate to generate new individuals, and randomly change the genes of some individuals at a certain mutation rate to gradually evolve the population and obtain the final solution. When the preset number of iterations is reached, stop the iteration to obtain the final solution; Arrange DAS sensors according to the final solution and start collecting ocean seismic signals.

[0008] Furthermore, the calculation formula of the fitness function is: ; where, represents the fitness; represents the signal quality weight coefficient; represents the signal-to-noise ratio; represents the coverage weight coefficient; represents the area covered by the sensor arrangement scheme; represents the cost weight coefficient; represents the total cost required for sensor arrangement.

[0009] Furthermore, preprocess the collected seismic signals to construct a seismic waveform dataset, including: Perform mean removal, median removal, and trend removal on the collected seismic signal data to obtain preliminarily processed data; Perform band-pass filtering and Gaussian filtering on the preliminarily processed data to obtain the processed data; According to the processed data, obtain the energy of the data within the short window and the long window, and calculate the ratio of the short window to the long window; Perform event clipping and segmentation on the ratio of the short window to the long window to obtain the preprocessed data; Process the preprocessed data to construct a seismic waveform dataset with a resolution of 256×256 pixels.

[0010] Furthermore, according to the preprocessed data, perform feature extraction on the preprocessed seismic signal through the U-KAN network to obtain feature data, including: Construct the U-KAN network architecture; According to the U-KAN network architecture, perform feature extraction on the seismic signal of the preprocessed data to obtain seismic feature data; Decode the feature data to obtain the feature data.

[0011] Furthermore, perform identification and classification on the feature data to obtain P-waves and S-waves, including: According to the U-KAN network, map the feature data to the categories of P-waves and S-waves, and through the forward propagation of the network, obtain the probabilities of P-waves and S-waves at each time point; According to the obtained probabilities of P-waves and S-waves at each time point, set the warning threshold, and convert the probabilities into category labels to obtain P-waves and S-waves.

[0012] Furthermore, according to the identified P-waves and S-waves, calculate the time difference between the P-waves and S-waves, and calculate the earthquake source distance and magnitude according to the time difference and the seismic wave propagation speed, including: According to the identified P-waves and S-waves, extract the arrival times of the P-waves and S-waves, and calculate the difference between the arrival times of the P-waves and S-waves; According to the time difference and the seismic wave propagation speed, through Calculate the earthquake source distance and magnitude, where, is the estimated earthquake magnitude, and are the standard seismic wave amplitude and the observed seismic wave amplitude, 、 、 、 are correction coefficients, and are the propagation speeds of S-waves and P-waves in the seismic wave, and are the times when S-waves and P-waves arrive at the observation point, is the distance from the earthquake source to the observation point, is the energy released by the earthquake, is the starting energy of magnitude saturation, is the deviation of magnitude estimation caused by the magnitude saturation phenomenon.

[0013] In a second aspect, an ocean seismic signal processing system based on distributed fiber optic vibration sensing includes: An acquisition module for collecting ocean seismic signals through a distributed fiber optic vibration sensor; performing data preprocessing on the collected seismic signals to construct a seismic waveform dataset; An identification module for extracting features from the preprocessed seismic signals through a U-KAN network according to the preprocessed data to obtain feature data; identifying and classifying the feature data to obtain P-waves and S-waves; calculating the time difference between the P-waves and S-waves according to the identified P-waves and S-waves, and estimating the earthquake source distance and magnitude according to the time difference and the seismic wave propagation speed; An early warning module for setting a threshold for earthquake early warning, comparing the earthquake source distance and magnitude with the early warning threshold to obtain a comparison result; sending an earthquake early warning signal according to the comparison result to achieve earthquake early warning.

[0014] In a third aspect, a computing device includes: One or more processors; A storage device for storing one or more programs, which when executed by the one or more processors cause the one or more processors to implement the method described above.

[0015] In a fourth aspect, a computer-readable storage medium stores a program that implements the method when executed by a processor.

[0016] The above solution of the present invention has at least the following beneficial effects: By combining preprocessing technology and the U-KAN deep learning model, the present invention realizes a high degree of automation in seismic signal processing, greatly reducing manual intervention and processing time; by accurately extracting and classifying P-waves and S-waves using the U-KAN network, the present invention can accurately determine the arrival times of these two types of seismic waves, providing key time data support for earthquake monitoring and early warning; by calculating the time difference between P-waves and S-waves and developing a corresponding early warning system, effective earthquake early warning information can be provided within a very short time, helping people take timely risk avoidance and disaster reduction measures. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 is a flowchart of a method for processing ocean seismic signals based on distributed fiber optic vibration sensing; Figure 2 is a flowchart of an ocean seismic signal processing system based on distributed fiber optic vibration sensing; Figure 3 is the flowchart of marine seismic signal processing; Figure 4 is the effect diagram of preprocessing of marine seismic signals; Figure 5 is the diagram for making the marine seismic signal dataset; Figure 6 is the diagram for making the data labels of marine seismic signals; Figure 7 is the structure diagram of the U-KAN network; Figure 8 is the data processing flowchart of the U-KAN network; Figure 9 is the graph of the loss (Loss) and intersection over union (IoU) of the U-KAN network training and validation varying with Epoch. Specific implementation manners

[0018] The seismic signal processing method of the distributed fiber optic vibration sensing (DAS) technology will be described in more detail below with reference to the accompanying drawings. The specific implementation manners are shown in the drawings and how to use the seismic signal processing method based on the distributed fiber optic vibration sensing (DAS) technology to achieve seismic monitoring and early warning is described in detail.

[0019] As Figure 1 shown, a marine seismic signal processing method based on distributed fiber optic vibration sensing includes the following steps: Step 11, collecting marine seismic signals through a distributed fiber optic vibration sensor; Step 12, preprocessing the collected seismic signals to construct a seismic waveform diagram dataset; Step 13, extracting features of the preprocessed seismic signals through the U-KAN network according to the preprocessed data to obtain feature data; Step 14, identifying and classifying the feature data to obtain P-waves and S-waves; Step 15, calculating the time difference between the P-waves and S-waves according to the identified P-waves and S-waves, and calculating the earthquake source distance and magnitude according to the time difference and the seismic wave propagation speed; Step 16, setting a threshold for earthquake early warning, and comparing the earthquake source distance and magnitude with the early warning threshold to obtain a comparison result; Step 17, sending an earthquake early warning signal according to the comparison result to achieve earthquake early warning.

[0020] In the embodiments of the present invention, through a distributed fiber optic vibration sensor, marine seismic signals can be comprehensively and continuously collected to ensure the integrity and accuracy of data. The collected seismic signals are preprocessed to effectively remove noise and interference and improve data quality. The U-KAN network is used to extract features from the preprocessed seismic signals, which can accurately capture the key features in the seismic signals and provide strong support for the identification of P-waves and S-waves. The feature data is identified and classified to accurately distinguish P-waves and S-waves. According to the identified P-waves and S-waves, the time difference between them is calculated, and combined with the propagation speed of seismic waves, the distance and magnitude of the earthquake source can be accurately calculated. A reasonable earthquake warning threshold is set, and the distance and magnitude of the earthquake source are compared with the warning threshold to timely judge the potential threat of the earthquake. According to the comparison result, an earthquake warning signal is sent in real time, which can timely notify relevant personnel and institutions before the earthquake occurs and take necessary risk avoidance measures to reduce the losses caused by the earthquake disaster.

[0021] In a preferred embodiment of the present invention, the above step 11 may include: Step 111, determining the optimization objective and the key parameters of the genetic algorithm; Step 112, according to the key parameters of the genetic algorithm, taking the positions of the sensor arrangements as the individuals of the genetic algorithm, and randomly generating an initial population according to the coding scheme, where each individual represents a sensor arrangement scheme; Step 113, setting a fitness function according to the initial population, and selecting individuals to enter the next generation through roulette wheel selection according to the fitness function; Step 114, exchanging part of the genes of two excellent individuals at a certain crossover rate to generate new individuals, and randomly changing the genes of some individuals at a certain mutation rate to gradually evolve the population to obtain the final solution. When the preset number of iterations is reached, the iteration is stopped to obtain the final solution; Step 115, arranging DAS sensors according to the final solution and starting to collect marine seismic signals.

[0022] In the embodiments of the present invention, optimizing the sensor arrangement positions through the genetic algorithm can determine the optimal sensor arrangement scheme; arranging DAS sensors through the final solution generated by the genetic algorithm can make full use of the performance of the sensors and improve the acquisition efficiency and accuracy of seismic signals; by optimizing the sensor arrangement scheme, the performance of the sensors can be maximally utilized, the acquisition efficiency and accuracy of marine seismic signals are improved, an intelligent and automated processing process is realized, which has strong adaptability and flexibility, and at the same time saves resources and costs.

[0023] In a specific embodiment of the present invention, the specific steps include: Step 111, determine the optimization goal to ensure that the sensor layout scheme can collect marine seismic signals most effectively; determine the key parameters of the genetic algorithm, including population size, crossover rate, mutation rate, and number of iterations; Step 112, regard the positions of sensor layout as individuals of the genetic algorithm, each individual represents a specific sensor layout scheme, and according to the coding scheme, randomly generate an initial population. Each individual in the population is a possible representation of the sensor layout scheme; Step 113, based on the initial population, set the fitness function to evaluate the quality of each individual, which is usually closely related to the optimization goal. Select individuals to enter the next generation through roulette wheel selection, that is, select individuals with higher fitness as parents to generate more advantageous offspring.

[0024] Step 114, exchange part of the genes of two excellent individuals at a certain crossover rate to generate new individuals, randomly change the genes of some individuals at a certain mutation rate to introduce new genetic information, and repeat the crossover and mutation operations to gradually evolve the population and continuously approach the optimal solution; Step 115, when the preset number of iterations is reached or other stopping conditions are met, stop the iteration to obtain the final solution. Connect the DAS device to the existing optical cable in the sea area to collect and save seismic signal data in real time. According to the seismic station catalog and known seismic activity information, screen out the signals related to marine earthquakes as the original data for subsequent processing.

[0025] In a preferred embodiment of the present invention, in the above step 113, the calculation formula of the fitness function is: ; where, represents the fitness; represents the signal quality weight coefficient; represents the signal-to-noise ratio; represents the coverage weight coefficient; represents the area covered by the sensor layout scheme; represents the cost weight coefficient; represents the total cost required for sensor layout.

[0026] In the embodiments of the present invention, by introducing a fitness function, multiple key factors such as signal quality, coverage range, and cost are comprehensively considered, and the sensor layout scheme is comprehensively optimized; this ensures that the sensor layout can not only meet the requirements of seismic signal acquisition but also control costs and improve overall efficiency; the fitness function includes a signal quality weight coefficient and a signal-to-noise ratio. By selecting a layout scheme with a higher signal-to-noise ratio, the clarity and accuracy of the collected seismic signals can be effectively improved; the fitness function also considers a coverage range weight coefficient and the area covered by the sensor layout scheme. By introducing a cost weight coefficient and the total cost required for sensor layout, the present invention can fully consider cost factors when optimizing the sensor layout; each weight coefficient in the fitness function can be adjusted according to actual needs, making the optimization strategy more flexible and adjustable.

[0027] In a preferred embodiment of the present invention, step 12 may include: Step 121, perform mean removal, median removal, and trend removal processing on the collected seismic signal data to obtain preliminarily processed data; Step 122, perform band-pass filtering and Gaussian filtering processing on the preliminarily processed data to obtain processed data; Step 123, obtain the energy of the data within a short window and a long window according to the processed data, and calculate the ratio of the short window to the long window; Step 124, perform event clipping and segmentation processing on the ratio of the short window to the long window to obtain preprocessed data; Step 125, process the preprocessed data to construct a seismic waveform map dataset with a resolution of 256×256 pixels.

[0028] In the embodiments of the present invention, through mean removal, median removal, and trend removal processing, interference components such as DC offset, outliers, and long-term trends in the seismic signals are effectively removed, improving the purity of the signals; by performing band-pass filtering and Gaussian filtering processing on the preliminarily processed data, the effective frequency bands in the seismic signals can be retained while suppressing noise and interference, enhancing the characteristics of the signals; by calculating the energy of the data within a short window and a long window and obtaining their ratio, the energy characteristics of seismic events can be accurately extracted, which helps to distinguish seismic events of different scales and intensities; by performing event clipping and segmentation processing on the ratio of the short window to the long window, the effective events in the seismic signals can be effectively separated from noise and interference.

[0029] In a specific embodiment of the present invention, the specific steps include: Step 121, perform mean removal on the collected seismic signal data to eliminate the DC offset in the signal, perform median removal to remove outliers or prominent points in the signal, and implement detrending to eliminate the long-term trend or drift in the signal, obtaining the preliminarily processed data; Step 122, perform band-pass filtering on the preliminarily processed data, select a frequency range of 1 - 10 Hz to retain the effective frequency band in the seismic signal while suppressing the noise and interference outside the frequency band, perform Gaussian filtering to further smooth the signal and reduce high-frequency noise, obtaining the processed data; Step 123, according to the processed data, set the time lengths of the short window and the long window, calculate the energy value of the data within the short window, calculate the energy value of the data within the long window, and obtain the ratio of the short-window and long-window energy values as a characteristic index of the seismic event; Step 124, according to the ratio of the short window and the long window, set a threshold or criterion for identifying seismic events. When the ratio exceeds the threshold, a seismic event has occurred. Perform event cropping, extract the signal segment related to the event, and perform segmentation on the cropped event signal segment for subsequent feature extraction, classification, or further analysis, obtaining the preprocessed data; Step 125, according to the preprocessed data, starting from the first channel of the data array, select 100 consecutive channels in sequence each time. Merge the data of these 100 channels to obtain a new merged seismic waveform data. If the number of remaining channels is less than 100, these remaining data can be discarded, or special processing can be performed according to specific requirements. Convert the merged seismic waveform data into an image. During the conversion process, adjust the data to a resolution of 256×256 pixels. An appropriate image processing library can be used to draw the seismic waveform as an image and save it in a common image format. Continuously repeat the above steps until all 7091 channels in the data array are traversed, thereby constructing a complete seismic waveform diagram dataset.

[0030] In a preferred embodiment of the present invention, the above step 13 may include: Step 131, construct a U-KAN network architecture; Step 132, according to the U-KAN network architecture, perform feature extraction on the seismic signal of the preprocessed data to obtain seismic feature data; Step 133, decode the feature data to obtain the feature data.

[0031] In a preferred embodiment of the present invention, the above step 14 may include: Step 141, according to the U-KAN network, map the feature data to the categories of P-waves and S-waves, and through the forward propagation of the network, obtain the probabilities of P-waves and S-waves at each time point; Step 142: Set an early warning threshold according to the obtained P-wave and S-wave probabilities at each time point, convert the probabilities into class labels to obtain P-waves and S-waves.

[0032] In the embodiment of the present invention, through the U-KAN network, the feature data can be effectively mapped to the classes of P-waves and S-waves. The forward propagation mechanism of the network can accurately calculate the P-wave and S-wave probabilities at each time point. According to the P-wave and S-wave probabilities at each time point, an early warning threshold is set. When the probability exceeds the threshold, the probability can be immediately converted into a class label to realize the real-time identification and early warning of P-waves and S-waves; As an advanced neural network architecture, the U-KAN network has strong learning ability and adaptability. It can process various complex seismic signal features and maintain stable identification performance under different geological conditions and noise environments. Accurately identifying P-waves and S-waves is an important basis for seismic signal analysis. Through the accurate identification results provided by the U-KAN network, a reliable data basis is provided for subsequent seismic event location, phase analysis, seismic source parameter estimation, etc.

[0033] In a specific embodiment of the present invention, the specific steps include: Step 141: Input the preprocessed seismic signal feature data into the U-KAN network, perform forward propagation processing on the input feature data, and map the feature data to the classes of P-waves and S-waves through the calculations of each layer of the network; At the output layer of the network, obtain the P-wave and S-wave probabilities at each time point; Step 142: Set an early warning threshold to identify whether the seismic signal belongs to a P-wave or an S-wave. Compare the P-wave and S-wave probabilities at each time point with the set early warning threshold, convert the class labels, and obtain the P-wave and S-wave information in the seismic signal according to the converted class labels; Use the obtained P-wave and S-wave information to analyze the seismic event; Perform phase identification on the identified P-waves and S-waves, and output the analysis results and identification results in the form of a report or a chart.

[0034] Step 151: According to the identified P-waves and S-waves, extract the arrival times of the P-waves and S-waves, and calculate the difference between the arrival times of the P-waves and S-waves; Step 152: According to the time difference and the seismic wave propagation speed, through calculate the seismic source distance and magnitude, where is the estimated seismic magnitude, and are the standard seismic wave amplitude and the observed seismic wave amplitude, , , , are correction factors, and are the propagation speeds of S-waves and P-waves in seismic waves, and are the arrival times of S-waves and P-waves at the observation point, is the distance from the earthquake source to the observation point, is the energy released by the earthquake, is the starting energy of magnitude saturation, is the deviation amount of magnitude estimation caused by the magnitude saturation phenomenon.

[0035] In the embodiments of the present invention, by accurately identifying P-waves and S-waves and extracting their arrival times, the present invention can calculate the difference between the arrival times of P-waves and S-waves. This time difference is a key parameter for calculating the distance and magnitude of the earthquake source. When calculating the earthquake magnitude by analyzing the distance and magnitude of the earthquake source, a correction coefficient and the consideration of the magnitude saturation phenomenon are introduced, which can more accurately reflect the actual situation of magnitude calculation during large earthquakes and avoid the deviation of magnitude calculation caused by the magnitude saturation phenomenon. By considering the magnitude saturation phenomenon, the present invention can provide a more accurate earthquake magnitude estimation result, providing more powerful support for earthquake disaster assessment and emergency response; the distributed fiber optic vibration sensing technology has the advantages of high sensitivity and wide monitoring range, and is suitable for marine earthquake monitoring. Combining this technology with the analysis method of the arrival time difference between P-waves and S-waves can improve the accuracy and real-time performance of marine earthquake monitoring.

[0036] By real-time monitoring and analyzing marine earthquake signals, earthquake events can be detected in a timely manner, providing strong guarantee for marine disaster early warning and emergency response.

[0037] In a specific embodiment of the present invention, the specific steps include: Step 151, according to the identified positions of P-waves and S-waves, extract their arrival times and record the arrival time P of the P-wave and the arrival time of the S-wave.

[0038] Step 152, according to the time difference and the propagation speed of seismic waves, calculate the distance and magnitude of the earthquake source through the formula.

[0039] In a preferred embodiment of the present invention, the above step 16 may include: Step 161, according to historical earthquake data, set the warning thresholds for the distance and magnitude of the earthquake source; Step 162, compare the calculated distance and magnitude of the earthquake source with the set warning thresholds. If the distance of the earthquake source ≤ the distance threshold and the magnitude ≥ the magnitude threshold, trigger an earthquake warning signal to obtain the comparison result.

[0040] In the embodiments of the present invention, by introducing the setting of warning thresholds for the source distance and magnitude of earthquakes, as well as a warning trigger mechanism for comparing the calculation results with the thresholds, the accuracy and timeliness of earthquake warnings have been significantly improved, the reliability and practicality of the warning system have been enhanced, the resource allocation and emergency response strategies have been optimized, and the public's earthquake safety awareness and self-help ability have been improved. These beneficial effects make the present invention have broad application prospects and practical value in the field of marine earthquake monitoring and warning.

[0041] In a specific embodiment of the present invention, the specific steps include: Step 161, according to the analysis results of historical earthquake data, combined with the theories and experiences of seismology, set reasonable warning thresholds for the source distance and magnitude of earthquakes. The warning threshold for the source distance should be set according to the specific conditions of the marine earthquake monitoring area and the propagation characteristics of seismic waves, and the warning threshold for the magnitude should be set according to the magnitude distribution law of historical earthquakes and the degree of impact of earthquakes on the marine environment and human activities, so as to ensure the accuracy and effectiveness of the warning signal; Step 162, compare the calculated source distance and magnitude with the set warning thresholds. If the source distance is less than or equal to the distance threshold and the magnitude is greater than or equal to the magnitude threshold, it is determined that the earthquake event may have a greater impact on human activities or the marine environment. After triggering the earthquake warning signal, the warning information is quickly transmitted to relevant departments and the public through the communication network so that they can take corresponding measures in a timely manner.

[0042] In a preferred embodiment of the present invention, the above step 17 may include: sending an earthquake warning signal according to the comparison result to achieve earthquake warning.

[0043] In the embodiments of the present invention, special emphasis is placed on the function of sending an earthquake warning signal according to the comparison result to achieve earthquake warning. This design brings the following significant beneficial effects: By comparing the calculated source distance and magnitude with the set warning thresholds in real time, based on the distributed fiber optic vibration sensing technology, marine earthquake signals can be monitored in real time, and comparative analysis can be carried out in combination with the warning thresholds set by historical earthquake data; This analysis method combining real-time monitoring and historical data improves the accuracy and reliability of the warning system and reduces the risks of false alarms and missed alarms; By timely issuing earthquake warning signals, the public can learn about the possibility of an earthquake occurring and thus take necessary self-help measures to reduce the threat to personal safety caused by the earthquake; The earthquake warning signal sending mechanism not only provides technical support for earthquake monitoring and warning, but also provides valuable data resources for earthquake scientific research; By analyzing the triggering situation of warning signals, the distribution law of earthquake events, etc., seismologists can more deeply understand the mechanism and law of earthquake occurrence and provide a scientific basis for earthquake prediction and disaster prevention and mitigation.

[0044] Such asFigure 2 As shown in the figure, an embodiment of the present invention further provides an ocean seismic signal processing system 20 based on distributed fiber optic vibration sensing, including: An acquisition module 21, configured to collect ocean seismic signals through a distributed fiber optic vibration sensor; perform data preprocessing on the collected seismic signals to construct a seismic waveform dataset; An identification module 22, configured to extract features from the preprocessed seismic signals through a U-KAN network according to the preprocessed data to obtain feature data; identify and classify the feature data to obtain P-waves and S-waves; calculate the time difference between the P-waves and S-waves according to the identified P-waves and S-waves, and estimate the earthquake source distance and magnitude according to the time difference and the seismic wave propagation speed; An early warning module 23, configured to set a threshold for earthquake early warning, and compare the earthquake source distance and magnitude with the early warning threshold to obtain a comparison result; issue an earthquake early warning signal according to the comparison result to achieve earthquake early warning.

[0045] The ocean seismic signal processing method based on distributed fiber optic vibration sensing as shown in the figure, the specific implementation steps include: 1. Collect seismic signals: Connect the DAS device to the existing optical cable in the sea area, and collect and save seismic signal data in real time. According to the seismic station catalog and known earthquake activity information, screen out the signals related to ocean earthquakes as the original data for subsequent processing.

[0046] 2. Data preprocessing: Perform preprocessing operations such as mean removal, median removal, and trend removal on the collected original seismic signals to eliminate DC offset, outliers, and low-frequency trends, and highlight the short-term dynamic changes of the seismic signals; apply band-pass filtering technology to select a frequency range of 1-10 Hz to retain the main features of the seismic signals and remove high-frequency noise and low-frequency interference; use the STA / LTA technology to calculate the ratio of the short-term energy to the long-term energy of the signal, and distinguish seismic signals from background noise by setting a threshold to effectively suppress noise; use Gaussian filtering technology to further smooth the data, reduce random noise, and retain the main features of the signal at the same time.

[0047] 3. Production of the dataset: a. Channel extraction and merging: Starting from the first channel of the data array, select 100 consecutive channels in sequence each time. Merge the data of these 100 channels to obtain a new merged seismic waveform data. If the number of remaining channels is less than 100, these remaining data can be discarded, or special processing can be performed according to specific requirements.

[0048] b. Construct seismic waveform diagrams: Convert the merged seismic waveform data into images. During the conversion process, adjust the data to a resolution of 256×256 pixels. You can use a suitable image processing library (such as matplotlib) to draw the seismic waveforms as images and save them in common image formats (PNG, JPEG, etc.).

[0049] c. Repeat the operation: Continuously repeat steps a and b until all 7091 channels in the data array are traversed, thereby constructing a complete dataset of seismic waveform diagrams.

[0050] 4. Seismic signal recognition: The present invention uses the KAN network as the core framework for seismic signal recognition and utilizes its powerful feature learning ability to accurately classify and identify seismic signals. The specific process is as follows: a. Input layer data processing The input is a 256×256 seismic waveform diagram. Since it is a single-channel (grayscale image), the input feature dimension is set to [batch_size, 1, 256, 256]. Here, batch_size represents the number of samples input into the network at one time, 1 represents a single channel, and 256×256 is the spatial dimension of the image. These preprocessed seismic waveform diagrams enter the KAN network as initial data.

[0051] b. Encoder part processing (ConvLayer stage) encoder1: Receive input features of [batch_size, 1, 256, 256]. Process them using ConvLayer. Inside ConvLayer, first perform a convolution operation through nn.Conv2d to convert the single-channel input into kan_input_dim / / 8 channels. The convolution kernel is 3x3, and the padding value is set to 1. The purpose of padding is to maintain the size of the feature map and avoid excessive reduction in size during the convolution process. Then use nn.BatchNorm2d for batch normalization to normalize the features, accelerate the network convergence speed, and stabilize the training. Then use nn.ReLU(inplace=True) as the activation function to set negative values to 0, introducing non-linearity so that the network can learn more complex features. Finally, after the max pooling operation (pooling kernel is 2x2) in ConvLayer, the output feature dimension becomes [batch_size, kan_input_dim / / 8, 128, 128]. The pooling operation reduces the spatial dimension of the feature map while retaining the main feature information.

[0052] encoder2: The input feature dimension is [batch_size, kan_input_dim / / 8, 128, 128]. Similar to encoder1, it again uses ConvLayer for feature transformation and processing, changing the number of channels from kan_input_dim / / 8 to kan_input_dim / / 4, and also performing convolution, normalization, and activation operations. After max pooling (2x2), the output feature dimension is [batch_size, kan_input_dim / / 4, 64, 64].

[0053] encoder3: The input feature dimension is [batch_size, kan_input_dim / / 4, 64, 64]. It continues to be processed using ConvLayer, converting the number of channels to kan_input_dim, and successively completing convolution, normalization, and activation operations. After max pooling (2x2), the output feature dimension is [batch_size, kan_input_dim, 32, 32].

[0054] c. Processing in the TokenizedKAN stage patch_embed3: The input feature dimension is [batch_size, kan_input_dim, 32, 32]. It is processed using the PatchEmbed module. The nn.Conv2d inside this module performs convolution projection operations, projecting the input features of kan_input_dim channels onto embed_dims[1] channels, and at the same time changing the feature map size by adjusting the convolution kernel size, stride, and padding parameters. The stride affects the downsampling degree of the feature map. Finally, the output feature dimension is [batch_size, embed_dims[1], 16, 16].

[0055] block1: Both the input and output feature dimensions are [batch_size, embed_dims[1], 16, 16]. It is processed using KANLayer in KANBlock. KANLayer is a custom module that contains multiple KANLinear layers or nn.Linear layers, as well as depthwise separable convolution and nn.Dropout layers. KANLayer performs linear transformations through layers such as fc1, fc2, fc3, etc., uses depthwise separable convolution layers such as dwconv_1, dwconv_2, dwconv_3, etc. to perform spatial processing on the features, and finally uses nn.Dropout for random inactivation operations to effectively prevent overfitting.

[0056] patch_embed4: The input feature dimension is [batch_size, embed_dims[1], 16, 16]. Apply PatchEmbed again for feature projection, project the features of embed_dims[1] channels to embed_dims[2] channels, and adjust the feature map size. The output feature dimension becomes [batch_size, embed_dims[2], 8, 8].

[0057] block2: The input and output feature dimensions are [batch_size, embed_dims[2], 8, 8]. Similar to block1, use the KANLayer in KANBlock to process the features, and extract more advanced features through a series of linear and depthwise separable convolution operations.

[0058] d. Decoder part processing decoder1: The input feature dimension is [batch_size, embed_dims[2], 8, 8]. Use D_ConvLayer for processing. D_ConvLayer is similar to ConvLayer, including convolution, normalization, and activation operations. Enlarge the feature map size through upsampling (such as bilinear interpolation) and perform feature transformation at the same time. The output feature dimension is [batch_size, embed_dims[1], 16, 16].

[0059] dblock1: The input and output feature dimensions are [batch_size, embed_dims[1], 16, 16]. Use the KANLayer in KANBlock to process the features, and further extract and transform the features through complex operations inside the KANLayer.

[0060] decoder2: The input feature dimension is [batch_size, embed_dims[1], 16, 16]. Apply D_ConvLayer for upsampling and feature transformation, further enlarge the feature map size and adjust the number of channels. The output feature dimension is [batch_size, embed_dims[0], 32, 32].

[0061] dblock2: The input and output feature dimensions are [batch_size, embed_dims[0], 32, 32]. Use the KANLayer in KANBlock to process the features and continuously extract and transform the features.

[0062] decoder3: The input feature dimension is [batch_size, embed_dims[0], 32, 32]. Use D_ConvLayer for upsampling and feature transformation, adjust the number of channels to embed_dims[0] / / 4, and at the same time expand the feature map size. The output feature dimension is [batch_size, embed_dims[0] / / 4, 64, 64].

[0063] decoder4: The input feature dimension is [batch_size, embed_dims[0] / / 4, 64, 64]. Use D_ConvLayer for upsampling and feature transformation, adjust the number of channels to embed_dims[0] / / 8, and expand the feature map size. The output feature dimension is [batch_size, embed_dims[0] / / 8, 128, 128].

[0064] decoder5: The input feature dimension is [batch_size, embed_dims[0] / / 8, 128, 128]. Through D_ConvLayer for upsampling and feature transformation, keep the number of channels embed_dims[0] / / 8 unchanged, and restore the feature map size to be close to the input size. The output feature dimension is [batch_size, embed_dims[0] / / 8, 256, 256].

[0065] e. Output layer processing: final: The input feature dimension is [batch_size, embed_dims[0] / / 8, 256, 256]. Use nn.Conv2d for the final convolution operation to transform the features into the final output. The output feature dimension is [batch_size, 2, 256, 256], where 2 represents two image segmentation labels used to distinguish P-waves and S-waves, thus realizing the classification and recognition of seismic signals.

[0066] 5. Earthquake early warning: Calculate the early warning time based on the arrival time difference between P-waves and S-waves, and issue an early warning before the arrival of destructive seismic waves, providing valuable time for taking emergency response measures.

[0067] The above examples only show the preferred implementation schemes of the present invention, aiming to illustrate the working principle and processing method of the present invention, rather than limiting the protection scope of the present invention. The protection scope of the present invention should be determined according to the content clearly defined in the claims. Any equivalent replacement or technical improvement of the technical solutions described in the claims within the framework of the technical concept of the present invention should be considered to fall within the protection scope of the present invention.

Claims

1. A method for processing marine seismic signals based on distributed optical fiber vibration sensing, characterized in that: The method comprises: Collect marine seismic signals through distributed optical fiber vibration sensors; Perform data preprocessing on the collected seismic signals to construct a seismic waveform data set; According to the preprocessed data, the preprocessed seismic signal is subjected to feature extraction through the U-KAN network to obtain feature data; Identify and classify characteristic data to obtain P waves and S waves; Based on the identified P-wave and S-wave, the time difference between the P-wave and the S-wave is calculated, and the earthquake source distance and magnitude are estimated based on the time difference and the propagation speed of the seismic wave; Set the threshold of earthquake warning, and compare the earthquake source distance and magnitude with the threshold of warning to obtain the comparison result; An earthquake warning signal is issued based on the comparison results to achieve earthquake early warning.

2. The method for processing marine seismic signals based on distributed optical fiber vibration sensing according to claim 1, characterized in that: Collect marine seismic signals through distributed optical fiber vibration sensors, including: Determine the optimization objectives and key parameters of genetic algorithms; According to the key parameters of the genetic algorithm, the positions of the sensor arrangements are taken as individuals of the genetic algorithm, and the coding scheme is randomly generated into an initial population, wherein each individual represents a sensor arrangement scheme; According to the initial population, a fitness function is set, and individuals are selected to enter the next generation through roulette according to the fitness function; Two excellent individuals are exchanged with some of their genes at a certain crossover rate to generate new individuals. The genes of some individuals are randomly changed at a certain mutation rate to make the population gradually evolve and obtain the final solution. When the preset number of iterations is reached, the iteration is stopped and the final solution is obtained. Based on the final solution, DAS sensors are deployed to start collecting marine seismic signals.

3. The method for processing marine seismic signals based on distributed optical fiber vibration sensing according to claim 1, characterized in that: The calculation formula of the fitness function is: ； in, Indicates fitness; Represents the signal quality weight coefficient; represents the signal-to-noise ratio; represents the coverage weight coefficient; Indicates the area covered by the sensor layout scheme; represents the cost weight coefficient; Represents the total cost required for sensor deployment.

4. The method for processing marine seismic signals based on distributed optical fiber vibration sensing according to claim 1, characterized in that: The collected seismic signals are preprocessed to construct a seismic waveform dataset, including: The collected seismic signal data is processed by removing the mean, median and trend to obtain preliminary processed data; Performing bandpass filtering and Gaussian filtering on the initially processed data to obtain processed data; According to the processed data, the energy of the data in the short window and the long window is obtained, and the ratio of the short window to the long window is calculated; The ratio of the short window to the long window is subjected to event clipping and segmentation processing to obtain preprocessed data; The preprocessed data are processed to construct a seismic waveform dataset with a resolution of 256×256 pixels.

5. The method for processing marine seismic signals based on distributed optical fiber vibration sensing according to claim 1, characterized in that: According to the preprocessed data, the preprocessed seismic signal is feature extracted through the U-KAN network to obtain feature data, including; Build U-KAN network architecture; According to the U-KAN network architecture, feature extraction is performed on the seismic signal of the preprocessed data to obtain data on seismic features; The characteristic data is decoded to obtain the characteristic data.

6. The method for processing marine seismic signals based on distributed optical fiber vibration sensing according to claim 1, characterized in that: Identify and classify characteristic data to obtain P waves and S waves, including: According to the U-KAN network, the characteristic data is mapped to the categories of P waves and S waves, and the probability of P waves and S waves at each time point is obtained through forward propagation of the network; According to the probability of obtaining P waves and S waves at each time point, the warning threshold is set, and the probability is converted into a category label to obtain P waves and S waves.

7. The method for processing marine seismic signals based on distributed optical fiber vibration sensing according to claim 1, characterized in that: Based on the identified P-wave and S-wave, the time difference between the P-wave and the S-wave is calculated, and the earthquake source distance and magnitude are calculated based on the time difference and the propagation speed of the seismic wave, including: Based on the identified P wave and S wave, the arrival time of the P wave and the S wave are extracted, and the difference between the arrival time of the P wave and the S wave is calculated; According to the time difference and the propagation speed of seismic waves, Calculate the earthquake focal distance and magnitude, where: is the estimated earthquake magnitude, and is the standard seismic wave amplitude and the observed seismic wave amplitude, , , , is the correction factor, and is the propagation speed of S and P waves in seismic waves, and is the time when the S-wave and P-wave arrive at the observation point, is the distance from the earthquake source to the observation point, It is the energy released by the earthquake. is the initial energy of magnitude saturation, It is the amount of deviation in magnitude estimation due to the phenomenon of magnitude saturation.

8. A marine seismic signal processing system based on distributed optical fiber vibration sensing, the system implementing the method according to any one of claims 1 to 7, characterized in that: include: An acquisition module, used for collecting marine seismic signals through distributed optical fiber vibration sensors; Perform data preprocessing on the collected seismic signals to construct a seismic waveform data set; The recognition module is used to extract features of the preprocessed seismic signal through the U-KAN network according to the preprocessed data to obtain feature data; Identify and classify the characteristic data to obtain P waves and S waves; calculate the time difference between the P waves and S waves based on the identified P waves and S waves, and estimate the earthquake source distance and magnitude based on the time difference and the propagation speed of seismic waves; The early warning module is used to set the threshold of earthquake early warning, and compare the source distance and magnitude with the early warning threshold to obtain a comparison result; an earthquake early warning signal is issued according to the comparison result to realize earthquake early warning.

9. A computing device, characterized in that include: one or more processors; A storage device for storing one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors implement the method as claimed in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a program, which implements the method according to any one of claims 1 to 8 when executed by a processor.

Images