Seismic signal property classification method and system
By constructing a seismic signal property classification model through a Bayesian convolutional neural network, the problem of being unable to quantify the confidence of the classification results in the existing technology is solved, the uncertainty of the classification results is determined, and the reliability and accuracy of the model are improved.
Patent Information
- Application Number
- CN202510930288.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-07
- Publication Date
- 2025-09-26
AI Technical Summary
Existing seismic signal classification models are unable to quantify the confidence of the classification results, resulting in the uncertainty of the classification results being unable to be determined, affecting the reliability and accuracy of the model.
A Bayesian convolutional neural network is used to construct a seismic signal property classification model, so that the model parameters follow a certain probability distribution. The seismic data features are extracted through the Bayesian convolution module, flattening module and fully connected module. The Adam optimizer is used to optimize the loss function, and the confidence of the classification result is determined by combining the uncertainty calculation formula.
On the basis of ensuring the accuracy of the classification results, the uncertainty of the classification results is quantified to avoid the model's overconfident misclassification and improve the reliability and accuracy of the model.
Smart Images
Figure CN120703834A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of seismic signal classification, and in particular to a seismic signal property classification method and system. Background Art
[0002] Over the past few decades, seismic observation networks have made significant advances, significantly increasing their monitoring capabilities and recording an increasing number of seismic events. These events include not only natural earthquakes but also non-natural seismic activity caused by human or natural factors, such as volcanic activity, mine blasting, landslides, and train traffic. This diverse record of seismic events provides a rich data set for seismological research, but it also poses new challenges for earthquake monitoring. Failure to exclude non-natural events from earthquake catalogs could lead to inaccurate estimates of b-values, which in turn could affect assessments of the probability of major earthquakes.
[0003] Identifying seismic signal properties in seismic networks has long been a challenging task. Seismologists have proposed various automated methods, which can be roughly divided into two categories: source parameter-based methods and waveform-based methods. The former relies on source information such as location and / or onset time. For example, explosions are often concentrated in known shallow quarries (<2 km) and occur almost exclusively during the day. However, source location information requires precise pre-analysis, which limits its application in real-time processing. In contrast, waveform-based methods rely solely on seismic waveform features extracted manually or automatically from the data by experienced experts. These manually extracted features, such as the P / S amplitude ratio and waveform complexity, are often subjective, and different experts may focus on different signal characteristics. Manually designed features may not fully capture the essential differences between different events, thus affecting the objectivity of classification. Deep learning methods are a typical example of automated waveform feature extraction. Deep learning can directly extract implicit features from the data, rather than using manually defined features. During the training process, features are greedily sought to map input waveforms to output classes. Therefore, automatically extracted waveform features may be more representative than manually defined features. However, although many deep learning networks have been developed to identify the properties of seismic signals, existing networks are unable to quantify the confidence of the classification results, that is, the uncertainty of the classification results, which is a key component of the reliability of the classification model.
[0004] It can be seen that traditional deep learning models used for seismic signal classification are all deterministic models, that is, each parameter in the model is a fixed value. This model has two disadvantages. First, the model is overconfident, resulting in low accuracy of identification results; second, when the model makes an incorrect prediction for an unknown sample, the model does not have other information to remind whether further verification is needed, that is, the uncertainty of the classification results.
[0005] Based on the above problems, there is an urgent need to provide a new seismic signal classification method that can quantify the confidence of the classification results while ensuring the accuracy of the seismic signal classification results, so as to determine the uncertainty of the classification results. Summary of the Invention
[0006] The purpose of this application is to provide a method and system for classifying seismic signal properties, which can quantify the confidence of the classification results on the basis of ensuring the accuracy of the seismic signal classification results, and determine the uncertainty of the classification results.
[0007] To achieve the above objectives, this application provides the following solutions:
[0008] In a first aspect, the present application provides a method for classifying properties of seismic signals, the method comprising:
[0009] Obtain different types of earthquake data and corresponding labels; and build a dataset; labels include natural earthquakes or explosions;
[0010] According to the data set, a seismic signal property classification model is constructed based on a Bayesian convolutional neural network; the Bayesian convolutional neural network includes: a Bayesian convolution module, a flattening module, and a fully connected module;
[0011] According to the seismic data to be classified, the trained seismic signal property classification model is used to determine the classification results and the corresponding uncertainty.
[0012] Optionally, the seismic data includes: seismic waveform, sampling rate, waveform start time and end time, and P-wave arrival time.
[0013] Optionally, the step of obtaining different types of seismic data and corresponding labels further includes:
[0014] Different types of seismic data are preprocessed; the preprocessing includes removing the mean, removing linear trends, removing pinch-outs, bandpass filtering, resampling to a set sampling rate, amplitude normalization, fixing the P-wave first arrival time to a set time, and truncating the waveforms to the same length.
[0015] Optionally, the number of Bayesian convolution modules is 4, the number of flattening modules is 1, and the number of fully connected modules is 2.
[0016] Optionally, the Bayesian convolution module includes: a 1-dimensional Bayesian convolution layer, a ReLU activation function, and a 1-dimensional maximum pooling layer connected in sequence.
[0017] Optionally, the fully connected module includes: a first fully connected layer, a ReLu activation function layer, a second fully connected layer and a Softmax activation function layer connected in sequence.
[0018] Optionally, the loss function of the trained seismic signal property classification model is:
[0019]
[0020] in, is the loss function, P(w) is the prior probability density distribution function of the parameter w of the seismic signal property classification model, P(D|w) is the likelihood function, which is used to characterize the probability of observing data D given the parameter w, q(w|θ) is the distribution function of the approximate posterior probability density function P(w|D), θ is the parameter of the variational distribution, which is used to determine the shape of the variational distribution, KL(q(w|θ)||P(w)) is the divergence between q(w|θ) and P(w), which is used to characterize the similarity between the distributions of q(w|θ) and P(w), and E q(w|θ) [logP(D|w)] is the expectation of logP(D|w) under the distribution of q(w|θ).
[0021] Optionally, during the training process of the seismic signal property classification model, an Adam optimizer is used to optimize the loss function.
[0022] Optionally, the method of determining the classification result and the corresponding uncertainty based on the seismic data to be classified using a trained seismic signal property classification model specifically includes:
[0023] Using the formula Determine the classification results
[0024] Using the formula Determining the uncertainty of the classification result
[0025] Among them, w t is the parameter of the seismic signal property classification model sampling at the t-th prediction, P(y * =c|x * ,w t ) is the sample x * The predicted probability of belonging to category c at the tth prediction, C is the sample x obtained at the tth time * The total number of categories, N is the number of tests.
[0026] In a second aspect, the present application provides a seismic signal property classification system, the seismic signal property classification system comprising:
[0027] The dataset construction module is used to obtain different types of earthquake data and corresponding labels and construct datasets; labels include natural earthquakes or explosions;
[0028] A seismic signal property classification model construction module is used to construct a seismic signal property classification model based on the data set and based on a Bayesian convolutional neural network; the Bayesian convolutional neural network includes: a Bayesian convolution module, a flattening module, and a fully connected module;
[0029] The seismic data classification module is used to determine the classification results and corresponding uncertainties based on the seismic data to be classified using the trained seismic signal property classification model.
[0030] According to the specific embodiments provided in this application, this application has the following technical effects:
[0031] The present application provides a method and system for classifying seismic signal properties, which constructs a seismic signal property classification model based on a Bayesian convolutional neural network. Based on the Bayesian convolutional neural network, the parameters of the seismic signal property classification model are no longer deterministic values, but follow a certain probability distribution. On the basis of ensuring the accuracy of the seismic signal classification results, the confidence of the classification results can be quantified to determine the uncertainty of the classification results. BRIEF DESCRIPTION OF THE DRAWINGS
[0032] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.
[0033] Figure 1 This is a flow chart of a method for classifying properties of seismic signals in one embodiment of the present application;
[0034] Figure 2 Schematic diagram of the Bayesian convolutional neural network structure. DETAILED DESCRIPTION
[0035] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.
[0036] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.
[0037] In an exemplary embodiment, Figure 1As shown, a method for classifying properties of seismic signals is provided, which includes the following S101 to S103.
[0038] S101, obtain different types of seismic data and corresponding labels; and construct a data set; the labels include natural earthquakes or explosions; the seismic data includes: seismic waveform, sampling rate, waveform start time and end time, and P wave arrival time.
[0039] As a specific embodiment, earthquake data and corresponding labels are obtained from data centers such as the National Earthquake Data Center, the IRIS Earthquake Data Center, or the Southern California Seismic Network.
[0040] In order to ensure the training, verification and testing of the seismic signal property classification model, the dataset is divided into training set, verification set and test set in a ratio of 8:1:1.
[0041] In order to improve the accuracy of training the seismic signal property classification model, the seismic data is preprocessed; the preprocessing includes removing the mean, removing the linear trend, removing the pinch-out, bandpass filtering, resampling to a set sampling rate, amplitude normalization, fixing the P wave first arrival time to a set time, and truncating the waveform to the same length.
[0042] S102, based on the data set and the Bayesian convolutional neural network, a seismic signal property classification model is constructed; Figure 2 As shown, the Bayesian convolutional neural network includes: a Bayesian convolution module, a flattening module and a fully connected module;
[0043] Specifically, the number of Bayesian convolution modules is 4, the number of flattening modules is 1, and the number of fully connected modules is 2.
[0044] The Bayesian convolution module includes: a 1-dimensional Bayesian convolution layer, a ReLU activation function, and a 1-dimensional maximum pooling layer connected in sequence. The Bayesian convolution module is used to extract convolution features of seismic data;
[0045] The flattening module is used to flatten the extracted convolution features of the seismic data into a one-dimensional vector.
[0046] The fully connected module includes: a first fully connected layer, a ReLu activation function layer, a second fully connected layer and a Softmax activation function layer connected in sequence.
[0047] The loss function of the trained seismic signal property classification model is:
[0048]
[0049] in, is the loss function, P(w) is the prior probability density distribution function of the parameter w of the seismic signal property classification model, P(D|w) is the likelihood function, which is used to characterize the probability of observing data D given the parameter w, q(w|θ) is the distribution function of the approximate posterior probability density function P(w|D), θ is the parameter of the variational distribution, which is used to determine the shape of the variational distribution, KL(q(w|θ)||P(w)) is the divergence between q(w|θ) and P(w), which is used to characterize the similarity between the distributions of q(w|θ) and P(w), and E q(w|θ) [logP(D|w)] is the expectation of logP(D|w) under the distribution of q(w|θ).
[0050] During the training process of the seismic signal property classification model, the Adam optimizer is used to optimize the loss function so that the loss function reaches the minimum value.
[0051] As a specific embodiment, the Adam optimizer learning rate is set to 0.001, the learning rate decay parameter is set to 0.006, the batch size of the model training is set to 64, the model parameters are set to be uniformly distributed between -0.5 and 0.5 during initialization, the model parameter deviation is set to 5, and the maximum training rounds are set to 100;
[0052] S103, based on the seismic data to be classified, using the trained seismic signal property classification model, determine the classification result and the corresponding uncertainty.
[0053] The parameters of the seismic signal property classification model are no longer fixed values, but follow a certain probability distribution. Therefore, during the test, the number of test rounds is first set to 100 (it can be set by yourself, but in theory, the more times, the longer it takes, but the more accurate the uncertainty of the prediction results). Then, according to the weight parameter distribution in the seismic signal property classification model, the parameters of the seismic signal property classification model are sampled 100 times to obtain 100 models. Then, the data set is sent to the 100 models for testing, and the classification result is obtained based on the average value of the probability value of the 100 test results. The classification result is obtained as follows:
[0054]
[0055] The information entropy of 100 test results is calculated as the uncertainty of the test results. The formula for calculating information entropy is as follows:
[0056]
[0057] Among them, w t is the parameter of the seismic signal property classification model sampling at the t-th prediction, P(y * =c|x* ,w t ) is the sample x * The predicted probability y of belonging to category c at the tth prediction * , C is the sample x obtained for the tth time * The total number of categories, N is the number of tests.
[0058] When testing collapse earthquake waveforms, conventional convolutional neural networks will also misclassify them as earthquakes or explosions. However, based on the classification as earthquakes or explosions, this application outputs a high uncertainty, indicating that the classification result for the collapse sample is unreliable, reflecting the cognitive uncertainty of the earthquake signal property classification model, and avoiding overconfidence in the earthquake signal property classification model. In addition, the earthquake signal property classification model proposed in this application can not only give the classification result but also the uncertainty of the classification result. Experiments show that when the sample is misclassified, a high uncertainty is also given. Then, based on the uncertainty value of the classification result, other methods or manual review can be used to avoid misclassification.
[0059] Based on the same inventive concept, embodiments of the present application also provide a seismic signal property classification system for implementing the aforementioned seismic signal property classification method. The solution provided by this system is similar to the solution described in the aforementioned method. Therefore, the specific limitations in one or more of the following seismic signal property classification system embodiments can be found in the aforementioned limitations on the seismic signal property classification method and will not be further elaborated here.
[0060] In an exemplary embodiment, a seismic signal property classification system is provided, comprising:
[0061] The dataset construction module is used to obtain different types of earthquake data and corresponding labels and construct datasets; labels include natural earthquakes or explosions;
[0062] A seismic signal property classification model construction module is used to construct a seismic signal property classification model based on the data set and based on a Bayesian convolutional neural network; the Bayesian convolutional neural network includes: a Bayesian convolution module, a flattening module, and a fully connected module;
[0063] The seismic data classification module is used to determine the classification results and corresponding uncertainties based on the seismic data to be classified using the trained seismic signal property classification model.
[0064] In an exemplary embodiment, a computer device is provided, which may be a server or a terminal. The computer device includes a processor, a memory, an input / output (I / O) interface, and a communication interface. The processor, the memory, and the I / O interface are connected via a system bus, and the communication interface is connected to the system bus via the I / O interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The I / O interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, it implements a method for classifying seismic signal properties.
[0065] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.
[0066] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, database or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory may include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).
[0067] The databases involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processors involved in the various embodiments provided herein may include, but are not limited to, general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic units, data processing logic units based on quantum computing, and the like.
[0068] In this application, all actions to obtain signals, information or data are carried out in compliance with the relevant data protection laws and policies of the country where they are located and with the authorization given by the owner of the corresponding device.
[0069] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0070] This document uses specific examples to illustrate the principles and implementation methods of this application. The description of the above examples is only intended to help understand the method and core concept of this application. At the same time, for those skilled in the art, based on the concept of this application, there may be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.
Claims
1. A method for classifying properties of seismic signals, characterized in that: The seismic signal property classification method includes: Obtain different types of earthquake data and corresponding labels; and build a dataset; labels include natural earthquakes or explosions; According to the data set, a seismic signal property classification model is constructed based on a Bayesian convolutional neural network; the Bayesian convolutional neural network includes: a Bayesian convolution module, a flattening module, and a fully connected module; According to the seismic data to be classified, the trained seismic signal property classification model is used to determine the classification results and the corresponding uncertainty.
2. The seismic signal property classification method according to claim 1, characterized in that: The seismic data includes: seismic waveform, sampling rate, waveform start time and end time, and P wave arrival time.
3. The seismic signal property classification method according to claim 1, characterized in that: The method of obtaining different types of earthquake data and corresponding labels further includes: Different types of seismic data are preprocessed; the preprocessing includes removing the mean, removing linear trends, removing pinch-outs, bandpass filtering, resampling to a set sampling rate, amplitude normalization, fixing the P-wave first arrival time to a set time, and truncating the waveforms to the same length.
4. The seismic signal property classification method according to claim 1, characterized in that: The number of Bayesian convolution modules is 4, the number of flattening modules is 1, and the number of fully connected modules is 2.
5. The seismic signal property classification method according to claim 1, characterized in that: The Bayesian convolution module includes: a 1-dimensional Bayesian convolution layer, a ReLU activation function and a 1-dimensional maximum pooling layer connected in sequence.
6. The seismic signal property classification method according to claim 1, characterized in that: The fully connected module includes: a first fully connected layer, a ReLu activation function layer, a second fully connected layer and a Softmax activation function layer connected in sequence.
7. The seismic signal property classification method according to claim 1, characterized in that: The loss function of the trained seismic signal property classification model is: in, is the loss function, P(w) is the prior probability density distribution function of the parameter w of the seismic signal property classification model, P(D|w) is the likelihood function, which is used to characterize the probability of observing data D given the parameter w, q(w|θ) is the distribution function of the approximate posterior probability density function P(w|D), θ is the parameter of the variational distribution, which is used to determine the shape of the variational distribution, KL(q(w|θ)||P(w)) is the divergence between q(w|θ) and P(w), which is used to characterize the similarity between the distributions of q(w|θ) and P(w), and E q ( w|θ )[logP(D|w)] is the expectation of logP(D|w) under the distribution of q(w|θ).
8. The seismic signal property classification method according to claim 7, characterized in that: During the training process of the seismic signal property classification model, the Adam optimizer is used to optimize the loss function.
9. The seismic signal property classification method according to claim 7, characterized in that: The method of determining the classification result and the corresponding uncertainty based on the seismic data to be classified using the trained seismic signal property classification model specifically includes: Using the formula Determine the classification results Using the formula Determining the uncertainty of the classification result Among them, w t is the parameter of the seismic signal property classification model sampling at the t-th prediction, P(y * =c|x * ,w t ) is the sample x * The predicted probability of belonging to category c at the tth prediction, C is the sample x obtained at the tth time * The total number of categories, N is the number of tests.
10. A seismic signal property classification system, characterized in that: The seismic signal property classification system includes: The dataset construction module is used to obtain different types of earthquake data and corresponding labels and construct datasets; labels include natural earthquakes or explosions; A seismic signal property classification model construction module is used to construct a seismic signal property classification model based on the data set and based on a Bayesian convolutional neural network; the Bayesian convolutional neural network includes: a Bayesian convolution module, a flattening module, and a fully connected module; The seismic data classification module is used to determine the classification results and corresponding uncertainties based on the seismic data to be classified using the trained seismic signal property classification model.