A phase identification method and system based on a classification algorithm

Abstract

The present application relates to the technical field of data processing, in particular to a kind of phase identification method and system based on classification algorithm, method includes: obtaining original seismic wave data, according to the waveform characteristics of seismic wave data Generation standard seismic wave data, form feature vector set;From the standard seismic wave data in feature vector set Extraction multi-dimensional classification features, based on multi-dimensional classification features Construction feature vector set;Through feature vector set to decision tree training, obtain trained decision tree;Through feature vector set to Naive Bayes classifier training, obtain trained classifier;The trained decision tree and trained classifier are weighted and fused, to obtain fusion model;Based on fusion model to seismic wave data Phase identification, output standard phase and the identification result of abnormal phase;The present application improves the accuracy and efficiency of phase identification.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to a seismic phase recognition method and system based on classification algorithm. BACKGROUND

[0002] In traditional seismic phase recognition methods, artificial experience and simple threshold judgment are often relied on, which is low in efficiency and limited in accuracy. With the increasing amount of seismic monitoring data, the complexity of data is also increasing, and the traditional method is difficult to meet the demand of high-precision recognition of seismic phase. In addition, the seismic wave signal will be disturbed by various factors in the propagation process, such as geological structure, noise, etc., which further increases the difficulty of seismic phase recognition. In order to more accurately and efficiently recognize seismic phase and provide more reliable earthquake warning and analysis, it is of great practical significance to study a seismic phase recognition method and system based on classification algorithm. SUMMARY

[0003] The present application aims to provide a seismic phase recognition method and system based on classification algorithm, which aims to improve the accuracy and efficiency of seismic phase recognition.

[0004] In order to achieve the above-mentioned purpose, the present application provides the following technical solutions:

[0005] In a first aspect, the present application provides a seismic phase recognition method based on classification algorithm, which comprises the following steps:

[0006] S100, obtaining original seismic wave data, generating standard seismic wave data according to the waveform characteristics of the seismic wave data, labeling standard seismic phase and abnormal seismic phase in the standard seismic wave data, and forming a feature vector set;

[0007] S200, extracting multi-dimensional classification features from the standard seismic wave data in the feature vector set, and constructing the feature vector set based on the multi-dimensional classification features;

[0008] S300, training a decision tree through the feature vector set, mining the nonlinear association rules between each feature in the feature vector set by using the decision tree, and obtaining a trained decision tree;

[0009] S400, training a naive Bayes classifier through the feature vector set, calculating the global probability of seismic phase type by using the naive Bayes classifier, and obtaining a trained classifier;

[0010] S500, weighting and fusing the trained decision tree and the trained classifier to obtain a fusion model; based on the fusion model, recognizing the seismic phase of the seismic wave data, and outputting the recognition result of the standard seismic phase and the abnormal seismic phase.

[0011] Optionally, in S100, the original seismic wave data is acquired, standard seismic wave data is generated according to waveform characteristics of the seismic wave data, standard seismic phases and abnormal seismic phases in the standard seismic wave data are labeled, and a complete feature vector set is formed, including:

[0012] In S110, the original seismic wave data is acquired, and the seismic wave data is filtered according to waveform characteristics of the seismic wave data to obtain filtered seismic wave data.

[0013] In S120, the filtered seismic wave data is converted into a unified standard format to obtain standard seismic wave data.

[0014] In S130, standard seismic phases and abnormal seismic phases in the standard seismic wave data are labeled according to waveform characteristics of the standard seismic wave data.

[0015] In S140, the labeled standard seismic wave data is formed into a complete feature vector set.

[0016] Optionally, in S200, multi-dimensional classification features are extracted from the standard seismic wave data in the feature vector set, and the feature vector set is constructed based on the multi-dimensional classification features, including:

[0017] The standard seismic wave data is converted into a structured feature matrix, the feature matrix is subjected to feature discretization processing, a plurality of discrete features are obtained as the feature vector set, and the feature matrix contains dynamic features and statistical features.

[0018] Optionally, in S300, the decision tree is trained by using the feature vector set, the decision tree is used to mine non-linear association rules between features in the feature vector set, and a trained decision tree is obtained, including:

[0019] In S310, an empty decision tree is created, a node of the decision tree represents a feature, a branch represents a value of the feature, a leaf node is used to represent a classification result, and the classification result represents a phase type.

[0020] In S320, an information gain ratio of each feature is calculated, a feature with the largest information gain ratio is selected as a split attribute of a current node, and a plurality of child nodes are generated according to discretization thresholds of the selected split attribute.

[0021] In S330, it is determined whether the generated child nodes satisfy a termination condition, if yes, S330 is executed; otherwise, S320 is executed; the termination condition is that a sample number of the child nodes is less than a set number threshold or a purity reaches a set proportion threshold.

[0022] In S340, a pessimistic error pruning strategy is used to prune the constructed decision tree, a path from a root node to a leaf node forms a classification rule, and a trained decision tree is obtained.

[0023] Optionally, in S400, the training of the Naive Bayes classifier by the feature vector set, the calculation of the global probability of the seismic phase type by the Naive Bayes classifier, and the obtaining of the trained classifier, comprise:

[0024] S410, data preprocessing is performed on the feature vector set, and the feature vector set is divided into a training set and a test set;

[0025] S420, the conditional probability and the prior probability of each feature under different seismic phase categories are calculated according to the feature appearance frequency of each seismic phase in the training set, the posterior probability of each seismic phase type is obtained by multiplying the prior probability and the conditional probability of each feature according to the Bayes formula, and the global probability of the seismic phase type is obtained based on the posterior probability of each seismic phase type;

[0026] S430, the performance of the trained Naive Bayes classifier is evaluated by using the test set until the evaluation index is met, and the trained classifier is obtained.

[0027] Optionally, in S500, the seismic phase recognition of the seismic wave data based on the fusion model and the output of the recognition result of the standard seismic phase and the abnormal seismic phase, comprise:

[0028] S510, the newly collected seismic wave data is input into the fusion model, and the class probability output by the decision tree and the posterior probability output by the Naive Bayes are obtained;

[0029] S520, the class probability and the posterior probability are weighted and fused to obtain a fused probability; and the class with the highest probability is selected as the final classification result.

[0030] Optionally, after S500, the method further comprises:

[0031] A confidence threshold of the abnormal seismic phase is set, when the classification results of the seismic wave data by the trained decision tree and the trained classifier are consistent, and the weighted probability reaches a probability threshold, the seismic wave data is marked as a high-confidence abnormal signal.

[0032] In a second aspect, an embodiment of the present application provides a seismic phase recognition system based on a classification algorithm, the system comprising:

[0033] at least one processor;

[0034] at least one memory for storing at least one program;

[0035] When the at least one program is executed by the at least one processor, the at least one processor implements the method of any one of the above.

[0036] The beneficial effects of the present application are: the present application can fully mine the association rules between data features, comprehensively utilize the advantages of different algorithms, and effectively improve the accuracy and efficiency of phase identification by acquiring original seismic wave data, generating standard seismic wave data, constructing a feature vector set, and then weighting and fusing the trained decision tree and naive Bayes classifier. Compared with the traditional method which relies on artificial experience and simple threshold judgment, the method can better cope with increasingly complex seismic monitoring data and interference factors in the signal propagation process, and provide more reliable basis for earthquake warning and analysis. The present application can quickly and accurately identify standard phases and abnormal phases. BRIEF DESCRIPTION OF DRAWINGS

[0037] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments will be briefly introduced below. Obviously, the drawings in the following description only constitute some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0038] Figure 1 is a flowchart of the phase identification method based on the classification algorithm in the embodiments of the present application;

[0039] Figure 2 is a structural diagram of the phase identification system based on the classification algorithm in the embodiments of the present application. DETAILED DESCRIPTION

[0040] The concept, specific structure and technical effects of the present application will be described clearly and completely below by combining the embodiments and the drawings, so as to fully understand the purpose, scheme and effect of the present application. It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict.

[0041] Referring to Figure 1 , the present application provides a phase identification method based on a classification algorithm, which comprises the following steps:

[0042] S100, acquiring original seismic wave data, generating standard seismic wave data according to the waveform characteristics of the seismic wave data, labeling standard phases and abnormal phases in the standard seismic wave data, and forming a feature vector set;

[0043] S200, extracting multi-dimensional classification features from the standard seismic wave data in the feature vector set, and constructing a feature vector set based on the multi-dimensional classification features;

[0044] S300, training a decision tree through the feature vector set, mining the nonlinear association rules between the features in the feature vector set by using the decision tree, and obtaining a trained decision tree;

[0045] S400, training a Naive Bayes classifier by using the feature vector set, calculating a global probability of the phase type by using the Naive Bayes classifier, and obtaining a trained classifier;

[0046] S500, weighting and fusing the trained decision tree and the trained classifier to obtain a fusion model; and performing phase identification on the seismic wave data based on the fusion model and outputting an identification result of the standard phase and the abnormal phase.

[0047] In the embodiments provided in the application, firstly, according to the waveform characteristics of the original seismic wave data, the data preprocessing is performed to eliminate noise interference and unify the data format, and a feature vector set containing a standard phase (such as S and ScS) and an abnormal phase (such as SdS and PdP) is labeled; then, multi-dimensional classification features are extracted from the preprocessed data, including dynamic and statistical parameters such as wave peak, wave trough, amplitude, period, cross-correlation coefficient and relative travel time difference of the phase, and a feature vector set is constructed; secondly, the principle of "double-mode cooperation and probability weighting" is adopted, the decision tree (C4.5 algorithm) is used to mine the nonlinear association rules between features, and the Naive Bayes classifier is used to calculate the global probability distribution of the phase type, and the two are optimized by a weighted fusion strategy; finally, the fused model is applied to the actual seismic wave data, and the identification result of the standard phase and the abnormal phase is automatically output, and the high-confidence abnormal signal is marked, the rapid screening and classification of the seismic data are completed, and the efficient identification of the complex phase and the accurate detection of the deep structure of the earth are finally realized.

[0048] "Double-mode cooperation and probability weighting": that is, the hierarchical logical relationship between features (such as the coupling effect of travel time difference and amplitude) is analyzed by the decision tree model, and the global distribution law of the phase category (such as the prior probability of SdS) is quantified from the probability point of view by using the Naive Bayes model, and after the two are trained, the prediction results are weighted (the decision tree accounts for 60%, and the Naive Bayes accounts for 40%), so as to balance the feature correlation and statistical robustness, and thus the generalization identification ability of the model to the extra phase is improved.

[0049] In some embodiments, in S100, the original seismic wave data is obtained, the standard seismic wave data is generated according to the waveform characteristics of the seismic wave data, the standard phase and the abnormal phase in the standard seismic wave data are labeled, and a feature vector set is formed, including:

[0050] S110, obtaining the original seismic wave data, filtering the seismic wave data according to the waveform characteristics of the seismic wave data, and obtaining filtered seismic wave data;

[0051] First, collect the raw seismic wave data, which may come from different seismic monitoring stations and different types of seismic events. Then, select appropriate filters based on the frequency characteristics of the seismic wave data and the noise frequency range. For example, for high-frequency noise, a low-pass filter can be used to pass seismic wave data signals below a certain frequency while blocking high-frequency noise; for low-frequency noise, a high-pass filter is used. By setting a reasonable cutoff frequency, the noise component is effectively removed, making the seismic wave data signal clearer.

[0052] S120, convert the filtered seismic wave data into a unified standard format to obtain standard seismic wave data;

[0053] Specifically, check the data format of the raw seismic wave data from different sources. These data may be stored in different file formats, such as text files, binary files, etc., and the storage structure and encoding method of the data may also be different. Establish a unified data format standard. This includes determining the storage file format of the data (such as the commonly used universal seismic data format), specifying the arrangement order of the data (such as arranging in time sequence), the data type (such as integer, floating point), and the data precision, etc. Convert seismic wave data of different formats into a unified standard format. During the conversion process, ensure the integrity and accuracy of the data, and avoid data loss or errors.

[0054] S130, according to the waveform characteristics of the standard seismic wave data, label the standard seismic phase and abnormal seismic phase in the standard seismic wave data;

[0055] Obtain the waveform characteristics of the seismic wave data, including analyzing the amplitude, frequency, period, phase, etc. Parameters of the waveform, understand the characteristic performance of different seismic phases (standard seismic phase and abnormal seismic phase) on these parameters. For example, the standard seismic phase S wave usually has a specific propagation speed and waveform characteristics, and its frequency and amplitude have regular changes within a certain range; abnormal seismic phase SdS may have unique distortion or delay on the waveform, through the research and comparison of a large amount of historical data, summarize these characteristic rules, provide the basis for subsequent processing and labeling.

[0056] Use the waveform characteristics of the standard seismic phase to identify the standard seismic phase (such as S, ScS) in the standard seismic wave data. Rule-based methods can be used to judge according to the propagation time, frequency range, amplitude change, etc. Characteristics of the seismic phase. For example, according to the location of the earthquake and the location of the monitoring station, calculate the theoretical propagation time of the S wave, and find the waveform segment that meets the frequency and amplitude characteristics of the S wave near this time. Label the identified standard seismic phase. The labeling information usually includes the type of seismic phase (such as S or ScS), the start time, the end time, etc. Associate the labeling information with the corresponding seismic wave data to form the labeling data of the standard seismic phase.

[0057] Also according to the waveform characteristics of abnormal seismic phases, abnormal seismic phases (such as SdS and PdP) are identified in seismic wave data. Due to the relatively small number of abnormal seismic phases and more complex characteristics, more prior knowledge and machine learning algorithms may be needed to identify them. For example, a machine learning classifier can be used to classify seismic wave data and determine whether it is an abnormal seismic phase. The identified abnormal seismic phases are labeled, and the labeling content is similar to that of standard seismic phases, including phase type, start time, end time, and the like. Ensure the accuracy and consistency of the labeling to facilitate subsequent model training.

[0058] S140, form a complete feature vector set from the labeled standard seismic wave data.

[0059] In some embodiments, in S200, the multi-dimensional classification features are extracted from the standard seismic wave data in the feature vector set, and the feature vector set is constructed based on the multi-dimensional classification features, including:

[0060] The standard seismic wave data is converted into a structured feature matrix, and the feature matrix is subjected to feature discretization processing to obtain a plurality of discrete features as the feature vector set. The feature matrix includes dynamic features and statistical features.

[0061] Specifically, the dynamic features include peak / trough amplitude, phase period, and waveform energy decay rate; the statistical features include correlation coefficient between adjacent phases, relative travel time difference, and waveform similarity index. The feature matrix is subjected to feature discretization processing, and the continuous features (such as amplitude and travel time difference) are subjected to binning processing to convert them into discrete values.

[0062] In some embodiments, in S300, the decision tree is trained by the feature vector set, the non-linear association rules between the features in the feature vector set are mined by the decision tree, and the trained decision tree is obtained, including:

[0063] S310, create an empty decision tree, the nodes of the decision tree represent features, the branches represent the values of the features, and the leaf nodes represent classification results, and the classification results represent phase types;

[0064] S320, calculate the information gain ratio of each feature, select the feature with the largest information gain ratio as the split attribute of the current node, and generate a plurality of child nodes according to the discretization threshold for the selected split attribute;

[0065] Specifically, the information gain ratio of each feature when dividing the feature vector set at the current node is calculated. The greater the information gain ratio, the greater the influence of the feature on classification, and the more suitable it is as the split feature of the current node. The formula for calculating the information gain ratio (Gain Ratio) of each feature is: information gain ratio = information gain / intrinsic information; the feature with the largest information gain ratio is selected as the split attribute of the current node, which has the strongest ability to distinguish different phase types.

[0066] According to the selected split attribute, the feature vector set is divided to generate nodes and branches of the decision tree. At each node, the feature vector set is divided into different subsets according to the value of the current best split feature, and each subset corresponds to a branch. For example, if the current best split feature is peak amplitude, and a threshold amplitude is set, the samples in the data set with peak amplitude greater than the threshold will be divided into one branch, and the samples with peak amplitude less than the threshold will be divided into another branch.

[0067] S330, determining whether the generated sub-node meets the termination condition, if yes, executing S330; otherwise, executing S320; the termination condition is that the number of samples of the sub-node is less than a set threshold or the purity reaches a set proportion threshold;

[0068] The step S230 is repeatedly executed at each sub-node until the termination condition is met (e.g., the number of samples of the sub-node is less than a set threshold or the purity reaches 95%). The above splitting process is recursively performed on the divided subsets until the stopping condition is met. The stopping condition is that all samples in the subset belong to the same category, at which time a leaf node is created and marked as the category; or there is no remaining feature to be selected, at which time a leaf node is created and marked as the category with the highest frequency in the subset. When the stopping condition is met, the node becomes a leaf node and is marked as the corresponding category (e.g., a standard phase or an abnormal phase). For the selected split attribute, sub-nodes are generated according to the discretization threshold, and each sub-node corresponds to a value range of the feature, for example, cross-correlation coefficient ≤ 0.6 is one category and cross-correlation coefficient > 0.6 is another category.

[0069] S340, pruning the constructed decision tree using a pessimistic error pruning strategy, the path from the root node to the leaf node forms a classification rule, and a trained decision tree is obtained.

[0070] The generated decision tree is pruned to prevent overfitting. The decision tree after pruning is the trained decision tree, which can mine the non-linear association rules between features according to the input feature vector set, and thus classify and predict phases.

[0071] Specifically, a pessimistic error pruning (PEP) strategy is adopted to remove sub-trees that contribute little to classification, preventing overfitting. The path from the root node to the leaf node forms a classification rule, for example: if the travel time difference between S and ScS phases satisfies t_S < t_SdS < t_ScS, and the waveform is highly similar to S phase (correlation coefficient > 0.6), then it is determined as SdS phase. Through the above steps, the nonlinear association rules between features are mined, providing a basis for subsequent fusion with the results of the Naive Bayes classifier.

[0072] The following is an example of decision tree analysis of the hierarchical logical relationship between features;

[0073] Example: Analysis of the coupling effect of travel time difference and amplitude;

[0074] Data samples: containing 600 seismic wave records, of which 150 are labeled as abnormal phases (SdS).

[0075] Core feature correction of SdS phase;

[0076] Position feature: between S wave and ScS wave (t_S < t_SdS < t_ScS);

[0077] Cause feature: reflection phase caused by D" discontinuity anomaly;

[0078] Waveform feature:

[0079] Amplitude: usually 40-70% of S wave;

[0080] Waveform: highly similar to S wave (correlation coefficient > 0.6);

[0081] Example of decision tree splitting process:

[0082]

[0083] Splitting process:

[0084] Root node (600 samples):

[0085] Split feature: t_S < t < t_ScS;

[0086] Split result:

[0087] Satisfy window: 400 samples (including 142 SdS);

[0088] Not satisfied: 200 samples (including 8 SdS), determined as non-SdS;

[0089] First layer splitting (400 samples):

[0090] Split feature: t_rel ∈ [0.2, 0.8];

[0091] Split result:

[0092] Satisfy: 350 samples (including 140 SdS) (excluding 20% area at both ends of the window to reduce boundary effect)

[0093] Not satisfy: 50 samples (including 2 SdS), determined as S wave tail wave;

[0094] Second layer split (350 samples):

[0095] Split feature: A_ratio ∈ [0.4, 0.7];

[0096] Split result:

[0097] Satisfy: 300 samples (including 135 SdS);

[0098] Not satisfy: 50 samples (including 5 SdS), determined as noise;

[0099] Third layer split (300 samples):

[0100] Split feature: CC_S > 0.6;

[0101] Split result:

[0102] Satisfy: 280 samples (including 132 SdS);

[0103] Not satisfy: 20 samples (including 3 SdS), determined as instrument interference;

[0104] Fourth layer split (280 samples):

[0105] Split feature: CC_noise < 0.3;

[0106] Split result:

[0107] Satisfy: 260 samples (including 130 SdS), determined as SdS;

[0108] Not satisfy: 20 samples (including 2 SdS), determined as environmental noise.

[0109] In some embodiments, in S400, the training of the Naive Bayes classifier by the feature vector set, the calculation of the global probability of the seismic phase type by the Naive Bayes classifier, and the obtaining of the trained classifier, include:

[0110] S410, data preprocessing is performed on the feature vector set, and the feature vector set is divided into a training set and a test set;

[0111] It should be noted that the training set is used to train the Naive Bayes classifier, and the test set is used to evaluate the performance of the classifier. First, analyze the data size and distribution of the feature vector set, and according to the characteristics of the data, divide the feature vector set randomly according to a certain proportion (such as 70% as the training set and 30% as the test set). During the division process, the data distribution of the training set and the test set should be as similar as possible to ensure the accuracy of the evaluation results. For example, if the data proportion of the standard seismic phase and the abnormal seismic phase in the feature vector set is 3:1, then the proportion should be maintained as much as possible in the training set and the test set.

[0112] S420, according to the feature frequency of each type of seismic phase in the training set, the conditional probability and the prior probability of each feature under different seismic phase categories are calculated, the prior probability and the conditional probability of each feature are multiplied according to the Bayes formula, and the posterior probability of each seismic phase type is obtained, and the global probability of the seismic phase type is obtained based on the posterior probability of each seismic phase type;

[0113] During the training process, it is assumed that each feature is independent of each other, and according to the feature frequency of each type of seismic phase (standard seismic phase and abnormal seismic phase) in the training set, the conditional probability of each feature under different seismic phase categories is calculated. For example, for a certain feature (such as peak amplitude), the probability of the amplitude value appearing under the standard seismic phase and the abnormal seismic phase is calculated. Taking the peak amplitude as an example, the number of samples in which the peak amplitude is in a certain interval in the S seismic phase sample is counted, and the number of samples in which the peak amplitude is in the same interval in the SdS seismic phase sample is counted, and then the conditional probability of the peak amplitude under the S seismic phase and the SdS seismic phase is calculated. For other features, such as seismic phase period, wave energy decay rate, etc., the conditional probability under different seismic phase types is also calculated in the same way. Then, combined with the prior probability of each type of seismic phase (i.e. the probability of each type of seismic phase appearing in the overall data), for example, if there are 1000 samples in the feature vector set, of which 600 are S seismic phase and 150 are SdS seismic phase, then the prior probability of S seismic phase is 60%, and the prior probability of SdS seismic phase is 15%, and the rest is 25%.

[0114] For a given seismic wave data sample, its feature vector contains peak amplitude, seismic phase period and other features. By multiplying the conditional probability of these features under different seismic phase types with the prior probability of the corresponding seismic phase type, the posterior probability of the sample belonging to different seismic phase types is obtained. The seismic phase type with the maximum posterior probability is the classification result of the Naive Bayes classifier for the sample. Through statistical analysis of different seismic phase types and features in the feature vector set, the global probability distribution of the seismic phase type can be accurately calculated.

[0115] S430, use the test set to evaluate the performance of the trained Naive Bayes classifier until the evaluation index is met, and obtain the trained classifier.

[0116] Specifically, the evaluation indicators include accuracy, recall rate, F1 value, etc. The feature vectors in the test set are input into the trained classifier to obtain the classification prediction results. By comparing with the actual phase type in the test set, the accuracy (the proportion of correctly predicted samples to the total number of samples), recall rate (the proportion of correctly predicted samples of a certain phase type to the actual number of samples of that phase type), and F1 value (the harmonic mean of accuracy and recall rate) are calculated. Through these indicators, the performance of the classifier in identifying phase types is evaluated to determine whether the classifier achieves the expected effect. If the performance is not ideal, the training parameters can be adjusted or retrained.

[0117] After training and evaluation, a trained classifier is obtained, which can calculate the global probability distribution of phase types according to the input feature vectors, providing a basis for subsequent weighted fusion with the results of the decision tree. For example, for a new seismic wave data feature vector, the trained classifier can output the probability that the data belongs to the standard phase and the abnormal phase, thereby providing an important basis for comprehensive judgment of phase types.

[0118] In some embodiments, the prior probability formula is: ; wherein C i represents the i-th phase type, represents the number of samples of phase type in the feature vector set, and N represents the total number of samples.

[0119] In some embodiments, the calculation formula of the posterior probability is:

[0120] ;

[0121] wherein X is the input feature vector, X j is the j-th feature, and C i is the posterior probability of phase type.

[0122] In some embodiments, in S500, the fusion model is used to identify the phase of the seismic wave data, and the identification results of the standard phase and the abnormal phase are output, including:

[0123] S510, input the newly collected seismic wave data into the fusion model to obtain the class probability output by the decision tree and the posterior probability output by the naive Bayes;

[0124] S520, the class probability and the posterior probability are weighted and fused to obtain the fused probability; the class with the highest probability is selected as the final classification result.

[0125] Specifically, first, weight distribution is performed, for example, decision tree weight 60% (focusing on complex correlation between features), Naive Bayes weight 40% (focusing on global statistical distribution). Then, weighted voting is performed, and the fused probability = 0.6 x class probability + 0.4 x posterior probability; finally, the class with the highest probability is selected as the final classification result.

[0126] In some embodiments, after S500, the method further comprises:

[0127] The confidence threshold of the abnormal seismic phase is set, and when the classification results of the trained decision tree and the trained classifier on the seismic wave data are consistent, and the weighted probability reaches the probability threshold, the seismic wave data is marked as a high-confidence abnormal signal.

[0128] In the embodiments provided by the application, the identification results of the standard seismic phase and the abnormal seismic phase can be automatically output, and the high-confidence abnormal signal can be marked. The implementation mode of the automatic marking rule of the high-confidence abnormal signal is as follows:

[0129] The confidence threshold is set: according to the verification result of the test set, the confidence threshold of the abnormal seismic phase (such as SdS) is set, for example, 90%; if the SdS probability output by the model is greater than or equal to 90%, it is marked as a “high-confidence abnormal signal”.

[0130] Multi-model collaborative verification: when the classification results of the decision tree and the Naive Bayes on a certain sample are consistent (both are SdS), and the weighted probability is greater than or equal to 85%, it is directly marked.

[0131] Example:

[0132] Input sample features: travel time difference t_S < t_SdS < t_ScS, amplitude = 0.6, cross-correlation coefficient = 0.6;

[0133] Decision tree output: SdS probability 95%;

[0134] Naive Bayes output: SdS probability 88%;

[0135] Weighted probability: 0.6 x 0.95 + 0.4 x 0.88 = 0.922;

[0136] Determination result: high-confidence abnormal signal (marked as a red warning).

[0137] Corresponding to the method of Figure 1 , with reference to Figure 2 , the embodiments of the application provide a seismic phase identification system based on a classification algorithm, comprising:

[0138] At least one processor;

[0139] at least one memory for storing at least one program;

[0140] When the at least one program is executed by the at least one processor, the at least one processor implements the above method.

[0141] It can be seen that the contents in the above method embodiments are all applicable to the present system embodiment, the present system embodiment specifically implements the same functions as the above method embodiments, and achieves the same beneficial effects as the above method embodiments.

[0142] In addition, the embodiment of the present application further discloses a computer program product or a computer program, which is stored in a computer readable storage medium. The processor of the computer device can read the computer program from the computer readable storage medium, and the processor executes the computer program, so that the computer device executes the above method. Similarly, the contents in the above method embodiments are all applicable to the present storage medium embodiment, the present storage medium embodiment specifically implements the same functions as the above method embodiments, and achieves the same beneficial effects as the above method embodiments.

[0143] Those skilled in the art can understand that all or some of the above disclosed methods, systems can be implemented as software, firmware, hardware and appropriate combinations thereof. Some or all physical components can be implemented as software executed by a processor, such as a central processing unit, a digital signal processor or a microprocessor, or as hardware, or as an integrated circuit, such as an application specific integrated circuit. Such software can be distributed on a computer readable medium, which can include computer storage media (or non-transitory media) and communication media (or transitory media). As known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and can be accessed by a computer. In addition, as known to those skilled in the art, communication media generally includes computer readable instructions, data structures, program modules or other data in modulated data signals such as carrier waves or other transmission mechanisms, and can include any information delivery medium.

[0144] The above is a specific description of the preferred implementation of the present disclosure, but the present disclosure is not limited to the above implementation, and those skilled in the art can make various equivalent modifications or replacements without departing from the spirit of the present disclosure, and these equivalent modifications or replacements are all included in the scope defined by the claims of the present disclosure.

Claims

1. A method for phase identification based on a classification algorithm, characterized in that, The method comprises the following steps: S100, acquiring original seismic wave data, generating standard seismic wave data according to the waveform characteristics of the seismic wave data, labeling standard seismic phases and abnormal seismic phases in the standard seismic wave data, and forming a complete feature vector set; S200, extracting multi-dimensional classification features from the standard seismic wave data in the complete feature vector set, and constructing a feature vector set based on the multi-dimensional classification features; S300, training a decision tree through the feature vector set, mining non-linear association rules between features in the feature vector set by using the decision tree, and obtaining a trained decision tree; S400, training a naive Bayes classifier through the feature vector set, calculating a global probability of a seismic phase type by using the naive Bayes classifier, and obtaining a trained classifier; S500, weighting and fusing the trained decision tree and the trained classifier to obtain a fusion model; performing seismic phase recognition on seismic wave data based on the fusion model, and outputting a recognition result of standard seismic phases and abnormal seismic phases; In S500, the seismic phase recognition on the seismic wave data based on the fusion model and the output of the recognition result of the standard seismic phases and the abnormal seismic phases comprise: S510, inputting newly collected seismic wave data into the fusion model to obtain a class probability output by the decision tree and a posterior probability output by the naive Bayes; S520, weighting and fusing the class probability and the posterior probability to obtain a fused probability; and selecting a class with the highest probability as a final classification result.

2. The method of claim 1, wherein, In S100, the acquisition of the original seismic wave data, the generation of the standard seismic wave data according to the waveform characteristics of the seismic wave data, the labeling of the standard seismic phases and the abnormal seismic phases in the standard seismic wave data, and the formation of the complete feature vector set comprise: S110, acquiring original seismic wave data, filtering the seismic wave data according to the waveform characteristics of the seismic wave data to obtain filtered seismic wave data; S120, converting the filtered seismic wave data into a unified standard format to obtain standard seismic wave data; S130, labeling standard seismic phases and abnormal seismic phases in the standard seismic wave data according to the waveform characteristics of the standard seismic wave data; S140, forming a complete feature vector set from the labeled standard seismic wave data.

3. The method of claim 1, wherein, In S200, the extraction of the multi-dimensional classification features from the standard seismic wave data in the complete feature vector set and the construction of the feature vector set based on the multi-dimensional classification features comprise: Converting the standard seismic wave data into a structured feature matrix, performing feature discretization processing on the feature matrix to obtain a plurality of discrete features as the feature vector set, and the feature matrix containing dynamic features and statistical features.

4. The method of claim 1, wherein, In S300, the training of the decision tree through the feature vector set and the mining of non-linear association rules between features in the feature vector set by using the decision tree to obtain the trained decision tree comprise: S310, creating an empty decision tree, wherein a node of the decision tree represents a feature, a branch represents a value of the feature, a leaf node is used to represent a classification result, and the classification result represents a seismic phase type; S320, calculate the information gain ratio of each feature, select the feature with the largest information gain ratio as the split attribute of the current node, and generate a plurality of child nodes according to the discretization threshold for the selected split attribute; S330, determine whether the generated child nodes meet a termination condition, if yes, perform S340; otherwise, perform S320; the termination condition is that the number of samples of the child nodes is less than a set number threshold or the purity reaches a set proportion threshold; S340, pruning the constructed decision tree using a pessimistic error pruning strategy, the path from the root node to the leaf node forms a classification rule, and a trained decision tree is obtained.

5. The method of claim 1, wherein, In S400, the naive Bayes classifier is trained by the feature vector set, and the global probability of the phase type is calculated by using the naive Bayes classifier, and a trained classifier is obtained, including: S410, data preprocessing is performed on the feature vector set, and the feature vector set is divided into a training set and a test set; S420, according to the feature frequency of each type of seismic phase in the training set, the conditional probability and the prior probability of each feature under different seismic phase categories are calculated, the prior probability and the conditional probability of each feature are multiplied according to the Bayes formula, the posterior probability of each seismic phase type is obtained, and the global probability of the seismic phase type is obtained based on the posterior probability of each seismic phase type; S430, the performance of the trained naive Bayes classifier is evaluated using the test set until the evaluation index is met, and a trained classifier is obtained.

6. The method of claim 1, wherein, After S500, the method further includes: setting a confidence threshold of the abnormal seismic phase, when the classification results of the trained decision tree and the trained classifier on the seismic wave data are consistent, and the weighted probability reaches a probability threshold, the seismic wave data is marked as a high-confidence abnormal signal.

7. A phase identification system based on a classification algorithm, characterized in that, The system includes: at least one processor; at least one memory for storing at least one program; when the at least one program is executed by the at least one processor, the at least one processor implements the method of any one of claims 1 to 6.

Images