Microearthquake weak signal identification method, system and device and storage medium

Abstract

The invention provides a microseismic signal identification method, system and device and a storage medium, and the method comprises the steps: obtaining a training sample set, and constructing a feature weight calculation model and a classification dual model through employing training samples; acquiring a real-time acquisition signal, and performing feature extraction on the real-time acquisition signal to obtain a plurality of features; calculating the accumulative gradient of each feature, and calculating the weight iteration quantity of each feature according to the accumulative gradient; updating a feature weight calculation model according to the weight iteration quantity, and calculating the weight of each feature; and substituting the weight into the classification dual model, classifying the real-time acquired signals, and judging whether the signals are micro-seismic signals or not. The problems that in the prior art, due to the fact that a reservoir area environment interference source is complex, the micro-seismic signal intensity is weak, and the micro-seismic signal is easily covered by noise in the environment are solved.

Description

Technical Field

[0001] The present invention relates to the field of geological disaster monitoring and early warning, and particularly to a method, system, device and storage medium for identifying weak microseismic signals. Background Art

[0002] Weak microseismic signals induced by rock mass rupture (usually with a Richter magnitude of -3 to +1) are interfered by the complex near-surface rock and soil mass structure and shallow landslides underground during propagation. Their signal intensity is weak and is extremely easy to be masked by the noise in the environment. Therefore, accurately identifying the weak signals generated by rock rupture from numerous noise sources has become a technical problem to be solved urgently in the current field of geological engineering. Summary of the Invention

[0003] Aiming at the deficiencies in the prior art, the present invention provides a method, system, device and storage medium for identifying weak microseismic signals, which solves the problem that the weak microseismic signals induced by the rupture of the rock mass on the reservoir bank slope are extremely easy to be masked by the noise in the environment due to the complex interference sources in the reservoir area environment in the prior art.

[0004] According to an embodiment of the present invention, a method for identifying weak microseismic signals includes:

[0005] Obtaining a training sample set, and using the training samples to construct a feature weight calculation model and a classification dual model;

[0006] Obtaining real-time acquisition signals, and performing feature extraction on the real-time acquisition signals to obtain multiple features;

[0007] Calculating the cumulative gradient of each feature, and calculating the weight iteration amount of each feature according to the cumulative gradient;

[0008] Updating the feature weight calculation model according to the weight iteration amount, and calculating the weight of each feature;

[0009] Substituting the weights into the classification dual model, classifying the real-time acquisition signals, and determining whether they are microseismic signals.

[0010] Preferably, the construction method of the feature weight calculation model includes:

[0011] S1: Classifying all samples in the training sample set to obtain multiple classification sample sets, and at the same time performing feature extraction on all samples;

[0012] S2: Randomly taking out a sample R from all samples, taking out k nearest neighbor samples from the classification sample set to which the sample R belongs, and then taking out k nearest neighbor samples from other classification sample sets;

[0013] S3: Updating the weights of the features of all nearest neighbor samples;

[0014] S4: Repeat steps S2 - S3 until the weights are assigned to the features of all samples.

[0015] Preferably, the method for constructing the classification dual model includes:

[0016] Map the training data to a high - dimensional space to obtain a hyperplane function, and transform the hyperplane function into a Lagrangian function.

[0017] Substitute the partial derivative function of the Lagrangian function and the Gaussian radial basis kernel function into the original NL - SVM model to obtain the classification dual model.

[0018] Preferably, the calculation formula of the classification dual model is as follows:

[0019]

[0020] where K is the Gaussian radial basis kernel function, b is the bias, λ is the Lagrange multiplier, and y i is the corresponding data label.

[0021] Preferably, the method for calculating the cumulative gradient of each feature and calculating the weight iteration amount of each feature according to the cumulative gradient is as follows:

[0022] Set different attenuation coefficients according to the importance of various features of the micro - seismic signal.

[0023] Calculate the moving step of each feature according to the attenuation coefficient.

[0024] Calculate the cumulative gradient of each feature, and calculate the weight iteration amount of each feature according to the cumulative gradient and the moving step.

[0025] Preferably, the update formula of the feature weight calculation model is as follows:

[0026]

[0027] where diff(A, R1, R2) represents the difference between samples R1 and R2 on feature A, Hi represents the i - th nearest neighbor sample in the classification sample set to which sample R belongs, Mi(C) represents the i - th nearest neighbor sample in other classification sample sets, m represents the number of nearest neighbor samples of the same classification as sample R, and P represents the prior probability of samples in the classification sample set or other sample sets to which sample R belongs.

[0028] Preferably, after the feature weight calculation model is updated, take the minimum value of the feature weight calculation model as the weight of the corresponding feature.

[0029] On the other hand, according to an embodiment of the present invention, a micro - seismic signal recognition system is further provided. This system uses the above - mentioned micro - seismic signal recognition method, and includes:

[0030] A data acquisition module, which is used to acquire real-time acquisition signals;

[0031] A modeling module, which is used to construct a feature weight calculation model and a classification dual model;

[0032] A calculation module, which is used to extract features from training samples and real-time acquisition signals, and at the same time calculate the weight iteration amount and update the feature weight calculation model;

[0033] A judgment module, which is used to substitute the weights into the classification dual model, classify the real-time acquisition signals, and judge whether they are microseismic signals.

[0034] On the other hand, according to an embodiment of the present invention, there is also provided a computer device, including a memory and a processor. The memory stores a computer program. When the computer program is executed by the processor, the processor executes the above-mentioned comprehensive evaluation method for a scholar's ability.

[0035] On the other hand, according to an embodiment of the present invention, there is also provided a computer storage medium, storing a computer program. When the computer program is executed by a processor, the processor executes the above-mentioned comprehensive evaluation method for a scholar's ability.

[0036] Compared with the prior art, the present invention has the following beneficial effects:

[0037] Use training samples with clearly known microseismic signals to train the feature weight calculation model and the classification dual model, so that the classification dual model can accurately distinguish whether the input signal is a microseismic signal induced by rock fracture. Then collect real-time acquisition signals, set different weight coefficients according to the importance of various features of the microseismic signal, and update the feature weight calculation model accordingly, so that the feature weight calculation model can assign corresponding weights according to the importance of various features of the microseismic signal, thereby excluding the interference of other environmental signals, highlighting the features of the microseismic signal induced by rock fracture, and avoiding being masked by noise in a complex environment. Then use the trained classification dual model to judge whether the real-time acquisition signal is a microseismic signal. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is a flowchart of microseismic signal recognition according to an embodiment of the present invention.

[0039] Figure 2 It is a flowchart of iterative calculation of microseismic signal feature weights according to an embodiment of the present invention

[0040] Figure 3The micro-seismic monitoring data induced by rock fracture collected in real time in the embodiments of the present invention.

[0041] Figure 4 The micro-seismic signal classification result diagram of the embodiments of the present invention. Specific implementation manners

[0042] The technical solutions in the present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0043] As Figure 1 shown, the embodiments of the present invention propose a method for identifying weak micro-seismic signals, including:

[0044] Obtaining a training sample set, and using the training samples to construct a feature weight calculation model and a classification dual model;

[0045] The training sample set includes a training database of 7 types of signals such as natural earthquakes, hydraulic fracturing micro-seismic, rock slope fractures, wind, lightning, automobile engines, and mine blasting. Six features such as the amplitude, phase, polarity, waveform duration, energy, and frequency of each type of signal are extracted. Here, the training data set {(x i , y i )} is defined, where xi is the input sample, and y i is the corresponding output data label (+1 or -1).

[0046] In order to obtain the weights of different features, a feature weight calculation model is used to calculate the weights of each feature. Different weights are assigned to the features according to the correlation between each feature and the category. Features with weights less than a certain threshold will be removed. The greater the weight of a feature, the stronger the classification ability of the feature. On the contrary, it means that the classification ability of the feature is weaker.

[0047] First, the training samples are classified to obtain multiple classification sample sets. A sample R is randomly selected from the training sample set, and then k nearest neighbor samples of R are found from the sample set of the same category as R. k nearest neighbor samples are found from each sample set of different categories from R, and then the weights of each feature are updated as shown in the following formula:

[0048]

[0049] Among them, diff(A, R1, R2) represents the difference between samples R1 and R2 on feature A, and H i represents the i-th nearest neighbor sample in the classification sample set to which sample R belongs, and M i (C) represents the i-th nearest neighbor sample in other classification sample sets, m represents the number of nearest neighbor samples of the same classification as sample R, and P represents the prior probability of the samples in the classification sample set or other sample sets to which sample R belongs.

[0050] In addition,

[0051] After that, a classification dual model is constructed. At this time, the mapping function (x) is introduced to map the original input data to a higher-dimensional feature space. The goal is to find a separable hyperplane in the feature space. This hyperplane can be expressed as:

[0052] ω·φ(x)+b = 0

[0053] Among them, ω represents the weight, b represents the bias, and the optimization goal is to obtain the minimized weight. Using the Lagrange multiplier method, the hyperplane function and the constraint conditions are transformed into the Lagrangian function:

[0054]

[0055] In the formula, λ is the Lagrange multiplier, i = 1, 2,..., n,

[0056] In the case of non-linearly separable, the original model of NL-SVM is

[0057]

[0058] s.t.C > 0, ζ i ≥ 0, ζ i ≥ 1 - y i (w T φ(x i ) + b)

[0059] Take the partial derivative functions of ω and b in the Lagrangian function and set them equal to zero, and substitute the results into the original model formula of NL-SVM, the dual model of the non-linear support vector machine can be solved:

[0060]

[0061] Then introduce the following Gaussian radial basis kernel function K:

[0062] K(x i , x j ) = φ(x i ) T φ(x j )

[0063]

[0064] σ is a parameter that controls the width of the Gaussian radial basis kernel function. Substitute the kernel function into the dual model of the non-linear support vector machine to obtain the final classification dual model:

[0065]

[0066] Substitute the weight coefficient ω and the kernel function K into the above formula to obtain the decision function as follows:

[0067]

[0068] As Figure 2 shown, the embodiment of the present invention proposes a method for iteratively calculating the feature weights of microseismic weak signals, including:

[0069] Obtain the real-time acquisition signal, and perform feature extraction on the real-time acquisition signal to obtain multiple features;

[0070] Select the microseismic monitoring data of a certain landslide as the real-time acquisition signal, and extract 6 features including amplitude, phase, polarity, waveform duration, energy, and frequency of the monitoring data respectively.

[0071] In order to enable the update of the feature weights to be adaptively adjusted according to different geological structures, different lithologies, and different landslide modes, the update method of the feature weights is improved.

[0072] First, calculate the cumulative gradient of each feature within each sampling period:

[0073]

[0074] Second, calculate the update iteration amount of the feature weights:

[0075]

[0076] Third, calculate the moving step size term:

[0077] S t+1 = ρS t +(1 - ρ)ΔW t

[0078] Update the feature weight calculation model according to the weight iteration amount to obtain the new feature weight W(A) t+1 :

[0079]

[0080] After that, according to the new feature weight calculation model, take the minimum value of the feature weight calculation model as the weight of the corresponding feature, and then substitute it into the classification dual model. If the output result of the classification dual model is 1, it is a microseismic signal; otherwise, if it is -1, it is not a microseismic signal.

[0081] Select the microseismic monitoring data of a certain landslide to test the performance of the method proposed by the present invention. A total of 18 channels of data are selected as the test data set, and 6 channels of data are selected as the validation set. The monitoring data is as Figure 3As shown in the figure. Six features, namely amplitude, phase, polarity, waveform duration, energy, and frequency, of the monitoring data are respectively extracted as the features for NL-SVM classification. Using the algorithm described in the present invention, the signals are divided into two categories, where +1 represents the signal and -1 represents the interference source. Figure 4 The classification results are shown. Among them, the left figure (a) is the result of training the model using the test data, and the right figure (b) is the verification result of the training result of the model using the validation set. Since the microseismic signals induced by rock fractures are very weak, it is difficult to identify the weak microseismic signals if linear or polynomial kernel functions are used for classification. By using the method proposed in the present invention for classification, the microseismic signals can be effectively classified and recognized, and the correct rate reaches 96%.

[0082] On the other hand, an embodiment of the present invention further provides a microseismic signal recognition system, which uses the above-mentioned microseismic signal recognition method, including:

[0083] A data acquisition module, which is used to acquire real-time acquisition signals;

[0084] A modeling module, which is used to construct a feature weight calculation model and a classification dual model;

[0085] A calculation module, which is used to extract features from training samples and real-time acquisition signals, and at the same time calculate the weight iteration amount and update the feature weight calculation model;

[0086] A judgment module, which is used to substitute the weights into the classification dual model, classify the real-time acquisition signals, and judge whether they are microseismic signals.

[0087] On the other hand, an embodiment of the present invention further provides a computer device, including a memory and a processor. The memory stores a computer program, and when the computer program is executed by the processor, the processor executes the above-mentioned comprehensive evaluation method for scholars' capabilities.

[0088] On the other hand, an embodiment of the present invention further provides a computer storage medium, storing a computer program, and when the computer program is executed by a processor, the processor executes the above-mentioned comprehensive evaluation method for scholars' capabilities.

[0089] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the purpose and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

Claims

1. A method for identifying weak microseismic signals, characterized in that: Including: Obtain a training sample set, and use the training samples to construct a feature weight calculation model and a classification dual model; Obtain real-time acquisition signals, and perform feature extraction on the real-time acquisition signals to obtain multiple features; Calculate the cumulative gradient of each feature, and calculate the weight iteration amount of each feature according to the cumulative gradient; Update the feature weight calculation model according to the weight iteration amount, and calculate the weight of each feature; Substitute the weights into the classification dual model, classify the real-time acquisition signals, and determine whether they are microseismic signals.

2. A method for identifying weak microseismic signals according to claim 1, characterized in that: The construction method of the feature weight calculation model includes: S1: Classify all samples in the training sample set to obtain multiple classified sample sets, and at the same time perform feature extraction on all samples; S2: Randomly take out a sample R from all samples, take out k nearest neighbor samples from the classified sample set to which the sample R belongs, and then also take out k nearest neighbor samples from other classified sample sets; S3: Update the weights of the features of all nearest neighbor samples; S4: Repeat steps S2 - S3 until the weights are assigned to the features of all samples.

3. A method for identifying weak microseismic signals according to claim 1, characterized in that: The construction method of the classification dual model includes: Map the training data to a high-dimensional space to obtain a hyperplane function, and transform the hyperplane function into a Lagrangian function; Substitute the partial derivative function of the Lagrangian function and the Gaussian radial basis kernel function into the NL-SVM original model to obtain the classification dual model.

4. A method for identifying weak microseismic signals according to claim 3, characterized in that: The calculation formula of the classification dual model is as follows: Among them, K is the Gaussian radial basis kernel function, b is the bias, λ is the Lagrange multiplier, and y i is the corresponding data label.

5. A method for identifying weak microseismic signals according to claim 2, characterized in that: The method for calculating the cumulative gradient of each feature and calculating the weight iteration amount of each feature according to the cumulative gradient is as follows: Set different attenuation coefficients according to the importance of various features of the microseismic signal; Calculate the moving step of each feature according to the attenuation coefficient; Calculate the cumulative gradient of each feature, and calculate the weight iteration amount of each feature according to the cumulative gradient and the moving step.

6. A method for identifying weak microseismic signals according to claim 5, characterized in that: The update formula of the feature weight calculation model is as follows: Where diff(A, R1, R2) represents the difference between samples R1 and R2 on feature A, Hi represents the i-th nearest neighbor sample in the classified sample set to which sample R belongs, Mi(C) represents the i-th nearest neighbor sample in other classified sample sets, m represents the number of nearest neighbor samples of the same classification as sample R, and P represents the prior probability of samples in the classified sample set to which sample R belongs or other sample sets.

7. A method for identifying weak microseismic signals according to claim 6, characterized in that: After the feature weight calculation model is updated, take the minimum value of the feature weight calculation model as the weight of the corresponding feature.

8. A microseismic weak signal recognition system, characterized in that: This system uses a method for identifying microseismic signals according to any one of claims 1 - 7, including: A data acquisition module, and the data acquisition module is used to acquire real-time acquisition signals; A modeling module, which is used to build a feature weight calculation model and a classification dual model; A calculation module, which is used to extract features from training samples and real-time acquisition signals, and at the same time calculate the weight iteration amount and update the feature weight calculation model; A judgment module, which is used to substitute the weight into the classification dual model, classify the real-time acquisition signal, and judge whether it is a microseismic signal.

9. A computer device, characterized in that: It includes a memory and a processor. The memory stores a computer program. When the computer program is executed by the processor, the processor executes a comprehensive evaluation method for a scholar's ability according to any one of claims 1 to 7.

10. A computer storage medium, characterized in that: Stores a computer program. When the computer program is executed by a processor, the processor executes a comprehensive evaluation method for a scholar's ability according to any one of claims 1 to 7.

Images