Coal rock hydraulic fracturing signal identification method based on MLP neural network model

Abstract

The invention discloses a coal rock hydraulic fracturing signal identification method based on an MLP neural network model, and the method comprises the steps: firstly screening out micro-seismic characteristic parameters which can represent different coal rock lithology, then obtaining the micro-seismic characteristic parameters corresponding to different types of coal rock lithology, and dividing the data into a training set and a verification set; then, an MLP neural network model of a specific structure is constructed, training learning is conducted on the model through the training set, meanwhile, the weight and bias terms of the MLP neural network model are optimized through a BP algorithm, parameters are updated, and after optimization training is finally completed, all micro-seismic data in the hydraulic fracturing process are input into the model. Coal rock lithology labels corresponding to each group of data can be accurately output after model identification, and the coal rock hydraulic fracturing signal distinguishing process is completed; therefore, the lithology and horizon information of hydraulic fracturing coal rock fracture are judged, the search range of microseismic positioning inversion is narrowed, and finally the efficiency and accuracy of microseismic positioning inversion are effectively improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of microseismic monitoring, and particularly relates to a method for identifying coal-rock hydraulic fracturing signals based on an MLP neural network model. Background Art

[0002] Hydraulic fracturing is one of the key technologies to increase the gas permeability of coal seams and improve the gas extraction effect. Microseismic monitoring is an important means for detecting the scope of hydraulic fracturing. Using microseismic monitoring technology, the crack propagation state of underground coal seam hydraulic fracturing can be dynamically observed, the effect of hydraulic fracturing on permeability enhancement can be evaluated, which is of great significance for the accurate implementation of underground coal seam hydraulic fracturing and efficient gas extraction. Microseismic source location is one of the core technologies of microseismic monitoring. Accurately detecting the scope of coal seam hydraulic fracturing is of great significance for scientifically evaluating the effect of hydraulic fracturing on increasing permeability and improving the efficiency of coal seam gas extraction.

[0003] At present, the thickness of coal seams where hydraulic fracturing for permeability enhancement is implemented in most underground coal mines is generally within 5 meters. When implementing hydraulic fracturing in such thin coal seams, it is extremely easy to cause the roof and floor of the coal seam to be affected by fracturing, resulting in the fracturing of the surrounding rock strata. This leads to the fact that the signals collected by microseismic monitoring not only include microseismic signals of coal seam fracturing, but also microseismic signals generated by the fracturing of the surrounding rock strata of the roof and floor. Furthermore, when using microseismic monitoring technology to evaluate the scope of coal-rock hydraulic fracturing for permeability enhancement, due to the mixing of various microseismic signals, the fracture signals of other rocks will interfere with the coal seam hydraulic fracturing signals during the data analysis process, resulting in a low microseismic positioning accuracy.

[0004] Therefore, how to provide a new method that can identify the collected microseismic signals, accurately distinguish the hydraulic fracturing signals corresponding to different types of coal-rock lithologies, and is relatively fast is the research direction required by the present invention. Summary of the Invention

[0005] In view of the problems existing in the above-mentioned prior art, the present invention provides a method for identifying coal-rock hydraulic fracturing signals based on an MLP neural network model, which can identify the collected microseismic signals, accurately distinguish the hydraulic fracturing signals corresponding to different types of coal-rock lithologies, and is relatively fast.

[0006] To achieve the above object, the technical solution adopted by the present invention is: a method for identifying coal-rock hydraulic fracturing signals based on an MLP neural network model, comprising the following steps:

[0007] Step 1: First, screen out microseismic characteristic parameters that can represent different coal-rock lithologies, and then collect coal samples of the coal seam to be hydraulically fractured and corresponding rock samples of its roof and floor strata; each type of sample includes multiple ones. For all samples of different types, simulate the hydraulic fracturing process, obtain the microseismic characteristic parameters of various samples, and assign different labels to different types of samples;

[0008] Step 2: Select a part of the multi - group microseismic characteristic parameters of each sample in Step 1 as the training set, and the rest as the validation set;

[0009] Step 3: Construct an MLP (Multi - Layer Perceptron) neural network model, which consists of an input layer, a hidden layer, and an output layer;

[0010] Step 4: Set the maximum number of iterations, initialize the weights and bias terms of the MLP neural network model, and use the BP algorithm to learn the weight parameters of the MLP neural network model;

[0011] Step 5: Input the training set in Step 2 into the MLP neural network model with the parameters set in Step 4 for iterative training. After reaching the maximum number of iterations, input all the data in the validation set in Step 2 into the trained MLP neural network model for recognition, and then output the label corresponding to each group of data. If the error value is less than the set threshold, go to Step 7; otherwise, go to Step 6;

[0012] Step 6: Use the BP algorithm to optimize and update the weights and bias terms of the MLP neural network model in Step 5, and after completion, repeat Step 5;

[0013] Step 7: Conduct a hydraulic fracturing experiment on the coal - rock sample, collect all the microseismic data during the hydraulic fracturing process, input each group of microseismic data into the MLP neural network model for recognition in turn, so as to output the label corresponding to each group of data, and complete the process of distinguishing coal - rock hydraulic fracturing signals.

[0014] Further, the microseismic characteristic parameters that can represent different coal - rock lithologies in Step 1 include amplitude, rise time, ring count, energy, centroid frequency, and peak frequency. These parameters are obtained by the inventors through research. By combining these parameter combinations with the subsequent model, different types of coal - rock lithologies can be accurately identified.

[0015] Further, the simulation of hydraulic fracturing in Step 1 is carried out in the laboratory, and the same hydraulic fracturing parameters are used for each sample. This can ensure that the microseismic data obtained for each sample is from the same hydraulic fracturing, and guarantee the accuracy of data collection.

[0016] Further, the MLP neural network model in Step 2 has 4 layers, including 1 input layer with 6 neurons, 2 hidden layers with 64 neurons in each layer, using ReLU as the activation function, and 1 output layer with 4 neurons, using Softmax as the activation function. Using these model parameters can effectively ensure the accuracy of model recognition.

[0017] Further, the maximum number of iterations in Step 4 is 3000 times.

[0018] Compared with the prior art, the present invention has the following advantages:

[0019] 1. The present invention constructs an MLP neural network model with a specific structure and combines various microseismic characteristic parameters obtained through research that can represent different coal and rock lithologies. The MLP neural network model with a specific structure has a powerful self-adaptive learning ability, enabling it to train and learn the microseismic characteristic parameters of different coal and rock lithologies. After completion, it extracts the microseismic signal characteristics and performs intelligent identification of coal and rock during hydraulic fracturing, outputs the coal and rock lithologies corresponding to different hydraulic fracturing signals, thereby identifying the layer information of coal and rock fractures during hydraulic fracturing, narrowing the search range of microseismic location inversion, and ultimately effectively improving the efficiency and accuracy of microseismic location inversion.

[0020] 2. The MLP neural network model with a specific structure in the present invention does not require adjusting the hyperparameters of the gradient descent algorithm, which brings the greatest possible convenience to users. Therefore, the usability of this method is relatively high. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 is a schematic diagram of the overall process of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0022] The present invention will be further described below.

[0023] As Figure 1 shown, this embodiment includes the following steps:

[0024] Step 1: First, screen out the microseismic characteristic parameters that can represent different coal and rock lithologies, including amplitude, rise time, ring count, energy, centroid frequency, and peak frequency. These parameters are obtained by the inventor through research. Using the combination of these parameters and the subsequent model can accurately identify different types of coal and rock lithologies. Then, collect the coal samples of the coal seam to be hydraulically fractured and the corresponding rock samples of the roof and floor rock strata, which are 4 types: coal, sandstone, shale, and mudstone. Each type of sample contains multiple samples. The simulated hydraulic fracturing process is carried out on all samples of the 4 types in the laboratory, and the same hydraulic fracturing parameters are used for each type of sample. This can ensure that the microseismic data obtained for each type of sample is obtained by the same hydraulic fracturing, ensuring the accuracy of data collection. Furthermore, obtain the microseismic characteristic parameters of all samples of each type. One sample in each type corresponds to a set of microseismic characteristic parameters, so the same type of sample has multiple sets of microseismic characteristic parameters. And different labels are assigned to different types of samples. The labels for coal, sandstone, shale, and mudstone are "0", "1", "2", and "3" respectively.

[0025] Step 2: Use 500 sets of microseismic characteristic parameters of each type of sample in Step 1 as the training set as shown in Table 1, and the remaining 50 sets of each type of sample as the verification set as shown in Table 2;

[0026] Table 1 Training Set of Microseismic Signals for Hydraulic Fracturing of Coal and Rock

[0027]

[0028]

[0029] Table 2 Validation Set of Microseismic Signals for Hydraulic Fracturing of Coal and Rock

[0030]

[0031] Step 3: Construct an MLP (Multi-Layer Perceptron) neural network model. This model has 4 layers, including 1 input layer with 6 neurons, 2 hidden layers with 64 neurons in each layer, using ReLU as the activation function, and 1 output layer with 4 neurons, using Softmax as the activation function. Using such model parameters can effectively ensure the accuracy of model recognition.

[0032] Step 4: Set the maximum number of iterations of the MLP neural network model to 3000 times, initialize the weights and bias terms, and use the BP algorithm to learn the weight parameters of the MLP neural network model;

[0033] Step 5: Input the training set in Step 2 into the MLP neural network model with the parameters set in Step 4 for iterative training. After reaching the maximum number of iterations, input all the data in the validation set in Step 2 into the trained MLP neural network model for recognition, and output the labels corresponding to each group of data. If the error value is less than the set threshold, go to Step 7; otherwise, go to Step 6;

[0034] Step 6: Use the BP algorithm to optimize and update the weights and bias terms of the MLP neural network model in Step 5, and repeat Step 5 after completion;

[0035] Step 7: The optimized MLP neural network model recognizes the validation sets of various types of samples. Through actual test verification, it is known that the recognition accuracy of this model reaches 96%, indicating that the model of the present invention can accurately identify and distinguish the hydraulic fracturing signals of different types of coal and rock lithologies.

[0036] The above are only the preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art in this technical field, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention.

Claims

1. A method for identifying coal-rock hydraulic fracturing signals based on an MLP neural network model, characterized in that, It includes the following steps: Step 1: First, screen out the microseismic characteristic parameters that can represent different coal and rock lithologies. Then, collect coal samples of the coal seam to be hydraulically fractured and the corresponding rock samples of its roof and floor strata; there are multiple samples of each type. Simulate the hydraulic fracturing process for all samples of different types, obtain the microseismic characteristic parameters of various samples, and assign different labels to different types of samples; Step 2: Select a part of the multiple groups of microseismic characteristic parameters of each sample in Step 1 as the training set, and the rest as the validation set; Step 3: Build an MLP neural network model, which consists of an input layer, a hidden layer, and an output layer; Step 4: Set the maximum number of iterations, initialize the weights and bias terms of the MLP neural network model, and use the BP algorithm to learn the weight parameters of the MLP neural network model; Step 5: Input the training set in Step 2 into the MLP neural network model with the parameters set in Step 4 for iterative training. After reaching the maximum number of iterations, input all the data in the validation set in Step 2 into the trained MLP neural network model for recognition, and output the label corresponding to each group of data. If the error value is less than the set threshold, go to Step 7; Otherwise, go to Step 6; Step 6: Use the BP algorithm to optimize and update the weights and bias terms of the MLP neural network model in Step 5. After completion, repeat Step 5; Step 7: Conduct a hydraulic fracturing experiment on the coal and rock specimens, collect all microseismic data during the hydraulic fracturing process, input each group of microseismic data into the MLP neural network model for recognition in turn, so as to output the label corresponding to the type of each group of data, and complete the process of distinguishing coal and rock hydraulic fracturing signals.

2. The method for identifying coal-rock hydraulic fracturing signals based on the MLP neural network model according to claim 1, wherein The microseismic characteristic parameters that can represent different coal and rock lithologies in Step 1 include amplitude, rise time, ring count, energy, centroid frequency, and peak frequency.

3. The method for identifying coal-rock hydraulic fracturing signals based on the MLP neural network model according to claim 1, wherein The simulation of hydraulic fracturing in Step 1 is carried out in the laboratory, and the same hydraulic fracturing parameters are used for each sample.

4. The method for identifying coal-rock hydraulic fracturing signals based on the MLP neural network model according to claim 1, wherein The MLP neural network model in Step 2 has 4 layers, including 1 input layer with 6 neurons; 2 hidden layers, each with 64 neurons, with ReLU as the activation function; 1 output layer with 4 neurons, with Softmax as the activation function.

5. The coal and rock hydraulic fracturing signal recognition method based on the MLP neural network model according to claim 1, characterized in that The maximum number of iterations in Step 4 is 3000 times.

Images