A method for predicting rock porosity based on simultaneous inversion of drilling data from small samples.

Abstract

This method provides a rock porosity prediction method based on inversion of simultaneous drilling data from small samples. Simultaneous drilling information, including parameters such as torque M, thrust force F, rotational speed N, thrust velocity V, drill pipe amplitude A, and vibration acceleration a, is collected during the drilling process in the laboratory or field. Different types of simultaneous drilling parameter data, after multiple denoising, are input into a BP-GA model. Discrete point removal, data enhancement, and iterative calculations are performed to generate a new simultaneous drilling dataset. Time-frequency domain feature maps of simultaneous drilling parameters for rocks with different porosities are input into a VG-CNN convolutional neural network prediction model. Through training and learning, an inversion model between the time-frequency domain features of simultaneous drilling parameters and porosity is obtained, ultimately achieving inversion prediction for rock porosity based on real-time simultaneous drilling parameters in the field. This method can rapidly, accurately, and quantitatively predict rock porosity based on small amounts of simultaneous drilling data.

Description

[Technical Field]

[0001] This invention relates to a method for predicting the inversion of rock porosity in subsurface engineering, and more specifically, to a method for predicting rock porosity based on simultaneous inversion of drilling data from small-scale samples. [Background technology]

[0002] Porosity is an important indicator representing physical and mechanical parameters such as rock strength, elastic modulus, and permeability. Rapid measurement of rock porosity in underground tunnel construction is crucial for selecting support methods and determining support parameters. Conventional methods for measuring rock porosity required processing rock samples after core extraction and conducting laboratory test analyses. In fields such as oil drilling, analyzing log data allows for detailed characterization and quantitative inversion of the porosity of oil and gas storage layers. However, for shallow tunnel construction (subways, tunnels, underground tunnels, etc.), logging and log work are economically costly and cumbersome. Therefore, there is a need to propose a quantitative, rapid, and economical in-situ measurement method for rock porosity.

[0003] Currently, simultaneous drilling measurement technology is gradually becoming a hot spot in the field of intelligent exploration for rockwork. Simultaneous drilling measurement technology utilizes the characteristics of changes in simultaneous drilling parameters during the drilling process to reflect the mechanical properties of surrounding rocks, enabling the identification of lithology and structural features of geological formations. It effectively compensates for the time delay of conventional measurement methods and is a convenient and rapid in-situ measurement method that does not affect on-site construction work. Related research results in the field of simultaneous drilling measurement have rudimentarily verified the feasibility of reversing rock strength and identifying rock body structural surfaces based on simultaneous drilling parameters. However, due to the complexity of the drill bit and rock cutting process, and the small amount of effective data and susceptibility to external environmental influences in the on-site collection process of simultaneous drilling parameters, predicting the reversal of rock porosity using simultaneous drilling parameters presents significant difficulties. Existing literature has few methods for predicting the reversal of porosity of rocks surrounding tunnels, and no method for accurately predicting the reversal of rock porosity using simultaneous drilling data from small samples has been reported. Therefore, it is necessary to establish a scientific and rational method for predicting the reversal of rock porosity during simultaneous drilling. [Overview of the project] [Problems that the invention aims to solve]

[0004] To address the technical shortcomings described above, the object of the present invention is to provide a rock porosity prediction method based on reversal of simultaneous drilling data of small samples, which can rapidly and quantitatively predict rock porosity by utilizing simultaneous drilling data of small samples in a laboratory or field drilling process. [Means for solving the problem]

[0005] To solve the above technical problems, the present invention adopts the following technical solution.

[0006] This invention provides a method for predicting rock porosity based on simultaneous inversion of drilling data from small samples. Step S1 involves collecting simultaneous drilling information during the drilling process in a laboratory or underground tunnel using an excavator, and obtaining key simultaneous drilling parameters such as torque M, thrust force F, rotational speed N, thrust speed V, drill pipe amplitude A, and vibration acceleration a. Step S2 involves performing multiple denoising on different types of collected simultaneous drilling parameters, inputting the denoised simultaneous drilling parameter values ​​into the "BP-GA" model, and obtaining a new simultaneous drilling parameter derived dataset. Step S3 involves analyzing the change curves of different types of simultaneous drilling parameters with respect to drilling time t within a new simultaneous drilling parameter dataset, obtaining time-frequency domain diagrams of simultaneous drilling parameters, and obtaining a time-frequency domain image library of different types of simultaneous drilling parameters through the data enhancement function of a neural network. Step S4 includes constructing a convolutional neural network prediction model for rock porosity, training and learning a time-frequency domain feature image library of simultaneous drilling parameters in the drilling process of rocks with different porosities, determining the initial learning rate and decay coefficient of the model, and obtaining a prediction result for rock porosity.

[0007] Preferably, the realization process for the multiple noise reduction method for different types of simultaneous drilling parameters in step S2 is: Based on raw data of different types of simultaneous drilling parameters collected continuously, the extreme value points m of the raw data of each type of simultaneous drilling parameter are identified. i , mean n i , and local amplitude a i Calculate,

number

number

number

number

number

[0008] Preferably, in step S2, a "BP-GA" model is constructed to address the insufficient sample data for the time-frequency domain diagram due to the difficulty in collecting field data for the six types of simultaneous drilling parameters mentioned above. The specific method is as follows: Step 2.1: Determine the simultaneous excavation information samples collected using the neural network BP and the expected output values, and determine the initial weights and thresholds of the BP neural network corresponding to each type of simultaneous excavation parameter through an adaptation algorithm, thereby removing unreasonable data. Step 2.2: For the six types of simultaneous excavation parameters after removing unreasonable data, use the genetic algorithm to perform iteration, crossover, and data mutation, with a crossover probability of 0.75 and a mutation coefficient of 0.031. The newly generated simultaneous excavation parameter data groups after encoding are M1, M2, ……, M6 respectively. Step 2.3: Construct the relationship curves between the newly generated simultaneous excavation parameter data and the excavation time t respectively, thereby obtaining the time-frequency domain feature diagrams of different types of simultaneous excavation parameters. Step 2.4: Utilize the data enhancement function of the neural network to perform random flipping, clockwise or counterclockwise rotation, brightness change, and pixel value improvement operations on the time-frequency domain feature diagrams of each type of simultaneous excavation parameter, expand the time-frequency domain image library of the simultaneous excavation parameters, and stop the iterative calculation of the neural network until the number of time-frequency domain feature images corresponding to each type of simultaneous excavation parameter is ≥ 1200.

[0009] Preferably, in Step 3, the method for obtaining the time-frequency domain feature diagrams of different types of simultaneous excavation parameters is as follows: Based on the simultaneous excavation information after multiple noise removals and "BP-GA" model processing, construct the change curves of different types of simultaneous excavation parameters with respect to the excavation time t, that is, Step 3.1 of obtaining the time domain feature diagrams of the simultaneous excavation parameters. Use the discrete Fourier transform to transform the time domain feature diagram of the simultaneous excavation parameter into a frequency domain feature diagram, and its transformation formula is as follows:

Equation

[0010] Preferably, in Step 4, the specific prediction process of the convolutional neural network prediction model for rock porosity is as follows: Step 4.1: First, input the time-frequency domain feature diagrams of the simultaneous drilling information of rocks with different porosities into the first convolutional layer of the VG-CNN prediction model. This convolutional layer is composed of 42 convolutional kernels with a size of 3×3. Step 4.2: Next, input the feature diagram processed by the first convolutional layer into the first pooling layer of the VG-CNN prediction model. This pooling layer is composed of a 2×2 pooling window. Step 4.3: Repeat Steps 4.1 and 4.2 until the learning rate of the VG-CNN prediction model is ≤ 3×10 -5 and the decay coefficient is ≤ 1×10 -6 At this time, stop the model training. Then, substitute the rock porosity prediction confusion matrix data into the following formula to verify the prediction accuracy of the VG-CNN prediction model for rock porosity.

[0011]

Equation

Advantages of the Invention

[0012] The beneficial effects of this invention are as follows:

[0013] 1. This method acquires limited real-time simultaneous drilling parameter data during the drilling process, applies multiple noise reduction processing, and then performs iterative derivation using a "neural network + genetic algorithm." The newly formed simultaneous drilling dataset is characterized by its large data volume, high reliability, and high robustness, effectively solving the technical challenges of the data acquisition process at the simultaneous drilling experimental site, such as the large impact of external interference, the small amount of usable data, and the high discreteness of the data.

[0014] 2. The simultaneous drilling parameters employed in this method are rational and scientific, and can truly reflect rock porosity. By optimizing the structural parameters of the convolutional neural network, the accuracy of rock porosity reversal is improved, resolving the field delay caused by the need for related laboratory tests in conventional quantitative evaluation of rock porosity. Therefore, this method can be applied to predicting and evaluating the porosity of surrounding rocks in underground construction projects with diverse geological conditions and limited space, such as coal mine tunnels, subways, and other underground structures. [Brief explanation of the drawing]

[0015] To more clearly illustrate embodiments of the present invention or technical concepts in the prior art, the following briefly introduces the accompanying drawings necessary for describing the embodiments or prior art. Clearly, these drawings represent only some embodiments of the present invention, and those skilled in the art can derive other drawings from these without any creative effort. [Figure 1] This is a flowchart of a rock porosity prediction method based on simultaneous drilling data inversion of small samples provided by an embodiment of the present invention. [Figure 2] This is a structural diagram of a BP-GA intelligent algorithm derived model. [Modes for carrying out the invention]

[0016] Hereinafter, the technical proposals in embodiments of the present invention will be clearly and completely described in relation to the drawings of the embodiments of the present invention. Clearly, the embodiments described are only some embodiments of the present invention, not all embodiments. All other embodiments obtained by those skilled in the art without creative work based on the embodiments of the present invention are within the scope of the protection of the present invention.

[0017] As shown in Figure 1, a method for predicting rock porosity based on simultaneous drilling data inversion of small samples includes the following steps: Step S1: Using sensors, collect simultaneous drilling information of the drilling process in a laboratory or field tunnel, and obtain key simultaneous drilling parameters such as torque M, thrust force F, rotational speed N, thrust speed V, drill pipe amplitude A, and vibration acceleration a. Step S2: Perform multiple denoising on the different types of simultaneous drilling parameters collected, input the denoised simultaneous drilling parameter values ​​into the "BP-GA" intelligent algorithm derived model, and obtain a new simultaneous drilling parameter dataset. Step S3: Within the simultaneous drilling parameter dataset, analyze the change curves of different types of simultaneous drilling parameters with respect to drilling time t, obtain time-frequency domain diagrams of simultaneous drilling parameters, and obtain a time-frequency domain image library of different types of simultaneous drilling parameters through the data enhancement function of a neural network. Step S4: Construct a convolutional neural network (VG-CNN) prediction model for rock porosity, train and learn a time-frequency domain feature image library of simultaneous drilling parameters in the drilling process of rocks with different porosity levels, determine the initial learning rate and decay coefficient of the model, and obtain prediction results for rock porosity.

[0018] Here, using a laboratory test as an example, we will perform rock porosity prediction based on the inversion of simultaneous drilling parameters. First, rocks with different porosities, such as sandstone, mudstone, limestone, granite, and marble, are uniformly processed into 15cm × 15cm × 15cm cubic test pieces, and divided into 5 groups according to lithology, with 9 rock samples in each group, for a total of 45 samples. The simultaneous drilling test experiment employs the control variable method, that is, the drilling speed V and rotation speed N are kept constant during the drilling process, and real-time data of other simultaneous drilling parameters of the drilling process are monitored. The drilling speed is set to 60 mm / min, and the drill bit rotation speed is set to 200 r / min. The drilling depth is uniformly preset to 120 mm. When the predetermined drilling depth is reached, the simultaneous drilling test device automatically stops drilling.

[0019] First, noise reduction processing is performed on simultaneous drilling parameters such as torque M, thrust force F, rotational speed N, thrust speed V, drill pipe amplitude A, and vibration acceleration a, collected by sensors, using the multi-layer noise reduction method proposed in this patent. Since the drilling speed V and rotational speed N are kept constant during the experimental process, particular emphasis is placed on noise reduction for torque M, thrust force F, drill pipe amplitude A, and vibration acceleration a. The wavelet noise reduction function used for torque M is the db3 wavelet noise reduction function, the wavelet noise reduction function used for thrust force is the coif3 wavelet noise reduction function, and the wavelet noise reduction function used for drill pipe amplitude A and vibration acceleration a is the bior3.3 wavelet noise reduction function. Noise reduction is stopped until the raw simultaneous drilling data x(t) of each type is reduced to a monotonic function or has ≤ 3 extreme values. Once noise reduction is complete, the datasets corresponding to each type of simultaneous drilling parameter are N1, N2, ..., N6, respectively.

[0020] Figure 2 shows the structure of the "BP-GA" intelligent algorithm derived model. Based on the simultaneous drilling experiment results, the initial weights and selection thresholds of the sample data are determined, and highly discrete data points are extracted by calculating the polar differences and mean values ​​of the data among simultaneous drilling parameters of the same type. For the six types of simultaneous drilling parameters after removing irrational data, iteration, crossover, and data mutation are performed using a genetic algorithm. The crossover probability is 0.75, and the mutation coefficient is 0.031, and the new sets of simultaneous drilling parameter data generated after encoding are M1, M2, ..., M6, respectively.

[0021] Furthermore, relationship curves between newly generated drilling simultaneous parameter data and drilling time t are constructed to obtain time-frequency domain feature diagrams for different types of drilling simultaneous parameters. The time-domain feature diagrams of the drilling simultaneous parameters are converted to frequency-domain feature diagrams using discrete Fourier transforms, and then the time-domain and frequency-domain diagrams of the drilling simultaneous parameters are merged using short-time Fourier transforms to obtain time-frequency domain feature images for different types of drilling simultaneous parameters. Using the data enhancement capabilities of the neural network, operations such as random inversion, clockwise or counterclockwise rotation, brightness adjustment, and pixel value enhancement are performed on the time-frequency domain feature diagrams of each type of drilling simultaneous parameter to expand the time-frequency domain image library of drilling simultaneous parameters, and the iterative calculation of the neural network is stopped until there are ≥ 1200 time-frequency domain feature images corresponding to each type of drilling simultaneous parameter.

[0022] Finally, time-frequency domain feature maps of simultaneous drilling information for rocks with different porosities are input into the first convolutional layer of the rock porosity VG-CNN prediction model. This convolutional layer consists of 42 3x3 convolutional kernels. The feature maps processed in the first convolutional layer are input into the first pooling layer of the VG-CNN prediction model. This pooling layer consists of a 2x2 pooling window. Steps 4.1 and 4.2 are repeated until the learning rate of the VG-CNN prediction model is ≤3×10 -5 , the damping coefficient is ≤ 1 × 10 -6Model training is stopped until this point is reached. At this point, the optimal hyperparameters for the model are shown in Table 1.

[0023] [Table 1]

[0024] After the VG-CNN model training is complete, a new coarse-grained sandstone sample is selected, processed into a 15cm × 15cm × 15cm cubic specimen, and a laboratory simultaneous drilling test is performed. The drilling speed is set to 60mm / min, and the drill bit rotation speed is set to 200r / min. The drilling depth is uniformly pre-set to 120mm. Real-time simultaneous drilling parameters and time-frequency domain feature maps collected in the laboratory are input into the VG-CNN prediction model, and the porosity of the coarse-grained sandstone obtained by inversion is 23.15%. At the same time, a correlation test is performed against the true porosity of the coarse-grained sandstone using a laboratory mercury intrusion test, and based on the test results, the porosity of the coarse-grained sandstone is 25.37%. It can be seen that the rock porosity obtained using the inversion of the present invention is relatively close to the true porosity measured in the laboratory. The present invention makes it possible to accurately predict rock porosity based on simultaneous drilling parameter inversion.

[0025] Clearly, those skilled in the art can make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, if such changes or modifications of the present invention are included in the claims and equivalents of the present invention, the present invention is intended to encompass such changes and modifications.

Claims

1. A method for predicting rock porosity based on simultaneous drilling data inversion of small-scale samples, Step S1 involves collecting simultaneous drilling information during the drilling process in a laboratory or underground tunnel using an excavator, and obtaining key simultaneous drilling parameters such as torque M, thrust force F, rotational speed N, thrust speed V, drill pipe amplitude A, and vibration acceleration a. Step S2 involves performing multiple denoising on different types of collected simultaneous drilling parameters, inputting the denoised simultaneous drilling parameter values ​​into the "BP-GA" model, and obtaining a new simultaneous drilling parameter derived dataset. The implementation process for a multi-noise reduction method for simultaneous drilling parameters of different types is: Based on raw data of different types of simultaneous drilling parameters collected continuously, the extreme point m i, mean n i, and local amplitude a i of the raw data of each type of simultaneous drilling parameter are calculated. [Math 1] [Math 2] The raw drilling simultaneous data x(t) is separated using the local mean function a(t), and then the raw drilling simultaneous data x(t) is demodulated using the envelope estimation function m(t). [Math 3] [Math 4] The above operation is repeated until the raw drilling simultaneous data x(t) is reduced in dimension to a monotonic function or has ≤ 3 extreme values, at which point the raw drilling simultaneous data x(t) is decomposed into k decision components PF and one decomposed residual (step 1.1). Kurtosis H1, correlation coefficient H2, and root mean square value H3 are used as selection criteria for the PF component, and the evaluation index Vij for the PF component is expressed by the following formula: [Math 5] Among them, j takes on values ​​of 1, 2, 3, ..., k. Here, step 1.2 defines PF components with V ij > 90% or more as active ingredients, Step 1.3 involves performing further noise reduction processing on the selected PF components using a wavelet noise reduction method, and performing wavelet noise reduction on the effective PF components of torque M, thrust force F, rotational speed N, thrust speed V, drill pipe amplitude A, and vibration acceleration a, wherein the wavelet noise reduction function used for torque and rotational speed is the db3 wavelet noise reduction function, the wavelet noise reduction function used for thrust force and thrust speed is the coif3 wavelet noise reduction function, and the wavelet noise reduction function used for drill pipe amplitude and vibration acceleration is the bior3.3 wavelet noise reduction function. Step 1.4 includes inputting the simultaneous drilling parameter data after multiple noise reduction into the "BP-GA" model, performing iteration, crossover, and variation to achieve data enhancement and data derivation. The specific method for constructing the "BP-GA" model is as follows: Step 2.1: Determine the collected drilling-simultaneous information samples and expected output values ​​using the neural network BP, determine the initial weights and thresholds of the BP neural network corresponding to each type of drilling-simultaneous parameter through an adaptive algorithm, thereby removing irrational data. Step 2.2: After removing irrational data, the six types of simultaneous drilling parameters were subjected to iteration, crossover, and data mutation using a genetic algorithm, with a crossover probability of 0.75 and a mutation coefficient of 0.

031. The new sets of simultaneous drilling parameter data generated after encoding were M1, M2, ..., M6, respectively. Step 2.3: Construct relationship curves between the newly generated simultaneous drilling parameter data and drilling time t, thereby obtaining time-frequency domain feature diagrams for different types of simultaneous drilling parameters. Step 2.4: Using the neural network's data enhancement function, random inversion, clockwise or counterclockwise rotation, brightness adjustment, and pixel value enhancement operations are performed on the time-frequency domain feature diagrams of each type of simultaneous drilling parameter to expand the time-frequency domain image library of simultaneous drilling parameters. The iterative calculation of the neural network is stopped until there are ≥ 1200 time-frequency domain feature images corresponding to each type of simultaneous drilling parameter. Step S3 involves analyzing the change curves of different types of simultaneous drilling parameters with respect to drilling time t within a new simultaneous drilling parameter dataset, obtaining time-frequency domain diagrams of simultaneous drilling parameters, and obtaining a time-frequency domain image library of different types of simultaneous drilling parameters through the data enhancement function of a neural network. A method for predicting rock porosity based on inversion of simultaneous drilling data for a small sample, comprising step S4: constructing a convolutional neural network prediction model for rock porosity, learning and training a time-frequency domain feature image library of simultaneous drilling parameters in the drilling process of rocks with different porosities, determining the initial learning rate and decay coefficient of the model, and obtaining a prediction result for rock porosity.

2. In step 3, the method for obtaining time-frequency domain feature diagrams of different types of simultaneous drilling parameters is as follows: Step 3.1 involves constructing curves of change over time t for different types of simultaneous drilling parameters, based on the simultaneous drilling information after multiple noise reduction and "BP-GA" model processing, i.e., obtaining a time-domain feature map of the simultaneous drilling parameters. Using the discrete Fourier transform, the time-domain feature map of the drilling simultaneous parameters is transformed into a frequency-domain feature map, and the transformation equation is as follows: [Math 6] In step 3.2, Y(n) represents a finite-length discrete frequency sequence of length L, where n = 0, 1, ..., L-1, Hz, and y(t) represents a discrete-time sequence of time T. A method for predicting rock porosity based on inversion of simultaneous drilling data for a small sample, as described in claim 1, comprising step 3.3, which involves fusing time-domain and frequency-domain diagrams of simultaneous drilling parameters by short-time Fourier transform, i.e., obtaining time-frequency domain feature images of different types of simultaneous drilling parameters with time on the horizontal axis and the collection frequency and instantaneous energy change of simultaneous drilling parameters on the vertical axis, and inputting the expanded time-frequency domain image library of simultaneous drilling parameters into a convolutional neural network prediction model for rock porosity, i.e., a VG-CNN prediction model.

3. In Step 4, the specific prediction process of the convolutional neural network prediction model for rock porosity is as follows: Step 4.1: First, time-frequency domain feature maps of simultaneous drilling information for rocks with different porosities are input into the first convolutional layer of the VG-CNN prediction model, which consists of 42 3x3 convolutional kernels. Step 4.2: Next, the feature map processed in the first convolutional layer is input to the first pooling layer of the VG-CNN prediction model, which consists of a 2x2 pooling window. Step 4.3: Repeat Steps 4.1 and 4.2, ensuring the learning rate of the VG-CNN prediction model is ≤ 3 × 10⁻⁶. －５ , the damping coefficient is ≤ 1 × 10 －６ Model training is stopped until this point is reached, and at this time, the rock porosity prediction confusion matrix data is substituted into the following equation to verify the prediction accuracy of the VG-CNN prediction model for rock porosity. [Number 7] Eventually P Ｔ This indicates a true positive, N Ｔ True negative, P Ｆ This is a false positive, N Ｆ A method for predicting rock porosity based on simultaneous drilling data inversion of a small sample, as described in claim 2, characterized in that it represents a false negative.

Images