一种基于全卷积神经网络钻孔外围砂岩层富水性预测方法

Abstract

本发明公开了一种基于全卷积神经网络钻孔外围砂岩层富水性预测方法，包括：获取实测原始数据；实测原始数据为钻孔内接收点按采样时间序列采集的钻孔外围含水砂岩地质体响应的感应电动势；对实测原始数据进行特征提取，生成特征信息数据；将特征信息数据输入全卷积神经网络中，根据预设富水程度等级分类标准，输出钻孔外围含水砂岩地质体的富水性分级预测信息；富水性分级预测信息为富水性类别，包括：不含水围岩、弱富水性围岩、中等富水性围岩、强富水性围岩和极强富水性围岩。可有效提高钻孔孔壁外围径向方向砂岩含水层富水程度的分级预测精度。

Description

Technical Field

[0001] This invention relates to the field of water-bearing prediction technology for sandstone layers surrounding boreholes, and particularly to a method for predicting water-bearing properties of sandstone layers surrounding boreholes based on a fully convolutional neural network. Background Technology

[0002] Mine water hazards are a major safety hazard threatening mine production. Hidden geological features such as faults, collapse columns, goafs, and fissure zones that may exist ahead of tunnels are direct factors causing sudden water hazard accidents. To ensure safe mining, it is necessary to investigate the water-bearing capacity of underground aquifers in the mining area, the hydraulic connections between aquifers, and the water conductivity of geological structures. In particular, determining the water-bearing capacity of sandstone layers and making reasonable predictions and assessments of their hazards is of paramount guiding significance and practical value for mine safety production.

[0003] Currently, the hydrogeological assessment of the water-bearing capacity of sandstone layers generally adopts a combination of drilling and pumping tests. First, boreholes are drilled in the exploration area, and then pumping tests are conducted. Based on the results of the pumping tests, the water inflow of the sandstone layer is determined, and its water-bearing capacity is predicted and evaluated. This method requires drilling and pumping tests, which is costly and time-consuming. Furthermore, it can only determine the water-bearing capacity of the sandstone layer at the borehole location and cannot accurately predict the water-bearing capacity of the sandstone layer surrounding the borehole location. This method does not meet the requirements of relevant regulations for safe production in mines.

[0004] Advanced detection of water hazards ahead of underground tunnels mainly includes drilling and geophysical methods. The borehole transient electromagnetic method (BEM) utilizes a coil to emit a pulsed primary electromagnetic field into the borehole wall strata, and another coil to receive the secondary eddy current electromagnetic field induced by the pulsed electromagnetic field in the water-bearing body of the surrounding strata. The water-bearing capacity of the strata is predicted by analyzing the spatial and temporal distribution of this secondary field. The BEM uses a linear array of transmitting and receiving points within the borehole and can be applied to advanced detection in underground mining and the detection of water-bearing geological bodies in sandstone strata. However, current BEM methods have limited analysis of the transient electromagnetic response characteristics of different water-bearing capacities in sandstone strata, and can only qualitatively analyze the strength of water-bearing capacity through inverted resistivity profiles, failing to achieve accurate classification and prediction of the water-bearing capacity of geological bodies.

[0005] Therefore, how to improve the accuracy of predicting the water-bearing properties of sandstone layers surrounding boreholes has become a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] In view of the above problems, the present invention proposes a method for predicting the water-bearing capacity of sandstone layers around boreholes based on a fully convolutional neural network, which at least solves some of the above technical problems. This method can achieve accurate classification and prediction of the water-bearing capacity of water-bearing geological bodies in the radial direction around the borehole wall.

[0007] This invention provides a method for predicting the water-bearing capacity of sandstone layers surrounding boreholes based on a fully convolutional neural network, comprising:

[0008] Acquire the measured raw data; the measured raw data is the induced electromotive force of the water-bearing sandstone geological body outside the borehole collected by the receiving point according to the sampling time sequence; the receiving point is the receiving point generated by placing the transmitting coil and the receiving coil in the borehole and moving them simultaneously at all times.

[0009] Feature extraction is performed on the measured raw data to generate feature information data;

[0010] The feature information data is input into a fully convolutional neural network, and the water-bearing grade prediction information of the water-bearing sandstone geological body surrounding the borehole is output according to the preset water-bearing degree classification standard. The water-bearing grade prediction information is a water-bearing category, including: waterless surrounding rock, weakly water-bearing surrounding rock, moderately water-bearing surrounding rock, strongly water-bearing surrounding rock, and extremely strongly water-bearing surrounding rock.

[0011] Furthermore, feature extraction is performed on the measured raw data to generate feature information data, including:

[0012] Divide the measured raw data by the magnetic moment to obtain the normalized induced electromotive force;

[0013] The normalized induced electromotive force is processed by a decay rate program to obtain the decay rate.

[0014] The normalized induced electromotive force is used to calculate the apparent resistivity of the entire region through an optimized binary search algorithm;

[0015] Based on the apparent resistivity of the entire area, the detection distance is calculated using the time-depth conversion formula derived from the electromagnetic field smoke ring theory.

[0016] The normalized induced electromotive force, decay rate, apparent resistivity of the entire area, detection distance, and sampling time together constitute the feature information data.

[0017] Furthermore, the fully convolutional neural network includes an encoding part and a decoding part.

[0018] Furthermore, the encoding portion consists of five stages; wherein the first stage includes a first network substructure Conv3-8; the second stage includes a second network substructure Conv3-16; the third stage includes a third network substructure Conv3-32; the fourth stage includes a fourth network substructure Conv3-64; and the fifth stage includes a fifth network substructure Conv3-128.

[0019] Furthermore, the first, second, third, fourth, and fifth network substructures all include convolutional layers, batch normalization layers, and ReLU activation function layers.

[0020] Furthermore, the decoding part consists of five stages; the first stage includes the sixth network substructure Conv3-64; the second stage includes the seventh network substructure Conv3-32; the third stage includes the eighth network substructure Conv3-16; the fourth stage includes the ninth network substructure Conv3-8; and the fifth stage includes a convolutional layer Conv3-1.

[0021] Furthermore, the first, second, third, fourth, and fifth stages are all connected by pooling layers that downsample the input feature data.

[0022] Furthermore, the downsampling operation skips to a preset layer in the fully convolutional neural network; the preset layer refers to a layer with the same number of channels as the downsampling operation.

[0023] Furthermore, an upsampling operation is performed after the downsampling operation;

[0024] After the upsampling operation, the feature data are linearly fused into an output dimension for output, and the mean squared error loss function is used as the standard to measure the quality of the network output prediction; the output dimension is the same size as the output layer of the fully convolutional neural network.

[0025] Furthermore, the preset water abundance level classification standard is obtained in the following way:

[0026] The unit water inflow rate corresponding to the resistivity of the water-bearing sandstone geological body per unit thickness around the borehole is calculated using the following formula:

[0027]

[0028] In the above formula, ρ is the resistivity of water-bearing sandstone; ρ w ρ is the formation water resistivity; m is the cementation index of the rock; n is the saturation index; a is the lithology coefficient related to the rock; g is the gravitational acceleration; v is the kinematic viscosity; d is the average diameter of the filling particles; q is the unit inflow rate of the drainage hole; S is the unit drawdown of the drainage hole; S w h' is the water saturation level; h' is the unit aquifer thickness; r is the wellbore radius. Radius of borehole influence; K is the permeability coefficient;

[0029] Based on the unit water inflow corresponding to the resistivity of the water-bearing sandstone geological body per unit thickness surrounding the borehole, the preset water-bearing degree classification standard is constructed.

[0030] The beneficial effects of the above-described technical solutions provided in the embodiments of the present invention include at least the following:

[0031] This invention provides a method for predicting the water-bearing capacity of sandstone layers surrounding a borehole based on a fully convolutional neural network. The method includes: acquiring measured raw data; the measured raw data consists of the induced electromotive force of the water-bearing sandstone geological body surrounding the borehole, collected from the receiving point according to a sampling time sequence; extracting features from the measured raw data to generate feature information data; inputting the feature information data into the fully convolutional neural network, and outputting water-bearing capacity classification prediction information for the water-bearing sandstone geological body surrounding the borehole according to a preset water-bearing capacity classification standard; the water-bearing capacity classification prediction information is a water-bearing capacity category, including: no water-bearing surrounding rock, weakly water-bearing surrounding rock, moderately water-bearing surrounding rock, strongly water-bearing surrounding rock, and extremely strongly water-bearing surrounding rock. This method can effectively improve the accuracy of the classification prediction of the water-bearing capacity of sandstone aquifers in the radial direction surrounding the borehole wall.

[0032] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings.

[0033] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0034] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0035] Figure 1 A flowchart of the water-bearing capacity prediction method for sandstone layers surrounding boreholes based on a fully convolutional neural network, provided in an embodiment of the present invention;

[0036] Figure 2 This is a schematic diagram of the topology of a fully convolutional neural network provided in an embodiment of the present invention;

[0037] Figure 3 This is a schematic diagram of the borehole transient electromagnetic detection setup provided in an embodiment of the present invention. Detailed Implementation

[0038] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0039] This invention provides a method for predicting the water-bearing capacity of sandstone layers surrounding boreholes based on a fully convolutional neural network, referring to... Figure 1 As shown, it includes:

[0040] Obtain the measured raw data; the measured raw data is the induced electromotive force of the water-bearing sandstone geological body around the borehole collected by the receiving point according to the sampling time sequence; the receiving point is the receiving point generated by placing the transmitting coil and the receiving coil in the borehole and moving them simultaneously at all times.

[0041] Feature extraction is performed on the measured raw data to generate feature information data;

[0042] The feature information data is input into a fully convolutional neural network, and the water-bearing grade prediction information of the water-bearing sandstone geological body around the borehole is output according to the preset water-bearing degree classification standard. The water-bearing grade prediction information is a water-bearing category, including: waterless surrounding rock, weakly water-bearing surrounding rock, moderately water-bearing surrounding rock, strongly water-bearing surrounding rock, and extremely strongly water-bearing surrounding rock.

[0043] This embodiment provides a method for predicting the water-bearing capacity of sandstone strata surrounding boreholes based on fully convolutional neural networks. By utilizing borehole transient electromagnetic detection data and employing a fully convolutional neural network method, it effectively improves the accuracy of graded prediction of the water-bearing capacity of sandstone aquifers in the radial direction surrounding the borehole wall. This method can effectively guide mine water hazard prevention and control, reduce the occurrence of mine water inrush accidents, and ensure safe mining operations.

[0044] This method is implemented through the following steps:

[0045] Step 1: For sandstone formations, based on Archie's formula, Kozeny–Carman (KC) formula, hydraulic conductivity formula, and unit yield formula, establish the mathematical relationship between the resistivity of the aquifer per unit thickness and the unit yield of the sandstone aquifer:

[0046]

[0047] In the above formula, ρ is the resistivity of water-bearing sandstone, Ω·m; ρ w Let ρ be the resistivity of the formation water. w = 1Ω·m; m is the cementation index of the rock, taken as m = 2; n is the saturation index, taken as n = 2; a is the lithology coefficient related to the rock, taken as a = 1; g is the acceleration due to gravity, m / s² 2 v is the kinematic viscosity, m 2 / s; d is the average diameter of the filling particles, taken as d = 0.5 mm; σ = 1 / ρ; q is the unit flow rate of the discharge hole, L / (s·m); S is the unit drawdown of the discharge hole, taken as S = 10 m; S w Let S be the water saturation level.w =1; h' is the unit aquifer thickness, taken as h' = 1m; r is the well radius, taken as r = 0.0455m; Radius of influence of the borehole is in meters; K is the permeability coefficient, in meters per second.

[0048] The unit water inflow corresponding to the resistivity of the water-bearing body per unit thickness is calculated, which can realize the accurate classification and prediction of the water-bearing degree of the rock strata, as shown in Table 1 below.

[0049] Table 1 Relationship between resistivity and unit flow rate

[0050]

[0051] The water-bearing level of sandstone aquifers per unit thickness can be predicted based on the resistivity of the sandstone aquifers per unit thickness in the table above.

[0052] If ρ≥60 and q≤0.1, then the water-bearing geological body is classified as weakly water-bearing.

[0053] If 13 < ρ < 60 and 0.1 < q ≤ 1.0, then the water-bearing geological body is classified as moderately water-bearing.

[0054] If 7≤ρ≤13 and 1.0<q≤5.0, then the water-bearing geological body is classified as strongly water-bearing.

[0055] If 4 ≤ ρ < 7 and q > 5.0, then the water-bearing geological body is classified as extremely water-bearing.

[0056] Step 2: Obtain raw data on the water-bearing properties of the water-bearing sandstone layer surrounding the borehole based on the full-space three-dimensional time-domain transient electromagnetic forward modeling.

[0057] Based on the full-space three-dimensional time-domain transient electromagnetic forward modeling method, the transmitting and receiving coils are placed in the borehole and moved simultaneously for detection. The induced electromotive force (EMF) collected by the receiving point according to the sampling time sequence is used as the raw data. The raw data is then divided by the magnetic moment to obtain the normalized induced EMF. The normalized induced EMF data is processed by an attenuation rate program to obtain the attenuation rate. The normalized induced EMF is then used to calculate the apparent resistivity of the entire area using an optimized binary search algorithm. The detection distance is calculated using the time-depth conversion formula derived from the electromagnetic field smoke ring theory, specifically by calculating the detection distance based on time and the apparent resistivity of the entire area. From the smoke ring effect of electromagnetic field propagation, it is known that the propagation speed v of the electromagnetic field in the rock medium is time-dependent. Therefore, the propagation depth of the electromagnetic field can be considered as a function related to the propagation speed and time, as expressed below:

[0058]

[0059] In the formula, D is the propagation depth of the electromagnetic field, v is the propagation speed of the electromagnetic field, ρ is the true resistivity of the rock stratum, and t is the propagation time. In actual data processing, the apparent resistivity ρ is calculated. s Therefore, the above formula can be written as:

[0060]

[0061] The above data processing flow completes the extraction of feature information from the original data. The decay rate program is designed to calculate the derivative value of the induced electromotive force (EMF) at the sampling point time, based on the fluctuations in the response values ​​of the induced EMF collected at the receiving point in both aquifer and aquifer geological bodies. This yields the decay rate of the induced EMF.

[0062] Step 3: Use the aforementioned characteristic information data that reflects the water-bearing capacity of sandstone layers to construct and train a fully convolutional neural network model.

[0063] The fully convolutional neural network (WCNN) structure consists of two parts: encoding and decoding. It mainly includes an input layer, convolutional layers, pooling layers, and an output layer. The input layer contains five predetermined transient electromagnetic features from borehole drilling. The convolutional layers use 3×3 kernels, the pooling layers employ max-pooling, and the output layer classifies the water-bearing capacity of sandstone strata. The basic network topology is as follows: Figure 2 As shown, the input feature parameters Pi (i = 1, 2, 3, 4, 5) for each measuring point correspond to the induced electromotive force, apparent resistivity, induced electromotive force decay rate, sampling time, and detection distance, respectively. The output parameter T represents the water-bearing capacity category of the aquifer, where "0" represents no water-bearing surrounding rock, "1" represents weak water-bearing capacity, "2" represents moderate water-bearing capacity, "3" represents strong water-bearing capacity, and "4" represents extremely strong water-bearing capacity. The input and output of the fully convolutional neural network are both in matrix form, that is, the matrix form composed of the 5 input feature attribute parameters is P = [P1 P2 P3 P4 P5], and the output matrix form is T = [T].

[0064] Specifically, the encoding portion of a fully convolutional neural network has five stages: The first stage contains a substructure Conv3-8 (meaning the convolution kernel size is 3x3 with 8 channels), which includes convolutional layers, batch normalization layers, and ReLU activation function layers. The second stage contains a substructure Conv3-16. The third stage contains a substructure Conv3-32. The fourth stage contains a substructure Conv3-64. The fifth stage contains a substructure Conv3-128.

[0065] The decoding part of the fully convolutional neural network has a network structure divided into five stages: the first stage contains a substructure Conv3-64, the second stage contains a substructure Conv3-32, the third stage contains a substructure Conv3-16, the fourth stage contains a substructure Conv3-8, and the fifth stage contains a convolutional layer Conv3-1. Each stage is connected by pooling layers to downsample the input feature data. During downsampling, certain layers in the neural network are skipped, connecting feature data from different skipping paths. This allows for the fusion of detailed information from shallow features and semantic features from deeper networks, improving the accuracy of network prediction. Then, upsampling is performed, and finally, the feature data is linearly fused to a dimension with the same size as the output dimension. The mean squared error loss function is used as a metric for evaluating the network's prediction performance. "certain layers" refers to layers with the same number of channels as those in the upsampling process (the channels corresponding to upsampling and downsampling).

[0066] The specific network parameters for the encoding and decoding parts of the network are shown in the table below:

[0067] Table 2. Structure Parameters of Fully Convolutional Neural Networks

[0068]

[0069] Step 4: Apply the fully convolutional neural network model constructed above to predict the water-bearing properties of the sandstone strata.

[0070] Specifically, the borehole transient electromagnetic detection device includes a main unit, a transmitter, and receiver points, with the main unit connected to the transmitter and receiver points. The transmitter and receiver points are arranged in a straight line within the borehole. Different detection arrangements are established by changing the relative positions of the transmitter and receiver points. Different arrangements are selected based on the actual geological task. Detection arrangements include coplanar and coaxial arrangements of the transmitter and receiver points. The transmitter and receiver points move along the borehole axis at a set interval, such as... Figure 3 Normalized induced electromotive force (EMF) data of different water-bearing sandstone geological bodies are collected from the receiving point in a time series as the measured raw data and stored in the host computer. Step 2 processes the measured raw data, inputting the obtained normalized EMF, decay rate, sampling time, apparent resistivity of the entire area, and detection distance into the model. The model outputs water-bearing classification prediction information. The output water-bearing categories are: "0" represents waterless surrounding rock, "1" represents weak water-bearing, "2" represents moderate water-bearing, "3" represents strong water-bearing, and "4" represents extremely strong water-bearing.

[0071] The water-bearing capacity prediction method for sandstone aquifers around boreholes based on a fully convolutional neural network provided in this embodiment differs from existing transient electromagnetic borehole methods, which can only determine the relative strength of water-bearing capacity in geological bodies. The fully convolutional neural network model used in this embodiment exhibits strong adaptability to feature parameters and high computational efficiency, enabling graded prediction of the water-bearing capacity of sandstone aquifers in the radial direction surrounding the borehole wall. This method can accurately and effectively predict the water-bearing capacity of sandstone aquifers during mining operations, providing safety warnings and a theoretical basis for the design of mine drainage systems. It solves the problem of inaccurate graded prediction of water-bearing capacity in existing mine water hazard prevention methods, effectively ensuring safe mining operations.

[0072] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for predicting the water-bearing capacity of sandstone layers surrounding boreholes based on a fully convolutional neural network, characterized in that, include: Acquire the measured raw data; the measured raw data is the induced electromotive force of the water-bearing sandstone geological body around the borehole collected by the receiving point according to the sampling time sequence; the receiving point is the receiving point generated by placing the transmitting coil and the receiving coil in the borehole and moving them simultaneously at all times. Feature extraction is performed on the measured raw data to generate feature information data; The feature information data is input into a fully convolutional neural network, and the water-bearing grade prediction information of the water-bearing sandstone geological body around the borehole is output according to the preset water-bearing degree classification standard. The water-bearing grade prediction information is a water-bearing category, including: waterless surrounding rock, weakly water-bearing surrounding rock, moderately water-bearing surrounding rock, strongly water-bearing surrounding rock, and extremely strongly water-bearing surrounding rock. The process of extracting features from the measured raw data to generate feature information data includes: Divide the measured raw data by the magnetic moment to obtain the normalized induced electromotive force; The normalized induced electromotive force is processed by a decay rate program to obtain the decay rate. The normalized induced electromotive force is used to calculate the apparent resistivity of the entire region through an optimized binary search algorithm; Based on the apparent resistivity of the entire area, the detection distance is calculated using the time-depth conversion formula derived from the electromagnetic field smoke ring theory. The normalized induced electromotive force, decay rate, apparent resistivity of the entire area, detection distance, and sampling time together constitute the feature information data. The preset water abundance level classification standard is obtained in the following way: The unit water inflow rate corresponding to the resistivity of the water-bearing sandstone geological body per unit thickness around the borehole is calculated using the following formula: ； In the above formula, Resistivity of water-bearing sandstone; The resistivity of formation water; The cementation index of the rock; It is the saturation index; Lithology coefficients related to rocks; It is the acceleration due to gravity; Kinematic viscosity; The average diameter of the filling particles; The unit flow rate of water from the drain hole; The unit water level drawdown of the drain hole; Water saturation; The unit is the thickness of the aquifer. r The radius of the wellbore; , where is the radius of influence of the borehole; It is the permeability coefficient; Based on the unit water inflow corresponding to the resistivity of the water-bearing sandstone geological body per unit thickness surrounding the borehole, the preset water-bearing degree classification standard is constructed.

2. The method for predicting the water-bearing capacity of sandstone layers surrounding boreholes based on a fully convolutional neural network as described in claim 1, characterized in that, The fully convolutional neural network includes an encoding part and a decoding part.

3. The method for predicting the water-bearing capacity of sandstone layers surrounding boreholes based on a fully convolutional neural network as described in claim 2, characterized in that, The encoding part consists of five stages; the first stage includes a first network substructure Conv3-8; the second stage includes a second network substructure Conv3-16; the third stage includes a third network substructure Conv3-32; the fourth stage includes a fourth network substructure Conv3-64; and the fifth stage includes a fifth network substructure Conv3-128.

4. The method for predicting the water-bearing capacity of sandstone layers surrounding boreholes based on a fully convolutional neural network as described in claim 3, characterized in that, The first, second, third, fourth, and fifth network substructures all contain convolutional layers, batch normalization layers, and ReLU activation function layers.

5. The method for predicting the water-bearing capacity of sandstone layers surrounding boreholes based on a fully convolutional neural network as described in claim 2, characterized in that, The decoding part consists of five stages; the first stage contains the sixth network substructure Conv3-64; the second stage contains the seventh network substructure Conv3-32; the third stage contains the eighth network substructure Conv3-16; the fourth stage contains the ninth network substructure Conv3-8; and the fifth stage contains a convolutional layer Conv3-1.

6. The method for predicting the water-bearing capacity of sandstone layers surrounding boreholes based on a fully convolutional neural network as described in claim 5, characterized in that, The first, second, third, fourth, and fifth stages are connected by pooling layers that downsample the input feature data.

7. The method for predicting the water-bearing capacity of sandstone layers surrounding boreholes based on a fully convolutional neural network as described in claim 6, characterized in that, The downsampling operation skips to a preset layer in the fully convolutional neural network; the preset layer refers to a layer with the same number of channels as the downsampling operation.

8. The method for predicting the water-bearing capacity of sandstone layers surrounding boreholes based on a fully convolutional neural network as described in claim 6, characterized in that, The downsampling operation is followed by the upsampling operation; After the upsampling operation, the feature data are linearly fused into an output dimension for output, and the mean squared error loss function is used as the standard to measure the quality of the network output prediction; the output dimension is the same size as the output layer of the fully convolutional neural network.

Images