CN115308803A

CN115308803A - Coal seam thickness prediction method, device, equipment and medium

Info

Publication number: CN115308803A
Application number: CN202211115529.9A
Authority: CN
Inventors: 姜绪超; 王峰; 张守祥; 周如林; 高思伟; 曹宁宁; 李再峰; 宋国利
Original assignee: Beijing Meike Tianma Automation Technology Co Ltd; Beijing Tianma Intelligent Control Technology Co Ltd
Current assignee: Beijing Meike Tianma Automation Technology Co Ltd; Beijing Tianma Intelligent Control Technology Co Ltd
Priority date: 2022-09-14
Filing date: 2022-09-14
Publication date: 2022-11-08

Abstract

The disclosure provides a method, a device, equipment and a medium for predicting coal seam thickness, and relates to the technical field of artificial intelligence. The method comprises the following steps: detecting the coal rock layer to be detected by adopting a ground penetrating radar to obtain a radar signal to be detected; performing feature extraction on the radar signal to be detected by adopting a first sub-network in the coal rock recognition model to obtain a prediction feature for indicating the spatial distribution of a coal bed and a rock stratum in the coal rock stratum to be detected; and predicting the coal seam thickness of the prediction characteristics by adopting a second sub-network in the coal rock recognition model so as to obtain the first thickness distribution of the coal seam in the coal rock layer to be detected on the detection path of the ground penetrating radar. Therefore, the thickness of the coal bed (namely the thickness distribution of the coal bed in the coal rock stratum to be detected) at each position point on the detection path of the ground penetrating radar is predicted based on the deep learning technology, the accuracy of a prediction result can be improved, the prediction speed is high, and real-time prediction can be realized. And parameters such as the relative dielectric constant of the coal seam, the propagation speed and the like do not need to be measured, and the workload of the prediction process can be reduced.

Description

Coal seam thickness prediction method, device, equipment and medium

Technical Field

The disclosure relates to the technical field of artificial intelligence, in particular to a method, a device, equipment and a medium for predicting coal seam thickness.

Background

The coal rock (coal bed and rock stratum) identification is a technology for identifying the lithology of the geologic body, and the coal rock identification technology has important significance for intelligent mining, improvement of coal mining operation environment and realization of safe and efficient production of coal mines. The thickness of the coal bed in the coal rock is identified, and the requirement for accurately detecting the coal rock in unmanned intelligent mining of a mine can be met.

In the related technology, a ground penetrating radar can be adopted to detect mine coal and rock, radar data is obtained, the radar data is preprocessed, then a seed position selecting and tracking method (or a position tracking method and a coal and rock position tracking method) is adopted to calculate a coal and rock position, finally, a coal and rock position coordinate is calculated by using algorithms such as a road correlation coefficient and the like and the position of a radar antenna, and thickness information of a coal bed is obtained from the radar data.

However, the above coal petrography horizon locating method needs to determine the propagation velocity of the electromagnetic wave in the measured coal seam or the relative dielectric constant of the measured coal seam in advance, and the above parameters need to be measured additionally, which increases the workload of coal seam thickness prediction. In addition, the measurement needs sampling measurement, and the prediction of the coal seam thickness cannot be automatically completed.

Disclosure of Invention

The present disclosure is directed to solving, at least to some extent, one of the technical problems in the related art.

The invention provides a method, a device, equipment and a medium for predicting the thickness of a coal bed, which are used for predicting the thickness of the coal bed (namely the thickness distribution of the coal bed in a coal rock stratum to be detected) at each position point on a detection path of a ground penetrating radar based on a deep learning technology. Moreover, compared with the method for calculating the thickness of the coal seam by using a physical formula in the prior art, the method and the device do not need to measure the relative dielectric constant of the coal seam, calculate parameters such as propagation speed and the like, and can reduce the workload of the prediction process. In addition, the thickness distribution of the coal bed can be directly predicted according to the radar signal to be measured without measuring the relative dielectric constant, and the automatic prediction of the coal bed thickness can be realized.

An embodiment of the first aspect of the disclosure provides a method for predicting coal seam thickness, which includes:

detecting the coal rock layer to be detected by adopting a ground penetrating radar to obtain a radar signal to be detected;

performing feature extraction on the radar signal to be detected by adopting a first sub-network in a coal rock recognition model to obtain a prediction feature; the prediction features are used for indicating the spatial distribution of coal beds and rock strata in the coal rock strata to be detected;

and predicting the thickness of the coal bed by adopting a second sub-network in the coal rock recognition model to the prediction characteristics so as to obtain the first thickness distribution of the coal bed in the coal rock layer to be detected on the detection path of the ground penetrating radar.

According to the method for predicting the coal seam thickness, the coal rock layer to be detected is detected by adopting the ground penetrating radar, so that a radar signal to be detected is obtained; performing feature extraction on the radar signal to be detected by adopting a first sub-network in the coal rock recognition model to obtain a prediction feature; the prediction characteristics are used for indicating the spatial distribution of the coal bed and the rock stratum in the coal rock stratum to be detected; and predicting the coal seam thickness of the prediction characteristics by adopting a second sub-network in the coal rock recognition model so as to obtain the first thickness distribution of the coal seam in the coal rock layer to be detected on the detection path of the ground penetrating radar. Therefore, the thickness of the coal bed (namely the thickness distribution of the coal bed in the coal rock stratum to be detected) at each position point on the detection path of the ground penetrating radar is predicted based on the deep learning technology, on one hand, the accuracy of a prediction result can be improved, on the other hand, the prediction speed is high, and real-time prediction can be achieved. Moreover, compared with the method for calculating the thickness of the coal seam by using a physical formula in the prior art, the method and the device do not need to measure the relative dielectric constant of the coal seam, calculate parameters such as propagation speed and the like, and can reduce the workload of the prediction process. In addition, the thickness distribution of the coal bed can be directly predicted according to the radar signal to be measured without measuring the relative dielectric constant, and the automatic prediction of the coal bed thickness can be realized.

An embodiment of a second aspect of the present disclosure provides a device for predicting coal seam thickness, including:

the detection module is used for detecting the coal rock layer to be detected by adopting a ground penetrating radar so as to obtain a radar signal to be detected;

the extraction module is used for extracting the characteristics of the radar signal to be detected by adopting a first sub-network in the coal rock recognition model so as to obtain predicted characteristics; the prediction features are used for indicating the spatial distribution of coal beds and rock strata in the coal rock strata to be detected;

and the prediction module is used for predicting the thickness of the coal bed by adopting a second sub-network in the coal rock recognition model to obtain the first thickness distribution of the coal bed in the coal rock layer to be detected on the detection path of the ground penetrating radar.

The coal seam thickness prediction device of the embodiment of the disclosure adopts a ground penetrating radar to detect a coal rock layer to be detected so as to obtain a radar signal to be detected; performing feature extraction on the radar signal to be detected by adopting a first sub-network in the coal rock recognition model to obtain a prediction feature; the prediction characteristics are used for indicating the spatial distribution of the coal bed and the rock stratum in the coal rock stratum to be detected; and predicting the coal seam thickness of the prediction characteristics by adopting a second sub-network in the coal rock recognition model so as to obtain the first thickness distribution of the coal seam in the coal rock layer to be detected on the detection path of the ground penetrating radar. Therefore, the thickness of the coal bed (namely the thickness distribution of the coal bed in the coal rock stratum to be detected) at each position point on the detection path of the ground penetrating radar is predicted based on the deep learning technology, on one hand, the accuracy of a prediction result can be improved, on the other hand, the prediction speed is high, and real-time prediction can be achieved. Moreover, compared with the method for calculating the thickness of the coal seam by using a physical formula in the prior art, the method and the device do not need to measure the relative dielectric constant of the coal seam, calculate parameters such as propagation speed and the like, and can reduce the workload of the prediction process. In addition, the thickness distribution of the coal bed can be directly predicted according to the radar signal to be measured without measuring the relative dielectric constant, and the automatic prediction of the coal bed thickness can be realized.

An embodiment of a third aspect of the present disclosure provides an electronic device, including: the device comprises a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein the processor executes the program to realize the coal seam thickness prediction method provided by the embodiment of the first aspect of the disclosure.

A fourth aspect of the present disclosure provides a non-transitory computer-readable storage medium, on which a computer program is stored, where the computer program, when executed by a processor, implements a method for predicting coal seam thickness as set forth in the first aspect of the present disclosure.

An embodiment of a fifth aspect of the present disclosure provides a computer program product, wherein when the instructions of the computer program product are executed by a processor, the method for predicting the thickness of the coal seam as set forth in the embodiment of the first aspect of the present disclosure is performed.

Additional aspects and advantages of the disclosure will be set forth in part in the description which follows and, in part, will be obvious from the description, or may be learned by practice of the disclosure.

Drawings

The foregoing and/or additional aspects and advantages of the present disclosure will become apparent and readily appreciated from the following description of the embodiments, taken in conjunction with the accompanying drawings of which:

fig. 1 is a schematic flow chart of a method for predicting coal seam thickness according to an embodiment of the present disclosure;

fig. 2 (a) is a schematic diagram of a ground penetrating radar data collection provided in the embodiment of the present disclosure;

fig. 2 (b) is a schematic diagram of a ground penetrating radar data acquisition according to an embodiment of the present disclosure;

FIG. 3 is a schematic flow chart illustrating another method for predicting coal seam thickness according to an embodiment of the present disclosure;

fig. 4 is a first schematic structural diagram of a coal rock identification model provided in an embodiment of the present disclosure;

fig. 5 is a schematic structural diagram ii of a coal rock identification model provided in the embodiment of the present disclosure;

FIG. 6 is a diagram of data according to an embodiment of the present disclosure;

fig. 7 is a schematic diagram of a training process of a coal rock recognition model according to an embodiment of the present disclosure;

fig. 8 is a schematic diagram of a training process of another coal rock recognition model provided in the embodiment of the present disclosure;

fig. 9 is a schematic structural diagram of a coal seam thickness prediction apparatus according to an embodiment of the present disclosure.

Detailed Description

Reference will now be made in detail to the embodiments of the present disclosure, examples of which are illustrated in the accompanying drawings, wherein like or similar reference numerals refer to the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the drawings are exemplary and intended to be illustrative of the present disclosure, and should not be construed as limiting the present disclosure.

Currently, the thickness of a coal seam in a coal rock can be determined in the following manner:

in the first mode, a ground penetrating radar is adopted to detect mine coal and rock, radar data are obtained, the radar data are preprocessed, then a seed position selecting and tracking method (or a position tracking method and a coal and rock position tracking method) is adopted to calculate a coal and rock position, finally, an algorithm such as a channel correlation coefficient and the like and the position of a radar antenna are utilized to calculate a coal and rock position coordinate, and thickness information of a coal bed is obtained from the radar data.

In the second mode, a ground penetrating radar is used for detecting the interface of the top and bottom plates of the coal rock, then the relative dielectric constant is corrected through drilling data, and finally the thickness of the coal bed is calculated by using radar matching software.

However, in the first method, the propagation speed of the electromagnetic wave in the measured coal seam or the relative dielectric constant of the measured coal seam needs to be determined in advance, and the parameters need to be obtained through additional measurement, so that the workload of predicting the coal seam thickness is increased, and the automation of the coal seam thickness prediction is not realized.

The second method does not involve a specific algorithm of data processing software, and requires human intervention in the calculation process. In addition, when the method is used for predicting the coal seam thickness, a drilling map is needed to obtain the relative dielectric constant of the coal seam, manual participation is needed for predicting the coal seam thickness, and automatic identification and calculation cannot be achieved.

Therefore, in order to solve at least one of the above problems, the present disclosure provides a method, an apparatus, an electronic device, and a storage medium for predicting a coal seam thickness.

Methods, apparatus, devices and media for predicting coal seam thickness according to embodiments of the present disclosure are described below with reference to the accompanying drawings. Before describing embodiments of the present disclosure in detail, for ease of understanding, common terminology will be introduced first:

a ground penetrating radar: a geophysical method for detecting the characteristics and distribution rule of the substance in medium features that the antenna is used to emit and receive high-frequency electromagnetic waves. The ground penetrating radar technology is a geophysical method for detecting the characteristics and distribution rule of substances in a medium, and can be used for identifying the interfaces of coal seams and rock stratums.

Deep learning: a machine learning method can independently construct basic rules according to example data in a learning process.

Fig. 1 is a schematic flow chart of a method for predicting coal seam thickness according to an embodiment of the present disclosure.

In the embodiment of the disclosure, the method for predicting the coal seam thickness can be applied to any electronic equipment, so that the electronic equipment can execute a function of predicting the coal seam thickness.

The electronic device may be any device having a computing capability, for example, a PC (Personal Computer), a mobile terminal, a server, and the like, and the mobile terminal may be a hardware device having various operating systems, touch screens, and/or display screens, such as a mobile phone, a tablet Computer, a Personal digital assistant, and a wearable device.

As shown in fig. 1, the method for predicting the thickness of the coal seam may include the following steps:

step 101, detecting the coal rock layer to be detected by adopting a ground penetrating radar to obtain a radar signal to be detected.

In the embodiment of the disclosure, a ground penetrating radar may be used to detect the coal rock layer to be detected, so as to obtain a radar signal to be detected.

As an example, the radiation direction of the ground penetrating radar may be vertically upward and toward a coal formation (such as a coal formation to be measured), the moving direction is a horizontal direction, and the moving manner is a uniform movement. For example, the moving mode and the moving direction of the ground penetrating radar may be as shown in fig. 2 (a), and the ground penetrating radar may move a certain distance at a constant horizontal speed to collect a radar signal to be detected. For example, under the condition that the ground penetrating radar moves at a constant speed, a fixed number of frames can be collected every second, so that the intervals of all position points in the collected radar signal image are equal.

As another example, the radiation direction of the ground penetrating radar may be vertically upward and toward the coal formation (e.g., the coal formation to be detected), the moving direction may be a horizontal direction, and the moving manner may be a uniform movement or a non-uniform movement. For example, the ground penetrating radar may move in the direction shown in fig. 2 (b), and the ground penetrating radar may be connected to a distance measuring device (e.g., a distance measuring wheel, a laser distance measuring device, a displacement sensor, etc.), wherein the distance measuring device may move at a constant speed or may move at a non-constant speed, which is not limited by the present disclosure. And controlling the ground penetrating radar to collect data once when the distance measuring device moves for a set interval.

102, extracting the characteristics of the radar signal to be detected by adopting a first sub-network in the coal rock recognition model to obtain predicted characteristics; the prediction features are used for indicating the spatial distribution of the coal bed and the rock stratum in the coal rock stratum to be detected.

In the embodiment of the present disclosure, the coal rock recognition model is a trained model, and the coal rock recognition model learns the correspondence between the radar signal and the coal seam thickness.

In the embodiment of the disclosure, a first sub-network (or called lithology recognition sub-network) in a coal rock recognition model may be adopted to perform feature extraction on a radar signal to be detected to obtain a prediction feature; the prediction features are used for indicating the spatial distribution of the coal bed and the rock stratum in the coal rock stratum to be detected. Namely, the first sub-network can perform semantic segmentation on the radar signal to be detected to obtain an interface of the coal bed and the rock stratum.

And 103, predicting the coal seam thickness of the prediction characteristics by adopting a second sub-network in the coal rock recognition model to obtain a first thickness distribution of the coal seam in the coal rock layer to be detected on the detection path of the ground penetrating radar.

In the embodiment of the present disclosure, a second sub-network (or referred to as a coal thickness prediction sub-network) in the coal petrography recognition model may be used to predict the coal thickness of the prediction feature, so as to obtain the coal thickness of each position point on the detection path of the ground penetrating radar, that is, the thickness distribution of the coal seam in the coal petrography to be detected, which is referred to as a first thickness distribution in the present disclosure.

According to the coal seam thickness prediction method, the coal rock layer to be detected is detected by adopting the ground penetrating radar to obtain a radar signal to be detected; performing feature extraction on the radar signal to be detected by adopting a first sub-network in the coal rock recognition model to obtain a prediction feature; the prediction characteristics are used for indicating the spatial distribution of the coal bed and the rock stratum in the coal rock stratum to be detected; and predicting the coal seam thickness of the prediction characteristics by adopting a second sub-network in the coal rock recognition model so as to obtain the first thickness distribution of the coal seam in the coal rock layer to be detected on the detection path of the ground penetrating radar. Therefore, the thickness of the coal bed (namely the thickness distribution of the coal bed in the coal rock stratum to be detected) at each position point on the detection path of the ground penetrating radar is predicted based on the deep learning technology, on one hand, the accuracy of a prediction result can be improved, on the other hand, the prediction speed is high, and real-time prediction can be achieved. Moreover, compared with the method for calculating the thickness of the coal seam by using a physical formula in the prior art, the method and the device do not need to measure the relative dielectric constant of the coal seam, calculate parameters such as propagation speed and the like, and can reduce the workload of the prediction process. In addition, the thickness distribution of the coal bed can be directly predicted according to the radar signal to be measured without measuring the relative dielectric constant, and the automatic prediction of the coal bed thickness can be realized.

In order to clearly illustrate how the first thickness distribution of the coal seam in the coal rock layer to be detected on the detection path of the ground penetrating radar is predicted based on the coal rock recognition model in the above embodiment, the disclosure further provides a coal seam thickness prediction method.

Fig. 3 is a schematic flow chart of another method for predicting coal seam thickness according to an embodiment of the present disclosure.

As shown in fig. 3, the method for predicting the thickness of the coal seam may include the following steps:

and 301, detecting the coal rock layer to be detected by adopting a ground penetrating radar to obtain a radar signal to be detected.

For the explanation of step 301, reference may be made to relevant descriptions in any embodiment of the present disclosure, and details are not described herein.

Step 302, performing feature extraction on the radar signal to be detected by using a first feature extraction layer in a first sub-network in the coal rock recognition model to obtain a first intermediate feature.

In the embodiment of the disclosure, a first feature extraction layer in a first sub-network in a coal rock recognition model may be used to perform feature extraction on a radar signal to be detected to obtain a first intermediate feature.

Taking the structure of the coal rock identification model as illustrated in fig. 4 or fig. 5, the first feature extraction layer may include 3 convolutional layers and 2 max pooling layers, where the convolutional layers are used to extract features of the radar signal image to be detected, and reduce the number of parameters. The maximum pooling layer includes the following two functions: firstly, the amount of calculation is reduced, overfitting is prevented, and secondly, the receptive field is increased, so that the subsequent convolution kernels can learn more global information.

The radar signal image to be detected can be sequentially input into the convolution layer, the maximum pooling layer, the convolution layer, the maximum pooling layer and the convolution layer of the first feature extraction layer, so as to obtain a first intermediate feature output by the last convolution layer in the first feature extraction layer.

And step 303, performing feature extraction on the first intermediate features by using a second feature extraction layer in the first sub-network to obtain second intermediate features, wherein the scale of the second intermediate features is smaller than that of the first intermediate features.

In the embodiment of the present disclosure, feature extraction may be further performed on the first intermediate feature by using a second feature extraction layer in the first sub-network to obtain a second intermediate feature, where a scale of the second intermediate feature is smaller than that of the first intermediate feature.

Still taking the structure of the coal petrography recognition model as illustrated in fig. 4 or fig. 5, the second feature extraction layer may include 1 maximum pooling layer and 1 convolutional layer, and the first intermediate features may be sequentially input into the maximum pooling layer and the convolutional layer of the second feature extraction layer to obtain second intermediate features output by the convolutional layer of the second feature extraction layer, wherein the second intermediate features have a smaller scale than the first intermediate features. For example, referring to fig. 4 and 5, the first intermediate feature may be a 128 (width W) × 16 (height H) × 64 (channel C or feature dimension) stereo feature and the second intermediate feature may be a 64 × 8 × 128 stereo feature.

And step 304, performing feature extraction on the second intermediate features by using a third feature extraction layer in the first sub-network to obtain third intermediate features.

In an embodiment of the disclosure, feature extraction may be performed on the second intermediate features using a third feature extraction layer in the first sub-network to obtain third intermediate features, where the third intermediate features have a smaller scale than the second intermediate features.

Still taking the structure of the coal petrography recognition model as illustrated in fig. 4 or fig. 5, the third feature extraction layer may include 1 maximum pooling layer and 2 convolutional layers, and the second intermediate features may be sequentially input into the maximum pooling layer, the convolutional layer and the convolutional layer of the third feature extraction layer to obtain third intermediate features output by the last convolutional layer of the third feature extraction layer, wherein the third intermediate features have a smaller scale than the second intermediate features. For example, referring to fig. 4 and 5, the second intermediate feature may be a 64 × 8 × 128 stereo feature and the third intermediate feature may be a 32 × 4 × 128 stereo feature.

The third intermediate feature is upsampled 305 to obtain a fourth intermediate feature.

In embodiments of the present disclosure, the third intermediate feature may be upsampled to obtain a fourth intermediate feature.

Still taking the structure of the coal-rock identification model as illustrated in fig. 4 or 5, the third intermediate feature may be upsampled by an upsampling layer in the first sub-network, and the resulting fourth intermediate feature may be a 64 × 8 × 128 stereo feature.

Where the upsampling layer is used to restore features to a larger size for further computation.

And step 306, generating a prediction characteristic according to the fourth intermediate characteristic, wherein the prediction characteristic is used for indicating the spatial distribution of the coal bed and the rock stratum in the coal rock stratum to be detected.

In the embodiment of the present disclosure, a predicted feature may be generated according to the fourth intermediate feature, where the predicted feature is used to indicate spatial distribution of coal seams and rock strata in the coal rock stratum to be detected.

In a possible implementation manner of the embodiment of the present disclosure, the fourth intermediate feature and the second intermediate feature may be fused to obtain the first fused feature, for example, a scale of the fourth intermediate feature may be matched with a scale of the second intermediate feature, so that the scale-matched fourth intermediate feature and the second intermediate feature may be spliced to obtain the first fused feature. Therefore, the two characteristics with the same dimensionality are fused, and the deep network can be prevented from losing shallow characteristic information.

For example, taking the structure of the coal petrography recognition model as illustrated in fig. 4, the fourth intermediate feature may be fused with the second intermediate feature through a hopping connection layer in the first sub-network to obtain the first fused feature. And the jump connection layer is used for introducing the characteristic information on the corresponding scale into an up-sampling process so as to solve the problem of gradient disappearance.

Thereafter, feature extraction may be performed on the first fused feature using a fourth feature extraction layer in the first subnetwork to obtain a fifth intermediate feature.

Still taking the structure of the coal petrography recognition model as illustrated in fig. 4 for example, the fourth feature extraction layer may include 1 convolutional layer, and the first fused feature may be input into the convolutional layer of the fourth feature extraction layer to obtain a fifth intermediate feature output by the convolutional layer.

The fifth intermediate feature may then be upsampled to obtain a sixth intermediate feature. Still taking the structure of the coal-rock identification model as illustrated in fig. 4 as an example, the fifth intermediate feature may be upsampled by an upsampling layer in the first sub-network, and the obtained sixth intermediate feature may be a 128 × 16 × 64 stereo feature, for example.

The sixth intermediate feature may then be fused with the first intermediate feature to obtain a second fused feature, for example, the scale of the sixth intermediate feature may be matched with the scale of the first intermediate feature, so that the scale-matched sixth intermediate feature may be spliced with the first intermediate feature to obtain the second fused feature. Finally, a fifth feature extraction layer in the first subnetwork can be used to perform feature extraction on the second fused feature to obtain the predicted feature.

Still taking the structure of the coal petrography recognition model as illustrated in fig. 4 for example, the fifth feature extraction layer may include 1 convolutional layer, and the second fused feature may be input into the convolutional layer of the fifth feature extraction layer to obtain the predicted feature output by the convolutional layer.

In another possible implementation of the embodiment of the present disclosure, the fourth intermediate feature is feature-extracted by using a fourth feature extraction layer in the first sub-network to obtain a seventh intermediate feature.

For example, taking the structure of the coal rock recognition model as shown in fig. 5 as an example, the fourth feature extraction layer may include 1 convolutional layer, and the fourth intermediate features may be input into the convolutional layer of the fourth feature extraction layer to obtain seventh intermediate features output by the convolutional layer.

Thereafter, the seventh intermediate feature may be upsampled to obtain an eighth intermediate feature. Still taking the structure of the coal rock recognition model as exemplified in fig. 5, the eighth intermediate feature may be, for example, a 128 × 16 × 64 solid feature.

Finally, a fifth feature extraction layer in the first sub-network may be used to perform feature extraction on the eighth intermediate features to obtain the predicted features.

Still taking the structure of the coal petrography recognition model as illustrated in fig. 5 for example, the fifth feature extraction layer may include 1 convolution layer, and the eighth intermediate feature may be input into the convolution layer of the fifth feature extraction layer to obtain the predicted feature output by the convolution layer.

Therefore, the prediction characteristics used for indicating the spatial distribution of the coal bed and the rock stratum in the coal rock stratum to be detected can be extracted from the radar signals to be detected through the first sub networks with different structures, and the flexibility and the applicability of the method can be improved.

And 307, predicting the coal seam thickness of the prediction characteristics by using a second sub-network in the coal rock recognition model to obtain a first thickness distribution of the coal seam in the coal rock layer to be detected on the detection path of the ground penetrating radar.

For the explanation of step 307, reference may be made to the related description in any embodiment of the present disclosure, which is not described herein again.

In a possible implementation manner of the embodiment of the present disclosure, a sixth feature extraction layer in the second sub-network is used to perform feature extraction on the predicted feature to obtain a first candidate feature, where a scale of the first candidate feature is smaller than that of the predicted feature.

Taking the structure of the coal rock identification model as illustrated in fig. 4 or fig. 5, the sixth feature extraction layer may include 2 maximum pooling layers and 1 convolutional layer, and the predicted features may be sequentially input into the maximum pooling layer, the convolutional layer and the maximum pooling layer of the sixth feature extraction layer to obtain first candidate features output by the last maximum pooling layer of the sixth feature extraction layer, where a scale of the first candidate features is smaller than a scale of the predicted features. For example, the first candidate feature may be a stereo feature of 32 × 4 × 64.

Then, each element in the first candidate feature may be stretched or flattened to obtain a second candidate feature, that is, the multidimensional first candidate feature may be subjected to one-dimensional transformation to obtain the second candidate feature. Still taking the structure of the coal petrography recognition model as illustrated in fig. 4 or fig. 5, the first candidate feature may be flattened by a flattening layer in the second sub-network, and the resulting second candidate feature may be a one-dimensional feature of 8192.

Finally, the second candidate features may be sequentially input into the multiple fully-connected layers in the second sub-network to determine the first thickness distribution of the coal seam in the coal seam to be measured according to the output of the last fully-connected layer in the multiple fully-connected layers.

Still taking the structure of the coal rock identification model as illustrated in fig. 4 or fig. 5, the second sub-network may include 2 fully-connected layers, and the second candidate features may be sequentially input into the 2 fully-connected layers in the second sub-network, so as to determine the first thickness distribution of the coal bed in the coal rock layer to be measured according to the output of the last fully-connected layer.

Wherein the fully connected layer is used to map the learned feature representation to a label space of the sample.

The down-sampling module in fig. 4 and 5 is configured to extract features in the radar signal image to be detected to obtain lithological information and structural information; and the upsampling module is used for carrying out spatial reconstruction on the extracted lithology information so as to reflect lithology spatial distribution characteristics, namely spatial distribution of coal seams and rock stratums. And the second sub-network is used for extracting coal seam structure characteristics in the lithologic spatial distribution characteristics and mapping the coal seam structure characteristics into coal seam thickness distribution.

As an example, the image of the radar signal to be measured acquired by the ground penetrating radar may be as shown in fig. 6 (a), the spatial distribution of the coal seam and the rock formation in the coal formation to be measured, which is indicated by the prediction feature output by the first sub-network, may be as shown in fig. 6 (b), and the coal seam thickness distribution output by the second sub-network may be as shown in fig. 6 (c), where fig. 6 (c) is used to indicate the thickness of the coal seam at different positions on the detection path of the ground penetrating radar.

The horizontal axes of fig. 6 (a), 6 (b) and 6 (c) all represent the spatial positions of the position points on the detection path of the ground penetrating radar, the vertical axis of fig. 6 (a) represents time, for example, 0-10ns, in the example, 10ns is divided into 512 sampling points to correspond to the input size of the coal rock identification model, the vertical axis of fig. 6 (b) represents the height, and the vertical axis of fig. 6 (c) represents the thickness.

The method for predicting the thickness of the coal bed can predict the thickness of the coal bed in the coal rock layer to be tested based on the deep learning technology, and can improve the accuracy of a prediction result.

In order to clearly illustrate how the coal rock recognition model is trained in the above embodiments, the present disclosure also provides a training method of the coal rock recognition model.

Fig. 7 is a schematic diagram of a training process of a coal rock recognition model according to an embodiment of the present disclosure.

As shown in fig. 7, on the basis of the embodiment shown in fig. 1 or fig. 3, the coal rock recognition model can be obtained by training through the following steps:

and 701, detecting the sample coal rock layer by adopting a ground penetrating radar to obtain a sample radar signal.

In the embodiment of the present disclosure, the sample coal formation and the coal formation to be measured may be the same or different, and the present disclosure does not limit this. For example, the sample coal formation may be a coal formation with a known coal seam thickness in a certain region, and the coal formation to be measured may be a coal formation with an unknown coal seam thickness.

In the embodiment of the disclosure, a ground penetrating radar may be used to detect a sample coal formation to obtain a sample radar signal.

As an example, the radiation direction of the ground penetrating radar can be vertically upward and towards the sample coal stratum, the moving direction is the horizontal direction, and the moving speed is uniform movement.

And 702, measuring the thickness of the coal seam at each position point on the detection path of the ground penetrating radar to obtain a second thickness distribution.

In the embodiment of the disclosure, the thickness of the coal seam at each position point on the detection path of the ground penetrating radar can be measured by using the related measurement device, so as to obtain the second thickness distribution of the coal seam.

As an example, some coal strata are exposed, for example, when mining equipment is used for mining coal in the previous time, part of the strata is cut or naturally falls off to expose the front coal strata, and at this time, the boundary between the coal seam and the strata can be directly observed visually, so that the thickness of the coal seam can be directly measured through a measuring ruler, and the second thickness distribution of the coal seam can be obtained.

As another example, a core drill may be used to core the coal seam at set intervals (e.g., 0.5 meters) along the path of the ground penetrating radar survey to obtain a second thickness profile for the coal seam.

And 703, marking the sample radar signal according to the second thickness distribution to obtain a training sample.

In an embodiment of the present disclosure, the sample radar signal may be labeled according to the second thickness distribution to obtain a training sample. That is, the sample radar signal may be used as an input to the model, and the second thickness distribution may be used as annotation information for the sample radar signal.

Step 704, inputting the training sample into the initial coal rock recognition model to obtain a third thickness distribution output by the initial coal rock recognition model.

In the embodiment of the present disclosure, the training sample may be input into the initial coal rock recognition model to obtain a third thickness distribution output by the initial coal rock recognition model. That is, the first sub-network in the initial coal rock identification model may be used to perform feature extraction on the sample radar signal to obtain predicted sample features, and the second sub-network in the initial coal rock model may be used to perform coal seam thickness prediction on the predicted sample features to obtain the coal seam thickness at each position on the detection path of the ground penetrating radar, that is, the third thickness distribution of the coal seam in the sample coal rock layer.

Step 705, training the initial coal and rock recognition model according to the third thickness distribution and the second thickness distribution to obtain a trained coal and rock recognition model.

In this embodiment of the present disclosure, the initial coal petrography recognition model may be trained according to a difference between the third thickness distribution and the second thickness distribution to obtain a trained coal petrography recognition model.

As an example, a calculated value of the loss function (denoted as a loss value in this disclosure) may be determined from a difference between the third thickness distribution and the second thickness distribution, where the loss function may include, but is not limited to, a square loss function, a mean square error loss function, a root mean square error loss function, a smoothed mean absolute error loss function, an absolute value loss function, a mean absolute percentage error loss function, a root mean square Log error loss function, an L1 norm (Huber) loss function, a regression (Log-Cosh) loss function, a Quantile (Quantile) loss function, and so on.

Therefore, in the method and the device, the model parameters in the initial coal rock recognition model can be adjusted according to the loss value, so that the loss value is minimized, and the trained coal rock recognition model is obtained.

It should be noted that, the above example is performed by taking only the termination condition of the model training as the minimization of the loss value, and in practical application, other termination conditions may also be set, for example, the number of times of training reaches the set number of times, the training duration reaches the set duration, the loss value converges, and the like, which is not limited by the present disclosure.

In conclusion, the trained coal petrography recognition model is obtained by training the initial coal petrography recognition model in advance, so that the thickness distribution of the coal bed is predicted by adopting the trained coal petrography recognition model, and the accuracy of the prediction result can be improved.

When the number of the training samples is multiple, in order to clearly illustrate how the initial coal rock recognition model is trained in the above embodiment, the present disclosure further provides a training method of the coal rock recognition model.

Fig. 8 is a schematic diagram of a training process of another coal rock recognition model according to an embodiment of the present disclosure.

As shown in fig. 8, on the basis of the embodiment shown in fig. 1 or fig. 3, the coal rock recognition model can be obtained by training through the following steps:

step 801, detecting a sample coal rock layer by using a ground penetrating radar to obtain a sample radar signal.

And 802, measuring the thickness of the coal seam at each position point on the detection path of the ground penetrating radar to obtain a second thickness distribution.

And 803, labeling the sample radar signal according to the second thickness distribution to obtain a training sample.

For the explanation of steps 801 to 803, reference may be made to the related description in the above embodiments, which is not repeated herein.

And step 804, respectively inputting the plurality of training samples into the initial coal rock recognition model to obtain third thickness distribution of the plurality of training samples output by the initial coal rock recognition model.

In this embodiment of the present disclosure, a plurality of training samples may be respectively input to the initial coal rock recognition model to obtain a third thickness distribution of the plurality of training samples output by the initial coal rock recognition model. That is, for any one of the training samples, the training sample may be input into the initial coal rock recognition model to obtain a third thickness distribution of the training sample output by the initial coal rock recognition model. Specifically, feature extraction may be performed on the sample radar signal in the training sample by using a first sub-network in the initial coal-rock recognition model to obtain a predicted sample feature, and a coal seam thickness prediction may be performed on the predicted sample feature by using a second sub-network in the initial coal-rock model to obtain a third thickness distribution of the coal seam in the sample coal rock layer on the ground penetrating radar detection path.

Step 805, training the initial coal rock recognition model according to the third thickness distributions marked on the plurality of training samples and the second thickness distributions of the plurality of training samples to obtain a trained coal rock recognition model.

In the embodiment of the disclosure, the initial coal rock recognition model may be trained according to the second thickness distributions marked on the plurality of training samples and the third thickness distributions of the plurality of training samples, so as to obtain a trained coal rock recognition model.

As a possible implementation manner, for any training sample in the plurality of training samples, a difference between the third thickness distribution and the second thickness distribution corresponding to the training sample may be determined, and a calculated value of the loss function (referred to as a loss value in this disclosure) may be determined according to the difference between the third thickness distribution and the second thickness distribution of the plurality of training samples, where the loss function may include, but is not limited to, a square loss function, a mean square error loss function, an absolute value loss function, and the like.

For example, when the loss function is a squared loss function, the loss value L may be determined by the following equation:

wherein, y _i Second thickness distribution, x, for the ith training sample _i For the ith training sample, f (x) _i ) And outputting a third thickness distribution of the ith training sample for the initial coal rock recognition model, wherein n is the number of the training samples.

For another example, when the loss function is a mean square error loss function, the loss value L may be determined by the following equation:

for another example, when the loss function is an absolute value loss function, the loss value L may be determined by the following formula:

According to the method and the device, the spatial distribution of the coal bed and the rock stratum is directly predicted through a deep learning technology, and the thickness of the coal bed is calculated, so that the automatic identification and prediction of the boundary of the coal bed and the rock stratum and the thickness of the coal bed can be realized. After the model training is finished, the prediction of the coal bed is automatically finished by the electronic equipment with computing capability without manual participation, and the coal rock intelligent identification requirement of unmanned intelligent mining of the mine is met. In addition, through a deep learning method, the coal rock horizon selection and dielectric constant correction method is optimized, and the prediction result is more accurate.

Compared with the first mode in the related art, in the prediction stage, parameters such as the relative dielectric constant of the coal seam, the propagation speed of the electromagnetic wave in the measured coal seam and the like do not need to be measured, instead, the deep learning technology is used for extracting the characteristic information in the radar signal to be measured, and the spatial distribution of geologic bodies with different relative dielectric constants can be directly predicted according to the characteristic information, so that the division or identification of the interfaces of the coal seam and the rock stratum and the prediction of the thickness of the coal seam are realized.

Compared with the second mode in the related art, in the prediction stage, the deep learning technology is adopted to directly predict the coal seam thickness without manual participation or acquiring the relative dielectric constant parameters of the drilling map, so that the prediction workload can be reduced, and the manual participation degree can be reduced.

Specifically, the prediction of the coal seam thickness can be realized through the following steps:

step A, collecting a training set for training a coal rock recognition neural network (which is recorded as a coal rock recognition model in the present disclosure).

In the step A1, the radiation direction of the ground penetrating radar may be vertically upward and toward a coal rock layer (for example, a sample coal rock layer), the moving direction is a horizontal direction, and the moving mode may be uniform movement or non-uniform movement. As an example, the movement mode and the movement direction of the ground penetrating radar may be as shown in fig. 2 (a), and the ground penetrating radar may move a certain distance at a constant horizontal speed to collect radar signals (denoted as sample radar signals in this disclosure). As another example, the ground penetrating radar may move in the direction shown in fig. 2 (b), and the ground penetrating radar may be connected to the ranging device, wherein the ranging device may move at a constant speed or may move at a non-constant speed. And controlling the ground penetrating radar to collect data once when the distance measuring device moves for a set interval.

And A2, measuring the thickness of the coal seam at each position point on the ground penetrating radar detection path to obtain the thickness distribution of the coal seam, which is marked as second thickness distribution in the disclosure.

And step A3, marking the sample radar signals by adopting the second thickness distribution to obtain training data (marked as a training sample in the disclosure), namely, taking the second thickness distribution of the coal seam as a prediction target and taking the sample radar signals as input of the coal rock recognition model.

And step A4, repeating the steps A1-A3, collecting not less than 1000 pieces of training data, and generating a training set according to the collected training data, wherein the number of the training data in the training set is 90%, and the rest 10% of the training data is used as a verification set.

And step B, constructing a coal rock recognition model.

Constructing a coal petrography recognition model consisting of a first sub-network (or called lithology recognition sub-network) and a second sub-network (or called coal bed thickness prediction sub-network), and setting model parameters (or called network parameters).

As an example, the structure of the coal petrography recognition model may be as shown in fig. 4 or fig. 5, for example. The first sub-network is used for extracting features of the sample radar signals, wherein the features are used for indicating the spatial distribution of the coal seam and the rock stratum, namely the first sub-network can carry out semantic segmentation on the sample radar signals to obtain interfaces of the coal seam and the rock stratum. The second sub-network consists of a maximum pooling layer, a rolling layer, a flattening layer and a full-connection layer, and the thickness (denoted as a third thickness distribution in the disclosure) of the coal seam at each position point on the ground penetrating radar collecting path or the detecting path is output.

And step C, training a coal rock recognition model.

And inputting training data in the training set into the coal rock recognition model, and training the coal rock recognition model for not less than 100 times by adopting a random gradient descent method, an RMSProp (root mean square propagation gradient descent) optimization method, an Adam (Adaptive Moment Estimation) optimization method, an Ftrl (Follow-the-regulated-Leader) optimization method and the like to obtain the trained coal rock recognition model.

And D, deploying a coal rock recognition model.

And transmitting the trained coal rock recognition model to a storage medium of computer equipment under the mine.

And E, collecting radar signals to be detected of the coal rock layer to be detected.

And connecting the ground penetrating radar with computer equipment under the mine, and acquiring a radar signal to be detected in real time through the computer equipment.

And F, predicting and outputting the thickness of the coal bed.

And inputting the acquired radar signal to be detected into the trained coal rock recognition model, predicting and outputting the thickness distribution (recorded as first thickness distribution in the disclosure) of the coal bed at each position point on the ground penetrating radar acquisition path or the detection path by using the coal rock recognition model.

As an example, the structure of the coal rock identification model is illustrated as fig. 4, and the coal rock identification model may be composed of 9 convolutional layers, 6 pooling layers, 2 upsampling layers, 2 jump connection layers, 1 flattening layer, and 2 full connection layers.

As another example, the structure of the coal rock identification model is illustrated as shown in fig. 5, and the coal rock identification model may be composed of 9 convolutional layers, 6 pooling layers, 2 upsampling layers, 1 flattening layer, and 2 fully-connected layers.

The convolution layer convolution kernel size can be 3 multiplied by 3, the step length is all 1, and the filling parameter is the filling and space dimension reservation; the sizes of the maximum pooling layer and the up-sampling layer can be 2 multiplied by 2, the step length of the maximum pooling layer is 1, and the maximum pooling layer is not filled; the size of the 2 fully connected layers is 1024 and 160 (or 1024 and n), respectively, where n is the number of sample points on the probe path.

Optionally, 1 Dropout layer may also be provided after the flattening layer and after the first fully connected layer, respectively, with the loss rate set to 0.4. The purpose of the Dropout layer is to mitigate the occurrence of overfitting and to some extent to achieve the regularization effect.

The last full-connection layer of the coal rock recognition model does not use an activation function, and the rest of the convolutional layers and the full-connection layers can adopt the following activation functions: sigmoid function, hyperbolic tangent function (Tanh function), linear rectification function (ReLU function), leakage ReLU function, ELU function, etc.

Alternatively, the coal petrography recognition model may employ a mean square error as a loss function, and optimize the loss function using an SGD (Stochastic Gradient Descent) optimizer.

In summary, a deep learning method is used to construct a coal rock recognition neural network (which is recorded as a coal rock recognition model in the present disclosure), and a radar signal (such as a radar signal to be detected) acquired by a ground penetrating radar in a coal rock recognition scene is processed or recognized based on the coal rock recognition model to obtain the coal seam thickness, and the automatic prediction of the coal seam thickness can be realized without manual operation and without predicting or measuring the coal seam lithology (such as a relative dielectric constant) in the prediction process. The first sub-network of the coal petrography recognition model (or called lithology recognition sub-network) is used for predicting the spatial distribution of the coal seam and the rock stratum, and the first sub-network can be a neural network structure similar to a 'UNet' structure or a neural network structure similar to a 'SegNet' structure. A second sub-network of the coal petrography identification model (or called a sub-network of coal thickness prediction), which may be a sequential structure, is connected after the first sub-network (sub-network of lithology identification), is used to predict the thickness distribution of the coal seam.

Compared with the scheme in the related art, the method has the following advantages that:

firstly, the automatic prediction of the coal seam thickness can be realized, the deep learning technology is used, the characteristics in the radar signal to be detected are directly extracted to predict the coal seam thickness, and the radar signal to be detected does not need to be preprocessed and analyzed manually;

secondly, in the prediction stage, the model can output the thickness distribution of the coal seam only by inputting the radar signal to be detected, compared with the traditional method that the thickness of the coal seam is calculated by using a physical formula and the relative dielectric constant of the coal seam needs to be measured, the workload in the prediction process can be reduced, and the thickness distribution (namely, the first thickness distribution) of the coal seam can be directly predicted according to the radar signal to be detected without measuring the relative dielectric constant, so that the automatic prediction of the thickness of the coal seam can be realized;

thirdly, the prediction calculation speed is high, and real-time prediction can be realized. The reason is that: the trained model is adopted to predict the thickness of the coal seam, the required computing resources are less, and the traditional method (such as full waveform inversion) needs a larger amount of computation.

It should be noted that, although it is possible to add a dielectric constant tester to the acquisition system based on the first mode in the related art, so as to achieve accurate calculation of the coal rock horizon. However, the measurement of the dielectric constant requires sampling, and real-time and in-situ measurement cannot be realized. Moreover, the quality of the radar signal is influenced by a plurality of field factors, and the coal rock horizon is difficult to automatically select only by a computer program, so that the application difficulty is high.

In addition, it is now possible to predict the permittivity distribution image of a tunnel geologic body based on a neural network, and predict the spatial distribution of steel reinforcements, water-free fissures, and water-containing defects in the tunnel geologic body from the permittivity distribution image. Although the method can be popularized to the field of coal rock identification, the interface of the coal bed and the rock stratum is identified based on the neural network, so that the thickness of the coal bed is predicted according to the interface of the coal bed and the rock stratum. However, such a neural network is very complicated in training data acquisition, and because the prediction result is a dielectric constant distribution diagram image of a geologic body, when a training set is made, not only the thickness of the coal seam but also the dielectric constants of the coal seam and the rock stratum at different positions need to be measured. However, if the dielectric constant is freely defined through computer modeling and the radar signal is obtained through model calculation, although the acquisition difficulty of acquiring training data can be reduced, the acquired training data cannot reflect the influence of the real world on the radar signal, and the prediction effect is poor.

Corresponding to the method for predicting the thickness of the coal seam provided in the embodiments of fig. 1 to 8, the present disclosure also provides a device for predicting the thickness of the coal seam, and since the device for predicting the thickness of the coal seam provided in the embodiments of the present disclosure corresponds to the method for predicting the thickness of the coal seam provided in the embodiments of fig. 1 to 8, the implementation manner of the method for predicting the thickness of the coal seam is also applicable to the device for predicting the thickness of the coal seam provided in the embodiments of the present disclosure, and will not be described in detail in the embodiments of the present disclosure.

As shown in fig. 9, the coal seam thickness prediction apparatus 900 may include: a first detection module 901, an extraction module 902, and a prediction module 903.

The first detection module 901 is configured to detect a coal rock layer to be detected by using a ground penetrating radar to obtain a radar signal to be detected.

An extraction module 902, configured to perform feature extraction on a radar signal to be detected by using a first sub-network in a coal rock recognition model to obtain a prediction feature; the prediction features are used for indicating the spatial distribution of the coal bed and the rock stratum in the coal rock stratum to be detected.

And the predicting module 903 is used for predicting the coal seam thickness of the predicted characteristic by using a second sub-network in the coal rock recognition model so as to obtain a first thickness distribution of the coal seam in the coal rock layer to be detected on the detection path of the ground penetrating radar.

In a possible implementation manner of the embodiment of the present disclosure, the extracting module 902 is specifically configured to: performing feature extraction on the radar signal to be detected by adopting a first feature extraction layer in a first sub-network to obtain a first intermediate feature; performing feature extraction on the first intermediate features by adopting a second feature extraction layer in the first sub-network to obtain second intermediate features, wherein the scale of the second intermediate features is smaller than that of the first intermediate features; performing feature extraction on the second intermediate features by using a third feature extraction layer in the first sub-network to obtain third intermediate features, wherein the scale of the third intermediate features is smaller than that of the second intermediate features; upsampling the third intermediate feature to obtain a fourth intermediate feature; and generating a predicted feature according to the fourth intermediate feature.

In a possible implementation manner of the embodiment of the present disclosure, the extracting module 902 is specifically configured to: fusing the fourth intermediate feature with the second intermediate feature to obtain a first fused feature; performing feature extraction on the first fusion feature by using a fourth feature extraction layer in the first sub-network to obtain a fifth intermediate feature; upsampling the fifth intermediate feature to obtain a sixth intermediate feature, wherein the scale of the sixth intermediate feature matches the scale of the first intermediate feature; fusing the sixth intermediate feature with the first intermediate feature to obtain a second fused feature; and performing feature extraction on the second fusion feature by using a fifth feature extraction layer in the first sub-network to obtain a predicted feature.

In a possible implementation manner of the embodiment of the present disclosure, the extracting module 902 is specifically configured to: performing feature extraction on the fourth intermediate features by using a fourth feature extraction layer in the first sub-network to obtain seventh intermediate features; upsampling the seventh intermediate feature to obtain an eighth intermediate feature; and performing feature extraction on the eighth intermediate feature by using a fifth feature extraction layer in the first sub-network to obtain a predicted feature.

In a possible implementation manner of the embodiment of the present disclosure, the prediction module 903 is specifically configured to: performing feature extraction on the predicted feature by adopting a sixth feature extraction layer in the second sub-network to obtain a first candidate feature, wherein the scale of the first candidate feature is smaller than that of the predicted feature; stretching each element in the first candidate feature to obtain a second candidate feature; and sequentially inputting the second candidate features into the multiple fully-connected layers in the second sub-network so as to determine the first thickness distribution of the coal bed in the coal rock layer to be detected according to the output of the last fully-connected layer in the multiple fully-connected layers.

In a possible implementation manner of the embodiment of the present disclosure, the coal rock recognition model is obtained by training through the following modules:

the second detection module is used for detecting the sample coal rock layer by adopting a ground penetrating radar to obtain a sample radar signal;

the measuring module is used for measuring the thickness of the coal seam at each position point on the detection path of the ground penetrating radar to obtain second thickness distribution;

the marking module is used for marking the sample radar signal according to the second thickness distribution so as to obtain a training sample;

the input module is used for inputting the training sample into the initial coal rock recognition model so as to obtain a third thickness distribution output by the initial coal rock recognition model;

and the training module is used for training the initial coal rock recognition model according to the third thickness distribution and the second thickness distribution so as to obtain a trained coal rock recognition model.

In a possible implementation manner of the embodiment of the present disclosure, the number of training samples is multiple, and the training module is specifically configured to: determining a difference between a third thickness distribution and a second thickness distribution corresponding to any training sample in the plurality of training samples; generating a loss value according to the difference of the plurality of training samples; and adjusting model parameters in the initial coal rock identification model according to the loss value so as to minimize the loss value.

In order to implement the foregoing embodiments, the present disclosure further provides an electronic device, where the electronic device may be any device with computing capability, and the electronic device includes: the device comprises a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein the processor executes the program to realize the coal seam thickness prediction method provided by any one of the previous embodiments of the disclosure.

To achieve the above embodiments, the present disclosure also proposes a non-transitory computer-readable storage medium having stored thereon a computer program which, when executed by a processor, implements a method of predicting a thickness of a coal seam as proposed in any of the preceding embodiments of the present disclosure.

To achieve the above embodiments, the present disclosure further provides a computer program product, wherein when the instructions in the computer program product are executed by a processor, the method for predicting coal seam thickness as set forth in any of the previous embodiments of the present disclosure is performed.

In the description of the present specification, reference to the description of "one embodiment," "some embodiments," "an example," "a specific example," or "some examples" or the like means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the present disclosure. In this specification, the schematic representations of the terms used above are not necessarily intended to refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, various embodiments or examples and features of different embodiments or examples described in this specification can be combined and combined by one skilled in the art without contradiction.

Furthermore, the terms "first", "second" and "first" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of the feature. In the description of the present disclosure, "a plurality" means at least two, e.g., two, three, etc., unless explicitly specifically limited otherwise.

Any process or method descriptions in flow charts or otherwise described herein may be understood as representing modules, segments, or portions of code which include one or more executable instructions for implementing steps of a custom logic function or process, and alternate implementations are included within the scope of the preferred embodiment of the present disclosure in which functions may be executed out of order from that shown or discussed, including substantially concurrently or in reverse order, depending on the functionality involved, as would be understood by those reasonably skilled in the art of the embodiments of the present disclosure.

The logic and/or steps represented in the flowcharts or otherwise described herein, e.g., an ordered listing of executable instructions that can be considered to implement logical functions, can be embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, processor-containing system, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions. For the purposes of this description, a "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of the computer-readable medium would include the following: an electrical connection (electronic device) having one or more wires, a portable computer diskette (magnetic device), a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium could even be paper or another suitable medium upon which the program is printed, as the program can be electronically captured, via for instance optical scanning of the paper or other medium, then compiled, interpreted or otherwise processed in a suitable manner if necessary, and then stored in a computer memory.

It should be understood that portions of the present disclosure may be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, the various steps or methods may be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. If implemented in hardware, as in another embodiment, any one or combination of the following techniques, which are well known in the art, may be used: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, an application specific integrated circuit having an appropriate combinational logic gate circuit, a Programmable Gate Array (PGA), a Field Programmable Gate Array (FPGA), or the like.

It will be understood by those skilled in the art that all or part of the steps carried by the method for implementing the above embodiments may be implemented by hardware related to instructions of a program, which may be stored in a computer readable storage medium, and when the program is executed, the program includes one or a combination of the steps of the method embodiments.

In addition, functional units in the embodiments of the present disclosure may be integrated into one processing module, or each unit may exist alone physically, or two or more units are integrated into one module. The integrated module can be realized in a hardware mode, and can also be realized in a software functional module mode. The integrated module, if implemented in the form of a software functional module and sold or used as a stand-alone product, may also be stored in a computer readable storage medium.

The storage medium mentioned above may be a read-only memory, a magnetic or optical disk, etc. Although embodiments of the present disclosure have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present disclosure, and that changes, modifications, substitutions and alterations may be made to the above embodiments by those of ordinary skill in the art within the scope of the present disclosure.

Claims

1. A method for predicting coal seam thickness, the method comprising:

detecting the coal and rock layer to be detected by adopting a ground penetrating radar to obtain a radar signal to be detected;

2. The method according to claim 1, wherein the performing feature extraction on the radar signal to be tested by using a first sub-network in the coal-rock recognition model to obtain a predicted feature comprises:

performing feature extraction on the radar signal to be detected by using a first feature extraction layer in the first sub-network to obtain a first intermediate feature;

performing feature extraction on the first intermediate features by using a second feature extraction layer in the first sub-network to obtain second intermediate features, wherein the second intermediate features have a smaller scale than the first intermediate features;

performing feature extraction on the second intermediate features by using a third feature extraction layer in the first sub-network to obtain third intermediate features, wherein the scale of the third intermediate features is smaller than that of the second intermediate features;

upsampling the third intermediate feature to obtain a fourth intermediate feature;

generating the predicted feature based on the fourth intermediate feature.

3. The method of claim 2, wherein the generating the predicted feature from the fourth intermediate feature comprises:

fusing the fourth intermediate feature with the second intermediate feature to obtain a first fused feature;

performing feature extraction on the first fusion feature by using a fourth feature extraction layer in the first sub-network to obtain a fifth intermediate feature;

upsampling the fifth intermediate feature to obtain a sixth intermediate feature, wherein a scale of the sixth intermediate feature matches a scale of the first intermediate feature;

fusing the sixth intermediate feature with the first intermediate feature to obtain a second fused feature;

and performing feature extraction on the second fusion feature by using a fifth feature extraction layer in the first sub-network to obtain the predicted feature.

4. The method of claim 2, wherein the generating the predicted feature from the fourth intermediate feature comprises:

performing feature extraction on the fourth intermediate feature by using a fourth feature extraction layer in the first sub-network to obtain a seventh intermediate feature;

upsampling the seventh intermediate feature to obtain an eighth intermediate feature;

and performing feature extraction on the eighth intermediate feature by using a fifth feature extraction layer in the first sub-network to obtain the predicted feature.

5. The method of claim 1, wherein the predicting the coal seam thickness of the predicted feature by using a second sub-network in the coal petrography recognition model to obtain a first thickness distribution of the coal seam in the coal petrography to be detected on the detection path of the ground penetrating radar comprises:

performing feature extraction on the predicted feature by using a sixth feature extraction layer in the second sub-network to obtain a first candidate feature, wherein the scale of the first candidate feature is smaller than that of the predicted feature;

stretching each element in the first candidate feature to obtain a second candidate feature;

and sequentially inputting the second candidate features into the multiple fully-connected layers in the second sub-network, so as to determine the first thickness distribution of the coal bed in the coal stratum to be detected according to the output of the last fully-connected layer in the multiple fully-connected layers.

6. The method of claim 1, wherein the coal rock recognition model is trained by:

detecting the sample coal rock layer by adopting the ground penetrating radar to obtain a sample radar signal;

measuring the thickness of the coal seam at each position point on the ground penetrating radar detection path to obtain a second thickness distribution;

labeling the sample radar signal according to the second thickness distribution to obtain a training sample;

inputting the training sample into an initial coal rock recognition model to obtain a third thickness distribution output by the initial coal rock recognition model;

and training the initial coal and rock recognition model according to the third thickness distribution and the second thickness distribution to obtain the trained coal and rock recognition model.

7. The method according to claim 6, wherein the number of the training samples is plural, and the training of the initial coal petrography recognition model according to the third thickness distribution and the second thickness distribution comprises:

for any training sample in a plurality of training samples, determining the difference between a third thickness distribution and a second thickness distribution corresponding to the training sample;

generating a loss value according to the difference of the plurality of training samples;

and adjusting model parameters in the initial coal rock identification model according to the loss value so as to minimize the loss value.

8. An apparatus for predicting the thickness of a coal seam, the apparatus comprising:

the detection module is used for detecting the coal rock layer to be detected by adopting a ground penetrating radar to obtain a radar signal to be detected;

the extraction module is used for extracting the characteristics of the radar signal to be detected by adopting a first sub-network in the coal rock recognition model so as to obtain predicted characteristics; the prediction features are used for indicating the spatial distribution of the coal bed and the rock stratum in the coal rock stratum to be detected;

9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the method according to any of claims 1-7 when executing the program.

10. A non-transitory computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the method according to any one of claims 1 to 7.