Coal mine drainage and drainage capacity prediction and drill site optimization design method based on neural network

Abstract

The invention discloses a neural network-based coal mine drainage and drainage quantity prediction and drill site optimization design method, which comprises the following steps of: deeply analyzing stratum characteristics such as sand shale proportion and lithology category by introducing a space occurrence characteristic quantitative model and a convolutional neural network; historical drainage water volume data and real-time monitoring information are dynamically learned in combination with the bidirectional long-short-term memory network, and the prediction precision is remarkably improved. Meanwhile, a drill site adaptive optimization module is constructed based on a reinforcement learning algorithm, and the drill hole spacing, depth and layout mode are automatically generated according to a prediction result. According to the method, sand shale category division and space occurrence characteristics are incorporated into a neural network input layer, and the problem that lithology heterogeneity modeling is insufficient in a traditional method is solved; and through dynamic coupling prediction and design links, real-time optimization of drilling parameters is realized, and the universality and flexibility of the technology are remarkably improved. The invention provides an efficient, accurate and economical technical means for coal mine water disaster prevention and control.

Description

Technical Field

[0001] The present invention relates to the technical field of coal mining, and in particular to a method for predicting coal mine drainage volume and optimizing drilling site design based on a neural network. Background Art

[0002] Coal occupies a pivotal position in my country's energy structure and is a key energy source for ensuring national energy security and stable economic development. For a long time, coal, as my country's primary energy source, has provided indispensable power for numerous sectors, including industrial production and power supply. With the continued development of the economy, the demand for coal resources has also been increasing, leading to an increase in the scale and depth of coal mining. However, as mining deepens, the geological conditions facing coal mining become increasingly complex, and water hazards have become increasingly prominent, becoming a major constraint on safe and efficient coal mining.

[0003] During coal mining, the complex geological structure, rock properties, and other factors combined with various water sources—groundwater, spent water, and surface water—are highly susceptible to flooding. These floods not only severely damage coal mine production equipment but also threaten the lives of miners, resulting in significant economic losses and social impacts for coal mining companies. To effectively prevent and control flooding accidents and ensure safe coal mining operations, drainage is crucial. Drainage reduces the water pressure and level in aquifers, minimizing the likelihood of flooding and creating a safe working environment for coal mining.

[0004] During coal mining, drainage of aquifers in the roof and floor of coal seams is a core measure for preventing and controlling water hazards. The purpose of drainage is to reduce aquifer pressure through drilling and prevent water inrush accidents. However, the successful implementation of drainage projects depends on accurate water volume prediction and scientific drill site design. Traditional drainage volume prediction methods rely primarily on empirical formulas (such as the large well method and analytical methods) or static geological models, which are difficult to adapt to complex and changing geological conditions. For example, in interbedded sandstone and mudstone formations, the permeability of sandstone and mudstone differs significantly. Traditional methods cannot accurately quantify the impact of the sandstone-mudstone ratio and its classification (such as the permeability difference between brittle sandstone and plastic mudstone) on drainage volume, resulting in significant deviations in prediction results. Furthermore, existing drill site designs often use fixed borehole spacing and layout patterns, lacking dynamic analysis of the spatial distribution characteristics of the coal seam roof and floor (such as heterogeneity and the degree of fracture development). This leads to irrational borehole layout, low drainage efficiency, and severe waste of resources.

[0005] At present, the main problems of drilling site design are as follows:

[0006] Insufficient geological parameter modeling: Traditional methods are insufficient in quantitative analysis of the sandstone-mudstone ratio and its classification, making it difficult to accurately reflect the impact of lithologic heterogeneity on drainage paths.

[0007] Poor design flexibility: The existing solution cannot dynamically adjust drilling parameters based on real-time geological data (such as the thickness of the sandstone-mudstone interlayer and changes in the lithologic interface), resulting in unsatisfactory drainage effects.

[0008] Lack of multi-factor coupling: Traditional methods do not fully consider the synergistic effect of sandstone and mudstone classification and hydrogeological conditions, and lack accurate modeling of drainage paths under complex geological conditions.

[0009] Low level of intelligence: Existing technologies mostly rely on manual experience and lack data-driven intelligent optimization methods, making it difficult to adapt to water drainage needs under complex geological conditions.

[0010] Existing technologies have yet to address the deep integration of sandstone and mudstone classification, spatial distribution characteristics, and dynamic hydrological monitoring data. They also lack an intelligent method for integrated optimization of drainage volume prediction and borehole placement. Therefore, a technical solution that overcomes these limitations and offers both high precision and flexibility is urgently needed to address the needs of coal mine water hazard prevention and control under complex geological conditions. Summary of the Invention

[0011] In response to the above-mentioned technical deficiencies, the purpose of the present invention is to provide a coal mine drainage volume prediction and drilling site optimization design method based on a neural network, which can solve the shortcomings of traditional drilling site design methods in terms of geological parameter limitations, design rigidity and lack of multi-factor coupling. This method incorporates sandstone and mudstone classification and spatial occurrence characteristics into the neural network input layer, solving the problem of insufficient modeling of lithologic heterogeneity by traditional methods; through dynamic coupling prediction and design links, real-time optimization of drilling parameters is achieved, significantly improving the universality and flexibility of the technology.

[0012] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0013] The present invention provides a method for predicting the amount of water discharged from a coal mine and optimizing the design of a drilling site based on a neural network, which specifically includes the following steps:

[0014] (1) Data collection: Collect multidimensional data of the mining area, including geological exploration data, spatial distribution characteristics of coal seam roof and floor, sandstone and mudstone ratio and lithology classification data, historical drainage volume data and real-time hydrological monitoring information;

[0015] (2) Data preprocessing: preprocessing the collected data, including data cleaning, denoising and standardization, to ensure the accuracy and consistency of the data;

[0016] (3) Feature quantification: Introducing a spatial occurrence feature quantification model to convert the heterogeneity of the coal seam roof and floor and the degree of fracture development into quantifiable indicators;

[0017] (4) Model construction: The indicators of sandstone and mudstone classification and spatial distribution characteristics are incorporated into the neural network input layer, and a hybrid neural network model consisting of a convolutional neural network and a bidirectional long-short-term memory network is constructed. The convolutional neural network is used to analyze the stratigraphic characteristics, and the hybrid neural network model of the bidirectional long-short-term memory network is used to learn the temporal dynamic characteristics of historical and real-time drainage volume data;

[0018] (5) Model training: training the hybrid neural network model based on the preprocessed data and optimizing the model parameters to minimize the prediction error;

[0019] (6) Model evaluation and optimization: Use cross-validation method to evaluate and optimize the trained hybrid neural network model to ensure that the model has good generalization ability and stability;

[0020] (7) Drilling site optimization design: A reinforcement learning algorithm is used to build a drilling site adaptive optimization module, which automatically generates the drilling hole spacing, depth, and layout pattern based on the drainage volume prediction results output by the neural network;

[0021] (8) Dynamic coupling: Dynamic coupling of water discharge prediction and drilling site optimization design, adjusting drilling parameters according to real-time geological conditions.

[0022] Preferably, during the data collection phase in step (1), the geological exploration data covers detailed information at different depths and in different areas, and is regularly updated daily to reflect the dynamic changes in the geological conditions of the mining area.

[0023] Preferably, in step (1), the spatial occurrence characteristic data includes the thickness of the sand-mudstone interlayer, the change of the lithologic interface, and the thin mudstone layer merging threshold, wherein the thin mudstone layer merging threshold is a set thickness value. When the thickness is lower than the threshold, the thin mudstone layer is merged with other thick rock layers. If the thickness of the mudstone layer is less than 1m and the ratio of sandstone to mudstone is greater than 5:1, the mudstone is classified as sandstone.

[0024] Preferably, in step (5), the model training adopts a dynamic learning rate adjustment strategy, which automatically adjusts the learning rate according to the training progress to accelerate convergence.

[0025] Preferably, the reinforcement learning algorithm in step (7) adopts a particle swarm optimization algorithm, which searches for the optimal drilling parameter combination in a high-dimensional solution space through a particle swarm, and the objective function is to maximize the drainage efficiency and minimize the drilling cost.

[0026] Preferably, when the drilling parameters are adjusted in real time in step (8), a monitoring data feedback mechanism is established to trigger the adjustment process of the drilling site design parameters when a significant change in geological conditions is detected.

[0027] Preferably, the particle update position formula in the particle swarm optimization algorithm is:

[0028]

[0029] in, is the current position of the ith particle in the tth generation, is the velocity of the particle in generation t+1;

[0030] The particle velocity update formula is:

[0031]

[0032] Among them, w is the inertia weight, which controls the influence of the particle's previous velocity, c1, c2 are acceleration constants, r1, r2 are random numbers, is the particle’s own best historical position, g (t) It is the best position among all particles; through the mutual cooperation of the particle group, a global optimal solution or an approximate optimal solution is found.

[0033] Preferably, a linear regression model is constructed in a high-dimensional feature space so that the model has strong generalization ability. The goal is to find an optimal regression function within the error range, maximize the interval, and minimize the error. Given a training data set, find the regression function:

[0034] f(x)=w T φ(x)+b (3)

[0035] Among them, φ(x) is a mapping function that maps the input to a high-dimensional feature space, w is the weight vector, and b is the bias term;

[0036] Minimize the following objective function:

[0037]

[0038] At the same time, the constraints are met:

[0039] |y i -(w T φ(x i )+b)|≤ε+ξ i (5)

[0040] Among them, ε is the allowable error, ξ i is the slack variable, which represents the deviation of the error.

[0041] Preferably, the influence radius of the drainage volume is determined based on the following formula:

[0042]

[0043] Where: Q is set as the underground drainage volume, m 3 / h; M0 is the groundwater recharge index (L / s·km 2 ), clastic rock is 0.3-1.0L / s·km 2 , and M0 is 0.5L / s·km 2 .

[0044] Preferably, the drilling site optimization design adopts the Q-Sw linear and curved principles to determine the effective drainage radius for superimposed water release, constructs an optimization objective function with the goal of maximizing drainage efficiency and minimizing drilling costs, and dynamically adjusts the drilling spacing, depth and layout pattern through the reinforcement learning algorithm.

[0045] The beneficial effects of the present invention are:

[0046] 1. Traditional methods rely on empirical formulas and static geological models, making them difficult to cope with complex geological conditions, resulting in low accuracy in water release predictions. This paper introduces a spatial occurrence feature quantification model and a convolutional neural network, which can deeply analyze stratigraphic characteristics such as sandstone and mudstone ratios and lithology categories. Combined with a bidirectional long-short-term memory network that dynamically learns historical and real-time data, it captures the temporal characteristics and dynamic changes of the data, thereby significantly improving the accuracy of water release predictions.

[0047] 2. Existing drill site designs often use fixed models and lack dynamic response to changing geological conditions. This invention builds a drill site adaptive optimization module based on a reinforcement learning algorithm. This module automatically generates borehole spacing, depth, and layout patterns based on predictions, and adjusts drilling parameters in real time based on real-time geological changes. This enables dynamic optimization of drill site design, improves drainage efficiency, and avoids resource waste.

[0048] 3. Traditional methods do not adequately model lithologic heterogeneity, making it difficult to accurately reflect the impact of sandstone and mudstone ratios and classification on drainage paths. This invention incorporates sandstone and mudstone classification and spatial distribution characteristics into the neural network input layer, effectively solving this problem and enabling more accurate simulation of drainage processes in complex formations.

[0049] 4. Existing technologies do not fully consider the deep integration of sandstone and mudstone classification, spatial distribution characteristics, and dynamic hydrological monitoring data, and their intelligence level is low. This invention achieves a deep integration of multiple factors through steps such as data collection, preprocessing, model construction, and training. It also adopts a data-driven intelligent optimization method, which improves the universality and flexibility of the technology and provides an efficient, accurate, and economical technical means for coal mine water hazard prevention and control. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0051] Figure 1 A schematic flow chart of a method for predicting coal mine drainage volume and optimizing drilling site design based on a neural network provided in an embodiment of the present invention;

[0052] Figure 2 a in the middle is the distribution diagram of unit water inflow of coal seam roof aquifer;

[0053] Figure 2 Middle b is a schematic diagram of the thickness of the impermeable layer between coal seams;

[0054] Figure 2 Middle c is a schematic diagram of the thickness of the aquifer between coal seams;

[0055] Figure 2 d in the middle is a schematic diagram of the distance between the 9th coal floor and the 11th coal roof;

[0056] Figure 2 Middle e is a schematic diagram of the ratio of aquifer to aquiclude between coal seams;

[0057] Figure 2 Middle f is the average water pressure diagram of the aquifer on the top of the working face;

[0058] Figure 3 A graph showing the prediction results of coal mine drainage volume provided by an embodiment of the present invention;

[0059] Figure 4 This is a coal mine working face drilling site design result diagram provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0060] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0061] See also Figure 1 This embodiment provides a method for predicting the amount of water discharged from a coal mine and optimizing the design of a drilling site based on a neural network, specifically including:

[0062] (1) Data collection: Collect multidimensional data of the mining area, including geological exploration data, spatial distribution characteristics of coal seam roof and floor, sandstone and mudstone ratio and lithology classification data, historical drainage volume data and real-time hydrological monitoring information;

[0063] See also Figure 2 The distribution patterns of various parameters of coal mine strata and aquifers shown in middle af; in this embodiment, the geological and hydrological data of the target coal mine working face are comprehensively collected, the aquifer-aquitard ratio between coal seams is calculated with the help of drill core data and logging data, the unit water inflow distribution of the coal seam roof aquifer is determined based on hydrological test data such as pumping tests, the distance from the 9th coal floor to the 11th coal roof is clarified through three-dimensional geological modeling or drilling data, the average water pressure of the working face roof aquifer is monitored in real time using pressure sensors or obtained based on existing hydrological monitoring data, and information such as the thickness of the aquifer and the thickness of the aquitard between coal seams is extracted from the drilling data and geological profiles.

[0064] Geological exploration data covers detailed information at different depths and in different areas, and is updated regularly to reflect the dynamic changes in geological conditions in the mining area.

[0065] The spatial occurrence characteristic data include the thickness of the sand-mudstone interlayer, the change of the lithologic interface and the merging threshold of the thin mudstone layer. The merging threshold of the thin mudstone layer is a set thickness value. When the thickness is lower than the threshold, the thin mudstone layer is merged with other thick rock layers.

[0066] (2) Data preprocessing: preprocessing the collected data, including data cleaning, denoising and standardization, to ensure the accuracy and consistency of the data; eliminating dimensional differences to make data of different magnitudes comparable;

[0067] (3) Feature quantification: A spatial occurrence feature quantification model is introduced to convert the heterogeneity of the coal seam roof and floor and the degree of fracture development into quantifiable indicators; the data is divided into a training set and a test set according to a certain ratio. The training set is used for training the neural network, and the test set is used for verifying the neural network.

[0068] (4) Model construction: The indicators of sandstone and mudstone classification and spatial distribution characteristics are incorporated into the neural network input layer, and a hybrid neural network model consisting of a convolutional neural network and a bidirectional long-short-term memory network is constructed. The convolutional neural network is used to analyze the stratigraphic characteristics, and the hybrid neural network model of the bidirectional long-short-term memory network is used to learn the temporal dynamic characteristics of historical and real-time drainage volume data;

[0069] In this embodiment, a multi-layer neural network model is constructed, and the collected geological and hydrological parameters such as the aquifer-aquiclude ratio, unit water yield distribution, coal seam spacing, average water pressure of the aquifer, aquifer thickness and aquiclude thickness are introduced as input variables into the input layer. The hidden layer adopts a neural network combined with a bidirectional long-short-term memory network to fully explore the nonlinear relationship and spatiotemporal characteristics in the input data to enhance the learning ability of the model. The output layer outputs the predicted value of the water discharge volume of the target working face.

[0070] (5) Model training: The hybrid neural network model is trained based on the preprocessed data, and the model parameters are optimized to minimize the prediction error. The model training adopts a dynamic learning rate adjustment strategy, which automatically adjusts the learning rate according to the training progress to accelerate convergence.

[0071] In this embodiment, the constructed neural network is trained using training set data, and the mean square error is used as the loss function. The model parameters are continuously optimized through the back propagation algorithm, and training is continued until the model converges to ensure that the model can accurately learn the patterns in the data.

[0072] (6) Model evaluation and optimization: Use cross-validation method to evaluate and optimize the trained hybrid neural network model to ensure that the model has good generalization ability and stability;

[0073] In this embodiment, after the model is trained, the geological and hydrological data such as the aquifer-aquiclude ratio, unit water yield distribution, coal seam spacing, average water pressure of the aquifer, aquifer thickness and aquiclude thickness of the target working face are accurately input into the trained neural network model. The neural network performs complex calculations based on the input data to obtain a predicted value of the drainage volume and generate a drainage volume distribution map, and outputs a prediction result including the drainage volume at different locations and its spatial distribution characteristics.

[0074] (7) Drilling site optimization design: A reinforcement learning algorithm is used to build a drilling site adaptive optimization module, which automatically generates the drilling hole spacing, depth, and layout pattern based on the drainage volume prediction results output by the neural network;

[0075] The reinforcement learning algorithm adopts a particle swarm optimization algorithm to search for the optimal drilling parameter combination in a high-dimensional solution space through a particle swarm. The objective function is to maximize the drainage efficiency and minimize the drilling cost.

[0076] (8) Dynamic coupling: Dynamic coupling of water discharge prediction and drilling site optimization design, adjusting drilling parameters according to real-time geological conditions.

[0077] See also Figure 3-Figure 4In this embodiment, based on the drainage volume prediction results, a reinforcement learning algorithm is used to optimize the drilling site design. The Q-Sw linear and curved principle is used to determine the effective drainage radius for superimposed drainage. That is, drilling sites are continued to be set up at the edge of the effective impact radius for superimposed drainage. A reasonable optimization objective function is constructed with the goals of maximizing drainage efficiency and minimizing drilling costs, providing a clear direction for subsequent optimization work. The drilling spacing, depth, and layout pattern are dynamically adjusted through the reinforcement learning algorithm. While considering cost and construction convenience, the optimal drilling plan is generated through multiple iterative calculations. The optimized drilling plan is then applied to the target working face to verify its drainage effect and economy, ensuring the feasibility and effectiveness of the plan.

[0078] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A method for predicting coal mine drainage volume and optimizing drilling site design based on neural network, characterized in that: The specific steps include: (1) Data collection: Collect multidimensional data of the mining area, including geological exploration data, spatial distribution characteristics of coal seam roof and floor, sandstone and mudstone ratio and lithology classification data, historical drainage volume data and real-time hydrological monitoring information; (2) Data preprocessing: preprocessing the collected data, including data cleaning, denoising and standardization, to ensure the accuracy and consistency of the data; (3) Feature quantification: Introducing a spatial occurrence feature quantification model to convert the heterogeneity of the coal seam roof and floor and the degree of fracture development into quantifiable indicators; (4) Model construction: The indicators of sandstone and mudstone classification and spatial distribution characteristics are incorporated into the neural network input layer, and a hybrid neural network model consisting of a convolutional neural network and a bidirectional long-short-term memory network is constructed to analyze the stratigraphic characteristics and learn the temporal dynamic characteristics of historical and real-time drainage volume data; (5) Model training: training the hybrid neural network model based on the preprocessed data and optimizing the model parameters to minimize the prediction error; (6) Model evaluation and optimization: Use cross-validation method to evaluate and optimize the trained hybrid neural network model to ensure that the model has good generalization ability and stability; (7) Drilling site optimization design: A reinforcement learning algorithm is used to build a drilling site adaptive optimization module, which automatically generates the drilling hole spacing, depth, and layout pattern based on the drainage volume prediction results output by the neural network; (8) Dynamic coupling: Dynamic coupling of water discharge prediction and drilling site optimization design, adjusting drilling parameters according to real-time geological conditions.

2. The method for predicting coal mine drainage volume and optimizing drilling site design based on neural network according to claim 1, characterized in that: During the data collection phase in step (1), geological exploration data covers detailed information at different depths and in different areas, and is updated regularly to reflect the dynamic changes in geological conditions in the mining area.

3. The method for predicting coal mine drainage volume and optimizing drilling site design based on neural network according to claim 1, characterized in that: In step (1), the spatial occurrence characteristic data include the thickness of the sand-mudstone interlayer, the change of the lithologic interface and the thin mudstone layer merging threshold, wherein the thin mudstone layer merging threshold is a set thickness value. When its thickness is lower than the threshold, the thin mudstone layer is merged with other thick rock layers.

4. The method for predicting coal mine drainage volume and optimizing drilling site design based on neural network according to claim 1, characterized in that: In step (5), the model training adopts a dynamic learning rate adjustment strategy, which automatically adjusts the learning rate according to the training progress to accelerate convergence.

5. The method for predicting coal mine drainage volume and optimizing drilling site design based on neural network according to claim 1, characterized in that: The reinforcement learning algorithm described in step (7) adopts a particle swarm optimization algorithm, which searches for the optimal drilling parameter combination in a high-dimensional solution space through a particle swarm, and the objective function is to maximize the drainage efficiency and minimize the drilling cost.

6. The method for predicting coal mine drainage volume and optimizing drilling site design based on neural network according to claim 1, characterized in that: When adjusting the drilling parameters in real time in step (8), a monitoring data feedback mechanism is established to trigger the adjustment process of the drilling site design parameters when a significant change in geological conditions is detected.

7. The method for predicting coal mine drainage volume and optimizing drilling site design based on neural network according to claim 1, characterized in that: The particle update position formula in the particle swarm optimization algorithm is: in, is the current position of the ith particle in the tth generation, is the velocity of the particle in generation t+1; The particle velocity update formula is: Among them, w is the inertia weight, which controls the influence of the particle's previous velocity, c1, c2 are acceleration constants, r1, r2 are random numbers, is the particle’s own best historical position, g (t) It is the best position among all particles; through the mutual cooperation of the particle group, a global optimal solution or an approximate optimal solution is found.

8. The method for predicting coal mine drainage volume and optimizing drilling site design based on neural network according to claim 1, characterized in that: Construct a linear regression model in a high-dimensional feature space so that the model has strong generalization ability. The goal is to find an optimal regression function within the error range, maximize the interval, and minimize the error. Given a training data set, find the regression function: f(x)=w T φ(x)+b (3) Among them, φ(x) is a mapping function that maps the input to a high-dimensional feature space, w is the weight vector, and b is the bias term; Minimize the following objective function: At the same time, the constraints are met: |y i -(w T φ(x i )+b)|≤ε+ξ i (5) Among them, ε is the allowable error, ξ i is the slack variable, which represents the deviation of the error.

9. The method for predicting coal mine drainage volume and optimizing drilling site design based on neural network according to claim 1, characterized in that: The influence radius of the drainage volume is determined based on the following formula: Where: Q is set as the underground drainage volume, m 3 / h; M0 is the groundwater recharge index (L / s·km 2 ).

10. The method for predicting coal mine drainage volume and optimizing drilling site design based on neural network according to claim 1, characterized in that: The drilling site optimization design uses the Q-Sw linear and curved principles to determine the effective drainage radius for superimposed water release. The optimization objective function is constructed with the goal of maximizing drainage efficiency and minimizing drilling costs. The drilling spacing, depth and layout pattern are dynamically adjusted through the reinforcement learning algorithm.

Images