Method for predicting water inflow of open pit coal mine

Abstract

The invention discloses an opencast coal mine water inflow prediction method, and belongs to the technical field of opencast coal mine mining hydrogeology. The method comprises the following steps: constructing a three-dimensional geology-hydrology coupling model integrating a stratigraphic structure, fault distribution and the like; deploying a multi-source dynamic monitoring network in places such as a stope slope, and collecting related data in real time; 12 core parameters are determined, and a standardized data set is constructed; developing a three-layer coupling model architecture comprising a short-term dynamic layer (LSTM neural network), a medium-term mechanism layer (improved Green Anine function, introduced rock mass damage correction term) and a long-term trend layer (unsteady state well flow equation combined with finite element method), determining weight through Bayesian optimization, and integrating and outputting results; and dynamic feedback correction triggered by double thresholds is implemented to form closed-loop control. In addition, the invention further provides a computer readable storage medium for implementing the method.

Description

Technical Field

[0001] The present invention relates to the field of open-pit coal mining hydrogeology technology, and in particular to an open-pit coal mine water inflow prediction method integrating three-dimensional coupling modeling, multi-source monitoring and hybrid algorithm. Background Art

[0002] In open-pit coal mining, dynamic prediction of water inflow is a core challenge to ensure safe production and rational use of water resources. The existing technology has the following main drawbacks:

[0003] Insufficient model accuracy: Traditional analytical methods rely on simplified hydrogeological models, which are difficult to reflect complex three-dimensional geological structures and non-steady-state seepage processes; although numerical simulation methods can handle complex scenarios, parameter inversion is time-consuming and relies on manual experience.

[0004] Poor dynamic adaptability: There is a lack of real-time quantification of the impact of dynamic factors such as rock damage, dewatering and pressure reduction on water inflow during the mining process. Existing methods are mostly based on static parameters and cannot adapt to changes in mining progress.

[0005] Insufficient use of multi-source data: Multi-dimensional data such as precipitation, extraction rate, and groundwater dynamics are not effectively coupled, and insufficient coverage of the monitoring network leads to incomplete extraction of driving factors.

[0006] Therefore, there is an urgent need for a water inflow prediction method that integrates three-dimensional geological modeling, dynamic monitoring and intelligent algorithms to solve the problems of prediction accuracy and real-time performance of existing technologies under complex mining conditions. Summary of the Invention

[0007] A method for predicting water inflow in open-pit coal mines is provided, which integrates a three-dimensional geological-hydrological coupling model, multi-source dynamic monitoring and a hybrid prediction algorithm to achieve high-precision dynamic prediction of water inflow.

[0008] In order to achieve the above-mentioned purpose, a method for predicting water inflow in open pit coal mines is provided, which is characterized by comprising the following steps: (1) constructing a three-dimensional geological-hydrological coupling model: integrating the stratum structure, fault distribution, aquifer parameters and aquiclude spatial morphology of the open pit mine area, and establishing a three-dimensional seepage model including anisotropic permeability tensor; (2) deploying a multi-source dynamic monitoring network: arranging pressure sensors, flow meters, rain gauges and microseismic monitoring devices on the side slopes, pit bottoms and drainage wells of the mining area to collect hydrological, meteorological and rock damage data in real time; (3) establishing a water inflow driving factor system: determining the factors including meteorological factors, (4) Develop and construct a standardized data set with spatiotemporal characteristics based on 12 core parameters including mining factors, hydrological factors, geological factors, and dynamic coupling factors; (5) Develop and construct a three-layer coupling model architecture: including a short-term dynamic layer, a medium-term mechanism layer, and a long-term trend layer, which output water inflow prediction values ​​from the short-term, medium-term, and long-term perspectives respectively. The weighted average method is used to integrate the output results of the three layers, and the weight coefficients are determined by Bayesian optimization; (6) Implement dynamic feedback correction: By calculating the deviation between real-time monitoring data and predicted values, the model weight coefficients and hydrogeological parameters are automatically updated to form a closed-loop control of prediction-verification-correction.

[0009] Furthermore, the construction of the three-dimensional geological-hydrological coupling model in step (1) includes:

[0010] Construct a 3D stratigraphic grid based on borehole data with a resolution of ≤5m to distinguish between aquifers and aquicludes;

[0011] Import seismic exploration data to establish fault geometry models and annotate water conductivity parameters;

[0012] Through pumping tests and model inversion, the aquifer parameters such as permeability coefficient and porosity are obtained, and the anisotropic permeability tensor is constructed.

[0013] Furthermore, the multi-source dynamic monitoring network in step (2) includes:

[0014] Vibrating wire pore water pressure sensors are arranged on the side slopes of the stope with a spacing of 20-50m;

[0015] The electromagnetic flow meter is installed in the drainage pipe at the bottom of the pit to monitor the water inflow in real time;

[0016] The microseismic monitoring system is deployed in the mining area to monitor the development of cracks caused by rock damage.

[0017] Furthermore, among the 12 core parameters in step (3), meteorological factors include: precipitation intensity and evaporation;

[0018] Mining factors include mining advancement rate and stope exposure area; hydrological factors include dewatering pressure reduction amplitude and aquifer permeability coefficient;

[0019] Geological factors include: rock damage factor and fault hydraulic conductivity;

[0020] Dynamic coupling factors include: precipitation-groundwater response time and mining disturbance impact radius.

[0021] Furthermore, the specific method of constructing the three-layer coupling model architecture in step (4) is as follows:

[0022] Short-term dynamic layer: A long short-term memory neural network (LSTM) is used to process time series data. The input layer has 12 neurons, corresponding to 12 driving factors, and there are two hidden layers, each with 64 neurons. The output layer is the predicted water inflow value for the next 1-7 days. The Adam optimizer is used for training, and the loss function is the root mean square error (RMSE).

[0023] Mid-term mechanism layer: Improved Green’s function is used to predict water inflow. The rock damage correction term is introduced into the improved Green’s function. The formula is as follows:

[0024] Q(t): water inflow, K: permeability coefficient, M: aquifer thickness, S: water storage coefficient, α: damage coefficient, ranging from 0.1 to 0.5, D: rock damage degree, T: hydraulic conductivity, t: time, r w : pumping well radius;

[0025] Long-term trend layer: The unsteady-state well flow equation is used to calculate the total recharge within the dewatering influence range. Combined with the boundary conditions of the three-dimensional model, the finite element method is used to solve the groundwater seepage field and output the trend value of the water inflow in the next 30-90 days.

[0026] Furthermore, the short-term, medium-term and long-term prediction weights are determined by Bayesian optimization, with the weight range being 0.4-0.6 for the short term, 0.2-0.4 for the medium term and 0.1-0.3 for the long term. The short-term dynamic layer (LSTM neural network) inputs 12 driving factors (such as precipitation intensity, mining advancement rate, etc.) that are monitored in real time. The data is updated frequently, which can directly reflect the immediate impact of current mining activities and meteorological conditions on water inflow. The short-term prediction time span is small, and the model relies on a shorter time series data chain. It is less affected by long-term uncertainty factors such as geological parameter drift and boundary condition changes, and the prediction results are more reliable. In view of the immediate impact of high-frequency dynamic factors such as sudden meteorological changes (such as heavy rain) and mining disturbances (such as slope blasting) on ​​the water inflow of open-pit coal mines, the short-term model captures nonlinear characteristics through real-time monitoring data (update frequency ≤ 1 hour), and its prediction The results provide direct guidance for on-site drainage scheduling, so the weight is set at 0.4-0.6. The medium-term mechanism layer (improved Green's function) is based on hydrogeological and physical processes and focuses on the lagged effects of factors such as rock damage accumulation and dewatering and pressure reduction. The intensity of these factors gradually releases over time, and predictions require integrating historical data with geological evolution patterns. Its weight is lower than the short-term but higher than the long-term. While the medium-term model incorporates a rock damage correction term to improve accuracy, it relies on inverted parameters from multiple sources such as pumping tests and microseismic monitoring, resulting in some uncertainty. The medium-term model focuses on the lagged effects of geological evolution and describes the gradual change in water inflow over time through physical equations. Despite some uncertainty in its parameter inversion, it effectively connects short-term dynamics with long-term trends. A weight of 0.2-0.4 is set. Compared to the real-time nature of the short-term model, the medium-term prediction is less timely, so its weight is assigned a moderate value. The long-term trend layer simulates the groundwater seepage field and involves long-term predictions of slowly changing parameters such as aquifer recharge and fault conductivity. However, geological conditions can mutate during open-pit coal mining, causing model boundary conditions to deviate from reality and significantly accumulating prediction errors over time. Long-term models are primarily used for macro-planning, and their value lies in revealing trends in water inflow rather than accurately capturing daily fluctuations. Therefore, their impact on short-term dynamic results needs to be mitigated in weight allocation to avoid interference with real-time prediction accuracy due to trend extrapolation errors. Long-term models, based on the macro-balance of the groundwater system, predict recharge trends within the dewatering zone. However, due to factors such as sudden changes in geological conditions and parameter drift, prediction errors increase significantly over time. Their core role is to assist in planning drainage facility capacity, so the weight is set at 0.1-0.3.

[0027] Furthermore, the dynamic feedback correction in step (5) adopts a dual-threshold trigger mechanism:

[0028] When the deviation between the monitored value and the predicted value is greater than 15%, the weight update is initiated; Bayesian optimization can automatically increase the short-term weight to 0.6, strengthening the LSTM model's response to real-time data; if the long-term trend shows a continued increase in aquifer recharge, the weight can be fine-tuned (increased from 0.1 to 0.2) to make the prediction result more in line with the macro trend.

[0029] When the deviation is greater than 25% for three consecutive times, the hydrogeological parameter field is reconstructed.

[0030] Furthermore, the three-dimensional geological modeling software is GMS or COMSOL, with a grid resolution of ≤5m, which can simulate the enhanced effect of the fault zone permeability coefficient.

[0031] Furthermore, the multi-source data preprocessing includes:

[0032] The precipitation intensity was processed using a 5-min sliding average, and the evaporation was calculated using the Penman-Monteith formula;

[0033] The mining advancement rate is converted from the daily mining volume, and the exposed area of ​​the mining site is updated in real time through 3D laser scanning.

[0034] On the other hand, the present invention further provides a computer-readable storage medium storing program instructions, characterized in that when the instructions are executed by a processor, the steps of the above method are implemented.

[0035] Beneficial effects

[0036] Multi-dimensional coupled modeling: The three-dimensional geological-hydrological model accurately depicts the complex seepage field and solves the defects of the traditional model in simplifying the boundary.

[0037] Strong dynamic adaptability: Real-time integration of dynamic factors such as mining disturbance and rock damage greatly improves prediction accuracy compared to traditional methods.

[0038] Intelligent algorithm optimization: LSTM captures short-term nonlinear characteristics, improved analytical methods embed physical mechanisms, and non-steady-state equations process long-term trends, forming a complementary prediction system.

[0039] Full-process closed-loop control: A dynamic feedback mechanism of monitoring, prediction, and correction enables adaptive updating of model parameters to meet the multi-stage mining needs of open-pit coal mines. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a flow chart of the method for predicting water inflow in open-pit coal mines in the present invention. DETAILED DESCRIPTION

[0041] The following is a clear and complete description of the technical solutions in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts are within the scope of protection of the present invention.

[0042] Figure 1 This is a flow chart of the method for predicting water inflow in an open-pit coal mine in this embodiment, which includes the following steps:

[0043] (1) Construction of a three-dimensional geological-hydrological coupling model

[0044] Use 3D geological modeling software (such as GMS, COMSOL) to integrate open pit mine data:

[0045] Stratigraphic structure: A three-dimensional stratigraphic grid is constructed based on borehole data with a resolution of ≤5m to distinguish between Quaternary unconsolidated layers, bedrock aquifers (such as Jurassic sandstones), and aquicludes (such as mudstones).

[0046] Fault distribution: Import seismic exploration data, establish fault geometry model, and annotate hydraulic conductivity parameters (permeability enhancement factor 0.5-10).

[0047] Aquifer parameters: permeability coefficient (K = 1-50m / d), porosity, etc. The anisotropic permeability tensor is constructed by adjusting parameters such as water degree (μ = 0.05-0.2).

[0048] The following steps are involved:

[0049] (1) Construction of a three-dimensional geological-hydrological coupling model

[0050] Use 3D geological modeling software (such as GMS, COMSOL) to integrate open pit mine data:

[0051] Stratigraphic structure: A three-dimensional stratigraphic grid is constructed based on borehole data with a resolution of ≤5m to distinguish between Quaternary unconsolidated layers, bedrock aquifers (such as Jurassic sandstones), and aquicludes (such as mudstones).

[0052] Fault distribution: Import seismic exploration data, establish fault geometry model, and annotate hydraulic conductivity parameters (permeability enhancement factor 0.5-10).

[0053] Aquifer parameters: permeability coefficient (K = 1-50m / d), porosity, etc. The anisotropic permeability tensor is constructed by adjusting parameters such as water degree (μ = 0.05-0.2).

[0054] Aquiclude morphology: Identify the spatial distribution of the aquiclude at the roof and floor of the coal seam and set boundary conditions (such as aquiclude boundary and constant head boundary).

[0055] Deploy monitoring equipment at key locations in the mining area:

[0056] Stope slope: Bury vibrating wire pore water pressure sensors (accuracy ±0.1% FS) at intervals of 20-50m to monitor changes in shallow groundwater pressure.

[0057] Pit bottom aquifer: Install an electromagnetic flowmeter (accuracy ±0.5%) on the main drainage pipe to collect water inflow data in real time; arrange temperature-conductivity sensors (accuracy ±0.1℃, ±0.5%) to monitor groundwater runoff characteristics.

[0058] Dewatering well group: equipped with pressure transmitter (accuracy ±0.25% FS) and water level gauge (accuracy ±1cm) to monitor the water level dynamics during the dewatering and pressure reduction process.

[0059] Surface monitoring: Set up tipping bucket rain gauges (resolution 0.1mm) and ultrasonic anemometers (accuracy ±0.1m / s) to simultaneously collect meteorological data; deploy a microseismic monitoring system (number of channels ≥64) to monitor the development of cracks caused by rock damage.

[0060] (3) Establishing a water inflow driving factor system

[0061]

[0062]

[0063] (4) Develop hybrid prediction algorithms

[0064] Build a three-layer coupled model architecture:

[0065] Short-term dynamic layer: A long short-term memory neural network (LSTM) is used to process time series data. The input layer has 12 neurons (corresponding to 12 driving factors), and there are two hidden layers (64 neurons each). The output layer is the predicted value of water inflow in the next 1-7 days. The Adam optimizer (learning rate 0.001) is used for training, and the loss function is the root mean square error (RMSE).

[0066] Mid-term mechanism layer: The improved Greene function introduces the rock damage correction term, and the formula is as follows: Q(t): water inflow, K: permeability coefficient, M: aquifer thickness, S: water storage coefficient, α: damage coefficient, ranging from 0.1 to 0.5, D: rock damage degree, T: hydraulic conductivity, t: time, r w : Pumping well radius.

[0067] Long-term trend layer: The unsteady-state well flow equation (Theis equation) calculates the total recharge within the dewatering influence range. Combined with the boundary conditions of the three-dimensional model, the finite element method is used to solve the groundwater seepage field and output the trend value of the water inflow in the next 30-90 days.

[0068] Fusion strategy: The three-layer outputs are integrated using the weighted average method, and the weight coefficients are determined by Bayesian optimization (short-term: medium-term: long-term = 0.5:0.3:0.2).

[0069] (5) Implement dynamic feedback correction

[0070] The dynamic feedback correction adopts a dual-threshold trigger mechanism:

[0071] When the deviation between the monitored value and the predicted value is greater than 15%, the weight update is initiated; Bayesian optimization can automatically increase the short-term weight to 0.6, strengthening the LSTM model's response to real-time data; if the long-term trend shows a continued increase in aquifer recharge, the weight can be fine-tuned (increased from 0.1 to 0.2) to make the prediction result more in line with the macro trend.

[0072] When the deviation is greater than 25% for three consecutive times, the hydrogeological parameter field is reconstructed.

[0073] Specific embodiment: prediction of water inflow in an open-pit coal mine

[0074] Mining area overview

[0075] An open-pit coal mine is located in the Ordos Basin. The main coal seam is buried at a depth of 150-300m. The aquifer is Middle Jurassic sandstone (K = 8m / d), the aquiclude is mudstone (K = 0.01m / d), the average annual precipitation is 450mm, and the mining advance rate is 15m / d.

[0076] Specific implementation steps

[0077] 3D model construction

[0078] GMS software was used to import 200 borehole data, construct a 0.5m×0.5m×0.5m grid model, identify three water-conducting faults (permeability coefficient enhanced by 3 times), and set the boundary condition to constant head recharge from the Yellow River on the north (h=+10m).

[0079] The inversion results show that the permeability tensor of the aquifer is Kx = 10 m / d, Ky = 8 m / d, and Kz = 5 m / d.

[0080] Monitoring network deployment

[0081] 20 pore water pressure sensors (model: vibrating string VWP-200) are arranged on the slope of the mining area, 3 electromagnetic flow meters (diameter DN300, accuracy ±0.3%) are installed at the bottom of the pit, and 15 sets of water level-pressure monitors (resolution 1cm / 0.1kPa) are deployed in the dewatering well group.

[0082] The microseismic monitoring system is equipped with 8 seismic sensors with a positioning accuracy of ≤10m, which can monitor the rock damage caused by blasting in real time (cumulative microseismic events >500 times / day).

[0083] Driving factor extraction

[0084] Calculation of rock damage factor D: Select 30 days of microseismic data, total energy E = 2 × 10 6 J, monitoring volume V = 1 × 10 6 m 3 , D = 2 × 10 6 / (1×10 6 × 100) = 0.02 (reference energy E0 = 100 J / m 3 ).

[0085] The precipitation-groundwater response time τ = 6 h, and the lag time between the precipitation peak and the water inflow peak was determined by cross-correlation analysis.

[0086] Hybrid algorithm implementation

[0087] LSTM model: The training data contains 1825 days of data from 2020 to 2024, the input sequence length is 14 days, the validation set accounts for 20%, and the optimal parameters are 64-32 neurons in the hidden layer and a dropout rate of 0.2.

[0088] Improved Green's function: damage coefficient α = 0.3 (fitted by the 2023 rainy season data), and a correction term D = 0.02 is introduced when calculating the water yield of a single well in the dewatering well group.

[0089] Unsteady well flow equation: Calculate the 30-day drainage impact radius R = 230m (using the bisection method to solve the distance corresponding to a 0.01m drop in water level), the total recharge Q total =12000m 3 / d.

[0090] Dynamic feedback correction

[0091] The measured water inflow on July 15, 2024 is Q mon =8500m 3 / d, predicted value Q pred =7800m 3 / d, deviation ε=8.2%<15%, parameter update is not triggered.

[0092] After the heavy rainfall on July 20, ε = 22%. The system automatically adjusted the LSTM forget gate weight (Δw = 0.05) and inverted and updated the aquifer permeability coefficient K = 8.5 m / d. After correction, the prediction accuracy increased to 92%.

[0093] Prediction results

[0094] Compared with the data of the rainy season (July-September) in 2024, the root mean square error (RMSE) of this method is 450m 3 / d, compared with the traditional numerical simulation method (RMSE = 720m 3 / d) increased by 37.5%, with a mean absolute error (MAE) of 380m 3 / d, meeting the design accuracy requirements of open-pit coal mine drainage systems (error ≤ 15%).

[0095] The above specific embodiments will allow those skilled in the art to easily implement the present invention. However, it should be understood that the present invention is not limited to the above specific embodiments. Based on the disclosed embodiments, those skilled in the art can arbitrarily combine different technical features to implement different technical solutions.

[0096] Except for the technical features described in the specification, all other technical features are known technologies to those skilled in the art.

Claims

1. A method for predicting water inflow in open-pit coal mines, characterized in that The following steps are involved: (1) Construct a three-dimensional geological-hydrological coupling model: Integrate the stratigraphic structure, fault distribution, aquifer parameters and spatial morphology of the open-pit mine area to establish a three-dimensional seepage model including anisotropic permeability tensors; (2) Deploy a multi-source dynamic monitoring network: Arrange pressure sensors, flow meters, rain gauges and microseismic monitoring devices on the slopes, pit bottoms and drainage wells of the mining area to collect hydrological, meteorological and rock damage data in real time; (3) Establish a water inflow driving factor system: Determine 12 core parameters including meteorological factors, mining factors, hydrological factors, geological factors and dynamic coupling factors, and construct a standardized data set including spatiotemporal characteristics; (4) Develop and construct a three-layer coupling model architecture: Include short-term dynamic layer, medium-term mechanism layer and long-term trend layer, output water inflow prediction values ​​from the short-term, medium-term and long-term respectively, integrate the output results of the three layers by weighted average method, and determine the weight coefficient by Bayesian optimization; (5) Implement dynamic feedback correction: Automatically update the model weight coefficient and hydrogeological parameters by calculating the deviation between real-time monitoring data and predicted values, forming a closed-loop control of prediction-verification-correction.

2. The method according to claim 1, characterized in that The construction of the three-dimensional geological-hydrological coupling model in step (1) includes: Construct a 3D stratigraphic grid based on borehole data with a resolution of ≤5m to distinguish between aquifers and aquicludes; Import seismic exploration data to establish fault geometry models and annotate water conductivity parameters; Through pumping tests and model inversion, the aquifer parameters such as permeability coefficient and porosity are obtained, and the anisotropic permeability tensor is constructed.

3. The method according to claim 1, characterized in that The multi-source dynamic monitoring network in step (2) includes: Vibrating wire pore water pressure sensors are arranged on the side slopes of the stope with a spacing of 20-50m; The electromagnetic flow meter is installed in the drainage pipe at the bottom of the pit to monitor the water inflow in real time; The microseismic monitoring system is deployed in the mining area to monitor the development of cracks caused by rock damage.

4. The method according to claim 1, wherein Among the 12 core parameters described in step (3), meteorological factors include: precipitation intensity and evaporation; Mining factors include mining advancement rate and stope exposure area; hydrological factors include dewatering pressure reduction amplitude and aquifer permeability coefficient; Geological factors include: rock damage factor and fault hydraulic conductivity; Dynamic coupling factors include: precipitation-groundwater response time and mining disturbance impact radius.

5. The method according to claim 1, wherein The specific method of constructing the three-layer coupling model architecture in step (4) is as follows: Short-term dynamic layer: A long short-term memory neural network (LSTM) is used to process time series data. The input layer has 12 neurons, corresponding to 12 driving factors, and there are two hidden layers, each with 64 neurons. The output layer is the predicted water inflow value for the next 1-7 days. The Adam optimizer is used for training, and the loss function is the root mean square error (RMSE). Mid-term mechanism layer: Improved Green’s function is used to predict water inflow. The rock damage correction term is introduced into the improved Green’s function. The formula is as follows: Q(t): water inflow, K: permeability coefficient, M: aquifer thickness, S: water storage coefficient, α: damage coefficient, ranging from 0.1 to 0.5, D: rock damage degree, T: hydraulic conductivity, t: time, r w : pumping well radius; Long-term trend layer: The unsteady-state well flow equation is used to calculate the total recharge within the dewatering influence range. Combined with the boundary conditions of the three-dimensional model, the finite element method is used to solve the groundwater seepage field and output the trend value of the water inflow in the next 30-90 days.

6. The method according to claim 1, characterized in that The dynamic feedback correction in step (5) adopts a dual-threshold trigger mechanism: When the deviation between the monitored value and the predicted value is greater than 15%, the weight update is initiated; When the deviation is greater than 25% for three consecutive times, the hydrogeological parameter field is reconstructed.

7. The method according to claim 1, characterized in that The short-term, medium-term, and long-term forecast weights are determined through Bayesian optimization, with the weight range being 0.4–0.6 for the short term, 0.2–0.4 for the medium term, and 0.1–0.3 for the long term.

8. The method according to claim 1, characterized in that The three-dimensional geological modeling software is GMS or COMSOL, with a grid resolution of ≤5m, which can simulate the enhanced effect of the fault zone permeability coefficient.

9. The method according to claim 1, characterized in that The multi-source data preprocessing includes: The precipitation intensity was processed using a 5-min sliding average, and the evaporation was calculated using the Penman-Monteith formula; The mining advancement rate is converted from the daily mining volume, and the exposed area of ​​the mining site is updated in real time through 3D laser scanning.

10. A computer-readable storage medium storing program instructions, characterized in that: When the instructions are executed by a processor, the steps of the method according to any one of claims 1 to 9 are implemented.

Images