Method and system for dynamically monitoring, regulating and controlling sealing and storage of water depth of mine

Abstract

The invention discloses a dynamic monitoring, regulating and controlling method and system for sealing and storage of a water depth of a mine, and relates to the technical field of sealing and storage monitoring of mine water. The method comprises the following steps: monitoring network deployment: installing related sensors at key positions to form a three-dimensional monitoring network; acquisition and transmission: acquiring original data of sensors in the three-dimensional monitoring network, and performing data preprocessing; aI model construction and training: collecting historical data to construct a training set, and constructing an LSTM neural network prediction model; dynamic risk assessment: constructing a risk assessment model based on an XGBoost algorithm, and calculating a sequestration layer risk index in real time through the risk assessment model; and intelligent regulation and control: generating a regulation and control strategy according to the output of the LSTM neural network prediction model. According to the method and the system, monitoring of the sealing and storage area of the water depth of the mine can be accurately realized in real time, and early warning of risks can be realized.

Description

Technical Field

[0001] The present invention relates to the technical field of mine water sequestration monitoring, and particularly to a method and system for dynamic monitoring and regulation of deep mine water sequestration. Background Art

[0002] During the mining of mineral resources such as coal, a large amount of mine water is generated. Traditionally, mine water is often directly discharged, which not only wastes precious water resources but also may cause environmental pollution, such as soil salinization, water eutrophication and other problems. Among them, deep sequestration technology has gradually become an important way to treat high salinity mine water due to its environmental protection and economic advantages.

[0003] In the prior art, the following problems mainly exist in the deep sequestration of mine water: ① The monitoring means during the sequestration process are single, mainly relying on manual regular sampling and analysis, with poor real-time performance; ② Lack of dynamic prediction ability for the pressure and water quality changes of the sequestration layer; ③ The regulation of the sequestration system lags behind and cannot make a rapid response based on real-time data; ④ The safety assessment of the sequestration mainly relies on empirical formulas, with insufficient accuracy; ⑤ The data of each monitoring subsystem are isolated and lack the ability of comprehensive analysis. Summary of the Invention

[0004] The technical problem to be solved by the present invention is how to provide a method and system for dynamic monitoring and regulation of deep mine water sequestration that can monitor the deep mine water sequestration area in real time and accurately and give early warnings of risks.

[0005] To solve the above technical problems, the technical solution adopted by the present invention is: A method for dynamic monitoring and regulation of deep mine water sequestration, comprising the following steps: 1) Deployment of the monitoring network: Install relevant sensors at injection wells, monitoring wells, sequestration layers, injection pipelines and set key positions to form a three-dimensional monitoring network; 2) Data collection and transmission: Collect the original data of the sensors in the three-dimensional monitoring network, perform data preprocessing, and transmit the processed data to relevant models; 3) Construction and training of the AI model: Collect historical data to construct a training set, construct an LSTM neural network prediction model, train the LSTM neural network prediction model, and use the trained network prediction model to predict the evolution trend of the parameters of the sequestration layer; 4) Dynamic risk assessment: Construct a risk assessment model based on the XGBoost algorithm, and calculate the risk index of the sequestration layer in real time through the risk assessment model; Trigger an early warning when the risk index exceeds the set threshold; 5) Intelligent regulation: Generate a regulation strategy according to the output of the LSTM neural network prediction model; Adjust the frequency of the injection pump and the opening of the valve through a PID controller; Start a pressure relief circuit or stop water injection in case of emergency according to the evaluation result of the risk assessment model.

[0006] The present invention also discloses a dynamic monitoring and control system for deep - water storage in a mine, including: Multi - parameter monitoring network: a three - dimensional monitoring network composed of a variety of sensors, used to monitor injection wells, monitoring wells, storage layers, injection pipelines, and set key positions; Edge computing node: a data processing unit deployed underground, used to implement data pre - processing and feature extraction; AI analysis platform: including an LSTM neural network prediction model based on deep learning and a risk assessment model, used to predict the evolution trend of storage layer parameters by the prediction model and calculate the risk index of the storage layer respectively; Intelligent control system: including an execution structure, used to generate a control strategy according to the output of the LSTM neural network prediction model; adjust the frequency of the injection pump and the opening of the valve through a PID controller, and start the pressure relief circuit or stop water injection according to the evaluation result of the risk assessment model in case of emergency.

[0007] The beneficial effects of adopting the above - mentioned technical solutions are as follows: The method and system of the present invention can realize real - time, comprehensive, and accurate dynamic monitoring of the deep - water storage area in the mine, and timely master the changes of key parameters such as water level, water quality, and pressure during the storage process of mine water. By constructing an intelligent control model based on artificial intelligence and automatically analyzing and generating a scientific and reasonable control strategy according to the monitoring data, the accurate and efficient control of the mine water storage process is realized, ensuring the safety and effect of storage. In addition, the degree of intelligence and automation level of the method and system reduce manual intervention, lower labor costs and the risk of human errors. Brief Description of the Drawings

[0008] The present invention will be further described in detail below with reference to the drawings and specific embodiments.

[0009] Figure 1 is the overall flowchart of the method described in the embodiment of the present invention; Figure 2 is the schematic diagram of the LSTM model in the method described in the embodiment of the present invention; Figure 3 is the visualization graph of the prediction result of the LSTM model in the method described in the embodiment of the present invention; Figure 4 is the schematic diagram of the XGBoost risk assessment model in the method described in the embodiment of the present invention; Figure 5 is the curve graph of the data prediction result of the LSTM model in the method described in the embodiment of the present invention; Figure 6 is the principle block diagram of the system described in the embodiment of the present invention. Detailed Description of the Embodiments

[0010] Combined with the accompanying drawings in the embodiments of the present invention, the technical solutions in the embodiments of the present invention are clearly and completely described. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of protection of the present invention.

[0011] Many specific details are set forth in the following description in order to provide a thorough understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein. Those skilled in the art can make similar extensions without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0012] Generally, as Figure 1 shown, the embodiments of the present invention disclose a method for dynamic monitoring and regulation of deep water storage in a mine, including the following steps: S1. Monitoring network deployment: Install relevant sensors at injection wells, monitoring wells, storage layers, injection pipelines, and set key positions to form a three-dimensional monitoring network; S2. Data acquisition and transmission: Collect the original data of the sensors in the three-dimensional monitoring network, perform data preprocessing, and transmit the processed data to relevant models; S3. AI model construction and training: Collect historical data to construct a training set, construct an LSTM neural network prediction model, train the LSTM neural network prediction model, and use the trained network prediction model to predict the evolution trend of the parameters of the storage layer; S4. Dynamic risk assessment: Construct a risk assessment model based on the XGBoost algorithm, and calculate the risk index of the storage layer in real time through the risk assessment model; When the risk index exceeds the set threshold, an alarm is triggered; S5. Intelligent regulation: Generate a regulation strategy according to the output of the LSTM neural network prediction model; Adjust the frequency of the injection pump and the opening of the valve through a PID controller; Start the pressure relief circuit or stop the injection in case of emergency according to the evaluation result of the risk assessment model.

[0013] The above steps are described in detail below in combination with specific content: S1. Monitoring network deployment: 1) Install 36 distributed fiber optic sensors in injection wells and monitoring wells to monitor the changes in the temperature field and acoustic wave field in real time; 2) Arrange 8 microseismic monitoring stations around the storage layer to monitor the signals emitted by rock microfractures; 3) Install multiple sets of water quality sensors on the injection pipeline to monitor pH, water pressure, and flow rate in real time; 4) Install pore water pressure sensors and displacement monitors at key surface locations.

[0014] S2, Data collection and transmission: Collect the original sensor data at the edge nodes at a frequency of 10 s, screen, denoise, and extract features from the collected original data, and then transmit the processed data to the ground platform through an industrial ring network.

[0015] S3, Construct an LSTM neural network prediction model: 1) Data preparation and preprocessing: Data collection: Collect the operation data of mine water sealing in the mining area in the past five years, including real-time monitoring data (pH, water pressure, flow rate), geological parameters (permeability, porosity), fiber optic data (temperature, acoustic wave), and microseismic data, etc.

[0016] Data preprocessing: Use interpolation or mean filling methods to process missing data; arrange the data in chronological order to construct a time series data set.

[0017] 2) Construct an LSTM neural network prediction model: Input layer: Input the constructed time series data set, including real-time monitoring data (pH, water pressure, flow rate), geological parameters (permeability, porosity), geological parameters (permeability, porosity), fiber optic data (temperature, acoustic wave), and microseismic data, etc.

[0018] Hidden layer 1: Adopt a multi-layer LSTM structure, with each layer containing multiple LSTM units, used to capture long-term dependencies in time series data.

[0019] Hidden layer 2: Add a fully connected layer after the LSTM layer, used to map the output of the LSTM to the target variable.

[0020] Output layer: Finally, output the predicted value of the target variable.

[0021] The structure of the constructed LSTM neural network prediction model is as Figure 2 shown.

[0022] 3) Model training: Calculation of the loss function: The calculation of this value can be used to measure the difference between the predicted result of the model and the real result. In the present invention, the mean square error (MSE) is calculated to measure the predicted result. The smaller the MSE value, the better the prediction effect of the model, that is, the better the fitting effect of the model.

[0023] Hyperparameter Tuning: If the value of the mean squared error (MSE) loss function is large and the prediction effect is not ideal, hyperparameters need to be adjusted through grid search to optimize the prediction effect of the model. Select the number of layers, time step length, learning rate, and batch size as parameters, and use the exhaustive search method of grid search to train and evaluate all possible hyperparameter combinations within the specified hyperparameter value range, and finally select the hyperparameter combination with the best performance on the validation set.

[0024] ; where n is the number of samples, is the i-th true value, is the i-th predicted value.

[0025] 4) Model Evaluation: Evaluation Metrics: Use metrics such as root mean squared error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE) to evaluate the model performance. The lower the values of these three metrics, the better the model performance. Corresponding evaluation metrics can be selected according to the target task. If there are outliers in the data and it is sensitive to large errors, RMSE is preferred; if more attention is paid to the average error and it is not sensitive to outliers, MAE is an important reference; if you want to understand the relative error for comparison between different data, MAPE is more appropriate.

[0026] ; ; ; Validation and Testing: Divide the dataset into a training set, a validation set, and a testing set. Adjust the model parameters through the validation set and use the testing set to evaluate the final model performance.

[0027] 5) Visualization and Application: Visualization of Prediction Results: Plot a comparison graph of predicted values and true values to intuitively show the prediction effect of the model, as Figure 3 shown.

[0028] Model Deployment: Deploy the trained model into practical applications to monitor the prediction operation status of the mining area in real time and conduct risk assessment.

[0029] S4. Build a Risk Assessment Model: Build a risk assessment model based on the XGBoost algorithm, calculate the risk index of the storage system in real time, and trigger an alarm when the risk index exceeds the threshold.

[0030] 1) Data Preparation and Preprocessing: Data Collection: Collect data related to the storage system, such as equipment operation parameters, environmental data, historical fault records, etc.

[0031] Data cleaning: Handle missing values and outliers to ensure data quality.

[0032] Feature engineering: Extract feature parameters related to risk assessment, such as equipment operation time, temperature change rate, pressure change, etc.

[0033] Data partitioning: Divide the dataset into a training set and a test set, generally in a ratio of 70:30.

[0034] 2) Model construction: Use XGBClassifier to input the feature parameters related to risk assessment in the mining area in the past five years for model training, and the output result is directly used as the prediction result of the corresponding risk probability.

[0035] The specific operations are as follows: Initial model training: Use the original training set data to construct the first CART tree to obtain Model 1.

[0036] Calculate the deviation: Model 1 will generate a prediction deviation on the training set, and a new training set 1 is constructed based on this deviation.

[0037] Iterative training: Use training set 1 to construct the second CART tree to obtain Model 2. Then calculate the deviation of Model 2, construct training set 2 based on this deviation, and then use training set 2 to construct the third CART tree to obtain Model 3. This process is continuously iterated, and each new tree attempts to fit the deviation of the previous tree, as Figure 4 shown.

[0038] Model fusion: Sum the prediction results of multiple models such as Model 1, Model 2, and Model 3 to obtain the final prediction result, and then use the test set to evaluate the performance of the final model.

[0039] 3) Risk warning mechanism: Set the threshold of the risk index to 0.7 according to requirements and risk tolerance. When the real-time calculated risk index exceeds the set threshold, the warning mechanism will be triggered.

[0040] Table 1 Risk warning values

[0041] S5, Intelligent regulation: Generate a regulation strategy according to the output results of the LSTM model and the risk assessment model, and adjust the injection pump frequency and valve opening (start the pressure relief circuit or stop water injection in case of emergency).

[0042] Regulation strategy generation: Based on the prediction results of the LSTM neural network prediction model (such asFigure 5 As shown in the figure, based on the results predicted by the risk assessment model, a control strategy is generated by combining an expert system and machine learning algorithms. A large amount of professional knowledge and experience rules regarding the deep underground storage of mine water are pre-stored in the expert system. When the risk assessment model detects an abnormal situation, a preliminary judgment is first made according to the expert rules. Then, the historical data and current monitoring data are trained through machine learning algorithms to continuously optimize the control strategy to adapt to different geological conditions and storage situations.

[0043] Actuator control: After the control strategy is generated, the control of the actuator is used to regulate the process of mine water storage. The actuators include devices such as valves and pumps. For example, when it is detected that the pressure in the storage area is too high, the system automatically controls the valve to open to release some pressure; when the water quality shows abnormal changes, the pump is started to adjust the injected or extracted water volume to maintain the water quality stability.

[0044] Correspondingly, as Figure 6 shown in the figure, an embodiment of the present invention also discloses a dynamic monitoring and control system for deep underground storage of mine water, including: Multi-parameter monitoring network 101: It includes a three-dimensional monitoring network composed of a variety of sensors, which is used to monitor injection wells, monitoring wells, storage layers, injection pipelines, and set key positions; the sensors include distributed optical fiber sensors, microseismic monitoring arrays, on-line water quality analyzers, etc., and the specific types of sensors can be selected according to actual needs; Edge computing node 102: A data processing unit deployed underground, which is used to realize data preprocessing and feature extraction; AI analysis platform 103: It includes an LSTM neural network prediction model based on deep learning and a risk assessment model, which are used to predict the evolution trend of storage layer parameters by the prediction model and calculate the risk index of the storage layer respectively; Intelligent control system 104: It includes an execution structure, which is used to generate a control strategy according to the output of the LSTM neural network prediction model; the frequency of the injection pump and the opening degree of the valve are adjusted through a PID controller, and in case of an emergency, the pressure relief circuit is started or the water injection is stopped according to the evaluation result of the risk assessment model.

[0045] For the specific implementation method of the above system, reference can be made to the aforementioned method for dynamic monitoring and control of deep underground storage of mine water, which will not be elaborated here.

[0046] The method and system described in this application organically integrate a variety of advanced sensor technologies for the first time, achieving real-time dynamic monitoring of all aspects and multiple parameters in the deep water sealing area of the mine, greatly improving the accuracy and comprehensiveness of monitoring. A complete intelligent system architecture is constructed, and all links from data collection, transmission, analysis and processing to intelligent control cooperate closely, realizing the high automation and intelligence of the system, effectively reducing manual intervention, and reducing costs and risks.

Claims

1. A method for dynamic monitoring and regulation of deep water storage in a mine, characterized in that, It includes the following steps: 1) Monitoring network deployment: Install relevant sensors at injection wells, monitoring wells, storage layers, injection pipelines, and set key positions to form a three-dimensional monitoring network; 2) Data collection and transmission: Collect the original data of sensors in the three-dimensional monitoring network, perform data preprocessing, and transmit the processed data to relevant models; 3) AI model construction and training: Collect historical data to construct a training set, construct an LSTM neural network prediction model, train the LSTM neural network prediction model, and use the trained network prediction model to predict the evolution trend of storage layer parameters; 4) Dynamic risk assessment: Construct a risk assessment model based on the XGBoost algorithm, and calculate the risk index of the storage layer in real time through the risk assessment model; Trigger an alarm when the risk index exceeds the set threshold; 5) Intelligent regulation: Generate a regulation strategy according to the output of the LSTM neural network prediction model; Adjust the injection pump frequency and valve opening through a PID controller; Start the pressure relief circuit or stop water injection in case of emergency according to the evaluation result of the risk assessment model.

2. The method for dynamic monitoring and regulation of deep water seal in a mine as claimed in claim 1, wherein The monitoring network deployment includes the following steps: Install several distributed fiber optic sensors in injection wells and monitoring wells to monitor the changes in temperature field and acoustic field in injection wells and monitoring wells in real time; Arrange several microseismic monitoring stations around the storage layer to monitor the signals emitted by rock microfractures; Install several sets of water quality sensors in the injection pipeline to monitor the pH value, water pressure, and flow rate of water in the main pipeline in real time; Install pore water pressure sensors and displacement monitors at key surface points.

3. The dynamic monitoring and regulation method for deep water sealing in a mine as described in claim 1, characterized in that, The data collection and transmission includes the following steps: Collect the original data of sensors at the edge node at a set frequency, perform data screening, noise reduction, and feature extraction on the collected original data, and then transmit the processed data to relevant models through an industrial ring network.

4. The dynamic monitoring and regulation method for deep water seal in a mine as described in claim 1, characterized in that, The method for constructing the LSTM neural network prediction model includes the following steps: 1) Data preparation and preprocessing: Data collection: Collect the operation data of mine water storage in the mining area in the past five years; Data preprocessing: Use interpolation or mean filling methods to process missing data, arrange the data in chronological order, and construct a time series data set; 2) Model construction: Input layer: Input the constructed time series data set; Hidden layer 1: Adopt a multi-layer LSTM structure, each layer includes multiple LSTM units, which are used to capture the long-term dependencies in time series data; Hidden layer 2: Add a fully connected layer after the LSTM layer to map the output of the LSTM to the target variable; Output layer: Finally, output the predicted value of the target variable; 3) Model training: Calculation of the loss function: Measure the prediction result by calculating the mean square error MSE, and measure the difference between the prediction result of the model and the real result; Adjustment of hyperparameters: If the value of the mean squared error (MSE) of the loss function is large and the prediction effect is not ideal, the hyperparameters need to be adjusted through grid search to optimize the prediction effect of the model. Select the number of layers, time step length, learning rate, and batch size as parameters, and use the exhaustive search method of grid search to train and evaluate all possible combinations of hyperparameters within the specified range of hyperparameter values. Finally, select the combination of hyperparameters with the best performance on the validation set. ； Where n is the number of samples, is the i-th true value, is the i-th predicted value; 4) Model evaluation: Evaluation metrics: Use the root mean squared error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE) to evaluate the performance of the model. ； ； ； 5) Validation and testing: Divide the dataset into a training set, a validation set, and a testing set. Adjust the model parameters through the validation set and use the testing set to evaluate the performance of the final model. 6) Visualization and application: Visualization of prediction results: Plot a comparison graph of the predicted values and the true values to show the prediction effect of the model. Model deployment: Deploy the trained model to the relevant platform to monitor the prediction operation status of the mining area in real time and conduct risk assessment.

5. The method for dynamic monitoring and regulation of deep water storage in a mine according to claim 1, characterized in that, The method for constructing the risk assessment model includes the following steps: 1) Data preparation and preprocessing: Data collection: Collect data related to the sealing system. Data cleaning: Process the missing values and outliers in the collected data. Feature engineering: Extract the feature parameters related to risk assessment. Data division: Divide the dataset into a training set and a testing set. 2) Model construction: Initial model training: Use the original training set data to construct the first CART tree to obtain Model 1. Calculate the bias: Model 1 will generate a prediction bias on the training set. Based on this bias, construct a new training set 1. Iterative training: Use training set 1 to construct the second CART tree to obtain Model 2. Then calculate the bias of Model 2, based on this bias construct training set 2, and then use training set 2 to construct the third CART tree to obtain Model 3. Repeat the above process for continuous iteration. Model fusion: Sum the prediction results of the multiple obtained models to get the final prediction result, and then use the testing set to evaluate the performance of the final model. 3) Risk warning mechanism: Set the threshold of the risk index according to the requirements and risk tolerance. When the real-time calculated risk index exceeds the set threshold, the warning mechanism will be triggered.

6. The method for dynamic monitoring and regulation of deep water storage in a mine as described in claim 1, characterized in that, The intelligent regulation includes the following steps: Regulation strategy generation: Based on the prediction results of the LSTM model and the risk assessment model, generate regulation strategies by combining the expert system and machine learning algorithms. A large amount of professional knowledge and experience rules about the deep sealing of mine water are pre-stored in the expert system. When the risk assessment model discovers abnormal situations, first make a preliminary judgment according to the expert rules. Then, train the historical data and current monitoring data through machine learning algorithms to continuously optimize the regulation strategies to adapt to different geological conditions and sealing situations. Execution mechanism control: After the regulation strategy is generated, control the execution mechanism to realize the regulation of the mine water sealing process.

7. A dynamic monitoring and control system for deep water storage in a mine, characterized in that, Including: Multi-parameter monitoring network: It includes a three-dimensional monitoring network composed of various sensors, and is used to monitor injection wells, monitoring wells, storage layers, injection pipelines, and set key positions; Edge computing node: A data processing unit deployed underground, used to implement data preprocessing and feature extraction; AI analysis platform: It includes an LSTM neural network prediction model and a risk assessment model based on deep learning, and is used to predict the evolution trend of the parameters of the storage layer by the model prediction and calculate the risk index of the storage layer respectively; Intelligent regulation system: It includes an execution structure, and is used to generate a regulation strategy according to the output of the LSTM neural network prediction model; Adjust the frequency of the injection pump and the opening of the valve through a PID controller, and start the pressure relief circuit or stop the water injection in case of emergency according to the evaluation result of the risk assessment model.

Images