Mine water inflow prediction method based on optimized artificial neural network model
By optimizing the artificial neural network model, combining data processing and multimodal data fusion, the problem of large error in water inflow prediction of coal mines is solved, and high-precision water inflow prediction and energy consumption optimization are achieved.
Patent Information
- Application Number
- CN202510436146.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-09
- Publication Date
- 2025-07-22
AI Technical Summary
The existing technology has large errors in the prediction of mine water in coal mines, and traditional methods are difficult to accurately predict, which affects the prevention of water inrush accidents.
The optimized artificial neural network model is adopted, combining data statistics and preprocessing, BP neural network training, particle swarm optimization algorithm and data-model dual-driven update, integrating timing and spatial data, and synthesis of rare event samples by generating adversarial networks to perform multimodal data fusion prediction.
It improves the accuracy of the mine water inrush prediction, reduces prediction errors, improves the prediction accuracy and economicality of the drainage system, and reduces energy consumption.
Smart Images
Figure CN120355017A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of mine water disaster prevention and control, and specifically to a prediction method for mine water inflow based on an optimized artificial neural network model. Background Technique
[0002] Among various coal mine disaster accidents, the number of water inrush accidents ranks second. Accurately predicting the mine water inflow is a necessary link to reduce the probability of water inrush accidents. During the coal mine mining process, traditional methods such as the large shaft method are commonly used to predict the mine water inflow. However, there is still a certain gap between its prediction results and the measured data. With the rapid development of artificial neural networks, taking the measured data as sample data and inputting it into a trained BP neural network time series prediction model based on the particle swarm optimization algorithm, and then predicting the mine water inflow, relatively accurate mine water inflow prediction data can be obtained. Summary of the Invention
[0003] Aiming at the deficiencies of the existing technology, the purpose of the present invention is to provide a prediction method for mine water inflow based on an optimized artificial neural network model to solve the problems raised in the above background technique. The present invention predicts the mine water inflow through an optimized artificial neural network system, reducing the errors brought by traditional water inflow prediction methods when predicting the water inflow.
[0004] To achieve the above purpose, the present invention is realized through the following technical solutions: A prediction method for mine water inflow based on an optimized artificial neural network model, including the following steps:
[0005] Step 1: Data statistics and preprocessing. According to the mine water inflow record ledger, organize the water consumption data in chronological order. Unify the unit of water inflow records, integrate hydrogeological monitoring data, geological structure data, and mining activity records, use GIS technology to achieve spatial interpolation, establish an association matrix between the time series and spatial coordinates, and identify outliers through a clustering algorithm for outliers, and perform correction or elimination;
[0006] Step 2: Establish an initial BP neural network training model, use the Network-in-Network technology to dynamically adjust the number of hidden layers, set the number of nodes according to the Kolomogorov theorem, and construct a parallel model cluster;
[0007] Step 3: Establish a particle swarm optimization algorithm, generate an initial particle swarm, use the Halton sequence to replace random numbers, evaluate the prediction error through fitness calculation, and adopt a dynamic adjustment strategy for the update mechanism;
[0008] Step 4: Input the actual observed data of the water inflow, train the model until it meets the accuracy, and use the data division strategy to divide the data set into a training set, a validation set, and a test set;
[0009] Step 5: Use the trained neural network model to predict the future water inflow of the mine and establish a data-model dual-driven update: When the prediction error exceeds the threshold three times in a row, trigger model retraining;
[0010] Step 6: Input the water inflow prediction result into the optimal scheduling model of the drainage system and calculate the equipment start-stop plan corresponding to the prediction results with different confidence levels.
[0011] Furthermore, the hydrogeological monitoring data in Step 1 includes dynamic water levels and water quality parameters, the geological structure data includes fault distribution and fracture development degree, and the mining activity records include mining progress and drainage system parameters.
[0012] Furthermore, the processing of outliers identifies outliers through the DBSCAN clustering algorithm, uses the ARIMA-LSTM hybrid model for multi-scale interpolation of missing periods, and combines the isolation forest algorithm with the manual review mechanism to retain the real abnormal data during geological mutation periods.
[0013] Furthermore, in Step 2, the input layer of the dynamic network structure includes lagged water inflow, cumulative rainfall, and mining depth change rate, and uses a residual block structure. The node number formula is:
[0014]
[0015] where dynamic adjustment (α ∈ [5, 10]), N h is the number of hidden layer nodes, which determines the network's ability to capture the spatio-temporal characteristics of water inflow, N i is the number of input layer nodes, including time series features and spatial feature data, N o represents the number of output layer nodes, which is 1 or n, α is the adjustment factor, and the number of hidden layer nodes is used to compensate for the non-linear loss and prevent model underfitting.
[0016] Furthermore, in Step 3, the velocity formula for each particle to update in each iteration is as follows:
[0017]
[0018] The position formula for each particle to update in each iteration is as follows:
[0019]
[0020] where w is the inertia weight, c1 / c2 are the learning factors, r1 / r2 ∈ [0, 1] are random numbers, pbest is the individual historical optimum, and gbest is the global optimum.
[0021] Furthermore, the fitness calculation uses weighted MAE, increases the error weight of recent data, introduces a velocity compression factor in the dynamic adjustment strategy to prevent divergence, and sets the termination condition for water inrush prediction. The termination condition uses multi-index judgment and simultaneously satisfies including iteration > 500 times or fitness variance < 1e-6 or gbest has no improvement for 50 consecutive times.
[0022] Furthermore, step three also includes constraint handling. In the constraint handling process, boundary constraints are added to physical parameters such as permeability coefficient, and the Metropolis criterion of the simulated annealing algorithm is introduced in the later iteration.
[0023] Furthermore, the training set includes typical working conditions and extreme events; the validation set includes the latest data with time continuity; the test set includes representative samples with spatial distribution.
[0024] Furthermore, in step five, it also includes synthesizing rare water inrush event samples through a generative adversarial network (GAN), applying wavelet transform to separate the trend term and fluctuation term of water inrush data, and the prediction types include short-term prediction, medium-term prediction, and long-term prediction.
[0025] Furthermore, in step six, it also includes economic evaluation, calculating the drainage energy consumption savings brought by the improvement of prediction accuracy:
[0026]
[0027] where Qt is the drainage volume in period t; H t is the head; η is the comprehensive efficiency of the pump unit; P elec is the electricity price per period; Δt is the operation duration.
[0028] Advantages of the present invention:
[0029] 1. This method for predicting mine water inrush based on an optimized artificial neural network model makes the water inrush prediction model based on actual mine water inrush observation data, effectively reduces the prediction error brought by traditional water inrush prediction methods, improves the prediction accuracy of mine water inrush in production activities, and has a guiding role in aspects such as mine drainage design.
[0030] 2. This method for predicting mine water inrush based on an optimized artificial neural network model integrates time series data (historical water inrush curve), spatial data (three-dimensional coordinate coding of mining areas), and environmental data (rainfall, mining intensity), and dynamically weights the contribution degrees of different features through an attention mechanism, enabling it to have the ability of multi-modal data fusion, achieving a breakthrough improvement in prediction accuracy, being able to capture the non-linear correlation between water level mutations and mining activities, and having a smaller peak appearance time error. Description of the drawings
[0031] Figure 1This is the algorithm flowchart of the mine water inrush prediction method based on the optimized artificial neural network model of the present invention. Detailed implementation manners
[0032] To make the technical means, creative features, achieved purposes and effects of the present invention easy to understand, the present invention will be further described below in conjunction with the detailed implementation manners.
[0033] Please refer to Figure 1 , the present invention provides the following technical solutions: a mine water inrush prediction method based on an optimized artificial neural network model, including the following steps:
[0034] Step 1, data statistics and preprocessing. According to the mine water inrush record ledger, the water consumption data is sorted in chronological order. The unit of the water inrush record is unified, the hydrogeological monitoring data, geological structure data and mining activity records are integrated, spatial interpolation is realized by using GIS technology, a correlation matrix of time series and spatial coordinates is established, and outliers are identified by a clustering algorithm for correction or elimination.
[0035] The hydrogeological monitoring data includes dynamic water levels and water quality parameters, the geological structure data includes fault distribution and fracture development degree, and the mining activity records include mining progress and drainage system parameters.
[0036] The processing of outliers identifies outliers through the DBSCAN clustering algorithm, and a multi-scale interpolation is performed on the missing time period by using an ARIMA-LSTM hybrid model. At the same time, the isolation forest algorithm and the manual review mechanism are combined to retain the real abnormal data in the geological mutation period.
[0037] Step 2, establish an initial BP neural network training model, use the Network-in-Network technology to dynamically adjust the number of hidden layers, the number of nodes is set according to the Kolomogorov theorem, and a parallel model cluster is constructed.
[0038] The input layer of the dynamic network structure includes the lagged water inrush, the cumulative rainfall value, and the change rate of the mining depth, and adopts a residual block structure. The formula for the number of nodes is:
[0039]
[0040] where the dynamic adjustment (α ∈ [5, 10]), N h is the number of hidden layer nodes, which is used to determine the ability of the network to capture the spatio-temporal characteristics of the water inrush. N i is the number of input layer nodes, including time series features and spatial feature data. N o represents the number of output layer nodes, which is 1 or n. α is an adjustment factor, and the number of hidden layer nodes is used to compensate for the non-linear loss and prevent the model from being underfitted;
[0041] Step 3: Establish a particle swarm optimization algorithm to generate an initial particle swarm. Use the Halton sequence to replace random numbers, evaluate the prediction error through fitness calculation, and adopt a dynamic adjustment strategy for the update mechanism.
[0042] The velocity formula for each particle to update in each iteration is as follows:
[0043]
[0044] The position formula for each particle to update in each iteration is as follows:
[0045]
[0046] Among them, w is the inertia weight, c1 / c2 are the learning factors, r1 / r2 ∈ [0, 1] are random numbers, pbest is the individual historical optimum, and gbest is the global optimum.
[0047] The fitness calculation uses weighted MAE, increases the weight of recent data errors, introduces a velocity compression factor in the dynamic adjustment strategy to prevent divergence, and sets the termination condition for water inflow prediction. The termination condition uses multi-index judgment and simultaneously satisfies including iteration > 500 times or fitness variance < 1e-6 or gbest has no improvement for 50 consecutive times.
[0048] In the constraint handling process, boundary constraints are added to physical parameters such as permeability coefficient, and the Metropolis criterion of the simulated annealing algorithm is introduced in later iterations;
[0049] Step 4: Input the actual observed data of water inflow, train the model to meet the accuracy, and the data partitioning strategy divides the data set into a training set, a validation set, and a test set. The training set contains typical working conditions and extreme events; the validation set contains the latest data with time continuity; the test set contains representative samples with spatial distribution;
[0050] Step 5: Use the trained neural network model to predict the future water inflow of the mine, and establish a data-model dual-driven update: When the prediction error exceeds the threshold for 3 consecutive times, trigger model retraining.
[0051] Synthesize rare water inrush event samples through a generative adversarial network (GAN), apply wavelet transform to separate the trend term and fluctuation term of water inflow data, and the prediction types include short-term prediction, medium-term prediction, and long-term prediction;
[0052] Step 6: Input the water inflow prediction results into the optimized scheduling model of the drainage system, and calculate the equipment start-stop schemes corresponding to the prediction results with different confidence levels.
[0053] Calculate the drainage energy consumption savings brought by the improvement of prediction accuracy:
[0054]
[0055] where Qt is the drainage volume in period t (m 3 / h);
[0056] H t is the head;
[0057] η is the comprehensive efficiency of the pump group (the range in this embodiment is 0.6 - 0.8);
[0058] P elec is the time-of-use electricity price (yuan / kWh);
[0059] Δt is the operation duration (h).
[0060] This embodiment also provides a calculation of the carbon reduction amount generated by the above method, and the formula is as follows:
[0061] ΔCO2 = ΔE × EF grid
[0062] where EF grid = 0.583 kgCO2 / kWh).
[0063] The above shows and describes the basic principles, main features and advantages of the present invention. For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and without departing from the spirit or basic features of the present invention, the present invention can be implemented in other specific forms.
[0064] In addition, it should be understood that although this specification is described according to embodiments, not every embodiment only contains an independent technical solution. This narrative way of the specification is only for clarity. Those skilled in the art should regard the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.
Claims
1. A prediction method for mine water inrush based on an optimized artificial neural network model, characterized in that: It includes the following steps: Step 1: Data statistics and preprocessing. According to the mine water inflow record ledger, organize the water consumption data in chronological order, unify the unit of water inflow records, and integrate hydrogeological monitoring data, geological structure data, and mining activity records; Step 2: Establish an initial BP neural network training model. Use the Network-in-Network technology to dynamically adjust the number of hidden layers. The number of nodes is set according to the Kolomogorov theorem, and a parallel model cluster is constructed; Step 3: Establish a particle swarm optimization algorithm, generate an initial particle swarm, use the Halton sequence to replace random numbers, evaluate the prediction error through fitness calculation, and adopt a dynamic adjustment strategy for the update mechanism; Step 4: Input the actual observed data of water inflow, train the model until it meets the accuracy requirements. The data partitioning strategy divides the data set into a training set, a validation set, and a test set; Step 5: Use the trained neural network model to predict the future water inflow of the mine, and establish a data-model dual-driven update: when the prediction error exceeds the threshold three times consecutively, trigger model retraining; Step 6: Input the water inflow prediction results into the optimized scheduling model of the drainage system, and calculate the equipment start-stop schemes corresponding to the prediction results with different confidence levels.
2. The mine water inflow prediction method based on the optimized artificial neural network model according to claim 1, characterized in that: The hydrogeological monitoring data in Step 1 includes dynamic water levels and water quality parameters. The geological structure data includes fault distribution and fracture development degree. The mining activity records include mining progress and drainage system parameters.
3. The mine water inflow prediction method based on an optimized artificial neural network model according to claim 2, characterized in that: The processing of outliers identifies outliers through the DBSCAN clustering algorithm, and at the same time combines the isolation forest algorithm with the manual review mechanism to retain the real abnormal data during the geological mutation period.
4. The mine water inrush prediction method based on an optimized artificial neural network model according to claim 1, characterized in that: In Step 2, the input layer of the dynamic network structure includes the lagged water inflow, cumulative rainfall value, and mining depth change rate, and adopts a residual block structure. The formula for the number of nodes is: Among them, dynamic adjustment (α ∈ [5, 10]), N h is the number of hidden layer nodes, which is used to determine the network's ability to capture the spatio-temporal characteristics of water inflow, N i is the number of input layer nodes, including time series features and spatial feature data, N o represents the number of output layer nodes, which is 1 or n. α is a regulatory factor, and the number of hidden layer nodes is used to compensate for non-linear losses and prevent the model from underfitting.
5. The mine water inrush prediction method based on an optimized artificial neural network model according to claim 1, characterized in that In Step 3, the formula for updating the velocity of each particle in each iteration is as follows: The formula for updating the position of each particle in each iteration is as follows: Among them, w is the inertia weight, c1 / c2 are the learning factors, r1 / r2 ∈ [0,1] are random numbers, pbest is the individual historical optimum, and gbest is the global optimum.
6. The mine water inrush prediction method based on the optimized artificial neural network model according to claim 5, wherein: The fitness calculation uses weighted MAE, increases the weight of recent data errors, introduces a velocity compression factor in the dynamic adjustment strategy to prevent divergence, and sets the termination condition for water inflow prediction.
7. The mine water inflow prediction method based on an optimized artificial neural network model according to claim 6, characterized in that: Step 3 also includes constraint processing. During the constraint processing, boundary constraints are added to physical parameters such as permeability coefficient, and the Metropolis criterion of the simulated annealing algorithm is introduced in the later iteration.
8. The mine water inrush prediction method based on the optimized artificial neural network model according to claim 1, characterized in that: The training set contains typical working conditions and extreme events; the validation set contains the latest data with time continuity; the test set contains representative samples with spatial distribution.
9. The method for predicting mine water inflow based on an optimized artificial neural network model according to claim 8, characterized in that: Step 5 also includes synthesizing rare water inrush event samples through a generative adversarial network (GAN), applying wavelet transform to separate the trend term and fluctuation term of water inflow data, and the prediction types include short-term prediction, medium-term prediction, and long-term prediction.
10. The mine water inflow prediction method based on the optimized artificial neural network model according to claim 9, wherein: Step 6 also includes economic evaluation, calculating the drainage energy consumption savings brought by the improvement of prediction accuracy: where Qt is the drainage volume in period t; H t is the head; η is the comprehensive efficiency of the pump unit; P elec is the electricity price per period; Δt is the operation duration.