An optimization method for multi-stage PRB in-situ remediation of acidic mine water in coal mines

Abstract

The present invention discloses an optimization method for multi-stage PRB in-situ remediation of acidic mine water in coal mines, which relates to the technical field of numerical simulation. The method includes: constructing a two-dimensional geometric model for multi-stage PRB to treat acidic mine water in coal mines; assigning dimensions and hydrogeological parameters to the fillers of each stage of PRB to obtain a numerical simulation model; batch-processing the numerical simulation model to generate multiple groups of initial design parameters for multi-stage PRB and inputting them into the optimization numerical model for operation to obtain multiple objective function values; predicting each objective function value through a backpropagation neural network model, and using the backpropagation neural network model with the best prediction effect as an alternative model for the numerical simulation model; coupling the alternative model with a fast non-dominated sorting genetic algorithm to obtain a Pareto optimal solution set. The present invention solves the problem that the existing technology lacks clear guiding principles and evaluation criteria, and improves the accuracy of the multi-stage PRB optimization method in practical engineering applications.

Description

Technical Field

[0001] The present invention relates to the technical field of numerical simulation, and particularly relates to an optimization method for multi-stage PRB in-situ remediation of coal mine acidic mine water. Background Art

[0002] China is rich in coal resources with a relatively high sulfur content. After the coal seams are exposed due to mining, high-sulfur coal will react in the coal seams to produce coal mine acidic mine water. Coal mine acidic mine water is a kind of groundwater pollution that deeply endangers the ecological environment and the health and safety of humans. As an in-situ remediation method for mine water, multi-stage PRB (Permeable Reactive Barriers) is a current hot technology in the field of mine water pollution remediation. By setting multi-stage PRB in coal mine tunnels, it can effectively treat characteristic pollutants such as Fe and SO4 in coal mine acidic mine water and improve the water quality of the mine water effluent. 2- and other characteristic pollutants, and improve the water quality of the mine water effluent.

[0003] At present, there is a lack of multi-objective optimization methods for multi-stage PRB to treat high-sulfur coal mine acidic mine water. Most optimization methods are based on statistical methods such as the response surface method. The theoretical research on the multi-objective optimization of multi-stage PRB is relatively scarce, and a systematic theoretical framework and methodology have not been formed. This leads to a lack of clear guiding principles and evaluation criteria in practical applications; and the model construction is difficult, and there is a lack of a model that meets the actual needs and is easy to solve. Therefore, it is difficult to meet the accurate and efficient multi-stage PRB optimization method required in practical engineering applications according to the current research. Summary of the Invention

[0004] Based on this, it is necessary to provide an optimization method for multi-stage PRB in-situ remediation of coal mine acidic mine water in view of the above technical problems.

[0005] An embodiment of the present invention provides an optimization method for multi-stage PRB in-situ remediation of coal mine acidic mine water, including:

[0006] Obtain the mine size parameters of the study area, determine the numerical simulation boundary and structure of multi-stage PRB according to the mine size parameters, so as to construct a two-dimensional geometric model for multi-stage PRB to treat coal mine acidic mine water; assign values to the sizes and hydrogeological parameters of the fillers of each stage of PRB in the two-dimensional geometric model, and construct a numerical simulation model for simulating the migration and diffusion of characteristic pollutants in coal mine acidic mine water in the fillers of each stage of PRB; and batch process the numerical simulation model to generate multiple groups of initial design parameters of multi-stage PRB;

[0007] Taking the service life and pollutant treatment capacity of multi-stage PRB as the objective function, and taking the hydrogeological parameters of the fillers of multi-stage PRB and the concentration load of characteristic pollutants as the constraint conditions, construct an optimization numerical model;

[0008] Input the initial design parameters of multiple groups and multiple levels of PRBs into the optimization numerical model for calculation to obtain multiple objective function values corresponding to the initial design parameters of multiple groups and multiple levels of PRBs; predict each objective function value through the backpropagation neural network model BPNN to obtain the prediction results corresponding to the objective function values;

[0009] Evaluate the prediction effect through the correlation coefficient between each objective function value and the prediction results corresponding to the objective function values, and use the backpropagation neural network model BPNN with the best prediction effect as the alternative model of the numerical simulation model;

[0010] Couple the alternative model of the numerical simulation model with the fast non-dominated sorting genetic algorithm NSGA-II to obtain the Pareto optimal solution set for treating coal mine acidic mine water by multiple-level PRBs, and select the optimal multiple-level PRB design scheme for in-situ remediation of coal mine acidic mine water in the study area from the Pareto optimal solution set.

[0011] Optionally, construct a numerical simulation model for treating coal mine acidic mine water by multiple-level PRBs that simulates the migration and diffusion of characteristic pollutants of coal mine acidic mine water in the fillers of each level of PRBs, specifically including:

[0012] Take limestone, coconut shell biochar, and anion exchange resin D201 as the fillers of multiple-level PRBs, and obtain the hydrogeological parameters of the fillers of each level; the hydrogeological parameters include: filler porosity, permeability coefficient, dispersion degree, and chemical reaction parameters;

[0013] Assign values to the filler porosity, permeability coefficient, dispersion degree, and chemical reaction parameters of the fillers of each level in the numerical simulation software, construct a groundwater seepage model and a reactive transport model for coal mine acidic mine water in each filling material of multiple-level PRBs, simulate the migration and diffusion of characteristic pollutants of coal mine acidic mine water in the fillers of each level, and obtain the numerical simulation model.

[0014] Optionally, the numerical simulation model includes a groundwater seepage model and a reactive transport model; the groundwater seepage model, and its formula is:

[0015] ;

[0016] Among them, K is the aquifer permeability coefficient, H is the water level elevation of the aquifer, S s is the water storage rate of the porous medium, t is the time, x and y are both independent variables;

[0017] The reactive transport model, and its formula is:

[0018] ;

[0019] wherein, c is the solute concentration, D L is the hydrodynamic dispersion tensor, u is the actual groundwater flow velocity, c p is the pollutant concentration adsorbed per unit dry weight of the solid, ρ b is the dry density of the solid medium, ρ b c p is the mass of the solute adsorbed on the porous medium; k b c p is the mass of the chemical reaction of the solute in the porous medium.

[0020] Optionally, with the service life and pollutant treatment capacity of the multi-stage PRB as the objective function, and the hydrogeological parameters of the multi-stage PRB filler and the concentration load of the characteristic pollutants as the constraints, specifically including:

[0021] Maximizing the effective service life of the multi-stage PRB f 1 Maximizing the effective water treatment volume of the multi-stage PRB f 2 Maximizing the amount of Fe treated by the multi-stage PRB f 3 Maximizing the amount of SO4 2- treated by the multi-stage PRB f 4 as the objective function, with the length of limestone L PRB1 the length of biochar L PRB2 the length of anion exchange resin L PRB3 the initial water head H in the porosity of limestone n PRB1 the porosity of anion exchange resin n PRB2 the porosity of biochar n PRB3 the permeability coefficient of limestone K PRB1 the permeability coefficient of anion exchange resin K PRB2 the permeability coefficient of biochar K PRB3 the dispersivityα , influent Fe concentration c Fe , influent SO4 2- concentration c SO4 2- are used as constraint conditions, and the calculation formulas are as follows:

[0022] Max f 1 = min( D Fe , D SO4 2- );

[0023] Max f 2 = v × f 1 × A ;

[0024] Max f 3 = f 2 × c Fe ;

[0025] Max f 4 = f 2 × c SO4 2- ;

[0026] s.t.{∑ L PRBi = 6; L PRBi , n PRBi , K PRBi , H in , c Fe ∈0.01×N; α , c SO4 2- ∈0.1×N, i = 1,2,3};

[0027] Among them, D Fe and D SO4 2- are the iThe time it takes for the concentration of a characteristic pollutant to exceed the groundwater quality standard limit from the start of the simulation to the effluent v is the Darcy velocity at the effluent A is the cross-sectional area in the vertical water flow direction c Fe is the influent Fe concentration c SO4 2- is the influent SO4 2- concentration, and N is the set of reference values

[0028] Optionally, the prediction effect is evaluated by the correlation coefficient between the objective function value and the prediction result, specifically including:

[0029] Construct a backpropagation neural network model BPNN in Matlab, and divide the samples into a training set and a test set according to a set ratio;

[0030] Input the training set into the backpropagation neural network model BPNN for training, and input the test set into the trained backpropagation neural network model BPNN for prediction;

[0031] Calculate the determination coefficient R 2 between the prediction result and the numerical simulation result, root mean square error RMSE, mean absolute error MAE, and mean bias error MBE to evaluate the prediction effect of the backpropagation neural network model BPNN;

[0032] Save the backpropagation neural network model BPNN with the best prediction effect and use it as an alternative model for the numerical simulation model.

[0033] Optionally, couple the alternative model of the numerical simulation model with the fast non-dominated sorting genetic algorithm NSGA-II, specifically including:

[0034] Use the Latin hypercube sampling method LHS in Matlab software to generate an initial population;

[0035] Input the initial population into the alternative model of the numerical simulation model for prediction, output the corresponding objective function, and calculate the Pareto front of the objective function through the fast non-dominated sorting genetic algorithm NSGA-II to output the Pareto optimal solution set for the multi-stage PRB treatment of coal mine acid mine water.

[0036] The above optimization method for in-situ remediation of multi-stage PRB of coal mine acid mine water provided by the embodiments of the present invention has the following beneficial effects compared with the prior art:

[0037] In the process of constructing the optimized numerical model of the present invention, the hydrogeological parameters of the most representative multi-stage PRB filler and the concentration load of characteristic pollutants during the in-situ remediation of multi-stage PRB are selected as constraint conditions, which are directly related to the performance of the PRB filler and the removal effect of pollutants in mine water, accurately reflecting the actual situation of mine water treatment, solving the problem that the existing technology lacks clear guiding principles and evaluation criteria, and improving the accuracy of the optimization method of multi-stage PRB in practical engineering applications.

[0038] In addition, the present invention introduces the backpropagation neural network model BPNN to predict the objective function value. By learning and simulating complex functional relationships through the backpropagation neural network model BPNN, rapid prediction is achieved, avoiding the problems of large computational amount and long computational time faced by traditional simulation-optimization methods, and having wider applicability. Description of the Drawings

[0039] Figure 1 It is a flow chart of an optimization method for in-situ remediation of multi-stage PRB of acidic mine water in coal mines provided in an embodiment;

[0040] Figure 2 It is the objective function of prediction for in-situ remediation of multi-stage PRB of acidic mine water in coal mines provided in an embodiment f 1 Graph of the linear fitting result between the value of the objective function of prediction and the numerical simulation value for in-situ remediation of multi-stage PRB of acidic mine water in coal mines provided in an embodiment;

[0041] Figure 3 It is the objective function of prediction for in-situ remediation of multi-stage PRB of acidic mine water in coal mines provided in an embodiment f 2 The linear fitting result between the value of the objective function of prediction and the numerical simulation value for in-situ remediation of multi-stage PRB of acidic mine water in coal mines provided in an embodiment;

[0042] Figure 4 It is a schematic diagram of the result of an optimization method for in-situ remediation of multi-stage PRB of acidic mine water in coal mines provided in an embodiment. Detailed Embodiments

[0043] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0044] The present invention solves the problem that the prior art lacks clear guiding principles and evaluation criteria, and avoids the limitations of single-objective optimization. A set of optimal solutions are calculated, which meet the requirements under multiple constraints, providing richer choices for the design of multi-stage PRB treatment processes. By introducing the backpropagation neural network model BPNN, the problems of large computational amount and long computational time faced by traditional simulation-optimization methods are avoided, thus having a wider range of applicability.

[0045] In one embodiment, an optimization method for in-situ remediation of multi-stage PRB of acidic mine water in coal mines is provided. As Figure 1 shown, the method includes:

[0046] Obtain the mine size parameters of the study area, determine the numerical simulation boundaries and structures of the multi-stage PRB according to the mine size parameters, so as to construct a two-dimensional geometric model for treating acidic mine water in coal mines with the multi-stage PRB. Assign values to the sizes and hydrogeological parameters of the PRB fillers at each stage in the two-dimensional geometric model, and construct a numerical simulation model for simulating the migration and diffusion of characteristic pollutants in acidic mine water in coal mines in the PRB fillers at each stage. And batch-process the numerical simulation model to generate multiple sets of initial design parameters for the multi-stage PRB.

[0047] Taking the hydrogeological parameters of the multi-stage PRB fillers and the concentration load of characteristic pollutants as constraint conditions, and taking the service life and pollutant treatment capacity of the multi-stage PRB as objective functions, construct an optimization numerical model.

[0048] Input multiple sets of initial design parameters of the multi-stage PRB into the optimization numerical model for calculation to obtain multiple objective function values corresponding to the multiple sets of initial design parameters of the multi-stage PRB. Use the backpropagation neural network model BPNN (Back Propagation Neural Network) to predict each objective function value to obtain a prediction result corresponding to the objective function value.

[0049] Evaluate the prediction effect through the correlation coefficient between each objective function value and the prediction result corresponding to the objective function value, and use the backpropagation neural network model BPNN with the best prediction effect as the alternative model of the numerical simulation model.

[0050] Couple the alternative model of the numerical simulation model with the fast non-dominated sorting genetic algorithm NSGA-II to obtain the Pareto optimal solution set for treating acidic mine water in coal mines with the multi-stage PRB, and select the optimal multi-stage PRB design scheme for in-situ remediation of acidic mine water in coal mines in the study area from the Pareto optimal solution set.

[0051] The specific implementation process includes:

[0052] S1 Obtain the mine size parameters in the study area, design the numerical simulation boundaries and structures of the multi-stage PRB, construct a two-dimensional geometric model for treating coal mine acidic mine water with the multi-stage PRB, assign values to the sizes and hydrogeological parameters of the fillers of each stage of the PRB in Comsol (COMSOL Multiphysics, a multi-physics simulation software), and generate a numerical model for treating coal mine acidic mine water with the multi-stage PRB.

[0053] The numerical simulation model for treating acidic mine water from high-sulfur coal mines with the multi-stage PRB provided by the present invention. The numerical simulation model of coal mine acidic mine water includes a groundwater seepage model and a reactive transport model. The specific construction steps are as follows:

[0054] Obtain the mine size parameters in the study area, design the numerical simulation boundaries and structures of the multi-stage PRB, construct a two-dimensional geometric model for treating coal mine acidic mine water with the multi-stage PRB, assign values to the sizes of the fillers of each PRB in Comsol software, and generate a multi-stage PRB geometric model.

[0055] Select limestone, coconut shell biochar, and anion exchange resin D201 as the filling materials for the multi-stage PRB, obtain the relevant hydrogeological parameters of the fillers of each stage of the PRB. The hydrogeological parameters include: filler porosity, permeability coefficient, dispersivity, and chemical reaction parameters. Assign values to the filler porosity, permeability coefficient, dispersivity, and chemical reaction parameters of the fillers of each stage of the PRB in the numerical simulation software, construct a groundwater seepage model and a reactive transport model for coal mine acidic mine water in each filling material of the multi-stage PRB, simulate the migration and diffusion of characteristic pollutants in coal mine acidic mine water in the fillers of each stage, and obtain a numerical simulation model.

[0056] Among them, the groundwater seepage model:

[0057] ;

[0058] In the formula: K is the aquifer permeability coefficient, m / d; H is the water level elevation of the aquifer, m; S s is the storage rate of the porous medium, 1 / m, that is, the amount of water released when the aquifer drops by one unit of water head; t is the time, d (d represents days); x and y are both independent variables.

[0059] Reactive transport model:

[0060] ;

[0061] In the formula: c is the solute concentration, mg / L; DL is the hydrodynamic dispersion tensor, m 2 / d; u is the actual groundwater flow velocity, m / d; c p is the pollutant concentration adsorbed per unit dry weight of solid, mg / kg; ρ b is the dry density of the solid medium, kg / m 3 , ρ b c p is the mass of solute adsorbed on the porous medium; k b c p is the chemical reaction mass of the solute in the porous medium.

[0062] S2 saves the constructed numerical simulation model for treating coal mine acid mine drainage with a multi-stage PRB, connects Comsol using the Comsol with Matlab interface, writes code using Matlab (Matrix Laboratory), generates 500 sets of initial design parameters for the multi-stage PRB, inputs them into the saved Comsol model for calculation, and outputs the corresponding objective function values to achieve batch processing of the numerical simulation model.

[0063] S3 selects the service life and pollutant treatment capacity of the multi-stage PRB as the objective function, and selects the hydrogeological parameters of the multi-stage PRB filler and the concentration load of the characteristic pollutants as the constraint conditions to construct an optimization numerical model.

[0064] The relevant parameters of the constraint conditions and the objective function include:

[0065] Taking the maximization of the effective service life of the multi-stage PRB ( f 1 , unit: d), the maximization of the effective water treatment capacity of the multi-stage PRB ( f 2 , unit: m 3 ), the maximization of the amount of Fe treated by the multi-stage PRB ( f 3 , unit: kg), the maximization of the amount of SO4 2- treated by the multi-stage PRB ( f 4 , unit: kg) as the objective function, and taking the length of limestone ( L PRB1 , unit: m), the length of biochar ( L PRB2 , unit: m), the length of anion exchange resin ( L PRB3 , unit: m), the initial water head (H in , unit: m), limestone porosity ( n PRB1 , dimensionless), anion exchange resin porosity ( n PRB2 , dimensionless), biochar porosity ( n PRB3 , dimensionless), limestone permeability ( K PRB1 , unit: m / d), anion exchange resin permeability ( K PRB2 , unit: m / d), biochar permeability ( K PRB3 , unit: m / d), dispersivity ( α , unit m), influent Fe concentration ( c Fe , unit: mol / m 3 ), influent SO4 2- concentration ( c SO4 2- , unit: mol / m 3 ) as constraints, the calculation formula is as follows:

[0066] Max f 1 = min( D Fe , D SO4 2- );

[0067] Max f 2 = v × f 1 × A ;

[0068] Max f 3 = f 2 × c Fe ;

[0069] Max f 4 = f 2 × c SO4 2- ;

[0070] s.t.{∑ L PRBi = 6;L PRBi , n PRBi , K PRBi , H in , c Fe ∈ 0.01×N; α , c SO4 2- ∈ 0.1×N, i = 1,2,3};

[0071] Where: D Fe and D SO4 2- is the i th characteristic pollutant from the start of the simulation to the time when the concentration at the effluent exceeds the groundwater quality standard limit value (Fe is 2 mg / L, SO4 2- is 350 mg / L), unit: d; v is the Darcy velocity at the effluent, unit m / d; A is the cross-sectional area in the vertical water flow direction, unit m 2 ; c Fe and c SO4 2- are the influent Fe and SO4 2- concentrations respectively, unit mol / m 3 , N is the set of reference values.

[0072] As Figure 2 shown, Figure 2 is the linear fitting result of the predicted objective function f 1 (maximizing the effective service life of the multi-stage PRB) value and the numerical simulation value. The abscissa in the figure is the numerical simulation value, the ordinate is the predicted value of the BPNN model, and the color scale in the figure represents the point density, which can characterize the concentration degree of the data. It can be seen from the figure that f 1 The determination coefficient R 2 between the numerical simulation value and the predicted value of f 1 is 0.986, and the error indicators such as MAE, MBE and RMSE are relatively low, indicating that the machine learning model is relatively accurate in predicting

[0073] As Figure 3 shown, Figure 3 is the predicted objective function f 2The linear fitting result of the value of (maximizing the effective treatment water volume of multi-stage PRB) and the numerical simulation value. The abscissa in the figure is the numerical simulation value, and the ordinate is the predicted value of the BPNN model. The color scale in the figure represents the point density, which can characterize the concentration degree of the data. It can be seen from the figure that f 2 The determination coefficient R between the numerical simulation value and the predicted value of 2 is 0.993, and the error indexes such as MAE, MBE and RMSE are relatively low, indicating that this machine learning model is f 2 accurately predicted.

[0074] S4 uses the back propagation neural network model BPNN (Back Propagation Neural Network) to predict the output result of the optimized numerical model, and calculates and evaluates the prediction result. After multiple calculations, the neural network with the best prediction effect is saved.

[0075] The multi-objective optimization result of treating acidic mine water in high-sulfur coal mines by multi-stage PRB provided by the present invention includes the following steps:

[0076] Save the result calculated in S3 for subsequent prediction. Build a back propagation neural network model BPNN in Matlab, divide 500 samples into a training set and a test set according to a set ratio (8:2), predict the saved result, and calculate the determination coefficient R 2 of the prediction result and the numerical simulation result, root mean square error RMSE, mean absolute error MAE and mean bias error MBE to evaluate the prediction effect of the back propagation neural network model BPNN. By debugging the hyperparameters of the back propagation neural network model BPNN multiple times, save the back propagation neural network model BPNN with the best prediction effect and use it as an alternative model for the numerical simulation model.

[0077] As Figure 4 shown, Figure 4 is the 500 groups of Pareto optimal solution sets calculated by the trained BPNN model, which are represented in the form of a parallel coordinate plot. In the figure f 1 — f 4 ( f 1 is to maximize the effective service life of the multi-stage PRB, f 2 is to maximize the effective treatment water volume of the multi-stage PRB, f 3 is to maximize the amount of Fe treated by the multi-stage PRB, f 4 is to maximize the treatment of SO4 by the multi-stage PRB 2-The values (of the quantity) respectively represent the values of four objective functions, where f 1 The value range of is 1440 - 5235; f 2 The value range of is 910 - 1390; f 3 The value range of is 19 - 42; f 4 The value range of is 1428 - 3586. Each curve in the figure represents the objective function values corresponding to each Pareto optimal solution, and the color of the curve represents f 1 The color scale of, and the correlation relationship between the objective functions can be judged according to the color of the color scale. For example f 1 The larger, f 2 the smaller. It can be seen from Figure 4 that the predicted values R of the objective functions by the trained BPNN model 2 are 0.986 and 0.993 respectively, indicating that the model has a good prediction effect and can be used as an alternative model for the multi - level PRB numerical simulation model.

[0078] S5 uses the Latin Hypercube Sampling method LHS in Matlab software to generate the initial population, inputs it into the alternative model of the numerical simulation model (the saved Back Propagation Neural Network model BPNN) for prediction, outputs the corresponding objective function values, and calculates the Pareto front of the objective function values through the Non - dominated Sorting Genetic Algorithm II NSGA - II, and outputs the Pareto optimal solution set of the objective function (Pareto Optimal Set, also known as the Pareto optimal collection).

[0079] Set the initial population to 500 and the number of iterations to 1000. Perform a quick non - dominated sort on the population and calculate the corresponding crowding degree, and generate the offspring population through selection, crossover, and mutation operations. Repeat the above steps until the specified number of iterations is reached, and output the Pareto optimal solution set of the objective function. Save the calculated Pareto optimal solution set of the objective function and represent the predicted objective function values through a parallel coordinate plot.

[0080] The above - described embodiments only represent several implementation manners of the present invention. The description is relatively specific and detailed, but it should not be construed as a limitation on the scope of the invention patent. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can still be made, and these all belong to the protection scope of the present invention.

Claims

1. An optimization method for multi-stage PRB in-situ remediation of acidic mine water in coal mines, characterized in that, Including: Construct a numerical simulation model for treating coal mine acid mine drainage with a multi-stage PRB according to the mine size parameters in the study area; Batch process the numerical simulation model to generate multiple sets of initial design parameters for the multi-stage PRB; Construct an optimization numerical model with the service life and pollutant treatment capacity of the multi-stage PRB as the objective function and the hydrogeological parameters of the multi-stage PRB filler and the concentration load of characteristic pollutants as the constraint conditions; Among them, taking the service life and pollutant treatment capacity of the multi-stage PRB as the objective function and the hydrogeological parameters of the multi-stage PRB filler and the concentration load of characteristic pollutants as the constraint conditions specifically include: To maximize the effective service life of the multi-stage PRB f 1 、To maximize the effective water treatment capacity of the multi-stage PRB f 2 、To maximize the amount of Fe treated by the multi-stage PRB f 3 、To maximize the amount of SO4 treated by the multi-stage PRB 2- amount f 4 As the objective function, with the limestone length L PRB1 、Biochar length L PRB2 、Anion exchange resin length L PRB3 、Initial head H in 、Limestone porosity n PRB1 、Anion exchange resin porosity n PRB2 、Biochar porosity n PRB3 、Limestone permeability K PRB1 、Anion exchange resin permeability K PRB2 、Biochar permeability K PRB3 、Dispersivity α 、Inlet Fe concentration c Fe 、Inlet SO4 2- concentration c SO4 2- As the constraint conditions, the calculation formula is as follows: Max f 1 = min( D Fe , D SO4 2- ); Max f 2 = v × f 1 × A ; Max f 3 = f 2 × c Fe ; Max f 4 = f 2 × c SO4 2- ; s.t. {∑ L PRBi = 6; L PRBi 、 n PRBi 、 K PRBi 、 H in 、 c Fe ∈0.01×N; α 、 c SO4 2- ∈0.1×N, i = 1, 2, 3}; Among them, D Fe and D SO4 2- is the time taken for the concentration of the i th characteristic pollutant to exceed the groundwater quality standard limit from the start of the simulation to the effluent, v is the Darcy velocity at the effluent, A is the cross-sectional area in the vertical water flow direction, c Fe is the influent Fe concentration, c SO4 2- is the influent SO4 2- concentration, and N is the set of reference values Input multiple sets of initial design parameters of the multi-stage PRB into the optimization numerical model for calculation to obtain multiple objective function values; predict each objective function value through a backpropagation neural network model to obtain prediction results; evaluate the prediction effect through the correlation coefficient between each objective function value and the prediction results, and use the backpropagation neural network model with the best prediction effect as the alternative model of the numerical simulation model; Couple the alternative model of the numerical simulation model with the fast non-dominated sorting genetic algorithm to obtain the Pareto optimal solution set for treating coal mine acid mine drainage with the multi-stage PRB, and select the optimal multi-stage PRB design scheme for in-situ remediation of coal mine acid mine drainage in the study area from the Pareto optimal solution set.

2. The optimized method for multi-stage PRB in-situ remediation of acidic mine water in coal mines according to claim 1, wherein The specific steps of constructing a numerical simulation model for treating coal mine acid mine drainage with a multi-stage PRB according to the mine size parameters in the study area include: Determine the numerical simulation boundary and structure of the multi-stage PRB according to the mine size parameters to construct a two-dimensional geometric model for treating coal mine acid mine drainage with the multi-stage PRB; Assign values to the sizes and hydrogeological parameters of the PRB fillers at all levels in the two-dimensional geometric model to construct a numerical simulation model for treating coal mine acid mine drainage with the multi-stage PRB; the numerical simulation model is used to simulate the migration and diffusion of characteristic pollutants in coal mine acid mine drainage in the PRB fillers at all levels.

3. The optimized method for multi-stage PRB in-situ remediation of acidic mine water in coal mines according to claim 2, characterized in that, The specific steps of assigning values to the sizes and hydrogeological parameters of the PRB fillers at all levels in the two-dimensional geometric model to construct a numerical simulation model for treating coal mine acid mine drainage with the multi-stage PRB include: Use limestone, coconut shell biochar, and anion exchange resin D201 as the PRB fillers at all levels to obtain the hydrogeological parameters of the PRB fillers at all levels; the hydrogeological parameters include: filler porosity, permeability coefficient, dispersion degree, and chemical reaction parameters; Assign values to the filler porosity, permeability coefficient, dispersion degree, and chemical reaction parameters of the PRB fillers at all levels in the numerical simulation software to construct a groundwater seepage model and a reactive transport model for coal mine acid mine drainage in each filling material of the multi-stage PRB, simulate the migration and diffusion of characteristic pollutants in coal mine acid mine drainage in the fillers at all levels, and obtain the numerical simulation model.

4. The optimized method for multi-stage PRB in-situ remediation of acidic mine water in coal mines according to claim 3, characterized in that, The numerical simulation model includes a groundwater seepage model and a reactive transport model; the formula of the groundwater seepage model is: ； Among them, K is the permeability coefficient of the aquifer, H is the water level elevation of the aquifer, S s is the water storage rate of the porous medium, t is time, x and y are both independent variables; The formula of the reactive transport model is: ； Among them, c is the solute concentration, D L is the hydrodynamic dispersion tensor, u is the actual groundwater flow velocity, c p is the pollutant concentration adsorbed per unit dry weight of solid, ρ b is the dry density of the solid medium, ρ b c p is the mass of solute adsorbed on the porous medium; k b c p is the chemical reaction mass of the solute in the porous medium.

5. The optimization method for multi-stage PRB in-situ remediation of acidic mine water in coal mines according to claim 1, characterized in that The specific steps of evaluating the prediction effect through the correlation coefficient between each objective function value and the prediction results include: Build a backpropagation neural network model in Matlab, and divide the samples into a training set and a test set according to a set ratio; Input the training set into the backpropagation neural network model for training, and input the test set into the trained backpropagation neural network model for prediction; Calculate the coefficient of determination R between the prediction result and the numerical simulation result 2 , root mean square error RMSE, mean absolute error MAE, and mean bias error MBE to evaluate the prediction effect of the backpropagation neural network model; Save the backpropagation neural network model with the best prediction effect and use it as an alternative model for the numerical simulation model.

6. The optimization method for multi-stage PRB in-situ remediation of acidic mine water in coal mines according to claim 1, wherein Coupling the alternative model of the numerical simulation model with the fast non-dominated sorting genetic algorithm specifically includes: Use the Latin hypercube sampling method LHS in Matlab software to generate the initial population; Input the initial population into the alternative model of the numerical simulation model for prediction, output the corresponding objective function values, and calculate the Pareto front of the objective function values through the fast non-dominated sorting genetic algorithm, and output the Pareto optimal solution set for treating coal mine acid mine water by multi-stage PRB.

Images