Permeable reaction dam structure for preventing and controlling acid heavy metal leaching water in coal gangue stockyard

Abstract

This invention relates to the fields of geological disaster prevention and control and artificial intelligence technology, and discloses a permeable reactive barrier dam structure for controlling acidic heavy metal leaching water in coal gangue dumps. The structure includes: a rigid load-bearing dam body for bearing structural loads and intercepting solid particles; replaceable reactive filter units, detachably installed in a modular filter unit installation compartment for purifying the flowing leaching water; the modular filter unit installation compartment being built into the non-load-bearing space within the rigid load-bearing dam body; a sensor system deployed on the rigid load-bearing dam body for real-time data acquisition; and a smart operation and maintenance platform, communicatively connected to the sensor system, for predicting the performance degradation trend of the replaceable reactive filter units based on operational data, and generating and outputting operation and maintenance instructions based on the prediction results. This invention achieves integrated and coordinated prevention and control of geological disasters and water pollution in coal gangue dumps, significantly improving the system's engineering feasibility, operational stability, and service life.

Description

Technical Field

[0001] This invention belongs to the fields of geological disaster prevention and control and artificial intelligence technology. Specifically, it relates to a permeable reactive barrier dam structure for the prevention and control of acidic heavy metal leaching water in coal gangue dumps. It is particularly suitable for coal gangue dumps with stable slopes and no risk of large-scale deep sliding, but with continuous rainfall leaching water pollution, accompanied by small shallow gangue slides and silt erosion. Background Technology

[0002] With the rapid development of my country's coal industry, coal gangue, as a major solid waste generated during coal mining and washing, has accumulated to a huge amount, forming numerous large-scale coal gangue dumps. In rainy areas such as southwest my country, the oxidative hydrolysis of sulfide minerals in the coal gangue under long-term rainfall leaching produces acidic mine wastewater with high acidity and high concentration of heavy metals, which seeps into surrounding water bodies, causing serious soil and aquatic ecological pollution. At the same time, the long-term accumulation of coal gangue in these dumps can easily lead to slope instability, posing a risk of sudden geological disasters such as landslides and debris flows, thus posing a dual threat to the ecological security and the safety of people's lives and property in the surrounding areas.

[0003] Currently, the remediation technologies for coal gangue dumps mainly fall into two categories: geological disaster prevention and control, and water pollution control. These two approaches generally suffer from functional fragmentation and poor synergy. Firstly, traditional rigid protective structures such as concrete retaining dams and grid dams can only intercept particulate matter and prevent geological disasters; they cannot purify dissolved acidic substances and heavy metal ions. Polluted water can still overflow and infiltrate into the downstream environment, failing to fundamentally solve the ecological pollution problem. Secondly, conventional permeable reactive barriers (PRBs) can only achieve in-situ remediation of groundwater pollution. They are typically buried in underground aquifers, resulting in high construction costs, susceptibility to pore blockage during long-term operation, and difficulty in replacing the reactive medium after deactivation. The existing PRB systems have several drawbacks. First, they have a short service life and lack the ability to protect slopes and prevent geological disasters. Second, current technologies lack integrated engineering structures that can simultaneously address both sudden geological disasters and persistent chemical pollution, making it impossible to achieve coordinated prevention and control of disasters and pollution. Third, the operation and maintenance of existing PRB systems are mostly passive, only addressing issues after media blockage or failure. They lack precise prediction methods for the degradation of media performance. Furthermore, existing machine learning prediction models are mostly data-driven and do not incorporate the physical mechanisms of material degradation, making them prone to overfitting and non-physical predictions. They cannot accurately extrapolate long-term performance using early short-cycle operating data, making it difficult to support intelligent operation and maintenance throughout the system's entire lifecycle. Summary of the Invention

[0004] The purpose of this invention is to overcome the aforementioned defects and shortcomings in the existing technology and provide a permeable reactive barrier dam structure for the prevention and control of acidic heavy metal leaching water in coal gangue dumps, integrating slope stability protection, long-term purification of acidic heavy metal leaching water, and intelligent operation and maintenance throughout the entire life cycle. This invention solves the inherent contradiction between physical barrier and chemical infiltration in traditional structures by completely decoupling the rigid load-bearing structure and the purification functional unit; it addresses the pain points of difficult replacement and high maintenance costs of traditional PRB media through a modular and replaceable filter element design; it achieves accurate prediction of long-term media performance by innovatively proposing a physical constraint neural network model based on material degradation kinetics; and it realizes integrated and coordinated prevention and control of geological disasters and water pollution in coal gangue dumps through a collaborative management closed loop of "replaceable modular filter element + intelligent air-water backwashing + AI full life cycle prediction," significantly improving the system's engineering feasibility, operational stability, and service life.

[0005] To achieve the above objectives, the following technical solution is adopted:

[0006] This invention provides a permeable reactive barrier dam structure for controlling acidic heavy metal leaching water in coal gangue dumps, comprising:

[0007] The main body of the rigid load-bearing dam is used to bear structural loads and intercept solid particulate matter.

[0008] A replaceable reaction filter unit, at least one of which is detachably installed in a modular filter installation chamber, is used to purify the leached water flowing through it; the modular filter installation chamber is built into the rigid load-bearing dam body and located in a non-load-bearing space that does not bear structural loads;

[0009] The sensor system is deployed on the main body of the rigid load-bearing dam to collect operational data in real time;

[0010] The intelligent operation and maintenance platform communicates with the sensor system and is used to predict the performance degradation trend of the replaceable reaction filter unit based on the collected operating data, and generate and output operation and maintenance instructions for the replaceable reaction filter unit according to the prediction results.

[0011] Furthermore, the main body of the rigid load-bearing dam includes:

[0012] The double-layer inlet bar unit, modular filter cartridge installation compartment, drainage guide layer and downstream reinforcement support layer are arranged sequentially along the water flow direction;

[0013] The dual-layer water inlet grille unit is used to intercept solid particles of different sizes.

[0014] The modular filter cartridge installation compartment is an independent compartment formed within a rigid frame, and has a standardized installation slot inside for installing the replaceable reaction filter cartridge unit.

[0015] The drainage and shower layer is equipped with a collection and drainage pipe network for collecting the purified water.

[0016] The downstream reinforcement support layer is rigidly connected to the dam foundation to resist water pressure and impact loads, ensuring the overall structural stability of the dam.

[0017] Furthermore, the rigid load-bearing dam body also includes an overflow emergency flood discharge unit installed on its top. The overflow emergency flood discharge unit includes an overflow channel, a hydraulically operable emergency flood discharge gate, and a liquid level sensor, which is used to automatically open the flood discharge when the backwater in front of the dam exceeds the warning threshold.

[0018] Furthermore, the replaceable reaction filter unit includes a hollow filter housing, the wall of which is a porous, water-permeable structure. The interior of the filter housing is sequentially filled with a first filtration layer, a second reaction layer, and a third neutralization layer, forming a functionally graded composite reaction medium, along the water flow direction.

[0019] The first filter layer is composed of high-strength inert aggregate;

[0020] The second reaction layer is composed of porous ceramsite adsorbent material prepared from modified coal-based solid waste. The porous ceramsite adsorbent material is porous ceramsite loaded with nano-iron sulfide, which is made from fly ash and coal gangue as the main raw materials through alkali activation, foaming and sintering.

[0021] The third neutralizing layer is composed of a mixture of limestone and dolomite alkaline aggregates.

[0022] Furthermore, it also includes an anti-clogging backwashing system;

[0023] The anti-clogging backwashing system includes backwashing pipes and aeration pipes installed at the bottom of the drainage layer or modular filter cartridge installation chamber; the backwashing pipes and aeration pipes are connected to a clean water pump and an air compressor.

[0024] The sensor system also includes a differential pressure monitoring module for monitoring the hydraulic pressure difference between the inlet and outlet of the dam.

[0025] The backwashing system responds to the instructions of the intelligent operation and maintenance platform to clean the replaceable reaction filter unit.

[0026] Furthermore, the intelligent operation and maintenance platform embeds a physical constraint neural network model based on material degradation dynamics;

[0027] The input features of the physical constraint neural network model are the inherent material characteristics and operating environment characteristics of the medium in the replaceable reactive filter unit, and the output is a number of attenuation parameters with clear physical meaning.

[0028] The physical constraint neural network model calculates the predicted value of the effective adsorption capacity by substituting the multiple attenuation parameters into the pre-constructed attenuation equation, and applies a physical constraint penalty term during model training to force the model to output a prediction result that conforms to physical laws.

[0029] Furthermore, the decay equation is a logarithmic decay equation, used to describe the decay law of the effective adsorption capacity of the reaction medium over time, as shown in the following formula:

[0030]

[0031] In the formula: for Effective adsorption capacity of a unit mass of reaction medium at any given time; This represents the initial effective adsorption capacity. This represents the maximum effective adsorption capacity. The continuous operating time of the reaction medium; The characteristic time scale; For shape parameters.

[0032] Furthermore, the physical constraint neural network model adopts an end-to-end architecture, specifically including:

[0033] Input layer: Receives a normalized feature vector of dimension N. The feature vector covers the inherent material characteristics of the reaction medium and the operating environment characteristics, and does not take the running time t as an input feature.

[0034] Multilayer perceptron module: connected to the input layer, used to learn the nonlinear mapping relationship between input features and decay parameters, adopting a 3-layer fully connected hidden layer structure, with each layer set with ReLU nonlinear activation function and Dropout layer;

[0035] Output layer and physical range scaling layer: These layers output the attenuation parameters and scale them to a reasonable range that conforms to physical laws. The output layer has four neurons, each corresponding to one of the attenuation parameters. , , , The scaling layer imposes hard physical range constraints on the output parameters.

[0036] Physical constraint loss module: This module is used to substitute the scaled decay parameters and running time t into the decay equation to calculate the predicted effective adsorption capacity, construct the total loss function by combining the measured values, and add a physical inequality constraint penalty term.

[0037] Furthermore, the physical constraint neural network model is obtained through a two-stage training method:

[0038] In the first stage, a pre-trained dataset containing laboratory accelerated test data and publicly available literature data was used to perform general pre-training on the model in order to learn the general physical laws of media performance degradation.

[0039] In the second stage, for the target coal gangue stockpile project, the early short-cycle operation data collected on-site by the sensor system is used to transfer and fine-tune the pre-trained model in order to achieve accurate prediction of the long-term performance of the medium in the project.

[0040] Furthermore, the intelligent operation and maintenance platform is used to construct a closed-loop collaborative management and control system covering the entire lifecycle of "structural decoupling - performance prediction - targeted maintenance," specifically including:

[0041] Based on the real-time operational data collected by the sensor system, the physical constraint neural network model is used to predict the attenuation trend of the effective adsorption capacity of the reaction medium in the replaceable reaction filter unit and the growth trend of the hydraulic pressure difference of the barrier dam.

[0042] When the predicted hydraulic pressure difference will reach 150% of the design threshold within a set time period, the anti-clogging backwashing system is triggered in advance to clean the corresponding replaceable reaction filter unit.

[0043] When the predicted effective adsorption capacity will drop to less than 30% of the initial effective adsorption capacity, or the hydraulic permeability cannot be restored to 60% of the design value after backwashing, an early warning for the replacement of the corresponding replaceable reaction filter unit will be issued.

[0044] Compared with the prior art, the present invention achieves the following beneficial effects:

[0045] 1. This invention completely decouples the rigid load-bearing structure from the purification functional unit. The rigid concrete frame undertakes all structural load-bearing, impact load resistance, and geological disaster prevention functions, while the filter element unit only undertakes the water purification function. This completely solves the natural contradiction between the rigidity requirement of physical barrier and the permeability requirement of chemical purification in traditional structures. For extreme geological disasters such as landslides and debris flows, the rigid frame can serve as the last line of defense. Even if the filter element is damaged by impact, it can still block solid particles, preventing dam failure and disaster expansion. It takes into account both the continuous purification of leachate under normal working conditions and the structural safety protection under extreme working conditions, achieving coordinated prevention and control of disasters and pollution.

[0046] 2. The core adsorption medium of this invention is prepared using coal-based solid waste such as fly ash and coal gangue as the main raw materials, realizing the high-value resource utilization of mining solid waste. The material cost is reduced by more than 60% compared with commercial adsorption materials, realizing waste treatment with waste, and achieving both environmental and economic benefits. At the same time, the complete preparation process and reasonable performance indicators of modified porous ceramsite are clarified, solving the engineering pain point of easy detachment and agglomeration of nano-iron sulfide particles under high loading, ensuring the batch stability and engineering reproducibility of material performance, and exhibiting excellent adsorption, reduction and removal effects on various heavy metal ions. The material performance is stable, controllable and highly reproducible.

[0047] 3. This invention adopts a guide rail type pull-out prefabricated modular filter element unit, realizing the standardization and factory prefabrication of the purification unit, and facilitating on-site installation. When the medium is saturated and deactivated, it can be quickly replaced by a single person without large-scale excavation. The modular and standardized design makes construction and operation and maintenance convenient and efficient, completely solving the industry pain points of traditional PRB medium replacement difficulties, high construction costs, and long downtime, and significantly reducing the long-term operation and maintenance difficulty and cost of the system.

[0048] 4. This invention is the first to apply a Physically Constrained Neural Network (PENN) model based on material degradation kinetics to PRB media decay prediction and mine wastewater operation and maintenance scenarios. It is also the first to propose a general logarithmic decay equation for adsorption capacity applicable to acidic heavy metal wastewater systems. By embedding physical constraints, it avoids the overfitting and non-physical prediction problems of traditional pure data-driven models from the root. It adopts a two-stage training scheme of "pre-training + transfer fine-tuning", which only requires 200-500 hours of early operation data from the target project site to achieve accurate prediction of long-term media performance for 5,000-10,000 hours. This solves the industry pain point of no long-term field data for new projects. The model's prediction accuracy and generalization ability are significantly better than existing technologies.

[0049] 5. This invention drives a collaborative operation and maintenance closed loop of "replaceable modular filter element + intelligent air-water backwashing system" through the PENN model, realizing a mode upgrade from "passive response operation and maintenance" to "proactive predictive operation and maintenance". By predicting the media decay trend and blockage process in advance, it realizes intelligent optimization of the backwashing procedure and accurate early warning of filter element replacement, which can delay the physical blockage of the media to the maximum extent and accurately control the chemical life of the media. Compared with the traditional fixed cycle operation and maintenance mode, the effective service life of the media is extended by more than 2 times, the overall service life of the dam can reach more than 10 years, and the operation and maintenance cost of the whole life cycle is significantly reduced.

[0050] It should be understood that the description in the Summary of the Invention is not intended to limit the key or essential features of the embodiments of the present invention, nor is it intended to restrict the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0051] The above and other features, advantages, and aspects of the various embodiments of the present invention will become more apparent from the accompanying drawings and the following detailed description. The drawings are provided for a better understanding of the invention and are not intended to limit the invention. In the drawings, the same or similar reference numerals denote the same or similar elements, wherein:

[0052] Figure 1 This is a schematic diagram illustrating the construction process of a permeable reactive barrier dam structure for controlling acidic heavy metal leaching water in coal gangue dumps, according to an embodiment of the present invention.

[0053] Figure 2 This is a schematic diagram of the overall structure of a permeable reactive barrier dam for controlling acidic heavy metal leaching water in coal gangue dumps, according to an embodiment of the present invention.

[0054] Figure 3 This is a schematic diagram of the architecture and training process of the physical constraint neural network model in an embodiment of the present invention.

[0055] In the diagram: 110 - Rigid load-bearing dam body, 111 - Double-layer inlet grid unit, 112 - Modular filter cartridge installation compartment, 113 - Drainage and shower layer, 114 - Downstream reinforcement support layer, 115 - Overflow emergency flood discharge unit, 120 - Replaceable reactive filter cartridge unit, 130 - Anti-clogging backwashing system, 140 - Sensor system, 150 - Smart operation and maintenance platform. Detailed Implementation

[0056] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0057] Furthermore, the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0058] The purpose of this invention is to overcome the shortcomings of existing coal gangue dump leachate control technologies, such as limited functionality, difficult construction and maintenance, separation of pollution and disaster management, and lack of long-term intelligent operation and maintenance methods. This invention provides a permeable reactive barrier dam structure for controlling acidic heavy metal leachate in coal gangue dumps, integrating slope stability protection, long-term purification of acidic heavy metal leachate, and intelligent operation and maintenance throughout the entire lifecycle. This invention is particularly suitable for scenarios where the overall slope of the coal gangue dump is stable and there is no risk of large-scale deep slippage, but there is continuous rainfall-induced leachate pollution, accompanied by small-scale shallow gangue slippage and sediment erosion. For extreme landslides, debris flows, and other geological disasters, the rigid load-bearing frame of this invention can serve as the last line of physical defense to prevent the disaster from escalating. Simultaneously, it achieves integrated and coordinated prevention and control of disaster and pollution, significantly improving system operational stability and industrial feasibility, and reducing the total lifecycle operation and maintenance cost.

[0059] Figure 1 This is a schematic diagram illustrating the construction process of a permeable reactive barrier dam structure for controlling acidic heavy metal leaching water in coal gangue dumps, according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the overall structure of a permeable reactive barrier dam for controlling acidic heavy metal leaching water in coal gangue dumps, according to an embodiment of the present invention. Figure 1 and Figure 2 As shown, the present invention provides a permeable reactive barrier dam structure for controlling acidic heavy metal leaching water in coal gangue dumps, comprising a rigid load-bearing dam body 110, a replaceable reactive filter unit 120, an anti-clogging backwashing system 130, a sensor system 140, and a smart operation and maintenance platform 150. Details are as follows:

[0060] The rigid load-bearing dam body 110 is used to bear structural loads and intercept solid particulate matter.

[0061] Step S1: Construct the main body of the rigid load-bearing dam 110

[0062] The main body of the rigid load-bearing dam 110 is a precast reinforced concrete frame or integral cast gravity rigid load-bearing structure, which bears all lateral loads, impact loads and structural stability protection functions; the main body of the dam is arranged with functional units coaxially along the water flow direction (from the water-facing side to the back water-facing side), and an integrated overflow emergency flood discharge unit is set at the top of the dam.

[0063] Furthermore, the rigid load-bearing dam body 110 includes: a double-layer inlet grille unit 111, a modular filter cartridge installation chamber 112, a drainage and shower layer 113, a downstream reinforcement support layer 114, and an overflow emergency flood discharge unit 115 arranged sequentially along the water flow direction; wherein, the double-layer inlet grille unit 111 is used to intercept solid particles of different sizes; the modular filter cartridge installation chamber 112 is an independent chamber formed within a rigid frame, with standardized installation slots inside for installing the replaceable reactive filter cartridge unit; the drainage and shower layer 113 has a collection and drainage pipe network inside for collecting purified water; the downstream reinforcement support layer 114 is rigidly connected to the dam foundation to resist water pressure and impact loads, ensuring the overall structural stability of the dam body. The overflow emergency flood discharge unit 115 includes an overflow trough, a hydraulically operable emergency flood discharge gate, and a liquid level sensor, used to automatically open the flood discharge when the backwater in front of the dam exceeds the warning threshold. Specific sub-steps are as follows:

[0064] S11: Double-layer inlet bar unit 111

[0065] The double-layer water intake grid unit is located on the upstream water-facing side of the dam body. It is divided into two levels, front and rear, and is detachably connected to the rigid frame of the dam body by bolts. Specifically, it includes: (1) front coarse grid: made of high-strength corrosion-resistant stainless steel with a grid aperture of 15~20mm, used to intercept large-diameter gangue fragments and shallowly sliding block gangue that migrate with the leachate from the coal gangue stockpile; (2) rear fine grid: made of high-strength corrosion-resistant stainless steel with a grid aperture of 5~10mm, used to intercept fine-particle silt and suspended matter to prevent clogging of the downstream filter media pores.

[0066] S12: Modular filter cartridge installation compartment 112

[0067] The modular filter cartridge installation chamber 112 is located downstream of the double-layer inlet bar unit. It is an independent chamber formed by a rigid concrete frame. One or more standardized guide rail installation slots are reserved inside the chamber. The replaceable reaction filter cartridge unit can be pulled out and embedded into the installation slot through the guide rail. The modular filter cartridge installation chamber 112 only provides installation and protection space for the filter cartridge. It does not rely on the filter cartridge medium to bear the structural load-bearing function, completely isolates the structural load-bearing function from the water purification function, and avoids damage to the filter cartridge by impact load.

[0068] S13: Drainage and shower layer 113

[0069] The drainage layer 113 is located downstream of the modular filter cartridge installation compartment. It is equipped with perforated water guide holes and a collection and drainage pipe network to uniformly collect the water after it has been purified by the filter cartridge and stably discharge it to the downstream water body.

[0070] S14: Downstream Reinforcement Support Layer 114

[0071] The downstream reinforcement support layer 114 is located on the downstream backwater side of the dam body. It is integrally cast with C30 or higher strength concrete and is rigidly connected with the rigid frame of the modular filter cartridge installation chamber and the dam foundation to form a complete load-bearing system. It is used to resist upstream water pressure, gangue accumulation pressure and impact load under extreme working conditions, and ensure the overall structural stability of the dam body.

[0072] S15: Overflow Emergency Discharge Unit 115

[0073] The overflow emergency flood discharge unit 115 is installed on the top of the dam body and is integrally cast with the downstream reinforcement support layer 114. It includes an overflow channel and an emergency flood discharge gate that can be hydraulically opened and closed. A liquid level sensor is installed in front of the gate. When the water level in front of the dam exceeds the preset warning threshold (corresponding to rainstorm / debris flow conditions such as grid blockage and insufficient flow capacity), the flood discharge gate will automatically open to quickly discharge high sediment-laden floodwater and prevent the dam body from being breached by backwater. The emergency flood discharge gate is equipped with a trash rack, which can block solid particles under large flow conditions and ensure the ecological safety of the downstream area.

[0074] A replaceable reaction filter unit 120, at least one of which is detachably installed in a modular filter installation chamber, is used to purify the leached water flowing through it; the modular filter installation chamber is built into the rigid load-bearing dam body and is located in a non-load-bearing space that does not bear structural loads;

[0075] Step S2: Preparation and assembly of replaceable reaction filter unit 120

[0076] The replaceable reaction filter unit 120 is a prefabricated standardized modular box made of high-strength corrosion-resistant fiberglass. The box wall has a porous permeable structure with an opening rate of ≥30%. The box is adapted to the mounting groove via guide rails, allowing for quick single-person pull-out replacement. The interior of the box is filled with functional gradient composite reaction media in layers along the water flow direction. The media layers are separated by permeable geotextile to avoid mixing.

[0077] Furthermore, the replaceable reaction filter unit 120 includes a hollow filter housing with a porous, permeable wall. The interior of the filter housing is sequentially filled along the water flow direction with a first filtration layer, a second reaction layer, and a third neutralization layer, forming a functionally gradient composite reaction medium. The first filtration layer is composed of high-strength inert aggregate. The second reaction layer is composed of porous ceramsite adsorbent material prepared from modified coal-based solid waste. This porous ceramsite adsorbent material is porous ceramsite loaded with nano-iron sulfide, which is formed by alkali activation, foaming, and sintering using fly ash and coal gangue as the main raw materials. The third neutralization layer is composed of a mixed alkaline aggregate of limestone and dolomite. Specific sub-steps are as follows:

[0078] S21: Prefabrication of filter cartridge housing

[0079] Based on the dimensions of the modular filter cartridge installation compartment, a standardized prefabricated fiberglass box is constructed. The side walls of the box are evenly permeable with an opening rate of ≥30%. Sliding components that are compatible with the guide rails of the installation groove are installed on both sides of the box. Permeable geotextile pads are installed at the front and rear ends of the box to prevent media loss.

[0080] S22: Functionally graded composite reactive media with layered filling

[0081] S221: First filter layer filling

[0082] The front end of the tank is filled with high-strength inert basalt aggregate with a particle size of 10~20mm and a filling thickness of 100~300mm. This is used to further homogenize the water flow, intercept fine suspended particles that are not intercepted by the inlet bar unit, and protect the downstream functional reaction medium.

[0083] S222: Second reaction layer filling

[0084] A porous ceramsite adsorbent material prepared from modified coal-based solid waste is filled at the rear end of the first filter layer, with a filling thickness of 300~800mm. The porous ceramsite is made from fly ash and coal gangue as the main raw materials, and is prepared by alkali activation, foaming and sintering, and loading with nano-iron sulfide (FeS). The mass ratio of nano-FeS is 5%~15%, and the specific surface area of ​​the ceramsite is ≥50m² / g. It is used to remove various heavy metal ions such as cadmium, arsenic, mercury, lead, and zinc from the leaching water through adsorption, reduction, and precipitation.

[0085] The preparation method of modified coal-based solid waste porous ceramsite is as follows:

[0086] (1) Matrix sintering: using fly ash (60%~70% by mass) and coal gangue (10%~20% by mass) as the main raw materials, adding 5%~10% sodium hydroxide alkaline activator and 3%~5% sodium bicarbonate foaming agent, ball milling to below 200 mesh, dry mixing evenly, adding deionized water to roll into raw material balls with a diameter of 5~10 mm; drying the raw material balls at 105℃ for 2 h, then heating them to 1100~1200℃ in a muffle furnace at a heating rate of 5℃ / min, holding for 2 h, and naturally cooling to room temperature to obtain a porous ceramsite matrix;

[0087] (2) Nano FeS loading: The ceramic matrix was placed in a 0.5 mol / L ferrous sulfate solution and vacuum impregnated for 2 h to allow the solution to fully penetrate into the pores of the ceramic. Then, under a nitrogen protective atmosphere, an equimolar concentration of sodium sulfide solution was added and the reaction was carried out at a constant temperature of 60°C in a water bath for 4 h. After the reaction was completed, the ceramic was washed with oxygen-free deionized water until the filtrate was neutral and then vacuum dried at 60°C for 12 h to obtain modified porous ceramic with nano FeS uniformly loaded on the surface and in the pores.

[0088] S223: Third neutralizing layer filler

[0089] The rear end of the second reaction layer is filled with a mixture of limestone and dolomite aggregate with a particle size of 5-10 mm and a mass ratio of (2-4):1, with a filling thickness of 200-500 mm. This is used to continuously neutralize the acidity of the leaching water, raise the pH value of the water to 6-9, promote the hydrolysis and precipitation of heavy metal ions, and provide the optimal pH environment for the adsorption and reduction reaction of the second reaction layer.

[0090] S23: Filter element assembly and sealing

[0091] After the medium filling is completed, the housing is sealed and encapsulated. The filter element unit is then embedded into the corresponding mounting slot of the modular filter element mounting compartment via guide rails, completing the pipeline connection and sealing test.

[0092] Preferably, in some embodiments of the present invention, the permeable reactive barrier dam structure for controlling acidic heavy metal leaching water in coal gangue dumps of the present invention further includes: an anti-clogging backwashing system 130; the anti-clogging backwashing system 130 includes a backwashing pipe and an aeration pipe arranged at the bottom of the drainage layer or the modular filter cartridge installation compartment; the backwashing pipe and the aeration pipe are connected to a clean water pump and an air compressor; the anti-clogging backwashing system 130 responds to the instructions of the intelligent operation and maintenance platform 150 to clean the replaceable reactive filter cartridge unit.

[0093] Step S3: Install the anti-clogging backwashing system 130

[0094] The anti-clogging backwashing system 130 is used to restore the hydraulic conductivity of the medium, delay the increase in dam pressure differential caused by physical clogging, and extend the effective service life of the medium. However, it cannot reverse the chemical deactivation of the medium caused by adsorption site saturation, nano-FeS oxidation consumption, and reaction product encapsulation. The specific sub-steps are as follows:

[0095] S31: Deploying the hardware system

[0096] Pre-embedded backwashing and aeration pipes are installed at the bottom of the drainage and shower layer and the modular filter cartridge installation compartment, and equipped with a clean water pump, air compressor, differential pressure monitoring module, and automatic control module. Perforated nozzles are evenly distributed on the backwashing and aeration pipes, with each nozzle corresponding to a filter cartridge unit, enabling independent full-section air-water combined backwashing of a single filter cartridge. Differential pressure monitoring modules are respectively installed at the front end of the double-layer water inlet grid unit and the rear end of the drainage and shower layer to monitor the hydraulic pressure difference at the inlet and outlet of the dam in real time.

[0097] S32: Set backflushing operation logic

[0098] Triggering conditions: When the pressure difference between the inlet and outlet exceeds 150% of the design threshold, or when a backwashing command is received from the intelligent operation and maintenance platform, the system will automatically start the backwashing program of the corresponding filter element.

[0099] Backwashing process: The process of "air washing first, air-water combined washing, and then rinsing with clean water" is adopted to break up the caking filter media, remove the trapped suspended solids, and restore the hydraulic permeability of the medium.

[0100] Failure detection: When the hydraulic permeability of the medium cannot be restored to 60% of the design value after backwashing, the system will automatically trigger a filter replacement warning.

[0101] Sensor system 140 is deployed on the main body of the rigid load-bearing dam to collect operational data in real time;

[0102] Step S4: Deploy the sensor system 140

[0103] Water quality sensors (for online monitoring of pH and heavy metal concentration), differential pressure sensors, flow sensors, temperature sensors, and level sensors are installed at various locations on the rigid load-bearing dam body 110 to collect operational data in real time and transmit it to the database of the intelligent operation and maintenance platform via 4G / 5G or wired network. Among them, the differential pressure sensor is used to monitor the hydraulic pressure difference between the dam body inlet and outlet.

[0104] The intelligent operation and maintenance platform 150 is connected to the sensor system 140 to predict the performance degradation trend of the replaceable reaction filter unit 120 based on the collected operating data, and to generate and output operation and maintenance instructions for the replaceable reaction filter unit 120 according to the prediction results.

[0105] Step S5: Build a smart operation and maintenance platform 150

[0106] This invention establishes a smart operation and maintenance platform 150 based on a Physically Enforced Neural Network (PENN) for material degradation kinetics. This invention is the first to apply this model to predict the adsorption capacity decay of permeable reactive media, and the first to propose a universal logarithmic decay equation for adsorption capacity applicable to acidic heavy metal wastewater systems. By embedding physical constraints, it avoids the non-physical prediction results of traditional machine learning models, achieving accurate extrapolation from early short-cycle operational data to long-term performance, providing core support for intelligent operation and maintenance of barrier dams throughout their entire lifecycle. The smart operation and maintenance platform 150 includes a data acquisition module, a data storage module, a PENN prediction module, an intelligent decision-making module, and a visualization interaction module.

[0107] The physical constraint neural network model takes as input the inherent material characteristics and operating environment features of the medium in the replaceable reactive filter unit, and outputs multiple decay parameters with clear physical meaning. The model calculates the predicted effective adsorption capacity by substituting these decay parameters into a pre-constructed decay equation, and applies a physical constraint penalty term during model training to force the model to output predictions that conform to physical laws. These decay parameters include at least the initial effective adsorption capacity, the limiting effective adsorption capacity, the characteristic time scale, and the shape parameter. Specific sub-steps are as follows:

[0108] S51: Construction of the Core Mechanism of the PENN Model

[0109] This invention addresses the decay of the effective adsorption capacity of the reaction medium over time in acidic heavy metal leaching water environments. Based on a multi-process coupled degradation kinetic mechanism, a logarithmic decay equation is constructed as the physical constraint core of the PENN model. This decay equation can uniformly describe the performance degradation trend under different medium formulations and operating conditions:

[0110] ;

[0111] In the formula: : Effective adsorption capacity per unit mass of reaction medium at time t, in units of It is a quantitative characterization of the medium's comprehensive adsorption capacity and acidity neutralization capacity for heavy metal ions in the acidic leachate of coal gangue. The logarithm of the effective adsorption capacity per unit mass of the reaction medium at time t is the core dependent variable of the physical constraint core equation of the Physical Constraint Neural Network (PENN) model of this invention. Initial effective adsorption capacity: This refers to the theoretical maximum effective adsorption capacity per unit mass of the virgin reaction medium under the design conditions, expressed in units of... ; : Logarithm of the initial effective adsorption capacity; Limiting effective adsorption capacity refers to the residual effective adsorption capacity of the medium after long-term operation to reach adsorption saturation and chemical deactivation, measured in units of... ; : The logarithm of the limiting effective adsorption capacity; : Continuous operating time of the reaction medium, in hours; Characteristic timescale, i.e., effective adsorption capacity from decay to and The runtime required for the intermediate value, in hours, is used to quantify the overall decay rate of the medium. Shape parameter, dimensionless, determines the steepness of the adsorption capacity decay curve and can distinguish the decay mode of the medium (gradual decay or steep decay).

[0112] Based on the above equations, the common attenuation master curves under different media and operating conditions can be obtained through normalization. The normalization formula is as follows:

[0113] ;

[0114] The normalized variable is defined as follows:

[0115] ;

[0116] ;

[0117] In the formula: Normalized effective adsorption capacity is the ratio of the difference between the measured effective adsorption capacity and the limit value to the difference between the initial value and the limit value. It is used to eliminate the characteristic differences of different media formulations and different operating conditions. The normalized logarithmic value of the effective adsorption capacity is the core dependent variable of the universal decay master curve of this invention, used to present the general law of medium decay behavior under different media-operating condition combinations. Normalized time variable, which is the ratio of runtime to the characteristic time scale. The power is used to eliminate the difference in decay rate between different media, so as to make the decay curves of different media-operating condition combinations converge to the general master curve.

[0118] After normalization, the attenuation curves of all media-condition combinations can converge to the above general master curve, proving that the equation can capture the general law of media performance attenuation under different systems, and providing a physical basis for the model's cross-material and cross-condition generalization ability.

[0119] S52: PENN Model Architecture Design

[0120] like Figure 3 The diagram shown illustrates the architecture and training process of the Physical Constraint Neural Network (PENN) model in this embodiment of the invention. The PENN model designed in this invention employs an end-to-end architecture of "feature encoding - Multi-Layer Perceptron (MLP) - physical parameter output - physical constraint loss," completely resolving the shortcomings of traditional models that use time as an input feature and fragment the continuous decay process of the same medium. The specific architecture is as follows:

[0121] (1) Input layer: receives normalized feature vectors with dimension N. The feature vectors cover two major categories: the inherent material characteristics of the reaction medium and the operating environment characteristics. The running time t is not used as the input feature to avoid splitting the same medium samples at different times into independent samples, ensuring that the model learns the continuous decay dynamics.

[0122] (2) Multilayer perceptron (MLP) module: It is connected to the input layer and is used to learn the nonlinear mapping relationship between input features and decay parameters. It adopts a 3-layer fully connected hidden layer structure, and each layer is set with ReLU nonlinear activation function and Dropout layer to learn the nonlinear mapping relationship between input features and decay physical parameters.

[0123] (3) Output layer and physical range scaling layer: used to output the attenuation parameter and scale the attenuation parameter to a reasonable range that conforms to physical laws; the output layer is set with 4 neurons, each corresponding to 4 attenuation parameters with clear physical meaning. , , , The scaling layer imposes hard physical range constraints on the output parameters, mapping them to a reasonable range that conforms to engineering realities. , , , This structurally avoids the model outputting non-physical results.

[0124] (4) Physical constraint loss module: Substitute the scaled four decay parameters and the running time t into the core decay equation to calculate the predicted effective adsorption capacity. The loss function is constructed by combining the measured values, and a physical inequality constraint penalty term is added to ensure that the model training process conforms to physical laws.

[0125] S53: Two-Stage Training and Engineering Deployment of the PENN Model

[0126] To address the industry pain point of lacking long-term field data for new construction projects, this invention adopts a two-stage training scheme of "pre-training + transfer fine-tuning" to balance the generalization of the model with its adaptability to the field.

[0127] like Figure 3 As shown, the physical constraint neural network model is trained using a two-stage method: In the first stage, a pre-training dataset containing laboratory accelerated testing data and publicly available literature data is used to perform general pre-training on the model to learn the general physical laws governing the degradation of medium performance. In the second stage, for the target coal gangue dump project, early short-cycle operational data collected on-site by the sensor system is used to perform transfer fine-tuning on the pre-trained model to achieve accurate prediction of the long-term performance of the medium in the project. The specific sub-steps are as follows:

[0128] S531: Phase 1: Construction of General Pre-trained Model

[0129] 1. Construction of pre-trained dataset: The pre-trained dataset covers time series data from multiple sources and scenarios, ensuring data diversity and coverage, specifically including:

[0130] (1) Laboratory test data: Through batch static adsorption experiments and dynamic column breakthrough experiments, data on the change of effective adsorption capacity over time under different media formulations, different influent water quality, different hydraulic loads, different temperatures, and different dry and wet alternation conditions were obtained. The data covered individual and composite systems of modified coal-based solid waste ceramsite, limestone, dolomite and other media, with a total of no less than 6,000 data points and a time span of 0h to 10,000h.

[0131] (2) Public literature data: Time series performance data were extracted from published academic literature on acidic heavy metal wastewater treatment, adsorption material decay, and operation and maintenance of permeable reactive barriers (PRBs) using the WebPlotDigitizer tool. Samples of different materials and different working conditions were added, with a total of no less than 4,000 data points. The overall pre-training dataset has a total of no less than 10,000 data points, covering more than 120 unique media formulation-working condition combinations of complete time series decay curves, covering complex working conditions such as temperature fluctuations, water quality changes, and alternating wet and dry conditions commonly encountered in field engineering.

[0132] 2. Data preprocessing:

[0133] (1) Time axis alignment: The zero point t=0 of the time axis of each sample is defined as the time point when the medium reaches a stable working state (i.e. the time when the effluent water quality first reaches the design discharge standard), eliminating the data deviation caused by the initial activation and wetting effect of the medium, and ensuring that the decay start point of all samples is consistent.

[0134] (2) Outlier removal: The 3σ criterion is used to remove outliers in experimental and monitoring data to avoid noise interference with model training;

[0135] (3) Feature normalization: normalize all continuous input features to the [0,1] interval to ensure the numerical stability of model training.

[0136] (4) Feature Engineering: The input feature vectors are divided into two main categories, comprehensively covering the core factors affecting medium attenuation, specifically including:

[0137] Material inherent characteristics: medium formulation composition (material type and mass ratio of each layer of medium), nano-FeS loading of modified ceramsite, specific surface area, porosity, initial ion exchange capacity (IEC), calcium carbonate content of limestone, magnesium-calcium ratio of dolomite, medium packing density, layer thickness, gradation, etc.

[0138] Environmental operating conditions: influent pH, initial concentration of each heavy metal ion, sulfate concentration, total dissolved solids (TDS), operating ambient temperature, hydraulic retention time (HRT), hydraulic load, flow velocity, rainfall intensity, wet-dry cycle, etc.

[0139] 3. Pre-training and performance verification:

[0140] (1) Data set partitioning: The pre-training dataset is divided into training set, validation set and test set in a ratio of 8:1:1 by using a hierarchical partitioning method based on medium-operating condition combination. The test set consists of a brand new medium formula or operating condition that has not been used in training, which is used to verify the generalization ability of the model. At the same time, 5-fold cross-validation and 5 sets of random seeds are used to ensure the statistical robustness of the model performance.

[0141] (2) Baseline model construction: Gaussian process regression (GPR) and ordinary fully connected neural network (NN) were constructed simultaneously as baseline models. Both GPR and NN used time t as input feature to directly predict the effective adsorption capacity, which was used to compare and verify the performance advantage of the PENN model.

[0142] (3) Hyperparameter optimization: The Optuna Bayesian optimization framework is used to globally optimize the model hyperparameters. The optimization range includes: learning rate (1×10⁻⁶). -4 ~1×10 -3 ), number of neurons in the hidden layer (512~1024 in the first layer, 128~512 in the second layer, 64~128 in the third layer), Dropout rate (0.1~0.5), physical constraint weight ω (0.00~0.50, step size 0.01).

[0143] (4) Training configuration: The Adam optimizer is used, and the learning rate scheduler is set: when the validation loss does not decrease for 100 consecutive rounds, the learning rate is halved, with a minimum lower bound of 1×10. -6 Set an early stopping strategy: terminate training when the validation loss does not decrease for 200 consecutive rounds to prevent the model from overfitting.

[0144] (5) Construction of loss function: total loss function It consists of two parts: the core prediction loss and the physical constraint penalty term, as shown in the following formula:

[0145]

[0146] In the formula: The total loss function is the objective optimization function for training the PENN model in this invention, comprehensively measuring the model's prediction accuracy and its conformity to physical laws; MSE ( Mean Squared Error (MSE) is the mean squared error between the effective adsorption capacity predicted by the model and the measured effective adsorption capacity. It is the core prediction loss term of the loss function and is used to quantify the deviation between the model prediction and the actual value. The model predicts the effective adsorption capacity, in units of... , is the core predicted value output by the PENN model; Measured effective adsorption capacity, in units of It is the true value of the effective adsorption capacity obtained through laboratory / on-site monitoring; Physical constraint weights are dimensionless and their optimal values ​​are determined through Optuna Bayesian optimization. They are used to adjust the weight ratio of physical constraint penalty terms in the total loss function. Physical constraint penalty term, dimensionless, applies a penalty to parameters in the model output that do not conform to physical laws. The meanings of the constraints are as follows:

[0147] Constraint 1: The initial effective adsorption capacity output by the model must be no less than [amount missing]. The measured effective adsorption capacity at any given time avoids non-physical results where the initial capacity prediction is lower than the measured value;

[0148] Constraint 2: The model output should show a limiting effective adsorption capacity that is no greater than the longest running time. The measured effective adsorption capacity at any given time avoids non-physical results where the predicted limiting capacity is higher than the measured value;

[0149] Constraint 3: The model output features must have positive time scales, which aligns with the physical principle that "decay time is positive".

[0150] Constraint 4: The model output shape parameters must be positive to ensure the physical rationality of the adsorption capacity decay curve and avoid the non-physical trend of "negative decay rate".

[0151] (6) Pre-trained model performance: The coefficient of determination (COP) of the trained general pre-trained PENN model on the test set. ≥0.98), significantly better than GPR ( ) and NN The baseline model; at the same time, the normalized decay curves of all test samples can converge to the general master curve, proving that the model successfully captures the general law of medium performance decay.

[0152] S532: Phase Two: Target Project Migration, Fine-tuning, and Implementation

[0153] For specific new retaining dam projects, there is no need to retrain the model. Accurate predictions can be achieved simply by fine-tuning the pre-trained model using a small amount of field data. The specific steps are as follows:

[0154] (1) On-site pilot test data collection: Pilot test was carried out at the target project site, using the same medium formula and actual leachate as the project design, and continuously collected 200-500 hours of operating data, including core data such as influent water quality, effluent water quality, hydraulic pressure difference, and environmental conditions.

[0155] (2) Model fine-tuning: Freeze the MLP backbone network parameters of the pre-trained model and only perform incremental fine-tuning on the parameters of the output layer and scaling layer. The fine-tuning dataset is 200~500h pilot data collected on site to quickly adapt to the on-site working conditions of the target project.

[0156] (3) Predictive Performance Verification: Using a finely tuned Physically Constrained Neural Network (PENN) model, based on early data from the field (200-500 hours), the effective adsorption capacity decay trend of the medium over the next 5000-10000 hours can be accurately predicted. In the field pilot verification of this invention, a continuous flow pilot device was built for the actual leachate (pH=3.2, total heavy metal concentration 12.8 mg / L) of a coal gangue dump in Southwest China. The device operated continuously and stably for a total of 5300 hours. The PENN model was migrated and finely tuned using the field pilot measured data from the first 300 hours to predict the effective adsorption capacity decay trend of the medium over the next 5000 hours. The prediction results were compared and verified with the actual continuous operation monitoring data over the entire 5000-hour cycle. In this verification, the average relative error was defined as the arithmetic mean of the relative errors between the predicted and measured values ​​at all time points in the full prediction time series, and the coefficient of determination was... Table 1 shows the comparison of the core performance indicators of different models to measure the goodness of fit between the predicted and actual monitored values ​​over a 5000-hour full prediction time series. The results show that the average relative error between the PENN model (fine-tuned) and the actual continuous data is as low as 7.2%, far superior to the prediction errors of over 25% for ordinary neural networks (NN) and Gaussian process regression (GPR) models, thus verifying the model's engineering applicability.

[0157] Table 1. Comparison of prediction performance of different models for long-term degradation of the reaction medium.

[0158]

[0159] Note: The values ​​in Table 1 are the measured verification results from a pilot test at a coal gangue stockpile in Southwest China. After 5-fold cross-validation, the PENN model of this invention, after fine-tuning, has a prediction determination coefficient under similar coal gangue leaching water treatment conditions. All values ​​are ≥0.98, and the average relative error is stable at ≤8%.

[0160] The core features and performance differences of each model are explained below: The fine-tuned PENN model of this invention embeds the physical constraint equations of the degradation kinetics of the reaction medium in the coal gangue leaching water environment. Instead of using runtime as a direct input feature, it achieves long-term prediction by learning decay parameters with clear physical meaning. During training, it avoids non-physical prediction results through physical constraint penalty terms, effectively avoiding the overfitting problem common in traditional models and significantly reducing the risk of overfitting. It can accurately capture the inflection point and long-term asymptotic trend of the medium decay curve. Therefore, even with only 300 hours of early data input, it can still achieve high-precision prediction over a long period of 5000 hours. Extrapolation and prediction: Ordinary fully connected neural network models directly use time as input feature to fit time series data without introducing any physical constraints. During training, they are prone to overfitting to early short-period data. When extrapolating to long periods, they are prone to non-physical results such as the adsorption capacity increasing over time, which does not conform to the decay law of the medium. Therefore, the prediction error is large and the long-period generalization ability is poor. The GPR model, which uses a conventional smooth kernel function (such as radial basis function RBF), relies on the kernel function to achieve smooth fitting. It will excessively weaken the abrupt characteristics of the medium decay curve and cannot accurately capture the decay inflection point. The fitting of the long-term decay trend deviates significantly from the actual operating data. Therefore, the long-period prediction accuracy is insufficient.

[0161] S54: Intelligent Operation and Maintenance Platform Achieves Collaborative Management and Control Closed Loop

[0162] This invention forms a collaborative operation and maintenance closed loop through "replaceable modular filter cartridges + intelligent air-water backwashing + PENN-driven full lifecycle prediction." These three elements are not simply a combination of functions, but rather form a comprehensive intelligent management and control system encompassing "prediction-control-replacement," constructing a collaborative lifecycle management and control closed loop of "structural decoupling-performance prediction-targeted maintenance." The specific operating logic is as follows:

[0163] (1) Real-time data acquisition and transmission: The operation data is collected in real time through the 140 sensor systems deployed on the dam body, including water quality sensors (pH, heavy metal concentration online monitoring), differential pressure sensors, flow sensors, temperature sensors and liquid level sensors, and transmitted to the database of the smart operation and maintenance platform via 4G / 5G or wired network.

[0164] (2) Performance degradation and clogging trend prediction: The platform will input the real-time collected working condition data and the inherent material characteristics of the corresponding filter element into the finely adjusted PENN model to predict the degradation trend of the effective adsorption capacity of the medium and the growth trend of the hydraulic pressure difference in real time, and predict the physical clogging process and chemical deactivation time point in advance.

[0165] (3) Intelligent operation and maintenance decision output:

[0166] (a) Intelligent optimization of backwashing: When the model predicts that the pressure difference of the dam body will exceed 150% of the design threshold within the next 7 days, the backwashing procedure is triggered in advance (that is, the anti-clogging backwashing system is triggered in advance to clean the corresponding replaceable reaction filter unit), realizing "predictive cleaning" instead of the traditional "cleaning after clogging", which delays physical clogging to the maximum extent and extends the service life of the medium.

[0167] (b) Precise filter replacement warning: When the model predicts that the effective adsorption capacity of the medium will drop to less than 30% of the initial value, or the hydraulic permeability cannot be restored to 60% of the design value after backwashing, a filter replacement warning will be issued 30 days in advance, and a suitable replacement plan will be generated.

[0168] (c) Emergency control under extreme conditions: In response to extreme rainfall conditions such as rainstorms, the model can quickly predict changes in the liquid level in front of the dam and the trend of medium decay under short-term high hydraulic load impact. When the risk of blockage and backflow is predicted, the emergency flood discharge gate can be opened in advance to ensure the safety of the dam structure and the stability of the effluent water quality.

[0169] In summary, the permeable reactive barrier dam structure for controlling acidic heavy metal leaching water in coal gangue dumps, as described in this embodiment of the invention, completely decouples the rigid load-bearing structure from the purification functional unit. The concrete skeleton bears all structural load-bearing and disaster barrier functions, while the filter element unit only performs water purification, thus completely resolving the inherent contradiction between physical barrier and chemical infiltration in traditional structures. For extreme geological disasters, the rigid skeleton can serve as a last line of defense; even if the filter element is damaged by impact, solid barrier function can still be achieved, preventing dam failure and disaster expansion, thus balancing continuous purification under normal operating conditions with safety protection under extreme conditions. The core reaction medium uses fly ash, coal gangue, and other coal-based solid waste as the main raw materials, achieving high-value resource utilization of solid waste, and significantly reducing material costs compared to commercial adsorption materials. Simultaneously, the complete preparation process and reasonable performance indicators of the modified ceramsite are clearly defined, solving the engineering pain points of nanoparticle detachment and agglomeration under high loads, ensuring batch stability and engineering reproducibility of material performance. Employing a rail-mounted, pull-out, prefabricated modular filter element unit, rapid single-person replacement is possible when the filter medium becomes saturated and deactivated, eliminating the need for large-scale excavation and construction. This completely solves the pain points of difficult and costly replacement of traditional PRB media, significantly reducing long-term operation and maintenance difficulty and downtime. Furthermore, a PENN model based on material degradation kinetics is proposed and applied for the first time to PRB media attenuation prediction and mine wastewater operation and maintenance. A general attenuation equation applicable to acidic heavy metal systems is proposed, fundamentally avoiding overfitting and non-physical prediction results inherent in traditional machine learning models through physical constraints. A two-stage approach of "pre-training + transfer fine-tuning" is adopted, requiring only 200-500 hours of early field data to achieve accurate long-term performance predictions for thousands of hours, addressing the industry pain point of lacking long-term data for new projects. The model's prediction accuracy and generalization ability are significantly superior to existing technologies. By using the PENN model to drive the "replaceable filter + backwashing system" to form a collaborative operation and maintenance closed loop, the operation and maintenance mode has been upgraded from "passive response" to "proactive prediction". This can delay physical blockage of the medium to the maximum extent, accurately control the chemical life of the medium, and extend the effective service life of the medium by more than 2 times compared with the traditional fixed cycle operation and maintenance mode. The overall service life of the dam can reach more than 10 years, and the operation and maintenance cost of the whole life cycle is significantly reduced.

[0170] It should also be noted that, in the embodiments of this application, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0171] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined in the embodiments of this application may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown in this application, but is to be accorded the widest scope consistent with the principles and novel features disclosed in the embodiments of this application.

Claims

1. A permeable reactive barrier dam structure for controlling acidic heavy metal leaching water in coal gangue dumps, characterized in that, include: The main body of the rigid load-bearing dam is used to bear structural loads and intercept solid particulate matter. A replaceable reaction filter unit, at least one of which is detachably installed in a modular filter installation chamber, is used to purify the leached water flowing through it; the modular filter installation chamber is built into the rigid load-bearing dam body and located in a non-load-bearing space that does not bear structural loads; The sensor system is deployed on the main body of the rigid load-bearing dam to collect operational data in real time; The intelligent operation and maintenance platform communicates with the sensor system and is used to predict the performance degradation trend of the replaceable reaction filter unit based on the collected operating data, and generate and output operation and maintenance instructions for the replaceable reaction filter unit according to the prediction results.

2. The permeable reactive barrier dam structure for controlling acidic heavy metal leaching water in coal gangue dumps according to claim 1, characterized in that, The main body of the rigid load-bearing dam includes: The double-layer inlet bar unit, modular filter cartridge installation compartment, drainage guide layer and downstream reinforcement support layer are arranged sequentially along the water flow direction; The dual-layer water inlet grille unit is used to intercept solid particles of different sizes. The modular filter cartridge installation compartment is an independent compartment formed within a rigid frame, and has a standardized installation slot inside for installing the replaceable reaction filter cartridge unit. The drainage and shower layer is equipped with a collection and drainage pipe network for collecting the purified water. The downstream reinforcement support layer is rigidly connected to the dam foundation to resist water pressure and impact loads, ensuring the overall structural stability of the dam.

3. The permeable reactive barrier dam structure for controlling acidic heavy metal leaching water in coal gangue dumps according to claim 2, characterized in that, The rigid load-bearing dam body also includes an overflow emergency flood discharge unit installed on its top. The overflow emergency flood discharge unit includes an overflow channel, a hydraulically operable emergency flood discharge gate, and a liquid level sensor, which is used to automatically open the flood discharge when the backwater in front of the dam exceeds the warning threshold.

4. The permeable reactive barrier dam structure for controlling acidic heavy metal leaching water in coal gangue dumps according to claim 1, characterized in that, The replaceable reaction filter unit includes a hollow filter housing with a porous, permeable wall. The interior of the filter housing is sequentially filled with a first filtration layer, a second reaction layer, and a third neutralization layer along the water flow direction to form a functionally graded composite reaction medium. The first filter layer is composed of high-strength inert aggregate; The second reaction layer is composed of porous ceramsite adsorbent material prepared from modified coal-based solid waste. The porous ceramsite adsorbent material is porous ceramsite loaded with nano-iron sulfide, which is made from fly ash and coal gangue as the main raw materials through alkali activation, foaming and sintering. The third neutralizing layer is composed of a mixture of limestone and dolomite alkaline aggregate.

5. The permeable reactive barrier dam structure for controlling acidic heavy metal leaching water in coal gangue dumps according to claim 1, characterized in that, It also includes an anti-clogging backwashing system; The anti-clogging backwashing system includes backwashing pipes and aeration pipes installed at the bottom of the drainage layer or modular filter cartridge installation chamber; the backwashing pipes and aeration pipes are connected to a clean water pump and an air compressor. The sensor system also includes a differential pressure monitoring module for monitoring the hydraulic pressure difference between the inlet and outlet of the dam. The backwashing system responds to the instructions of the intelligent operation and maintenance platform to clean the replaceable reaction filter unit.

6. The permeable reactive barrier dam structure for controlling acidic heavy metal leaching water in coal gangue dumps according to claim 5, characterized in that, The intelligent operation and maintenance platform has an embedded physical constraint neural network model based on material degradation dynamics. The input features of the physical constraint neural network model are the inherent material characteristics and operating environment characteristics of the medium in the replaceable reactive filter unit, and the output is a number of attenuation parameters with clear physical meaning. The physical constraint neural network model calculates the predicted value of the effective adsorption capacity by substituting the multiple attenuation parameters into the pre-constructed attenuation equation, and applies a physical constraint penalty term during model training to force the model to output a prediction result that conforms to physical laws.

7. The permeable reactive barrier dam structure for controlling acidic heavy metal leaching water in coal gangue dumps according to claim 6, characterized in that, The decay equation is a logarithmic decay equation, used to describe the decay of the effective adsorption capacity of the reaction medium over time, and the formula is as follows: ； In the formula: for Effective adsorption capacity of a unit mass of reaction medium at any given time; This represents the initial effective adsorption capacity. This represents the maximum effective adsorption capacity. The continuous operating time of the reaction medium; The characteristic time scale; For shape parameters.

8. The permeable reactive barrier dam structure for controlling acidic heavy metal leaching water in coal gangue dumps according to claim 7, characterized in that, The physical constraint neural network model adopts an end-to-end architecture, specifically including: Input layer: Receives a normalized feature vector of dimension N. The feature vector covers the inherent material characteristics of the reaction medium and the operating environment characteristics, and does not take the running time t as an input feature. Multilayer perceptron module: connected to the input layer, used to learn the nonlinear mapping relationship between input features and decay parameters, adopting a 3-layer fully connected hidden layer structure, with each layer set with ReLU nonlinear activation function and Dropout layer; Output layer and physical range scaling layer: These layers output the attenuation parameters and scale them to a reasonable range that conforms to physical laws. The output layer has four neurons, each corresponding to one of the attenuation parameters. , , , The scaling layer imposes hard physical range constraints on the output parameters. Physical constraint loss module: This module is used to substitute the scaled decay parameters and running time t into the decay equation to calculate the predicted effective adsorption capacity, construct the total loss function by combining the measured values, and add a physical inequality constraint penalty term.

9. The permeable reactive barrier dam structure for controlling acidic heavy metal leaching water in coal gangue dumps according to claim 6, characterized in that, The physical constraint neural network model is obtained through a two-stage training method: In the first stage, a pre-trained dataset containing laboratory accelerated test data and publicly available literature data was used to perform general pre-training on the model in order to learn the general physical laws of media performance degradation. In the second stage, for the target coal gangue stockpile project, the early short-cycle operation data collected on-site by the sensor system is used to transfer and fine-tune the pre-trained model in order to achieve accurate prediction of the long-term performance of the medium in the project.

10. The permeable reactive barrier dam structure for controlling acidic heavy metal leaching water in coal gangue dumps according to claim 6, characterized in that, The intelligent operation and maintenance platform is used to construct a closed-loop collaborative management and control system covering the entire lifecycle of "structural decoupling - performance prediction - targeted maintenance", specifically including: Based on the real-time operational data collected by the sensor system, the physical constraint neural network model is used to predict the attenuation trend of the effective adsorption capacity of the reaction medium in the replaceable reaction filter unit and the growth trend of the hydraulic pressure difference of the barrier dam. When the predicted hydraulic pressure difference will reach 150% of the design threshold within a set time period, the anti-clogging backwashing system is triggered in advance to clean the corresponding replaceable reaction filter unit. When the predicted effective adsorption capacity will drop to less than 30% of the initial effective adsorption capacity, or the hydraulic permeability cannot be restored to 60% of the design value after backwashing, an early warning for the replacement of the corresponding replaceable reaction filter unit will be issued.

Images