Intelligent ecological blind ditch for underground water seepage of water-rich fill slope and control method

Abstract

The invention discloses an intelligent ecological blind ditch for underground water seepage of a water-rich fill slope and a control method, and the intelligent ecological blind ditch comprises a blind ditch, a water circulation module, a control valve module, a seepage monitoring module and an intelligent decision-making module, the control method comprises the steps of slope design parameter extraction, multi-dimensional real-time parameter collection, monitoring data preprocessing, prediction analysis based on a physical constraint neural network, control instruction generation and execution and system self-optimization. Key parameters in a slope body are captured in real time through the seepage monitoring module, weather forecast information is combined, the seepage change trend and the slope stability are pre-judged in advance by means of a neural network model with physical constraints, meanwhile, underground water is directionally recycled for vegetation maintenance, and unification of slope engineering safety and ecological health is achieved.

Description

Technical Field

[0001] This invention belongs to the field of slope engineering technology and relates to an intelligent ecological blind ditch and control method for groundwater seepage in water-rich fill slopes. Background Technology

[0002] Embankment slopes, especially those in water-rich environments, present key challenges in geotechnical and environmental engineering for long-term stability and ecological maintenance. Groundwater seepage is one of the core factors affecting slope stability. During periods of heavy rainfall, the groundwater level rises rapidly, increasing pore water pressure and significantly reducing soil shear strength, making landslides and instability highly likely. Conversely, during periods of drought, low soil moisture content leads to vegetation degradation and surface soil cracking, ultimately impacting the long-term durability of the slope.

[0003] Currently, the treatment of groundwater seepage on slopes largely relies on blind ditch drainage systems. Traditional blind ditches are typically constructed using permeable materials such as gravel and permeable pipes, making them passive drainage facilities. Their drainage process depends entirely on the hydraulic gradient, failing to dynamically adjust to environmental changes or effectively utilize water resources. Traditional blind ditch drainage systems have the following limitations: 1) Lack of regulation capacity: Traditional blind ditches directly discharge groundwater into the slope, failing to collect and reuse water resources, resulting in waste; 2) Delayed response and passivity: The system's drainage function is fixed after construction, only responding after the water level rises. It cannot anticipate and proactively control based on weather forecasts or internal conditions, making it poorly able to cope with sudden heavy rainfall events; 3) Conflict between safety and ecology: Traditional designs prioritize rapid drainage and safety, but continuous drainage during the dry season leads to excessively low soil moisture content, affecting vegetation health, weakening root soil stabilization, and exacerbating shallow erosion, ultimately impacting overall stability in the long run.

[0004] In recent years, with the development of sensing technology, the Internet of Things, and artificial intelligence, slope monitoring methods have become increasingly advanced, and numerous automated monitoring solutions have emerged. However, most of these systems have failed to achieve closed-loop management from data acquisition to control execution, that is, they lack the intelligent decision-making and execution capabilities to transform real-time monitoring and forecasting information into regulatory actions.

[0005] Therefore, there is an urgent need for an intelligent blind drain system that can balance safety and ecological needs and has the capabilities of sensing, prediction, regulation and self-optimization, so as to achieve precise control of groundwater seepage on water-rich fill slopes and rational utilization of water resources. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention proposes an intelligent ecological blind ditch and control method for groundwater seepage in water-rich embankment slopes, aiming to solve the technical problems of uncontrollable drainage, delayed response, and lack of coordination with ecological needs in traditional blind ditches.

[0007] The intelligent ecological blind ditch for groundwater seepage in water-rich embankment slopes includes a blind ditch, a water circulation module, a control valve module, a seepage monitoring module, and an intelligent decision-making module. The blind ditch is constructed of geotextile, gravel filling material and permeable pipes, forming a continuous main drainage channel inside; the side walls and bottom of the blind ditch are provided with water inlet channels to collect seepage water from inside the slope; The water circulation module includes a drainage pipe, a water storage well, and a water supply pipe. All components are sealed together via standardized flange interfaces or pipe fittings, forming a water circulation pathway that can be dynamically switched according to the slope seepage status and ecological needs. The drainage pipe connects the blind ditch to the natural water body, discharging water from the blind ditch into the natural water body. The water storage well connects to the blind ditch, temporarily storing water from the blind ditch. The water supply pipe connects to the water storage well, directionally transporting the groundwater stored in the water storage well to the root zone of the slope vegetation. The control valve module includes a first electromagnetic flow valve and a second electromagnetic flow valve, which control the direction and flow rate of water through the opening and closing of the valves; wherein, the first electromagnetic flow valve is installed at the connection between the blind ditch and the drain pipe and the water storage well; the second electromagnetic flow valve is installed at the connection between the water storage well and the water supply pipe; The seepage monitoring module consists of multiple types of sensors deployed inside the slope to collect key seepage parameters in real time. The seepage monitoring module includes a head sensor for monitoring the water head inside the slope soil, a flow sensor for measuring the real-time flow in the blind ditch, and a soil moisture sensor for monitoring the moisture content of the slope soil. The core of the intelligent decision-making module is a neural network model based on physical constraints. It predicts and makes intelligent decisions on the seepage state of the slope based on the slope design parameters and real-time parameters, and generates control commands for the first and second electromagnetic flow valves. The control strategy takes into account both slope stability and ecological water conservation requirements throughout the process.

[0008] Preferably, the neural network model adopts a hierarchical design, including an input layer, a feature extraction layer, a physical constraint layer, and an output layer connected in sequence; each layer achieves functional connection through parameter passing, which not only ensures the independence of the technical features of each layer, but also improves the model's prediction accuracy and decision reliability for slope seepage state through synergistic effect; The input layer adopts a fully connected structure, with the number of neurons matching the dimension of the input vector, and is responsible for mapping the input data to a high-dimensional feature space. The feature extraction layer consists of multiple hidden layers, which are used to automatically learn and fuse the complex nonlinear relationship between the slope seepage process and stability state in the spatiotemporal dimension contained in the input data to extract deep features. The physical constraint layer, located between the feature extraction layer and the output layer, introduces physical constraints based on seepage mechanics and slope stability principles during model training. By embedding physical laws into the loss function as constraint terms, it forces the model output to conform to the physical mechanism, thereby enhancing the model's generalization ability and prediction reliability. The physical constraint terms include seepage field control equation constraints and static equilibrium equation constraints. The seepage field control equation constraints are shown in the following equation:

[0009] in, h The total head is in meters (m). K ( h ) represents the unsaturated hydraulic conductivity, in m / s; S represents the source-sink term, reflecting rainfall infiltration or evaporation, in s. -1 ; i ( h () represents the volumetric water content; t For time, s; The constraints of the static equilibrium equations are shown below:

[0010] That is, the resultant force of all forces F The resultant torque of all torques is 0. M =0; The output layer is configured to generate the final prediction result and control parameters based on the fusion features extracted by the feature extraction layer and the requirements of the physical constraint layer; the output layer uses a linear activation function, and the output values ​​of its neurons must satisfy the boundary conditions defined by the physical constraint layer. The model's total loss function L total Data loss L data Constraint loss of the control equation of the seepage field L influent and static equilibrium equation constraint loss L static It consists of, that is, satisfying the following expression:

[0011] in, L data The mean squared error between the model's predicted values ​​and the actual monitored values ​​is used to ensure the model's ability to fit the monitoring data. L influent and L static These are all physical constraint error terms, used to ensure that the model output conforms to physical common sense; m 1 , m 2 and m3 All are weighting coefficients, and satisfy the following conditions: m 1 + m 2 + m 3 =1, the specific value of which is determined through cross-validation; The training sample set consists of historical engineering measured data and virtual samples based on physical mechanism simulation. The virtual samples are generated by numerical simulation methods and can cover a variety of extreme working conditions, significantly enhancing the model's generalization ability and robustness when measured data is insufficient or has not experienced extreme conditions.

[0012] A method for controlling groundwater seepage in water-rich fill slopes using intelligent ecological blind drains includes the following steps: S1. Extraction of slope design parameters: Slope geometric parameters and slope soil physical and mechanical parameters are extracted from design data for model building and predictive analysis. The slope geometric parameters include slope height, slope ratio, slope orientation, and fill layer thickness. The slope soil physical and mechanical parameters include cohesion. c s internal friction angle f s Compression modulus E s and permeability coefficient k s ; S2. Real-time acquisition of multi-dimensional parameters: Real-time seepage parameters of the slope are collected by seepage monitoring modules deployed inside the blind drain and in the surrounding soil, including water head data within the slope soil. H ( x , y , z , t Soil moisture content N ( x , y , z , t ) and real-time flow rate in blind drains Q ( t Simultaneously, rainfall forecast information for the slope location is obtained from the local meteorological information platform through the data interface, including short-term and medium-term rainfall forecast information, including rainfall probability, peak hourly rainfall intensity, and cumulative rainfall; in addition, vegetation information on the slope surface is obtained through high-resolution satellite remote sensing image interpretation, including vegetation distribution density, vegetation type, and growth status information. S3. Preprocessing of monitoring data: Based on the slope geometric parameters extracted in step S1, a three-dimensional spatiotemporal grid model matching the actual spatial structure of the slope is constructed; the slope soil physical and mechanical parameters, the real-time slope seepage parameters, the rainfall forecast information and the vegetation distribution information are denoised and normalized, and combined on a unified spatiotemporal grid model to construct a spatiotemporally aligned multidimensional feature vector, providing good input features for the neural network model; Preferably, the noise reduction and normalization processing in step S3 specifically includes the following steps: S301, Data Cleaning: Wavelet denoising algorithm is used to filter out high-frequency interference noise in the parameter data, according to 3 s The criteria identify outliers and fill them with linear interpolation based on the parameter change trends of adjacent time points and adjacent spatial nodes; S302. Data Normalization: The cleaned parameters are mapped to a unified [0,1] interval to eliminate scale differences caused by different physical units, ensuring that the weights of each parameter are comparable in the neural network model; the normalized parameter values... X norm The calculation expression is:

[0013] in, X These are the parameter values ​​after cleaning. X max and X min These are the global maximum and minimum values ​​of the parameter recorded in the historical engineering database and the current monitoring period, respectively. S303, Spatiotemporal Alignment: The cleaned and normalized parameters are aligned according to their corresponding spatiotemporal coordinates. x , y , z , t The model is registered and synchronized on a three-dimensional spatiotemporal grid model, and integrated to form a spatiotemporally aligned multidimensional feature vector, which is adapted to the input requirements of the neural network model.

[0014] S4. Predictive analysis based on physical constraint neural networks: The multidimensional feature vector constructed in step S3 is input into the trained physical constraint-based neural network model, and the prediction results are output through model calculation. The prediction results include the water head value and its rate of change, the soil moisture content and its rate of change, the flow rate change curve and peak flow rate in the blind ditch, and the slope stability coefficient. S5. Generation and execution of control commands: Based on the multi-scenario prediction results obtained in step S4, the intelligent decision-making module combines the preset slope stability safety threshold and ecological water demand threshold, and uses a multi-objective optimization algorithm to generate control commands in real time, dynamically adjusting the opening and flow direction of the first electromagnetic flow valve and the second electromagnetic flow valve. Based on the predicted rainfall scenario, the control mode adaptively switches between the heavy rainfall period control mode, the rainfall interval period control mode, and the drought period control mode. The heavy rainfall control mode aims to rapidly reduce pore water pressure and ensure slope stability. The intelligent decision-making module calculates and outputs instructions to make the first electromagnetic flow valve open to the maximum extent to discharge the water in the blind ditch directly to the natural water body through the drainage pipe; at the same time, the second electromagnetic flow valve is kept closed to stop ecological water replenishment and fully ensure drainage efficiency. The rainfall interval control mode aims to regulate water resources and maintain stable water levels. The intelligent decision-making module controls the first electromagnetic flow valve to switch the water flow to the water storage well and dynamically adjusts the inflow rate to maintain the water level in the water storage well at 80% to 90% of its rated capacity. If the water storage well is full, it automatically switches to drainage mode. At the same time, based on feedback from the soil moisture sensor, if the moisture content of the root layer of the slope vegetation is lower than the ecological threshold, the second electromagnetic flow valve is controlled to open at an appropriate time to replenish water locally. The drought control mode takes ecological water conservation and maintaining vegetation health as its core objectives. The intelligent decision-making module controls the first electromagnetic flow valve to divert all the water flow into the storage well for storage. The second electromagnetic flow valve is precisely opened according to the spatial distribution data of soil moisture, and prioritizes compensatory irrigation for water-deficient areas in the early morning or evening to keep the soil moisture content in the root zone within a suitable range. The intelligent decision-making module sends valve opening degree, flow direction and control timing commands to the field control unit through a wireless communication network, driving the actuator to act, and realizing intelligent and adaptive seepage control; S6. System self-optimization: The environmental data, decision instructions, and execution results in each control loop are integrated into structured cases and stored in a historical database to expand the training sample set. The newly added samples are used periodically to incrementally train the physical constraint neural network, optimize the model parameters, and enhance its prediction accuracy and decision reliability under different climatic conditions and geological environments, so as to achieve continuous evolution of system performance.

[0015] In summary, compared with existing technologies, the beneficial effects of this invention are as follows: To achieve precise control and rational utilization of water resources for groundwater seepage in water-rich fill slopes, an intelligent ecological blind ditch and control method for groundwater seepage in water-rich fill slopes are proposed. This intelligent ecological blind ditch includes a blind ditch, a water circulation module, a control valve module, a seepage monitoring module, and an intelligent decision-making module. The control method includes the extraction of slope design parameters, the acquisition of multi-dimensional real-time parameters, the preprocessing of monitoring data, predictive analysis based on physical constraint neural networks, the generation and execution of control commands, and system self-optimization, achieving the following breakthrough improvements: 1) Real-time perception and intelligent control of seepage status: The seepage monitoring module captures key parameters such as water head, soil moisture content and blind ditch flow rate in real time. Combined with meteorological forecast information, the seepage change trend and slope stability are predicted in advance by using a neural network model with physical constraints. In the face of extreme scenarios such as heavy rainfall, the intelligent decision module can quickly drive the first electromagnetic flow valve to operate at maximum opening, and quickly discharge the water in the blind ditch to natural water bodies through the drainage pipe. This effectively reduces the pore water pressure of the slope and avoids the risk of landslide caused by a sudden rise in groundwater level from the root. It completely solves the problem of the lag of traditional blind ditches and provides active protection for slope safety. 2) Balancing slope stability and ecological sustainability: By temporarily storing seepage water in wells during rainfall intervals and precisely replenishing the root zone of vegetation based on soil moisture feedback, and by storing water during drought periods and providing compensatory irrigation according to region and time period, the soil moisture content is kept within an appropriate range. This mechanism not only prevents vegetation degradation and topsoil cracking caused by excessive drainage during the dry season, but also enhances the slope's resistance to erosion by maintaining vegetation health, thereby achieving a balance between slope engineering safety and ecological health. 3) Improve the utilization efficiency of groundwater resources: By utilizing the temporary storage function of water storage wells for seepage water and combining it with the precise control of the second electromagnetic flow valve, groundwater can be reused in a targeted manner for vegetation maintenance, effectively improving the utilization rate of water resources. This not only reduces unnecessary water discharge but also provides stable water source support for slope ecological restoration, making it particularly suitable for slope treatment projects in arid or water-scarce areas. 4) Achieve adaptive system management: By integrating the seepage monitoring module, intelligent decision-making module, and control valve module, a complete intelligent control closed loop is formed; the monitoring module provides real-time data input, the physical constraint neural network model ensures prediction accuracy and decision reliability, the control module dynamically executes valve control commands, and the case data of each control cycle is stored in the database for incremental model training, thereby achieving intelligent and adaptive seepage management. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the intelligent ecological blind ditch for groundwater seepage in water-rich fill slopes as described in an embodiment of the present invention; Figure 2 This is a schematic diagram of the blind drain structure described in an embodiment of the present invention; Figure 3 This is a flowchart of the intelligent ecological blind ditch control method for groundwater seepage in water-rich fill slopes, as described in an embodiment of the present invention. Among them, 1-blind drain, 11-geotextile, 12-gravel filler, 13-permeable pipe, 21-drainage pipe, 22-water storage well, 23-water supply pipe, 31-first electromagnetic flow valve, 32-second electromagnetic flow valve. Detailed Implementation

[0017] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments given herein are for illustration and explanation only and are not intended to limit the present invention.

[0018] The first aspect of this application discloses as follows: Figure 1-2 The intelligent ecological blind ditch shown is for groundwater seepage in water-rich fill slopes, including blind ditch 1, water circulation module, control valve module, seepage monitoring module and intelligent decision-making module; The blind ditch 1 is constructed of geotextile 11, crushed stone filling material 12 and permeable pipe 13, forming a continuous main drainage channel inside; the side walls and bottom of the blind ditch 1 are provided with water inlet channels for collecting seepage water inside the slope. The water circulation module includes a drainage pipe 21, a water storage well 22, and a water supply pipe 23. Each component is sealed and connected through standardized flange interfaces or pipe connectors to form a water circulation path that can be dynamically switched according to the slope seepage status and ecological needs. The drainage pipe connects the blind ditch and the natural water body to discharge the water from the blind ditch into the natural water body. The water storage well connects to the blind ditch to temporarily store the water from the blind ditch. The water supply pipe connects to the water storage well to directionally transport the groundwater stored in the water storage well to the root system layer of the slope vegetation. The control valve module includes a first electromagnetic flow valve 31 and a second electromagnetic flow valve 32, which control the direction and flow rate of water through the opening and closing of the valves; wherein, the first electromagnetic flow valve 31 is disposed at the connection between the blind ditch 1 and the drain pipe 21 and the water storage well 22; the second electromagnetic flow valve 32 is disposed at the connection between the water storage well 22 and the water supply pipe 23; The seepage monitoring module consists of multiple types of sensors deployed inside the slope to collect key seepage parameters in real time. The seepage monitoring module includes a head sensor for monitoring the water head inside the slope soil, a flow sensor for measuring the real-time flow in the blind ditch, and a soil moisture sensor for monitoring the moisture content of the slope soil. The core of the intelligent decision-making module is a neural network model based on physical constraints. It predicts and makes intelligent decisions on the seepage state of the slope based on the slope design parameters and real-time parameters, and generates control commands for the first and second electromagnetic flow valves. The control strategy takes into account both slope stability and ecological water conservation requirements throughout the process.

[0019] In specific implementation, the neural network model adopts a hierarchical design, including an input layer, a feature extraction layer, a physical constraint layer, and an output layer connected in sequence; each layer achieves functional connection through parameter passing, which not only ensures the independence of the technical features of each layer, but also improves the model's prediction accuracy and decision reliability for slope seepage state through synergistic effect. The input layer adopts a fully connected structure, with the number of neurons matching the dimension of the input vector, and is responsible for mapping the input data to a high-dimensional feature space. The feature extraction layer consists of multiple hidden layers, which are used to automatically learn and fuse the complex nonlinear relationship between the slope seepage process and stability state in the spatiotemporal dimension contained in the input data to extract deep features. The physical constraint layer, located between the feature extraction layer and the output layer, introduces physical constraints based on seepage mechanics and slope stability principles during model training. By embedding physical laws into the loss function as constraint terms, it forces the model output to conform to the physical mechanism, thereby enhancing the model's generalization ability and prediction reliability. The physical constraint terms include seepage field control equation constraints and static equilibrium equation constraints. The seepage field control equation constraints are shown in the following equation: (1) in, h The total head is in meters (m). K ( h ) represents the unsaturated hydraulic conductivity, in m / s; S represents the source-sink term, reflecting rainfall infiltration or evaporation, in s. -1 ; i ( h () represents the volumetric water content; t For time, s; The constraints of the static equilibrium equations are shown below: (2) That is, the resultant force of all forces F The resultant torque of all torques is 0. M =0; The output layer is configured to generate the final prediction result and control parameters based on the fusion features extracted by the feature extraction layer and the requirements of the physical constraint layer; the output layer uses a linear activation function, and the output values ​​of its neurons must satisfy the boundary conditions defined by the physical constraint layer. The model's total loss function L total Data lossL data Constraint loss of the control equation of the seepage field L influent and static equilibrium equation constraint loss L static It consists of, that is, satisfying the following expression: (3) in, L data The mean squared error between the model's predicted values ​​and the actual monitored values ​​is used to ensure the model's ability to fit the monitoring data. L influent and L static These are all physical constraint error terms, used to ensure that the model output conforms to physical common sense; m 1 , m 2 and m 3 All are weighting coefficients, and satisfy the following conditions: m 1 + m 2 + m 3 =1, the specific value of which is determined through cross-validation; The training sample set consists of historical engineering measured data and virtual samples based on physical mechanism simulation. The virtual samples are generated by numerical simulation methods and can cover a variety of extreme working conditions, significantly enhancing the model's generalization ability and robustness when measured data is insufficient or has not experienced extreme conditions.

[0020] A method for controlling groundwater seepage in water-rich fill slopes using intelligent ecological blind drains includes the following steps: S1. Extraction of Slope Design Parameters: Slope geometric parameters and slope soil physical and mechanical parameters are extracted from design data for model building and predictive analysis. The slope geometric parameters include slope height, slope ratio, slope orientation, and fill layer thickness. The slope soil physical and mechanical parameters include cohesion. c s internal friction angle f s Compression modulus E s and permeability coefficient k s ; S2. Real-time acquisition of multi-dimensional parameters: Real-time seepage parameters of the slope are acquired by seepage monitoring modules deployed inside the blind ditch and in the surrounding soil, including water head data inside the slope soil. H ( x , y ,z , t Soil moisture content N ( x , y , z , t ) and real-time flow rate in blind drains Q ( t Simultaneously, rainfall forecast information for the slope location is obtained from the local meteorological information platform through the data interface, including short-term and medium-term rainfall forecast information, including rainfall probability, peak hourly rainfall intensity, and cumulative rainfall; in addition, vegetation information on the slope surface is obtained through high-resolution satellite remote sensing image interpretation, including vegetation distribution density, vegetation type, and growth status information. S3. Preprocessing of monitoring data: Based on the slope geometric parameters extracted in step S1, a three-dimensional spatiotemporal grid model matching the actual spatial structure of the slope is constructed; the slope soil physical and mechanical parameters, the real-time slope seepage parameters, the rainfall forecast information and the vegetation distribution information are denoised and normalized, and combined on a unified spatiotemporal grid model to construct a spatiotemporally aligned multidimensional feature vector, providing good input features for the neural network model; The specific implementation includes the following steps: S301, Data Cleaning: Wavelet denoising algorithm is used to filter out high-frequency interference noise in the parameter data, according to 3 s The criteria identify outliers and fill them with linear interpolation based on the parameter change trends of adjacent time points and adjacent spatial nodes; S302. Data Normalization: The cleaned parameters are mapped to a unified [0,1] interval to eliminate scale differences caused by different physical units, ensuring that the weights of each parameter are comparable in the neural network model; the normalized parameter values... X norm The calculation expression is: (4) in, X These are the parameter values ​​after cleaning. X max and X min These are the global maximum and minimum values ​​of the parameter recorded in the historical engineering database and the current monitoring period, respectively. S303, Spatiotemporal Alignment: The cleaned and normalized parameters are aligned according to their corresponding spatiotemporal coordinates. x , y , z , t The model is registered and synchronized on a three-dimensional spatiotemporal grid model, and integrated to form a spatiotemporally aligned multidimensional feature vector, which is adapted to the input requirements of the neural network model.

[0021] S4. Predictive analysis based on physical constraint neural network: Input the multidimensional feature vector constructed in step S3 into the trained physical constraint based neural network model, and output the prediction results through model calculation; the prediction results include water head value and its rate of change, soil moisture content and its rate of change, flow rate change curve and peak flow rate in blind ditch, and slope stability coefficient. In practice, the water head at the foot of the slope is predicted. H The depth is 4.2m, and the water content of the shallow layer of the slope is... N The flow rate was 22%, close to the ecological threshold of 20%; the flow sensor at the blind drain inlet monitored the real-time flow rate. Q ( t ) is 0.8m 3 / h; The instantaneous stability coefficient of the slope calculated using the simplified Bishop method will temporarily decrease from 1.35 to 1.05 during the peak rainfall period, which is very close to the safety threshold of 1.0, indicating a potential risk of instability.

[0022] S5. Generation and execution of control commands: Based on the multi-scenario prediction results obtained in step S4, the intelligent decision-making module combines the preset slope stability safety threshold and ecological water demand threshold, and uses a multi-objective optimization algorithm to generate control commands in real time, dynamically adjusting the opening and flow direction of the first electromagnetic flow valve and the second electromagnetic flow valve; according to the predicted rainfall scenario, the control mode adaptively switches between the heavy rainfall period control mode, the rainfall interval period control mode and the drought period control mode. The heavy rainfall control mode aims to rapidly reduce pore water pressure and ensure slope stability. The intelligent decision-making module calculates and outputs commands to allow the first electromagnetic flow valve to open at maximum, directly discharging water from the blind ditch into natural water bodies via the drainage pipe. Simultaneously, the second electromagnetic flow valve remains closed, halting ecological water replenishment and maximizing drainage efficiency. The rainfall interval control mode aims to regulate water resources and maintain stable water levels. The intelligent decision-making module controls the first electromagnetic flow valve, switching water flow to the storage well and dynamically adjusting the inflow rate to maintain the well's water level at 80% to 90% of its rated capacity. If the well is full, it automatically switches to drainage mode. Simultaneously, based on soil moisture sensing… The system receives feedback that if the root layer moisture content of the slope vegetation is lower than the ecological threshold, the second electromagnetic flow valve will be opened at an appropriate time to provide localized water replenishment. The drought control mode focuses on ecological water conservation and maintaining vegetation health. The intelligent decision-making module controls the first electromagnetic flow valve to divert all water flow into the storage well. The second electromagnetic flow valve is precisely opened according to the spatial distribution data of soil moisture, prioritizing compensatory irrigation for water-deficient areas in the early morning or evening to maintain the root layer soil moisture content within a suitable range. The intelligent decision-making module sends the valve opening degree, flow direction, and control timing commands to the field control unit through a wireless communication network, driving the actuator to achieve intelligent and adaptive seepage control. In practice, after receiving the prediction results from S4, the intelligent decision-making module immediately activates its built-in multi-objective optimization algorithm, decides to switch to the heavy rainfall control mode, and generates specific control commands, including switching the first electromagnetic flow valve to the path leading to the drainage pipe and setting the opening degree to 100%. At the same time, the water storage well is maintained at 50% water level to prepare for possible subsequent regulation or emergency situations.

[0023] S6. System self-optimization: Integrate environmental data, decision instructions, and execution results from each control loop into structured cases, store them in the historical database, and expand the training sample set; periodically use the newly added samples to incrementally train the physical constraint neural network, optimize model parameters, enhance its prediction accuracy and decision reliability under different climatic conditions and geological environments, and achieve continuous evolution of system performance.

[0024] The above describes one or more embodiments of the present invention in a relatively specific and detailed manner, but it should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this patent should be determined by the appended claims.

Claims

1. An intelligent ecological blind ditch for groundwater seepage on water-rich fill slopes, characterized in that, It includes blind drains, water circulation modules, control valve modules, seepage monitoring modules, and intelligent decision-making modules; The blind ditch is constructed of geotextile, gravel filling material and permeable pipes, forming a continuous drainage channel inside; the side walls and bottom of the blind ditch are provided with water inlet channels for collecting seepage water from inside the slope. The water circulation module includes a drainage pipe, a water storage well, and a water supply pipe, which are sealed together by flange interfaces or pipe connectors to form a water circulation path. The drainage pipe connects the blind ditch and the natural water body to discharge water from the blind ditch into the natural water body. The water storage well connects to the blind ditch to temporarily store water from the blind ditch. The water supply pipe connects to the water storage well to directionally transport the groundwater stored in the water storage well to the root system layer of the slope vegetation. The control valve module includes a first electromagnetic flow valve and a second electromagnetic flow valve, which control the direction and flow rate of water through the opening and closing of the valves; wherein, the first electromagnetic flow valve is installed at the connection between the blind ditch and the drain pipe and the water storage well; the second electromagnetic flow valve is installed at the connection between the water storage well and the water supply pipe; The seepage monitoring module consists of multiple types of sensors deployed inside the slope to collect key seepage parameters in real time. The seepage monitoring module includes a head sensor for monitoring the water head inside the slope soil, a flow sensor for measuring the real-time flow in the blind ditch, and a soil moisture sensor for monitoring the moisture content of the slope soil. The core of the intelligent decision-making module is a neural network model based on physical constraints. It predicts and makes intelligent decisions on the seepage state of the slope based on the slope design parameters and real-time parameters, and generates control commands for the first and second electromagnetic flow valves. The control strategy takes into account both slope stability and ecological water conservation requirements throughout the process.

2. The intelligent ecological blind ditch for groundwater seepage in water-rich fill slopes according to claim 1, characterized in that, The neural network model adopts a hierarchical design, including an input layer, a feature extraction layer, a physical constraint layer, and an output layer connected sequentially. The input layer uses a fully connected structure. The feature extraction layer consists of multiple hidden layers, used to automatically learn and fuse the complex nonlinear relationship between the slope seepage process and stability state in the spatiotemporal dimension contained in the input data, extracting deep features. The physical constraint layer is located between the feature extraction layer and the output layer, used to introduce physical constraints based on seepage mechanics and slope stability principles during model training. By embedding physical laws into the loss function in the form of constraint terms, the model output results are forced to conform to the physical mechanism, thereby enhancing the model's generalization ability and prediction reliability. The physical constraint terms include seepage field control equation constraints and static equilibrium equation constraints. The seepage field control equation constraints are shown in the following equation: in, h The total head is in meters (m). K ( h ) represents the unsaturated hydraulic conductivity, in m / s; S represents the source-sink term, reflecting rainfall infiltration or evaporation, in s. -1 ; θ ( h () represents the volumetric water content; t For time, s; The constraints of the static equilibrium equations are shown below: That is, the resultant force of all forces F The resultant torque of all torques is 0. M =0; The output layer is configured to generate the final prediction result and control parameters based on the fused features extracted by the feature extraction layer and the requirements of the physical constraint layer; the output layer uses a linear activation function, and the output values ​​of its neurons must satisfy the boundary conditions defined by the physical constraint layer; the model's total loss function... L total Data loss L data Constraint loss of the control equation of the seepage field L influent and static equilibrium equation constraint loss L static It consists of, that is, satisfying the following expression: in, L data The mean squared error between the model's predicted values ​​and the actual monitored values ​​is used to ensure the model's ability to fit the monitoring data. L influent and L static These are all physical constraint error terms, used to ensure that the model output conforms to physical common sense; μ 1 , μ 2 and μ 3 All are weighting coefficients, and satisfy the following conditions: μ 1 + μ 2 + μ 3 =1, the specific value of which is determined through cross-validation; The training sample set consists of historical engineering measured data and virtual samples based on physical mechanism simulation; the virtual samples are generated through numerical simulation methods.

3. A method for controlling groundwater seepage in water-rich fill slopes, characterized in that, The intelligent ecological blind ditch for groundwater seepage in water-rich embankment slopes as described in claim 1 includes the following steps: S1. Extraction of Slope Design Parameters: Slope geometric parameters and slope soil physical and mechanical parameters are extracted from design data for model building and predictive analysis. The slope geometric parameters include slope height, slope ratio, slope orientation, and fill layer thickness. The slope soil physical and mechanical parameters include cohesion. c s internal friction angle φ s Compression modulus E s and permeability coefficient k s ; S2. Real-time acquisition of multi-dimensional parameters: Real-time seepage parameters of the slope are collected by seepage monitoring modules deployed inside the blind ditch and in the surrounding soil, including water head data inside the slope soil. H ( x , y , z , t Soil moisture content N ( x , y , z , t ) and real-time flow rate in blind drains Q ( t Simultaneously, rainfall forecast information for the slope location is obtained from the local meteorological information platform through the data interface, including short-term and medium-term rainfall forecast information, including rainfall probability, peak hourly rainfall intensity, and cumulative rainfall; in addition, vegetation information on the slope surface is obtained through high-resolution satellite remote sensing image interpretation, including vegetation distribution density, vegetation type, and growth status information. S3. Preprocessing of monitoring data: Based on the slope geometric parameters extracted in step S1, a three-dimensional spatiotemporal grid model matching the actual spatial structure of the slope is constructed; the slope soil physical and mechanical parameters, the real-time slope seepage parameters, the rainfall forecast information and the vegetation distribution information are denoised and normalized, and combined on a unified spatiotemporal grid model to construct a spatiotemporally aligned multidimensional feature vector, providing good input features for the neural network model; S4. Predictive analysis based on physical constraint neural network: Input the multidimensional feature vector constructed in step S3 into the trained physical constraint based neural network model, and output the prediction results through model calculation; the prediction results include water head value and its rate of change, soil moisture content and its rate of change, flow rate change curve and peak flow rate in blind ditch, and slope stability coefficient. S5. Generation and execution of control commands: Based on the multi-scenario prediction results obtained in step S4, the intelligent decision-making module combines the preset slope stability safety threshold and ecological water demand threshold, and uses a multi-objective optimization algorithm to generate control commands in real time, dynamically adjusting the opening degree and flow direction of the first electromagnetic flow valve and the second electromagnetic flow valve. Based on the predicted rainfall scenario, the control mode adaptively switches between the heavy rainfall period control mode, the rainfall interval period control mode, and the drought period control mode. The heavy rainfall control mode aims to rapidly reduce pore water pressure and ensure slope stability. The intelligent decision-making module calculates and outputs instructions to make the first electromagnetic flow valve open to the maximum extent to discharge the water in the blind ditch directly to the natural water body through the drainage pipe; at the same time, the second electromagnetic flow valve is kept closed to stop ecological water replenishment and fully ensure drainage efficiency. The rainfall interval control mode aims to regulate water resources and maintain stable water levels. The intelligent decision-making module controls the first electromagnetic flow valve to switch the water flow to the water storage well and dynamically adjusts the inflow rate to maintain the water level in the water storage well at 80% to 90% of its rated capacity. If the water storage well is full, it automatically switches to drainage mode. At the same time, based on feedback from the soil moisture sensor, if the moisture content of the root layer of the slope vegetation is lower than the ecological threshold, the second electromagnetic flow valve is controlled to open at an appropriate time to replenish water locally. The drought control mode takes ecological water conservation and maintaining vegetation health as its core objectives. The intelligent decision-making module controls the first electromagnetic flow valve to divert all the water flow into the storage well for storage. The second electromagnetic flow valve is precisely opened according to the spatial distribution data of soil moisture, and prioritizes compensatory irrigation for water-deficient areas in the early morning or evening to keep the soil moisture content in the root zone within a suitable range. The intelligent decision-making module sends valve opening degree, flow direction and control timing commands to the field control unit through a wireless communication network, driving the actuator to act, and realizing intelligent and adaptive seepage control; S6. System self-optimization: Integrate environmental data, decision instructions, and execution results from each control loop into structured cases, store them in the historical database, and expand the training sample set; periodically use the newly added samples to incrementally train the physical constraint neural network, optimize model parameters, enhance its prediction accuracy and decision reliability under different climatic conditions and geological environments, and achieve continuous evolution of system performance.

4. The control method for intelligent ecological blind drains for groundwater seepage in water-rich fill slopes according to claim 3, characterized in that, The noise reduction and normalization process in step S3 specifically includes the following steps: S301, Data Cleaning: Wavelet denoising algorithm is used to filter out high-frequency interference noise in the parameter data, according to 3 σ The criteria identify outliers and fill them with linear interpolation based on the parameter change trends of adjacent time points and adjacent spatial nodes; S302. Data Normalization: The cleaned parameters are mapped to a unified [0,1] interval to eliminate scale differences caused by different physical units, ensuring that the weights of each parameter are comparable in the neural network model; the normalized parameter values... X norm The calculation expression is: in, X These are the parameter values ​​after cleaning. X max and X min These are the global maximum and minimum values ​​of the parameter recorded in the historical engineering database and the current monitoring period, respectively. S303, Spatiotemporal Alignment: The cleaned and normalized parameters are aligned according to their corresponding spatiotemporal coordinates. x , y , z , t The model is registered and synchronized on a three-dimensional spatiotemporal grid model, and integrated to form a spatiotemporally aligned multidimensional feature vector, which is adapted to the input requirements of the neural network model.

Images