Intelligent early warning and cooperative prevention and control method and system for karst underground water

Abstract

The invention relates to the technical field of hidden danger early warning, in particular to a karst groundwater intelligent early warning and cooperative prevention and control method and system.The method comprises the steps that cooperative monitoring is conducted on target pollutants and multi-element physical and chemical parameters, and multi-source heterogeneous data are collected; constructing a digital twinborn model, fusing multi-source heterogeneous data, introducing a periodic active hydraulic disturbance diagnosis mechanism, dynamically updating the state of the model, and realizing simulation and prediction; comparing a prediction result of the model with a dynamic threshold library, and automatically triggering a grading early warning signal; in response to the grading early warning signal, reversely positioning the position of the pollution source by utilizing an adjoint equation method and combining a particle backtracking algorithm, comprehensively performing cross validation, and generating a pollution source probability distribution diagram; according to the early warning level and the pollution source probability distribution diagram, calculating an optimal prevention and control scheme, and generating a specific execution instruction; and the execution mechanism receives and executes the instruction. Therefore, the problems of pollution discovery lagging, tracing difficulty, high treatment difficulty and the like are solved.

Description

Technical Field

[0001] This application relates to the field of hazard early warning technology, and in particular to intelligent early warning and collaborative control methods and systems for karst groundwater. Background Technology

[0002] Karst springs are unique hydrogeological phenomena in carbonate rock distribution areas. They not only provide vital water for production and daily life in the surrounding areas but also maintain a unique karst ecosystem, possessing irreplaceable ecological, economic, and social value. However, the management of karst springs is a significant challenge. The core issue stems from the extreme complexity of the karst aquifer itself. First, the aquifer is composed of various media such as dissolution channels, fissures, and pores, with vastly different permeability across space. Water flow preferentially chooses large channels, resulting in highly concentrated and unpredictable pollutant transport paths. Second, the water flow velocity in karst channels is extremely high. Once pollutants enter the dominant runoff channels, they rapidly diffuse to the spring mouth, and the transport process occurs underground, making it difficult to detect. Furthermore, the recharge area of ​​the spring is vast, and the pollution sources are dispersed, making pollution source tracing extremely difficult. The vast underground space and strong heterogeneity make it difficult for traditional treatment methods (such as injecting remediation agents) to effectively distribute and contact with pollutants, resulting in high treatment difficulty and a high likelihood of recurrence. Summary of the Invention

[0003] This application provides a method and system for intelligent early warning and collaborative control of karst groundwater to solve problems such as delayed pollution detection, difficulty in tracing sources, and high difficulty in remediation.

[0004] The first aspect of this application provides a method for intelligent early warning and collaborative control of karst groundwater, including the following steps: S1. Deploy pollution-triggered intelligent tracer capsule arrays, distributed optical fiber sensor networks and multi-parameter sensor node groups to collaboratively monitor target pollutants and multi-dimensional physicochemical parameters of karst groundwater and collect multi-source heterogeneous data. S2. Construct a digital twin model, integrate the multi-source heterogeneous data, and introduce a periodic active hydraulic disturbance diagnosis mechanism. Dynamically update the state of the digital twin model through data assimilation technology to achieve real-time simulation and multi-step advance prediction of the water flow field and solute concentration field. S3. Compare the prediction results of the digital twin model with the dynamic threshold library to automatically trigger graded early warning signals with three levels: attention, warning, and alarm. S4. In response to the graded early warning signal, the location of the pollution source is located in reverse by using the adjoint equation method and the particle backtracking algorithm, and the trigger signals of the intelligent tracer capsule array are cross-validated to generate a pollution source probability distribution map. S5. Based on the triggered warning level and the pollution source probability distribution map, a model predictive control optimization algorithm is used to perform multi-objective optimization to calculate the optimal control scheme with the coordinated scheduling of the dam group as the core, and specific execution instructions are generated. S6. The actuator receives the execution command and precisely adjusts the gate opening or pump station speed to achieve the governance goal of hydraulic coordinated resistance control.

[0005] Optionally, the pollution-triggered smart tracer capsule array has capsule shells made of environmentally responsive materials, containing built-in fluorescent tracers, and the degradation rate of the capsule shells is positively correlated with the concentration of the target pollutant, thereby triggering release when pollution exceeds the standard; the distributed optical fiber sensor network is used to collect distributed physical field data; the multi-parameter sensor node group is used to collect water quality and hydrological data at the point location; the multi-source heterogeneous data includes the distributed physical field data and the water quality and hydrological data, wherein the distributed physical field data includes temperature field data and vibration field data; the water quality and hydrological data includes water level, pH (hydrogen ion concentration), conductivity, dissolved oxygen, turbidity, and target ion concentration data.

[0006] Optionally, the periodic active hydraulic disturbance diagnosis mechanism includes: By setting a cycle or triggering an emergency, a short-term, quantitative pumping or injection experiment is conducted at a preset location downstream to artificially create a hydraulic stimulus. The digital twin model analyzes the water level response, water quality response, and tracer response of each upstream monitoring point to this stimulus, and inverts and updates the permeability field and porosity field parameters of the karst aquifer to achieve model calibration. The standard response curves measured during the healthy period of disturbance are stored in the database. In subsequent periodic disturbances, the newly measured response curves are compared with the standard response curves to identify anomalies and achieve early anomaly localization.

[0007] Optionally, the warning thresholds in the dynamic threshold library are dynamically calculated and generated by the digital twin model based on the current water flow state, historical background concentration, and the uncertainty of the prediction itself. The core calculation formula is:

[0008] in, For the final dynamic threshold, As the background baseline value, For safety reasons, The standard deviation of the predicted value. The correlation coefficient, The current simulated flow rate, This represents the historical average flow velocity.

[0009] Optionally, the adjoint equation method obtains a sensitivity field by constructing and solving the adjoint equation of the concentration field at the observation point on the release intensity of the pollution source, and then normalizes the sensitivity field to obtain a probability distribution map of the pollution source contribution; the particle backtracking algorithm simulates the reverse migration of the particles by releasing a large number of virtual particles at the warning point and using the reverse flow field provided by the digital twin model, statistically analyzes the spatial distribution of all particles at the end of the simulation, and generates a probability distribution map; the cross-validation generates a final high-confidence pollution source probability distribution map by comparing the pollution source contribution probability distribution map and the probability distribution map with the actual triggering position of the smart tracer capsule.

[0010] Optionally, the model predictive control optimization algorithm performs multi-objective optimization solutions including: A multi-objective cost function is constructed with the objectives of simultaneously minimizing pollution concentration, gate operation volume, and treatment economic cost, and is solved through rolling optimization within a finite time domain. The multi-objective cost function is as follows:

[0011] in, To minimize overall cost, To minimize the deviation of pollution concentration from the target value, To minimize the magnitude of change in the control variables, To minimize the economic cost of governance, , , These are the weighting coefficients.

[0012] Optionally, after step S6, there is also an effect feedback and adaptive optimization step: the execution status of the implementing agency and the environmental response data after governance are fed back to the digital twin model in real time to evaluate governance performance and to perform incremental learning and optimization of the model.

[0013] A second aspect of this application provides an intelligent early warning and collaborative system for karst groundwater, comprising: an intelligent sensing module, a data fusion and communication module, a digital twin and early warning module, an intelligent source tracing module, a collaborative decision engine, an execution control module, and an adaptive module, wherein... The intelligent sensing module includes an intelligent tracer capsule array, a distributed optical fiber sensor network, and a group of multi-parameter sensor nodes, which are used to execute an active hydraulic disturbance mechanism and collect multi-source heterogeneous data. The data fusion and communication module is used to clean, align, and transmit the multi-source heterogeneous data; The digital twin and early warning module has a built-in digital twin model, which is used to simulate, predict and trigger graded early warnings; The intelligent source tracing module is used to run the adjoint equation method and particle backtracking algorithm to generate a pollution source probability distribution map; The collaborative decision engine is used to run model predictive control optimization algorithms to generate optimal resistance and control schemes and execution instructions; The execution control module is used to receive and execute the instructions, and control the execution mechanism to complete the action; The adaptive module is used to complete governance performance evaluation and incremental model learning.

[0014] A third aspect of this application provides an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being configured to perform the intelligent early warning and collaborative control method for karst groundwater as described in the above embodiments.

[0015] A fourth aspect of this application provides a computer-readable storage medium having a computer program stored thereon, which is executed by a processor to implement the intelligent early warning and collaborative control method for karst groundwater as described in the above embodiments.

[0016] Therefore, this application has at least the following beneficial effects: This application's embodiments achieve 24-hour real-time monitoring and minute-level data acquisition by deploying a pollution-triggered intelligent tracer capsule array, a distributed fiber optic sensor network, and a multi-parameter sensor node group. It also revolutionizes the early warning mode from "alarm after exceeding the standard" to "predictive early warning before pollution occurs or in the early stages of diffusion," buying valuable time for response. Through a periodic active hydraulic disturbance diagnostic mechanism, it inverts and updates the aquifer permeability field, assesses channel patency, and achieves early diagnosis of system health. Furthermore, by integrating multiple source tracing technologies such as the adjoint equation method and particle backtracking algorithm, combined with on-site evidence from intelligent tracers, it achieves... The rapid, accurate, and reliable location of pollution sources overcomes the significant challenge of tracing pollution sources in karst channels. Through digital twins and model predictive control optimization algorithms, multi-objective optimization solutions can be achieved, automatically simulating the treatment effects and economic costs of different schemes and providing a scientifically optimal solution, fundamentally improving the scientific rigor and accuracy of decision-making. Abandoning the extensive treatment model, a precise treatment strategy of "hydraulic synergistic control" is proposed. By intelligently scheduling dams and pumping stations within the spring area, clean water flows are used to seize channels, block pollution plumes, and guide them to pre-designated treatment areas, achieving low-cost, high-efficiency, and environmentally friendly treatment goals.

[0017] This solved the technical problems of delayed pollution detection, difficulty in tracing the source, and great difficulty in pollution control.

[0018] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0019] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a flowchart of a method for intelligent early warning and collaborative control of karst groundwater according to an embodiment of this application; Figure 2 This is a schematic diagram of a karst groundwater intelligent early warning and collaborative control system provided in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of an electronic device provided according to an embodiment of this application. Detailed Implementation

[0020] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0021] The following describes, with reference to the accompanying drawings, an intelligent early warning and collaborative control method and system for karst groundwater according to embodiments of this application. Addressing the problems of delayed pollution detection, difficulty in tracing sources, and high remediation difficulty mentioned in the background art, this application provides an intelligent early warning and collaborative control method for karst groundwater. In this method, 24-hour real-time monitoring and minute-level data acquisition are achieved by deploying a pollution-triggered intelligent tracer capsule array, a distributed optical fiber sensor network, and a multi-parameter sensor node group. The early warning mode is innovated from "alarm after exceeding the standard" to "predictive early warning before pollution occurs or in the early stages of diffusion," buying valuable time for response. Through a periodic active hydraulic disturbance diagnostic mechanism, the aquifer permeability field is updated by inversion, and channel patency is assessed, achieving early diagnosis of system health. Furthermore, by integrating the adjoint equation method and particle... By employing backtracking algorithms and other source tracing technologies, combined with on-site evidence from intelligent tracers, rapid, accurate, and reliable location of pollution sources is achieved, overcoming the significant challenge of tracing pollution sources in karst channels. Through digital twins and model predictive control optimization algorithms, multi-objective optimization solutions are implemented, automatically simulating the treatment effects and economic costs of different schemes and providing scientifically optimal solutions, fundamentally improving the scientific rigor and accuracy of decision-making. Abandoning the extensive treatment model, a precise treatment strategy of "hydraulic synergistic control" is proposed. This involves intelligently scheduling dams and pumping stations within the spring area, using clean water flow to seize channels, block pollution plumes, and guide them to pre-designated treatment areas, achieving low-cost, high-efficiency, and environmentally friendly treatment goals. This solves the technical problems of delayed pollution detection, difficulty in source tracing, and high treatment complexity.

[0022] Specifically, Figure 1 This is a flowchart illustrating a method for intelligent early warning and collaborative control of karst groundwater provided in an embodiment of this application.

[0023] like Figure 1 As shown, the intelligent early warning and collaborative control method for karst groundwater includes the following steps: In step S101, a pollution-triggered smart tracer capsule array, a distributed optical fiber sensor network, and a multi-parameter sensor node group are deployed to collaboratively monitor target pollutants and multi-dimensional physicochemical parameters in karst groundwater and collect multi-source heterogeneous data.

[0024] Among them, the pollution-triggered smart tracer capsule array has a capsule shell made of environmentally responsive material, contains a fluorescent tracer, and the degradation rate of the capsule shell is positively correlated with the concentration of the target pollutant, thereby triggering release when pollution exceeds the standard; a distributed fiber optic sensor network is used to collect distributed physical field data; a multi-parameter sensor node group is used to collect water quality and hydrological data at the point location; multi-source heterogeneous data includes distributed physical field data and water quality and hydrological data, wherein the distributed physical field data includes temperature field data and vibration field data; the water quality and hydrological data includes water level, pH, conductivity, dissolved oxygen, turbidity, and target ion concentration data.

[0025] Pollution-triggered smart tracer capsule arrays primarily serve as targeted early warning and source tracing triggers, acting as "chemical signal flares" deployed before risk points. They solve the problem that traditional sensors cannot be deployed at all potential risk points and cannot provide early warnings before pollution occurs. Because the capsule shell is made of an environmentally responsive material, such as an environmentally responsive hydrogel, the cross-linking degree of this gel chemically reacts with the concentration of a specific target pollutant. When the target pollutant diffuses around the capsule and reaches a threshold concentration, it undergoes ion exchange or integration with the gel network, causing the cross-linking bonds of the gel network to break or the structure to become loose, thus rendering the capsule shell ineffective. The high concentration of easily detectable fluorescent tracer encapsulated in the capsule shell is immediately released into the groundwater. The released tracer moves with the water flow. This sudden signal can be easily captured by fluorescent sensors at downstream springs or monitoring wells.

[0026] In one feasible approach, the selection of environmentally responsive hydrogels includes: pH-sensitive hydrogels, ion-sensitive hydrogels, and enzyme-catalyzed hydrogels. pH-sensitive hydrogels can be used to monitor acidic mine wastewater or alkaline industrial wastewater; ion-sensitive hydrogels specifically respond to target pollutants such as nitrate, chromate, and ammonium; and enzyme-catalyzed hydrogels can respond to specific organic pollutants. The internal tracer is selected from fluorescent substances with high fluorescence intensity, low background value, good stability, and environmental friendliness, such as sodium fluorescein and rhodamine WT. Sodium fluorescein is a green fluorescent substance with low cost and the widest application; rhodamine WT is a red fluorescent substance with better stability and strong anti-interference ability.

[0027] Understandably, the pollution-triggered smart tracer capsule array is normally in a dormant state. Once a specific pollutant exceeds the standard, the capsules will "self-destruct" and release easily traceable "dye," signaling to downstream users, "Something's happened here!" This provides the most direct and immediate pollution event alarm signal, thus achieving early warning. The capsule's placement location itself is the most direct suspected source of pollution, and the tracer concentration curve it releases provides verification data for subsequent source tracing.

[0028] It should be noted that the target pollutant trigger threshold for environmentally responsive materials is determined based on the water quality standards of the protected target, and the degradation threshold of the capsule shell can be customized in cooperation with material suppliers.

[0029] In one implementation, the smart tracer capsule array is deployed using a drilling implantation technique to fix the smart tracer capsules in karst fissures or pipe inlets downstream of potential pollution sources, ensuring full contact with groundwater. Depending on the risk level, 3-5 smart tracer capsules are deployed downstream of each key source to form a monitoring array and prevent omissions.

[0030] In one implementation, the distributed fiber optic sensor network is deployed by burying optical cables underground at the bottom of riverbeds or in key areas where leakage may occur, to monitor surface water-groundwater exchange.

[0031] In one implementation, the deployment of the multi-parameter sensor node group involves placing sensors at different depths (shallow, medium, and deep) in key monitoring wells to reveal vertical water quality stratification; and placing nodes at springs, water intakes at water sources, at the boundaries of different aquifers, and upstream and downstream of polluted areas. A star network topology is used to ensure reliable data transmission.

[0032] In step S102, a digital twin model is constructed, multi-source heterogeneous data is integrated, and a periodic active hydraulic disturbance diagnosis mechanism is introduced. The state of the digital twin model is dynamically updated through data assimilation technology to achieve real-time simulation and multi-step advance prediction of the water flow field and solute concentration field.

[0033] The periodic active hydraulic disturbance diagnosis mechanism includes: artificially creating a hydraulic excitation by conducting short-term, quantitative pumping or injection experiments at a preset downstream location through periodic setting or emergency triggering; the digital twin model analyzes the water level response, water quality response, and tracer response of each upstream monitoring point to this excitation, and inversely updates the permeability coefficient field and porosity field parameters of the karst aquifer to achieve model calibration; the standard response curves measured during the disturbance period in the healthy period are stored in the database, and in subsequent periodic disturbances, the newly measured response curves are compared with the standard response curves to identify anomalies and achieve early anomaly location.

[0034] It should be noted that the cycle setting method can be customized to a suitable time according to needs. For example, in one embodiment, it is preset to be carried out once every quarter; in another embodiment, it is preset to be carried out once a month during the period of stable water flow to establish a "health baseline". The emergency triggering method refers to the additional disturbance manually triggered by the administrator during high-risk periods, such as after heavy rain or after a pollution accident in the vicinity.

[0035] Because karst groundwater systems are like extremely complex, invisible "black boxes," traditional monitoring can only passively wait for changes to occur within this black box (such as pollution events) before reacting. The periodic active hydraulic disturbance diagnostic mechanism does not passively wait for natural changes, but rather periodically and actively pumps water at downstream springs or major discharge points for short periods and in fixed quantities, artificially creating a water level fluctuation of known intensity and regularity. This fluctuation propagates upstream of the aquifer in the form of water pressure waves. The disturbance time and pumping flow rate are artificially set. The pumping flow rate needs to be large enough to generate a measurable water level response that propagates upstream, but not so large as to cause problems such as land subsidence. This is usually determined through preliminary experiments; for example, the pumping flow rate is set between 50-100 m³ / h. The disturbance time is typically set to 6-24 hours to ensure that the drawdown cone can fully develop and propagate. For example, pumping water at a flow rate of 1000 m³ / h for 6 hours at a major downstream spring.

[0036] Due to downstream hydraulic disturbances, upstream water levels will exhibit corresponding water level, water quality, and tracer responses in response to these disturbances. The water level response refers to the time, magnitude, and pattern of water level decline at each upstream monitoring point; the water quality response refers to the synchronous changes in water quality parameters (such as conductivity and temperature) at each upstream point; and the tracer response refers to the abrupt changes in the migration velocity of tracers if a tracer signal is present during the disturbance.

[0037] Based on the above response, during the pumping period and for a period of time after it ends, all upstream monitoring points synchronously collect multi-source heterogeneous data at a high frequency (e.g., once per minute) to form the "water level-time" drop curve and "water quality-time" change curve for the entire field.

[0038] The digital twin model receives all data from the perturbation experiment, including pumping volume, water level-time drop curves at various points, and water quality-time change curves. Through an inversion algorithm, it continuously and automatically adjusts parameters such as permeability and porosity in each region of the digital twin model until the simulated water level response curve closely matches the actually observed water level response curve, thus achieving model calibration. After multiple perturbation calibrations, when the simulated water level response curve perfectly matches the actually observed water level response curve, this embodiment of the application obtains a permeability field and porosity field that infinitely approximates the real situation, thereby completing high-precision calibration.

[0039] After undergoing high-precision calibration, the high-fidelity digital twin model has more accurate early warning capabilities and more reliable traceability capabilities.

[0040] Understandably, active hydraulic disturbance is an innovation in data acquisition methods, providing digital twins with high-value "dynamic health check data" that traditional passive monitoring cannot obtain for calibrating models; digital twins are an innovation in data analysis and modeling methods, utilizing disturbance data to greatly improve their simulation accuracy, thereby providing more reliable early warning, tracing, and decision support.

[0041] Specifically, the state of the digital twin model is dynamically updated through data assimilation technology, enabling real-time simulation and multi-step advance prediction of the water flow field and solute concentration field. The goal of real-time simulation is to make the state of the digital twin model infinitely close to the current state of the real world, primarily achieved through data assimilation technology. After data assimilation ensures the current state is as accurate as possible, the model can then perform multi-step predictions. The specific steps of real-time simulation include: starting from the current time t, the digital twin model, based on its internal physical equations (such as groundwater flow equations and solute transport equations) and the current best estimated state, calculates a very short time step (e.g., the next 10 minutes) to obtain a predicted state value; at time t+1, new real-time observation data is transmitted from sensors (water level, water quality, etc.) located in various locations; the data assimilation algorithm begins to work, calculating the difference between the predicted state value and the new real-time observation data, and then, based on the uncertainties of both (model error and observation error), optimally "absorbing" or "injecting" the new information from the observations into the model; finally, a new, corrected optimal estimated state value is output. This optimal estimated state value is closer to the real situation than either a simple model prediction or a simple observation, thus completing an update of the model state and achieving real-time simulation.

[0042] It should be noted that data assimilation algorithms aim to optimally fuse observed data (real-world measurements) with dynamic models (such as digital twin models) to obtain the most accurate and complete estimate of the current state of the system. The core idea is to neither completely trust the model nor completely trust the data.

[0043] Optionally, in some embodiments, the data assimilation algorithm is the Kalman filter method. The algorithm flow is as follows: Starting from the current time step, each member of the model set (e.g., 100) runs the model independently, forecasting forward to the next moment when observation data is available. At this time, each member will obtain a forecast value, and the distribution of all members represents the uncertainty range of the model forecast. When new observation data arrives, assimilation begins: the forecast value of each member is compared with the actual observation value, and the difference is calculated. Based on the statistical relationship of all members, an optimal gain matrix is ​​calculated. This matrix value determines the extent to which the observation value should be used to correct the model forecast value in this embodiment. If the observation is very reliable (small error) and the model forecast uncertainty is large, then the matrix value is large, and the correction magnitude is large; if the observation noise is large (large error) and the model forecast is confident, then the matrix value is small, and the correction magnitude is small. Each member in the set is corrected using the following formula: ; in, This represents the optimal gain matrix value.

[0044] After all model members' states have been corrected, their average value is the optimal estimate of the system's current state, and their dispersion represents the uncertainty of the updated state. Using the updated state as the new initial field, the above steps are repeated iteratively.

[0045] After ensuring the current state is as accurate as possible, the model can begin multi-step forward prediction. The optimal estimated state value, after data assimilation and correction, is used as the initial field for prediction. This is a prerequisite for accurate prediction. When running the prediction model, the digital twin model will no longer receive new observation data, but will freely calculate forward based purely on its internal physical mechanisms and initial state. In this process, it will incorporate known or predicted external driving conditions, such as meteorological forecast data including rainfall and evaporation for the next 72 hours, as well as known human activity plans such as industrial and agricultural water intake plans and reservoir scheduling plans, and boundary conditions such as river water levels and the exchange volume of other aquifers. The model will calculate the hydraulic head and pollutant concentration values ​​at various locations throughout the entire simulation area at every future time step (e.g., once per hour), ultimately generating a "spatiotemporal evolution animation of the water flow field and solute concentration field," which visually demonstrates how the pollution plume moves, diffuses, and dilutes.

[0046] Understandably, through data assimilation technology, digital twin models can continuously synchronize with the real world, ensuring that their virtual state is consistent with reality. Based on this, they can scientifically and quantitatively predict future water flow and pollutant transport, thus achieving a revolutionary shift from "passive monitoring" to "proactive early warning," providing decision-makers with valuable lead time.

[0047] In step S103, the prediction results of the digital twin model are compared with the dynamic threshold library to automatically trigger a graded early warning signal that includes three levels: attention, warning, and alarm.

[0048] The warning thresholds in the dynamic threshold library are dynamically calculated and generated by the digital twin model based on the current water flow status, historical background concentration, and the uncertainty of the prediction itself. The core calculation formula is:

[0049] in, For the final dynamic threshold, As the background baseline value, For safety reasons, The standard deviation of the predicted value. The correlation coefficient, The current simulated flow rate, This represents the historical average flow velocity.

[0050] It should be noted that the background baseline value represents the natural background concentration or historical normal level of a certain pollutant in a water body under conditions of no human interference or low-level human activity. It is typically taken as the higher quantile (e.g., 95th percentile) of historical concentration data from the same location and time period (such as the same quarter) within the last 3-5 years. The background baseline value serves as the baseline for dynamic thresholds, and the background baseline value itself can be adjusted according to the season and hydrological period to form a more refined background database. The term represents the uncertainty increment, indicating the uncertainty in the model's prediction. It measures the model's confidence in predicting the concentration at a future point in time. A large value indicates low reliability and high uncertainty in the prediction results. The values ​​are obtained through ensemble forecasts. For example, running a digital twin model 100 times (each time with slightly different parameters or initial conditions) yields 100 predicted values; the standard deviation of these predicted values ​​is... Safety factor It is a multiplier set according to risk preference, usually with a value of 2 or 3. It is the hydrological condition increment; the current simulated velocity is the groundwater pore velocity or Darcy velocity at the current moment, provided by the digital twin model; the historical average velocity is the average velocity at this location during the same period or over a long period of history; the correlation coefficient is a coefficient obtained through regression analysis of historical data, representing the sensitivity of velocity changes to pollutant concentrations, and is usually a positive number.

[0051] It's important to explain that the digital twin model, after data assimilation and multi-step prediction, outputs a prediction result that is not a single number, but a multi-dimensional, spatiotemporal data product containing uncertain information. This mainly includes: a pollutant concentration field, the probability distribution or confidence interval for each predicted value, a water flow field, and a pollution plume transport path. In this embodiment, the probability distribution or confidence interval for each predicted value is given in the form of standard deviation. The water flow field predicts future changes in water level and velocity, which helps in understanding the driving forces and directions of pollutant transport. The pollution plume transport path refers to the three-dimensional spatial trajectory and influence range of pollutants in groundwater as they move, diffuse, and migrate downstream from the pollution source. In this embodiment, the pollution plume transport path is visualized, clearly showing the movement trajectory, diffusion range, and arrival time of the pollution plume from the source to the discharge point.

[0052] The prediction results are compared with a dynamic threshold library. The specific steps are as follows: For each target location to be monitored (such as a spring or water source intake), at each future prediction time point, the system dynamically calculates a corresponding threshold based on the previously introduced formula. Then, the prediction result is compared point-by-point over time with the corresponding dynamic threshold. Finally, uncertainty analysis is performed. Uncertainty analysis means comparing not only the "best predicted value" but also the entire prediction interval. For example, even if the best predicted value is slightly lower than the threshold, if the upper limit of its 90% confidence interval far exceeds the threshold, the system will still consider it a high-risk situation.

[0053] Based on the above comparison results, a tiered early warning system is implemented. Specifically, when the predicted concentration exceeds 70% of the threshold and shows an upward trend, a warning is triggered, and a notification (e.g., flashing yellow) is displayed on the management platform interface, notifying management personnel to pay closer attention and arranging for manual verification. When the predicted concentration continues to approach the threshold, or uncertainty analysis indicates a high risk, source tracing analysis is automatically initiated, a preliminary control plan is generated, and an early warning message is sent to management personnel via SMS / App, initiating emergency preparedness. When the predicted concentration is confirmed to exceed the dynamic threshold or definitive evidence is received from the site (capsule trigger), a collaborative control mechanism is automatically triggered, issuing control instructions to the implementing agency (e.g., opening dams). Simultaneously, all relevant responsible persons are notified via the highest priority method (telephone, SMS).

[0054] Understandably, the dynamic threshold library comparison method, where dynamic thresholds change with environmental conditions and model confidence, makes the comparison scientific and reasonable; comparison based on prediction results can provide early warnings before pollution occurs, giving managers valuable response time; and different response procedures are initiated according to the risk level, avoiding "overreacting to minor issues" or "missing major events," thus optimizing resource allocation.

[0055] In step S104, in response to the graded early warning signal, the location of the pollution source is located in reverse using the adjoint equation method combined with the particle backtracking algorithm, and the trigger signals of the intelligent tracer capsule array are cross-validated to generate a pollution source probability distribution map.

[0056] The adjoint equation method constructs and solves the adjoint equation of the concentration field at the observation point to the release intensity of the pollution source, obtaining a sensitivity field. This sensitivity field is then normalized to derive a probability distribution map of the pollution source contribution. The particle backtracking algorithm releases a large number of virtual particles at the warning point and utilizes the reverse flow field provided by a digital twin model to simulate the reverse migration of particles. It then statistically analyzes the spatial distribution of all particles at the end of the simulation and generates a probability distribution map. Cross-validation compares the pollution source contribution probability distribution map and the probability distribution map with the actual triggering location of the smart tracer capsule to generate a final high-confidence pollution source probability distribution map. Specifically, the algorithm process of the adjoint equation method is as follows: Step 1: Define a new variable (x, y, z, t) are called the accompanying state variables. The physical meaning is that, per unit time, a unit mass of pollutant is released at point (x, y, z), affecting the downstream observation point (x, y, z). , The contribution of the final concentration at point ). Step 2: Construct and solve the adjoint equation. Since the adjoint equation is the dual form of the original solute transport equation (convection-dispersion equation), the original solute transport equation is:

[0057] in, Let the dispersion coefficient tensor be... For concentration, For the velocity vector, The coefficient of the reaction term, Sources and sinks (pollution sources).

[0058] Therefore, the adjoint equation is:

[0059] in, Let be the dispersion coefficient tensor. For new variables, For the velocity vector, The coefficient of the reaction term, Let be the Dirac function, representing the impulse input that occurs at the downstream observation point. and observation time , For downstream observation points, The observation time.

[0060] Step 3: Solve the above adjoint equation to obtain the final time t= The accompanying state field ( , , , This state-related field is the required sensitivity field. The higher its value, the greater the contribution of pollutant released at that point to the concentration at downstream detection points.

[0061] Step 4: Assign the accompanying state field ( , , , After normalization, the probability distribution map of pollution source contribution can be obtained.

[0062] Understandably, the adjoint equation method only requires solving the partial differential equation once in both directions to obtain the sensitivity of all locations in the entire field, which is extremely efficient. In contrast, traditional methods require simulating pollution sources at each location, and the computational load increases linearly with the number of source points.

[0063] The particle backtracking algorithm is a stochastic simulation method based on the Lagrange framework. It can be visualized as releasing a large number of virtual particles at the location where pollution is detected (downstream), allowing them to wander randomly against time and the direction of water flow. The areas where these particles eventually stop or accumulate are the most likely locations of the pollution source.

[0064] Specifically, the algorithm process is as follows: Step 1: Using the calibrated digital twin model, simulate the steady flow field of the study area during (or currently) the pollution event, i.e. the velocity vector of each grid cell.

[0065] Step Two: At the downstream point where pollution is detected (e.g., at a spring), release a large number (e.g., 10,000) of virtual particles, simulate time stepping in the negative direction, and calculate the displacement of each virtual particle p in each time step. The calculation formula is:

[0066] in, The displacement of each virtual particle p in each time step. For the reverse advection term, For random diffusion term, For time steps.

[0067] It should be noted that the stochastic diffusion term is used to simulate the mechanical dispersion and molecular diffusion effects experienced by pollutants during transport. This is typically achieved using a stochastic model, such as:

[0068] in, Let be a random vector that follows a standard normal distribution. The hydrodynamic dispersion coefficient, For time steps.

[0069] Step 3: Stop the simulation when a particle reaches the set simulation start point (e.g., backtracking 30 days), reaches the model boundary, or comes to a standstill. Record the final coordinates of each particle when it stops.

[0070] Step 4: Divide the entire model area into a 3D grid, count the number of particles that eventually become stationary in each grid cell, and calculate the probability value for each grid cell. The probability value is the ratio of the number of stationary particles to the total number of particles. Visualize the calculated probability values ​​to obtain the pollution source probability distribution map. The darker the color (the denser the particles), the higher the probability that the area is a pollution source.

[0071] Cross-validation involves overlaying and comparing the "pollution source contribution probability distribution map" obtained by the adjoint equation method with the "probability distribution map" obtained by the particle backtracking method. If both indicate that the same area (area A) has the highest probability, then the confidence of area A is greatly increased. The overlapping high-probability area (area A) is then compared with the actual triggering location of the smart tracer capsule. If the capsule triggering point is exactly within area A, a complete chain of evidence—theoretical inference + numerical simulation + field evidence—is formed, resulting in the highest confidence level for the conclusion. A final high-confidence pollution source probability distribution map is then generated, clearly identifying the most likely source area. If the capsule triggers in another area, area B, while the theoretical simulation points to area A, the algorithm will identify this inconsistency and initiate uncertainty analysis. It might determine that area B is a currently active pollution source, while area A is a potential old or secondary pollution source. The final generated probability map will comprehensively reflect this judgment and provide managers with multiple possibility analyses.

[0072] Understandably, by automatically comparing different results through algorithms, the conclusions are no longer black-box outputs, but rather verified and interpretable. The final generated "Pollution Source Probability Distribution Map" is a visualized heat map, allowing even non-experts to easily identify key investigation areas, greatly improving the efficiency and accuracy of source tracing work.

[0073] In step S105, based on the triggered warning level and the probability distribution map of the pollution source, a model predictive control optimization algorithm is used to perform multi-objective optimization to calculate the optimal control scheme with the coordinated scheduling of the dam group as the core, and specific execution instructions are generated.

[0074] It is understood that the warning level determines the weighting of optimization objectives. At the alarm level, water quality safety is prioritized at all costs, while economic costs and equipment stability are given lower weights. At the warning level, the effectiveness of treatment, economic efficiency, and stability are weighed. At the attention level, there may be a greater emphasis on low-cost, small-scale monitoring and scheduling. The pollution source probability distribution map provides crucial spatial information, determining the main areas for treatment. It tells the optimization algorithm which area the pollution is most likely to originate from, and the algorithm will prioritize scheduling upstream or downstream water conservancy facilities in that area, rather than blindly operating across the entire basin.

[0075] Model predictive control optimization algorithms do not perform one-time static programming, but rather a rolling optimization strategy. The optimization revolves around a mathematical function that quantifies the overall cost of a "good" solution:

[0076] in, To minimize overall cost, It is a water quality target, aimed at minimizing the concentration of pollutants at sensitive points (such as springs). The goal is to ensure stability, thereby preventing gates and pumping stations from operating too frequently or violently, protecting equipment, and ensuring system stability. The goal is to achieve economic objectives while minimizing governance costs. , , These are weighting coefficients. These three coefficients are dynamically determined by the warning level. When the warning level is alarm, The weight is extremely large. , The weights are very small; when the warning level is "warning," the three weights are equal, seeking a balance; when the warning level is "attention," the weights are small. The weight is quite large.

[0077] Specifically, the model predictive control optimization algorithm constructs a multi-objective cost function with the objectives of simultaneously minimizing pollution concentration, gate operation volume, and treatment economic cost, and performs rolling optimization within a finite time domain. The multi-objective cost function is as follows:

[0078] in, To minimize overall cost, To minimize the deviation of pollution concentration from the target value, To minimize the magnitude of change in the control variables, To minimize the economic cost of governance, , , These are the weighting coefficients.

[0079] It is easy to associate with, For the above , For the above , For the above .

[0080] The optimization solver finds the control sequence that minimizes the total cost from thousands of possible scheduling combinations. Optimal control schemes typically include a combination of the following strategies: upstream dilution with clean water: scheduling upstream reservoirs or dams to increase the discharge flow, using clean water to flush and dilute pollutants, thus blocking the flow path; downstream throttling and blocking: scheduling downstream dams to close or reduce the discharge flow in a timely manner, artificially raising the local groundwater level, slowing or even temporarily blocking the advance of the pollution plume towards the spring, buying time for treatment; lateral guidance: through scheduling, guiding the polluted water to pre-designated artificial wetlands or ecological ponds for natural treatment, preventing it from directly entering sensitive water areas.

[0081] The final generated execution instructions are specific and executable. For example: Instruction 1: Adjust gate A linearly from the current 30% opening to 55% opening within 600 seconds; Instruction 2: Start pump station B and set its speed to 1200 RPM with a target flow rate of 40 m³ / h; Instruction 3: After 24 hours, restore gate A to 40% opening.

[0082] These instructions are sent directly to the PLC (Programmable Logic Controller) in the field through industrial control systems (such as SCADA systems) for automatic execution.

[0083] Understandably, the generated decision-making instructions are based on model simulation and optimization algorithms rather than empiricism, making the generation of instructions more scientific; the water conservancy facilities of the entire basin are treated as a whole for coordinated scheduling, achieving "unified" governance; from early warning to instruction generation and execution, the entire process is automated, achieving unprecedented response speed and processing accuracy.

[0084] In step S106, the actuator receives the execution command and precisely adjusts the gate opening or pump station speed to achieve the governance goal of hydraulic coordinated resistance control.

[0085] Specifically, the actuator receives the precise instructions described above and adjusts the gate opening or pump station speed precisely according to the instructions. For example, the gate A is linearly adjusted from the current 30% opening to 55% opening within 600 seconds, and pump station B is started, with its speed set to 1200 RPM and the target flow rate to 40 m³ / h.

[0086] It should be noted that after step S106 above, there is also an effect feedback and adaptive optimization step: the execution status of the implementing agency and the environmental response data after governance are fed back to the digital twin model in real time to evaluate governance performance and to incrementally learn and optimize the model.

[0087] It's important to understand that the data in the execution status feedback includes the actual opening degree, actual rotational speed, actual flow rate, current, and voltage of the actuators (gates, pumping stations). The purpose is to verify whether the commands are executed accurately. For example, if the output command is for the gate to open to 50%, but the actual feedback is only 48%, it indicates a possible mechanical jam or error, and this deviation will be recorded. The environmental response feedback data includes changes in environmental parameters such as groundwater level, water quality (pollutant concentration), and temperature after the treatment begins. The purpose is to assess the actual effects of the treatment actions on the real environment. This is the ultimate standard for verifying the correctness of the decision.

[0088] After receiving the feedback data, the corresponding module will automatically conduct a quantitative assessment of the governance performance: based on the set evaluation indicators such as concentration reduction rate, pollution plume control efficiency, cost-effectiveness ratio, and command execution compliance, the assessment will be carried out, and a performance report of this governance action will be automatically generated, for example: "This control action took 12 hours, the nitrate concentration at the spring mouth decreased from 25mg / L to 12mg / L, the reduction rate was 52%, the total electricity consumption was XXX kWh, and the average command execution compliance rate was 99.5%." Simultaneously, the actual water level and water quality response data during the remediation period were used as new, high-quality inputs and re-input into the digital twin model for data assimilation. Backward optimization was performed to adjust model parameters (such as permeability coefficient K and dispersion) to make the model's simulation of the system response during this remediation process more accurate. This means that through this practical experience, the model has gained a better understanding of groundwater behavior.

[0089] In analyzing "instruction-effect" data, it may be discovered that the weighting coefficients in the cost function are not set appropriately. For example, it may be found that slightly higher electricity costs ( The weight can be appropriately reduced) in exchange for governance effectiveness. With a significant increase in weight, the system will fine-tune the weight to make better decisions in the next iteration.

[0090] Understandably, by optimizing the feedback loop, predictions and decisions can become increasingly aligned with the characteristics of the actual watershed, resulting in continuous improvement in accuracy. The model can adapt to changes in the environment (such as new engineering structures altering hydrogeological conditions or the emergence of new pollution sources) and adjust itself through self-learning, without the need for manual reprogramming or large-scale model adjustments, allowing the model to evolve automatically and continuously.

[0091] The intelligent early warning and collaborative method for karst groundwater proposed in this application deploys a pollution-triggered intelligent tracer capsule array, a distributed optical fiber sensor network, and a multi-parameter sensor node group to achieve 24-hour real-time monitoring and minute-level data acquisition. It also revolutionizes the early warning mode from "alarm after exceeding the standard" to "predictive early warning before pollution occurs or in the early stages of diffusion," buying valuable time for response. Through a periodic active hydraulic disturbance diagnostic mechanism, it inverts and updates the aquifer permeability field, assesses channel patency, and achieves early diagnosis of system health. Furthermore, it integrates multiple source tracing technologies, such as the adjoint equation method and particle backtracking algorithm, and combines intelligent... The on-site evidence from tracers enables rapid, accurate, and reliable location of pollution sources, overcoming the significant challenge of tracing pollution sources in karst channels. Through digital twins and model predictive control optimization algorithms, multi-objective optimization solutions can be achieved, automatically simulating the treatment effects and economic costs of different schemes and providing a scientifically optimal solution, fundamentally improving the scientific rigor and accuracy of decision-making. Abandoning the extensive treatment model, a precise treatment strategy of "hydraulic synergistic control" is proposed. By intelligently scheduling dams and pumping stations within the spring area, clean water flows are used to seize channels, block pollution plumes, and guide them to pre-designated treatment areas, achieving low-cost, high-efficiency, and environmentally friendly treatment goals.

[0092] This solved the technical problems of delayed pollution detection, difficulty in tracing the source, and great difficulty in pollution control.

[0093] In some embodiments, a famous karst spring area in northern my country is an important source of drinking water and a tourist attraction. Its recharge area contains farmland and scattered villages. During the 2024 rainy season, the system successfully issued an early warning and handled an agricultural non-point source nitrate pollution incident.

[0094] Dozens of nitrate-triggered capsules (with shells sensitive to nitrate concentration and internal Rhodamine WT) were pre-embedded in karst fissures downstream of the main farmland areas and near village sewage outlets within the recharge zone. Furthermore, several kilometers of temperature / vibration measuring optical fibers were deployed along known major groundwater flow channels and around springs. Sensor nodes capable of real-time monitoring of water level, pH, conductivity, and nitrate concentration were installed at springs, water source intakes, and monitoring wells along major flow paths. All devices transmit data back to the central platform every 5 minutes via IoT technology.

[0095] The digital twin model assimilated rainfall forecast data and predicted an upward trend in nitrate concentration at the spring within the next 72 hours. The system executed a planned quarterly "active hydraulic disturbance," pumping water quantitatively at the spring for 6 hours. Analysis of the response curve revealed that the response speed at monitoring point A in the northern agricultural area was significantly slower than historical "health fingerprints," indicating decreased channel patency and potential siltation, making pollutants more likely to accumulate. At 10:00 AM on a certain day, the model predicted a continued rise in concentration, but it had not yet exceeded the standard. Due to the identification of anomaly point A, the system automatically triggered a "pay attention" level warning (yellow signal), prompting management personnel to be aware of the northern area.

[0096] At 2 p.m. that day, a smart capsule downstream of the northern farmland was triggered, releasing rhodamine WT tracer. Simultaneously, sensors showed a sharp increase in nitrate concentration in the area. The system immediately upgraded to "alert" level (orange signal).

[0097] Using the adjoint equation method, rapid calculations revealed that the sensitivity field of spring concentration had the highest contribution rate in region B, downstream of the northern farmland area. Simulations using the particle backtracking method showed that a large number of particles terminated in region B. Cross-validation revealed that region B highly overlapped with the location triggered by the smart capsule. An automatically generated "high-confidence pollution source probability distribution map" accurately pinpointed the pollution source to an area of ​​approximately 0.5 square kilometers in the northern farmland area.

[0098] The model predicts that the pollution plume will affect the spring water quality in 36 hours. The system automatically triggers the "alarm" level (red signal).

[0099] The optimal solution is finally calculated by using a model predictive control optimization algorithm to solve multi-objective optimization problems. Instruction 1: Immediately open the gate of the ecological reservoir G1 in the northern upstream area, increase the discharge flow by 20%, and dilute and flush away pollutants with clean water.

[0100] Instruction 2: Close the control gate Z2 downstream of the pollution plume by 30% to artificially raise the water level, forming a "hydraulic barrier" to slow the speed at which the pollution plume advances towards the spring.

[0101] Instruction 3: Divert a portion of the water to an artificial wetland on the east side, which will be transformed from an abandoned mine pit, for natural purification.

[0102] The instructions are sent to the PLCs of G1 Reservoir and Z2 Gate via the Industrial Internet of Things, and the actuators automatically complete precise adjustments.

[0103] The actual opening degree of the G1 gate matched the command 99%. The monitoring network showed that the concentration in the northern area began to decrease, and the migration speed of the pollution plume slowed down significantly. After 36 hours, the concentration at the spring outlet fluctuated only slightly, far below the warning line. The system assessment report showed: "This control was successful, avoiding a major pollution event, and the power cost was only XXX yuan." The data from the entire event (such as the actual effect of closing the Z2 gate by 30% on the water level) was used for incremental learning, which optimized the parameters of the digital twin model and the cost function of MPC, making the system "smarter".

[0104] Next, referring to the accompanying drawings, we describe the intelligent early warning and collaborative system for karst groundwater proposed according to the embodiments of this application.

[0105] In this embodiment, anomalies are detected through a three-dimensional monitoring network, early warnings are achieved through digital twin prediction and proactive diagnosis, pollution sources are quickly located through multi-technology fusion for source tracing, and precise location of pollution sources is achieved; hydraulic control is achieved through intelligent decision-making and scheduling of water conservancy facilities; and continuous optimization is achieved through feedback loop, making it increasingly intelligent with use.

[0106] Figure 2 This is a schematic diagram of the structure of the intelligent early warning and collaborative system for karst groundwater according to an embodiment of this application.

[0107] like Figure 2 As shown, the karst groundwater intelligent early warning and collaborative system 10 includes: an intelligent sensing module 100, a data fusion and communication module 200, a digital twin and early warning module 300, an intelligent tracing module 400, a collaborative decision engine 500, an execution control module 600, and an adaptive module 700.

[0108] The system includes: an intelligent sensing module 100, comprising an intelligent tracer capsule array, a distributed optical fiber sensor network, and a multi-parameter sensor node group, used to execute active hydraulic disturbance mechanisms and collect multi-source heterogeneous data; a data fusion and communication module 200, used to clean, align, and transmit multi-source heterogeneous data; a digital twin and early warning module 300, with a built-in digital twin model, used to perform simulation, prediction, and trigger graded early warnings; an intelligent source tracing module 400, used to run the adjoint equation method and particle backtracking algorithm to generate a pollution source probability distribution map; a collaborative decision engine 500, used to run model predictive control optimization algorithms to generate optimal control schemes and execution instructions; an execution control module 600, used to receive and execute instructions to control the actuators to complete actions; and an adaptive module 700, used to complete governance performance evaluation and incremental model learning.

[0109] It should be noted that the foregoing explanation of the embodiment of the intelligent early warning and collaborative method for karst groundwater also applies to the intelligent early warning and collaborative system for karst groundwater in this embodiment, and will not be repeated here.

[0110] The intelligent early warning and collaborative system for karst groundwater proposed in this application deploys a pollution-triggered intelligent tracer capsule array, a distributed fiber optic sensor network, and a multi-parameter sensor node group to achieve 24-hour real-time monitoring and minute-level data acquisition. It also revolutionizes the early warning mode from "alarm after exceeding the standard" to "predictive early warning before pollution occurs or in the early stages of diffusion," buying valuable time for response. Through a periodic active hydraulic disturbance diagnostic mechanism, it inverts and updates the aquifer permeability field, assesses channel patency, and achieves early diagnosis of system health. Furthermore, it integrates multiple source tracing technologies, such as the adjoint equation method and particle backtracking algorithm, combined with intelligent... The on-site evidence from tracers enables rapid, accurate, and reliable location of pollution sources, overcoming the significant challenge of tracing pollution sources in karst channels. Through digital twins and model predictive control optimization algorithms, multi-objective optimization solutions can be achieved, automatically simulating the treatment effects and economic costs of different schemes and providing a scientifically optimal solution, fundamentally improving the scientific rigor and accuracy of decision-making. Abandoning the extensive treatment model, a precise treatment strategy of "hydraulic synergistic control" is proposed. By intelligently scheduling dams and pumping stations within the spring area, clean water flows are used to seize channels, block pollution plumes, and guide them to pre-designated treatment areas, achieving low-cost, high-efficiency, and environmentally friendly treatment goals.

[0111] This solved the technical problems of delayed pollution detection, difficulty in tracing the source, and great difficulty in pollution control.

[0112] Figure 3 A schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include: The memory 301, the processor 302, and the computer program stored on the memory 301 and capable of running on the processor 302.

[0113] When the processor 302 executes the program, it implements the intelligent early warning and collaborative method for karst groundwater provided in the above embodiments.

[0114] Furthermore, electronic devices also include: Communication interface 303 is used for communication between memory 301 and processor 302.

[0115] The memory 301 is used to store computer programs that can run on the processor 302.

[0116] The memory 301 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0117] If the memory 301, processor 302, and communication interface 303 are implemented independently, then the communication interface 303, memory 301, and processor 302 can be interconnected via a bus to complete communication between them. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of representation, Figure 3 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0118] Optionally, in a specific implementation, if the memory 301, processor 302, and communication interface 303 are integrated on a single chip, then the memory 301, processor 302, and communication interface 303 can communicate with each other through an internal interface.

[0119] Processor 302 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application.

[0120] This application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the above-described intelligent early warning and collaborative method for karst groundwater.

[0121] This application also provides a computer program product, which stores a computer program that, when executed by a processor, implements the above-mentioned intelligent early warning and collaborative method for karst groundwater.

[0122] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0123] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0124] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more N executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0125] It should be understood that the various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0126] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

Claims

1. A method for intelligent early warning and collaborative control of karst groundwater, characterized in that, The following steps are involved: S1. Deploy pollution-triggered intelligent tracer capsule arrays, distributed optical fiber sensor networks and multi-parameter sensor node groups to collaboratively monitor target pollutants and multi-dimensional physicochemical parameters of karst groundwater and collect multi-source heterogeneous data. S2. Construct a digital twin model, integrate the multi-source heterogeneous data, and introduce a periodic active hydraulic disturbance diagnosis mechanism. Dynamically update the state of the digital twin model through data assimilation technology to achieve real-time simulation and multi-step advance prediction of the water flow field and solute concentration field. S3. Compare the prediction results of the digital twin model with the dynamic threshold library to automatically trigger graded early warning signals with three levels: attention, warning, and alarm. S4. In response to the graded early warning signal, the location of the pollution source is located in reverse by using the adjoint equation method and the particle backtracking algorithm, and the trigger signals of the intelligent tracer capsule array are cross-validated to generate a pollution source probability distribution map. S5. Based on the triggered warning level and the pollution source probability distribution map, a model predictive control optimization algorithm is used to perform multi-objective optimization to calculate the optimal control scheme with the coordinated scheduling of the dam group as the core, and specific execution instructions are generated. S6. The actuator receives the execution command and precisely adjusts the gate opening or pump station speed to achieve the governance goal of hydraulic coordinated resistance control.

2. The intelligent early warning and collaborative control method for karst groundwater according to claim 1, characterized in that, The pollution-triggered intelligent tracer capsule array has capsule shells made of environmentally responsive materials, containing fluorescent tracers. The degradation rate of the capsule shells is positively correlated with the concentration of the target pollutant, thus triggering release when pollution exceeds the standard. The distributed fiber optic sensor network is used to collect distributed physical field data. The multi-parameter sensor node group is used to collect water quality and hydrological data at point locations. The multi-source heterogeneous data includes the distributed physical field data and the water quality and hydrological data, wherein the distributed physical field data includes temperature field data and vibration field data; the water quality and hydrological data includes water level, pH, conductivity, dissolved oxygen, turbidity, and target ion concentration data.

3. The intelligent early warning and collaborative control method for karst groundwater according to claim 1, characterized in that, The periodic active hydraulic disturbance diagnosis mechanism includes: By setting a cycle or triggering an emergency, a short-term, quantitative pumping or injection experiment is conducted at a preset location downstream to artificially create a hydraulic stimulus. The digital twin model analyzes the water level response, water quality response, and tracer response of each upstream monitoring point to this stimulus, and inverts and updates the permeability field and porosity field parameters of the karst aquifer to achieve model calibration. The standard response curves measured during the healthy period of disturbance are stored in the database. In subsequent periodic disturbances, the newly measured response curves are compared with the standard response curves to identify anomalies and achieve early anomaly localization.

4. The intelligent early warning and collaborative control method for karst groundwater according to claim 1, characterized in that, The warning thresholds in the dynamic threshold library are dynamically calculated and generated by the digital twin model based on the current water flow state, historical background concentration, and the uncertainty of the prediction itself. The core calculation formula is: ； in, For the final dynamic threshold, As the background baseline value, For safety reasons, The standard deviation of the predicted value. The correlation coefficient is... The current simulated flow rate, This represents the historical average flow velocity.

5. The intelligent early warning and collaborative control method for karst groundwater according to claim 1, characterized in that, The adjoint equation method constructs and solves the adjoint equation of the concentration field at the observation point to the release intensity of the pollution source, obtains the sensitivity field, and normalizes the sensitivity field to obtain the pollution source contribution probability distribution map. The particle backtracking algorithm releases a large number of virtual particles at the warning point and uses the reverse flow field provided by the digital twin model to simulate the reverse migration of the particles, statistically analyzes the spatial distribution of all particles at the end of the simulation, and generates a probability distribution map. The cross-validation compares the pollution source contribution probability distribution map and the probability distribution map with the actual triggering position of the smart tracer capsule to generate the final high-confidence pollution source probability distribution map.

6. The intelligent early warning and collaborative control method for karst groundwater according to claim 1, characterized in that, The model predictive control optimization algorithm performs multi-objective optimization solutions including: A multi-objective cost function is constructed with the objectives of simultaneously minimizing pollution concentration, gate operation volume, and treatment economic cost, and is solved through rolling optimization within a finite time domain. The multi-objective cost function is as follows: ； in, To minimize overall cost, To minimize the deviation of pollution concentration from the target value, To minimize the magnitude of change in the control variables, To minimize the economic cost of governance, , , These are the weighting coefficients.

7. The intelligent early warning and collaborative control method for karst groundwater according to claim 1, characterized in that, Step S6 is followed by an effect feedback and adaptive optimization step: the execution status of the implementing agency and the environmental response data after governance are fed back to the digital twin model in real time to evaluate governance performance and to perform incremental learning and optimization of the model.

8. A karst groundwater intelligent early warning and collaborative system, used to implement the method described in any one of claims 1-7, characterized in that, include: The system comprises an intelligent sensing module, a data fusion and communication module, a digital twin and early warning module, an intelligent traceability module, a collaborative decision-making engine, an execution control module, and an adaptive module. The intelligent sensing module includes an intelligent tracer capsule array, a distributed optical fiber sensor network, and a group of multi-parameter sensor nodes, which are used to execute an active hydraulic disturbance mechanism and collect multi-source heterogeneous data. The data fusion and communication module is used to clean, align, and transmit the multi-source heterogeneous data; The digital twin and early warning module has a built-in digital twin model, which is used to simulate, predict and trigger graded early warnings; The intelligent source tracing module is used to run the adjoint equation method and particle backtracking algorithm to generate a pollution source probability distribution map; The collaborative decision engine is used to run model predictive control optimization algorithms to generate optimal resistance and control schemes and execution instructions; The execution control module is used to receive and execute the instructions, and control the execution mechanism to complete the action; The adaptive module is used to complete governance performance evaluation and incremental model learning.

9. An electronic device, characterized in that, include: The method includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the intelligent early warning and collaborative control method for karst groundwater as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, The program is executed by the processor to implement the intelligent early warning and collaborative control method for karst groundwater as described in any one of claims 1-7.

Images