Groundwater pumping and injection integrated intelligent management and control method and system based on internet of things

Abstract

This invention discloses an integrated intelligent control method and system for groundwater pumping and injection based on the Internet of Things (IoT). The method includes: conducting geological surveys of the target area and constructing a three-dimensional geological model, thereby planning the layout of pumping and injection well groups and sensor networks; deploying a monitoring network and collecting data based on this layout, and constructing a digital twin of the contaminated site by fusing the three-dimensional geological model; extracting groundwater status data from the digital twin, using an LSTM model to predict pollutant migration paths and water level dynamics, and outputting pollution scenario simulation information corrected for geological characteristics; using this simulation information as input, coupled with geological parameters as constraints, and employing the BLMFO two-layer multi-objective optimization algorithm to generate a collaborative control strategy set for the pumping and injection well group; finally, executing control commands, monitoring the real-time situation during the remediation process, judging and pushing strategy change suggestions, including flow rate adjustment, reagent optimization schemes, and equipment operating parameter corrections, to improve remediation efficiency and adapt to complex geological environments.

Description

Technical Field

[0001] This invention relates to the field of environmental protection and intelligent control technology, and in particular to an integrated intelligent management and control method and system for groundwater extraction and injection based on the Internet of Things. Background Technology

[0002] With the acceleration of industrialization, groundwater pollution is becoming increasingly serious. Traditional groundwater injection systems have the following significant drawbacks: Insufficient real-time performance: Relying on periodic manual monitoring and control, they cannot respond promptly to sudden changes in water quality (such as chemical leaks). High energy consumption and cost: Fixed equipment operating parameters and a lack of dynamic optimization mechanisms result in maintenance costs accounting for more than 40% of total repair costs. Poor geological adaptability: Low injection efficiency under complex geological conditions such as fissure water and karst areas.

[0003] While existing technologies propose a two-layer, multi-objective optimization framework, they only focus on the coupling between simulation and optimization algorithms, lacking the ability to achieve a closed-loop execution capability driven by real-time monitoring data. For example, the optimization results need to be manually imported into the equipment for execution, failing to achieve a second-level response time for "sudden water quality changes - parameter adjustments - execution feedback"; furthermore, they do not involve modular hardware integration design and solar hybrid power supply systems, making rapid deployment difficult in scenarios without external power or in complex terrain. In existing technologies, the application of IoT and deep learning is mostly limited to single-device control, lacking multi-objective collaborative optimization and closed-loop management throughout the entire process. Therefore, there is an urgent need for a groundwater pumping and injection system that integrates real-time monitoring, intelligent decision-making, and efficient execution to improve remediation efficiency, reduce costs, and adapt to complex geological environments. Summary of the Invention

[0004] This invention overcomes the shortcomings of the prior art and provides an integrated intelligent management and control method and system for groundwater extraction and injection based on the Internet of Things. Its main purpose is to improve restoration efficiency, reduce costs, and adapt to complex geological environments.

[0005] To achieve the above objectives, the first aspect of this invention provides an integrated intelligent management and control method for groundwater extraction and injection based on the Internet of Things, comprising:

[0006] Geological surveys were conducted on the target area and a three-dimensional geological model of the area was constructed. The three-dimensional geological model of the area was then used to plan the layout of the injection well group and the heterogeneous sensor monitoring network, resulting in a regional injection and remediation layout scheme.

[0007] Based on the aforementioned regional injection and remediation layout scheme, a heterogeneous sensor monitoring network is deployed to collect regional data, obtain spatial distribution data and water quality monitoring data of regional groundwater, and integrate them with the regional three-dimensional geological model to establish a digital twin of the contaminated site.

[0008] The groundwater status data of the target area is extracted by the digital twin of the contaminated site and geological constraints are set. The LSTM prediction model is used to predict the migration path of pollutants and the dynamic water level, and the pollution scenario simulation information corrected by geological characteristics is output.

[0009] Using the pollution scenario simulation information as input, coupled with the permeability coefficient and fracture distribution parameters as constraints, the BLMFO two-layer multi-objective optimization algorithm is used to perform multi-objective collaborative optimization calculations to generate a set of collaborative control strategies for the injection and extraction well group.

[0010] Based on the collaborative control strategy set of the injection and extraction well group, a sequence of control instructions is generated and sent to the corresponding execution mechanism for execution. During the injection and extraction repair process, the real-time operating status is monitored and it is determined whether the repair strategy needs to be changed. The repair strategy is then generated as a suggestion and pushed out.

[0011] In this scheme, the geological survey of the target area and the construction of a three-dimensional geological model of the area are carried out. The three-dimensional geological model is then used to plan the layout of the injection well group and heterogeneous sensor monitoring network, resulting in a regional injection remediation layout scheme. Specifically, this includes:

[0012] By drilling at gridded locations, undisturbed soil and rock samples at different depths were obtained and geophysical interpretation was performed. Remote sensing images and historical geological maps of the target area were collected as supplementary data. At the same time, layered hydrogeological parameters were obtained through field pumping tests, water injection tests, and water pressure tests. A regional geological exploration dataset containing spatial location, stratigraphic properties, and hydrogeological parameters was constructed.

[0013] Based on the regional geological exploration dataset, the Kriging interpolation algorithm is used to spatially interpolate discrete borehole lithology data and a three-dimensional geological structure framework model including stratigraphic lithology, tectonic faults and weathering zones is constructed by combining deterministic modeling methods. The sequential indicator simulation algorithm is used to perform random simulation of the spatial distribution of lithology in the three-dimensional geological structure framework model, and finally a regional three-dimensional geological model representing the geometric morphology of stratigraphic interfaces and the spatial heterogeneity of lithological parameters is obtained.

[0014] The three-dimensional geological model of the region was imported into the groundwater simulation software. The Delaunay triangulation method was used to discretize the unstructured grid to generate a finite volume computational grid suitable for complex geological boundary conditions. Each grid cell was assigned corresponding hydrogeological parameters, and hydraulic boundary conditions and source-sink terms were set.

[0015] The MODFLOW module is called to solve the control equations for groundwater flow, obtaining the streamline distribution and steady-state velocity field of the target area. Driven by the steady-state velocity field, the spatiotemporal migration and evolution process of pollutants under the control of complex geological structures is analyzed by solving the convection-diffusion-response equations, identifying the diffusion path of the pollution plume and potential high-concentration accumulation areas, and obtaining the simulation results of pollutant migration.

[0016] In this scheme, the step of conducting geological surveys of the target area and constructing a three-dimensional geological model of the area, and using the three-dimensional geological model to plan the layout of injection well groups and heterogeneous sensor monitoring networks to obtain a regional injection remediation layout scheme, also includes:

[0017] Based on the results of pollutant migration simulation, an optimization model is established with the objectives of maximizing the pollution capture rate and minimizing the construction cost. A multi-objective particle swarm optimization algorithm is used to randomly generate well site populations within the simulated pollution plume range. The pollution interception efficiency is evaluated by calculating the overlap volume between the well site capture zone and the pollution plume. The drilling cost is estimated by combining the well depth and formation lithology. After iterative calculation, the optimal solution set is output to generate the injection well group layout scheme.

[0018] The current frontal position, main migration path, and boundary of the hydraulic capture zone of the pollution plume are extracted from the pollutant migration simulation results. The differential deployment design of heterogeneous sensors is then carried out in conjunction with the injection well group layout scheme.

[0019] Distributed fiber optic sensors based on Brillouin optical time-domain reflectometers are laid along the line. The deployment trajectory is preferentially selected along the simulated and predicted pollution center axis or a cross section perpendicular to the groundwater flow direction. By monitoring the frequency shift of the backscattered light of the fiber optic cable, continuous temperature or strain information along the sensing cable is obtained to interpret the continuous changes in water quality.

[0020] For potential pollutant accumulation areas, areas with abrupt changes in hydrogeological conditions, and key verification points for remediation effectiveness, miniature spectral sensors are deployed. The specific installation locations are determined by performing grid calculations on the simulated concentration field using spatial interpolation algorithms, and the grid nodes with the largest concentration gradient changes are selected as the optimal deployment points.

[0021] The final solution is a regional injection repair layout. The communication distance between sensors is adjusted through network connectivity analysis to ensure that all sensor node data can be reliably transmitted to the converged edge computing node and cloud server via LoRaWAN or wired connection.

[0022] In this scheme, the deployment of a heterogeneous sensor monitoring network based on the regional injection remediation layout scheme and the acquisition of regional groundwater spatial distribution data and water quality monitoring data, combined with the regional three-dimensional geological model, to establish a digital twin of the contaminated site, specifically includes:

[0023] Obtain the regional injection and repair layout plan, and deploy the physical sensing layer based on the sensor deployment coordinates and communication paths determined in the regional injection and repair layout plan. Lay the distributed optical fiber sensors along the planned main monitoring axis and form a closed loop.

[0024] Miniature spectral sensors are fixedly installed at key monitoring nodes and connected to an independent power supply unit consisting of solar panels and battery packs, as well as a LoRaWAN wireless communication module. The working status of the sensing unit and network connectivity are verified through initialization tests, thus constructing a stable physical sensing layer infrastructure.

[0025] Regional data is collected through the physical sensing layer infrastructure. The distributed optical fiber sensing system emits probe light pulses through the demodulator and collects back Brillouin scattering signals. The original optical signals are processed by the frequency shift analysis algorithm. Combined with the pre-established calibration model, the frequency shift is converted into continuous distribution data of water quality parameters. At the same time, the miniature spectral sensor uses the spectral feature extraction algorithm to analyze the absorption spectrum of the water sample and calculate the pollutant concentration value.

[0026] Edge nodes perform digital filtering and signal enhancement on the collected raw data, and package the processed data with spatial location information and timestamps into transmission data units, which are then sent to the cloud data processing center through a wireless communication network. After receiving the multi-source monitoring data, the cloud platform performs spatiotemporal registration and data fusion.

[0027] A time series alignment algorithm is used to synchronize the measurement data of heterogeneous sensors, eliminating the time series deviation caused by the difference in sampling frequency. A spatial interpolation algorithm is used to generate a spatially continuous concentration distribution field from the concentration measurement values ​​of discrete points. At the same time, the physical quantity measurement values ​​of the distributed optical fiber are inverted into a water quality parameter distribution map using the calibration conversion relationship.

[0028] The water quality parameter distribution map is fused with a regional three-dimensional geological model. The two-dimensional water quality parameter field is vertically correlated with the stratigraphic structure and permeability coefficient field in the three-dimensional geological model through geostatistical analysis. The dynamics are then overlaid in the geological structure model as a three-dimensional cloud map using visualization technology. At the same time, the correlation mapping relationship between the sensor network and the monitoring data is established to realize interactive query and analysis of monitoring data and three-dimensional scene.

[0029] In this scheme, the extraction of groundwater status data of the target area through the digital twin of the contaminated site and the setting of geological condition constraints, the use of an LSTM prediction model to predict pollutant migration paths and water level dynamics, and the output of pollution scenario simulation information corrected for geological characteristics, specifically includes:

[0030] By querying the real-time data service interface of the digital twin, we can obtain the latest updated pollutant concentration spatial distribution matrix from the sensor network, the water level time series data of each monitoring point, and the instantaneous flow velocity field vector data obtained through numerical simulation.

[0031] Simultaneously, key geological attribute parameters pre-stored in the grid cells of the regional three-dimensional geological model are extracted, including the permeability tensor, effective porosity scalar, and spatial geometric and connectivity parameters of fracture development zones for each grid cell; the extracted geological parameters are matched and aligned with real-time monitoring data according to spatial coordinates to form an original dataset containing spatiotemporal dynamic data and static geological constraints.

[0032] Based on the original dataset, time-series data of water level and pollutant concentration are fused with corresponding geological parameters using an attention mechanism to generate a hybrid input vector containing temporal and static geological features. This vector is then segmented into fixed-length sample fragments, each containing observations from several past time steps and their corresponding geological features. Simultaneously, time-series data of external environmental factors are obtained using a big data network as an additional feature dimension added to the input vector, constructing a multi-dimensional training sample set with spatiotemporal correlation.

[0033] The multidimensional training sample set is input into the blank LSTM prediction model for supervised training. The input layer receives the time series data of multidimensional sample features, the hidden layer learns the complex long-term dependencies in the groundwater system through the forget gate, input gate and output gate, and the output layer generates the predicted values ​​of pollutant concentration and water level through a fully connected network, and finally obtains the trained LSTM prediction model.

[0034] The input for each time step includes the monitoring data at the current moment and the corresponding static geological parameters. The predicted output is calculated through forward propagation, and the gradient of the loss function with respect to the network weights is calculated through backpropagation using the BPTT algorithm. The network parameters are iteratively updated using an adaptive moment estimation optimizer.

[0035] The trained LSTM model is used for multi-step recursive prediction. The monitoring data at the current time is used as the initial state to predict the pollutant concentration distribution and water level field at the next time step. Then, the prediction results are used as new inputs to recursively predict and generate the spatial distribution sequence of pollutant concentration and the dynamic change sequence of water level at future time steps.

[0036] Geological constraints were applied to the generated spatial distribution sequence of pollutant concentration and dynamic change sequence of water level. Based on the spatial distribution tensor of permeability coefficient, the inverse distance weighted interpolation method was used to spatially register the permeability coefficient field with the predicted concentration field. Using the density field data of fracture development, the optimal path analysis method based on Dijkstra's algorithm was used to identify the dominant transport channels of pollutants in the fracture network. In the fracture development area, the convection-dispersion parameters of pollutants were adjusted according to the fracture density ratio. Finally, the pollution scenario simulation information corrected by geological characteristics was generated.

[0037] In this scheme, the pollution scenario simulation information is used as input, coupled with the permeability coefficient and fracture distribution parameters as constraints. A multi-objective collaborative optimization calculation is performed using the BLMFO two-layer multi-objective optimization algorithm to generate a collaborative control strategy set for the injection-extraction well group, specifically including:

[0038] A two-level multi-objective optimization algorithm for BLMFO is constructed, with three conflicting optimization objectives: minimizing the plume range, minimizing system operating energy consumption, and balancing equipment usage load. The flow distribution scheme, operating time series, and reagent dosage concentration of each injection well are used as decision variables to establish a mathematical model of the upper-level optimization problem that includes multi-objective functions and basic constraints.

[0039] Based on the established upper-level optimization problem model, a decision variable space is constructed by using a preset three-level coding strategy according to the real-time operation control strategy and operation monitoring data of the injection well group in the target area. An improved artificial bee colony optimization algorithm is introduced for upper-level solution calculation. Honey source initialization is performed by combining random selection and strategy selection. An initial Pareto optimal front is constructed through iterative optimization analysis to generate candidate control strategies.

[0040] The candidate control strategy is passed to the lower-level optimization model. In the lower-level model, a response surface model under geological constraints is established based on the spatial variation characteristics of the permeability coefficient and the fracture network connectivity parameters. The geological feasibility of each candidate strategy passed from the upper level is verified, and the verification results are fed back to the upper-level optimization process in the form of constraint violation degree.

[0041] Based on the geological constraint verification results fed back from the lower-level model, the upper-level optimization algorithm performs a co-evolution operation, using dynamically adjusted inertia weights and learning factors to update the velocity and position of particles, and eliminating inferior solutions that do not meet the geological conditions through a tournament selection mechanism. After a preset number of double-layer iterative calculations, the algorithm outputs the optimal solution set that achieves the best balance among multiple optimization objectives and meets the geological constraints, thus generating a set of collaborative control strategies for the injection and extraction well group.

[0042] In this scheme, a decision variable space is constructed through a preset three-level coding strategy, and an improved artificial bee colony optimization algorithm is introduced for upper-level solution calculation. Honey source initialization is performed by combining random selection and strategy selection. An initial Pareto optimal front is constructed through iterative optimization analysis, and candidate control strategies are generated. Specifically, this includes:

[0043] The three-level coding strategy consists of injection well codes, process codes, and monitoring codes. The injection well codes include the location and functional attributes of each injection well. The process codes include the running sequence and running strategy of the corresponding injection well. The monitoring codes are the running monitoring status characteristics of the corresponding injection well.

[0044] During the bee-hiring phase, a uniform crossover strategy is used to perform a neighborhood search on the current nectar source. A solution is randomly selected from the Pareto archive and swapped with the current solution for the code fragment. At the same time, a new candidate solution is generated at the machine coding layer using the strategy of minimizing cumulative processing time. The Pareto dominance principle is used to select a better solution from the old and new solutions and update the archive.

[0045] In the follower bee phase, the following probability of all nectar sources is calculated based on the non-dominance level and crowding of the nectar source. A roulette wheel method is used to select nectar sources that need to be searched further. The search strategy is consistent with that in the hired bee phase to ensure in-depth mining around high-quality solutions.

[0046] When the nectar source reaches the search limit but the quality does not improve, the scout bee uses a random exchange strategy to reconstruct the well group allocation code, randomly selects some well groups to be redistributed to other processing units, while maintaining the overall processing capacity balance, and regenerates the scheduling order and equipment parameters according to the initialization scheme, thereby increasing the population diversity.

[0047] During the iterative optimization process, the non-dominated solutions are recorded as the Pareto optimal solution set by updating the archive set. After each iteration, the current population is sorted by fast non-dominated sorting. Solutions ranked within the preset range are added to the archive set, and individuals dominated by the new solutions are removed from the archive set. After a preset number of iterations, the initial Pareto optimal frontier is output, and candidate control strategies are generated.

[0048] In this scheme, the step of generating a control command sequence based on the injection-extraction well group collaborative control strategy set and sending it to the corresponding execution mechanism for execution, monitoring the real-time operating status during the injection-extraction repair process and determining whether a repair strategy change is needed, generating a repair strategy as a suggestion and pushing it out, specifically including:

[0049] Based on the operating parameters and target set values ​​of each well contained in the coordinated control strategy set of the injection and extraction well group, the continuous control parameters in the strategy set are transformed into a sequence of instructions with temporal relationships through time discretization processing. Each instruction contains an equipment identifier, execution timestamp, control parameter set value and expected effect index. The executability and temporal rationality of each instruction within the equipment's capability range are verified by the instruction logic verification algorithm to form a verified control instruction sequence.

[0050] The verified control command sequence is sent to the corresponding actuators through the Internet of Things communication protocol. Each actuator immediately returns a confirmation signal after receiving the command and starts the command execution process. At the same time, the actual operating parameters of the equipment are collected in real time through built-in sensors, forming an execution feedback data stream that includes instantaneous flow rate, actual dosage of reagents and equipment operating status.

[0051] During the execution of the command, multi-dimensional operational status monitoring is initiated simultaneously. Groundwater dynamic data, including water level change sequences, pollutant concentration distribution data, and equipment operating status parameters, are collected in real time through a heterogeneous sensor network deployed in the injection well group. After data preprocessing, a high-quality real-time operational status dataset is generated.

[0052] Dynamic evaluation of remediation effectiveness is carried out based on real-time operational status dataset. The actual monitored pollutant concentration distribution is compared and analyzed with the remediation trajectory expected in the strategy set. The deviation between the actual and expected values ​​of remediation performance indicators is calculated. At the same time, by analyzing equipment operation status data, the workload and operating efficiency of each execution unit are evaluated, equipment performance degradation or abnormal operation are identified, and a regional remediation status assessment report is generated.

[0053] Based on the regional restoration status assessment report, a process for determining the necessity of strategy changes was initiated. A judgment logic based on multi-dimensional thresholds was established. When the actual rate of decrease in pollutant concentration is lower than the set percentage of the expected value, or when the pollutant concentration at key points shows a rebound trend, or when the equipment operating efficiency is continuously lower than the rated efficiency threshold, it is determined that a strategy change is necessary.

[0054] When a strategy change is deemed necessary, based on real-time operational characteristics and effect evaluation results, effective adjustment strategies under similar operating conditions are retrieved from the historical case database using a case-based reasoning method. Combined with the current equipment operating status and geological constraints, the generated flow adjustment range, reagent ratio optimization scheme, and equipment operating parameter correction values ​​are used to generate a repair strategy change suggestion, which is then pushed to the management terminal.

[0055] A second aspect of this invention provides an integrated intelligent control system for groundwater extraction and injection based on the Internet of Things (IoT). This system includes: a geological sensing and digital twin unit, an intelligent prediction and decision-making unit, an execution and closed-loop control unit, a data communication and edge computing unit, and an energy management and security unit.

[0056] The geological sensing and digital twin unit integrates geological survey data and real-time sensor network monitoring data to construct and continuously update a digital twin of the contaminated site, providing a comprehensive perception of the underground environment.

[0057] The intelligent prediction and decision-making unit predicts pollutant migration and water level dynamics based on data provided by the digital twin, and performs corrections by coupling geological constraints and formulates the optimal set of collaborative control strategies for the injection and extraction well groups.

[0058] The execution and closed-loop control unit translates decisions into actions and makes real-time adjustments based on feedback, driving field actuators to perform repair operations, while continuously monitoring equipment operating status and repair effects, quickly judging and generating strategy adjustment suggestions;

[0059] The data communication and edge computing unit is responsible for the transmission and initial processing of all data and instructions. It employs a hybrid communication network of LoRaWAN and 5G to connect all sensors, actuators, and the cloud platform.

[0060] The energy management and protection unit integrates solar photovoltaic panels, energy storage batteries, and mains power to form a hybrid power supply system. It is responsible for providing power supply and dynamically optimizing energy distribution strategies to ensure the normal operation of equipment under abnormal power supply conditions.

[0061] A third aspect of the present invention provides a computer-readable storage medium comprising a program for an integrated intelligent management and control method for groundwater extraction and injection based on the Internet of Things (IoT). When the program is executed by a processor, it implements the steps of the integrated intelligent management and control method for groundwater extraction and injection based on the IoT as described in any of the preceding claims. Attached Figure Description

[0062] To more clearly illustrate the technical solutions in the embodiments or examples of the present invention, the drawings used in the embodiments or examples will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained according to these drawings without creative effort.

[0063] Figure 1 A flowchart of the first method of an integrated intelligent control method for groundwater extraction and injection based on the Internet of Things provided in an embodiment of the present invention;

[0064] Figure 2 A flowchart of the second method of an integrated intelligent control method for groundwater extraction and injection based on the Internet of Things provided in an embodiment of the present invention;

[0065] Figure 3 A block diagram of an integrated intelligent control system for groundwater extraction and injection based on the Internet of Things provided in an embodiment of the present invention;

[0066] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0067] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0068] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0069] Figure 1 A flowchart of the first method of an integrated intelligent control method for groundwater extraction and injection based on the Internet of Things provided in an embodiment of the present invention;

[0070] like Figure 1 As shown, the present invention provides a first method flowchart of an integrated intelligent control method for groundwater extraction and injection based on the Internet of Things, comprising:

[0071] S102, Conduct geological surveys of the target area and construct a three-dimensional geological model of the area. Use the three-dimensional geological model of the area to plan the layout of the injection well group and heterogeneous sensor monitoring network, and obtain the regional injection and remediation layout scheme.

[0072] S104. Based on the regional injection and remediation layout scheme, deploy a heterogeneous sensor monitoring network and collect regional data to obtain spatial distribution data and water quality monitoring data of regional groundwater, and integrate them with the regional three-dimensional geological model to establish a digital twin of the contaminated site.

[0073] S106, extract groundwater status data of the target area through the digital twin of the contaminated site and set geological condition constraints, use LSTM prediction model to predict pollutant migration path and water level dynamics, and output pollution scenario simulation information corrected by geological characteristics.

[0074] S108, using the pollution scenario simulation information as input, coupled with the permeability coefficient and fracture distribution parameters as constraints, multi-objective collaborative optimization calculation is performed through the BLMFO two-layer multi-objective optimization algorithm to generate a set of collaborative control strategies for the injection-extraction well group;

[0075] S110, based on the collaborative control strategy set of the injection-extraction well group, a control command sequence is generated and sent to the corresponding execution mechanism for execution. During the injection-extraction repair process, the real-time operating status is monitored and it is determined whether the repair strategy needs to be changed. The repair strategy is generated as a suggestion and pushed out.

[0076] Furthermore, in a preferred embodiment of the present invention, the step of conducting geological surveys of the target area and constructing a three-dimensional geological model of the area, and using the three-dimensional geological model of the area to plan the layout of the injection well group and heterogeneous sensor monitoring network, to obtain a regional injection remediation layout scheme, specifically includes:

[0077] By drilling at gridded locations, undisturbed soil and rock samples at different depths were obtained and geophysical interpretation was performed. Remote sensing images and historical geological maps of the target area were collected as supplementary data. At the same time, layered hydrogeological parameters were obtained through field pumping tests, water injection tests, and water pressure tests. A regional geological exploration dataset containing spatial location, stratigraphic properties, and hydrogeological parameters was constructed.

[0078] Based on the regional geological exploration dataset, the Kriging interpolation algorithm is used to spatially interpolate discrete borehole lithology data and a three-dimensional geological structure framework model including stratigraphic lithology, tectonic faults and weathering zones is constructed by combining deterministic modeling methods. The sequential indicator simulation algorithm is used to perform random simulation of the spatial distribution of lithology in the three-dimensional geological structure framework model, and finally a regional three-dimensional geological model representing the geometric morphology of stratigraphic interfaces and the spatial heterogeneity of lithological parameters is obtained.

[0079] The three-dimensional geological model of the region was imported into the groundwater simulation software. The Delaunay triangulation method was used to discretize the unstructured grid to generate a finite volume computational grid suitable for complex geological boundary conditions. Each grid cell was assigned corresponding hydrogeological parameters, and hydraulic boundary conditions and source-sink terms were set.

[0080] The MODFLOW module is called to solve the control equations for groundwater flow, obtaining the streamline distribution and steady-state velocity field of the target area. Driven by the steady-state velocity field, the spatiotemporal migration and evolution process of pollutants under the control of complex geological structures is analyzed by solving the convection-diffusion-response equations, identifying the diffusion path of the pollution plume and potential high-concentration accumulation areas, and obtaining the simulation results of pollutant migration.

[0081] It should be noted that, based on the systematic geological exploration of the target area, undisturbed soil and rock samples at different depths were first obtained through gridded drilling. Simultaneously, geophysical exploration methods such as high-density resistivity analysis and seismic tomography were used to interpret the underground spatial structure, and remote sensing images and historical geological maps of the area were collected as spatial background data. On this basis, key hydrogeological parameters such as permeability coefficient, specific yield, and storage coefficient of each aquifer were obtained through stratified field pumping, injection, and pressure tests. Finally, the multi-source heterogeneous data were integrated to construct a standardized regional geological exploration dataset containing precise spatial location, complete stratigraphic properties, and hydrogeological parameters. Based on this regional geological exploration dataset, the Kriging spatial interpolation algorithm was used to spatially interpolate the lithological classification data from discrete boreholes, generating a probability distribution model of the main strata. Then, deterministic modeling methods were combined with the fault and weathering zone boundaries interpreted from geophysical exploration to construct a structural framework model containing the three-dimensional spatial distribution of stratigraphic lithology, tectonic faults, and weathering zones. Subsequently, a sequential indicator simulation algorithm was used to stochastically simulate the spatial distribution of lithology within the three-dimensional geological structure framework. The spatial correlation of lithological parameters was characterized by variogram analysis. Finally, a high-precision regional three-dimensional geological model that can accurately reflect the geometric morphology of stratigraphic interfaces and the spatial heterogeneity of soil and rock parameters was established.

[0082] Subsequently, the established three-dimensional geological model was imported into professional groundwater simulation software. The Delaunay triangulation method was used to discretize the complex geological body into an unstructured grid, generating a finite volume computational grid adapted to irregular boundaries. Each grid cell was assigned parameters extracted from the geological model, such as permeability, porosity, and dispersion. Based on regional hydrogeological conditions, constant head boundaries for rivers, zero flux boundaries for ridgelines, and source-sink terms were set to complete the parameterization of the numerical model. On this basis, the MODFLOW module was used to solve the groundwater flow control equations, obtaining the steady-state flow field characteristics of the target area under the set boundary conditions, including streamline distribution patterns and spatial velocity distribution. Using this steady-state flow field as the driving condition, the spatiotemporal migration and evolution of pollutants under the control of complex geological structures was simulated by solving the convection-dispersion-response equations. The expansion velocity, transport path, and concentration decay law of the pollution plume were quantitatively analyzed, accurately identifying the dominant diffusion channels and potential high-concentration accumulation areas of pollutants. This yielded complete pollutant migration simulation results, providing a scientific basis for subsequent remediation scheme formulation.

[0083] Furthermore, in a preferred embodiment of the present invention, the step of conducting geological surveys of the target area and constructing a three-dimensional geological model of the area, and using the three-dimensional geological model of the area to plan the layout of the injection well group and heterogeneous sensor monitoring network to obtain a regional injection remediation layout scheme, further includes:

[0084] Based on the results of pollutant migration simulation, an optimization model is established with the objectives of maximizing the pollution capture rate and minimizing the construction cost. A multi-objective particle swarm optimization algorithm is used to randomly generate well site populations within the simulated pollution plume range. The pollution interception efficiency is evaluated by calculating the overlap volume between the well site capture zone and the pollution plume. The drilling cost is estimated by combining the well depth and formation lithology. After iterative calculation, the optimal solution set is output to generate the injection well group layout scheme.

[0085] The current frontal position, main migration path, and boundary of the hydraulic capture zone of the pollution plume are extracted from the pollutant migration simulation results. The differential deployment design of heterogeneous sensors is then carried out in conjunction with the injection well group layout scheme.

[0086] Distributed fiber optic sensors based on Brillouin optical time-domain reflectometers are laid along the line. The deployment trajectory is preferentially selected along the simulated and predicted pollution center axis or a cross section perpendicular to the groundwater flow direction. By monitoring the frequency shift of the backscattered light of the fiber optic cable, continuous temperature or strain information along the sensing cable is obtained to interpret the continuous changes in water quality.

[0087] For potential pollutant accumulation areas, areas with abrupt changes in hydrogeological conditions, and key verification points for remediation effectiveness, miniature spectral sensors are deployed. The specific installation locations are determined by performing grid calculations on the simulated concentration field using spatial interpolation algorithms, and the grid nodes with the largest concentration gradient changes are selected as the optimal deployment points.

[0088] The final solution is a regional injection repair layout. The communication distance between sensors is adjusted through network connectivity analysis to ensure that all sensor node data can be reliably transmitted to the converged edge computing node and cloud server via LoRaWAN or wired connection.

[0089] It should be noted that, based on the obtained pollutant migration simulation results, a multi-objective optimization model is established with the core objectives of maximizing the pollutant capture rate and minimizing construction costs. This model employs a multi-objective particle swarm optimization algorithm, randomly generating an initial well site population within the simulated spatial distribution range of the pollutant plume. The interception efficiency is evaluated by calculating the overlap volume between the hydraulic capture zone and the spatial distribution of the pollutant plume at each well site. Simultaneously, a cost estimation model is established by combining drilling depth and lithological and mechanical parameters of the strata traversed. After multiple iterative calculations, a Pareto solution set that achieves the optimal balance between interception efficiency and construction costs is output, thus generating a scientifically reasonable spatial layout scheme for the injection and extraction well clusters. Based on the determined well cluster layout, key characteristic parameters of the pollutant plume are further extracted from the pollutant migration simulation results, including the spatial location of the current pollution front, the direction of the main migration path, and the boundary range of the hydraulic capture zone. These characteristic parameters are then spatially overlaid with the determined injection and extraction well cluster layout to form a differentiated deployment scheme for a heterogeneous sensor network. The distributed fiber optic sensing system based on a Brillouin optical time-domain reflectometer is laid along the axis of the pollution center predicted by simulation. Auxiliary sensing lines are arranged at key monitoring sections perpendicular to the groundwater flow direction. By monitoring the frequency shift characteristics of the backscattered light in the sensing fiber in real time, the temperature or strain field information distributed along the fiber optic cable is inverted. Then, through the established calibration relationship, the continuous spatial variation of water quality parameters is interpreted. For potential high-concentration pollutant accumulation areas, abrupt changes in hydrogeological conditions, and key verification nodes for remediation effectiveness identified through numerical simulation, miniature spectroscopic sensors are deployed as point-based high-precision monitoring units. The specific installation locations are determined by rasterizing the simulated concentration field using a spatial interpolation algorithm. The raster nodes with the most significant concentration gradient changes are selected as the optimal deployment points to ensure the capture of key dynamic characteristics of the pollution plume. Ultimately, a complete regional injection and remediation layout scheme was formed. Through network connectivity analysis, the communication distance and data transmission path between sensor nodes were optimized to ensure that all monitoring data can be stably and reliably transmitted to edge computing nodes and cloud data processing centers via LoRaWAN low-power wide area network or wired communication, providing comprehensive data support for subsequent intelligent management and control.

[0090] Furthermore, in a preferred embodiment of the present invention, the step of deploying a heterogeneous sensor monitoring network based on the regional injection-remediation layout scheme and collecting regional data to obtain spatial distribution data and water quality monitoring data of regional groundwater, and fusing them with a regional three-dimensional geological model to establish a digital twin of the contaminated site, specifically includes:

[0091] Obtain the regional injection and repair layout plan, and deploy the physical sensing layer based on the sensor deployment coordinates and communication paths determined in the regional injection and repair layout plan. Lay the distributed optical fiber sensors along the planned main monitoring axis and form a closed loop.

[0092] Miniature spectral sensors are fixedly installed at key monitoring nodes and connected to an independent power supply unit consisting of solar panels and battery packs, as well as a LoRaWAN wireless communication module. The working status of the sensing unit and network connectivity are verified through initialization tests, thus constructing a stable physical sensing layer infrastructure.

[0093] Regional data is collected through the physical sensing layer infrastructure. The distributed optical fiber sensing system emits probe light pulses through the demodulator and collects back Brillouin scattering signals. The original optical signals are processed by the frequency shift analysis algorithm. Combined with the pre-established calibration model, the frequency shift is converted into continuous distribution data of water quality parameters. At the same time, the miniature spectral sensor uses the spectral feature extraction algorithm to analyze the absorption spectrum of the water sample and calculate the pollutant concentration value.

[0094] Edge nodes perform digital filtering and signal enhancement on the collected raw data, and package the processed data with spatial location information and timestamps into transmission data units, which are then sent to the cloud data processing center through a wireless communication network. After receiving the multi-source monitoring data, the cloud platform performs spatiotemporal registration and data fusion.

[0095] A time series alignment algorithm is used to synchronize the measurement data of heterogeneous sensors, eliminating the time series deviation caused by the difference in sampling frequency. A spatial interpolation algorithm is used to generate a spatially continuous concentration distribution field from the concentration measurement values ​​of discrete points. At the same time, the physical quantity measurement values ​​of the distributed optical fiber are inverted into a water quality parameter distribution map using calibration conversion relationship.

[0096] The water quality parameter distribution map is fused with a regional three-dimensional geological model. The two-dimensional water quality parameter field is vertically correlated with the stratigraphic structure and permeability coefficient field in the three-dimensional geological model through geostatistical analysis. The dynamics are then overlaid in the geological structure model as a three-dimensional cloud map using visualization technology. At the same time, the correlation mapping relationship between the sensor network and the monitoring data is established to realize interactive query and analysis of monitoring data and three-dimensional scene.

[0097] It should be noted that, based on the established regional injection and remediation layout plan, distributed fiber optic sensors are laid along the preset main monitoring axis to form a complete closed sensing loop. Simultaneously, miniature spectral sensors are securely installed at key monitoring nodes and connected to an independent power supply system composed of high-efficiency solar panels and high-capacity battery packs. These, along with LoRaWAN low-power wireless communication modules, form an autonomous monitoring unit. After hardware deployment, the operating parameters of each sensing unit and network communication quality are verified through a system initialization test process to ensure a stable and reliable physical sensing layer infrastructure. With the physical sensing layer operating normally, regional data acquisition is initiated. The distributed fiber optic sensing system periodically emits nanosecond-level probe light pulses through a demodulator, simultaneously acquiring the backscattered Brillouin signals generated along the fiber. Frequency shift analysis algorithms are used to process the raw optical data, and combined with a pre-established frequency shift-concentration calibration model, the frequency shift is converted into continuous spatial distribution data of water quality parameters. Meanwhile, the miniature spectral sensors use spectral feature extraction algorithms to analyze the absorption spectrum characteristics of the water sample. By identifying the intensity and position of characteristic absorption peaks, the concentration values ​​of specific pollutants are calculated, forming precise point-based monitoring data.

[0098] The raw monitoring data collected is preprocessed by edge computing nodes. Digital filtering algorithms are used to eliminate environmental noise interference, and signal enhancement technology is employed to improve data quality. The processed valid data, along with corresponding spatial coordinate information and precise timestamps, are combined into standardized transmission data units and transmitted to the cloud data processing center via the LoRaWAN wireless network. After receiving the multi-source monitoring data, the cloud platform first performs spatiotemporal registration to unify data from different sources under the same spatiotemporal reference. Subsequently, the data fusion processing stage begins. Dynamic time warping algorithms are used to synchronize the monitoring sequences of heterogeneous sensors, eliminating timing deviations caused by differences in sampling frequencies. Kriging spatial interpolation algorithms are used to generate a spatially continuous concentration distribution field from discrete concentration measurements. Simultaneously, calibration transformation relationships are used to invert the physical quantity measurements from distributed optical fibers into a water quality parameter distribution map. Finally, the water quality parameter distribution map obtained from the processing is deeply integrated with the regional three-dimensional geological model. The vertical correlation between the two-dimensional water quality parameter field and the stratigraphic structure and permeability coefficient field in the three-dimensional geological model is established through geostatistical analysis. The dynamic monitoring data is overlaid in the geological structure model in the form of a three-dimensional cloud map using visualization technology. The spatial mapping relationship between the sensor network and the monitoring data is established, realizing the interactive query and analysis function between the monitoring data and the three-dimensional scene, providing comprehensive data support for the precise control of contaminated sites.

[0099] Furthermore, in a preferred embodiment of the present invention, the step of extracting groundwater status data of the target area through the digital twin of the contaminated site and setting geological condition constraints, using an LSTM prediction model to predict pollutant migration paths and water level dynamics, and outputting pollution scenario simulation information corrected for geological features, specifically includes:

[0100] By querying the real-time data service interface of the digital twin, we can obtain the latest updated pollutant concentration spatial distribution matrix from the sensor network, the water level time series data of each monitoring point, and the instantaneous flow velocity field vector data obtained through numerical simulation.

[0101] Simultaneously, key geological attribute parameters pre-stored in the grid cells of the regional three-dimensional geological model are extracted, including the permeability tensor, effective porosity scalar, and spatial geometric and connectivity parameters of fracture development zones for each grid cell; the extracted geological parameters are matched and aligned with real-time monitoring data according to spatial coordinates to form an original dataset containing spatiotemporal dynamic data and static geological constraints.

[0102] Based on the original dataset, time-series data of water level and pollutant concentration are fused with corresponding geological parameters using an attention mechanism to generate a hybrid input vector containing temporal and static geological features. This vector is then segmented into fixed-length sample fragments, each containing observations from several past time steps and their corresponding geological features. Simultaneously, time-series data of external environmental factors are obtained using a big data network as an additional feature dimension added to the input vector, constructing a multi-dimensional training sample set with spatiotemporal correlation.

[0103] The multidimensional training sample set is input into the blank LSTM prediction model for supervised training. The input layer receives the time series data of multidimensional sample features, the hidden layer learns the complex long-term dependencies in the groundwater system through the forget gate, input gate and output gate, and the output layer generates the predicted values ​​of pollutant concentration and water level through a fully connected network, and finally obtains the trained LSTM prediction model.

[0104] The input for each time step includes the monitoring data at the current moment and the corresponding static geological parameters. The predicted output is calculated through forward propagation, and the gradient of the loss function with respect to the network weights is calculated through backpropagation using the BPTT algorithm. The network parameters are iteratively updated using an adaptive moment estimation optimizer.

[0105] The trained LSTM model is used for multi-step recursive prediction. The monitoring data at the current time is used as the initial state to predict the pollutant concentration distribution and water level field at the next time step. Then, the prediction results are used as new inputs to recursively predict and generate the spatial distribution sequence of pollutant concentration and the dynamic change sequence of water level at future time steps.

[0106] Geological constraints were applied to the generated spatial distribution sequence of pollutant concentration and dynamic change sequence of water level. Based on the spatial distribution tensor of permeability coefficient, the inverse distance weighted interpolation method was used to spatially register the permeability coefficient field with the predicted concentration field. Using the density field data of fracture development, the optimal path analysis method based on Dijkstra's algorithm was used to identify the dominant transport channels of pollutants in the fracture network. In the fracture development area, the convection-dispersion parameters of pollutants were adjusted according to the fracture density ratio. Finally, the pollution scenario simulation information corrected by geological characteristics was generated.

[0107] It should be noted that multi-source data was extracted from the constructed digital twin of the contaminated site to obtain the spatial distribution of pollutant concentrations and the water level change sequence at each monitoring point, updated in real time by the sensor network, reflecting the current situation. Instantaneous flow velocity field data obtained through numerical simulation was also coupled in. Simultaneously, key static geological attribute parameters extracted from the 3D geological model mesh, such as permeability coefficient, porosity, and geometric and connectivity parameters characterizing fracture channels, injected deep geological constraints into the prediction. These two types of data with different spatiotemporal scales (dynamic monitoring data and static geological parameters) were precisely matched and aligned according to spatial coordinates to form an original dataset that simultaneously contains both "current state" and "background constraints." Next, a training sample set suitable for LSTM model learning was constructed. To effectively utilize temporal information and geological features, an attention mechanism was introduced to fuse time-series data such as water level and pollutant concentration with corresponding static geological parameters, generating a hybrid input vector. Subsequently, a sliding window method was used to segment the continuous time-series data into fixed-length sample segments, each segment containing historical observations over a past period and its corresponding unchanging geological background information. Furthermore, external environmental factors (such as rainfall and evaporation) are incorporated as additional features, ultimately constructing a multidimensional training sample set with rich spatiotemporal correlations. The LSTM model is then trained in a supervised manner using this prepared sample set. The model's input layer receives a multidimensional feature sequence organized by time steps. Its core lies in the hidden layer's autonomous decision, through forget gates, input gates, and output gates, of which historical information needs to be retained or forgotten, and which new information needs to be updated, thereby effectively capturing the long-term dependencies existing in the groundwater system. The output layer generates predicted values ​​through a fully connected network. The training process employs the Backpropagation Through Time (BPTT) algorithm to calculate the gradient of the loss function, combined with an Adaptive Moment Estimator (Adam) optimizer to iteratively update the network parameters, thereby obtaining a predictive model that accurately simulates the system's behavior.

[0108] The trained model undergoes multi-step recursive prediction, using the latest monitoring data as the initial state to predict pollutant concentration and water level distribution for the next time step. This predicted value is then used as input to predict even more future time steps, and so on recursively, generating a complete spatial distribution sequence of pollutant concentration and dynamic water level change sequence over a future period. Finally, geological constraint correction is applied to the purely data-driven LSTM prediction results. Since the LSTM model may only learn patterns from the data and ignore physical constraints, this step aims to ensure that the prediction results are not only statistically accurate but also conform to the basic physical and geological principles of groundwater movement. Specifically, based on the spatial distribution data of permeability coefficients, the predicted concentration field is spatially registered with the geological attribute field to analyze the impact of permeability differences on pollutant transport; using fracture development data, the shortest path algorithm in graph theory (such as Dijkstra's algorithm) is used to identify the most likely dominant transport channels for pollutants in the fracture network, and the parameters of convection-diffusion in the model are dynamically adjusted according to fracture density in fracture-developed areas. The final pollution scenario simulation information not only inherits the adaptability of the data-driven model, but also enhances its physical rationality and geological credibility, making its pollution prediction in complex geological environments such as fractured aquifers more reliable.

[0109] Furthermore, in a preferred embodiment of the present invention, the step of generating a control command sequence based on the injection-extraction well group collaborative control strategy set and sending it to the corresponding execution mechanism for execution, monitoring the real-time operating status during the injection-extraction repair process and determining whether a repair strategy change is needed, generating a repair strategy as a suggestion and pushing it out, specifically includes:

[0110] Based on the operating parameters and target set values ​​of each well contained in the coordinated control strategy set of the injection and extraction well group, the continuous control parameters in the strategy set are transformed into a sequence of instructions with temporal relationships through time discretization processing. Each instruction contains an equipment identifier, execution timestamp, control parameter set value and expected effect index. The executability and temporal rationality of each instruction within the equipment's capability range are verified by the instruction logic verification algorithm to form a verified control instruction sequence.

[0111] The verified control command sequence is sent to the corresponding actuators through the Internet of Things communication protocol. Each actuator immediately returns a confirmation signal after receiving the command and starts the command execution process. At the same time, the actual operating parameters of the equipment are collected in real time through built-in sensors, forming an execution feedback data stream that includes instantaneous flow rate, actual dosage of reagents and equipment operating status.

[0112] During the execution of the command, multi-dimensional operational status monitoring is initiated simultaneously. Groundwater dynamic data, including water level change sequences, pollutant concentration distribution data, and equipment operating status parameters, are collected in real time through a heterogeneous sensor network deployed in the injection well group. After data preprocessing, a high-quality real-time operational status dataset is generated.

[0113] Dynamic evaluation of remediation effectiveness is carried out based on real-time operational status dataset. The actual monitored pollutant concentration distribution is compared and analyzed with the remediation trajectory expected in the strategy set. The deviation between the actual and expected values ​​of remediation performance indicators is calculated. At the same time, by analyzing equipment operation status data, the workload and operating efficiency of each execution unit are evaluated, equipment performance degradation or abnormal operation are identified, and a regional remediation status assessment report is generated.

[0114] Based on the regional restoration status assessment report, a process for determining the necessity of strategy changes was initiated. A judgment logic based on multi-dimensional thresholds was established. When the actual rate of decrease in pollutant concentration is lower than the set percentage of the expected value, or when the pollutant concentration at key points shows a rebound trend, or when the equipment operating efficiency is continuously lower than the rated efficiency threshold, it is determined that a strategy change is necessary.

[0115] When a strategy change is deemed necessary, based on real-time operational characteristics and effect evaluation results, effective adjustment strategies under similar operating conditions are retrieved from the historical case database using a case-based reasoning method. Combined with the current equipment operating status and geological constraints, the generated flow adjustment range, reagent ratio optimization scheme, and equipment operating parameter correction values ​​are used to generate a repair strategy change suggestion, which is then pushed to the management terminal.

[0116] It should be noted that the generated collaborative control strategy set is transformed into specific instructions executable by field devices through time discretization processing. Continuous strategies are decomposed into instruction sequences with strict temporal relationships, and each instruction is appended with detailed information such as device identifier, execution time, setpoint, and expected effect. Subsequently, an instruction logic verification algorithm verifies whether these instructions are within the physical capabilities of the equipment and whether there are any temporal conflicts between instructions, ensuring the safety and reliability of the execution process. Verified instruction sequences are sent to various actuators (such as water pumps and dosing pumps) via IoT protocols. Upon receiving the instructions, the equipment immediately returns an acknowledgment signal and begins execution. Simultaneously, built-in sensors collect real-time operating parameters such as flow rate, dosage, and energy consumption, forming an execution feedback data stream. The significance of this real-time feedback mechanism lies in establishing a precise mapping between "instruction and execution," enabling the system to accurately know whether the instructions are executed correctly and the actual execution effect. This provides real and timely first-hand data for subsequent effect evaluation, avoiding the misjudgment of "strategy failure" due to "execution deviation" caused by equipment failure or performance degradation.

[0117] Simultaneously with command execution, multi-dimensional operational status monitoring is initiated. A sensor network deployed on-site collects dynamic groundwater data and combines this data with equipment operating status parameters to generate a high-quality real-time operational status dataset. Based on this dataset, dynamic evaluation of remediation effectiveness is performed. The actual monitored pollutant concentration distribution is compared with the expected trajectory of the strategy, calculating deviations in remediation performance indicators (such as pollutant concentration deviation rate, water level control error, etc.), and assessing equipment workload and efficiency. The "execution results" are quantitatively compared with the "expected goals," objectively answering the two key questions of "whether the current strategy is effective" and "whether the equipment is in a healthy working state," providing data support for decision-making. Based on the regional remediation status report generated by the dynamic evaluation, a determination of the necessity for strategy change is initiated. This determination is based on preset multi-dimensional threshold logic; for example, when the actual pollutant decline rate is significantly lower than expected, concentration rebounds, or equipment efficiency remains excessively low, a strategy change is deemed necessary. This automatic judgment mechanism based on clear rules ensures that the system can promptly detect problems, avoiding remediation efficiency losses or environmental risks caused by delays in manual judgment. When a change is deemed necessary, the system uses a case-based reasoning approach to retrieve proven effective adjustment strategies from a historical case database under similar operating conditions. Combined with the current equipment status and geological constraints, it generates change suggestions including specific measures such as flow rate adjustments and reagent ratio optimization. This allows for faster identification of effective adjustment solutions, enhancing the system's decision-making intelligence and practicality. Finally, the generated repair strategy change suggestions are pushed to the management terminal, driving the system into the next "decision-execution-evaluation" optimization cycle, achieving closed-loop intelligent control.

[0118] Figure 2 A flowchart of the second method of an integrated intelligent control method for groundwater extraction and injection based on the Internet of Things provided in an embodiment of the present invention;

[0119] like Figure 2 As shown, the present invention provides a second method flowchart for an integrated intelligent control method for groundwater extraction and injection based on the Internet of Things, comprising:

[0120] S202, construct the upper-level optimization model of the BLMFO two-level multi-objective optimization algorithm, take the minimization of the pollution plume range, the minimization of system operating energy consumption and the equalization of equipment usage load as three conflicting optimization objectives, take the flow distribution scheme of each injection well, the operating time series and the reagent dosage concentration as decision variables, and establish a mathematical model of the upper-level optimization problem containing multi-objective functions and basic constraints.

[0121] S204, based on the established upper-level optimization problem model, constructs a decision variable space through a preset three-level coding strategy according to the real-time operation control strategy and operation monitoring data of the injection well group in the target area, and introduces an improved artificial bee colony optimization algorithm for upper-level solution calculation. It combines random selection and strategy selection for honey source initialization, constructs the initial Pareto optimal frontier through iterative optimization analysis, and generates candidate control strategies.

[0122] S206, the candidate control strategy is passed to the lower-level optimization model. In the lower-level model, a response surface model under geological constraints is established based on the spatial variation characteristics of the permeability coefficient and the fracture network connectivity parameters. The geological feasibility of each candidate strategy passed from the upper level is verified, and the verification result is fed back to the upper-level optimization process in the form of constraint violation degree.

[0123] S208, based on the geological constraint verification results fed back by the lower-level model, the upper-level optimization algorithm performs a co-evolution operation, uses dynamically adjusted inertia weights and learning factors to update the velocity and position of particles, eliminates inferior solutions that do not meet the geological conditions through a tournament selection mechanism, and outputs the optimal solution set that achieves the best balance among multiple optimization objectives and meets the geological constraints after a preset number of double-layer iterations, generating a set of collaborative control strategies for the injection and extraction well group.

[0124] It should be noted that the upper-level optimization model focuses on minimizing the pollution plume range, minimizing system operating energy consumption, and balancing equipment load. Operable parameters such as the flow rate, operating time, and reagent concentration of each injection well are set as decision variables. In the model solution phase, an improved artificial bee colony algorithm is employed. This algorithm's advantages lie in its powerful global search capability and good population diversity. Through a three-level coding strategy, the complex injection well scheduling problem is mapped into a solution space that the algorithm can handle. By employing local development by hired bees, focusing on key areas by follower bees, and conducting global exploration by scout bees, the algorithm iteratively approaches the Pareto optimal front, generating a batch of candidate control strategies that achieve a balance among multiple objectives. Subsequently, a lower-level geological feasibility verification step is performed. The upper-level optimization mainly focuses on system-level objectives and may neglect geological details. Therefore, the candidate strategies generated in the upper level need to be passed to the lower-level model. The lower-level model constructs a response surface model based on accurate geological parameters (such as spatial variation of permeability and fracture connectivity) to quickly simulate and verify each candidate strategy, calculating its "geological constraint violation degree." This approach injects physical constraints into the mathematical optimization problem, ensuring that the strategy is not only numerically optimal but also feasible and safe under actual geological conditions. This avoids the predicament of a theoretically perfect solution being unfeasible due to geological limitations. Finally, based on the verification results from the lower-level feedback, the upper-level algorithm performs a co-evolutionary operation. Through dynamic parameter adjustment and a tournament selection mechanism, the algorithm retains high-quality solutions that satisfy both multi-objective optimization and geological constraints, while eliminating inferior solutions. After a predetermined number of two-level iterations, the final output is a set of collaborative control strategies for the injection-extraction well group. The significance of this two-level iterative mechanism lies in its realization of a closed loop between "macro-optimization" and "micro-verification," ensuring that the final generated strategy set possesses both global optimality and field feasibility, significantly improving the success rate and economy of the remediation project.

[0125] Furthermore, in a preferred embodiment of the present invention, the step of constructing a decision variable space through a preset three-level coding strategy, introducing an improved artificial bee colony optimization algorithm for upper-level solution calculation, initializing nectar sources by combining random selection and strategy selection, constructing an initial Pareto optimal frontier through iterative optimization analysis, and generating candidate control strategies specifically includes:

[0126] The three-level coding strategy consists of injection well codes, process codes, and monitoring codes. The injection well codes include the location and functional attributes of each injection well. The process codes include the running sequence and running strategy of the corresponding injection well. The monitoring codes are the running monitoring status characteristics of the corresponding injection well.

[0127] During the bee-hiring phase, a uniform crossover strategy is used to perform a neighborhood search on the current nectar source. A solution is randomly selected from the Pareto archive and swapped with the current solution for the code fragment. At the same time, a new candidate solution is generated at the machine coding layer using the strategy of minimizing cumulative processing time. The Pareto dominance principle is used to select a better solution from the old and new solutions and update the archive.

[0128] In the follower bee phase, the following probability of all nectar sources is calculated based on the non-dominance level and crowding of the nectar source. A roulette wheel method is used to select nectar sources that need to be searched further. The search strategy is consistent with that in the hired bee phase to ensure in-depth mining around high-quality solutions.

[0129] When the nectar source reaches the search limit but the quality does not improve, the scout bee uses a random exchange strategy to reconstruct the well group allocation code, randomly selects some well groups to be redistributed to other processing units, while maintaining the overall processing capacity balance, and regenerates the scheduling order and equipment parameters according to the initialization scheme, thereby increasing the population diversity.

[0130] During the iterative optimization process, the non-dominated solutions are recorded as the Pareto optimal solution set by updating the archive set. After each iteration, the current population is sorted by fast non-dominated sorting. Solutions ranked within the preset range are added to the archive set, and individuals dominated by the new solutions are removed from the archive set. After a preset number of iterations, the initial Pareto optimal frontier is output, and candidate control strategies are generated.

[0131] It should be noted that a three-level coding strategy of "injection well code, process code, and monitoring code" is used to cleverly decompose and abstract the complex problem of injection well scheduling: the injection well code defines the identity attributes and functional roles of the well points, the process code specifies the operating rhythm and logic of these wells, and the monitoring code is associated with the feedback characteristics of their operating status. The significance of this hierarchical coding is that it transforms a large and highly coupled complex problem into three relatively independent but interconnected sub-problem spaces, greatly reducing the search dimensionality and complexity of the algorithm, enabling the algorithm to explore the solution space more effectively, and also making the generated solutions easier to understand and map to actual control commands.

[0132] In the hired bee phase, the algorithm focuses on local depth search to discover excellent genes in the neighborhood of high-quality solutions. It employs a uniform crossover strategy, randomly selecting a non-dominated solution from the Pareto archive and exchanging its encoded fragments (such as process code fragments) with the current solution. This mechanism is similar to gene recombination in biological genetics, effectively exploring the possible combinations of features from different excellent solutions. Simultaneously, at the machine allocation level, a strategy that minimizes cumulative processing time is used to generate new solutions, directly addressing the optimization goal of improving equipment operating efficiency through heuristic improvements. Subsequently, based on the Pareto dominance principle, the archive is updated by selecting the best solutions from the old and new sources. Through controlled local perturbations and greedy selection, the quality of solutions is rapidly improved, and the diversity of the Pareto front is enriched. In the follower bee phase, the algorithm calculates the following probability based on the non-dominated rank and crowding distance of the nectar source and uses a roulette wheel approach to select nectar sources for in-depth mining. The higher the non-dominated level (the smaller the Rank value) and the greater the crowding, the greater the probability of the solution being selected. Its search strategy is consistent with the hired bee stage. The purpose is to conduct a more refined search on these preferred regions, thereby ensuring that a local optimum solution is found in the region. This achieves a reasonable allocation of computing resources and balances the breadth and depth search capabilities of the algorithm.

[0133] When a nectar source shows no improvement after multiple searches, it indicates the algorithm may be stuck in a local optimum, at which point the scout bee mechanism is activated. It employs a random exchange strategy to reconstruct the well group assignment encoding and reinitialize relevant parameters, essentially introducing a completely new random solution into the current population. This strategy provides the algorithm with the crucial ability to escape local optima and stagnation, significantly enhancing population diversity by introducing randomness, thus ensuring the algorithm's continuous global exploration capability and preventing premature convergence. Throughout the iteration process, an external archive is used to dynamically maintain and update the currently found Pareto optimal solution set. After each iteration, the merged population is stratified using fast non-dominated sorting, and the first-rank (Rank 1) non-dominated solutions are added to the archive. Simultaneously, a crowding comparison operator is used to remove dominated individuals while maintaining the distribution of the solution set. After a predetermined number of iterations, the converged solution set becomes the initial Pareto optimal front.

[0134] Furthermore, the present invention provides an integrated intelligent management and control method for groundwater extraction and injection based on the Internet of Things, which further includes the following steps:

[0135] The integrated pumping and injection equipment system integrates a solar energy storage module consisting of a photovoltaic panel array, a lithium-ion battery pack, and an intelligent bidirectional converter. It optimizes the power generation efficiency of the photovoltaic panels in real time through maximum power point tracking technology and uses a battery management system to monitor the state of charge, health status, and remaining capacity parameters of the battery. It establishes an energy data acquisition system that includes solar power generation prediction, energy storage status monitoring, and grid electricity price information.

[0136] An initial energy dataset is generated by acquiring meteorological data, irradiance monitoring values, and photovoltaic power generation data within a preset period through an energy data acquisition system. Based on the initial energy dataset, the meteorological data and irradiance monitoring values ​​are fused to generate a fused feature sequence.

[0137] The fused feature sequence is divided into several feature sequence segments according to a preset time step. Photovoltaic power generation data is extracted from the initial energy dataset and a photovoltaic power generation sequence is generated. This sequence is then associated with the feature sequence segments to form an associated feature sequence. The associated feature sequence represents the relationship between environmental state and power generation within a monitoring period.

[0138] The associated feature sequence sets are generated from the associated feature sequences corresponding to multiple monitoring cycles. A monitoring cycle is divided into several time nodes. A state space is constructed through the associated feature sequence sets. The state transition probabilities and state transition matrices between different time nodes are calculated in the state space. Time nodes with similarity to meteorological data and irradiance monitoring values ​​are merged through similarity calculation to finally obtain an energy feature sample set.

[0139] Based on the energy feature sample set, each time step of a preset size is defined as a single node. The corresponding meteorological features and irradiance monitoring values ​​are used as the node representations of the corresponding nodes, and photovoltaic power generation is used as the node attribute. Several power generation status paths with a monitoring period as the time length are generated. The power generation status path represents the different photovoltaic power generation of different meteorological features and irradiance monitoring values ​​within a monitoring period.

[0140] A power generation situation prediction model is established, and the power generation situation prediction model is trained through the power generation situation path. After the model training is completed, meteorological data and irradiance monitoring values ​​obtained in real time are input to obtain the power generation situation prediction result at the current moment.

[0141] The predicted power generation situation is compared with a preset threshold. If the result is greater than the preset threshold, it is determined that a hybrid power supply strategy can be enabled. The deviation from the preset threshold is calculated to generate available energy information. Combined with the load demand curve and real-time electricity price information, a multi-objective optimization algorithm is used to formulate and control the hybrid power supply strategy.

[0142] It should be noted that the solar energy storage module is formed by integrating photovoltaic panels, batteries, and intelligent converters into the injection equipment system. Maximum power point tracking (MPPT) technology is used to improve the energy conversion efficiency of the photovoltaic panels, while the battery management system comprehensively monitors the health status and remaining power of the energy storage unit. In the data preprocessing and feature engineering stage, the collected meteorological, irradiance, and power generation data are fused to generate a feature sequence characterizing the relationship between the environment and power generation. Subsequently, by analyzing historical data from multiple monitoring periods, a state space is constructed and state transition probabilities are calculated. Time nodes with similar meteorological patterns are merged to ultimately generate a high-quality energy feature sample set. The variation patterns of power generation under different weather conditions are extracted from historical data, providing training samples that reflect the inherent causal relationships for subsequent prediction models.

[0143] In the model building and training phase, the concept of a "power generation status path" was proposed. This path links meteorological characteristics, irradiance, and corresponding power generation at different points in time within a complete monitoring cycle, forming a path that evolves over time. The power generation status prediction model is trained using this path, enabling it to learn the dynamic impact of continuously changing meteorological conditions on power generation, rather than simply an isolated point-to-point relationship. The trained model can predict future power generation status based on real-time monitored meteorological and irradiance data. Finally, in the strategy formulation and optimization phase, the prediction results are compared with preset thresholds to determine whether solar energy is sufficient to support system operation while retaining emergency energy. If the conditions are met, the available solar energy is further calculated, and combined with the load demand curve of the injection equipment and the time-of-use electricity price information of the grid, a multi-objective optimization algorithm is used to formulate the optimal hybrid power supply strategy. This strategy dynamically determines when and in what proportion to use solar energy, battery storage, or grid power. Its fundamental significance lies in achieving the optimal balance between energy economy, reliability, and cleanliness, ensuring that the integrated injection equipment can achieve efficient, low-carbon, and low-cost long-term operation under complex and changing conditions.

[0144] Figure 3 An embodiment of the present invention provides an integrated intelligent control system for groundwater extraction and injection based on the Internet of Things. The system includes: a geological sensing and digital twin unit, an intelligent prediction and decision-making unit, an execution and closed-loop control unit, a data communication and edge computing unit, and an energy management and security unit.

[0145] The geological sensing and digital twin unit integrates geological survey data and real-time sensor network monitoring data to construct and continuously update a digital twin of the contaminated site, providing a comprehensive perception of the underground environment.

[0146] The intelligent prediction and decision-making unit predicts pollutant migration and water level dynamics based on data provided by the digital twin, and performs corrections by coupling geological constraints and formulates the optimal set of collaborative control strategies for the injection and extraction well groups.

[0147] The execution and closed-loop control unit translates decisions into actions and makes real-time adjustments based on feedback, driving field actuators to perform repair operations, while continuously monitoring equipment operating status and repair effects, quickly judging and generating strategy adjustment suggestions;

[0148] The data communication and edge computing unit is responsible for the transmission and initial processing of all data and instructions. It employs a hybrid communication network of LoRaWAN and 5G to connect all sensors, actuators, and the cloud platform.

[0149] The energy management and protection unit integrates solar photovoltaic panels, energy storage batteries, and mains power to form a hybrid power supply system. It is responsible for providing power supply and dynamically optimizing energy distribution strategies to ensure the normal operation of equipment under abnormal power supply conditions.

[0150] A third aspect of the present invention provides a computer-readable storage medium comprising a program for an integrated intelligent management and control method for groundwater extraction and injection based on the Internet of Things (IoT). When the program is executed by a processor, it implements the steps of the integrated intelligent management and control method for groundwater extraction and injection based on the IoT as described in any of the preceding claims.

[0151] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A smart management and control method for integrated groundwater pumping and injection based on the Internet of Things, characterized in that, include: Geological surveys were conducted on the target area and a three-dimensional geological model of the area was constructed. The three-dimensional geological model of the area was then used to plan the layout of the injection well group and the heterogeneous sensor monitoring network, resulting in a regional injection and remediation layout scheme. Based on the aforementioned regional injection and remediation layout scheme, a heterogeneous sensor monitoring network is deployed to collect regional data, obtain spatial distribution data and water quality monitoring data of regional groundwater, and integrate them with the regional three-dimensional geological model to establish a digital twin of the contaminated site. The groundwater status data of the target area is extracted by the digital twin of the contaminated site and geological constraints are set. The LSTM prediction model is used to predict the migration path of pollutants and the dynamic water level, and the pollution scenario simulation information corrected by geological characteristics is output. Using the pollution scenario simulation information as input, coupled with the permeability coefficient and fracture distribution parameters as constraints, the BLMFO two-layer multi-objective optimization algorithm is used to perform multi-objective collaborative optimization calculations to generate a set of collaborative control strategies for the injection and extraction well group. Based on the aforementioned collaborative control strategy set for the injection and extraction well group, a sequence of control commands is generated and sent to the corresponding execution mechanism for execution. During the injection and extraction repair process, the real-time operating status is monitored and it is determined whether a repair strategy change is required. A repair strategy change suggestion is generated and pushed out. The integrated pumping and injection equipment system integrates a solar energy storage module consisting of a photovoltaic panel array, a lithium-ion battery pack, and an intelligent bidirectional converter. It optimizes the power generation efficiency of the photovoltaic panels in real time through maximum power point tracking technology and uses a battery management system to monitor the state of charge, health status, and remaining capacity parameters of the battery. It establishes an energy data acquisition system that includes solar power generation prediction, energy storage status monitoring, and grid electricity price information. An initial energy dataset is generated by acquiring meteorological data, irradiance monitoring values, and photovoltaic power generation data within a preset period through an energy data acquisition system. Based on the initial energy dataset, the meteorological data and irradiance monitoring values ​​are fused to generate a fused feature sequence. The fused feature sequence is divided into several feature sequence segments according to a preset time step. Photovoltaic power generation data is extracted from the initial energy dataset and a photovoltaic power generation sequence is generated. This sequence is then associated with the feature sequence segments to form an associated feature sequence. The associated feature sequence represents the relationship between environmental state and power generation within a monitoring period. The associated feature sequence sets are generated from the associated feature sequences corresponding to multiple monitoring cycles. A monitoring cycle is divided into several time nodes. A state space is constructed through the associated feature sequence sets. The state transition probabilities and state transition matrices between different time nodes are calculated in the state space. Time nodes with similarity to meteorological data and irradiance monitoring values ​​are merged through similarity calculation to finally obtain an energy feature sample set. Based on the energy feature sample set, each time step of a preset size is defined as a single node. The corresponding meteorological features and irradiance monitoring values ​​are used as the node representations of the corresponding nodes, and photovoltaic power generation is used as the node attribute. Several power generation status paths with a monitoring period as the time length are generated. The power generation status path represents the different photovoltaic power generation of different meteorological features and irradiance monitoring values ​​within a monitoring period. A power generation situation prediction model is established, and the power generation situation prediction model is trained through the power generation situation path. After the model training is completed, meteorological data and irradiance monitoring values ​​obtained in real time are input to obtain the power generation situation prediction result at the current moment. The predicted power generation situation is compared with a preset threshold. If the result is greater than the preset threshold, it is determined that a hybrid power supply strategy can be enabled. The deviation from the preset threshold is calculated to generate available energy information. Combined with the load demand curve and real-time electricity price information, a multi-objective optimization algorithm is used to formulate and control the hybrid power supply strategy.

2. The integrated intelligent control method for groundwater extraction and injection based on the Internet of Things as described in claim 1, characterized in that, The process involves conducting geological surveys of the target area and constructing a three-dimensional geological model of the area. Using this model, a layout plan for the injection well cluster and heterogeneous sensor monitoring network is developed to obtain a regional injection remediation layout scheme. Specifically, this includes: By drilling at gridded locations, undisturbed soil and rock samples at different depths were obtained and geophysical interpretation was performed. Remote sensing images and historical geological maps of the target area were collected as supplementary data. At the same time, layered hydrogeological parameters were obtained through field pumping tests, water injection tests, and water pressure tests. A regional geological exploration dataset containing spatial location, stratigraphic properties, and hydrogeological parameters was constructed. Based on the regional geological exploration dataset, the Kriging interpolation algorithm is used to spatially interpolate discrete borehole lithology data and a three-dimensional geological structure framework model including stratigraphic lithology, tectonic faults and weathering zones is constructed by combining deterministic modeling methods. The sequential indicator simulation algorithm is used to perform random simulation of the spatial distribution of lithology in the three-dimensional geological structure framework model, and finally a regional three-dimensional geological model representing the geometric morphology of stratigraphic interfaces and the spatial heterogeneity of lithological parameters is obtained. The three-dimensional geological model of the region was imported into the groundwater simulation software. The Delaunay triangulation method was used to discretize the unstructured grid to generate a finite volume computational grid suitable for complex geological boundary conditions. Each grid cell was assigned corresponding hydrogeological parameters, and hydraulic boundary conditions and source-sink terms were set. The MODFLOW module is called to solve the control equations for groundwater flow, obtaining the streamline distribution and steady-state velocity field of the target area. Driven by the steady-state velocity field, the spatiotemporal migration and evolution process of pollutants under the control of complex geological structures is analyzed by solving the convection-diffusion-response equations, identifying the diffusion path of the pollution plume and potential high-concentration accumulation areas, and obtaining the simulation results of pollutant migration.

3. The integrated intelligent control method for groundwater extraction and injection based on the Internet of Things as described in claim 1, characterized in that, The process of conducting geological surveys of the target area and constructing a three-dimensional geological model of the area, using the three-dimensional geological model to plan the layout of injection well groups and heterogeneous sensor monitoring networks, and obtaining a regional injection remediation layout scheme, also includes: Based on the results of pollutant migration simulation, an optimization model is established with the objectives of maximizing the pollution capture rate and minimizing the construction cost. A multi-objective particle swarm optimization algorithm is used to randomly generate well site populations within the simulated pollution plume range. The pollution interception efficiency is evaluated by calculating the overlap volume between the well site capture zone and the pollution plume. The drilling cost is estimated by combining the well depth and formation lithology. After iterative calculation, the optimal solution set is output to generate the injection well group layout scheme. The current frontal position, main migration path, and boundary of the hydraulic capture zone of the pollution plume are extracted from the pollutant migration simulation results. The differential deployment design of heterogeneous sensors is then carried out in conjunction with the injection well group layout scheme. Distributed fiber optic sensors based on Brillouin optical time-domain reflectometers are laid along the line. The deployment trajectory is preferentially selected along the simulated and predicted pollution center axis or a cross section perpendicular to the groundwater flow direction. By monitoring the frequency shift of the backscattered light of the fiber optic cable, continuous temperature or strain information along the sensing cable is obtained to interpret the continuous changes in water quality. For potential pollutant accumulation areas, areas with abrupt changes in hydrogeological conditions, and key verification points for remediation effectiveness, miniature spectral sensors are deployed. The specific installation locations are determined by performing grid calculations on the simulated concentration field using spatial interpolation algorithms, and the grid nodes with the largest concentration gradient changes are selected as the optimal deployment points. The final solution is a regional injection repair layout. The communication distance between sensors is adjusted through network connectivity analysis to ensure that all sensor node data can be reliably transmitted to the converged edge computing node and cloud server via LoRaWAN or wired connection.

4. The integrated intelligent control method for groundwater extraction and injection based on the Internet of Things as described in claim 1, characterized in that, The deployment of a heterogeneous sensor monitoring network based on the regional injection and remediation layout scheme, and the acquisition of regional groundwater spatial distribution data and water quality monitoring data, are then fused with a regional three-dimensional geological model to establish a digital twin of the contaminated site. Specifically, this includes: Obtain the regional injection and repair layout plan, and deploy the physical sensing layer based on the sensor deployment coordinates and communication paths determined in the regional injection and repair layout plan. Lay the distributed optical fiber sensors along the planned main monitoring axis and form a closed loop. Miniature spectral sensors are fixedly installed at key monitoring nodes and connected to an independent power supply unit consisting of solar panels and battery packs, as well as a LoRaWAN wireless communication module. The working status of the sensing unit and network connectivity are verified through initialization tests, thus constructing a stable physical sensing layer infrastructure. Regional data is collected through the physical sensing layer infrastructure. The distributed optical fiber sensing system emits probe light pulses through the demodulator and collects back Brillouin scattering signals. The original optical signals are processed by the frequency shift analysis algorithm. Combined with the pre-established calibration model, the frequency shift is converted into continuous distribution data of water quality parameters. At the same time, the miniature spectral sensor uses the spectral feature extraction algorithm to analyze the absorption spectrum of the water sample and calculate the pollutant concentration value. Edge nodes perform digital filtering and signal enhancement on the collected raw data, and package the processed data with spatial location information and timestamps into transmission data units, which are then sent to the cloud data processing center through a wireless communication network. After receiving the multi-source monitoring data, the cloud platform performs spatiotemporal registration and data fusion. A time series alignment algorithm is used to synchronize the measurement data of heterogeneous sensors, eliminating the time series deviation caused by the difference in sampling frequency. A spatial interpolation algorithm is used to generate a spatially continuous concentration distribution field from the concentration measurement values ​​of discrete points. At the same time, the physical quantity measurement values ​​of the distributed optical fiber are inverted into a water quality parameter distribution map using calibration conversion relationship. The water quality parameter distribution map is fused with a regional three-dimensional geological model. The two-dimensional water quality parameter field is vertically correlated with the stratigraphic structure and permeability coefficient field in the three-dimensional geological model through geostatistical analysis. The dynamics are then overlaid in the geological structure model as a three-dimensional cloud map using visualization technology. At the same time, the correlation mapping relationship between the sensor network and the monitoring data is established to realize interactive query and analysis of monitoring data and three-dimensional scene.

5. The integrated intelligent control method for groundwater extraction and injection based on the Internet of Things as described in claim 1, characterized in that, The process involves extracting groundwater status data of the target area using the digital twin of the contaminated site, setting geological constraints, and using an LSTM prediction model to predict pollutant migration paths and water level dynamics. The output is then adjusted for geological characteristics to simulate the pollution scenario. Specifically, this includes: By querying the real-time data service interface of the digital twin, we can obtain the latest updated pollutant concentration spatial distribution matrix from the sensor network, the water level time series data of each monitoring point, and the instantaneous flow velocity field vector data obtained through numerical simulation. Simultaneously, key geological attribute parameters pre-stored in the grid cells of the regional three-dimensional geological model are extracted, including the permeability tensor, effective porosity scalar, and spatial geometric and connectivity parameters of fracture development zones for each grid cell; the extracted geological parameters are matched and aligned with real-time monitoring data according to spatial coordinates to form an original dataset containing spatiotemporal dynamic data and static geological constraints. Based on the original dataset, time-series data of water level and pollutant concentration are fused with corresponding geological parameters using an attention mechanism to generate a hybrid input vector containing temporal and static geological features. This vector is then segmented into fixed-length sample fragments, each containing observations from several past time steps and their corresponding geological features. Simultaneously, time-series data of external environmental factors are obtained using a big data network as an additional feature dimension added to the input vector, constructing a multi-dimensional training sample set with spatiotemporal correlation. The multidimensional training sample set is input into the blank LSTM prediction model for supervised training. The input layer receives the time series data of multidimensional sample features, the hidden layer learns the complex long-term dependencies in the groundwater system through the forget gate, input gate and output gate, and the output layer generates the predicted values ​​of pollutant concentration and water level through a fully connected network, and finally obtains the trained LSTM prediction model. The input for each time step includes the monitoring data at the current moment and the corresponding static geological parameters. The predicted output is calculated through forward propagation, and the gradient of the loss function with respect to the network weights is calculated through backpropagation using the BPTT algorithm. The network parameters are iteratively updated using an adaptive moment estimation optimizer. The trained LSTM model is used for multi-step recursive prediction. The monitoring data at the current time is used as the initial state to predict the pollutant concentration distribution and water level field at the next time step. Then, the prediction results are used as new inputs to recursively predict and generate the spatial distribution sequence of pollutant concentration and the dynamic change sequence of water level at future time steps. Geological constraints were applied to the generated spatial distribution sequence of pollutant concentration and dynamic change sequence of water level. Based on the spatial distribution tensor of permeability coefficient, the inverse distance weighted interpolation method was used to spatially register the permeability coefficient field with the predicted concentration field. Using the density field data of fracture development, the optimal path analysis method based on Dijkstra's algorithm was used to identify the dominant transport channels of pollutants in the fracture network. In the fracture development area, the convection-dispersion parameters of pollutants were adjusted according to the fracture density ratio. Finally, the pollution scenario simulation information corrected by geological characteristics was generated.

6. The integrated intelligent control method for groundwater extraction and injection based on the Internet of Things as described in claim 1, characterized in that, The process uses the pollution scenario simulation information as input, coupled with permeability coefficient and fracture distribution parameters as constraints, and performs multi-objective collaborative optimization calculations using the BLMFO two-layer multi-objective optimization algorithm to generate a collaborative control strategy set for the injection-extraction well group, specifically including: A two-level multi-objective optimization algorithm for BLMFO is constructed, with three conflicting optimization objectives: minimizing the plume range, minimizing system operating energy consumption, and balancing equipment usage load. The flow distribution scheme, operating time series, and reagent dosage concentration of each injection well are used as decision variables to establish a mathematical model of the upper-level optimization problem that includes multi-objective functions and basic constraints. Based on the established upper-level optimization problem model, a decision variable space is constructed by using a preset three-level coding strategy according to the real-time operation control strategy and operation monitoring data of the injection well group in the target area. An improved artificial bee colony optimization algorithm is introduced for upper-level solution calculation. Honey source initialization is performed by combining random selection and strategy selection. An initial Pareto optimal front is constructed through iterative optimization analysis to generate candidate control strategies. The candidate control strategy is passed to the lower-level optimization model. In the lower-level model, a response surface model under geological constraints is established based on the spatial variation characteristics of the permeability coefficient and the fracture network connectivity parameters. The geological feasibility of each candidate strategy passed from the upper level is verified, and the verification results are fed back to the upper-level optimization process in the form of constraint violation degree. Based on the geological constraint verification results fed back from the lower-level model, the upper-level optimization algorithm performs a co-evolution operation, using dynamically adjusted inertia weights and learning factors to update the velocity and position of particles, and eliminating inferior solutions that do not meet the geological conditions through a tournament selection mechanism. After a preset number of double-layer iterative calculations, the algorithm outputs the optimal solution set that achieves the best balance among multiple optimization objectives and meets the geological constraints, thus generating a set of collaborative control strategies for the injection and extraction well group.

7. The integrated intelligent control method for groundwater extraction and injection based on the Internet of Things as described in claim 6, characterized in that, The process involves constructing a decision variable space using a pre-defined three-level coding strategy, introducing an improved artificial bee colony optimization algorithm for upper-level solution calculations, initializing nectar sources by combining random selection and strategy selection, constructing an initial Pareto optimal frontier through iterative optimization analysis, and generating candidate control strategies. Specifically, this includes: The three-level coding strategy consists of injection well codes, process codes, and monitoring codes. The injection well codes include the location and functional attributes of each injection well. The process codes include the running sequence and running strategy of the corresponding injection well. The monitoring codes are the running monitoring status characteristics of the corresponding injection well. During the bee-hiring phase, a uniform crossover strategy is used to perform a neighborhood search on the current nectar source. A solution is randomly selected from the Pareto archive and swapped with the current solution for the code fragment. At the same time, a new candidate solution is generated at the machine coding layer using the strategy of minimizing cumulative processing time. The Pareto dominance principle is used to select a better solution from the old and new solutions and update the archive. In the follower bee phase, the following probability of all nectar sources is calculated based on the non-dominance level and crowding of the nectar source. A roulette wheel method is used to select nectar sources that need to be searched further. The search strategy is consistent with that in the hired bee phase to ensure in-depth mining around high-quality solutions. When the nectar source reaches the search limit but the quality does not improve, the scout bee uses a random exchange strategy to reconstruct the well group allocation code, randomly selects some well groups to be redistributed to other processing units, while maintaining the overall processing capacity balance, and regenerates the scheduling order and equipment parameters according to the initialization scheme, thereby increasing the population diversity. During the iterative optimization process, the non-dominated solutions are recorded as the Pareto optimal solution set by updating the archive set. After each iteration, the current population is sorted by fast non-dominated sorting. Solutions ranked within the preset range are added to the archive set, and individuals dominated by the new solutions are removed from the archive set. After a preset number of iterations, the initial Pareto optimal frontier is output, and candidate control strategies are generated.

8. The integrated intelligent control method for groundwater extraction and injection based on the Internet of Things as described in claim 1, characterized in that, The process involves generating a control command sequence based on the coordinated control strategy set of the injection and extraction well group, sending it to the corresponding execution mechanism for execution, monitoring the real-time operating status during the injection and extraction repair process, determining whether a repair strategy change is necessary, generating a repair strategy change suggestion, and pushing it out. Specifically, this includes: Based on the operating parameters and target set values ​​of each well contained in the coordinated control strategy set of the injection and extraction well group, the continuous control parameters in the strategy set are transformed into a sequence of instructions with temporal relationships through time discretization processing. Each instruction contains an equipment identifier, execution timestamp, control parameter set value and expected effect index. The executability and temporal rationality of each instruction within the equipment's capability range are verified by the instruction logic verification algorithm to form a verified control instruction sequence. The verified control command sequence is sent to the corresponding actuators through the Internet of Things communication protocol. Each actuator immediately returns a confirmation signal after receiving the command and starts the command execution process. At the same time, the actual operating parameters of the equipment are collected in real time through built-in sensors, forming an execution feedback data stream that includes instantaneous flow rate, actual dosage of reagents and equipment operating status. During the execution of the command, multi-dimensional operational status monitoring is initiated simultaneously. Groundwater dynamic data, including water level change sequences, pollutant concentration distribution data, and equipment operating status parameters, are collected in real time through a heterogeneous sensor network deployed in the injection well group. After data preprocessing, a high-quality real-time operational status dataset is generated. Dynamic evaluation of remediation effectiveness is carried out based on real-time operational status dataset. The actual monitored pollutant concentration distribution is compared and analyzed with the remediation trajectory expected in the strategy set. The deviation between the actual and expected values ​​of remediation performance indicators is calculated. At the same time, by analyzing equipment operation status data, the workload and operating efficiency of each execution unit are evaluated, equipment performance degradation or abnormal operation are identified, and a regional remediation status assessment report is generated. Based on the regional restoration status assessment report, a process for determining the necessity of strategy changes was initiated. A judgment logic based on multi-dimensional thresholds was established. When the actual rate of decrease in pollutant concentration is lower than the set percentage of the expected value, or when the pollutant concentration at key points shows a rebound trend, or when the equipment operating efficiency is continuously lower than the rated efficiency threshold, it is determined that a strategy change is necessary. When a strategy change is deemed necessary, based on real-time operational characteristics and effect evaluation results, effective adjustment strategies under similar operating conditions are retrieved from the historical case database using a case-based reasoning method. Combined with the current equipment operating status and geological constraints, the generated flow adjustment range, reagent ratio optimization scheme, and equipment operating parameter correction values ​​are used to generate a repair strategy change suggestion, which is then pushed to the management terminal.

9. An integrated intelligent control system for groundwater extraction and injection based on the Internet of Things, characterized in that, The system includes: a geological sensing and digital twin unit, an intelligent prediction and decision-making unit, an execution and closed-loop control unit, a data communication and edge computing unit, and an energy management and security unit. The geological sensing and digital twin unit integrates geological survey data and real-time sensor network monitoring data to construct and continuously update a digital twin of the contaminated site, providing a comprehensive perception of the underground environment. The intelligent prediction and decision-making unit predicts pollutant migration and water level dynamics based on data provided by the digital twin, and performs corrections by coupling geological constraints and formulates the optimal set of collaborative control strategies for the injection and extraction well groups. The execution and closed-loop control unit translates decisions into actions and makes real-time adjustments based on feedback, driving field actuators to perform repair operations, while continuously monitoring equipment operating status and repair effects, quickly judging and generating strategy adjustment suggestions; The data communication and edge computing unit is responsible for the transmission and preliminary processing of all data and instructions. It adopts a hybrid communication network of LoRaWAN and 5G to connect all sensors, actuators and cloud platforms. The energy management and protection unit integrates solar photovoltaic panels, energy storage batteries, and mains power to form a hybrid power supply system. It is responsible for providing power supply and dynamically optimizing energy distribution strategies to ensure the normal operation of equipment under abnormal power supply conditions.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a program for an integrated intelligent management and control method for groundwater extraction and injection based on the Internet of Things (IoT). When the program is executed by a processor, it implements the steps of the integrated intelligent management and control method for groundwater extraction and injection based on the Internet of Things as described in any one of claims 1 to 8.

Images