A data-driven-based groundwater extraction and injection intelligent control method and system

Abstract

The application discloses a kind of data-driven based groundwater extraction injection intelligent control method and system, it is related to data processing field;The multidimensional data of acquisition place and local water quality purification parameter are carried out data processing to obtain unified data set;Based on historical unified data set, target prediction model is constructed in combination with target area hydrogeology basic data;Unified data set is substituted into target prediction model to obtain the pollution concentration variation trend of groundwater and pollution range diffusion risk level;Based on pollution concentration variation trend, pollution range diffusion risk level, in combination with historical pollution control effect data, the preset pollution concentration early warning threshold, pollution range boundary early warning threshold are dynamically updated;Actual pollution concentration in current unified data set, pollution range and updated early warning threshold are compared to obtain target control instruction, to carry out groundwater extraction injection intelligent control, reduce manpower input and operation error, improve groundwater extraction injection efficiency.

Description

Technical Field

[0001] This invention belongs to the field of data processing technology, specifically relating to a data-driven intelligent control method and system for groundwater extraction and injection. Background Technology

[0002] Existing extraction-injection systems are mostly single, manually controlled systems that extract and inject contaminated groundwater or perform recharge based on predetermined extraction-injection parameters. These parameters depend on groundwater contamination surveys and engineering designs; if contamination changes dynamically, the original design may fail to adapt, leading to performance losses and substandard results. This proposed solution integrates the extraction-injection system with front-end data collection and analysis modules. Automatic sampling, monitoring, and analysis data are transmitted to the extraction-injection system. Based on identified real-time contamination conditions or migration simulation trends, extraction-injection parameters are automatically set and corrected. During system operation, real-time monitoring and analysis are performed by the front-end data terminal, and the real-time data is retransmitted back to the extraction-injection system, achieving data linkage and transmission, and enabling automatic contamination identification and engineering control.

[0003] However, the current method of controlling the injection volume and pressure based on human experience is difficult to match with the dynamic changes in hydrogeology and cannot provide real-time feedback on the injection effect. This can easily lead to secondary pollution or waste of resources, ultimately resulting in low efficiency and high energy consumption in groundwater extraction and injection. Summary of the Invention

[0004] The purpose of this invention is to solve the problems of low extraction and injection efficiency and high energy consumption of groundwater extraction and injection, and to propose a data-driven intelligent control method and system for groundwater extraction and injection.

[0005] In a first aspect of this invention, a data-driven intelligent control method for groundwater extraction and injection is first proposed, the method comprising: Groundwater multi-dimensional data are collected according to a preset collection cycle; the multi-dimensional data includes target pollution concentration, pollution range, water level, water pressure, hydrogeological dynamic data, and water quality indicators; Simultaneously collect data on the dosage of water purification agents, the compliance status of treated water, and the effluent flow rate. A unified dataset is obtained by processing the multi-dimensional data and local water purification parameters. Based on a unified historical dataset, a target prediction model for spatiotemporal prediction of pollution concentration is constructed by combining basic hydrogeological data of the target area. Substituting the unified dataset into the target prediction model, we obtain the trend of groundwater pollution concentration change and the risk level of pollution spread within the next collection cycle at the current moment; Based on the trend of pollution concentration changes and the risk level of pollution spread, combined with historical pollution control effectiveness data, the preset pollution concentration warning threshold and pollution range boundary warning threshold are dynamically updated. The target control command is obtained by comparing the actual pollution concentration and pollution range in the current unified dataset with the updated warning threshold, and the intelligent control of groundwater extraction and injection is carried out according to the target control command.

[0006] By collecting multi-dimensional groundwater data and local water quality purification parameters at preset intervals and processing them to form a unified dataset, a predictive model is built based on historical data and hydrogeological data to accurately predict the trend of pollution concentration changes and the level of diffusion risk. The warning threshold is then dynamically updated, and control commands are generated based on the comparison between actual data and the updated threshold to achieve intelligent control of groundwater extraction and injection. This not only captures the dynamics of groundwater pollution and water quality purification in real time and blocks pollution migration paths in advance, significantly reducing the risk of pollution diffusion and treatment costs, but also replaces repetitive manual work, reduces manpower input and operational errors, improves management efficiency, stability, compliance and transparency, and increases the efficiency of groundwater extraction and injection.

[0007] Optionally, a target prediction model for pollution concentration and spatiotemporal prediction can be constructed based on a historical unified dataset and combined with basic hydrogeological data of the target area, including: A unified historical dataset and basic hydrogeological data of the target area are obtained. Feature association and filtering are performed on the two types of data to obtain the model training dataset. The unified historical dataset is a set of multi-dimensional groundwater data and local water quality purification parameters of the target area in the past after preprocessing. The basic hydrogeological data includes at least aquifer distribution, permeability and groundwater flow direction data. Substituting the training dataset into the time distribution model yields the time feature matrix; Substitute the training dataset into the spatial distribution model to obtain the spatial feature matrix; By fusing the temporal and spatial feature matrices, the trends in pollution concentration changes and the risk levels of pollution spread can be obtained. With the goal of minimizing the error between the predicted and actual pollution concentrations, the model parameters are iteratively optimized to obtain the target prediction model.

[0008] By integrating a unified historical dataset of the target area with key hydrogeological data such as aquifer distribution and permeability, a high-quality training dataset is formed through feature correlation screening. The parameters are iteratively optimized with the goal of minimizing the error between the predicted and actual pollution concentrations. This fully explores the spatiotemporal correlation information and core hydrogeological influencing factors in the data, enabling the model to accurately capture the evolution patterns of pollution. It also provides reliable support for predicting subsequent pollution concentration trends and diffusion risk levels, thereby providing a scientific basis for intelligent regulation of groundwater extraction and injection, and effectively improving the foresight and accuracy of pollution control.

[0009] Optionally, the time distribution model includes a bidirectional GRU model; the bidirectional GRU model has one layer for forward training and the other layer for backward training; Substitute the training dataset into the bidirectional GRU model to obtain the hidden state matrix; The hidden state features of the current time are used as the query vector, and the hidden state matrix is ​​used as the key vector and value vector. Attention is then performed to obtain the time feature matrix. The spatial feature matrix is ​​obtained by substituting the training dataset into the CNN model.

[0010] Optionally, based on the trend of pollution concentration changes, the risk level of pollution spread, and combined with historical pollution control effectiveness data, the preset pollution concentration warning threshold and pollution range boundary warning threshold can be dynamically updated, including: A trend adjustment coefficient is determined based on the pollution concentration change trend parameter, and a risk adjustment coefficient is determined based on the pollution range diffusion risk level parameter and basic hydrogeological parameters. The effectiveness of historical pollution control data sets was evaluated, and the proportions of effective control samples, ineffective control samples, and over-control samples were statistically analyzed. Based on the statistical results of the sample proportions, the historical effect feedback coefficients were determined. Based on the preset pollution concentration warning threshold and the preset pollution range boundary warning threshold, the updated pollution concentration warning threshold and pollution range boundary warning threshold are calculated by combining the trend adjustment coefficient, risk adjustment coefficient and historical effect feedback coefficient.

[0011] By combining the trend of pollution concentration changes to determine the trend adjustment coefficient, and by clarifying the risk adjustment coefficient based on the pollution range diffusion risk level and basic hydrogeological parameters, and then by evaluating the historical pollution control effect dataset and statistically analyzing the proportion of various control samples to obtain the historical effect feedback coefficient, the three types of coefficients are finally integrated to calculate and update the preset warning threshold. This approach fully considers the real-time evolution of pollution and the core impact of the geological environment, while also incorporating effective feedback from historical control experience. This makes the updated pollution concentration and range boundary warning thresholds more closely match actual working conditions, avoiding the limitations of fixed thresholds that are difficult to adapt to dynamic pollution scenarios. It can more accurately trigger subsequent extraction and injection of intelligent control commands, further improving the timeliness, pertinence, and scientific nature of pollution control, reducing ineffective or over-control situations, and lowering the risk of pollution diffusion and treatment costs.

[0012] Optionally, based on preset pollution concentration warning thresholds and preset pollution range boundary warning thresholds, combined with the trend adjustment coefficient, risk adjustment coefficient, and historical effect feedback coefficient, updated pollution concentration warning thresholds and pollution range boundary warning thresholds are calculated, including those obtained through formulas. The pollution urgency index is calculated; where E represents the pollution urgency index. and The weights are preset and the sum is 1. This represents the current pollution concentration. This represents the median value of the current pollution range. The preset pollution concentration baseline warning threshold, and These are the left and right boundary thresholds of the preset pollution range boundary warning threshold, respectively; based on the pollution urgency index, using the formula... Obtain the first weight using the formula. The second weight is obtained through the formula. A third weight is obtained; the trend adjustment coefficient, risk adjustment coefficient, and historical effect feedback coefficient are fused based on the first weight, the second weight, and the third weight to obtain a fusion coefficient; The preset pollution concentration warning threshold and the preset pollution range boundary warning threshold are updated based on the fusion coefficient to obtain the updated pollution concentration warning threshold and pollution range boundary warning threshold.

[0013] By quantitatively calculating the urgency of pollution, precise control of the current pollution situation is achieved. Through weight allocation and coefficient fusion, the dynamic trend of pollution changes, the level of diffusion risk, and historical control experience are fully considered, making the update of the warning threshold more data-supported and scientific. It effectively avoids the warning deviation caused by fixed thresholds or single factor considerations, so that the updated pollution concentration and range boundary warning thresholds can be dynamically adapted to the actual pollution scenario and geological environment, accurately triggering subsequent extraction and injection intelligent control, greatly improving the pertinence and timeliness of pollution control, reducing problems such as ineffective control, over-control, or control lag, and further reducing the risk of pollution diffusion and the cost of treatment.

[0014] In a second aspect of the present invention, a data-driven intelligent control system for groundwater extraction and injection is proposed, comprising: The data acquisition module is used to collect multi-dimensional data of groundwater according to a preset acquisition cycle; the multi-dimensional data includes target pollution concentration, pollution range, water level, water pressure, hydrogeological dynamic data and water quality indicators; The synchronous data acquisition module is used to synchronously collect data on the dosage of reagents used for local water purification, the compliance status of treated water, and the outflow rate. A unified dataset determination module is used to process the multi-dimensional data and local water quality purification parameters to obtain a unified dataset. The target prediction model building module is used to build a target prediction model for spatiotemporal prediction of pollution concentration based on a historical unified dataset and combined with basic hydrogeological data of the target area. The pollution prediction module is used to input the unified dataset into the target prediction model to obtain the trend of groundwater pollution concentration change and the risk level of pollution spread within the next collection cycle at the current moment. The threshold update module is used to dynamically update the preset pollution concentration warning threshold and pollution range boundary warning threshold based on the pollution concentration change trend, the pollution range spread risk level, and historical pollution control effect data. The intelligent control module is used to compare the actual pollution concentration and pollution range in the current unified dataset with the updated warning threshold to obtain the target control command, and to perform intelligent control of groundwater extraction and injection according to the target control command.

[0015] Optionally, the target prediction model construction module includes: The model training dataset determination module is used to acquire historical unified datasets and basic hydrogeological data of the target area, perform feature association and filtering on the two types of data to obtain the model training dataset; the historical unified dataset is a set of pre-processed and integrated groundwater multi-dimensional data and local water quality purification parameters of the target area in history; the basic hydrogeological data includes at least aquifer distribution, permeability and groundwater flow direction data. The time feature matrix determination module is used to substitute the training dataset into the time distribution model to obtain the time feature matrix; The spatial feature matrix determination module is used to substitute the training dataset into the spatial distribution model to obtain the spatial feature matrix. The feature fusion module is used to fuse the temporal feature matrix and the spatial feature matrix to obtain the pollution concentration change trend and the pollution range diffusion risk level; The target prediction model determination module is used to iteratively optimize model parameters to obtain the target prediction model with the goal of minimizing the error between the predicted pollution concentration and the actual concentration.

[0016] Optionally, the temporal distribution model includes a bidirectional GRU model; the bidirectional GRU model has one layer for forward training and another layer for backward training; the training dataset is substituted into the bidirectional GRU model to obtain a hidden state matrix; the hidden state features of the current time are used as query vectors, and the hidden state matrix is ​​used as key vectors and value vectors to perform attention calculations to obtain a temporal feature matrix; the training dataset is substituted into a CNN model to obtain a spatial feature matrix.

[0017] Optionally, the threshold update module includes: The adjustment coefficient determination module is used to determine the trend adjustment coefficient based on the pollution concentration change trend parameter, and to determine the risk adjustment coefficient based on the pollution range diffusion risk level parameter and the basic hydrogeological parameters. The feedback coefficient determination module is used to evaluate the effectiveness of historical pollution control effect datasets, statistically analyze the proportion of effective control samples, ineffective control samples, and over-control samples, and determine the historical effect feedback coefficients based on the sample proportion statistics. The updated threshold determination module is used to calculate, based on the preset pollution concentration warning threshold and the preset pollution range boundary warning threshold, combined with the trend adjustment coefficient, risk adjustment coefficient and historical effect feedback coefficient, to obtain the updated pollution concentration warning threshold and pollution range boundary warning threshold.

[0018] Optionally, the update threshold determination module includes: The pollution urgency index determination module is used to determine the urgency level of pollution through a formula. The pollution urgency index is calculated; where E represents the pollution urgency index. and The weights are preset and the sum is 1. This represents the current pollution concentration. This represents the median value of the current pollution range. The preset pollution concentration baseline warning threshold, and These are the left and right boundary thresholds of the preset pollution range boundary warning threshold, respectively; the weight calculation module is used to calculate the weight based on the pollution urgency index using the formula... Obtain the first weight using the formula. The second weight is obtained through the formula. The third weight is obtained; the fusion coefficient determination module is used to fuse the trend adjustment coefficient, risk adjustment coefficient and historical effect feedback coefficient according to the first weight, the second weight and the third weight to obtain the fusion coefficient; The threshold update calculation module is used to update the preset pollution concentration warning threshold and the preset pollution range boundary warning threshold according to the fusion coefficient, so as to obtain the updated pollution concentration warning threshold and pollution range boundary warning threshold.

[0019] The beneficial effects of this invention; This invention proposes a data-driven intelligent control method for groundwater extraction and injection. It collects multi-dimensional groundwater data and local water quality purification parameters at preset intervals, processing them to form a unified dataset. Then, based on the historical unified dataset and basic hydrogeological data of the target area, a predictive model is constructed to accurately predict the trend of groundwater pollution concentration changes and the risk level of pollution spread, thereby dynamically updating the pollution concentration and range boundary warning thresholds. Finally, the actual pollution concentration and pollution range in the current unified dataset are compared with the updated warning thresholds to generate target control commands to achieve intelligent control of groundwater extraction and injection. This significantly reduces the risk of pollution spread and subsequent treatment costs, and can replace repetitive tasks such as manual on-site inspections, parameter adjustments, and data recording and analysis, reducing manpower input and human error. It significantly improves management efficiency, stability, compliance, and transparency, effectively addressing the problem of low extraction efficiency in traditional extraction and injection systems. Attached Figure Description

[0020] The invention will now be further described with reference to the accompanying drawings.

[0021] Figure 1 A flowchart of a data-driven intelligent control method for groundwater extraction and injection provided for an embodiment of the present invention; Figure 2 This is a framework diagram of a data-driven intelligent control system for groundwater extraction and injection, provided for an embodiment of the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] This invention provides a data-driven intelligent control method for groundwater extraction and injection. See also... Figure 1 , Figure 1 A flowchart illustrating a data-driven intelligent control method for groundwater extraction and injection, provided as an embodiment of the present invention. The method includes the following steps; S101, collects multi-dimensional data of groundwater according to a preset collection cycle; S102, synchronously collect local water purification agent dosage, treated water quality compliance status and effluent flow data; S103, a unified dataset is obtained by processing multi-dimensional data and local water purification parameters; S104, a target prediction model for spatiotemporal prediction of pollution concentration is constructed based on a historical unified dataset and combined with basic hydrogeological data of the target area. S105, Substitute the unified dataset into the target prediction model to obtain the trend of groundwater pollution concentration change and pollution range diffusion risk level in the next collection cycle at the current moment; S106, based on the trend of pollution concentration change, the risk level of pollution range spread, and combined with historical pollution control effect data, dynamically update the preset pollution concentration warning threshold and pollution range boundary warning threshold. S107, compare the actual pollution concentration and pollution range in the current unified dataset with the updated warning threshold to obtain the target control command, and carry out intelligent control of groundwater extraction and injection according to the target control command; The multi-dimensional data includes target pollution concentration, pollution range, water level, water pressure, hydrogeological dynamics data, and water quality indicators.

[0024] In one implementation, the preset acquisition cycle is 15 to 30 minutes, which is determined by technical personnel; groundwater data acquisition is completed through monitoring terminals deployed in layers according to the core pollution area, pollution diffusion path, control boundary and aquifer level; hydrogeological dynamic data includes aquifer permeability and groundwater flow velocity, and water quality indicators include pH value, dissolved oxygen content and mineralization.

[0025] In one implementation, outliers are removed and missing values ​​are filled in, and non-numerical parameters are converted into numerical parameters; multi-source data are unified to a preset time sampling frequency and geographic coordinate system; based on groundwater flow direction and spatial distance, water quality purification parameters are spatiotemporally correlated with groundwater monitoring data in the corresponding affected areas; the fused data is standardized and scaled to eliminate dimensional differences, and finally a structured unified dataset is formed.

[0026] In one implementation, when the actual data does not exceed the updated warning threshold and there is no predicted risk of exceeding the standard or spreading, the system maintains the current operating parameters or remains in standby mode. When the actual data approaches or exceeds the updated warning threshold, or when it is predicted that the next cycle will approach or exceed the updated warning threshold, the system triggers an intelligent control command. The intelligent control command is based on the distribution location of the well groups, which are divided according to pollution zones and aquifer levels. The flow regulation error of the automatic control valve for a single well is ≤5%, and the pressure regulation error is ≤3%. When the actual pollution concentration is ≥1.5 times the updated concentration warning threshold or the pollution range exceeds the updated boundary warning threshold, the control command sets the injection-extraction ratio to 1:0.5 and initiates auxiliary extraction, control levels, and functional positioning of adjacent well groups, generating a quantitative extraction and injection command. The command includes parameters such as the injection-extraction ratio, injection volume, injection pressure, and runtime. The automatic control of a single well in the well group... The valve executes operations according to instructions, extracting water bodies with excessive pollution, while simultaneously controlling the injection well to inject water that has been treated to meet standards by the local water purification system or suitable water containing remediation agents. During the extraction and injection process, the pressure changes inside the well pipe are monitored in real time by a single-well pressure sensor. When the pressure fluctuation exceeds the preset fluctuation threshold of 0.02MPa, the pressure control tank is used for dynamic pressure relief or replenishment to maintain stable system operating pressure. The actual extraction and injection parameters and multi-dimensional data updated and collected by the monitoring terminal are continuously transmitted back to the cloud data analysis platform. The platform calls the built-in hydrogeological model, which is based on MODFLOW or VisualMODFLOW secondary development, to simulate the changes in pollution diffusion and generate visualized management results, including a heat map of the pollution range, a trend map of the extraction and injection effect, and a pollution evolution prediction map, which are used for pollution source tracing, optimization of treatment plans, and regulatory reporting.

[0027] In one embodiment, a target prediction model for predicting pollution concentration in time and space is constructed based on a historical unified dataset and combined with basic hydrogeological data of the target area, including: A unified historical dataset and basic hydrogeological data of the target area are obtained. Feature association and filtering are performed on the two types of data to obtain the model training dataset. The unified historical dataset is a set of multi-dimensional groundwater data and local water quality purification parameters of the target area in the past after preprocessing. The basic hydrogeological data includes at least aquifer distribution, permeability and groundwater flow direction data. Substituting the training dataset into the time distribution model yields the time feature matrix; Substitute the training dataset into the spatial distribution model to obtain the spatial feature matrix; By fusing the temporal and spatial feature matrices, the trends in pollution concentration changes and the risk levels of pollution spread can be obtained. With the goal of minimizing the error between the predicted and actual pollution concentrations, the model parameters are iteratively optimized to obtain the target prediction model.

[0028] In one implementation, by integrating multi-dimensional historical monitoring data and hydrogeological baseline data, and utilizing a modeling strategy of spatiotemporal decoupling and reconstruction, data silos are effectively broken down and physical constraints are introduced. This significantly improves the physical interpretability and spatiotemporal prediction accuracy of the model regarding pollutant migration and transformation patterns while eliminating noise interference. It achieves a leap from single concentration numerical prediction to diffusion risk level assessment, providing a highly reliable quantitative basis for precise groundwater environment management and scientific decision-making. In one embodiment, the time distribution model includes a bidirectional GRU model; one layer of the bidirectional GRU model is trained forward, and the other layer is trained backward. Substitute the training dataset into the bidirectional GRU model to obtain the hidden state matrix; The hidden state features of the current time are used as the query vector, and the hidden state matrix is ​​used as the key vector and value vector. Attention is then performed to obtain the time feature matrix. The spatial feature matrix is ​​obtained by substituting the training dataset into the CNN model.

[0029] In one implementation, both GRU layers in the bidirectional GRU model use the ReLU activation function. The forward GRU traverses the input data in chronological order, outputting a forward hidden state sequence to capture historical temporal correlations. The reverse GRU traverses the data in reverse chronological order, outputting a reverse hidden state sequence to mine future dependencies. The forward and reverse hidden state sequences at each time step are concatenated to obtain a hidden state matrix. Based on the hidden state matrix, the hidden state at the current time step is used as the query vector, and the hidden states at all time steps are used as the key and value vectors, respectively. The similarity is calculated using the dot product formula. The similarity is converted into attention weights in the 0-1 interval using the Softmax function. The attention weights are used as coefficients of the value vectors, and 1 minus the attention weights is used as coefficients of the hidden state matrix for weighted summation to obtain the output temporal feature matrix.

[0030] In one embodiment, based on the trend of pollution concentration change, the risk level of pollution range spread, and combined with historical pollution control effectiveness data, the preset pollution concentration warning threshold and pollution range boundary warning threshold are dynamically updated, including: The trend adjustment coefficient is determined based on the pollution concentration change trend parameter, and the risk adjustment coefficient is determined based on the pollution range diffusion risk level parameter and the basic hydrogeological parameters. The effectiveness of historical pollution control data sets was evaluated, and the proportions of effective control samples, ineffective control samples, and over-control samples were statistically analyzed. Based on the statistical results of the sample proportions, the historical effect feedback coefficients were determined. Based on the preset pollution concentration warning threshold and the preset pollution range boundary warning threshold, the updated pollution concentration warning threshold and pollution range boundary warning threshold are calculated by combining the trend adjustment coefficient, risk adjustment coefficient and historical effect feedback coefficient.

[0031] In one implementation, the trend adjustment coefficient α is determined based on the rate of change and acceleration of pollution concentration. Two core trend parameters are calculated: the rate of change of concentration and the acceleration of change of concentration. If the concentration is accelerating upward, with both the rate of change and the acceleration of change being positive, it indicates an extremely high risk of pollution running out of control. In this case, a small coefficient value is assigned, ranging from 0.7 to 0.8, with a default of 0.75. The specific value is determined by technical personnel. If the concentration is decelerating upward, with a positive rate of change but a negative acceleration of change, it indicates a moderate risk. In this case, a moderate coefficient value is assigned, ranging from 0.8 to 0.9, with a default of 0.85. The specific value is determined by technical personnel. If the concentration is stable or decreasing, with the rate of change being zero or less than zero, it indicates no risk. In this case, a large coefficient value is assigned, ranging from 1.0 to 1.1, with a default of 1.05. The specific value is determined by technical personnel.

[0032] In one implementation, the risk adjustment coefficient β is jointly determined by the pollution range diffusion risk level and basic hydrogeological parameters. Combining the current pollution range diffusion rate and the distance of the pollution plume from the sensitive point, the diffusion risk is divided into three levels: high, medium, and low. Further reading of the basic hydrogeological parameters of the target area is then performed. If the aquifer has high permeability (aquifer permeability greater than 10...), the risk adjustment coefficient is determined by... −4 In strata with high flow velocities (groundwater velocity greater than 0.5 m / h), pollution spreads faster, and the risk level will be further upgraded; if the aquifer has low permeability (permeability less than 10 cm / s), the pollution will spread faster, and the risk level will be further upgraded. −5 For strata with flow rates of cm / s and low velocity (groundwater flow velocity less than 0.2 m / h), the risk level remains unchanged or is appropriately lowered. Based on the revised final risk level, the system sets a risk adjustment coefficient; for high risk, the value is 0.6~0.7, with a default of 0.65, and the specific value is determined by technical personnel; for medium risk, the value is 0.8~0.9, with a default of 0.85, and the specific value is determined by technical personnel; for high risk, the value is 1.0~1.1, with a default of 1, and the specific value is determined by technical personnel.

[0033] In one implementation, the historical effect feedback coefficient γ is determined by screening historical control samples that match the current scenario. Each matching sample must include three types of core information: the α and β coefficients when the historical threshold is adjusted; the adjusted control effect, where the pollution attenuation efficiency is the concentration before control minus the concentration after control divided by the concentration before control; and whether secondary risks occur, such as excessive water level fluctuations or waste of reagents. If the pollution attenuation efficiency is greater than 30% and there are no secondary risks, it is considered a valid sample; if the pollution attenuation efficiency is less than or equal to 30% or secondary risks occur, it is considered an invalid sample. If the proportion of valid control samples is greater than or equal to 80%, the historical effect feedback coefficient is 1; if the proportion of valid control samples is greater than or equal to 60%, the historical effect feedback coefficient is 0.85; if the proportion of valid control samples is greater than or equal to 50%, the historical effect feedback coefficient is 0.7, otherwise it is 0.6.

[0034] In one embodiment, based on preset pollution concentration warning thresholds and preset pollution range boundary warning thresholds, combined with trend adjustment coefficients, risk adjustment coefficients, and historical effect feedback coefficients, updated pollution concentration warning thresholds and pollution range boundary warning thresholds are calculated, including those obtained through formulas. The pollution urgency index is calculated; where E represents the pollution urgency index. and The weights are preset and the sum is 1. This represents the current pollution concentration. This represents the median value of the current pollution range. The preset pollution concentration baseline warning threshold, and These are the left and right boundary thresholds of the preset pollution range boundary warning threshold, respectively; based on the pollution urgency index, using the formula... Obtain the first weight using the formula. The second weight is obtained through the formula. The third weight is obtained; the trend adjustment coefficient, risk adjustment coefficient, and historical effect feedback coefficient are fused based on the first, second, and third weights to obtain the fusion coefficient; The preset pollution concentration warning threshold and the preset pollution range boundary warning threshold are updated based on the fusion coefficient to obtain the updated pollution concentration warning threshold and pollution range boundary warning threshold.

[0035] In one implementation, the median value of the pollution range is the pollution value in the middle of the pollution area.

[0036] In one implementation, the trend adjustment coefficient, risk adjustment coefficient, and historical performance feedback coefficient are fused based on the first weight, second weight, and third weight to obtain the fusion coefficient, which is obtained through the formula... The fusion coefficient K is calculated, where α is the trend adjustment coefficient, β is the risk adjustment coefficient, and γ is the historical effect feedback coefficient; then, it is obtained through the formula... The updated pollution concentration warning threshold was obtained. Where k is the aquifer permeability of the target area, k0 is the standard permeability benchmark value (determined by technical personnel), v is the groundwater flow velocity of the target area, and v0 is the standard flow velocity benchmark value (determined by technical personnel); the pollution range boundary warning threshold is obtained by multiplying the fusion coefficient by the left boundary threshold and the right boundary threshold of the preset pollution range boundary warning threshold respectively.

[0037] In one implementation, the calculation of E in the formula incorporates the deviation between the core value of the current pollution range and the midpoint of the boundary, and combines the boundary span normalization process to accurately quantify the urgency of the pollution range relative to the two sides of the boundary.

[0038] Based on the same inventive concept, this invention also provides a data-driven intelligent control system for groundwater extraction and injection. See also Figure 2 , Figure 2 A framework diagram of a data-driven intelligent control system for groundwater extraction and injection provided for an embodiment of the present invention includes: The data acquisition module is used to collect multi-dimensional data of groundwater according to a preset acquisition cycle; the multi-dimensional data includes target pollution concentration, pollution range, water level, water pressure, hydrogeological dynamic data and water quality indicators; The synchronous data acquisition module is used to synchronously collect data on the dosage of reagents used for local water purification, the compliance status of treated water, and the outflow rate. The unified dataset determination module is used to process multi-dimensional data and local water quality purification parameters to obtain a unified dataset; The target prediction model building module is used to build a target prediction model for spatiotemporal prediction of pollution concentration based on a historical unified dataset and combined with basic hydrogeological data of the target area. The pollution prediction module is used to input a unified dataset into the target prediction model to obtain the trend of groundwater pollution concentration change and the risk level of pollution spread within the next collection cycle at the current moment. The threshold update module is used to dynamically update the preset pollution concentration warning threshold and pollution range boundary warning threshold based on the pollution concentration change trend, the pollution range spread risk level, and historical pollution control effect data. The intelligent control module is used to compare the actual pollution concentration and pollution range in the current unified dataset with the updated warning threshold to obtain the target control command, and to carry out intelligent control of groundwater extraction and injection according to the target control command.

[0039] This invention provides a data-driven intelligent control system for groundwater extraction and injection. It collects multi-dimensional groundwater data and local water quality purification parameters at preset intervals, processes them to form a unified dataset, and constructs a prediction model based on historical data and hydrogeological data. This model accurately predicts the trend of pollution concentration changes and the level of diffusion risk, dynamically updates the warning threshold, and finally generates control commands based on the comparison between actual data and the updated threshold to achieve intelligent control of groundwater extraction and injection. This system not only accurately captures the dynamics of groundwater pollution and water quality purification in real time, blocking pollution migration paths in advance and significantly reducing the risk of pollution diffusion and treatment costs, but also replaces repetitive manual work, reduces manpower input and operational errors, improves management efficiency, stability, compliance, and transparency, and increases the efficiency of groundwater extraction and injection.

[0040] The foregoing has described one embodiment of the present invention in detail, but this content is merely a preferred embodiment and should not be considered as limiting the scope of the present invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the scope of the claims of this invention.

Claims

1. A data-driven based intelligent control method for groundwater extraction and injection, characterized in that, The method includes; Collect multi-dimensional data on groundwater according to a preset collection cycle; The multi-dimensional data includes target pollution concentration, pollution range, water level, water pressure, hydrogeological dynamics data, and water quality indicators. Simultaneously collect data on the dosage of water purification agents, the compliance status of treated water, and the effluent flow rate. A unified dataset is obtained by processing the multi-dimensional data and local water purification parameters. Based on a unified historical dataset, a target prediction model for spatiotemporal prediction of pollution concentration is constructed by combining basic hydrogeological data of the target area. Substituting the unified dataset into the target prediction model, we obtain the trend of groundwater pollution concentration change and the risk level of pollution spread within the next collection cycle at the current moment; Based on the trend of pollution concentration changes and the risk level of pollution spread, combined with historical pollution control effectiveness data, the preset pollution concentration warning threshold and pollution range boundary warning threshold are dynamically updated. The target control command is obtained by comparing the actual pollution concentration and pollution range in the current unified dataset with the updated warning threshold, and the intelligent control of groundwater extraction and injection is carried out according to the target control command.

2. The data-driven based intelligent control method for groundwater extraction and injection according to claim 1, wherein, Based on a unified historical dataset, and combined with basic hydrogeological data of the target area, a target prediction model for spatiotemporal prediction of pollution concentration is constructed, including: A unified historical dataset and basic hydrogeological data of the target area are obtained. Feature association and filtering are performed on the two types of data to obtain the model training dataset. The unified historical dataset is a set of multi-dimensional groundwater data and local water quality purification parameters of the target area in the past after preprocessing. The basic hydrogeological data includes at least aquifer distribution, permeability and groundwater flow direction data. Substituting the training dataset into the time distribution model yields the time feature matrix; Substitute the training dataset into the spatial distribution model to obtain the spatial feature matrix; By fusing the temporal and spatial feature matrices, the trends in pollution concentration changes and the risk levels of pollution spread can be obtained. With the goal of minimizing the error between the predicted and actual pollution concentrations, the model parameters are iteratively optimized to obtain the target prediction model.

3. The data-driven intelligent control method for groundwater extraction and injection according to claim 2, characterized in that, The time distribution model includes a bidirectional GRU model; the bidirectional GRU model has one layer for forward training and the other layer for backward training. Substitute the training dataset into the bidirectional GRU model to obtain the hidden state matrix; The hidden state features of the current time are used as the query vector, and the hidden state matrix is ​​used as the key vector and value vector. Attention is then performed to obtain the time feature matrix. The spatial feature matrix is ​​obtained by substituting the training dataset into the CNN model.

4. The data-driven intelligent control method for groundwater extraction and injection according to claim 1, characterized in that, Based on the trend of pollution concentration changes and the risk level of pollution spread, combined with historical pollution control effectiveness data, the preset pollution concentration warning threshold and pollution range boundary warning threshold are dynamically updated. A trend adjustment coefficient is determined based on the pollution concentration change trend parameter, and a risk adjustment coefficient is determined based on the pollution range diffusion risk level parameter and basic hydrogeological parameters. The effectiveness of historical pollution control data sets was evaluated, and the proportions of effective control samples, ineffective control samples, and over-control samples were statistically analyzed. Based on the statistical results of the sample proportions, the historical effect feedback coefficients were determined. Based on the preset pollution concentration warning threshold and the preset pollution range boundary warning threshold, the updated pollution concentration warning threshold and pollution range boundary warning threshold are calculated by combining the trend adjustment coefficient, risk adjustment coefficient and historical effect feedback coefficient.

5. The data-driven intelligent control method for groundwater extraction and injection according to claim 4, characterized in that, Based on the preset pollution concentration warning threshold and the preset pollution range boundary warning threshold, combined with the trend adjustment coefficient, risk adjustment coefficient and historical effect feedback coefficient, the updated pollution concentration warning threshold and pollution range boundary warning threshold are calculated to obtain the following: Through formula The pollution urgency index was calculated. Among them, the E pollution urgency index, and The weights are preset and the sum is 1. This represents the current pollution concentration. This represents the median value of the current pollution range. The preset pollution concentration baseline warning threshold, and These are the left and right boundary thresholds of the preset pollution range boundary warning threshold, respectively; Based on the pollution urgency index, using the formula Obtain the first weight using the formula. The second weight is obtained through the formula. Obtain the third weight; The trend adjustment coefficient, risk adjustment coefficient, and historical effect feedback coefficient are fused together based on the first weight, the second weight, and the third weight to obtain a fusion coefficient. The preset pollution concentration warning threshold and the preset pollution range boundary warning threshold are updated based on the fusion coefficient to obtain the updated pollution concentration warning threshold and pollution range boundary warning threshold.

6. A data-driven intelligent control system for groundwater extraction and injection, characterized in that, The system includes; The data acquisition module is used to collect multi-dimensional data of groundwater according to a preset acquisition cycle; The multi-dimensional data includes target pollution concentration, pollution range, water level, water pressure, hydrogeological dynamics data, and water quality indicators. The synchronous data acquisition module is used to synchronously collect data on the dosage of reagents used for local water purification, the compliance status of treated water, and the outflow rate. A unified dataset determination module is used to process the multi-dimensional data and local water quality purification parameters to obtain a unified dataset. The target prediction model building module is used to build a target prediction model for spatiotemporal prediction of pollution concentration based on a historical unified dataset and combined with basic hydrogeological data of the target area. The pollution prediction module is used to input the unified dataset into the target prediction model to obtain the trend of groundwater pollution concentration change and the risk level of pollution spread within the next collection cycle at the current moment. The threshold update module is used to dynamically update the preset pollution concentration warning threshold and pollution range boundary warning threshold based on the pollution concentration change trend, the pollution range spread risk level, and historical pollution control effect data. The intelligent control module is used to compare the actual pollution concentration and pollution range in the current unified dataset with the updated warning threshold to obtain the target control command, and to perform intelligent control of groundwater extraction and injection according to the target control command.

7. The data-driven intelligent control system for groundwater extraction and injection according to claim 6, characterized in that, The target prediction model construction module includes: The model training dataset determination module is used to acquire historical unified datasets and basic hydrogeological data of the target area, perform feature association and filtering on the two types of data to obtain the model training dataset; the historical unified dataset is a set of pre-processed and integrated groundwater multi-dimensional data and local water quality purification parameters of the target area in history; the basic hydrogeological data includes at least aquifer distribution, permeability and groundwater flow direction data. The time feature matrix determination module is used to substitute the training dataset into the time distribution model to obtain the time feature matrix; The spatial feature matrix determination module is used to substitute the training dataset into the spatial distribution model to obtain the spatial feature matrix. The feature fusion module is used to fuse the temporal feature matrix and the spatial feature matrix to obtain the pollution concentration change trend and the pollution range diffusion risk level; The target prediction model determination module is used to iteratively optimize model parameters to obtain the target prediction model with the goal of minimizing the error between the predicted pollution concentration and the actual concentration.

8. The data-driven intelligent control system for groundwater extraction and injection according to claim 7, characterized in that, The temporal distribution model includes a bidirectional GRU model; the bidirectional GRU model has one layer for forward training and another layer for backward training; the training dataset is substituted into the bidirectional GRU model to obtain the hidden state matrix; the hidden state features of the current time are used as the query vector, and the hidden state matrix is ​​used as the key vector and value vector to perform attention calculation to obtain the temporal feature matrix; the training dataset is substituted into the CNN model to obtain the spatial feature matrix.

9. A data-driven intelligent control system for groundwater extraction and injection according to claim 6, characterized in that, The threshold update module includes: The adjustment coefficient determination module is used to determine the trend adjustment coefficient based on the pollution concentration change trend parameter, and to determine the risk adjustment coefficient based on the pollution range diffusion risk level parameter and the basic hydrogeological parameters. The feedback coefficient determination module is used to evaluate the effectiveness of historical pollution control effect datasets, statistically analyze the proportion of effective control samples, ineffective control samples, and over-control samples, and determine the historical effect feedback coefficients based on the sample proportion statistics. The updated threshold determination module is used to calculate, based on the preset pollution concentration warning threshold and the preset pollution range boundary warning threshold, combined with the trend adjustment coefficient, risk adjustment coefficient and historical effect feedback coefficient, to obtain the updated pollution concentration warning threshold and pollution range boundary warning threshold.

10. A data-driven intelligent control system for groundwater extraction and injection according to claim 9, characterized in that, The update threshold determination module includes: The pollution urgency index determination module is used to determine the urgency level of pollution through a formula. The pollution urgency index is calculated; where E represents the pollution urgency index. and The weights are preset and the sum is 1. This represents the current pollution concentration. This represents the median value of the current pollution range. The preset pollution concentration baseline warning threshold, and These are the left and right boundary thresholds of the preset pollution range boundary warning threshold, respectively; the weight calculation module is used to calculate the weight based on the pollution urgency index using the formula... Obtain the first weight using the formula. The second weight is obtained through the formula. The third weight is obtained; the fusion coefficient determination module is used to fuse the trend adjustment coefficient, risk adjustment coefficient and historical effect feedback coefficient according to the first weight, the second weight and the third weight to obtain the fusion coefficient; The threshold update calculation module is used to update the preset pollution concentration warning threshold and the preset pollution range boundary warning threshold according to the fusion coefficient, so as to obtain the updated pollution concentration warning threshold and pollution range boundary warning threshold.

Images