A dynamic intelligent monitoring method and system for groundwater pollution

Abstract

The application relates to the field of groundwater pollution monitoring, in particular to a groundwater pollution dynamic intelligent monitoring method and system. The method comprises the following steps: constructing and updating a groundwater solute transport model; generating a comprehensive pollution feature vector from obtained multi-source data related to groundwater pollution and containing multiple pollution indexes; respectively constructing a pollution source positioning model and a pollution diffusion model; triggering a pollution warning according to the comprehensive pollution feature vector; obtaining a pollution source position by using the pollution source positioning model according to the multi-source data and current hydrogeological data; respectively using the pollution diffusion model and the groundwater solute transport model for prediction simulation to obtain a first pollution distribution changing with time and a second pollution distribution changing with time; and weighting and fusing the first pollution distribution and the second pollution distribution to generate a predicted pollution distribution changing with time. The method solves the problems that groundwater pollution monitoring is not timely and accurate in the prior art.

Description

Technical Field

[0001] This invention relates to the field of groundwater pollution monitoring, specifically a dynamic intelligent monitoring method and system for groundwater pollution. Background Technology

[0002] Groundwater, as a vital water resource, presents pollution problems characterized by their concealment, long-term nature, and difficulty in remediation. Traditional groundwater pollution monitoring methods primarily rely on discretely distributed monitoring wells and manual sampling analysis, which have several insurmountable drawbacks. On the one hand, the manual sampling and laboratory analysis model results in long monitoring cycles, with sampling frequencies typically monthly or quarterly, making it impossible to capture real-time dynamic changes in pollution. For example, a sudden chemical raw material leak may form a large-scale pollution plume within days, but the periodicity of manual monitoring allows pollution to continue to spread in a concealed state, missing the golden window for early intervention. On the other hand, the sparse and fixed layout of monitoring wells makes it difficult to cover potential migration paths of groundwater pollution, such as complex hydrogeological areas like fault zones and fracture zones, creating numerous monitoring blind spots and preventing the timely capture of the spatial distribution details of pollution plumes.

[0003] In addition, traditional monitoring data has a single dimension, focusing on conventional water quality indicators such as COD and ammonia nitrogen. It is difficult to reveal the type, source and migration trend of pollutants through multi-indicator coupling analysis. Furthermore, data processing and pollution assessment rely entirely on manual interpretation, lacking intelligent analysis methods. The cycle from data collection to alarm triggering is long, resulting in low emergency response efficiency.

[0004] Meanwhile, groundwater pollution, situated in porous underground media, exhibits significant spatial concealment. Pollutants migrate slowly through pores and fissures, leaving no obvious trace on the surface. Traditional surface patrols struggle to detect the formation of underground pollution plumes. The time lag means that months to years can pass between a pollution source leak and the detection of abnormal concentrations in monitoring wells, allowing the pollution's impact to expand unnoticed and exponentially increasing remediation costs. Furthermore, the complexity of multi-component mixed pollution makes it impossible to accurately assess the overall pollution level with a single indicator, easily overlooking compound pollution risks and further exacerbating the reactive, "late-stage" approach to prevention and control. Therefore, overcoming the limitations of traditional manual monitoring methods and developing intelligent, high-precision groundwater pollution monitoring technologies has become an urgent need to address the concealment of pollution and the complexity of remediation. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a dynamic intelligent monitoring method and system for groundwater pollution, solving the problems of insufficient timeliness and accuracy in groundwater pollution monitoring in existing technologies.

[0006] To achieve the above objectives, the present invention provides a method for dynamic intelligent monitoring of groundwater pollution, comprising: acquiring real-time hydrogeological data of a target area, and constructing and updating a groundwater solute transport model based on the real-time hydrogeological data; constructing a pollution source location model and a pollution diffusion model respectively; acquiring multi-source data related to groundwater pollution, including multiple pollution indicators, and fusing the multi-source data to generate a comprehensive pollution feature vector; triggering a pollution alarm based on the comprehensive pollution feature vector, and extracting current hydrogeological data from the real-time hydrogeological data based on the pollution alarm; obtaining the pollution source location using the pollution source location model based on the multi-source data and the current hydrogeological data; predicting a first pollution distribution over time using the pollution diffusion model based on the current hydrogeological data, the pollution source location, and the multi-source data; simulating a second pollution distribution over time using the groundwater solute transport model based on the pollution source location and the multi-source data; and weighted fusing the first pollution distribution and the second pollution distribution to generate a predicted pollution distribution over time.

[0007] This invention constructs an updated groundwater solute transport model by acquiring real-time hydrogeological data, builds a pollution source location and pollution diffusion model, integrates multi-source pollution data to generate a comprehensive pollution feature vector, triggers a pollution alarm and extracts current hydrogeological data, then uses the model to sequentially obtain the pollution source location, the first pollution distribution, and the second pollution distribution, and finally weighted and fused to generate a predicted pollution distribution. The entire process realizes multi-technology collaboration and data-driven approach, overcomes the lag and data limitations of traditional manual monitoring, and improves the timeliness and accuracy of groundwater pollution monitoring.

[0008] Optionally, acquiring multi-source data related to groundwater pollution, including multiple pollution indicators, includes: determining potential pollution migration paths based on a groundwater solute transport model, and deploying distributed fiber optic sensors, microelectrode monitoring wells, and ground network monitoring nodes according to the potential pollution migration paths to form a monitoring network; using the monitoring network to obtain monitoring data, including groundwater flow velocity, groundwater temperature, multiple groundwater pollution concentrations, and multiple surface water pollution concentrations; and performing spatiotemporal alignment and feature extraction on the monitoring data to form multi-source data related to groundwater pollution, including multiple pollution indicators.

[0009] This invention identifies potential pollution migration paths using a groundwater solute transport model and deploys a monitoring network consisting of distributed fiber optic sensors, microelectrode monitoring wells, and ground network monitoring nodes. This network acquires monitoring data including groundwater flow velocity, groundwater temperature, various groundwater pollution concentrations, and various surface water pollution concentrations. After spatiotemporal alignment and feature extraction, multi-source data is obtained. This achieves precise coverage of pollution risk areas by the monitoring network, comprehensive capture of pollution characteristics by multi-dimensional data, and effective assurance of data availability through spatiotemporal alignment processing, significantly improving the scientific rigor and accuracy of groundwater pollution monitoring.

[0010] Optionally, the step of fusing the multi-source data to generate a comprehensive pollution feature vector includes: assigning index weights to the pollution indicators in the multi-source data; and generating a comprehensive pollution feature vector based on the index weights using the multi-source data.

[0011] This invention assigns weights to pollution indicators, enabling precise consideration based on their differences in importance and avoiding a one-size-fits-all approach to data processing. Weighted fusion processing effectively integrates multi-source data, suppresses the influence of secondary or interfering data, highlights key pollution information, and improves the accuracy of the comprehensive pollution feature vector in representing groundwater pollution.

[0012] Optionally, the construction of the pollution source location model and the pollution diffusion model includes: acquiring historical multi-source data, historical pollution source locations, and historical hydrogeological data; using the historical multi-source data, the historical pollution source locations, and the historical hydrogeological data to establish first sample data containing spatial correlation features and second sample data containing temporal features, respectively; constructing a pollution source location model framework based on a Bayesian network, and training the pollution source location model framework using the first sample data to obtain the pollution source location model; constructing a pollution diffusion model framework based on a long short-term memory network, and training the pollution diffusion model framework using the second sample data to obtain the pollution diffusion model.

[0013] This invention, by acquiring historical multi-source data, pollution source locations, and hydrogeological data, can fully mine relevant information about past pollution, providing rich and reliable samples for model construction. Using this data, a first sample dataset containing spatial correlation features and a second sample dataset containing temporal features are established, accurately adapting to different model requirements. A pollution source localization model based on Bayesian networks can effectively handle spatial uncertainty and achieve precise location of pollution sources; a pollution diffusion model based on Long Short-Term Memory networks can capture the dynamic changes in pollution over time, improving the scientific rigor of the model construction.

[0014] Optionally, assigning index weights to the pollution indicators in the multi-source data includes: constructing an initial matrix with each group of data in the pre-obtained historical multi-source dataset as rows and the pollution indicators in the multi-source data as columns; normalizing the initial matrix to obtain a normalized matrix; calculating the entropy value of each pollution indicator based on the normalized matrix; and calculating the index weight of each pollution indicator based on the entropy value.

[0015] This invention, by constructing an initial matrix and normalizing it, eliminates differences in the dimensions and magnitudes of various pollution indicators, making the data comparable. Based on the normalized matrix, entropy values ​​are calculated, and the principle of information entropy is used to quantify the degree of information disorder of each pollution indicator, objectively reflecting its relative importance in characterizing pollution status. Indicator weights are calculated based on entropy values, avoiding subjective arbitrariness and ensuring that weight allocation is based on the inherent characteristics of the data. The weights determined in this way can reasonably reflect the contribution of each pollution indicator when fusing multi-source data, improving the accuracy of comprehensive pollution assessment.

[0016] Optionally, generating a comprehensive pollution feature vector based on the indicator weights using the multi-source data includes: normalizing the multi-source data to obtain normalized multi-source data; and performing a weighted operation between the indicator weights and the normalized multi-source data to form a comprehensive pollution feature vector.

[0017] This invention eliminates dimensional differences by normalizing multi-source data, making different indicators comparable. Weighted calculations based on indicator weights highlight key pollution indicators and avoid subjective bias. The generated comprehensive pollution feature vector integrates multi-source data information to form a structured input, providing a reliable basis for pollution alarm triggering, pollution source location, and diffusion prediction, thereby improving the accuracy of the monitoring system's pollution status assessment and model prediction precision.

[0018] Optionally, triggering a pollution alarm based on the comprehensive pollution feature vector includes: calculating a comprehensive pollution probability using the comprehensive pollution feature vector; setting an alarm threshold and comparing the alarm threshold with the comprehensive pollution probability; and triggering a pollution alarm based on the comparison result.

[0019] This invention calculates pollution probability by integrating pollution feature vectors and combines them with dynamic thresholds to achieve automated alerts, reducing the response time from hours to minutes in traditional manual analysis and solving the problem of delayed pollution warnings. The quantitative probability assessment and threshold comparison logic avoids subjective judgment bias and improves the response time to groundwater pollution.

[0020] Optionally, calculating the comprehensive pollution probability using the comprehensive pollution feature vector includes: constructing a Gaussian mixture model based on pre-acquired historical multi-source data; performing parameter learning on the Gaussian mixture model using the expectation-maximization algorithm to obtain the mixing coefficients, mean vector, and covariance matrix for the clean state and the dirty state, respectively; substituting the mixing coefficients, the mean vector, the covariance matrix, and the comprehensive pollution feature vector into the Gaussian mixture model to obtain the probability density of the clean state and the probability density of the dirty state, respectively; and calculating the comprehensive pollution probability using the probability density of the clean state and the probability density of the dirty state.

[0021] This invention constructs a dual-state probabilistic model using historical multi-source data and incorporates an expectation-maximization algorithm to achieve adaptive learning of model parameters, enabling precise characterization of the distribution features of multi-dimensional indicators under clean and polluted states. This process quantifies the probability density of different states through mixing coefficients, mean vectors, and covariance matrices, avoiding the limitations of single-indicator threshold judgments and effectively capturing the coupling effects of multi-component pollution. The probability values ​​calculated based on the comprehensive pollution feature vector objectively reflect the likelihood of pollution occurrence. Combined with dynamic thresholds, this achieves a probabilistic assessment of pollution states, improving the scientific rigor of groundwater pollution assessment.

[0022] Optionally, the overall pollution probability satisfies the following formula:

[0023]

[0024] in, To comprehensively consider the probability of pollution, The mixing coefficient under polluted conditions. Let Gaussian probability density function be used. To form a comprehensive pollution feature vector, Let be the mean vector under polluted conditions. Let be the covariance matrix under pollution conditions. The mixing coefficient is the value of the mixture in the clean state. Let be the mean vector under clean conditions. Let be the covariance matrix under clean conditions.

[0025] The comprehensive pollution probability formula of this invention is based on a Gaussian mixture model, utilizing mixture coefficients, mean vector, and covariance matrix to scientifically quantify the likelihood of pollution states. By comprehensively considering the characteristics of multi-source data under both clean and polluted states, it overcomes the limitations of single indicators and improves the scientific rigor and accuracy of the comprehensive pollution probability calculation.

[0026] In another aspect, the present invention provides a dynamic intelligent monitoring system for groundwater pollution, comprising: a processor, an input device, an output device, and a memory, wherein the processor, the input device, the output device, and the memory are interconnected, wherein the memory is used to store a computer program, the computer program including program instructions, and the processor is configured to invoke the program instructions to execute a dynamic intelligent monitoring method for groundwater pollution as described in any of the preceding aspects of the present invention.

[0027] The present invention provides a groundwater pollution dynamic intelligent monitoring system with a compact structure, stable performance, high integration and simple configuration. It can stably execute the groundwater pollution dynamic intelligent monitoring method provided in the preceding aspect of the present invention, further improving the overall applicability and practical application capability of the present invention. Attached Figure Description

[0028] Figure 1 This is a flowchart of a method for dynamic intelligent monitoring of groundwater pollution according to an embodiment of the present invention;

[0029] Figure 2 This is a schematic diagram of the structure of a groundwater pollution dynamic intelligent monitoring system according to an embodiment of the present invention. Detailed Implementation

[0030] Specific embodiments of the present invention will now be described in detail. It should be noted that the embodiments described herein are for illustrative purposes only and are not intended to limit the invention. In the following description, numerous specific details are set forth in order to provide a thorough understanding of the invention. However, it will be apparent to those skilled in the art that these specific details are not necessary to practice the invention. In other instances, well-known circuits, software, or methods have not been specifically described to avoid obscuring the invention.

[0031] Throughout this specification, references to "an embodiment," "an embodiment," "an example," or "an example" mean that a particular feature, structure, or characteristic described in connection with that embodiment or example is included in at least one embodiment of the invention. Therefore, the phrases "in an embodiment," "in an embodiment," "an example," or "an example" appearing in various places throughout the specification do not necessarily refer to the same embodiment or example. Furthermore, specific features, structures, or characteristics can be combined in one or more embodiments or examples in any suitable combination and / or sub-combination. Moreover, those skilled in the art will understand that the illustrations provided herein are for illustrative purposes and are not necessarily drawn to scale.

[0032] Please see Figure 1 To address the problem of insufficient timeliness and accuracy in groundwater pollution monitoring in existing technologies, in one optional embodiment, such as Figure 1The method for dynamic intelligent monitoring of groundwater pollution, as shown, includes the following steps:

[0033] Step S1: Obtain real-time hydrogeological data of the target area, and construct and update the groundwater solute transport model based on the real-time hydrogeological data.

[0034] In this embodiment, the above steps include: acquiring real-time hydrogeological data of the target area, and constructing a hydrogeological model using the real-time hydrogeological data; and constructing the groundwater solute transport model using the hydrogeological model based on the test data of leaching and soaking tests.

[0035] The target area can be near industrial plants or areas with significant pollution risks. These areas are usually polluted to varying degrees. Real-time hydrogeological data such as groundwater flow velocity, water level, and permeability are collected using distributed fiber optic sensors, electromagnetic current meters, and other equipment. GIS tools such as ArcGIS are used to unify spatial coordinates and perform grid interpolation on the data. After outliers are removed and filtered, the data is input into MODFLOW-NWT software to construct a three-dimensional hydrogeological model. Darcy's law is used to solve the flow equation, generating a flow field that includes the flow direction and velocity distribution. Simultaneously, leaching and soaking tests are conducted to obtain key parameters such as pollutant release rate, soil-rock distribution coefficient, and dispersion coefficient. The test data are coupled with the hydrogeological model, and the flow field results from the hydrogeological model are called through MT3DMS software. A groundwater solute transport model is constructed by combining the convection-diffusion equation.

[0036] Step S2: Obtain multi-source data related to groundwater pollution, including various pollution indicators, and fuse the multi-source data to generate a comprehensive pollution feature vector.

[0037] The acquisition of multi-source data related to groundwater pollution, including various pollution indicators, specifically includes the following steps:

[0038] Step S211: Determine potential migration paths of pollution based on the groundwater solute transport model, and deploy distributed fiber optic sensors, microelectrode monitoring wells and ground network monitoring nodes according to the potential migration paths of pollution to form a monitoring network.

[0039] In this embodiment, based on the established groundwater solute transport model, the core path and sensitive areas of the pollution plume migration (such as weak points in the impermeable layer and spring outcrops) are extracted by simulating the spatiotemporal diffusion process of pollutants in groundwater. Based on this, a monitoring network is deployed: First, microelectrode monitoring wells are deployed every 50-100 meters along the water flow direction on the horizontal diffusion path of the pollution plume predicted by the model. The wells penetrate the entire thickness of the aquifer and are used to collect pollutant concentration data at different depths in real time. At the same time, the wells are densely deployed in the area 20% outside the pollution impact boundary in the vertical water flow direction to form a planar grid-like monitoring array.

[0040] Secondly, ground-based network monitoring nodes are deployed in the corresponding pollution plume projection area and around potential pollution sources, equipped with multi-parameter water quality analyzers (real-time monitoring of pH, conductivity, and characteristic pollutant concentrations) to construct a surface monitoring layer. Finally, distributed fiber optic sensors are laid along the flow direction at key groundwater flow control interfaces identified by the model (such as fault zones and lithological abrupt change zones). By monitoring changes in temperature field and flow velocity disturbances, continuous tracking of the vertical migration path of the pollution plume is achieved. Through the spatial collaborative deployment of these multi-dimensional devices, a monitoring network covering both surface and groundwater is formed. Its density is dynamically matched to the pollution risk level predicted by the model (monitoring spacing ≤ 50 meters in high-risk areas and ≤ 200 meters in low-risk areas), ensuring that the entire pollution migration process is monitorable and can be predicted.

[0041] Step S212: Use the monitoring network to obtain monitoring data, including groundwater flow velocity, groundwater temperature, concentrations of various groundwater pollutants, and concentrations of various surface water pollutants.

[0042] In this embodiment, the distributed nodes on the ground collect parameters such as pollution concentration (COD, ammonia nitrogen, heavy metals, etc.), pH value, and conductivity of surface water bodies (such as ditches and ponds) in real time through a multi-parameter water quality detector. The sampling frequency is once per hour, as well as rainfall.

[0043] The underground microelectrode monitoring well obtains groundwater samples at different depths through a stratified sampler and uses an online water quality sensor to monitor the concentration of characteristic pollutants such as nitrate, chloride, and volatile organic compounds in real time, with a sampling interval of 30 minutes.

[0044] Distributed fiber optic sensors monitor the Raman scattering signal of laser pulses in optical fibers, simultaneously acquiring groundwater temperature (accuracy ±0.1℃) and flow velocity disturbance data (resolution 0.01m / d) along the path, thus enabling continuous monitoring of groundwater dynamic parameters.

[0045] Step S213: Perform spatiotemporal alignment and feature extraction on the monitoring data to form multi-source data related to groundwater pollution, which includes multiple pollution indicators.

[0046] In this embodiment, firstly, a unified spatial reference is established, aligning the coordinates of the microelectrode monitoring wells and ground network monitoring nodes, and converting them to the National Geodetic Coordinate System 2000 to ensure a planar position error of ≤0.1 meters. Secondly, a regular spatial grid is constructed, dividing the target area into 20m × 20m units. For parameters such as pollution concentration and groundwater flow velocity from the microelectrode monitoring wells (point data) and ground nodes, continuous 20m grid data is generated based on Kriging interpolation with a groundwater flow direction weighting factor, expanding the point monitoring data into a planar distribution. For the monitoring data along the path of the distributed fiber optic sensors, equidistant sampling is performed at 20m intervals according to the spatial coordinates corresponding to the fiber optic laying path to match the overall grid accuracy.

[0047] For groundwater flow velocity data, the velocity disturbance coefficient (measured velocity / background velocity) is calculated, and grids with a disturbance coefficient > 1.2 are extracted as active pollution migration paths. For groundwater and surface water pollution concentration data, the exceedance multiple (measured concentration / standard limit) and concentration gradient direction (determined by the concentration difference between adjacent grids) are calculated, and concentration peak grids are marked. Simultaneously, temperature gradient abrupt change points (ΔT2 / ΔL > 0.5℃ / 100m) in groundwater temperature data are extracted and associated with the vertical migration interface of the pollution plume.

[0048] Finally, more than 20 pollution indicators, such as spatially aligned ΔT values, flow velocity disturbance coefficients, and exceedance multiples, were integrated into a multidimensional dataset according to the "spatial grid ID + pollution indicator" structure and normalized. Each grid contains hydrodynamic features and pollution concentration features, forming standardized multi-source data covering the entire region, providing structured input for subsequent comprehensive pollution feature vector generation and model training.

[0049] These aligned pollution indicators will then be time-aligned. Since different monitoring data have different time frequencies, the average value of the same indicator over a period of time (such as one hour, set according to the requirements of groundwater monitoring) will be calculated.

[0050] The step of fusing the multi-source data to generate a comprehensive pollution feature vector specifically includes the following sub-steps:

[0051] Step S221: Assign index weights to the pollution indicators in the multi-source data.

[0052] The specific steps of assigning index weights to the pollution indicators in the multi-source data include the following sub-steps:

[0053] Step S22101: Construct an initial matrix using each group of data in the pre-obtained historical multi-source dataset as rows and the pollution index in the multi-source data as columns.

[0054] In this embodiment, a large amount of historical multi-source datasets were obtained by utilizing the real-time monitoring function of the monitoring network. Therefore, the monitoring indicators of the multi-source data or historical multi-source data are the same. All monitoring indicators in the multi-source data are extracted as matrix columns, such as groundwater flow velocity and COD concentration, forming a set of fixed column names. Then, each spatiotemporal sample (spatial grid or monitoring well data at the same time point) in the historical data is defined as a matrix row. Each sample must contain a unique timestamp and spatial coordinates. The data is filled in according to chronological order and spatial distribution to form an initial matrix.

[0055] Step S22102: Normalize the initial matrix to obtain a normalized matrix.

[0056] In this embodiment, the maximum and minimum values ​​of each column index in the initial matrix are first calculated. Then, each element is normalized using the min-max normalization formula based on the maximum and minimum values, compressing the index value range to the [0,1] interval. If there is an extreme case where all index values ​​are 0, the default normalization is 0.5 to preserve the information entropy. Finally, the normalized matrix is ​​obtained.

[0057] Step S22103: Calculate the entropy value of each pollution index based on the normalization matrix.

[0058] The entropy value satisfies the following formula:

[0059]

[0060] in, For the first The entropy value of various pollution indicators The number of data points in the historical multi-source dataset. For the first The first sample Normalized values ​​of various pollution indicators.

[0061] Step S22104: Calculate the index weight of each pollution index based on the entropy value.

[0062] The weights of the indicators satisfy the following formula:

[0063]

[0064] in, For the first The weight of each pollution indicator For the first The entropy value of various pollution indicators The types of pollution indicators, For the first The entropy value of a pollution indicator.

[0065] Step S222: Generate a comprehensive pollution feature vector based on the multi-source data using the index weights.

[0066] Specifically, the step of generating a comprehensive pollution feature vector based on the indicator weights and the multi-source data includes the following sub-steps:

[0067] Step S22201: Normalize the multi-source data to obtain normalized multi-source data.

[0068] In this embodiment, for different types of pollution indicators (such as groundwater flow velocity, groundwater temperature, and pollutant concentration) in multi-source data, a corresponding normalization method (such as min-max normalization or Z-score standardization) is selected based on their physical dimensions and numerical ranges. Individual data for each pollution indicator is normalized one by one, scaling its value to a preset uniform range to eliminate the impact of dimensional differences on data fusion, thus forming normalized multi-source data with comparable values ​​across all indicator dimensions.

[0069] Step S22202: Perform a weighted operation on the indicator weights and the normalized multi-source data to form a comprehensive pollution feature vector.

[0070] In this embodiment, the value of each pollution indicator in the normalized multi-source data is multiplied by its corresponding indicator weight to obtain the weighted value of each indicator. Then, according to the indicator order preset when the weighted monitoring network is deployed (such as the fixed arrangement order of indicators such as groundwater flow velocity, groundwater temperature, and concentration of various pollutants), the weighted values ​​of all indicators are arranged in sequence to form a comprehensive pollution feature vector containing the weighted features of multiple pollution indicators.

[0071] Step S3: Construct the pollution source location model and the pollution diffusion model respectively.

[0072] The construction of the pollution source location model and the pollution diffusion model includes:

[0073] Step S301: Obtain historical multi-source data, historical pollution source locations, and historical hydrogeological data.

[0074] In this embodiment, historical multi-source data related to groundwater pollution in the target area over a historical period are first obtained. The historical multi-source data is consistent with the above-mentioned multi-source data types, including historical groundwater flow velocity, historical groundwater temperature, various historical groundwater pollution concentrations, and various historical surface water pollution concentrations. The historical multi-source data is then normalized.

[0075] The spatial coordinates of pollution sources actually recorded in historical time periods are acquired simultaneously, i.e., historical pollution source locations, as well as historical hydrogeological data consistent with the spatiotemporal benchmark of historical multi-source data. The historical hydrogeological data includes at least dynamic time-series parameters such as historical groundwater flow velocity and historical water level. The three are fully aligned on timestamps and spatial grids, and together they constitute a sample dataset for training pollution source location models and pollution diffusion models.

[0076] Step S302: Using the historical multi-source data, the historical pollution source locations, and the historical hydrogeological data, establish first sample data containing spatial correlation features and second sample data containing temporal features, respectively.

[0077] In this embodiment, spatial location-related features in historical multi-source data, such as the spatial distribution of pollution concentration at each monitoring node and the coordinates of monitoring wells, are associated with the spatial coordinates of historical pollution source locations. Spatial correlation features such as spatial distance and relative position in the direction of water flow are extracted. Combined with static spatial parameters in historical hydrogeological data (such as spatial distribution of permeability and aquifer structure), the first sample data is constructed to form a spatial correlation dataset.

[0078] Historical multi-source data is expanded according to time series, and the location of historical pollution sources and dynamic time series parameters (such as hourly groundwater flow velocity and water level fluctuation sequence) in historical hydrogeological data at corresponding time points are superimposed to construct a second sample data, forming a time-related dataset.

[0079] Step S303: Construct a pollution source localization model framework based on a Bayesian network, and train the pollution source localization model framework using the first sample data to obtain a pollution source localization model.

[0080] In this embodiment, a Bayesian network of node causal relationships is first constructed using historical hydrogeological data. This data includes dynamic time-series parameters such as historical groundwater flow velocity and water level, as well as static features such as the spatial distribution of permeability. This data is used to define the directed acyclic graph structure between network nodes. Core nodes include the location of the pollution source, pollution characteristics at the monitoring point, and groundwater flow direction. The causal chain is set according to hydrogeological principles, such as "groundwater flow direction → pollution source location → pollution characteristics at the monitoring point," ensuring that the pollution source location is always upstream of the monitoring point's water flow. This can be constrained by historical water flow direction data. Auxiliary nodes, such as spatial distance (Euclidean distance between the pollution source and the monitoring point) and hydraulic gradient, are extracted from historical data and used as intermediate variables to connect the core nodes, strengthening the spatial correlation logic.

[0081] Then, the first sample data is input into the Bayesian network framework for parameter training. The historical multi-source data is discretized (e.g., continuous variables such as pollution concentration and spatial distance are divided into different level intervals). The node causal relationship of the Bayesian network is constructed based on historical hydrogeological data. Then, the prior probability of the pollution source location (e.g., setting the probability of pollution occurrence in different areas according to historical distribution) and the probability distribution of the flow direction nodes are initialized. Then, using the first sample data, the conditional probability parameters of the Bayesian network (e.g., the probability of pollution characteristics appearing at monitoring points under different combinations of spatial distance and flow direction) are trained and optimized through algorithms such as maximum likelihood estimation. At the same time, causal edges that violate physical laws are removed by combining historical hydrogeological data, and finally, the pollution source location model is obtained.

[0082] Step S304: Construct a pollution diffusion model framework based on a long short-term memory network, and train the pollution diffusion model framework using the second sample data to obtain a pollution diffusion model.

[0083] The composition of the second sample data is clearly defined. It is a dataset composed of historical multi-source data (including hourly pollution index values), historical pollution source location coordinates (such as latitude and longitude) at corresponding time points, and historical hydrogeological dynamic parameters (such as hourly groundwater flow velocity and water level fluctuation sequences) aligned in time series. It needs to be divided into input features and output labels. The input features include pollution feature vectors, temporal changes in pollution source locations, and hydrological parameters. The output labels are the pollution distribution features at future times.

[0084] Next, the LSTM model framework is constructed: a multi-layer LSTM network structure is adopted, the number of neurons in the input layer corresponds to the dimension of the input features, the hidden layer is set with several LSTM units to capture temporal dependencies, and the output layer is set with the corresponding number of neurons and linear activation function according to the prediction target.

[0085] Then, the input data is preprocessed: continuous features such as pollution concentration and flow rate are normalized, timestamps are one-hot encoded to represent the temporal sequence, and the data is divided into sample sets with fixed time steps according to time windows.

[0086] During the training phase, the preprocessed second sample data is input into the model, the mean squared error is used as the loss function, the Adam optimizer is used to update the parameters, and the network weights are adjusted through the backpropagation algorithm. An early stopping mechanism is set to avoid overfitting, and the Dropout layer is used to randomly deactivate some neurons to enhance the model's generalization ability.

[0087] After training, the model performance is evaluated using a test set, and indices such as mean absolute error and coefficient of determination are calculated to verify its ability to capture the temporal characteristics of pollution diffusion. Finally, a pollution diffusion model is obtained that can predict the future distribution of pollution based on current hydrogeological data, pollution source location, and temporal pollution characteristics.

[0088] During the system's operation, a large amount of pollution data will be gradually obtained in the later stages. The pollution diffusion model will be updated and trained using the same training data format to gradually improve the predictive performance of the pollution diffusion model.

[0089] Step S4: Trigger a pollution alarm based on the comprehensive pollution feature vector, and extract the current hydrogeological data from the real-time hydrogeological data based on the pollution alarm.

[0090] The step of triggering a pollution alarm based on the comprehensive pollution feature vector includes:

[0091] Step S401: Calculate the comprehensive pollution probability using the comprehensive pollution feature vector.

[0092] The calculation of the comprehensive pollution probability using the comprehensive pollution feature vector specifically includes the following sub-steps:

[0093] Step S40101: Construct a Gaussian mixture model based on pre-acquired historical multi-source data.

[0094] In this embodiment, historical multi-source data containing multiple pollution indicators is extracted from historical multi-source data (the method is consistent with the construction of the comprehensive pollution feature vector described above), forming a historical comprehensive pollution feature vector. Combined with historical monitoring reports and groundwater quality standards (such as GB / T14848), the clean or polluted state labels corresponding to each feature vector are labeled using expert experience. The parameters of the Gaussian mixture model are initialized, and the model is set to contain two Gaussian components, representing the clean state and the polluted state, respectively. The mixing coefficients of each component are randomly initialized. ( ,satisfy ), mean vector and covariance matrix.

[0095] Step S40102: The Gaussian mixture model is trained using the expectation-maximization algorithm to obtain the mixing coefficients, mean vector and covariance matrix of the clean state and the dirty state, respectively.

[0096] In this embodiment, the Gaussian mixture model includes a clean state ( ) and pollution status ( The two Gaussian components, each composed of a mixing coefficient, are given by the following formula: Mean vector Covariance Matrix The process then proceeds in step E, for each integrated pollution feature vector from the historical multi-source data. Calculate the posterior probability of each state based on the current parameters. The calculation formula is as follows:

[0097]

[0098] For the first The sample belongs to the first The posterior probability of a state. For the first Historical multi-source data, For the first The mixing coefficient of the state For the first The mean vector of each state For the first The covariance matrix of each state This is a Gaussian probability density function, and the other subscripts are interpreted in the same way for the corresponding variables.

[0099] The Gaussian probability density function satisfies the following formula:

[0100]

[0101] Then, in the M-step, the parameters, including the mixing coefficients, are updated based on the posterior probabilities of all samples. The update formula is as follows:

[0102]

[0103] Mean vector The update formula is as follows:

[0104]

[0105] Mean vector The update formula is as follows:

[0106]

[0107] in, Given the total number of historical samples, repeat the E-step and M-step until the log-likelihood function converges or reaches the preset number of times, finally obtaining the mixing coefficients, mean vector, and covariance matrix of the clean and contaminated states.

[0108] Step S40103: Substitute the mixing coefficient, the mean vector, the covariance matrix, and the comprehensive pollution feature vector into the Gaussian mixture model to obtain the probability density of the clean state and the probability density of the dirty state, respectively.

[0109] In this embodiment, the probability density of the clean state satisfies the following formula:

[0110]

[0111] in, Let be the probability density of the clean state. To form a comprehensive pollution feature vector, Let be the mean vector of the clean state. Let be the covariance matrix of the clean state. Comprehensive pollution feature vector Dimensions for The determinant of the array, for The inverse matrix.

[0112] The probability density of the contaminated state satisfies the following formula:

[0113]

[0114] in, Let be the probability density of the clean state. To form a comprehensive pollution feature vector, Let be the mean vector of the contaminated state. Let be the covariance matrix of the pollution state. Comprehensive pollution feature vector Dimensions for The determinant of the array, for The inverse matrix.

[0115] Step S40104: Calculate the overall contamination probability using the probability density of the clean state and the probability density of the dirty state.

[0116] The overall pollution probability satisfies the following formula:

[0117]

[0118] in, To comprehensively consider the probability of pollution, The mixing coefficient under polluted conditions. Let Gaussian probability density function be used. To form a comprehensive pollution feature vector, Let be the mean vector under polluted conditions. Let be the covariance matrix under pollution conditions. The mixing coefficient is the value of the mixture in the clean state. Let be the mean vector under clean conditions. Let be the covariance matrix under clean conditions.

[0119] The fundamental basis for calculating the comprehensive pollution probability using the comprehensive pollution feature vector is that there are quantifiable probability distribution differences between the comprehensive pollution feature vectors under clean and polluted states in historical multi-source data. After learning the distribution parameters of these differences through a Gaussian mixture model, the probability of the current comprehensive pollution feature vector belonging to a polluted state can be inferred based on the probability density of the current comprehensive pollution feature vector in the two distributions.

[0120] Step S402: Set an alarm threshold and compare the alarm threshold with the comprehensive pollution probability.

[0121] In this embodiment, the 95th percentile can be calculated based on the comprehensive pollution probability distribution of pollution status in historical multi-source data, or the risk level can be divided in conjunction with standards such as the Groundwater Quality Standard (e.g., setting a low risk threshold of 0.5 and a high risk threshold of 0.8), and a dynamic adjustment mechanism can be established (e.g., updating the threshold quarterly based on new data or lowering it during the rainy season).

[0122] Step S403: Based on the comparison result, trigger a pollution alarm.

[0123] In this embodiment, after setting the threshold, the real-time calculated comprehensive pollution probability is compared with the threshold: if the comprehensive pollution probability is greater than or equal to the threshold, it is determined to be a polluted state, triggering the corresponding level alarm, and immediately extracting parameters such as current groundwater flow velocity and water level from real-time hydrogeological data; if the comprehensive pollution probability is less than the threshold, it is determined to be a low-risk or clean state, and continuous real-time monitoring is performed. This process achieves refined early warning of pollution risk and triggering of emergency response through precise threshold setting and dynamic comparison.

[0124] Upon receiving the alarm, the current hydrogeological data is extracted from the real-time acquired hydrogeological data and normalized. The current hydrogeological data refers to the hydrogeological data from the last update of the groundwater solute transport model.

[0125] Step S5: Based on the multi-source data and the current hydrogeological data, the pollution source location is obtained using the pollution source location model.

[0126] In this embodiment, multi-source data is input as observational evidence into a pre-trained pollution source location model. Simultaneously, dynamic parameters such as current groundwater flow direction and velocity are extracted from current hydrogeological data to update the state of corresponding nodes in the Bayesian network. Based on node causal relationships, the posterior probability distribution of the pollution source location nodes is calculated using a Bayesian inference algorithm. Combined with the physical constraints of the current hydrogeological data (e.g., the pollution source must be located upstream of the monitoring point), the spatial region with the highest probability is selected, and the coordinates of this region are determined as the current pollution source location. This process achieves probabilistic inference and spatial location of the pollution source by fusing real-time pollution characteristics with dynamic hydrological conditions.

[0127] Step S6: Based on the current hydrogeological data, the location of the pollution source, and the multi-source data, the pollution diffusion model is used to predict the first pollution distribution that changes over time.

[0128] In this embodiment, firstly, the real-time acquired hydrogeological data, the spatial coordinates of the pollution source obtained through the pollution source location model, and multi-source data are preprocessed to unify the timestamps and spatial coordinate systems to meet the model input requirements. Next, the preprocessed multidimensional data is input into a pollution diffusion model constructed based on a Long Short-Term Memory (LSTM) network. The model uses LSTM units to capture the evolution patterns of pollution characteristics over time, calculates the pollution migration path by combining the water flow direction and velocity parameters from the current hydrogeological data, and uses the pollution source location as an initial constraint on the diffusion starting point. Finally, the model outputs the first pollution distribution over a specified future time period, specifically represented as a sequence of predicted pollution index concentrations for each spatiotemporal grid point.

[0129] Step S7: Based on the location of the pollution source and the multi-source data, the groundwater solute transport model is used to simulate and obtain the second pollution distribution that changes over time.

[0130] The spatial coordinates of the pollution source obtained through the pollution source location model are used as the initial location for pollutant release. The initial release intensity and component proportions of the pollutants are then determined by combining multi-indicator pollution concentration values ​​from multi-source data. Next, a pre-built and updated groundwater solute transport model is invoked. This model uses software such as MODFLOW-NWT and MT3DMS to couple the calculation of the water flow field and solute transport. Based on the convection-diffusion equation, and combined with the initial and boundary conditions of the pollution source location (such as groundwater boundary velocity), the model simulates the migration and diffusion process of pollutants in groundwater over time, considering the influence of physicochemical processes such as adsorption, dispersion, and biodegradation on pollution distribution. Finally, the second pollution distribution over a specified future time period is output.

[0131] Step S8: The first pollution distribution and the second pollution distribution are weighted and fused to generate a predicted pollution distribution that changes over time.

[0132] The predicted pollution distribution satisfies the following formula:

[0133]

[0134] in, To predict pollution distribution, The first pollution distribution weight, The second pollution distribution, As the weight of the second pollution distribution, It is the first pollution distribution. For predicting hourly values.

[0135] In this embodiment, the weights of the first pollution distribution and the second pollution distribution can be assigned according to expert experience, or according to the maturity of the pollution diffusion model. Since the pollution diffusion model is gradually improved during the operation of this invention, the prediction effect will gradually increase. The maturity of the pollution diffusion model can be determined according to the test results of the pollution diffusion model.

[0136] In one alternative embodiment, based on past experience, the predictive effect of the groundwater solute transport model is quite good in the short term. However, the transfer of groundwater itself is a particularly complex process. Due to the imperfection of geological data collection, the predictive effect gradually decreases over time. Therefore, this invention also proposes a method for determining the weights. Based on experience, pollution diffusion can be divided into three days. In the early stage, the groundwater solute transport model is mainly relied upon, while in the later stage, the pollution diffusion model is mainly relied upon.

[0137] The weights can be determined by the following formula:

[0138]

[0139]

[0140] like Figure 2 As shown, the present invention also provides a dynamic intelligent monitoring system for groundwater pollution, comprising: a processor, an input device, an output device, and a memory, wherein the processor, the input device, the output device, and the memory are interconnected, wherein the memory is used to store a computer program, the computer program including program instructions, and the processor is configured to call the program instructions to execute relevant steps of a relevant embodiment of the dynamic intelligent monitoring method for groundwater pollution of the present invention.

[0141] This invention provides a dynamic intelligent monitoring system for groundwater pollution. The functional components can be integrated into a single processing unit, or each component can exist independently, or two or more components can be integrated into one unit. The integrated components can be implemented in hardware or software.

[0142] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. A method for dynamic intelligent monitoring of groundwater pollution, characterized in that, The method includes: Acquire real-time hydrogeological data of the target area, and construct and update the groundwater solute transport model based on the real-time hydrogeological data; Acquire multi-source data related to groundwater pollution, including various pollution indicators, and fuse the multi-source data to generate a comprehensive pollution feature vector; Construct pollution source location models and pollution diffusion models respectively, including: Acquire historical multi-source data, historical pollution source locations, and historical hydrogeological data; Using the historical multi-source data, the historical pollution source locations, and the historical hydrogeological data, a first sample data containing spatial correlation features and a second sample data containing temporal features are respectively established. A pollution source localization model framework is constructed based on a Bayesian network, and the pollution source localization model framework is trained using the first sample data to obtain a pollution source localization model. A pollution diffusion model framework is constructed based on a long short-term memory network, and the pollution diffusion model framework is trained using the second sample data to obtain a pollution diffusion model. A pollution alarm is triggered based on the comprehensive pollution feature vector, and the current hydrogeological data is extracted from the real-time hydrogeological data based on the pollution alarm. The location of the pollution source is obtained using the pollution source location model based on the multi-source data and the current hydrogeological data. Based on the current hydrogeological data, the location of the pollution source, and the multi-source data, the pollution diffusion model is used to predict the first pollution distribution that changes over time. Based on the pollution source locations and the multi-source data, the groundwater solute transport model was used to simulate the second pollution distribution that changes over time; The first pollution distribution and the second pollution distribution are weighted and fused to generate a predicted pollution distribution that changes over time.

2. The method for dynamic intelligent monitoring of groundwater pollution according to claim 1, characterized in that, The acquisition of multi-source data related to groundwater pollution, including various pollution indicators, includes: Based on the groundwater solute transport model, potential pollution migration paths are determined, and distributed fiber optic sensors, microelectrode monitoring wells, and ground network monitoring nodes are deployed according to the potential pollution migration paths to form a monitoring network; The monitoring network is used to obtain monitoring data, including groundwater flow velocity, groundwater temperature, concentrations of various groundwater pollutants, and concentrations of various surface water pollutants. The monitoring data is spatiotemporally aligned and feature extracted to form multi-source data related to groundwater pollution, which includes various pollution indicators.

3. The method for dynamic intelligent monitoring of groundwater pollution according to claim 1, characterized in that, The process of fusing the multi-source data to generate a comprehensive pollution feature vector includes: Assign index weights to the pollution indicators in the multi-source data; Based on the aforementioned indicator weights, a comprehensive pollution feature vector is generated using the multi-source data.

4. The method for dynamic intelligent monitoring of groundwater pollution according to claim 3, characterized in that, The process of assigning index weights to the pollution indicators in the multi-source data includes: An initial matrix is ​​constructed using each group of data in the pre-obtained historical multi-source dataset as rows and the pollution indicators in the multi-source dataset as columns; The initial matrix is ​​normalized to obtain a normalized matrix; Calculate the entropy value of each pollution index based on the normalized matrix; The index weight of each pollution index is calculated based on the entropy value.

5. The method for dynamic intelligent monitoring of groundwater pollution according to claim 3, characterized in that, The step of generating a comprehensive pollution feature vector based on the indicator weights and the multi-source data includes: The multi-source data is normalized to obtain normalized multi-source data; The weights of the indicators are weighted together with the normalized multi-source data to form a comprehensive pollution feature vector.

6. The method for dynamic intelligent monitoring of groundwater pollution according to claim 1, characterized in that, The step of triggering a pollution alarm based on the comprehensive pollution feature vector includes: The comprehensive pollution probability is calculated using the comprehensive pollution feature vector. Set an alarm threshold and compare the alarm threshold with the overall pollution probability; Based on the results of the comparison, a pollution alarm is triggered.

7. The method for dynamic intelligent monitoring of groundwater pollution according to claim 6, characterized in that, The calculation of the comprehensive pollution probability using the comprehensive pollution feature vector includes: A Gaussian mixture model is constructed based on pre-acquired historical multi-source data; The Gaussian mixture model was trained using the expectation-maximization algorithm to obtain the mixing coefficients, mean vector, and covariance matrix for the clean and dirty states, respectively. Substituting the mixing coefficient, the mean vector, the covariance matrix, and the comprehensive pollution feature vector into the Gaussian mixture model, the probability density of the clean state and the probability density of the dirty state are obtained respectively. The overall contamination probability is calculated using the probability density of the clean state and the probability density of the dirty state.

8. The method for dynamic intelligent monitoring of groundwater pollution according to claim 7, characterized in that, The overall pollution probability satisfies the following formula: in, To comprehensively consider the probability of pollution, The mixing coefficient under polluted conditions. Let Gaussian probability density function be used. To form a comprehensive pollution feature vector, Let be the mean vector under polluted conditions. Let be the covariance matrix under pollution conditions. The mixing coefficient is the value of the mixture in the clean state. Let be the mean vector under clean conditions. Let be the covariance matrix under clean conditions.

9. A dynamic intelligent monitoring system for groundwater pollution, characterized in that, include: The system includes a processor, an input device, an output device, and a memory, all interconnected. The memory stores a computer program, which includes program instructions. The processor is configured to invoke the program instructions to execute a dynamic intelligent monitoring method for groundwater pollution as described in any one of claims 1 to 8.

Images