An artificial intelligence-based groundwater level dynamic monitoring method and system

Abstract

The application provides a groundwater level dynamic monitoring method and system based on artificial intelligence, which constructs water head time sequence and water level change sequence of each layered aquifer in a target area through original pressure monitoring signals and static water level calibration data; extracts transient change characteristics of groundwater head of each layered aquifer, and determines hydraulic coupling characteristics between different layered aquifers according to the transient change characteristics; determines influence coefficients of different environmental factors on water level changes of each layered aquifer in the process of groundwater level change in the target area based on multi-source environmental dynamic data affecting the groundwater level fluctuation; and intelligently fuses the groundwater level of the target area by using the hydraulic coupling characteristics and all influence coefficients to generate a dynamic prediction result of the groundwater level of the internal transmission process and the external environment driving process of the groundwater. By using the scheme, the collaborative simulation and explainable accurate prediction of the internal transmission process and the external environment driving process of the groundwater can be completed under the condition of multi-process coupling.

Description

Technical Field

[0001] This application relates to the field of groundwater level monitoring technology, and more specifically, to a method and system for dynamic monitoring of groundwater levels based on artificial intelligence. Background Technology

[0002] Groundwater level monitoring is a fundamental task in water resource management, hydrogeological analysis, and ecological environmental protection. It primarily involves continuously observing changes in groundwater levels to understand groundwater recharge, runoff, and discharge, thereby providing data support for groundwater development and utilization, land subsidence control, and ecological security assessment. With accelerating urbanization and increasing human activity, groundwater levels are becoming more complex, driven by multiple factors, placing higher demands on monitoring accuracy and timeliness.

[0003] Existing groundwater level monitoring and prediction technologies are mainly divided into two categories: one is numerical simulation methods based on groundwater dynamics. These methods strictly follow the seepage control equations and have clear physical mechanisms, but they rely on a large amount of hydrogeological drilling and experimental data. They have long modeling cycles, poor adaptability to complex boundary conditions, and difficulty in quickly responding to real-time dynamic monitoring needs. The other category is data-driven artificial intelligence prediction methods. These methods can quickly fit dynamic changes in water levels, but they generally suffer from strong black-box characteristics, lack of physical mechanism constraints, poor interpretability, are prone to overfitting, and cannot accurately characterize the vertical hydraulic coupling process between different aquifer layers. In practical applications, existing technologies suffer from the following shortcomings: First, the stratified monitoring is not standardized, failing to isolate different aquifers through stratified water-stopping structures, leading to cross-layer interference, making it impossible to obtain accurate hydraulic head data for each aquifer and failing to reflect the mutual influence between aquifers at different depths. Second, the logic for determining interlayer hydraulic coupling is unreasonable, relying solely on the correlation of water level fluctuations to determine hydraulic connections, ignoring core control factors such as lithology and permeability coefficients, and failing to accurately characterize the vertical transport process within groundwater. Third, the nonlinear and hysteretic characteristics of groundwater systems are insufficiently characterized, often introducing environmental factors such as rainfall, evaporation, surface runoff, and artificial water intake as independent or simplified linear parameters, lacking a systematic characterization of the relationship between dynamic environmental factors and groundwater level response. Fourth, artificial intelligence models are disconnected from hydrogeological and physical mechanisms, simply splicing together general AI algorithms without substantial technical improvements, resulting in predictions lacking physical meaning and failing to guarantee prediction accuracy and reliability in complex scenarios. Therefore, how to achieve deep integration of hydrogeological and physical mechanisms with artificial intelligence models, and complete the collaborative simulation and interpretable and accurate prediction of the internal transport process of groundwater and the driving process of the external environment under the condition of multi-process coupling, has become a challenge for the industry. Summary of the Invention

[0004] This application provides a method and system for dynamic monitoring of groundwater level based on artificial intelligence, which can achieve deep integration of hydrogeological and physical mechanisms with artificial intelligence models, and complete the collaborative simulation and interpretable and accurate prediction of the internal transport process of groundwater and the driving process of the external environment under the condition of multi-process coupling.

[0005] In a first aspect, this application provides a method for dynamic monitoring of groundwater levels based on artificial intelligence, comprising the following steps: The original pressure monitoring signals and corresponding static water level calibration data of groundwater monitoring devices deployed by layered water-stopping structures in different aquifers of the target area are obtained. Based on the original pressure monitoring signals and static water level calibration data, the initial hydraulic head time sequence and water level change sequence of each aquifer in the target area are constructed. The transient change characteristics of groundwater head in each aquifer are extracted from the time series of water head and the water level change series. Based on the transient change characteristics, the layered layout structure of the monitoring device and the vertical hydrogeological parameters of the target area, the hydraulic coupling characteristics between different aquifers are determined. Based on multi-source environmental dynamic data affecting groundwater level fluctuations in the target area, and combined with the hysteresis and nonlinear characteristics of the groundwater system, a driving response model with physical constraints is constructed to determine the influence coefficients of different environmental factors on the water level changes of each aquifer when the groundwater level changes. By utilizing the hydraulic coupling characteristics and all the influence coefficients, a physical information graph neural network model with embedded groundwater seepage control equations is used to intelligently fuse the groundwater level in the target area, generating dynamic prediction results of groundwater level that reflect the internal transport process of groundwater and the driving process of the external environment. The dynamic prediction results of the groundwater level are sent to the groundwater monitoring and management platform.

[0006] In conjunction with the first aspect, in one possible implementation, constructing the initial hydraulic head time series and water level change series for each aquifer in the target area based on the original pressure monitoring signal and static water level calibration data specifically includes: Temperature compensation, zero-point drift correction and static water level calibration are performed on the original pressure monitoring signal to generate standardized head time series data for each aquifer. The standardized head time series data is interpolated to generate continuous, equally spaced head time series sequences for each aquifer layer. Based on the water head time series, the water head change and the corresponding regional comprehensive water level change at each layer of aquifer monitoring point between adjacent time points are determined. The initial water level change sequence of the target area is formed by the changes in water head at all monitoring points of the stratified aquifers within the target area and the overall water level change of the area.

[0007] In conjunction with the first aspect, in one possible implementation, extracting the transient change characteristics of groundwater head in each aquifer from the head time series and water level change series specifically includes: Based on the head time series and water level change series, the time domain and frequency domain feature vectors of the dynamic fluctuation of head in each stratified aquifer are extracted. By extracting all the feature vectors obtained, the transient variation characteristics of groundwater head in each aquifer under recharge, extraction and disturbance conditions are determined.

[0008] In conjunction with the first aspect, in one possible implementation, determining the hydraulic coupling characteristics between different aquifer layers based on the transient change characteristics, the layered deployment structure of the monitoring device, and the vertical hydrogeological parameters of the target area specifically includes: Any two stratified aquifers that are vertically adjacent are considered as aquifer pairs; The correlation degree of each aquifer to the dynamic response of hydraulic head is determined based on the transient change characteristics. The correlation degree is corrected by comparing the vertical hydrogeological parameters of the target area with the head difference between the aquifer pairs, resulting in the corrected correlation degree. Based on the corrected correlation between all aquifer pairs and the interlayer water exchange intensity calculated using the Darcy overflow formula, the hydraulic coupling characteristics between different aquifer layers are determined.

[0009] In conjunction with the first aspect, in one possible implementation, based on multi-source environmental dynamic data affecting groundwater level fluctuations in the target area, and considering the hysteresis and nonlinear characteristics of the groundwater system, a physically constrained driving response model is constructed to determine the influence coefficients of different environmental factors on the water level changes of each aquifer during groundwater level changes. Specifically, this includes: Acquire multi-source environmental dynamic data affecting groundwater level fluctuations in the target area; An interpretable nonlinear impact analysis model with hydrogeological and physical constraints is established, which takes the multi-source environmental dynamic data as time-delay input and the water level change sequence of each layer of aquifer in the target area as output. The impact analysis model serves as a driving response model with physical constraints. Solve the impact analysis model and extract the weight parameters corresponding to each environmental dynamic data input variable in the impact analysis model, which are used as the impact coefficients of different environmental factors on the water level changes of the corresponding aquifers when the groundwater level changes.

[0010] In conjunction with the first aspect, in one possible implementation, utilizing the hydraulic coupling characteristics and all influence coefficients, a physical information graph neural network model embedded with the groundwater seepage control equation is used to intelligently fuse the groundwater level in the target area, generating a dynamic prediction result of the groundwater level reflecting the internal transport process of groundwater and the driving process of the external environment. Specifically, this includes: Based on the aforementioned hydraulic coupling characteristics, a graph neural network model skeleton representing the vertical water transport process within the aquifer is constructed. All influence coefficients are embedded into the graph neural network model as attention weights and boundary condition constraints for regulating the input of external environment driving signals; By embedding the groundwater seepage control equation as a regularization term into the model loss function, a physical information graph neural network model is constructed to drive the model to perform spatiotemporal evolution inference and generate dynamic prediction results of groundwater level.

[0011] In conjunction with the first aspect, in one possible implementation, sending the dynamic prediction results of the groundwater level to the groundwater monitoring and management platform specifically includes: Based on the dynamic prediction results of groundwater level, and combined with the groundwater control threshold of each aquifer, a groundwater level change risk index for the target area is generated. The dynamic prediction results of groundwater level and the risk index of groundwater level change are sent to the groundwater monitoring and management platform.

[0012] In conjunction with the first aspect, in one possible implementation, the groundwater monitoring device is mainly composed of pressure-type water level sensors arranged in layers, equipped with float-type water level gauges or ultrasonic water level gauges for monitoring and calibrating the static water level in the well; wherein, the pressure-type water level sensors of each layer of aquifer are equipped with independent layered water-stopping structures to avoid cross-layer interference between aquifers.

[0013] In conjunction with the first aspect, in one possible implementation, the multi-source environmental dynamic data includes atmospheric recharge parameters, surface runoff data, groundwater extraction volume, local infiltration disturbance factors, and basic hydrogeological parameters of the target area.

[0014] Secondly, this application provides an artificial intelligence-based groundwater level dynamic monitoring system, comprising: The acquisition module is used to acquire the original pressure monitoring signals and corresponding static water level calibration data of the groundwater monitoring devices deployed by the layered water-stopping structure in different aquifers of the target area. Based on the original pressure monitoring signals and static water level calibration data, the initial hydraulic head time sequence and water level change sequence of each aquifer in the target area are constructed. The processing module is used to extract the transient change characteristics of groundwater head in each layered aquifer from the head time series and water level change series, and to determine the hydraulic coupling characteristics between different layersed aquifers based on the transient change characteristics, the layered layout structure of the monitoring device and the vertical hydrogeological parameters of the target area. The processing module is also used to construct a physically constrained driving response model based on multi-source environmental dynamic data affecting groundwater level fluctuations in the target area, combined with the hysteresis and nonlinear characteristics of the groundwater system, and to determine the influence coefficients of different environmental factors on the water level changes of each aquifer when the groundwater level changes. The processing module is also used to intelligently fuse the groundwater level in the target area based on the physical information graph neural network model embedded with the groundwater seepage control equation, using the hydraulic coupling characteristics and all the influence coefficients, to generate a dynamic prediction result of the groundwater level that reflects the internal transport process of groundwater and the driving process of the external environment. The execution module is used to send the dynamic prediction results of the groundwater level to the groundwater monitoring and management platform.

[0015] The technical solution provided in this application has the following beneficial effects: This application's approach, firstly, obtains real hydraulic head data for different aquifer layers through standardized monitoring with a layered water-stopping structure. This addresses the core issues of cross-layer interference and invalid data from the data source end. Based on the calibrated monitoring data, a hydraulic head time series and a water level change series are constructed, providing a reliable data foundation for subsequent analysis. Secondly, the transient change characteristics of hydraulic head for each aquifer layer are extracted from the hydraulic head time series. This process can characterize the abrupt changes, lags, and recovery characteristics of hydraulic head over short timescales, reflecting the differences in the immediate response of groundwater to recharge, extraction, or disturbances. This provides a basis for distinguishing between internal transport processes and external driving forces, avoiding misjudging short-term water level fluctuations as long-term trends, and enabling collaborative simulation to handle multiple coexisting processes. The ability to distinguish; secondly, based on the transient change characteristics, the layered layout of the monitoring device, and the vertical hydrogeological parameters of the target area, the hydraulic coupling characteristics between different aquifers are determined. This process abandons the logical fallacy of judging hydraulic connections solely based on fluctuation correlation. Combining core control factors such as stratigraphic lithology, permeability coefficient, and hydraulic head difference, the influence path and intensity of hydraulic head changes in vertical and interlayer propagation are clarified, characterizing the directionality and synergy of the vertical transmission process within groundwater. This effectively reveals the response linkage mechanism between different aquifers, providing structural constraints for the prediction model that conform to hydrogeological principles, avoiding the prediction model from relying solely on external factors to fit water level changes, and enhancing the ability to express the internal evolution law of groundwater; then, based on Based on multi-source environmental dynamic data and combined with the hysteresis and nonlinear characteristics of the groundwater system, a physically constrained driving response model is constructed to determine the influence coefficients of different environmental factors. This process can distinguish the degree of influence of factors such as rainfall, evaporation, surface runoff, and human activities on water level fluctuations at different stages and intensities, accurately characterizing the hysteresis and threshold effects of the groundwater system. It provides a clear and hydrologically consistent weight expression for the external driving process, enabling the model to dynamically adapt to changes in environmental conditions and avoid a single environmental variable dominating the prediction results. This improves the adaptability and stability of groundwater level prediction to complex external disturbance scenarios. Finally, utilizing the aforementioned hydraulic coupling characteristics and all influence coefficients, a driving response model is constructed based on the embedded groundwater seepage control equation. The physical information graph neural network model is intelligently fused to generate dynamic prediction results of groundwater level. This process integrates internal hydrogeological structure constraints and external driving intensity constraints, embedding the core control equations of groundwater dynamics into the AI ​​model. This completely solves the problems of black box fitting and lack of physical meaning in traditional AI models, avoids the bias caused by the separate modeling of internal and external processes, and enables the prediction results to simultaneously reflect the internal evolution logic of water level changes and the response characteristics of external disturbances. The intelligent fusion mechanism can dynamically adjust the contribution ratio of internal and external processes according to the dominant factors at different stages, thereby achieving coordinated simulation under multi-process coupling conditions and significantly improving the accuracy, reliability and interpretability of dynamic prediction of groundwater level in complex environments.In summary, this scheme enables deep integration of hydrogeological and physical mechanisms with artificial intelligence models, achieving collaborative simulation and interpretable, accurate prediction of internal groundwater transport processes and external environmental driving processes under multi-process coupling conditions. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is an exemplary flowchart of an artificial intelligence-based groundwater level dynamic monitoring method according to some embodiments of this application; Figure 2 This is an exemplary flowchart illustrating the determination of influence coefficients according to some embodiments of this application; Figure 3 This is an exemplary flowchart illustrating the generation of dynamic prediction results for groundwater levels according to some embodiments of this application; Figure 4 This is a schematic diagram of the structure of an artificial intelligence-based groundwater level dynamic monitoring system according to some embodiments of this application; Figure 5 This is a schematic diagram of the structure of a computer device for implementing an artificial intelligence-based method for dynamic monitoring of groundwater levels, according to some embodiments of this application. Detailed Implementation

[0018] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] refer to Figure 1 The figure is an exemplary flowchart of an artificial intelligence-based dynamic monitoring method for groundwater levels, according to some embodiments of this application. This artificial intelligence-based dynamic monitoring method for groundwater levels mainly includes the following steps: In step 101, the original pressure monitoring signals and corresponding static water level calibration data of the groundwater monitoring devices deployed by the layered water-stopping structure in different aquifers of the target area are obtained. Based on the original pressure monitoring signals and static water level calibration data, the initial water head time sequence and water level change sequence of each aquifer in the target area are constructed.

[0020] It should be noted that the main body of the groundwater monitoring device in this application is a layered pressure water level sensor, equipped with a float-type water level gauge or an ultrasonic water level gauge for monitoring the static water level calibration in the well; wherein, the pressure water level sensor of each layer of aquifer is equipped with an independent layered water-stopping structure to avoid cross-layer interference between aquifers.

[0021] In specific implementation, the acquisition of raw pressure monitoring signals and corresponding static water level calibration data from groundwater monitoring devices deployed with layered water-stopping structures in different aquifers within the target area can be achieved in the following manner: First, the vertical stratigraphic structure and aquifer distribution in the target area are determined through hydrogeological drilling. Within the monitoring wells in the target area, water-stopping materials such as expanding sealant and plugs are used for layered water-stopping isolation of different aquifers. Pressure level sensors are deployed in each aquifer, and float-type or ultrasonic level gauges are installed at the wellhead for static water level calibration within the monitoring well. The raw voltage, current, frequency, or pulse signals of each pressure level sensor and the static water level monitoring data from the float-type / ultrasonic level gauges are simultaneously acquired at a preset sampling frequency. Then, the acquired raw electrical signals undergo local preprocessing, which includes at least sensor-based preprocessing. Temperature compensation and zero-point drift correction of the instrument's factory calibration parameters are performed to eliminate the influence of ambient temperature and long-term sensor drift on the measurement. The analog electrical signal is converted into digital pressure readings via an analog-to-digital converter. Subsequently, the digital readings from different monitoring devices are formatted and timestamped, and converted into standardized data packets with unified physical units, time references, and data structures. Finally, the obtained standardized data packets are transmitted to the central processing unit via wired transmission, wireless IoT, or BeiDou satellite communication, and their validity is verified. The verification includes a reasonable range judgment based on historical data and hydrogeological knowledge, and a preliminary screening of abrupt change points to eliminate obvious outliers caused by equipment failure or communication interference. This results in the output of the quality-controlled raw pressure monitoring signal and static water level calibration data. Other methods may be used in other embodiments, which are not specifically limited here.

[0022] It should be noted that the original pressure monitoring signal in this application refers to standardized time-series data characterizing the groundwater pressure state in different aquifer layers of the target area; the static water level calibration data refers to the static water level elevation data in the monitoring well used to calibrate and calculate the total water head (water level) of the aquifer; the total water head (i.e., groundwater level) in this application strictly follows the hydrogeological definition, and the calculation formula is as follows: ,in Total water head (groundwater level). To monitor the water head elevation at the monitoring point location, Let p be the pressure head at the monitoring point, and p be the actual water pressure exerted by the groundwater on the pressure level sensor at the monitoring point. The density of water, This refers to gravitational acceleration, which will not be elaborated upon here.

[0023] In some embodiments, the initial hydraulic head time series and water level change series of each aquifer in the target area can be constructed based on the original pressure monitoring signal and static water level calibration data using the following steps: Temperature compensation, zero-point drift correction and static water level calibration are performed on the original pressure monitoring signal to generate standardized head time series data for each aquifer. The standardized head time series data is interpolated to generate continuous, equally spaced head time series sequences for each aquifer layer. Based on the water head time series, the water head change and the corresponding regional comprehensive water level change at each layer of aquifer monitoring point between adjacent time points are determined. The initial water level change sequence of the target area is formed by the changes in water head at all monitoring points of the stratified aquifers within the target area and the overall water level change of the area.

[0024] In specific implementation, the original pressure monitoring signal is processed for temperature compensation, zero-point drift correction, and static water level calibration to generate standardized head time series data for each aquifer. This can be achieved in the following way: temperature compensation and zero-point drift correction are performed based on sensor calibration parameters, and the corrected pressure reading is converted into pressure head; the total head (groundwater level) of the corresponding aquifer is calculated by combining the buried elevation (location head z) of the monitoring point; the calculated total head is calibrated using wellhead static water level monitoring data to eliminate system errors, and finally, standardized head time series data for each aquifer is generated; other calibration methods can also be used in other embodiments, which are not specifically limited here.

[0025] In specific implementation, the standardized head time series data is interpolated to generate continuous and equally spaced head time series sequences for each aquifer. This can be achieved in the following way: for data gaps caused by outliers in the standardized head time series data due to validity checks, time series interpolation methods are used to fill in the gaps to generate continuous head time series data. Specifically, for single-point gaps or short-term gaps, linear interpolation or cubic spline interpolation methods based on the historical head data of the monitoring point are used for filling. For large data gaps caused by long-term equipment failure, spatial interpolation methods, such as inverse distance weighted interpolation or ordinary kriging interpolation, are used to estimate and fill the gaps by combining the synchronous head data of adjacent monitoring wells in the same aquifer. After interpolation, a head time series sequence with continuous and equally spaced timestamps is output for each aquifer monitoring point. In a preferred embodiment, a water level prediction model based on a long short-term memory network can be used to interpolate complex missing patterns. In other embodiments, a seasonal decomposition autoregressive model can also be used for filling, which is not limited here.

[0026] In specific implementation, based on the head time series, determining the head change and corresponding regional comprehensive water level change of each stratified aquifer monitoring point between adjacent time points can be achieved in the following way: For each stratified aquifer monitoring point, the head time series is read in chronological order, and the difference between the observed head value at each sampling time and the observed head value at the previous adjacent sampling time is calculated. This difference is taken as the head change of the monitoring point at the current time. The head change at all sampling times is calculated sequentially to generate the head change sequence of the monitoring point. The head change at all monitoring points at the same time point and the same aquifer is spatially averaged to obtain the regional comprehensive water level change of the aquifer at the current time. The interval between adjacent time points is determined by the preset sampling frequency of the original monitoring signal, thereby ensuring that the calculation of the change at all monitoring points has a uniform time scale. Other methods can also be used in other embodiments, which are not limited here.

[0027] In specific implementation, the initial water level change sequence of the target area, composed of the head changes of all monitoring points in the layered aquifers and the overall water level changes of the area, can be achieved in the following way: The head change sequences of each monitoring point in the layered aquifers within the target area are arranged in layers according to the aquifer to which the monitoring points belong; within each layer, the change sequences of all monitoring points are organized according to their spatial coordinates; finally, along the three dimensions of time, vertical aquifer, and planar space, the head changes of all monitoring points and the overall water level changes of the area are integrated into a structured three-dimensional data array, which serves as the initial water level change sequence of the target area. The first dimension of the three-dimensional data array is the time axis, representing different sampling times; the second dimension is the vertical aquifer dimension, representing different layered aquifers; and the third dimension is the planar position axis, representing monitoring points at different horizontal positions within the same aquifer. As a preferred embodiment, a tensor data structure can be used to store and represent this three-dimensional data array. In other embodiments, nested lists or data cubes can also be used for organization, which is not limited here.

[0028] It should be noted that the head time series in this application refers to the set of total head (water level) observations from monitoring points of various stratified aquifers that are continuous in time and at equal intervals; the head change in this application refers to the numerical measure of the rise and fall of the total head between adjacent sampling times at the same monitoring point, which is used to characterize the instantaneous rate and direction of water level fluctuations over time in different stratified aquifers; the water level change series in this application refers to a three-dimensional data array that is structured and organized according to three dimensions: time, vertical direction of the aquifer, and planar spatial location, to form the head change of all monitoring points in the target area.

[0029] In step 102, the transient change characteristics of groundwater head in each aquifer are extracted from the head time series and water level change series. Based on the transient change characteristics, the layered layout of the monitoring device and the vertical hydrogeological parameters of the target area, the hydraulic coupling characteristics between different aquifers are determined.

[0030] In some embodiments, extracting the transient change characteristics of groundwater head in each aquifer from the head time series and water level change series can be achieved by the following steps: Based on the head time series and water level change series, the time domain and frequency domain feature vectors of the dynamic fluctuation of head in each stratified aquifer are extracted. By extracting all the feature vectors obtained, the transient variation characteristics of groundwater head in each aquifer under recharge, extraction and disturbance conditions are determined.

[0031] In specific implementation, based on the head time series and water level change series, the extraction of time-domain and frequency-domain feature vectors of the dynamic head fluctuations of each stratified aquifer can be achieved in the following way: First, the water level change series is divided into multiple water level change quantum sequences corresponding to different stratified aquifers along the vertical dimension of the aquifer; for each stratified aquifer subsequence, in the time domain, its statistical characteristics within a sliding time window are calculated as time-domain feature vectors, which at least include the mean, variance, range, zero-crossing rate, and abrupt change amplitude and duration of the change; in the frequency domain, the subsequence... The columns are subjected to Fast Fourier Transform or Wavelet Transform to extract the amplitude, frequency, and specified frequency band of their dominant frequency components, such as the energy proportion of a specific periodic frequency band reflecting the influence of rainfall or tides, as the frequency domain feature vector. Finally, the time domain and frequency domain feature vectors extracted from each layer of aquifer are combined to form the feature vector of the dynamic fluctuation of the water head of that layer of aquifer. In a preferred embodiment, the length of the sliding time window can be set according to the typical response time of groundwater to external disturbances. In other embodiments, empirical mode decomposition can also be used instead of wavelet transform to extract intrinsic fluctuation components, which is not limited here.

[0032] In specific implementation, the transient variation characteristics of groundwater head in each aquifer under recharge, extraction, and disturbance conditions can be determined by extracting all feature vectors. This can be achieved in the following way: the feature vectors corresponding to each aquifer are arranged and combined according to their vertical depth to construct a two-dimensional feature matrix. The row dimension of the matrix represents different aquifers, and the column dimension represents various time-domain and frequency-domain feature vectors. The row vectors of this two-dimensional feature matrix completely describe the transient fluctuation pattern of the head of a single aquifer, while the column vectors reveal the distribution pattern of the same feature vector in different aquifers. Finally, this two-dimensional feature matrix, abstracted and structured from spatiotemporal fluctuations, serves as the transient variation characteristics of groundwater head in each aquifer. In a preferred embodiment, the feature vectors of each aquifer can be Z-score standardized before constructing the feature matrix to eliminate the influence of dimensions. In other embodiments, principal component analysis can also be used to reduce the dimensionality of the feature matrix to obtain a subset of core features, which is not limited here.

[0033] It should be noted that the feature vector in this application refers to multidimensional feature data that characterizes the intensity, frequency distribution and change pattern of hydraulic head fluctuations in each layered aquifer per unit time; the transient change feature in this application refers to a two-dimensional feature matrix composed of the feature vectors of all layered aquifers arranged in order of vertical depth, which is used to comprehensively present the overall pattern and interlayer differences of the dynamic fluctuation behavior of hydraulic head in different aquifers in the vertical profile of the groundwater body in the target area.

[0034] In some embodiments, determining the hydraulic coupling characteristics between different aquifer layers based on the transient change characteristics, the layered layout of the monitoring device, and the vertical hydrogeological parameters of the target area can be achieved through the following steps: Any two stratified aquifers that are vertically adjacent are considered as aquifer pairs; The correlation degree of each aquifer to the dynamic response of hydraulic head is determined based on the transient change characteristics. The correlation degree is corrected by comparing the vertical hydrogeological parameters of the target area with the head difference between the aquifer pairs, resulting in the corrected correlation degree. Based on the corrected correlation between all aquifer pairs and the interlayer water exchange intensity calculated using the Darcy overflow formula, the hydraulic coupling characteristics between different aquifer layers are determined.

[0035] It should be noted that the aquifer pair in this application refers to a basic analytical unit consisting of two vertically adjacent layered aquifers with a weakly permeable / impermeable layer in between. In specific implementation, each layered aquifer can be combined with its vertically adjacent aquifers in sequence to generate all aquifer pairs that meet the conditions.

[0036] In specific implementation, determining the correlation degree of each aquifer to the dynamic head response based on the transient change characteristics can be achieved in the following way: For each aquifer pair, extract the two row vectors corresponding to the aquifer pair, i.e., the two-dimensional feature matrix, from the transient change characteristics; calculate the similarity measure between these two feature vectors, and use this measure value as the correlation degree of the aquifer to the dynamic head response; specifically, the similarity measure can be calculated using cosine similarity, Pearson correlation coefficient, or the reciprocal of Euclidean distance; for example, calculate the cosine value of the angle between the two feature vectors, the closer this value is to 1, the more similar the head fluctuation patterns of the two stratified aquifers are, and the higher the correlation degree; finally, output a correlation degree value corresponding to each aquifer pair; in other embodiments, the dynamic time warping algorithm can also be used to calculate the similarity of sequence morphology, which is not limited here.

[0037] In practice, the correlation is corrected by using the vertical hydrogeological parameters of the target area and the head difference between the aquifer pairs. The corrected correlation can be achieved as follows: The vertical hydrogeological parameters of the target area include the permeability coefficient, thickness, and storage coefficient of the weakly permeable layer. First, based on the results of hydrogeological drilling and pumping tests in the target area, the permeability coefficient, thickness, and storage coefficient of the weakly permeable layer between the aquifer pairs, as well as the long-term average head difference between the aquifer pairs, are obtained. Then, the theoretical overflow intensity of the aquifer pair is calculated based on the Darcy overflow formula. The overflow intensity calculation formula is: ,in For vertical overflow, The permeability coefficient of the weakly permeable layer, The difference in water head between aquifers, The thickness of the weakly permeable layer is used. The initial correlation is corrected based on the overflow intensity. The greater the overflow intensity, the higher the correction coefficient. When the overflow intensity is 0 (complete impermeable layer), the correction coefficient is 0. Finally, the initial correlation is multiplied by the correction coefficient to obtain the corrected correlation. In a preferred embodiment, the correction coefficient can be calibrated in combination with the tracer test results. In other embodiments, a piecewise linear function can also be used for approximate correction. This is not limited here.

[0038] In specific implementation, the hydraulic coupling characteristics between different aquifers can be determined by combining the corrected correlation degree between all aquifer pairs with the interlayer water exchange intensity calculated by the Darcy overflow formula. This can be achieved in the following way: construct a blank matrix with all aquifers as rows and columns; traverse all aquifer pairs and fill the matrix with the corrected correlation degree value of each aquifer pair at the intersection of the row and column of the corresponding two aquifer identifiers; fill the diagonal positions of the matrix, i.e., the autocorrelation of the same aquifer, with a fixed value of 1; finally, a symmetrical square matrix with elements of corrected correlation degree values ​​is obtained, which serves as the hydraulic coupling characteristics between different aquifers; the value of any element in this matrix intuitively represents the strength of the hydraulic connection between the corresponding two aquifers; as a preferred embodiment, this matrix can be defined as a hydraulic coupling matrix. In other embodiments, this matrix can also be converted into a graph structure, where the nodes are aquifers and the edge weights are the corrected correlation degrees, which is not limited here.

[0039] It should be noted that the correlation degree in this application refers to a scalar value used to quantify the similarity of the dynamic change patterns of the hydraulic head of any two adjacent aquifers, which is used to initially measure the potential hydraulic connection between the two aquifers; the corrected correlation degree in this application refers to the value obtained after correcting the correlation degree based on the hydrogeological parameters and overflow intensity of the aquifer pair, which is used to realistically characterize the actual hydraulic connection strength between the two aquifers; the hydraulic coupling feature in this application refers to a symmetric matrix organized by the corrected correlation degrees of all aquifer pairs, which is used to characterize the hydraulic connectivity network of the entire underground aquifer system in the target area on the vertical profile.

[0040] In step 103, based on multi-source environmental dynamic data affecting groundwater level fluctuations in the target area, and combined with the hysteresis and nonlinear characteristics of the groundwater system, a driving response model with physical constraints is constructed to determine the influence coefficients of different environmental factors on the water level changes of each aquifer when the groundwater level changes.

[0041] In some embodiments, reference Figure 2As shown, this figure is an exemplary flowchart for determining the influence coefficient in some embodiments of this application. In this embodiment, based on multi-source environmental dynamic data affecting groundwater level fluctuations in the target area, and combined with the hysteresis and nonlinear characteristics of the groundwater system, a driving response model with physical constraints is constructed. The influence coefficients of different environmental factors on the water level changes of each aquifer during groundwater level changes can be determined by the following steps: First, in step 1031, multi-source environmental dynamic data affecting groundwater level fluctuations in the target area are acquired; Secondly, in step 1032, an interpretable nonlinear impact analysis model with hydrogeological and physical constraints is established, which takes the multi-source environmental dynamic data as time-delay input and the water level change sequence of each layer of aquifer in the target area as output. The impact analysis model serves as a driving response model with physical constraints. Finally, in step 1033, the influence analysis model is solved, and the weight parameters corresponding to each environmental dynamic data input variable in the influence analysis model are extracted as the influence coefficients of different environmental factors on the water level changes of the corresponding stratified aquifers when the groundwater level changes.

[0042] In specific implementation, acquiring multi-source environmental dynamic data affecting groundwater level fluctuations in the target area can be achieved in the following way: Multi-source environmental dynamic data synchronized with the target area's water level change sequence can be collected from meteorological, hydrological, and human activity monitoring networks in the target area. This multi-source environmental dynamic data includes at least atmospheric recharge parameters (rainfall, evaporation), surface runoff data, groundwater extraction, local infiltration disturbance factors, and basic hydrogeological parameters of the target area. The collected data from each source are preprocessed, including imputation of missing values, removal of outliers, and resampling of all data to the same time resolution as the water level change sequence. Finally, the resampled sequence data are standardized to eliminate dimensional differences between different environmental factors, generating multi-source environmental dynamic data affecting groundwater level fluctuations in the target area. In a preferred embodiment, soil moisture remote sensing data or land use change data can also be included as supplementary environmental factors. In other embodiments, the types of data sources can be added or removed according to regional characteristics; no specific limitations are made here.

[0043] In specific implementation, an interpretable nonlinear impact analysis model with hydrogeological and physical constraints is established, using the multi-source environmental dynamic data as time-delay input and the water level change sequence of each aquifer in the target area as output. This impact analysis model, serving as a physically constrained driving response model, can be implemented as follows: For each aquifer, an independent impact analysis model is constructed to accommodate the different responses of different aquifers to environmental factors; the standardized multi-source environmental dynamic data is constructed into a time-delay feature vector containing observations from the past several days to capture the lag effects of the groundwater system; and a nonlinear model with interpretable parameters is selected as the model frame. The model can be structured using, for example, a constrained LASSO regression model, a ridge regression model, or a gradient boosting tree model. The time-delay feature vector is used as the input of the independent variable, and the water level change of the corresponding aquifer at that time is used as the output label of the dependent variable, thus constructing a training sample set. Hydrogeological and physical constraints are added to the model, such as a reasonable range for rainfall infiltration coefficients and a negative correlation constraint between extraction volume and water level changes. Finally, the physically constrained interpretable impact analysis model is established, which serves as the physically constrained driving response model. In other implementations, structured equation models can also be used to establish multi-factor path relationships; this is not limited here.

[0044] In specific implementation, the impact analysis model is solved, and the weight parameters corresponding to each environmental dynamic data input variable in the impact analysis model are extracted as the influence coefficients of different environmental factors on the water level changes of the corresponding aquifers during groundwater level changes. This can be achieved in the following way: using the above-mentioned training sample set, the impact analysis model is trained by the least squares method or gradient descent method to solve for the optimal model parameters that minimize the model prediction error; after training, the weight parameter values ​​corresponding to each input environmental feature variable are directly extracted from the solved model; the absolute value of the weight parameter directly represents the influence of the corresponding environmental factor on the water level of the aquifer. The degree of influence of the change is represented by positive or negative signs, indicating its driving direction (such as promoting the rise or fall of water level). Then, the weight parameters of each extracted environmental factor are normalized so that the sum of their absolute values ​​is 1. The final set of normalized weight parameters is used as the influence coefficient of different environmental factors on the water level change of the corresponding aquifer when the groundwater level changes. In a preferred embodiment, Bootstrapping-based repeated sampling training can be used to evaluate the stability of the weight parameters and calculate the confidence interval. In other embodiments, the SHAP value decomposition method can also be used to decompose the feature importance from the model as the influence coefficient. This is not limited here.

[0045] It should be noted that the multi-source environmental dynamic data in this application refers to a time-series dataset characterizing the external environmental driving forces on the groundwater system of the target area; the impact analysis model in this application refers to a mathematical model that can establish a quantitative mapping relationship between multi-source environmental dynamic data and the response of water level changes in stratified aquifers, and whose internal parameters have clear physical and statistical significance. It is used to analyze and quantify the magnitude and direction of the contribution of each environmental factor to water level changes; the impact coefficient in this application refers to the degree of influence of each environmental factor in driving the groundwater level changes of the corresponding stratified aquifers in a normalized numerical form.

[0046] In step 104, the groundwater level in the target area is intelligently fused based on the physical information graph neural network model embedded with the groundwater seepage control equation, using the hydraulic coupling characteristics and all the influence coefficients, to generate a dynamic prediction result of the groundwater level that reflects the internal transport process of groundwater and the driving process of the external environment.

[0047] In some embodiments, reference Figure 3 As shown in the figure, this is an exemplary flowchart for generating dynamic prediction results of groundwater level in some embodiments of this application. In this embodiment, the groundwater level in the target area is intelligently fused based on the physical information graph neural network model embedded with the groundwater seepage control equation, utilizing the hydraulic coupling characteristics and all influence coefficients, to generate dynamic prediction results of groundwater level reflecting the internal transport process of groundwater and the driving process of the external environment. This can be achieved by the following steps: First, in step 1041, a graph neural network model skeleton representing the vertical water transport process inside the aquifer is constructed based on the hydraulic coupling characteristics. Secondly, in step 1042, all the influence coefficients are embedded into the graph neural network model as attention weights and boundary condition constraints for regulating the input of external environment driving signals. Finally, in step 1043, the groundwater seepage control equation is embedded as a regularization term into the model loss function to construct a physical information graph neural network model, which drives the model to perform spatiotemporal evolution inference and generate dynamic prediction results of groundwater level.

[0048] In specific implementation, the graph neural network model skeleton representing the vertical water transport process within the aquifer based on the hydraulic coupling characteristics can be implemented in the following way: the hydraulic coupling characteristics are regarded as a weighted adjacency matrix, where each layered aquifer is defined as a node in the graph, and the element values ​​in the matrix, i.e., the corrected correlation, are defined as the weights of the connecting edges between corresponding nodes, thereby constructing an attribute graph describing the vertical connectivity structure of the aquifer; based on this attribute graph, a suitable neural network architecture for processing this graph structure is selected as the skeleton, such as a multi-layer graph convolutional network or a graph attention network; in this skeleton, each graph convolutional layer or graph attention layer is configured to aggregate and transmit the feature information of adjacent nodes, i.e., the layered aquifers, based on the edge weights, i.e., the hydraulic connection strength, thereby explicitly encoding the vertical transport process of groundwater in the model; as a preferred embodiment, a message-passing neural network with a gating mechanism can be used to simulate the nonlinear diffusion process of water flow. In other embodiments, recurrent neural network units can also be combined to handle time dependencies, which is not limited here.

[0049] In specific implementation, all influence coefficients are embedded into the graph neural network model. The attention weights and boundary condition constraints used to regulate the input of the external environmental driving signal can be implemented as follows: an environmental driving signal modulation module is added to the input layer of the graph neural network model skeleton. This module receives standardized multi-source environmental dynamic data as input and uses the influence coefficients of the corresponding stratified aquifers as attention weight vectors. Specifically, the data sequence of each environmental factor is multiplied by its corresponding influence coefficient (i.e., weight) to achieve weighted fusion based on importance, generating a weighted comprehensive environmental driving signal. Subsequently, this weighted signal is used as a boundary condition feature and concatenated with the spatiotemporal features of each node in the graph neural network (i.e., the stratified aquifer), thereby injecting external driving factors into the model in a quantitatively regulated manner. In a preferred embodiment, differentiated environmental driving influence modes can be set for different stratified aquifers. In other embodiments, a dynamic weight generation network based on influence coefficients can also be used; this is not limited here.

[0050] In practical implementation, the groundwater seepage control equation is embedded as a regularization term into the model's loss function. A physical information graph neural network model is constructed to drive the model to perform spatiotemporal evolution inference and generate dynamic prediction results for groundwater levels. This can be achieved in the following way: the three-dimensional groundwater seepage control equation is embedded as a physical constraint regularization term into the model's loss function. The seepage control equation is: ,in The water storage capacity of the aquifer. Let H be the partial derivative of the total head (groundwater level) with respect to time, where H is the total head of the aquifer (i.e., the groundwater level). For divergence operators, Let be the permeability tensor. For the head gradient, This is the Hamiltonian operator (gradient operator). The source and sink terms (including recharge, extraction, and overflow) are considered. The total loss function of the model consists of two parts: data fitting loss and physical constraint loss. The data fitting loss measures the deviation between the model's predicted value and the actual observed value, while the physical constraint loss measures the degree to which the model's predicted result satisfies the seepage control equation. The target area's water head time series from historical periods is used as the initial node feature input to the model, while the weighted comprehensive environmental driving signal for the future prediction period is used as the source and sink terms and boundary conditions input. The water head sequence from the previous historical moment and the current environmental driving signal are used as joint inputs to drive the physical information graph neural network model to perform spatiotemporal evolution inference, outputting the predicted values ​​of water head and water level changes for future multi-step periods, which serve as the dynamic prediction result of groundwater level reflecting the internal transport process and the external environmental driving process of groundwater. In a preferred embodiment, a supervised learning method based on historical observation data can be used to train the model end-to-end with the goal of minimizing the total loss function. In other embodiments, a sequence-to-sequence encoder-decoder architecture can also be used for multi-step prediction, which is not limited here.

[0051] It should be noted that the physical information graph neural network model in this application refers to a neural network model that constructs a graph structure based on hydraulic coupling characteristics and embeds the groundwater seepage control equation as a physical constraint. It is used to transform the vertical hydraulic connectivity of aquifers and the core laws of groundwater dynamics into a computable prediction model, realizing the physical interpretability of joint simulation and deduction of the spatiotemporal evolution of groundwater level. The attention weight in this application refers to the weight parameters used to weight the input signals of different environmental factors within the model, ensuring that the external environmental driving process is injected into the model with an intensity that matches its actual hydrological contribution. The dynamic prediction result of groundwater level in this application refers to the sequence of predicted water level values ​​for multiple time steps in the future for each layer of aquifer in the target area.

[0052] In step 105, the dynamic prediction results of the groundwater level are sent to the groundwater monitoring and management platform.

[0053] In some embodiments, sending the dynamic prediction results of groundwater level to the groundwater monitoring and management platform can be achieved by the following steps: Based on the dynamic prediction results of groundwater level, and combined with the groundwater control threshold of each aquifer, a groundwater level change risk index for the target area is generated. The dynamic prediction results of groundwater level and the risk index of groundwater level change are sent to the groundwater monitoring and management platform.

[0054] In specific implementation, based on the dynamic prediction results of groundwater level and combined with the groundwater control thresholds of each aquifer layer, the risk index of groundwater level change in the target area can be generated in the following way: First, according to the "Groundwater Management Regulations," local groundwater control indicators, and historical statistical data, safe water level thresholds (including red line water level and yellow line water level) are set for different aquifer layers; then, the predicted future water level value in the dynamic prediction results of groundwater level is compared with the safe threshold of the corresponding aquifer layer to calculate the magnitude and duration of the water level exceeding the safe range during the prediction period; then, based on the magnitude and duration of the exceedance and the aquifer layer... The risk score for each aquifer is calculated using a preset risk assessment function, taking into account its hydrogeological sensitivity level and water supply importance. Finally, the risk scores of each aquifer are weighted and aggregated, with weights allocated according to the aquifer's importance and vulnerability, to generate a single quantitative index that comprehensively reflects the risk level of future groundwater level changes in the target area. This index serves as the groundwater level change risk index for the target area. In a preferred embodiment, the risk assessment function can be a piecewise linear function or a fuzzy logic rule system. In other embodiments, the risk score can also be adjusted based on the uncertainty range of the prediction results; no specific limitations are imposed here.

[0055] In specific implementation, the dynamic prediction results of groundwater level and the risk index of groundwater level change can be sent to the groundwater monitoring and management platform in the following way: First, the dynamic prediction results of groundwater level, the risk index of groundwater level change, and their related metadata, including generation time, prediction period, target area code, aquifer number, and a brief description of risk level, can be packaged into a structured data packet that conforms to the platform's data exchange specifications, such as using JSON or XML format; then, the data packet can be sent to the data receiving application interface provided by the groundwater monitoring and management platform through a secure network transmission protocol, such as Hypertext Transfer Protocol based on Transport Layer Security (TLS), to present the monitoring and prediction information and risk warning information to the management personnel; as a preferred embodiment, asynchronous reliable transmission can be performed using a message queue. In other embodiments, lightweight geographic information system coordinate information can also be embedded in the data packet, which is not limited here.

[0056] It should be noted that the groundwater level change risk index in this application refers to a comprehensive risk assessment value that quantifies the degree to which future groundwater level changes in the target area deviate from the safety control threshold. It is used to provide managers with a basis for rapid risk situation assessment and decision-making priority ranking.

[0057] Furthermore, in another aspect of this application, in some embodiments, this application provides an artificial intelligence-based groundwater level dynamic monitoring system, with reference to... Figure 4 The figure is a schematic diagram of the structure of an AI-based groundwater level dynamic monitoring system according to some embodiments of this application. The AI-based groundwater level dynamic monitoring system 400 includes: an acquisition module 401, a processing module 402, and an execution module 403, which are described below: The acquisition module 401 in this application is mainly used to acquire the original pressure monitoring signals and corresponding static water level calibration data of the groundwater monitoring devices deployed by the layered water-stopping structure in different layers of aquifers in the target area. Based on the original pressure monitoring signals and static water level calibration data, the initial water head time sequence and water level change sequence of each layer of aquifer in the target area are constructed. Processing module 402 in this application is mainly used to extract the transient change characteristics of groundwater head of each layered aquifer from the head time series and water level change series, and to determine the hydraulic coupling characteristics between different layers of aquifer based on the transient change characteristics, the layered layout structure of the monitoring device and the vertical hydrogeological parameters of the target area. The processing module 402 described in this application is also used to construct a physically constrained driving response model based on multi-source environmental dynamic data affecting groundwater level fluctuations in the target area, combined with the hysteresis and nonlinear characteristics of the groundwater system, and to determine the influence coefficients of different environmental factors on the water level changes of each aquifer when the groundwater level changes. The processing module 402 described in this application is also used to utilize the hydraulic coupling characteristics and all the influence coefficients to intelligently fuse the groundwater level in the target area based on the physical information graph neural network model embedded with the groundwater seepage control equation, and generate a dynamic prediction result of the groundwater level that reflects the internal transport process of groundwater and the external environment driving process. The execution module 403 in this application is mainly used to send the dynamic prediction results of the groundwater level to the groundwater monitoring and management platform.

[0058] The foregoing has detailed examples of the AI-based groundwater level dynamic monitoring method and system provided in this application. It is understood that the corresponding apparatus, in order to achieve the above functions, includes hardware structures and / or software modules corresponding to the execution of each function. Those skilled in the art should readily recognize that, based on the units and algorithm steps of the examples described in conjunction with the embodiments disclosed herein, this application can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed by hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0059] In some embodiments, this application also provides a computer device, the computer device including a memory and a processor, the memory for storing a computer program, and the processor for calling and running the computer program from the memory, so that the computer device executes the above-described artificial intelligence-based groundwater level dynamic monitoring method.

[0060] In some embodiments, reference Figure 5 The dashed lines in the figure indicate that the unit or module is optional. This figure is a schematic diagram of the computer device implementing the AI-based dynamic groundwater level monitoring method of this application. The AI-based dynamic groundwater level monitoring method in the above embodiments can... Figure 5 The computer device 500 shown is used to implement this, and the computer device 500 includes at least one processor 501, a memory 502 and at least one communication unit 505. The computer device 500 may be a terminal device, a server or a chip.

[0061] The processor 501 can be a general-purpose processor or a special-purpose processor. For example, the processor 501 can be a central processing unit (CPU). The CPU can be used to control the computer device 500, execute software programs, and process data from the software programs. The computer device 500 may also include a communication unit 505 for inputting (receiving) and outputting (transmitting) signals.

[0062] For example, computer device 500 may be a chip, communication unit 505 may be the input and / or output circuit of the chip, or communication unit 505 may be the communication interface of the chip, and the chip may be a component of terminal device, network device or other device.

[0063] For example, computer device 500 may be a terminal device or a server, and communication unit 505 may be a transceiver of the terminal device or the server, or communication unit 505 may be a transceiver circuit of the terminal device or the server.

[0064] The computer device 500 may include one or more memories 502 storing a program 504. The program 504 can be executed by a processor 501 to generate instructions 503, causing the processor 501 to perform the methods described in the above method embodiments according to the instructions 503. Optionally, the memory 502 may also store data (such as a target audit model). Optionally, the processor 501 may also read data stored in the memory 502, which may be stored at the same storage address as the program 504, or the data may be stored at a different storage address than the program 504.

[0065] The processor 501 and memory 502 can be configured separately or integrated together, for example, integrated on the system-on-chip (SOC) of the terminal device.

[0066] It should be understood that each step of the above method embodiment can be completed by hardware logic circuits or software instructions in the processor 501. The processor 501 can be a CPU, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, such as discrete gates, transistor logic devices, or discrete hardware components.

[0067] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0068] For example, in some embodiments, this application also provides a computer-readable storage medium storing instructions or code that, when executed on a computer, cause the computer to implement the above-described artificial intelligence-based dynamic monitoring method for groundwater levels.

[0069] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0070] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for dynamic monitoring of groundwater levels based on artificial intelligence, characterized in that, Includes the following steps: The original pressure monitoring signals and corresponding static water level calibration data of groundwater monitoring devices deployed by layered water-stopping structures in different aquifers of the target area are obtained. Based on the original pressure monitoring signals and static water level calibration data, the initial hydraulic head time sequence and water level change sequence of each aquifer in the target area are constructed. The transient change characteristics of groundwater head in each aquifer are extracted from the time series of water head and the water level change series. Based on the transient change characteristics, the layered layout structure of the monitoring device and the vertical hydrogeological parameters of the target area, the hydraulic coupling characteristics between different aquifers are determined. Based on multi-source environmental dynamic data affecting groundwater level fluctuations in the target area, and combined with the hysteresis and nonlinear characteristics of the groundwater system, a driving response model with physical constraints is constructed to determine the influence coefficients of different environmental factors on the water level changes of each aquifer when the groundwater level changes. By utilizing the hydraulic coupling characteristics and all the influence coefficients, a physical information graph neural network model with embedded groundwater seepage control equations is used to intelligently fuse the groundwater level in the target area, generating dynamic prediction results of groundwater level that reflect the internal transport process of groundwater and the driving process of the external environment. The dynamic prediction results of the groundwater level are sent to the groundwater monitoring and management platform.

2. The method as described in claim 1, characterized in that, Based on the original pressure monitoring signal and static water level calibration data, the initial hydraulic head time series and water level change series of each aquifer in the target area are constructed, specifically including: Temperature compensation, zero-point drift correction and static water level calibration are performed on the original pressure monitoring signal to generate standardized head time series data for each aquifer layer. The standardized head time series data is interpolated to generate continuous, equally spaced head time series sequences for each aquifer layer. Based on the water head time series, the water head change and the corresponding regional comprehensive water level change at each layered aquifer monitoring point between adjacent time points are determined. The initial water level change sequence of the target area is formed by the changes in water head at all monitoring points of the stratified aquifers within the target area and the overall water level change of the area.

3. The method as described in claim 1, characterized in that, Extracting transient change characteristics of groundwater head in each aquifer from the aforementioned head time series and water level change series specifically includes: Based on the head time series and water level change series, the time domain and frequency domain feature vectors of the dynamic fluctuation of head in each stratified aquifer are extracted. By extracting all the feature vectors obtained, the transient variation characteristics of groundwater head in each aquifer under recharge, extraction and disturbance conditions are determined.

4. The method as described in claim 1, characterized in that, Based on the transient change characteristics, the layered layout of the monitoring devices, and the vertical hydrogeological parameters of the target area, the hydraulic coupling characteristics between different aquifer layers are determined, specifically including: Any two stratified aquifers that are vertically adjacent are considered as aquifer pairs; The correlation degree of each aquifer to the dynamic response of hydraulic head is determined based on the transient change characteristics. The correlation degree is corrected by comparing the vertical hydrogeological parameters of the target area with the head difference between the aquifer pairs, resulting in the corrected correlation degree. Based on the corrected correlation between all aquifer pairs and the interlayer water exchange intensity calculated using the Darcy overflow formula, the hydraulic coupling characteristics between different aquifer layers are determined.

5. The method as described in claim 1, characterized in that, Based on multi-source environmental dynamic data affecting groundwater level fluctuations in the target area, and considering the hysteresis and nonlinear characteristics of the groundwater system, a physically constrained driving response model is constructed to determine the influence coefficients of different environmental factors on the water level changes of each aquifer during groundwater level changes. Specifically, these include: Acquire multi-source environmental dynamic data affecting groundwater level fluctuations in the target area; An interpretable nonlinear impact analysis model with hydrogeological and physical constraints is established, which takes the multi-source environmental dynamic data as time-delay input and the water level change sequence of each layer of aquifer in the target area as output. The impact analysis model serves as a driving response model with physical constraints. Solve the impact analysis model and extract the weight parameters corresponding to each environmental dynamic data input variable in the impact analysis model, which are used as the impact coefficients of different environmental factors on the water level changes of the corresponding aquifers when the groundwater level changes.

6. The method as described in claim 1, characterized in that, Utilizing the aforementioned hydraulic coupling characteristics and all influence coefficients, a physical information graph neural network model embedded with the groundwater seepage control equation is used to intelligently fuse groundwater levels in the target area, generating dynamic prediction results of groundwater levels that reflect both the internal transport process and the external environmental driving process. Specifically, this includes: Based on the aforementioned hydraulic coupling characteristics, a graph neural network model skeleton representing the vertical water transport process within the aquifer is constructed. All influence coefficients are embedded into the graph neural network model as attention weights and boundary condition constraints for regulating the input of external environment driving signals; By embedding the groundwater seepage control equation as a regularization term into the model loss function, a physical information graph neural network model is constructed to drive the model to perform spatiotemporal evolution inference and generate dynamic prediction results of groundwater level.

7. The method as described in claim 1, characterized in that, Sending the dynamic prediction results of groundwater level to the groundwater monitoring and management platform specifically includes: Based on the dynamic prediction results of groundwater level, and combined with the groundwater control threshold of each aquifer, a groundwater level change risk index for the target area is generated. The dynamic prediction results of groundwater level and the risk index of groundwater level change are sent to the groundwater monitoring and management platform.

8. The method as described in claim 1, characterized in that, The groundwater monitoring device consists of a layered pressure level sensor, equipped with a float-type water level gauge or an ultrasonic water level gauge for monitoring and calibrating the static water level in the well. Each layer of the pressure level sensor is equipped with an independent layered water-stopping structure to avoid cross-layer interference between aquifers.

9. The method as described in claim 1, characterized in that, The multi-source environmental dynamic data includes atmospheric recharge parameters, surface runoff data, groundwater extraction volume, local infiltration disturbance factors, and basic hydrogeological parameters of the target area.

10. A groundwater level dynamic monitoring system based on artificial intelligence, characterized in that, include: The acquisition module is used to acquire the original pressure monitoring signals and corresponding static water level calibration data of the groundwater monitoring devices deployed by the layered water-stopping structure in different aquifers of the target area. Based on the original pressure monitoring signals and static water level calibration data, the initial hydraulic head time sequence and water level change sequence of each aquifer in the target area are constructed. The processing module is used to extract the transient change characteristics of groundwater head in each layered aquifer from the head time series and water level change series, and to determine the hydraulic coupling characteristics between different layersed aquifers based on the transient change characteristics, the layered layout structure of the monitoring device and the vertical hydrogeological parameters of the target area. The processing module is also used to construct a physically constrained driving response model based on multi-source environmental dynamic data affecting groundwater level fluctuations in the target area, combined with the hysteresis and nonlinear characteristics of the groundwater system, and to determine the influence coefficients of different environmental factors on the water level changes of each aquifer when the groundwater level changes. The processing module is also used to intelligently fuse the groundwater level in the target area based on the physical information graph neural network model embedded with the groundwater seepage control equation, using the hydraulic coupling characteristics and all the influence coefficients, to generate a dynamic prediction result of the groundwater level that reflects the internal transport process of groundwater and the driving process of the external environment. The execution module is used to send the dynamic prediction results of the groundwater level to the groundwater monitoring and management platform.

Images