A remote sensing enhanced system for groundwater level monitoring

Abstract

The application discloses a kind of remote sensing enhancement systems for groundwater level monitoring, including sensing acquisition module, feature enhancement module, groundwater inversion modeling module, predictive water level modeling module and monitoring sensing enhancement module.The application belongs to the technical field of groundwater level monitoring, specifically a kind of remote sensing enhancement systems for groundwater level monitoring, adopt the comprehensive intelligent idea of combining groundwater inversion modeling and predictive water level modeling, integrate groundwater flow velocity direction inversion data with multi-scale water level prediction results, can not only depict aquifer dynamics boundary condition, but also generate future evolution trend, so as to break through the limitation that traditional system can only make single trend prediction or static state monitoring;Adopt the anisotropic self-adaptive improved concomitant inversion method, carry out water flow velocity direction inversion;Adopt the multi-time-domain groundwater level prediction model improved by combining physical constraints and quantile consistency, carry out groundwater level prediction.

Description

Technical Field

[0001] This invention belongs to the field of groundwater monitoring technology, specifically a remote sensing enhancement system for groundwater level monitoring. Background Technology

[0002] The remote sensing enhancement system for groundwater level monitoring is a modern monitoring system integrating IoT sensing technology, wireless communication networks, and cloud computing platforms. Its core definition lies in its ability to automatically and continuously collect and analyze groundwater level data through intelligent water level sensors deployed in monitoring wells. It utilizes wireless networks for data management and interaction, thereby achieving efficient, accurate, and unmanned remote sensing of groundwater dynamics. This system not only improves the timeliness and accuracy of monitoring data, avoiding the lag and errors of manual measurements, but also provides strong data support and decision-making basis for the rational allocation of water resources, ecological environmental protection, and scientific research through the analysis and visualization of massive amounts of historical data. It is a key technological means to achieve sustainable water resource management and smart water conservancy.

[0003] However, existing remote sensing enhancement systems for groundwater level monitoring suffer from technical problems such as dispersed data sources, single-dimensional signal characteristics, and a lack of deep coupling between inversion and prediction information. Existing groundwater inversion modeling methods are unable to effectively handle aquifer anisotropy, and inversion results often become unstable or inconsistent with boundary conditions. Existing predictive water level modeling methods suffer from technical problems such as the disconnect between prediction models and hydrophysical mechanisms, and a lack of stability and reasonable responsiveness in prediction results. Summary of the Invention

[0004] To address the aforementioned issues and overcome the shortcomings of existing technologies, this invention provides a remote sensing enhancement system for groundwater level monitoring. Addressing the technical problems of existing remote sensing enhancement systems for groundwater level monitoring, such as scattered data sources, single-dimensional signal features, and a lack of deep coupling between inversion and prediction information, this solution creatively adopts a comprehensive intelligent approach combining groundwater inversion modeling and predictive water level modeling. It integrates groundwater velocity direction inversion data with multi-scale water level prediction results, thereby characterizing aquifer dynamic boundary conditions and generating future evolution trends, thus breaking through the limitations of traditional systems. This approach overcomes the limitations of traditional methods, which can only perform single-trend predictions or static state monitoring. It achieves high-precision, long-term, and remote enhanced sensing of groundwater dynamics. Addressing the technical problems in existing groundwater inversion modeling methods, such as the inability to effectively handle aquifer anisotropy and the frequent instability or inconsistency between inversion results and boundary conditions, this solution creatively adopts an adjoint inversion method that combines anisotropy adaptive improvement to perform inversion of water flow velocity direction. By integrating observation consistency terms, physical residual terms, anisotropy regularization terms, and boundary consistency terms, an improved objective function is constructed, and gradient is efficiently solved by combining the adjoint method. This method enhances the model's adaptability to complex aquifer conditions while ensuring physical plausibility, achieving accurate inversion of water flow velocity and direction, and significantly reducing parameter divergence and boundary distortion problems in traditional inversion methods. Addressing the technical issues of existing predictive water level modeling methods, such as the disconnect between the prediction model and hydrophysical mechanisms, and the lack of stability and reasonable responsiveness in prediction results, this scheme creatively employs a multi-temporal groundwater level prediction model that combines physical constraints and improved quantile consistency. By introducing physical residual constraints, monotonic response constraints, quantile consistency constraints, and temporal smoothing constraints on a shared encoder and multi-branch decoding network, an improved loss function is constructed and trained. This method not only ensures the consistency of prediction results across different time scales but also provides a reasonable response to changes in pumping and recharge, significantly improving the physical reliability and temporal stability of groundwater level prediction.

[0005] The technical solution adopted by the present invention is as follows: The present invention provides a remote sensing enhancement system for groundwater level monitoring, including a sensor acquisition module, a feature enhancement module, a groundwater inversion modeling module, a predictive water level modeling module, and a monitoring and sensing enhancement module;

[0006] The sensor acquisition module is used for multi-source sensor data acquisition. Through multi-source sensor data acquisition, raw sensor data is obtained and the raw sensor data is sent to the feature enhancement module.

[0007] The feature enhancement module is used for sensor feature enhancement. Through sensor feature enhancement, enhanced hydrological derived feature data is obtained, and the enhanced hydrological derived feature data is sent to the groundwater inversion modeling module and the predictive water level modeling module.

[0008] The groundwater inversion modeling module is used for inverting the direction of water flow velocity. By inverting the direction of water flow velocity, groundwater state inversion data is obtained, and the groundwater state inversion data is sent to the predictive water level modeling module and the monitoring and sensing enhancement module.

[0009] The predictive water level modeling module is used for groundwater level prediction. Through groundwater level prediction, multi-scale water level prediction data is obtained, and the multi-scale water level prediction data is sent to the monitoring and sensing enhancement module.

[0010] The monitoring and sensing enhancement module is used for remote sensing enhancement, and obtains remote sensing enhancement data through remote sensing enhancement.

[0011] Furthermore, the multi-source sensor data acquisition is used to collect and standardize multi-source sensor signals, specifically by obtaining raw sensor data through real-time multi-source sensor data acquisition.

[0012] The raw sensing data specifically includes groundwater well sensing data, meteorological environment sensing data, human activity interference data, and auxiliary positioning environmental parameters.

[0013] Furthermore, the sensor feature enhancement is used to extract and enhance hydrological derived features. Specifically, it involves preprocessing the original sensor data, extracting features and enhancing them at multiple scales, and combining them with groundwater physical constraints to perform consistency correction, thereby obtaining enhanced hydrological derived feature data. This includes the following steps: data preprocessing, basic feature extraction, multi-scale feature enhancement and physical consistency correction.

[0014] The data preprocessing specifically involves time alignment and missing measurement completion of the raw sensing data to ensure data consistency in both time and spatial dimensions, resulting in preprocessed data.

[0015] The basic feature extraction specifically involves extracting hydrological features based on the preprocessed data to obtain hydrological feature data; the hydrological features specifically include groundwater level change rate, periodic fluctuation characteristics, and rainfall lag response characteristics.

[0016] The multi-scale feature enhancement specifically involves decomposing and fusing hydrological features at different scales, enhancing short-term rapid fluctuation features and long-term trend features to obtain multi-scale enhanced features.

[0017] The physical consistency correction specifically involves introducing a water balance constraint based on the groundwater flow equation as a water flow boundary condition during the multi-scale feature enhancement process to perform physical consistency correction on the enhanced features. This is to avoid pure data-driven results that do not conform to the actual hydrological mechanism and to obtain enhanced hydrological derived feature data.

[0018] The enhanced hydrological derived feature data specifically includes basic hydrological derived features, time-domain and frequency-domain features, and spatial physical constraint derived features;

[0019] The basic hydrological derived features specifically include groundwater level change rate, periodic fluctuation characteristics, and rainfall lag response characteristics; the time-domain and frequency-domain features specifically include long-term trend characteristics and short-term rapid fluctuation characteristics; the spatial physical constraint derived features specifically include hydraulic gradient characteristics, permeability and porosity characteristics, aquifer storage coefficient characteristics, specific storage coefficient characteristics, and hydrodynamic boundary characteristics.

[0020] Furthermore, the flow velocity direction inversion is used to estimate the flow velocity and direction of the aquifer and to infer the groundwater flow boundary and hydrological parameter distribution. Specifically, based on the enhanced hydrological derived feature data, the adjoint inversion method combined with anisotropic adaptive improvement is used to perform flow velocity direction inversion to obtain groundwater state inversion data, including the following steps: data assimilation initialization, explicit model discretization modeling, improved objective function construction, adjoint method extraction calculation, vector field calculation, and flow velocity direction inversion.

[0021] The data assimilation initialization is used to construct the initial state for inversion. Specifically, it involves reading the feature data from the enhanced hydrological derived feature data, initializing the basic inversion parameters, and setting boundary conditions to obtain the initialized groundwater state inversion data.

[0022] The model is an explicit discrete modeling method used to establish an iterative forward evolution model. Specifically, based on the initial groundwater state inversion data, an explicit discrete groundwater evolution model is established based on the discrete expression of the groundwater flow equation. Forward calculation of the groundwater state is then performed to obtain a discrete forward evolution model of the groundwater state.

[0023] The improved objective function is constructed by establishing a comprehensive error index, jointly modeling the observation consistency term, physical residual term, anisotropic regularization term, and boundary consistency term, and weighting each term according to data quality and uncertainty to obtain the improved objective function.

[0024] The adjoint method extraction calculation specifically involves using the adjoint method to calculate the gradient information of the objective function with respect to the water level field and the anisotropic parameter field based on the groundwater state discrete forward evolution model and the improved objective function, and performing alternating updates until the parameter updates converge to obtain the updated groundwater state parameters.

[0025] The vector field calculation specifically involves calculating the groundwater velocity vector field based on the updated groundwater state parameters, the estimated water level gradient, and the anisotropic permeability parameters, and extracting the flow direction angle and velocity magnitude to obtain the vector field estimation data.

[0026] The vector field estimation data includes velocity vector field estimation data and flow direction angle estimation data;

[0027] The inversion of the water flow velocity direction specifically involves obtaining groundwater state inversion data by combining the vector field estimation data with boundary conditions to evaluate the consistency of the inversion results.

[0028] The groundwater state inversion data specifically includes water level field inversion data, anisotropic parameter field inversion data, groundwater flow velocity and direction inversion data, and boundary consistency assessment data.

[0029] Furthermore, the groundwater level prediction is used to generate groundwater level prediction results at different time scales, including short-term, medium-term, and long-term. Specifically, based on the enhanced hydrological derived feature data and the groundwater state inversion data, a multi-time-domain groundwater level prediction model that combines physical constraints and quantile consistency improvement is used to predict the groundwater level and obtain multi-scale water level prediction data. The steps include: input data integration, shared encoder construction, multi-time-domain decoding prediction, physical residual constraint optimization, monotonic response constraint optimization, improved loss model training, and groundwater level prediction.

[0030] The input data integration is used to unify and integrate the enhanced hydrological derived feature data and the groundwater state inversion data. Specifically, it is to pair the basic hydrological derived features and the inverted water level field inversion data, anisotropic parameter field inversion data and groundwater flow velocity and direction inversion data within the same observation period onto the same time axis and spatial reference to obtain a spatiotemporal sample sequence.

[0031] The shared encoder is constructed by building a standard convolutional long short-term neural network as the encoder structure based on the spatiotemporal sample sequence, and using unified convolutional layer and long short-term memory unit parameters to extract shared features from the spatiotemporal sample sequence to obtain multi-temporal shared feature encoding data.

[0032] The multi-temporal decoding prediction specifically involves constructing a multi-branch decoding network and, based on the multi-temporal shared feature encoding data, using a long short-term memory network decoder to perform short-term prediction decoding, medium-term prediction decoding, and long-term prediction decoding to obtain multi-temporal groundwater level prediction data.

[0033] The physical residual constraint optimization specifically involves constructing adaptive physical weights to dynamically penalize the physical constraint loss function of the groundwater level prediction model, thereby obtaining physical residual constraint weight data.

[0034] The monotonic response constraint optimization specifically involves constructing monotonic constraint loss functions for recharge and pumping rates, applying monotonic soft constraints for groundwater level prediction, and obtaining the monotonic response constraint loss function.

[0035] The improved loss model training specifically involves constructing an ensemble loss function, using the physical residual constraint weight data as the ensemble weights of the physical constraint loss function, and simultaneously introducing the monotonic response constraint loss function, the quantile consistency loss function, and the horizon-aware smoothing loss function to obtain an improved loss function. The improved loss function is then used in conjunction with the shared encoder and the multi-temporal decoder to train the groundwater level prediction model, resulting in a groundwater level prediction model that balances physical consistency, response rationality, quantile stability, and temporal smoothness.

[0036] The groundwater level prediction specifically involves using the groundwater level prediction model to predict the groundwater level based on the enhanced hydrological derived feature data and the groundwater state inversion data, thereby obtaining multi-scale water level prediction data.

[0037] Furthermore, the remote sensing enhancement is used to generate remote sensing enhancement information by comprehensively predicting data. Specifically, based on the groundwater state inversion data and the multi-scale water level prediction data, information fusion and visualization data analysis are performed by integrating the inversion data and prediction data, and remote sensing enhancement of groundwater level is performed through remote data interaction to obtain remote sensing enhancement data.

[0038] The beneficial effects achieved by the present invention using the above solution are as follows:

[0039] (1) In view of the technical problems of scattered data sources, single signal feature dimension, and lack of deep coupling between inversion and prediction information in the existing remote sensing enhancement system for groundwater level monitoring, this solution creatively adopts a comprehensive intelligent approach that combines groundwater inversion modeling and predictive water level modeling. It integrates groundwater velocity direction inversion data with multi-scale water level prediction results, which can not only characterize the dynamic boundary conditions of aquifers, but also generate future evolution trends. This breaks through the limitations of traditional systems that can only make single trend predictions or static state monitoring, and realizes high-precision, long-period, and remote enhanced sensing of groundwater dynamics.

[0040] (2) To address the technical problems in existing groundwater inversion modeling methods, such as the inability to effectively handle aquifer anisotropy and the frequent instability or inconsistency between inversion results and boundary conditions, this scheme creatively adopts an adjoint inversion method combined with anisotropic adaptive improvement to invert water flow velocity direction. By integrating observation consistency terms, physical residual terms, anisotropic regularization terms, and boundary consistency terms, an improved objective function is constructed, and gradient is efficiently solved by combining the adjoint method. This method enhances the model's adaptability to complex aquifer conditions while ensuring physical rationality, achieving accurate inversion of water flow velocity and direction, and significantly reducing the problems of parameter divergence and boundary distortion in traditional inversion.

[0041] (3) To address the technical problems in existing predictive water level modeling methods, such as the disconnect between the prediction model and hydrophysical mechanisms, and the lack of stability and reasonable responsiveness in the prediction results, this scheme creatively adopts a multi-time-domain groundwater level prediction model that combines physical constraints and quantile consistency improvements. Based on a shared encoder and a multi-branch decoding network, physical residual constraints, monotonic response constraints, quantile consistency constraints, and temporal smoothing constraints are introduced to construct an improved loss function and train it. This method not only ensures the consistency of prediction results across different time scales but also provides a reasonable response to changes in pumping and recharge, significantly improving the physical reliability and temporal stability of groundwater level prediction. Attached Figure Description

[0042] Figure 1 A schematic diagram of the structure of a remote sensing enhancement system for groundwater level monitoring provided by the present invention;

[0043] Figure 2 A flowchart illustrating the steps performed by the system provided for this invention;

[0044] Figure 3 A flowchart illustrating the steps performed during data fusion in the feature enhancement module;

[0045] Figure 4 A flowchart illustrating the steps performed by the groundwater inversion modeling module;

[0046] Figure 5 A flowchart illustrating the steps performed by the predictive water level modeling module.

[0047] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation

[0048] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0049] Example 1, see Figure 1 The technical solution adopted by the present invention is as follows: The present invention provides a remote sensing enhancement system for groundwater level monitoring, including a sensor acquisition module, a feature enhancement module, a groundwater inversion modeling module, a predictive water level modeling module, and a monitoring and sensing enhancement module;

[0050] The sensor acquisition module is used for multi-source sensor data acquisition. Through multi-source sensor data acquisition, raw sensor data is obtained and the raw sensor data is sent to the feature enhancement module.

[0051] The feature enhancement module is used for sensor feature enhancement. Through sensor feature enhancement, enhanced hydrological derived feature data is obtained, and the enhanced hydrological derived feature data is sent to the groundwater inversion modeling module and the predictive water level modeling module.

[0052] The groundwater inversion modeling module is used for inverting the direction of water flow velocity. By inverting the direction of water flow velocity, groundwater state inversion data is obtained, and the groundwater state inversion data is sent to the predictive water level modeling module and the monitoring and sensing enhancement module.

[0053] The predictive water level modeling module is used for groundwater level prediction. Through groundwater level prediction, multi-scale water level prediction data is obtained, and the multi-scale water level prediction data is sent to the monitoring and sensing enhancement module.

[0054] The monitoring and sensing enhancement module is used for remote sensing enhancement, and obtains remote sensing enhancement data through remote sensing enhancement.

[0055] By performing the above operations, this solution addresses the technical problems of existing remote sensing enhancement systems for groundwater level monitoring, such as scattered data sources, single signal feature dimensions, and lack of deep coupling between inversion and prediction information. It creatively adopts a comprehensive intelligent approach that combines groundwater inversion modeling and predictive water level modeling. This approach integrates groundwater velocity direction inversion data with multi-scale water level prediction results, which can both characterize the dynamic boundary conditions of aquifers and generate future evolution trends. This breaks through the limitations of traditional systems that can only perform single trend prediction or static state monitoring, and achieves high-precision, long-term, and remote enhanced sensing of groundwater dynamics.

[0056] Specifically, in traditional systems, the lack of flow direction information consistent with the groundwater dynamic boundary results in prediction accuracy remaining at the hourly to daily level for a long time, making it impossible to provide monthly-scale predictions. However, by adopting this scheme, flow velocity and boundary condition constraints are obtained through inversion modeling and combined with multi-scale prediction, water level fluctuation trend prediction can be achieved, reducing prediction bias.

[0057] Example 2, this example is based on the above example, see reference. Figure 1 , Figure 2 The multi-source sensor data acquisition is used to collect and standardize multi-source sensor signals, specifically to obtain raw sensor data through real-time multi-source sensor data acquisition.

[0058] The raw sensing data specifically includes groundwater well sensing data, meteorological environment sensing data, human activity interference data, and auxiliary positioning environmental parameters;

[0059] Preferably, the groundwater well sensing data specifically includes groundwater level data (m), water pressure data (Pa), water temperature data (°C), conductivity (mS / cm), dissolved oxygen (mg / L), and dissolved solids concentration (mg / L).

[0060] The meteorological environment sensing data specifically includes rainfall (mm), rainfall intensity (mm / h), atmospheric pressure (Pa), and estimated evapotranspiration data (mm).

[0061] The human activity disturbance data specifically includes pumping volume (m³ / h), injection volume (m³ / h), and river or reservoir water level (m).

[0062] The auxiliary positioning environment parameters specifically include wellhead geographic coordinates (GPS), wellhead elevation (m), data acquisition equipment health status parameters, and signal quality parameters;

[0063] The real-time multi-source sensor data acquisition specifically involves data collection through downhole water level sensors, meteorological sensors, flow sensors, surface water level sensors, and GPS positioning modules. The sampling period is set to 1 minute, 10 minutes, and 60 minutes for real-time acquisition. The acquired raw data is then timestamped and standardized in units. Basic anomaly detection and anomaly data processing are then used to obtain the original sensor data.

[0064] Example 3, this example is based on the above examples, see reference. Figure 1 , Figure 2 and Figure 3The sensor feature enhancement is used to extract and enhance hydrological derived features. Specifically, it involves preprocessing the original sensor data, extracting features and enhancing them at multiple scales, and combining them with groundwater physical constraints to perform consistency correction, thereby obtaining enhanced hydrological derived feature data. The process includes the following steps: data preprocessing, basic feature extraction, multi-scale feature enhancement and physical consistency correction.

[0065] The data preprocessing specifically involves time alignment and missing measurement completion of the raw sensing data to ensure data consistency in both time and spatial dimensions, resulting in preprocessed data.

[0066] The basic feature extraction specifically involves extracting hydrological features based on the preprocessed data to obtain hydrological feature data; the hydrological features specifically include groundwater level change rate, periodic fluctuation characteristics, and rainfall lag response characteristics.

[0067] Preferably, the rate of change of groundwater level is calculated using the adjacent time difference method;

[0068] The periodic fluctuation characteristics are specifically obtained by performing a Fourier transform on the water level time series in the preprocessed data to identify and extract the main periodic components.

[0069] The rainfall lag response characteristics are specifically constructed by analyzing rainfall and water level change sequences using cross-correlation functions.

[0070] The multi-scale feature enhancement specifically involves decomposing and fusing hydrological features at different scales, enhancing short-term rapid fluctuation features and long-term trend features to obtain multi-scale enhanced features.

[0071] Preferably, the multi-scale enhancement is specifically achieved by performing empirical mode decomposition on the hydrological feature data to obtain high-frequency features, mid-frequency features, and low-frequency features; the high-frequency features are used as short-term rapid fluctuation features, and the mid-frequency and low-frequency features are used as long-term trend features.

[0072] For the aforementioned short-term rapid fluctuation characteristics, the amplitude and duration of the sudden changes are specifically calculated to enhance the short-term sensitivity. For the aforementioned long-term trend characteristics, a moving average is specifically used to calculate the long-term characteristics, and the amplitude envelope and phase drift are calculated to reflect the medium-term process.

[0073] The physical consistency correction specifically involves introducing a water balance constraint based on the groundwater flow equation as a water flow boundary condition during the multi-scale feature enhancement process to perform physical consistency correction on the enhanced features. This is to avoid pure data-driven results that do not conform to the actual hydrological mechanism and to obtain enhanced hydrological derived feature data.

[0074] The enhanced hydrological derived feature data specifically includes basic hydrological derived features, time-domain and frequency-domain features, and spatial physical constraint derived features;

[0075] The basic hydrological derived features specifically include groundwater level change rate, periodic fluctuation characteristics, and rainfall lag response characteristics; the time-domain and frequency-domain features specifically include long-term trend characteristics and short-term rapid fluctuation characteristics; the spatial physical constraint derived features specifically include hydraulic gradient characteristics, permeability and porosity characteristics, aquifer storage coefficient characteristics, specific storage coefficient characteristics, and hydrodynamic boundary characteristics.

[0076] The formula for calculating the physical consistency correction of the enhanced features based on the groundwater flow equation, which introduces water balance constraints as flow boundary conditions, is as follows:

[0077] ;

[0078] In the formula, e t It is the water balance constraint parameter, h t This represents the groundwater level at time t, where t is the time index. S is the time step parameter, S is the specific water storage parameter, and R is the specific water storage parameter. t It is the supply quantity parameter, Q t It is the pumping capacity parameter, ET. t These are evapotranspiration parameters, and K is the permeability coefficient. It is the gradient operator. The overall parameter is the seepage divergence parameter, where, It is a groundwater level gradient parameter;

[0079] The calculation formula for multi-scale feature enhancement after introducing physical consistency correction is as follows:

[0080] ;

[0081] In the formula, f enh (t) represents data with enhanced hydrological derived features, f ms (t) is a multi-scale enhancement feature. It is an adaptive gain variable, specifically calculated based on the baseline coefficient and the variance of water level uncertainty, e t 1d is the water balance constraint parameter, and 1d is the feature channel vector.

[0082] Example 4, this example is based on the above examples, see below. Figure 1 and Figure 4The flow velocity direction inversion is used to estimate the flow velocity and direction of aquifer water and infer the groundwater flow boundary and hydrological parameter distribution. Specifically, based on the enhanced hydrological derived feature data, the adjoint inversion method combined with anisotropic adaptive improvement is used to perform flow velocity direction inversion to obtain groundwater state inversion data. The steps include: data assimilation initialization, explicit model discretization modeling, improved objective function construction, adjoint method extraction calculation, vector field calculation, and flow velocity direction inversion.

[0083] The data assimilation initialization is used to construct the initial state for inversion. Specifically, it involves reading the feature data from the enhanced hydrological derived feature data, initializing the basic inversion parameters, and setting boundary conditions to obtain the initialized groundwater state inversion data.

[0084] Preferably, the initial inversion basic parameters specifically include the initial value of groundwater level, aquifer parameters, external action parameters, and calculation step size;

[0085] The boundary conditions specifically include fixed water level boundaries and no-flow boundaries; the fixed water level boundary is used to represent a boundary that maintains a constant water level, and the no-flow boundary is used to represent a boundary through which groundwater flow cannot pass.

[0086] The model is an explicit discrete modeling method used to establish an iterative forward evolution model. Specifically, based on the initial groundwater state inversion data, an explicit discrete groundwater evolution model is established based on the discrete expression of the groundwater flow equation. Forward calculation of the groundwater state is then performed to obtain a discrete forward evolution model of the groundwater state.

[0087] The calculation formula for the discrete forward evolution model of groundwater state is as follows:

[0088] ;

[0089] In the formula, S is the specific storage parameter, used to represent the aquifer's response to water level changes, and h n+1 It is the water level at discrete time n+1, where n is the discrete time index, used as the time index for the discrete forward evolution model of the groundwater state, h n It represents the water level at discrete time n. It is the time step parameter, R n It is a forward evolution modeling supply term, Q n It is the pumping term in forward evolution modeling, ET n It is the evapotranspiration term in forward evolution modeling. The whole is due to the permeation tensor The resulting hydraulic conduction flux divergence is used as a spatial flow term characterizing groundwater flow. It is the gradient operator. These are the groundwater level gradient parameters corresponding to discrete time points;

[0090] The permeability tensor The calculation formula is:

[0091] ;

[0092] In the formula, It is the angle between the principal axis of aquifer permeability and the coordinate axis. It is the permeability anisotropy ratio parameter, and the specific calculation formula is as follows: ,in, It is the value of the permeability in the principal direction (i.e., the parallel direction). It is the permeability value perpendicular to the main direction. It is a rotation matrix;

[0093] The improved objective function is constructed by establishing a comprehensive error index, jointly modeling the observation consistency term, physical residual term, anisotropic regularization term, and boundary consistency term, and weighting each term according to data quality and uncertainty to obtain the improved objective function.

[0094] The formula for calculating the improved objective function is as follows:

[0095] ;

[0096] In the formula, It is an improved objective function, where h is the water level parameter. It is the angle between the principal axis of aquifer permeability and the coordinate axis. It is the permeability anisotropy ratio parameter, and AE is the observational consistency term. This is the weight of the physical residual term, with a default value of 0.5. PE represents the physical residual term. This is the weight of the anisotropic regularization term, with a default value of 0.1. DE is the anisotropic regularization term. BE is the boundary consistency term weight, with a default value of 0.4.

[0097] The formula for calculating the observation consistency term is:

[0098] ;

[0099] In the formula, AE is the observation consistency term, i is the sensor location index, n is the discrete time index, and w i,n It is the sensor weighting coefficient, h(x) i ,y i ,t n The whole represents groundwater level parameters, where x i The horizontal coordinate of the i-th sensor is y. iIt is the vertical coordinate of the i-th sensor, t n It is the observation time. It is the actual water level data observed;

[0100] The formula for calculating the physical residual term is:

[0101] ;

[0102] In the formula, PE is the rational residual term. It is a spatial identifier for the observation range. It is the divergence of hydraulic conduction flux;

[0103] The formula for calculating the anisotropic regularization term is as follows:

[0104] ;

[0105] In the formula, DE is the anisotropic regularization term. It is the gradient of the angle between the principal axis of aquifer permeability and the coordinate axis, used to represent the spatial variation of the direction of the principal axis of permeability. It is the canonical balance coefficient. It is a priori anisotropy ratio reference value, which is obtained from the actual observed value;

[0106] The formula for calculating the boundary consistency term is:

[0107] ;

[0108] In the formula, BE is the boundary consistency term. It is the spatial boundary identifier of the observation range. It is a boundary consistency function used to measure the deviation between the water level and permeability tensor and the boundary conditions;

[0109] The adjoint method extraction calculation specifically involves using the adjoint method to calculate the gradient information of the objective function with respect to the water level field and the anisotropic parameter field based on the groundwater state discrete forward evolution model and the improved objective function, and performing alternating updates until the parameter updates converge to obtain the updated groundwater state parameters.

[0110] The vector field calculation specifically involves calculating the groundwater velocity vector field based on the updated groundwater state parameters, the estimated water level gradient, and the anisotropic permeability parameters, and extracting the flow direction angle and velocity magnitude to obtain the vector field estimation data.

[0111] The vector field estimation data includes velocity vector field estimation data and flow direction angle estimation data;

[0112] The formula for calculating the velocity vector field estimation data is as follows:

[0113] ;

[0114] In the formula, This is flow velocity vector field estimation data, where x is the sensor's horizontal coordinate index, y is the sensor's vertical coordinate index, and t is the time index. It is the anisotropic permeability tensor. It is the angle between the principal axis of aquifer permeability and the coordinate axis. It is the permeability anisotropy ratio parameter. It is the water level gradient;

[0115] The formula for calculating the estimated flow direction angle data is as follows:

[0116] ;

[0117] In the formula, This is the estimated flow angle data, v y It is the perpendicular component of the velocity vector in a two-dimensional plane, v x It is the horizontal component of the velocity vector in a two-dimensional plane, and atan2 is the arctangent function with quadrant determination;

[0118] The inversion of the water flow velocity direction specifically involves obtaining groundwater state inversion data by combining the vector field estimation data with boundary conditions to evaluate the consistency of the inversion results.

[0119] The groundwater state inversion data specifically includes water level field inversion data, anisotropic parameter field inversion data, groundwater flow velocity and direction inversion data, and boundary consistency assessment data.

[0120] By performing the above operations, this solution addresses the technical problems of existing groundwater inversion modeling methods, such as the inability to effectively handle aquifer anisotropy and the frequent instability or inconsistency between inversion results and boundary conditions. It creatively employs an adjoint inversion method combined with anisotropic adaptive improvement for inverting water flow velocity direction. By integrating observation consistency terms, physical residual terms, anisotropic regularization terms, and boundary consistency terms, an improved objective function is constructed, and the adjoint method is used to efficiently solve for the gradient. This method enhances the model's adaptability to complex aquifer conditions while ensuring physical rationality, achieving accurate inversion of water flow velocity and direction, and significantly reducing parameter divergence and boundary distortion problems in traditional inversion methods.

[0121] Example 5, this example is based on the above examples, see below. Figure 1 and Figure 5The groundwater level prediction is used to generate groundwater level prediction results at different time scales, including short-term, medium-term and long-term. Specifically, based on the enhanced hydrological derived feature data and the groundwater state inversion data, a multi-time-domain groundwater level prediction model that combines physical constraints and quantile consistency improvement is used to predict the groundwater level and obtain multi-scale water level prediction data. The steps include: input data integration, shared encoder construction, multi-time-domain decoding prediction, physical residual constraint optimization, monotonic response constraint optimization, improved loss model training and groundwater level prediction.

[0122] The input data integration is used to unify and integrate the enhanced hydrological derived feature data and the groundwater state inversion data. Specifically, it is to pair the basic hydrological derived features and the inverted water level field inversion data, anisotropic parameter field inversion data and groundwater flow velocity and direction inversion data within the same observation period onto the same time axis and spatial reference to obtain a spatiotemporal sample sequence.

[0123] The shared encoder is constructed by building a standard convolutional long short-term neural network as the encoder structure based on the spatiotemporal sample sequence, and using unified convolutional layer and long short-term memory unit parameters to extract shared features from the spatiotemporal sample sequence to obtain multi-temporal shared feature encoding data.

[0124] Preferably, Table 1 is a parameter example table of the shared encoder. As shown in the table, in one specific embodiment, a shared convolutional long short-term neural network is used as the encoder. The shared characteristic is reflected in the fact that for spatiotemporal sample sequence data input from different monitoring points or different time scales, the same convolutional kernel parameters and LSTM unit parameters are uniformly used for processing. Instead of training an independent encoder for each task, feature extraction is achieved through a set of shared parameters. This can reduce the model parameter size, improve computational efficiency, and ensure the consistency of the feature space in the process of multi-task or multi-time domain decoding.

[0125] Table 1. Example of parameters for shared encoders

[0126]

[0127] The multi-temporal decoding prediction specifically involves constructing a multi-branch decoding network and, based on the multi-temporal shared feature encoding data, using a long short-term memory network decoder to perform short-term prediction decoding, medium-term prediction decoding, and long-term prediction decoding to obtain multi-temporal groundwater level prediction data.

[0128] In a preferred embodiment, the multi-temporal decoding prediction is achieved by constructing a multi-branch decoding network, which specifically includes a short-term prediction branch, a medium-term prediction branch, and a long-term prediction branch.

[0129] The short-term prediction branch uses a two-layer LSTM decoder with 64 hidden units per layer, and is mainly used to predict groundwater level changes in the next 1 to 3 days.

[0130] The intermediate prediction branch uses a two-layer LSTM decoder with 128 hidden units per layer, which is suitable for predicting groundwater level trends for the next week.

[0131] The long-term prediction branch uses a three-layer LSTM decoder with 256 hidden units per layer, which is suitable for predicting long-term evolution trends over the next month.

[0132] The physical residual constraint optimization specifically involves constructing adaptive physical weights to dynamically penalize the physical constraint loss function of the groundwater level prediction model, thereby obtaining physical residual constraint weight data.

[0133] The formula for calculating the physical residual constraint weight data is as follows:

[0134] ;

[0135] In the formula, It is physical residual constraint weight data. This is the basic weighting coefficient, with a default value of 0.4. At the predicted time The 90th quantile of the predicted distribution at time t, where t is the index of the discrete time. It is the time step of the predicted moment. At the predicted time The 10th percentile of the predicted distribution at that time. It is a smoothing constant;

[0136] The monotonic response constraint optimization specifically involves constructing monotonic constraint loss functions for recharge and pumping rates, applying monotonic soft constraints for groundwater level prediction, and obtaining the monotonic response constraint loss function.

[0137] The formula for calculating the monotonic response constraint loss function is as follows:

[0138] ;

[0139] In the formula, It is a monotonic response-constrained loss function. It is an increase in the supply quantity parameter The groundwater level value predicted by the model afterward. It is the perturbation step size parameter. This is the predicted water level value based on the baseline replenishment amount. It is due to the increase in pumping volume The groundwater level value predicted by the model afterward. This is the predicted water level value under the baseline pumping rate;

[0140] The improved loss model training specifically involves constructing an ensemble loss function, using the physical residual constraint weight data as the ensemble weights of the physical constraint loss function, and simultaneously introducing the monotonic response constraint loss function, the quantile consistency loss function, and the horizon-aware smoothing loss function to obtain an improved loss function. The improved loss function is then used in conjunction with the shared encoder and the multi-temporal decoder to train the groundwater level prediction model, resulting in a groundwater level prediction model that balances physical consistency, response rationality, quantile stability, and temporal smoothness.

[0141] The formula for calculating the improved loss function is as follows:

[0142] ;

[0143] In the formula, It is an improved loss function, L pred It is the basic prediction error loss, specifically using the mean squared error loss function. It is the physical residual constraint weight data, R phys It is the physical residual constraint loss, specifically the physical constraint loss based on the groundwater flow equation. These are monotonic response constraint weights. It is a monotonic response-constrained loss function. It is the quantile consistency weight. It is the quantile consistency loss function. It is a horizon-perceived smoothing weight.

[0144] It is the horizon-aware smoothing loss function;

[0145] The formula for calculating the quantile consistency loss function is as follows:

[0146] ;

[0147] In the formula, At the predicted time The 50th percentile of the predicted distribution at that time;

[0148] The calculation formula for the horizon-aware smoothing loss function is as follows:

[0149] ;

[0150] In the formula, H is the total prediction time step. It is a time-series smoothing weighting factor that decays with the prediction step size, and the specific calculation formula is as follows: ,in, This is the initial weighting coefficient, with a default value of 0.2. It is a time scale parameter used to control the rate of weight decay. It is the prediction time after introducing the time interval parameter. The 50th percentile of the predicted distribution at that time;

[0151] The groundwater level prediction specifically involves using the groundwater level prediction model to predict the groundwater level based on the enhanced hydrological derived feature data and the groundwater state inversion data, thereby obtaining multi-scale water level prediction data.

[0152] By performing the above operations, this solution addresses the technical problems in existing predictive water level modeling methods, such as the disconnect between the prediction model and hydrophysical mechanisms, and the lack of stability and reasonable responsiveness in the prediction results. It creatively employs a multi-temporal groundwater level prediction model that combines physical constraints and improved quantile consistency. This model is developed by introducing physical residual constraints, monotonic response constraints, quantile consistency constraints, and temporal smoothing constraints into a shared encoder and multi-branch decoding network, constructing an improved loss function and training it. This method not only ensures the consistency of prediction results across different time scales but also provides a reasonable response to changes in pumping and recharge, significantly improving the physical reliability and temporal stability of groundwater level predictions.

[0153] Example 6, this example is based on the above examples, see below. Figure 1 The remote sensing enhancement is used to generate remote sensing enhancement information by comprehensively predicting data. Specifically, based on the groundwater state inversion data and the multi-scale water level prediction data, information fusion and visualization data analysis are performed by integrating the inversion data and prediction data, and remote sensing enhancement of groundwater level is performed through remote data interaction to obtain remote sensing enhancement data.

[0154] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process or method.

[0155] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

[0156] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A remote perception enhancement system for groundwater level monitoring, characterized by: The system comprises a sensing collection module, a feature enhancement module, a groundwater inversion modeling module, a predictive water level modeling module and a monitoring awareness enhancement module. The sensing collection module is configured to collect multi-source sensing data, obtain sensing original data through the multi-source sensing data collection, and send the sensing original data to the feature enhancement module. The feature enhancement module is configured to enhance sensing features, obtain enhanced hydrological derived feature data through the sensing feature enhancement, and send the enhanced hydrological derived feature data to the groundwater inversion modeling module and the predictive water level modeling module. The groundwater inversion modeling module is configured to invert water flow velocity direction, obtain groundwater state inversion data through the water flow velocity direction inversion, and send the groundwater state inversion data to the predictive water level modeling module and the monitoring awareness enhancement module. The data assimilation initialization is configured to construct an initial state for inversion, specifically by reading feature data in the enhanced hydrological derived feature data, initializing inversion basic parameters and setting boundary conditions to obtain initialization groundwater state inversion data. The model explicit discrete modeling is configured to establish an iterative forward evolution model, specifically by establishing an explicit discrete groundwater evolution model based on a discrete expression of a groundwater flow equation according to the initialization groundwater state inversion data, performing groundwater state forward calculation, and obtaining a groundwater state discrete forward evolution model. The improved objective function construction is configured to establish a comprehensive error index, jointly model an observation consistent term, a physical residual term, an anisotropy regularization term and a boundary consistent term, and weight each term according to data quality and uncertainty to obtain an improved objective function. The calculation formula of the improved objective function is as follows: ； wherein, is the modified objective function, wherein h is the water level parameter, is the angle between the main axis of aquifer permeability and the coordinate axis, is the anisotropy ratio parameter of permeability, AE is the observation consistent term, is the weight of the physical residual term, PE is the physical residual term, is the weight of the anisotropy regularization term, DE is the anisotropy regularization term, is the weight of the boundary consistent term, BE is the boundary consistent term; The calculation formula of the physical residual term is as follows: ； where PE is a physical residual term, is an observed range spatial identifier, is the hydraulic conductive flux divergence; The calculation formula of the anisotropy regularization term is as follows: ； where DE is the anisotropic regularization term, is the gradient of the angle between the permeability principal axis of the aquifer and the coordinate axis, which is used to represent the variation of the permeability principal axis direction in space, is the regularized equilibrium coefficient, is the prior anisotropy ratio reference value, which is specifically obtained from the observed actual value; The calculation formula of the boundary consistent term is as follows: ； where BE is a boundary conformity term, is an observation range spatial boundary identifier, is a boundary conformity function that measures the deviation of the water level and the permeability tensor from the boundary conditions; The adjoint method extraction calculation is configured to calculate gradient information of a water level field and an anisotropy parameter field of the objective function by using the adjoint method according to the groundwater state discrete forward evolution model and the improved objective function, and perform alternating updates until parameter updates converge to obtain updated groundwater state parameters. The predictive water level modeling module is configured to predict groundwater level, obtain multi-scale water level prediction data through the groundwater level prediction, and send the multi-scale water level prediction data to the monitoring awareness enhancement module. The monitoring awareness enhancement module is configured to enhance remote awareness, obtain remote awareness enhancement data through the remote awareness enhancement.

2. A remote sensing enhanced system for monitoring groundwater level as claimed in claim 1 wherein: The multi-source sensing data collection is configured to collect and standardize multi-source sensing signals, specifically by collecting real-time multi-source sensing data to obtain sensing original data. The sensing original data specifically include groundwater well sensing data, meteorological environment sensing data, human activity interference data, and auxiliary positioning environment parameters.

3. A remote perception enhancement system for monitoring a groundwater level according to claim 2, characterized in that: The sensing feature enhancement is used for extracting and enhancing hydrological derived features, specifically for preprocessing, feature extraction, and multi-scale enhancement of the sensing original data, and combining with hydrological physical constraints for consistency correction to obtain enhanced hydrological derived feature data, including the following steps: data preprocessing, basic feature extraction, multi-scale feature enhancement, and physical consistency correction. The data preprocessing specifically includes time alignment and missing data completion of the sensing original data, which is used to ensure the consistency of the data in the time dimension and the spatial dimension, and obtains preprocessed data. The basic feature extraction specifically extracts hydrological features according to the preprocessed data to obtain hydrological feature data; the hydrological features specifically include groundwater level change rate, periodic fluctuation feature, and rainfall lag response feature. The multi-scale feature enhancement specifically includes scale decomposition and fusion of hydrological features, short-term rapid fluctuation feature and long-term trend feature enhancement, and obtains multi-scale enhanced features. The physical consistency correction specifically introduces a water balance constraint based on the groundwater flow equation as a water flow boundary condition to correct the enhanced features for physical consistency during the multi-scale feature enhancement process, which is used to avoid results inconsistent with actual hydrological mechanisms generated by pure data-driven, and obtains enhanced hydrological derived feature data. The enhanced hydrological derived feature data specifically include basic hydrological derived features, time domain and frequency domain features, and spatial physical constraint derived features.

4. A remote perception enhancement system for monitoring a water table according to claim 3, wherein: The flow velocity direction inversion is used for estimating the aquifer flow velocity and direction, and inferring the groundwater flow boundary and hydrological parameter distribution, specifically for using an anisotropic self-adaptive improved adjoint inversion method to perform flow velocity direction inversion according to the enhanced hydrological derived feature data, and obtaining groundwater state inversion data. The vector field calculation specifically calculates the groundwater flow velocity vector field according to the estimated water level gradient and anisotropic permeability parameter, extracts the flow direction angle and flow velocity, and obtains vector field estimation data according to the updated groundwater state parameters. The vector field estimation data includes flow velocity vector field estimation data and flow direction angle estimation data. The flow velocity direction inversion specifically obtains groundwater state inversion data by consistency evaluation of the inversion results combined with boundary conditions according to the vector field estimation data.

5. A remotely sensed augmented system for monitoring groundwater levels as claimed in claim 4, wherein: The groundwater state inversion data specifically include water level field inversion data, anisotropic parameter field inversion data, groundwater flow velocity and direction inversion data, and boundary consistency evaluation data.

6. A remotely sensed augmented system for monitoring groundwater levels as claimed in claim 5 wherein: The underground water level prediction is used for generating underground water level prediction results of different time scales of short, medium and long term, specifically, according to the enhanced hydrological derived feature data and the underground water state inversion data, an improved multi-time-domain underground water level prediction model combining physical constraints and quantile consistency is used for underground water level prediction, and multi-scale water level prediction data is obtained, including the following steps: input data integration, shared encoder construction, multi-time-domain decoding prediction, physical residual constraint optimization, monotonic response constraint optimization, improved loss model training and underground water level prediction. The physical residual constraint optimization specifically dynamically punishes the physical constraint loss function of the underground water level prediction model by constructing an adaptive physical weight to obtain physical residual constraint weight data. The calculation formula of the physical residual constraint weight data is: ； wherein is the physical residual constraint weight data, is the base weight coefficient, is the 90th percentile of the prediction distribution at prediction time t is the discrete time index, is the prediction time time step, is the 10th percentile of the prediction distribution at prediction time t is the discrete time index, is the smoothing constant; The monotonic response constraint optimization specifically constructs a monotonic constraint loss function of the recharge and pumping amount to perform monotonicity soft constraint of the underground water level prediction, and obtains a monotonic response constraint loss function. The improved loss model training specifically constructs an integrated loss function, uses the physical residual constraint weight data as the integrated weight of the physical constraint loss function, and simultaneously introduces a monotonic response constraint loss function, a quantile consistency loss function and a horizon perception smoothing loss function to obtain an improved loss model, and uses the improved loss model to train the underground water level prediction model in combination with the shared encoder and the multi-time-domain decoder to obtain an underground water level prediction model considering physical consistency, reasonable response, quantile stability and time sequence smoothness. The calculation formula of the improved loss model is: ； wherein, is an improved loss model, L pred is a base prediction error loss, specifically a mean squared error loss function, is a physical residual constraint weight data, R phys is a physical residual constraint loss, specifically a physical constraint loss based on a groundwater flow equation, is a monotonic response constraint weight, is a monotonic response constraint loss function, is a quantile consistency weight, is a quantile consistency loss function, is a horizon-aware smoothing weight, is a horizon-aware smoothing loss function; The calculation formula of the quantile consistency loss function is: ； wherein is the 50% quantile of the predictive distribution at the prediction time t. The calculation formula of the horizon perception smoothing loss function is: ； where H is the predicted total time step, is a timing smoothing weight factor decaying with the prediction step, which is calculated as where, is the initial weight coefficient, is a time scale parameter to control the speed of weight decay, is the 50% quantile of the predicted distribution at the prediction time after introducing the time interval parameter; The underground water level prediction specifically uses the underground water level prediction model to predict the underground water level according to the enhanced hydrological derived feature data and the underground water state inversion data, and obtains multi-scale water level prediction data.

7. A remotely sensed augmented system for monitoring groundwater levels as claimed in claim 6, wherein: The input data integration is used for uniformly integrating the enhanced hydrological derived feature data and the underground water state inversion data, specifically, the basic hydrological derived features and the inversion data of the water level field, the anisotropy parameter field and the underground water flow velocity and flow direction obtained by inversion in the same observation period are paired on the same time axis and spatial reference to obtain a time-space sample sequence. The shared encoder construction specifically constructs a standard convolutional long short-term neural network as an encoder structure according to the time-space sample sequence, and uses a unified convolutional layer and long short-term memory unit parameter to share feature extraction of the time-space sample sequence to obtain multi-time-domain shared feature encoding data. The multi-time-domain decoding prediction specifically constructs a multi-branch decoding network, uses a long short-term memory network decoder to respectively perform short-term prediction decoding, medium-term prediction decoding and long-term prediction decoding according to the multi-time-domain shared feature encoding data, and obtains multi-time-domain underground water level prediction data.

8. A remotely sensed augmented system for monitoring groundwater levels according to claim 7, wherein: The remote awareness enhancement is used for generating remote awareness enhancement information based on comprehensive prediction data, specifically, remote awareness enhancement data is obtained by performing information fusion and visual data analysis on the basis of the groundwater state inversion data and the multi-scale water level prediction data, and performing remote data interaction on the groundwater level remote awareness enhancement.

Images