An automatic parameter calibration method and system for a surface water-groundwater coupling model

Abstract

The present application relates to the technical field of hydrological model parameter calibration, in particular to a parameter automatic calibration method and system for a surface water-groundwater coupling model, comprising: obtaining multi-dimensional measured and model simulation output sequences of surface water and groundwater subsystems, fusing and constructing a coupling hydrological data entity containing time variation and spatial distribution characteristics, separating non-steady response characteristics according to the space-time deviation state after applying disturbance, filtering out the disturbance to form a steady-state data entity, and matching with a multi-mode parameter response template library. The adaptive structure decomposition generates a hierarchical model parameter feature cluster set, constructs a surface water and groundwater parameter dynamic constraint network, and obtains a preliminary adjustment scheme through synchronous optimization and cross-validation. After positioning the key conduction and boundary interaction parameter nodes, the optimization scheme path is optimized to generate a final parameter calibration instruction sequence. The method eliminates non-steady-state data interference, adapts to the parameter correlation characteristics of the coupling system, and realizes adaptive and accurate execution of parameter calibration.

Description

Technical Field

[0001] This invention relates to the field of hydrological model parameter calibration technology, and in particular to an automatic parameter calibration method and system for a surface water-groundwater coupled model. Background Technology

[0002] Existing methods for calibrating parameters of surface water-groundwater coupled models often directly compare single-dimensional measured observation data with model simulation output data. They process the parameters of the surface water subsystem and the groundwater subsystem separately using independent optimization modes, and perform parameter optimization based on fixed parameter response rules. They do not fuse multi-dimensional measured data with simulation output data, nor do they construct coupled hydrological data entities that integrate time-varying response characteristics and spatial distribution characteristics.

[0003] Existing calibration methods lack the ability to apply perturbations to coupled hydrological data and analyze spatiotemporal deviations. They cannot identify and separate unstable response features from the data, nor can they form de-perturbed steady-state data entities by filtering out interference features. Furthermore, they lack a matching mechanism between a multi-mode parameter response template library and steady-state data, meaning that unstable components in the data directly affect the accuracy of parameter matching. Parameter processing only employs fixed-structure decomposition, failing to achieve adaptive structural decomposition, thus preventing the formation of hierarchical model parameter feature clusters. They also lack a dynamic constraint network for surface water movement parameters and groundwater seepage parameters, do not support synchronous parameter optimization and cross-validation, cannot locate key transmission parameter nodes and boundary interaction parameter nodes, and cannot optimize parameter adjustment schemes. Ultimately, the generated parameter calibration commands are ill-suited to the interactive transmission characteristics of the coupled system.

[0004] It is necessary to achieve multi-source data fusion and separation of unstable features, complete the matching of steady-state data and parameter response templates, construct a dynamic constraint network through adaptive decomposition, and complete the path optimization of the calibration scheme based on key parameter nodes, so as to form an automatic parameter calibration process that adapts to the coupled system. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and to propose an automatic parameter calibration method and system for a surface water-groundwater coupling model.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: an automatic parameter calibration method for a surface water-groundwater coupled model, comprising:

[0007] Obtain multi-dimensional measured observation sequences and model simulation output sequences of the surface water subsystem and the groundwater subsystem;

[0008] By fusing multi-dimensional measured observation sequences and model simulation output sequences, a coupled hydrological data entity containing time-varying response characteristics and spatial distribution characteristics is constructed.

[0009] Perturbation is applied to coupled hydrological data entities, and unstable response features are identified and separated based on the deviation of the data in the spatial and temporal dimensions after perturbation.

[0010] Non-steady response features are filtered out from coupled hydrological data entities to form de-disturbed steady-state data entities, and these de-disturbed steady-state data entities are matched with a pre-established multi-mode parameter response template library.

[0011] Based on the matching results, adaptive structural decomposition is performed on the disturbed steady-state data entities to generate a set of model parameter feature clusters with hierarchical structure;

[0012] Based on the set of model parameter feature clusters, a dynamic constraint network is constructed between surface water movement parameters and groundwater seepage parameters.

[0013] A dynamic constraint network is used to simultaneously optimize and cross-validate the feature cluster set of model parameters to generate a preliminary parameter adjustment scheme.

[0014] Based on the preliminary parameter adjustment plan, key transmission parameter nodes and boundary interaction parameter nodes are located in the model parameter feature cluster set;

[0015] By combining key transmission parameter nodes and boundary interaction parameter nodes, the path of the preliminary parameter adjustment scheme is optimized to generate the final parameter calibration instruction sequence.

[0016] As a further aspect of the present invention, the multi-dimensional measured observation sequence and the model simulation output sequence are fused to construct a coupled hydrological data entity containing time-varying response characteristics and spatial distribution characteristics, including:

[0017] The measured data matrix is ​​constructed by extracting the cross-sectional flow, water level and velocity distribution of the river channel from the measured observation sequence of the surface water subsystem, and extracting the groundwater level, pore water pressure and water quality component concentration at different depths from the measured observation sequence of the groundwater subsystem.

[0018] Simultaneously, the model simulation data matrix that completely corresponds to the measured data matrix in the spatiotemporal dimension is extracted from the simulation output sequence of the surface water-groundwater coupling model;

[0019] The measured data matrix and the model simulation data matrix are interpolated and aligned within a unified spatiotemporal grid framework to form a spatiotemporally aligned comparison data field.

[0020] In the comparative data field, for each spatiotemporal grid point, the residual vector between the measured data and the simulated data is calculated, and the residual vector is bound to the measured data vector and the simulated data vector of the corresponding spatiotemporal grid point;

[0021] All spatiotemporal grid point data are bundled and spatially clustered according to the boundaries of hydrogeological units and surface catchment areas. A time evolution identifier is added to each spatial cluster unit to form a coupled hydrological data entity.

[0022] As a further aspect of the present invention, a perturbation is applied to the coupled hydrological data entity, and unstable response features are identified and separated based on the deviation of the perturbed data in the spatial and temporal dimensions, including:

[0023] For the model simulation data vector in each spatial clustering unit of the coupled hydrological data entity, a set of random noise with increasing intensity is injected, and the change trajectory of the residual vector between the bundled measured data vector and the simulated data vector after noise injection is observed.

[0024] Record the magnitude and direction of the residual vector at each spatiotemporal grid point under each intensity level to form a disturbance response surface;

[0025] Analyze the continuity of the disturbance response surface on the time axis. If the disturbance response surface shows abrupt changes or discontinuous jumps in adjacent time slices, the data response corresponding to the time slice is determined to be unstable.

[0026] Analyze the spatial diffusion range of the disturbance response surface. If the disturbance of a single grid point causes the residual change to remain significant beyond the preset spatial neighborhood, then the data response of the spatial region is determined to be unstable.

[0027] Extract the original bundled data corresponding to all time segments and spatial regions that are determined to be unstable, mark and separate them from the coupled hydrological data entities, and summarize them to form an unstable response feature set.

[0028] As a further aspect of the present invention, unstable response features are filtered from coupled hydrological data entities to form de-disturbed steady-state data entities, and the de-disturbed steady-state data entities are matched with a pre-established multi-mode parameter response template library, including:

[0029] All bundled data belonging to the unstable response feature set are removed from the coupled hydrological data entity, and the remaining part constitutes the de-disturbed steady-state data entity.

[0030] The multi-mode parameter response template library stores the standard response patterns that surface water and groundwater observation data should exhibit when the combination of key model parameters changes under different typical hydrogeological conditions. Each standard response pattern is defined by a set of feature vectors and weight matrices.

[0031] Iterate through each spatial clustering unit in the perturbation-free steady-state data entity, and calculate the multidimensional similarity between the residual vector, the measured data vector of the spatial clustering unit and the feature vector of each standard response mode in the multi-mode parameter response template library;

[0032] Standard response patterns with multidimensional similarity exceeding a set threshold are selected as candidate matching patterns for the spatial clustering unit, and the weight matrix of each candidate matching pattern is recorded.

[0033] The candidate matching patterns and their weight matrices of all spatial clustering units in the perturbation-free steady-state data entity are integrated according to the time evolution identifier to form a global matching relationship graph.

[0034] As a further aspect of the present invention, based on the matching results, an adaptive structural decomposition is performed on the perturbation-free steady-state data entities to generate a set of model parameter feature clusters with a hierarchical structure, including:

[0035] Based on the global matching relationship graph, identify spatial clustering unit clusters with high weights to the same standard response pattern in the disturbed steady-state data entities;

[0036] Taking each spatial clustering unit cluster as a unit, and combining its temporal evolution identifier, the model simulation data vector and the measured data vector in all bundled data within the spatial clustering unit cluster are differentially processed to generate the parameter-sensitive difference field of the spatial clustering unit cluster.

[0037] Multi-scale analysis of parameter-sensitive difference fields is performed to separate long-term trend components, periodic components, and random components on the time scale, and regional background field components and local anomaly field components on the spatial scale.

[0038] The components at different time and spatial scales are clustered and reorganized according to their correlation with the feature vectors defined in the standard response pattern to form multiple primary feature clusters.

[0039] Based on the spatiotemporal correlation between primary feature clusters, primary feature clusters with strong correlations are merged, and a hierarchy is assigned to the merged feature clusters. Finally, a set of model parameter feature clusters with a hierarchical structure is generated, in which high-level feature clusters reflect large-scale dominant parameter features, and low-level feature clusters reflect local parameter features.

[0040] As a further aspect of the present invention, a dynamic constraint network between surface water movement parameters and groundwater seepage parameters is constructed based on the model parameter feature cluster set, including:

[0041] The physical meaning of the hydrological process corresponding to each feature cluster in the set of analytical model parameter feature clusters is analyzed, and the feature clusters are mapped to a subset of surface water movement parameters or a subset of groundwater seepage parameters.

[0042] Identify hybrid feature clusters that simultaneously contain components from subsets of surface water movement parameters and subsets of groundwater seepage parameters, wherein the hybrid feature clusters indicate nodes of coupling between parameters;

[0043] Using each model parameter as a network node, and the coupling nodes between parameters and the physical connection relationships of hydrological processes as edges, a parameter association graph is initially constructed.

[0044] Based on the hierarchical structure of the model parameter feature cluster set, dynamic weights are assigned to each edge in the parameter association graph. The dynamic weights are determined by the hierarchy of the feature clusters to which the two connected parameters belong, the magnitude of the components within the feature clusters, and the time evolution trend, thus forming a dynamic constraint network.

[0045] In a dynamic constraint network, a state variable is defined for each node. The state variable consists of the integrated information of all feature clusters corresponding to the parameters. The edges between nodes define the transmission and constraint rules of the state variable.

[0046] As a further aspect of the present invention, a dynamic constraint network is used to simultaneously optimize and cross-validate the feature cluster set of model parameters to generate a preliminary parameter adjustment scheme, including:

[0047] Input the set of model parameter feature clusters into the dynamic constraint network to activate the state variables of each node in the network;

[0048] Based on the connection and weight of edges in the dynamic constraint network, the state variables are iteratively transmitted and negotiated between adjacent nodes. Each node updates the optimal estimate of its own state variables according to the received state information of adjacent nodes and its own feature cluster information.

[0049] During the network state update process, a consistency verification rule is established: for parameter node pairs connected through a hybrid feature cluster, the trend of their state variables must satisfy the physical coupling relationship defined by the hybrid feature cluster.

[0050] If the state changes of a parameter node pair do not meet the consistency check rules, a backtracking mechanism is triggered to adjust the contribution weights of the relevant feature clusters and to re-update and iterate the network state until all relevant parameter node pairs pass the consistency check.

[0051] When the overall state of the dynamic constraint network tends to stabilize, that is, when the change in the state variables of each node is lower than the convergence threshold, the optimal estimate of the final state variables of each node is extracted.

[0052] The optimal estimates of the state variables for each model parameter node are converted into suggestions for the direction and magnitude of parameter adjustments, and these are summarized to form a preliminary parameter adjustment plan.

[0053] As a further aspect of the present invention, based on the preliminary parameter adjustment scheme, key transmission parameter nodes and boundary interaction parameter nodes are located in the model parameter feature cluster set, including:

[0054] Analyze the preliminary parameter adjustment scheme, identify model parameters whose adjustment range exceeds the median of the global adjustment range, and mark the model parameters as highly sensitive parameters.

[0055] In a dynamic constrained network, the nodes corresponding to highly sensitive parameters are located, and the topological properties of these nodes in the network are analyzed.

[0056] Highly sensitive parameter nodes with high betweenness centrality in the dynamic constraint network are marked as key transmission parameter nodes. Key transmission parameter nodes are located at the intersection of multiple information transmission paths in the network.

[0057] In the dynamic constraint network, highly sensitive parameter nodes that have strong connection edges with both the surface water movement parameter subset and the groundwater seepage parameter subset nodes are marked as boundary interaction parameter nodes.

[0058] In the model parameter feature cluster set, locate all feature clusters associated with key transmission parameter nodes and boundary interaction parameter nodes, and extract detailed component composition and spatiotemporal distribution information of all feature clusters;

[0059] The process of combining key transmission parameter nodes and boundary interaction parameter nodes to optimize the path of the initial parameter adjustment scheme and generate the final parameter calibration instruction sequence includes:

[0060] For key transmission parameter nodes, analyze their main transmission paths in the dynamic constraint network. Based on the adjustment suggestions of upstream and downstream parameters along the transmission path, coordinate and optimize the adjustment range of key transmission parameter nodes to avoid excessive amplification or cancellation of parameter adjustments along the transmission path.

[0061] For boundary interaction parameter nodes, check whether the adjustment suggestions for the surface water movement parameters and groundwater seepage parameters connected to them meet the boundary conditions of mass conservation or energy conservation. If not, perform joint fine-tuning of the adjustment suggestions for the boundary interaction parameter nodes and their connected parameters until the preset boundary constraint conditions are met.

[0062] The proposed adjustments to all model parameters after coordination optimization and joint fine-tuning are sorted in the following order: priority is given to key transmission parameter nodes, followed by boundary interaction parameter nodes, and then the remaining parameters.

[0063] For each sorted parameter adjustment suggestion, add the feature cluster identifier of its source, the reason for the optimization, and the expected adjustment effective time window, and combine them to generate the final parameter calibration instruction sequence.

[0064] As a further aspect of the present invention, it also includes:

[0065] The final parameter calibration command sequence is injected into the surface water-groundwater coupled model to drive the model parameter update, and the feedback data stream generated by the updated model is captured and fed back to the acquisition process of multi-dimensional measured observation sequences and model simulation output sequences, including:

[0066] Modify the corresponding parameter values ​​of the surface water-groundwater coupling model in sequence according to the order, amplitude, and time window specified in the final parameter calibration instruction sequence.

[0067] After each parameter modification, the model is run to simulate a complete hydrological cycle, and the updated simulation output sequence of the model is collected.

[0068] The updated simulation output sequence is compared with the multi-dimensional measured observation sequence for the corresponding time period, and a new set of residual vectors is calculated.

[0069] The new set of residual vectors, the updated simulation output sequence and their corresponding parameter configurations are added as a new data sample to the previously obtained set of multi-dimensional measured observation sequences and model simulation output sequences.

[0070] When a certain number of new data samples are accumulated, the reconstruction of the coupled hydrological data entities is triggered, and a new round of automatic parameter calibration process is initiated.

[0071] As a further aspect of the present invention, the present invention also includes an automatic parameter calibration system for a surface water-groundwater coupled model. The system includes a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of the automatic parameter calibration method for a surface water-groundwater coupled model as described above.

[0072] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0073] By fusing multi-dimensional measured observation sequences with model simulation output sequences, a coupled hydrological data entity containing time-varying response characteristics and spatial distribution characteristics is constructed. Perturbation is applied to this data entity, and unstable response characteristics are identified and separated based on the deviation of the data in the spatial and temporal dimensions after perturbation. After filtering out the corresponding characteristics, a de-perturbed steady-state data entity is formed. The de-perturbed steady-state data entity is matched with a pre-established multi-mode parameter response template library, which can remove the non-steady-state disturbance components in the spatiotemporal dimensions of the data and retain the steady-state response attributes of the coupled hydrological system. The matching process between the data entity and the template library conforms to the dual characteristics of time-varying and spatial distribution data, avoids parameter matching offset caused by unstable responses, and makes the data matching results fit the actual operating state of the coupled hydrological system.

[0074] Based on the matching results, adaptive structural decomposition is performed on the disturbed steady-state data entities to generate a hierarchical set of model parameter feature clusters. A dynamic constraint network is constructed between surface water movement parameters and groundwater seepage parameters. The dynamic constraint network is used to complete the synchronous optimization and cross-validation of the parameter feature clusters. Key transmission parameter nodes and boundary interaction parameter nodes are located in the parameter feature clusters. The path of the preliminary parameter adjustment scheme is optimized by combining the two types of nodes, and the final parameter calibration instruction sequence is generated. This allows the parameter features to adapt to the parameter association logic of the coupled system in a hierarchical structure. The dynamic constraint network establishes the interaction relationship between the two types of parameters. Synchronous optimization and cross-validation reduce the ineffective steps of parameter adjustment. The location of key nodes clarifies the core path of parameter transmission and boundary interaction. The path optimization of the calibration scheme makes the instruction sequence fit the parameter interaction transmission logic of the coupled system. Attached Figure Description

[0075] Figure 1 This is a flowchart of an automatic parameter calibration method for a surface water-groundwater coupling model according to the present invention;

[0076] Figure 2 A flowchart for identifying and separating unstable response features;

[0077] Figure 3 Heatmap of dynamic constraint network weights;

[0078] Figure 4 A line graph illustrating the execution characteristics of the parameter calibration instruction sequence;

[0079] Figure 5 This is a graph showing the convergence curve of the model iteration residuals. Detailed Implementation

[0080] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0081] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0082] See Figure 1This invention provides an automatic parameter calibration method for a surface water-groundwater coupled model, the specific method including:

[0083] This process involves acquiring multi-dimensional measured observation sequences and model simulation output sequences for both the surface water and groundwater subsystems. These sequences are then fused to construct a coupled hydrological data entity that simultaneously contains time-varying response and spatial distribution characteristics. A perturbation is applied to this data entity, and the resulting spatiotemporal deviations are used to identify and separate unstable response features. After filtering out the unstable features, a de-perturbed steady-state data entity is formed and matched against a pre-generated multi-mode parameter response template library. Based on the matching results, an adaptive structural decomposition is performed on the steady-state data entity to generate a hierarchical set of model parameter feature clusters. Based on this set, a dynamic network characterizing the mutual constraints between surface water movement parameters and groundwater seepage parameters is constructed. This network is used to simultaneously optimize and cross-validate the feature cluster set, generating a preliminary parameter adjustment scheme. Key transmission parameter nodes and boundary interaction parameter nodes are located within the feature cluster set according to this scheme. The preliminary scheme is then optimized by combining these two types of key nodes, generating a final executable parameter calibration instruction sequence. The automatic parameter calibration method is a general process. Its core steps include: fusing multi-source data to construct coupled hydrological data entities; applying perturbations to separate unsteady features; matching with a multi-model parameter response template library; generating parameter feature clusters through adaptive decomposition; and constructing and utilizing a dynamic constraint network for collaborative optimization and path optimization. This method does not depend on a specific implementation of the surface water-groundwater coupling model, but is applicable to various surface water-groundwater coupling models that can provide the aforementioned multi-dimensional spatiotemporal output sequences. It is particularly suitable for handling parameter calibration problems under various typical hydrogeological conditions, such as homogeneous unconfined aquifers, layered confined aquifers, and riparian filter zones.

[0084] In one embodiment of the present invention, after acquiring multi-dimensional measured observation sequences and model simulation output sequences, the river cross-sectional flow, water level, and velocity distribution are extracted from the measured sequences of the surface water subsystem, and the groundwater level, pore water pressure, and water quality component concentration at different depths are extracted from the measured sequences of the groundwater subsystem. These data together constitute a measured data matrix. Simultaneously, a model simulation data matrix that completely corresponds to the measured data matrix in the spatiotemporal dimension is extracted from the simulation output sequence of the surface water-groundwater coupling model. The measured data matrix and the model simulation data matrix are interpolated and aligned within a unified spatiotemporal grid framework to form a spatiotemporally aligned comparative data field. In this comparative data field, for each spatiotemporal grid point, the residual vector between the measured data and the simulation data is calculated, and this residual vector is bound to the measured data vector and the simulation data vector corresponding to that grid point. Subsequently, the bound data of all spatiotemporal grid points are spatially clustered according to the known boundaries of hydrogeological units and surface catchment areas, and a time evolution identifier is added to each formed spatial cluster unit, ultimately constructing a coupled hydrological data entity.

[0085] See Figure 2 The constructed coupled hydrological data entity is perturbed. Within each spatial clustering unit of the entity, a set of random noise with increasing intensity is injected into the model simulation data vector. The trajectory of the residual vector change between the bound measured data vector and the noise-injected simulation data vector is observed. The amplitude and direction of the residual vector change at each spatiotemporal grid point are recorded at each noise intensity level, forming a series of perturbation response surfaces. The continuity of these perturbation response surfaces on the time axis is analyzed. If the surface shows abrupt changes or discontinuous jumps in adjacent time slices, the data response corresponding to that time slice is determined to be an unstable response. The spatial diffusion range of the perturbation response surfaces is analyzed. If the perturbation of a single grid point causes the residual change to exceed the preset spatial neighborhood range and still shows significant changes, the data response of that spatial region is determined to be an unstable response. The original bound data corresponding to all time slices and spatial regions determined to be unstable are extracted, marked and separated from the coupled hydrological data entity, and summarized to form an unstable response feature set.

[0086] In practical implementation, multi-dimensional measured observation sequences and model simulation output sequences of the surface water subsystem and groundwater subsystem are acquired. An example scenario involves the coupled simulation of surface water and groundwater in a watershed. The measured observation sequences of the surface water subsystem include the daily flow sequence, hourly water level sequence, and cross-sectional velocity distribution sequence for river section RW01. The measured observation sequences of the groundwater subsystem include the daily groundwater level sequence, hourly pore water pressure sequence, and daily sequence of water quality components such as chloride ion concentration at different depths (e.g., 10m, 20m, and 30m) of monitoring well GW05. These sequences together constitute the measured data matrix, where rows represent time points and columns represent different observed variables. Simultaneously, simulated flow sequence, simulated water level sequence, simulated velocity distribution sequence, simulated groundwater level sequence, simulated pore water pressure sequence, and simulated water quality component concentration sequence corresponding to the exact same spatiotemporal location and time point are extracted from the simulation output of the running surface water-groundwater coupled model. These sequences constitute the model simulation data matrix, which has the same dimensions and spatiotemporal coverage as the measured data matrix. In practice, the measured data matrix and the model simulation data matrix are interpolated and aligned within a unified spatiotemporal grid framework. This framework is defined as a regular grid with a spatial resolution of 500 meters by 500 meters covering the entire study area, with a time step of one day. The alignment process uses bilinear interpolation to interpolate the discrete measured station data and model output data to each grid point, forming a spatiotemporally aligned comparative data field. Within this comparative data field, for each spatiotemporal grid point, the residual vector between the measured data vector and the simulated data vector is calculated. The residual vector is expressed as:

[0087]

[0088] in: Indicates spatial location and time The residual vector at that point, Represents the measured data vector. The simulated data vector, along with the measured and simulated data vectors, contains multiple dimensions including flow rate, water level, flow velocity, groundwater level, pore water pressure, and water quality component concentration. The residual vector is bundled with the measured and simulated data vectors of the corresponding spatiotemporal grid points. This bundling operation encapsulates the three vectors into a single data structure unit. All bundled data from spatiotemporal grid points are spatially clustered according to pre-digitized hydrogeological unit boundary maps and surface catchment area boundary maps. Spatial clustering uses boundary polygon-based attribution judgment to classify grid points into different hydrogeological units and catchment areas. Each spatial cluster unit is appended with a time evolution identifier, consisting of a start timestamp and an end timestamp. This ultimately forms a coupled hydrological data entity, a structured dataset containing a list of spatial cluster units, a bundled data set within each unit, and a time identifier.

[0089] In the specific implementation, a perturbation is applied to the coupled hydrological data entity. Within each spatial clustering unit of the coupled hydrological data entity, a set of increasing random noise is injected into the model simulation data vector. The random noise follows a normal distribution with zero mean and a diagonal covariance matrix. The increasing intensity is achieved through a standard deviation sequence, set to 0.05, 0.1, 0.15, 0.2, and 0.25 times the standard deviation of each dimension of the simulated data vector. The trajectory of the residual vector between the bundled measured data vector and the simulated data vector after noise injection is observed. The amplitude and direction of the residual vector at each spatiotemporal grid point are recorded for each intensity level. The amplitude is the difference in the Euclidean norm of the residual vector, and the direction is the sign change of each component of the residual vector, forming a perturbation response surface. This perturbation response surface is a four-dimensional array storing different intensity levels, spatial locations, time points, and residual changes. The continuity of the disturbance response surface along the time axis is analyzed. Continuity is evaluated by calculating the magnitude of the second-order difference vector of the residual changes between adjacent time slices. If the magnitude of the second-order difference vector exceeds a threshold (e.g., 0.5), the disturbance response surface is considered to have abrupt changes or discontinuous jumps between adjacent time slices, thus classifying the data response corresponding to the time segment as unstable. The spatial diffusion range of the disturbance response surface is analyzed. The diffusion range is evaluated by calculating the spatial variogram of the residual changes. If a disturbance at a single grid point causes the residual change to exceed a preset spatial neighborhood (e.g., 500 meters), and the range of the spatial variogram is still less than the threshold, the data response of the spatial region is classified as unstable. The original bundled data corresponding to all time segments and spatial regions classified as unstable are extracted, marked, and separated from the coupled hydrological data entities, and summarized into an unstable response feature set. This unstable response feature set is an independent dataset containing unstable bundled data and their spatiotemporal coordinates.

[0090] In one embodiment of the present invention, all bundled data belonging to the unstable response feature set are removed from the coupled hydrological data entity, and the remaining part constitutes the de-disturbed steady-state data entity. A pre-generated multi-mode parameter response template library stores the standard response patterns that surface water and groundwater observation data should exhibit when the key model parameter combinations change under different typical hydrogeological conditions. Each standard response pattern is defined by a set of predefined feature vectors and weight matrices. Each spatial clustering unit in the de-disturbed steady-state data entity is traversed, and the multidimensional similarity between the residual vector, measured data vector, and feature vector of each standard response pattern in the template library is calculated. Standard response patterns with multidimensional similarity exceeding a set threshold are selected as candidate matching patterns for the spatial clustering unit, and the weight matrix corresponding to each candidate matching pattern is recorded. The candidate matching patterns and their weight matrices of all spatial clustering units in the de-disturbed steady-state data entity are integrated according to their temporal evolution identifiers to form a global matching relationship graph.

[0091] Based on this global matching graph, spatial clustering units with high weights to the same standard response pattern are identified within the disturbed steady-state data entities. For each identified spatial clustering unit, combined with its temporal evolution identifier, the model simulation data vectors and measured data vectors from all bundled data within the cluster are differentially analyzed to generate the cluster's parameter-sensitive difference field. Multi-scale analysis is then performed on this parameter-sensitive difference field, separating long-term trend components, periodic components, and random components at the time scale, and regional background field components and local anomaly field components at the spatial scale. The components separated at different time and spatial scales are clustered and reorganized based on their correlation with the feature vectors defined in the standard response pattern, forming multiple primary feature clusters. Then, based on the spatiotemporal correlation between these primary feature clusters, strongly correlated primary feature clusters are merged, and a hierarchy is assigned to the merged feature clusters, ultimately generating a hierarchical set of model parameter feature clusters. Higher-level feature clusters reflect large-scale dominant parameter features, while lower-level feature clusters reflect more local parameter features.

[0092] In practice, all bundled data belonging to the unstable response feature set are removed from the coupled hydrological data entity. The remaining portion constitutes the de-disturbed steady-state data entity. This de-disturbed steady-state data entity is a data structure containing multiple spatial clustering units. Each spatial clustering unit contains steady-state bundled data, spatial attribution information, and temporal evolution identifiers. A pre-established multi-mode parameter response template library stores the standard response patterns that surface water and groundwater observation data should exhibit when key model parameter combinations change under different typical hydrogeological conditions. Typical hydrogeological conditions include homogeneous unconfined aquifers, layered confined aquifers, and riparian filters. Each standard response pattern is defined by a set of feature vectors and weight matrices. The feature vectors are mathematical vectors describing the magnitude of water level changes, flow decay rates, and spatial distribution patterns of concentration gradients. The weight matrix characterizes the importance of each dimension of the feature vector in the matching process. Each spatial clustering unit in the de-disturbed steady-state data entity is traversed. The multi-dimensional similarity between the residual vector, the measured data vector of the spatial clustering unit, and the feature vector of each standard response pattern in the multi-mode parameter response template library is calculated. The multi-dimensional similarity calculation expression is:

[0093]

[0094] in: This represents the final multidimensional similarity scalar value. and It is a preset weighting coefficient and , Represents the residual vector of a spatial clustering unit. Represents the measured data vector of a spatial clustering unit. This represents a feature vector representing a standard response pattern from a multi-mode parametric response template library. The function represents the calculation of cosine similarity. Standard response patterns with multidimensional similarity exceeding a set threshold are selected as candidate matching patterns for spatial clustering units, and the weight matrix of each candidate matching pattern is recorded. The weight matrix is ​​a matrix directly extracted from the multi-mode parameter response template library and paired with the feature vector. The candidate matching patterns and their weight matrices of all spatial clustering units in the de-perturbed steady-state data entity are integrated according to time evolution identifiers to form a global matching relationship graph. The global matching relationship graph is a graph structure where nodes are spatial clustering units and standard response patterns, and edges represent matching relationships and include weight matrix information.

[0095] In practical implementation, based on the global matching relationship graph, spatial clustering unit clusters with high weights for the same standard response mode are identified among the disturbed steady-state data entities. The criterion for high weight is that the mean of the main diagonal elements in the weight matrix of the connecting edges is greater than a threshold. Taking each identified spatial clustering unit cluster as a unit, and combining the temporal evolution identifier of the spatial clustering unit cluster, the model simulation data vector and the measured data vector in all bundled data within the spatial clustering unit cluster are differentially analyzed to generate the parameter-sensitive difference field of the spatial clustering unit cluster. The parameter-sensitive difference field is a spatiotemporal field that expresses the difference between the model simulation data vector and the measured data vector at each grid point at different time points. Multi-scale analysis is performed on the parameter-sensitive difference field. On the time scale, the empirical mode decomposition method is used to separate the long-term trend component, periodic component, and random component. On the spatial scale, two-dimensional wavelet transform is used to separate the regional background field component and the local anomaly field component. Components at different time and spatial scales are clustered and recombined based on the strength of their correlation with the eigenvectors defined in the standard response model. The strength of the correlation is measured by calculating the absolute value of the correlation coefficient between the component sequence and the projected eigenvector, forming multiple primary feature clusters. Each primary feature cluster contains a set of components that are close in terms of correlation. Primary feature clusters with strong correlations are merged based on their spatiotemporal correlation. Spatiotemporal correlation is evaluated by calculating the maximum value of the time-series cross-correlation function and the spatial distribution covariance of the components contained in different primary feature clusters. The merged feature clusters are then assigned a hierarchy, ultimately generating a hierarchical set of model parameter feature clusters. In this hierarchical set, higher-level feature clusters reflect combinations of large-scale dominant parameter features corresponding to large spatial ranges and long-period components, while lower-level feature clusters reflect combinations of local parameter features corresponding to small spatial ranges and short-period components.

[0096] In some embodiments, the weighting coefficients in multidimensional similarity calculation and The parameters can be dynamically adjusted based on the objectives of the calibration phase. For example, if more attention is paid to residual matching in the early stages of calibration, the parameters can be increased. Value. It is understandable that the generation of parameter-sensitive difference fields can directly perform statistical analysis within spatial clustering units of the residual vector, without recalculating the difference between the model-simulated data vector and the measured data vector. Optionally, in multi-scale analysis over time, the separation of long-term trend components, periodic components, and random components can be achieved using Fourier transform combined with filtering techniques. In some embodiments, the merging of primary feature clusters can be spatially constrained based on whether the hydrogeological units corresponding to the feature clusters are adjacent, merging only spatially adjacent and highly correlated primary feature clusters. Optionally, the hierarchical allocation of the model parameter feature cluster set with a hierarchical structure can adopt a top-down splitting method, first treating all components as a top-level cluster, and then splitting layer by layer according to the spatiotemporal correlation threshold. It is understandable that the feature vectors of the standard response mode can be extracted from historically successful calibration cases through principal component analysis, constructing a multi-mode parameter response template library in a data-driven manner.

[0097] In one embodiment of the present invention, the physical meaning of the hydrological process corresponding to each feature cluster in the set of model parameter feature clusters is analyzed. The physical meaning of the hydrological process is the physical mechanism represented by the data change pattern reflected by the feature cluster in the actual water cycle process. Specifically, the feature clusters are mapped to a subset of surface water movement parameters or a subset of groundwater seepage parameters. The subset of surface water movement parameters includes the Manning roughness coefficient, which controls the resistance of river flow, and the river hydraulic conductivity coefficient, which characterizes the permeability of the riverbed. The subset of groundwater seepage parameters includes the aquifer permeability coefficient, which determines the flow capacity of groundwater, the specific yield, which reflects the water release characteristics of the aquifer, and the storage rate, which characterizes the water storage capacity. Mixed feature clusters that simultaneously contain components from both the subset of surface water movement parameters and the subset of groundwater seepage parameters are identified. These mixed feature clusters indicate the coupling nodes between parameters. Using each model parameter as a network node and the coupling nodes between parameters and the physical connection relationships of the hydrological process as edges, a parameter association graph is initially constructed. Based on the hierarchical structure of the model parameter feature cluster set, each edge in the parameter association graph is assigned a dynamic weight. This dynamic weight is determined by the hierarchy of the feature clusters to which the two parameters connected by the edge belong, the magnitude of the components within the feature cluster, and the time evolution trend, thus forming a dynamic constraint network. In this dynamic constraint network, a state variable is defined for each node. The state variable consists of the integrated information of all feature clusters corresponding to that parameter, and the edges between nodes define the rules for the propagation and constraint of the state variable.

[0098] The model parameter feature cluster set is input into the dynamic constraint network to activate the state variables of all nodes in the network. Based on the connection relationships and weights of the edges in the dynamic constraint network, the state variables are iteratively transmitted and negotiated between adjacent nodes. Each node updates its optimal estimate of its state variables based on the received state information of its adjacent nodes and its own feature cluster information. During the network state update process, a consistency verification rule is established. For parameter node pairs connected by mixed feature clusters, the trend of their state variable changes must satisfy the physical coupling relationship defined by the mixed feature cluster. If the state change of a parameter node pair does not satisfy the consistency verification rule, a backtracking mechanism is triggered, the contribution weights of the relevant feature clusters are adjusted, and the network state update iteration is repeated until all relevant parameter node pairs pass the consistency verification. When the overall state of the dynamic constraint network tends to stabilize, that is, when the change in the state variables of each node is lower than the preset convergence threshold, the final optimal estimate of the state variables of each node is extracted. The optimal estimate of the state variables of each model parameter node is converted into suggestions for the adjustment direction and magnitude of the model parameters, and these are summarized to form a preliminary parameter adjustment scheme.

[0099] In practical implementation, the physical meaning of the hydrological process corresponding to each feature cluster in the hierarchical model parameter feature cluster set is analyzed. The feature clusters are mapped to subsets of surface water movement parameters or groundwater seepage parameters. The surface water movement parameter subset includes the Manning roughness coefficient and the river hydraulic conductivity coefficient, while the groundwater seepage parameter subset includes aquifer permeability coefficient, specific yield, and storage rate. Hybrid feature clusters containing components from both the surface water movement parameter subset and the groundwater seepage parameter subset are identified. These hybrid feature clusters indicate the coupling nodes between parameters. For example, a hybrid feature cluster may simultaneously contain components related to river level changes and components related to adjacent groundwater level changes; this hybrid feature cluster indicates the coupling node between the river hydraulic conductivity coefficient and the aquifer permeability coefficient. Using each model parameter as a network node and the coupling nodes between parameters and the physical connections of the hydrological process as edges, a preliminary parameter association graph is constructed. The physical connections are determined based on the water cycle process; for example, the surface water river node and the groundwater aquifer node are connected through the riverbed sedimentary layer node.

[0100] In practical implementation, based on the hierarchical structure of the model parameter feature cluster set, dynamic weights are assigned to each edge in the parameter association graph, forming a dynamic constraint network. The assignment of dynamic weights is determined by the hierarchy of the feature clusters to which the two connected parameters belong, the magnitude of the components within the feature clusters, and the time evolution trend, specifically calculated through a weight function. In the dynamic constraint network, state variables are defined for each node. These state variables consist of integrated information from all feature clusters corresponding to the parameter. The integrated information includes the feature cluster hierarchy, the normalized vector of component magnitudes, and the time trend coefficient. The edges between nodes define the transmission and constraint rules of state variables. The transmission rules specify the weight decay method when state variable information is transmitted along the edge, and the constraint rules specify the hydraulic connection that the state variables connecting two nodes must satisfy in terms of change.

[0101] In practical implementation, a hierarchical set of model parameter feature clusters is input into the dynamic constraint network to activate the state variables of each node. Based on the edge connections and dynamic weights in the dynamic constraint network, state variables are iteratively transmitted and negotiated between adjacent nodes. Each node updates its optimal estimate of state variables based on the received state information of adjacent nodes and its own feature cluster information. The update process uses a combination of weighted averaging and gradient descent. During the network state update process, a consistency verification rule is established. For parameter node pairs connected by mixed feature clusters, the trend of their state variable changes must satisfy the physical coupling relationship defined by the mixed feature clusters. For example, the change in river water level and the change in adjacent groundwater level should maintain a certain hydraulic gradient relationship. If the state change of a parameter node pair does not satisfy the consistency verification rule, a backtracking mechanism is triggered, the contribution weights of the relevant feature clusters are adjusted, and the network state update iteration is repeated until all relevant parameter node pairs pass the consistency verification. When the overall state of the dynamic constraint network tends to stabilize, that is, when the change in the state variables of each node is lower than the convergence threshold, the final optimal estimate of the state variables of each node is extracted. The convergence threshold is set to the absolute value of the change of each component of the state variable being less than 1e-4. The optimal estimates of the state variables for each model parameter node are converted into suggestions for the direction and magnitude of parameter adjustments. These suggestions are then summarized to form a preliminary parameter adjustment scheme. The conversion is based on a linear mapping between the component magnitude vectors in the optimal estimates of the state variables and a preset parameter sensitivity coefficient matrix. See Table 1.

[0102] Table 1: Example Table of Model Parameters and Feature Cluster Mapping Relationship

[0103]

[0104] In some embodiments, state variable updates in a dynamic constraint network can be implemented using a message-passing algorithm, where each node sends and receives state information to its neighboring nodes. It is understood that the physical coupling relationships in the consistency verification rules can be expressed as a set of linear or nonlinear constraint equations, which are then added as hard constraints or soft penalty terms to the network state update process. Optionally, adjusting the contribution weights of relevant feature clusters in the backtracking mechanism can be achieved by reducing the weight coefficients of feature clusters that cause inconsistencies in the composition of state variables. In some embodiments, the parameter adjustment direction and magnitude suggestions in the initial parameter adjustment scheme can include confidence intervals, which are derived from the variance estimates of node state variables during the dynamic constraint network state update process. Optionally, the convergence determination of the dynamic constraint network can be based on the norm of the global state change vector, rather than just the change in the state variables of a single node.

[0105] See Figure 3 This is a dynamic constraint network weight heatmap, intuitively presenting the correlation strength of four core parameters and clearly identifying key coupling nodes, providing a topological basis for subsequent parameter optimization. The weight values ​​are calculated jointly by the feature cluster level to which the parameter belongs, component amplitude, and time evolution trend, serving as the core input for subsequent "state variable iterative transmission, consistency verification, and parameter adjustment." The parameters corresponding to high-weight edges are precisely the boundary interaction parameter nodes defined in the patent, which are the core optimization objects for parameter calibration. The higher the weight, the stronger the constraint during parameter adjustment, requiring simultaneous optimization of upstream and downstream parameters to avoid amplification or cancellation of adjustments. The high weights of the river channel hydraulic conductivity coefficient and aquifer permeability coefficient correspond to their jointly associated mixed feature cluster FC_07. This helps engineers quickly locate sensitive parameters in the model, optimize calibration efficiency, and improve the simulation accuracy of the surface water-groundwater coupling model.

[0106] In one embodiment of the present invention, a preliminary parameter adjustment scheme is analyzed to identify model parameters whose adjustment magnitude exceeds the median of the global adjustment magnitude, and these model parameters are marked as high-sensitivity parameters. In the dynamic constraint network, the nodes corresponding to these high-sensitivity parameters are located, and the topological properties of these nodes in the network are analyzed. High-sensitivity parameter nodes with high betweenness centrality in the dynamic constraint network are marked as key transmission parameter nodes; these nodes are located at the intersection of multiple information transmission paths in the network. High-sensitivity parameter nodes in the dynamic constraint network that have strong connections to both the surface water movement parameter subset and the groundwater seepage parameter subset are marked as boundary interaction parameter nodes. In the model parameter feature cluster set, all feature clusters associated with key transmission parameter nodes and boundary interaction parameter nodes are located, and the detailed component composition and spatiotemporal distribution information of these feature clusters are extracted.

[0107] For key transmission parameter nodes, their main transmission paths in the dynamic constraint network are analyzed. Based on the adjustment suggestions of upstream and downstream parameters along the transmission paths, the adjustment magnitude of key transmission parameter nodes is coordinated and optimized to avoid excessive amplification or cancellation of parameter adjustments along the transmission paths. For boundary interaction parameter nodes, it is checked whether the adjustment suggestions of the surface water movement parameters and groundwater seepage parameters connected to them satisfy the boundary conditions of mass conservation or energy conservation. If not, the adjustment suggestions of the boundary interaction parameter nodes and their connected parameters are jointly fine-tuned until the preset boundary constraint conditions are met. The adjustment suggestions of all model parameters after coordination optimization and joint fine-tuning are sorted in the order of priority for key transmission parameter nodes, followed by boundary interaction parameter nodes, and then the remaining parameters. For each sorted parameter adjustment suggestion, the feature cluster identifier of its source, the reason for optimization, and the expected adjustment effective time window are added to generate the final parameter calibration instruction sequence.

[0108] In the implementation, the preliminary parameter adjustment scheme is analyzed to identify model parameters whose adjustment range exceeds the median of the global adjustment range. These model parameters are then marked as highly sensitive parameters. The median of the global adjustment range is calculated based on the set of absolute values ​​of all suggested adjustment amounts for all parameters in the preliminary parameter adjustment scheme. The model parameters include the Manning roughness coefficient, the channel hydraulic conductivity coefficient, the aquifer permeability coefficient, and the specific yield. The suggested adjustment amounts for these parameters in the preliminary parameter adjustment scheme are +15%, -8%, +22%, and -5%, respectively. The set of absolute values ​​of all suggested adjustment amounts is {15, 8, 22, 5}, and the median is 11.5. Therefore, the model parameters whose adjustment range exceeds 11.5, namely the Manning roughness coefficient (+15%) and the aquifer permeability coefficient (+22%), are marked as highly sensitive parameters. In the dynamic constraint network, the nodes corresponding to the highly sensitive parameters are located, and the topological properties of the nodes in the dynamic constraint network are analyzed. The topological properties include the degree, betweenness centrality, and connection strength with nodes of different parameter subsets. Highly sensitive parameter nodes with high betweenness centrality in the dynamic constraint network are marked as key transmission parameter nodes. Betweenness centrality measures the frequency with which a node appears on all shortest paths in the network, and high betweenness centrality is defined as a betweenness value greater than the upper quartile of the betweenness values ​​of all nodes in the network. Highly sensitive parameter nodes with strong connections to both the surface water movement parameter subset and the groundwater seepage parameter subset in the dynamic constraint network are marked as boundary interaction parameter nodes. Strong connections are defined as edges whose dynamic weights are greater than the median weight of all edges in the network. In the model parameter feature cluster set, all feature clusters associated with key transmission parameter nodes and boundary interaction parameter nodes are located, and detailed component composition and spatiotemporal distribution information of the feature clusters are extracted. Detailed component composition includes the type, amplitude, and phase of the components, and spatiotemporal distribution information includes the time window of component activity and the main spatial region of influence.

[0109] In practical implementation, the initial parameter adjustment scheme is optimized by combining key transmission parameter nodes and boundary interaction parameter nodes. For key transmission parameter nodes, the main transmission paths of these nodes in the dynamic constraint network are analyzed. These main transmission paths are the sets of edges from the key transmission parameter node to its first- and second-level adjacent nodes. Based on the adjustment suggestions for upstream and downstream parameters along the main transmission paths, the adjustment magnitude of the key transmission parameter nodes is coordinated and optimized to avoid excessive amplification or cancellation of parameter adjustments along the transmission paths. This coordination optimization is achieved through a feedback adjustment function, which is:

[0110]

[0111] in: It refers to the adjustment range of key transmission parameter nodes after coordination and optimization. This refers to the initial adjustment range of key transmission parameter nodes in the preliminary parameter adjustment scheme. It is the adjustment coefficient. It is the set of direct upstream nodes of the key transmission parameter nodes on the main transmission path. This refers to the adjustment range of the upstream node. For boundary interaction parameter nodes, check whether the adjustment suggestions for surface water movement parameters and groundwater seepage parameters connected to the boundary interaction parameter nodes satisfy the boundary conditions of mass conservation or energy conservation. The boundary conditions are expressed as constraint equations. ,in Represents the surface water movement parameters of the connection. The parameters represent groundwater seepage parameters connected to the boundary. If the boundary conditions are not met, the proposed adjustments to the boundary interaction parameter nodes and their connection parameters are jointly fine-tuned. This joint fine-tuning is achieved by solving an optimization problem with the objective of minimizing the adjustment changes and the boundary conditions as constraints. The proposed adjustments to all model parameters after coordination optimization and joint fine-tuning are sorted in the following order: key transmission parameter nodes first, boundary interaction parameter nodes second, and other parameters last. For each sorted parameter adjustment proposal, the source feature cluster identifier, the reason for optimization, and the expected adjustment effective time window are added, and these are combined to generate a final parameter calibration instruction sequence, which is a structured list. See Table 2.

[0112] Table 2: Highly Sensitive Parameters and Their Node Attributes

[0113]

[0114] In some embodiments, the identification of key transmission parameter nodes can be combined with eigenvector centrality indices, rather than just betweenness centrality. Eigenvector centrality reflects the degree of connection between a node and important nodes. It is understood that the coordinated optimization of the adjustment magnitude of key transmission parameter nodes can be iteratively performed until the adjustment magnitude changes of all nodes on the main transmission path tend to stabilize. Optionally, the process of checking whether boundary conditions are satisfied can be transformed into calculating the difference between the left and right sides of the constraint equation and determining whether the difference is within an allowable tolerance range. In some embodiments, the sorting rule of the final parameter calibration instruction sequence can further consider the expected impact of parameter adjustments, prioritizing adjustments with greater expected impact. Optionally, the expected adjustment effective time window can be determined based on the temporal distribution information of the feature clusters associated with the parameters, such as the principal period of the periodic components in the feature clusters. It is understood that the joint fine-tuning optimization problem in path optimization can be solved using linear programming methods to ensure computational efficiency.

[0115] See Figure 4 This is a line graph showing the execution characteristics of the parameter calibration instruction sequence. The downward trend of the red line directly verifies the effectiveness of the sorting logic. Instructions 1-3 correspond to key transmission parameters / boundary interaction parameters, with the highest importance. Instructions 4-7 correspond to secondary sensitive parameters, with their importance gradually decreasing. Instructions 8-10 correspond to ordinary parameters, with their importance stabilizing in the 75%-80% range. The stability of the blue line indicates that the model simulation complexity of all parameter adjustment instructions is comparable, with no extremely time-consuming operations. The time cost of the calibration process is controllable, enabling efficient iteration. Prioritizing the execution of highly important parameters does not incur additional time overhead, balancing optimization effectiveness and efficiency. Quantifying the time cost of the calibration process provides a time dimension basis for engineering deployment and data support for subsequent optimization of the calibration process, allowing for further refinement of adjustment strategies for highly important parameters.

[0116] In one embodiment of the present invention, the final parameter calibration command sequence is injected into the surface water-groundwater coupled model. The corresponding parameter values ​​of the model are modified sequentially according to the order, amplitude, and time window specified in the command sequence. After each parameter modification, the model is run to simulate a complete hydrological cycle, and the updated simulation output sequence is collected. The updated simulation output sequence is compared with the corresponding multi-dimensional measured observation sequence, and a new set of residual vectors is calculated. The new set of residual vectors, the updated simulation output sequence, and their corresponding parameter configurations are added as a new data sample to the previously acquired set of multi-dimensional measured observation sequences and model simulation output sequences. When a certain number of new data samples accumulate, the reconstruction of the coupled hydrological data entity is triggered, and a new round of automatic parameter calibration is initiated.

[0117] In practice, the final parameter calibration command sequence is injected into the surface water-groundwater coupled model. Following the order, magnitude, and time window specified in the final parameter calibration command sequence, the corresponding parameter values ​​of the surface water-groundwater coupled model are modified sequentially. The final parameter calibration command sequence is an ordered list containing multiple commands. Each command specifies a model parameter identifier, the target value or adjustment increment, and the simulation period time window in which the adjustment takes effect. For example, one command increases the value of the parameter "Manning roughness coefficient" by 5% during the simulation period "2023-06-01 to 2023-08-31", while another command decreases the value of the parameter "aquifer permeability coefficient" by 3% during the simulation period "throughout 2023". After each parameter modification according to instructions, the surface water-groundwater coupled model is run to perform a complete hydrological cycle simulation. A complete hydrological cycle simulation refers to covering at least one complete wet and dry season cycle, with a simulation duration of one year. The updated simulation output sequence is collected, including the river cross-sectional flow sequence, water level sequence, and groundwater level sequence calculated by the model after parameter modification. The updated simulation output sequence is compared with the corresponding multi-dimensional measured observation sequence, and a new residual vector set is calculated. The new residual vector set is the set of differences between the updated simulated data and the measured data at the same spatiotemporal grid point. The new residual vector set, the updated simulation output sequence, and their corresponding parameter configurations are added as a new data sample to the previously acquired multi-dimensional measured observation sequence and model simulation output sequence set. The new data sample is a structured data package containing input parameters, model output, and corresponding observation residuals. When a certain number of new data samples are accumulated, the reconstruction of the coupled hydrological data entity is triggered, and a new round of automatic parameter calibration process is initiated. The triggering condition is determined by the sample quantity threshold or the rate of change of the overall model performance index.

[0118] In practice, calculating the new set of residual vectors involves spatiotemporal alignment and contrast. The updated simulation output sequence is interpolated onto spatiotemporal grid points consistent with the multidimensional measured observation sequence, and the difference vector is calculated point by point. The evaluation criteria for the new data samples can be formalized as follows:

[0119]

[0120] in: To represent a new data sample, This indicates the set of parameter configurations that triggered this model run, which is the part of the final parameter calibration instruction sequence that has already been executed. This represents the matrix formed by the updated simulation output sequence. This represents the new set of residual vectors. When a certain number of new data samples are accumulated, this number can be a preset integer N, for example, N=10, indicating that 10 sets of simulation results and residuals under different parameter configurations have been collected. Triggering the reconstruction of the coupled hydrological data entity means, based on the original coupled hydrological data entity, incorporating information from the new data samples, and re-executing the processes of data fusion, perturbation, feature separation, and matching decomposition. Starting a new round of automatic parameter calibration means, starting from the construction of a new coupled hydrological data entity, fully executing all steps of the automatic parameter calibration method, and generating a new set of final parameter calibration instruction sequences.

[0121] In some embodiments, the computation of the new residual vector set may employ a different residual norm than that used when constructing the initial coupled hydrological data entity, for example, using mean absolute error instead of Euclidean distance. It is understood that the storage of new data samples can be achieved by establishing a versioned database that records the complete chain of parameter modifications, model runs, and result comparisons. Optionally, in addition to the sample size threshold, the condition triggering reconstruction may also be that the overall norm mean of the new residual vector set exceeds the historical baseline by a certain proportion, indicating significant fluctuations in model performance. In some embodiments, before adding new data samples to the existing set, quality control can be performed on the updated simulation output sequence to remove abnormal outputs caused by model instability. Optionally, when starting a new round of automatic parameter calibration, some intermediate results obtained from the previous round of calculation can be inherited, such as a multi-mode parameter response template library, to improve computational efficiency. It is understood that the duration of a complete hydrological cycle simulation can be adjusted according to the climatic characteristics of the study area; for example, a longer simulation period may be needed in arid and semi-arid regions to include representative hydrological events.

[0122] See Figure 5 This is a convergence curve of model iteration residuals, measuring the deviation between the simulated and measured values. A smaller value indicates higher model accuracy. The continuous decline of both curves directly verifies the effectiveness of the automatic parameter calibration method. Each iteration of parameter adjustment effectively reduces the simulation error of the surface water and groundwater subsystems. The river flow residual is consistently lower than the groundwater level residual, consistent with hydrological patterns. The surface water system responds faster, and parameter adjustments significantly improve its accuracy. This directly demonstrates that the automatic parameter calibration method can continuously optimize model accuracy, providing core data support for the patented technology. Both curves exhibit an exponential downward trend, indicating fast algorithm convergence and good stability. Clearly distinguishing the differences in accuracy improvement between the surface water and groundwater subsystems provides a basis for subsequent targeted optimization.

[0123] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. An automatic parameter calibration method for a surface water-groundwater coupled model, characterized in that, include: Obtain multi-dimensional measured observation sequences and model simulation output sequences of the surface water subsystem and the groundwater subsystem; By fusing multi-dimensional measured observation sequences and model simulation output sequences, a coupled hydrological data entity containing time-varying response characteristics and spatial distribution characteristics is constructed. Perturbation is applied to coupled hydrological data entities, and unstable response features are identified and separated based on the deviation of the data in the spatial and temporal dimensions after perturbation. Non-steady response features are filtered out from coupled hydrological data entities to form de-disturbed steady-state data entities, and these de-disturbed steady-state data entities are matched with a pre-established multi-mode parameter response template library. Based on the matching results, adaptive structural decomposition is performed on the disturbed steady-state data entities to generate a set of model parameter feature clusters with hierarchical structure; Based on the set of model parameter feature clusters, a dynamic constraint network is constructed between surface water movement parameters and groundwater seepage parameters. A dynamic constraint network is used to simultaneously optimize and cross-validate the feature cluster set of model parameters to generate a preliminary parameter adjustment scheme. Based on the preliminary parameter adjustment plan, key transmission parameter nodes and boundary interaction parameter nodes are located in the model parameter feature cluster set; By combining key transmission parameter nodes and boundary interaction parameter nodes, the path of the preliminary parameter adjustment scheme is optimized to generate the final parameter calibration instruction sequence.

2. The automatic parameter calibration method for a surface water-groundwater coupled model according to claim 1, characterized in that, By fusing multi-dimensional measured observation sequences and model simulation output sequences, a coupled hydrological data entity containing time-varying response characteristics and spatial distribution characteristics is constructed, including: The measured data matrix is ​​constructed by extracting the cross-sectional flow, water level and velocity distribution of the river channel from the measured observation sequence of the surface water subsystem, and extracting the groundwater level, pore water pressure and water quality component concentration at different depths from the measured observation sequence of the groundwater subsystem. Simultaneously, the model simulation data matrix that completely corresponds to the measured data matrix in the spatiotemporal dimension is extracted from the simulation output sequence of the surface water-groundwater coupling model; The measured data matrix and the model simulation data matrix are interpolated and aligned within a unified spatiotemporal grid framework to form a spatiotemporally aligned comparison data field. In the comparative data field, for each spatiotemporal grid point, the residual vector between the measured data and the simulated data is calculated, and the residual vector is bound to the measured data vector and the simulated data vector of the corresponding spatiotemporal grid point; All spatiotemporal grid point data are bundled and spatially clustered according to the boundaries of hydrogeological units and surface catchment areas. A time evolution identifier is added to each spatial cluster unit to form a coupled hydrological data entity.

3. The automatic parameter calibration method for a surface water-groundwater coupled model according to claim 2, characterized in that, Perturbations are applied to coupled hydrological data entities, and unstable response features are identified and separated based on the spatial and temporal deviations of the perturbed data. These features include: For the model simulation data vector in each spatial clustering unit of the coupled hydrological data entity, a set of random noise with increasing intensity is injected, and the change trajectory of the residual vector between the bundled measured data vector and the simulated data vector after noise injection is observed. Record the magnitude and direction of the residual vector at each spatiotemporal grid point under each intensity level to form a disturbance response surface; Analyze the continuity of the disturbance response surface on the time axis. If the disturbance response surface shows abrupt changes or discontinuous jumps in adjacent time slices, the data response corresponding to the time slice is determined to be unstable. Analyze the spatial diffusion range of the disturbance response surface. If the disturbance of a single grid point causes the residual change to remain significant beyond the preset spatial neighborhood, then the data response of the spatial region is determined to be unstable. Extract the original bundled data corresponding to all time segments and spatial regions that are determined to be unstable, mark and separate them from the coupled hydrological data entities, and summarize them to form an unstable response feature set.

4. The automatic parameter calibration method for a surface water-groundwater coupled model according to claim 3, characterized in that, Non-steady-state response features are filtered out from coupled hydrological data entities to form de-disturbed steady-state data entities. These de-disturbed steady-state data entities are then matched with a pre-established multi-model parameter response template library, including: All bundled data belonging to the unstable response feature set are removed from the coupled hydrological data entity, and the remaining part constitutes the de-disturbed steady-state data entity. The multi-mode parameter response template library stores the standard response patterns that surface water and groundwater observation data should exhibit when the combination of key model parameters changes under different typical hydrogeological conditions. Each standard response pattern is defined by a set of feature vectors and weight matrices. Iterate through each spatial clustering unit in the perturbation-free steady-state data entity, and calculate the multidimensional similarity between the residual vector, the measured data vector of the spatial clustering unit and the feature vector of each standard response mode in the multi-mode parameter response template library; Standard response patterns with multidimensional similarity exceeding a set threshold are selected as candidate matching patterns for the spatial clustering unit, and the weight matrix of each candidate matching pattern is recorded. The candidate matching patterns and their weight matrices of all spatial clustering units in the perturbation-free steady-state data entity are integrated according to the time evolution identifier to form a global matching relationship graph.

5. The automatic parameter calibration method for a surface water-groundwater coupled model according to claim 4, characterized in that, Based on the matching results, adaptive structural decomposition is performed on the perturbation-free steady-state data entities to generate a hierarchical set of model parameter feature clusters, including: Based on the global matching relationship graph, identify spatial clustering unit clusters with high weights to the same standard response pattern in the disturbed steady-state data entities; Taking each spatial clustering unit cluster as a unit, and combining its temporal evolution identifier, the model simulation data vector and the measured data vector in all bundled data within the spatial clustering unit cluster are differentially processed to generate the parameter-sensitive difference field of the spatial clustering unit cluster. Multi-scale analysis of parameter-sensitive difference fields is performed to separate long-term trend components, periodic components, and random components on the time scale, and regional background field components and local anomaly field components on the spatial scale. The components at different time and spatial scales are clustered and reorganized according to their correlation with the feature vectors defined in the standard response pattern to form multiple primary feature clusters. Based on the spatiotemporal correlation between primary feature clusters, primary feature clusters with strong correlations are merged, and a hierarchy is assigned to the merged feature clusters. Finally, a set of model parameter feature clusters with a hierarchical structure is generated, in which high-level feature clusters reflect large-scale dominant parameter features, and low-level feature clusters reflect local parameter features.

6. The automatic parameter calibration method for a surface water-groundwater coupled model according to claim 5, characterized in that, Based on the model parameter feature cluster set, a dynamic constraint network is constructed between surface water movement parameters and groundwater seepage parameters, including: The physical meaning of the hydrological process corresponding to each feature cluster in the set of analytical model parameter feature clusters is analyzed, and the feature clusters are mapped to a subset of surface water movement parameters or a subset of groundwater seepage parameters. Identify hybrid feature clusters that simultaneously contain components from subsets of surface water movement parameters and subsets of groundwater seepage parameters, wherein the hybrid feature clusters indicate nodes of coupling between parameters; Using each model parameter as a network node, and the coupling nodes between parameters and the physical connection relationships of hydrological processes as edges, a parameter association graph is initially constructed. Based on the hierarchical structure of the model parameter feature cluster set, dynamic weights are assigned to each edge in the parameter association graph. The dynamic weights are determined by the hierarchy of the feature clusters to which the two connected parameters belong, the magnitude of the components within the feature clusters, and the time evolution trend, thus forming a dynamic constraint network. In a dynamic constraint network, a state variable is defined for each node. The state variable consists of the integrated information of all feature clusters corresponding to the parameters. The edges between nodes define the transmission and constraint rules of the state variable.

7. The automatic parameter calibration method for a surface water-groundwater coupled model according to claim 6, characterized in that, A dynamic constraint network is used to simultaneously optimize and cross-validate the feature cluster set of model parameters, generating a preliminary parameter adjustment scheme, including: Input the set of model parameter feature clusters into the dynamic constraint network to activate the state variables of each node in the network; Based on the connection and weight of edges in the dynamic constraint network, the state variables are iteratively transmitted and negotiated between adjacent nodes. Each node updates the optimal estimate of its own state variables according to the received state information of adjacent nodes and its own feature cluster information. During the network state update process, a consistency verification rule is established: for parameter node pairs connected through a hybrid feature cluster, the trend of their state variables must satisfy the physical coupling relationship defined by the hybrid feature cluster. If the state changes of a parameter node pair do not meet the consistency check rules, a backtracking mechanism is triggered to adjust the contribution weights of the relevant feature clusters and to re-update and iterate the network state until all relevant parameter node pairs pass the consistency check. When the overall state of the dynamic constraint network tends to stabilize, that is, when the change in the state variables of each node is lower than the convergence threshold, the optimal estimate of the final state variables of each node is extracted. The optimal estimates of the state variables for each model parameter node are converted into suggestions for the direction and magnitude of parameter adjustments, and these are summarized to form a preliminary parameter adjustment plan.

8. The automatic parameter calibration method for a surface water-groundwater coupled model according to claim 7, characterized in that, Based on the preliminary parameter adjustment plan, key transmission parameter nodes and boundary interaction parameter nodes are located in the model parameter feature cluster set, including: Analyze the preliminary parameter adjustment scheme, identify model parameters whose adjustment range exceeds the median of the global adjustment range, and mark the model parameters as highly sensitive parameters. In a dynamic constrained network, the nodes corresponding to highly sensitive parameters are located, and the topological properties of these nodes in the network are analyzed. Highly sensitive parameter nodes with high betweenness centrality in the dynamic constraint network are marked as key transmission parameter nodes. Key transmission parameter nodes are located at the intersection of multiple information transmission paths in the network. In the dynamic constraint network, highly sensitive parameter nodes that have strong connection edges with both the surface water movement parameter subset and the groundwater seepage parameter subset nodes are marked as boundary interaction parameter nodes. In the model parameter feature cluster set, locate all feature clusters associated with key transmission parameter nodes and boundary interaction parameter nodes, and extract detailed components of all feature clusters to form spatiotemporal distribution information; The process of combining key transmission parameter nodes and boundary interaction parameter nodes to optimize the path of the initial parameter adjustment scheme and generate the final parameter calibration instruction sequence includes: For key transmission parameter nodes, analyze their main transmission paths in the dynamic constraint network. Based on the adjustment suggestions of upstream and downstream parameters along the transmission path, coordinate and optimize the adjustment range of key transmission parameter nodes to avoid excessive amplification or cancellation of parameter adjustments along the transmission path. For boundary interaction parameter nodes, check whether the adjustment suggestions for the surface water movement parameters and groundwater seepage parameters connected to them meet the boundary conditions of mass conservation or energy conservation. If not, perform joint fine-tuning of the adjustment suggestions for the boundary interaction parameter nodes and their connected parameters until the preset boundary constraint conditions are met. The proposed adjustments to all model parameters after coordination optimization and joint fine-tuning are sorted in the following order: priority is given to key transmission parameter nodes, followed by boundary interaction parameter nodes, and then the remaining parameters. For each sorted parameter adjustment suggestion, add the feature cluster identifier of its source, the reason for the optimization, and the expected adjustment effective time window, and combine them to generate the final parameter calibration instruction sequence.

9. The automatic parameter calibration method for a surface water-groundwater coupled model according to claim 8, characterized in that, Also includes: The final parameter calibration command sequence is injected into the surface water-groundwater coupled model to drive the model parameter update, and the feedback data stream generated by the updated model is captured and fed back to the acquisition process of multi-dimensional measured observation sequences and model simulation output sequences, including: Modify the corresponding parameter values ​​of the surface water-groundwater coupling model in sequence according to the order, amplitude, and time window specified in the final parameter calibration instruction sequence. After each parameter modification, the model is run to simulate a complete hydrological cycle, and the updated simulation output sequence of the model is collected. The updated simulation output sequence is compared with the multi-dimensional measured observation sequence for the corresponding time period, and a new set of residual vectors is calculated. The new set of residual vectors, the updated simulation output sequence and their corresponding parameter configurations are added as a new data sample to the previously obtained set of multi-dimensional measured observation sequences and model simulation output sequences. When a certain number of new data samples are accumulated, the reconstruction of the coupled hydrological data entities is triggered, and a new round of automatic parameter calibration process is initiated.

10. An automatic parameter calibration system for a surface water-groundwater coupled model, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the automatic parameter calibration method for a surface water-groundwater coupling model as described in any one of claims 1 to 9.

Images