A frozen-thaw phase state constraint and water-blocking layer index-based groundwater level prediction method in cold regions

Abstract

This invention provides a groundwater level prediction method for cold regions based on freeze-thaw phase constraints and a water-blocking layer index. It uses multi-source data from monitoring wells in cold regions to label freeze-thaw stages and constructs a freeze-thaw water-blocking layer index, a melting connectivity coefficient, and a low-temperature capillary migration coefficient. Combining historical groundwater level sequences and external driving sequences, a bi-branch deep network model is used to output groundwater level predictions for multiple future time points, as well as predicted values ​​for at least one of the freeze-thaw water-blocking layer index, melting connectivity coefficient, and low-temperature capillary migration coefficient. The bi-branch deep network model is pre-trained based on monotonic constraints during the freezing period, infiltration lag constraints during the initial thawing period, and melting threshold constraints. In this scheme, freeze-thaw water blocking, melting connectivity, and low-temperature capillary migration are explicitly introduced into the prediction framework, which improves the ability to express the mechanisms in cold regions. By using monotonic constraints during the freezing period, infiltration lag constraints during the initial thawing period, and melting threshold constraints, the prediction accuracy of the freeze-thaw transition stage is improved.

Description

Technical Field

[0001] This invention relates to the field of groundwater monitoring and intelligent computing technology in cold regions, and more specifically, to a method for predicting groundwater levels in cold regions based on freeze-thaw phase constraints and aquitard index. Background Technology

[0002] Groundwater level changes in cold regions are significantly influenced by the permafrost environment. Compared to non-cold regions, groundwater levels in cold regions often exhibit distinct phased characteristics during the freezing-thawing transition, such as a continuous decline during the freezing period, a delayed rebound during the initial thawing period, a sudden increase during the rapid thawing period, and a return to stability after thawing. Most existing general groundwater level prediction methods directly regress or time-series model historical groundwater levels and meteorological factors, which are insufficient to accurately represent the unique control mechanisms of freezing-water blocking, thawing opening, and low-temperature capillary migration in cold regions. If historical groundwater levels and conventional meteorological factors are directly input into deep learning networks, the models typically struggle to explicitly identify freeze-thaw phase changes in cold regions at the feature level, resulting in increased prediction errors during the freezing-thawing transition phase, inaccurate identification of the inflection point during the initial thawing period, and insufficient characterization of the lag relationship between precipitation or snowmelt input and groundwater level response.

[0003] Therefore, it is necessary to provide a prediction scheme that can improve the prediction accuracy and engineering usability of groundwater levels in cold regions during the freezing-thawing transition phase. Summary of the Invention

[0004] The purpose of this invention is to provide a groundwater level prediction method for cold regions based on freeze-thaw phase constraints and aquitard index, to improve the accuracy of groundwater level prediction during the freeze-thaw transition stage.

[0005] In a first aspect, the present invention provides a method for predicting groundwater levels in cold regions based on freeze-thaw phase constraints and aquitard index, the method comprising: Multi-source data from monitoring wells in cold regions were collected, and freeze-thaw stage labels were defined based on the multi-source data. The freezing water barrier index, melting connectivity coefficient, and low-temperature capillary migration coefficient were constructed. The frozen water-blocking layer index, melting connectivity coefficient, low-temperature capillary migration coefficient, historical groundwater level sequence, exogenous driving sequence, and freeze-thaw stage labels are used as input data to construct a bi-branch deep network model. The bi-branch deep network model is pre-trained under the guidance of a constructed loss function, which includes a monotonic constraint during the freezing period, an infiltration lag constraint during the initial thawing period, and a melting threshold constraint. The dual-branch deep network model is used to extract features from the input data. Based on the extracted features, the predicted groundwater level for multiple future time periods is output, along with the predicted values ​​of at least one of the following: the frozen water barrier index, the melting connectivity coefficient, and the low-temperature capillary migration coefficient for multiple future time periods.

[0006] In an optional implementation, the dual-branch deep network model includes a freeze-thaw phase branch and a hydrodynamic branch; The step of extracting features from the input data using the dual-branch deep network model includes: Phase change control features are extracted based on the input data using the freeze-thaw phase branch, and replenishment and discharge features are extracted based on the input data using the hydrodynamic branch; The phase change control features and the replenishment and discharge features are fused to obtain the fused features.

[0007] In an optional implementation, the step of dividing freeze-thaw stage labels based on the multi-source data includes: Freezing degree days and thawing degree days are constructed based on the temperature sequences in the multi-source data; Based on the freezing days, thawing days, temperature below zero, soil temperature profile, and snow cover duration, freeze-thaw stage labels are defined. The freeze-thaw stage labels include three or more of the following: initial freezing stage, stable freezing stage, initial thawing stage, thawing stage, and complete thawing stage.

[0008] In an optional implementation, the freezing water barrier index is constructed in the following manner: The freezing depth, ice content of frozen soil, freezing degree days, snow cover correction amount and unfrozen water ratio are fused according to their corresponding weights and mapped to a set range to construct the frozen water barrier layer index. The frozen water barrier index is used to characterize the ability of the frozen layer to impede the downward infiltration of precipitation and snowmelt.

[0009] In an optional implementation, the fusion connectivity coefficient is constructed in the following manner: The melting depth, geothermal rise rate, local melting connectivity index, and frozen water-blocking layer index are fused according to their corresponding weights and mapped to a set range to construct the melting connectivity coefficient. The connectivity coefficient is used to characterize the degree of connectivity between precipitation or snowmelt and the underground aquifer after the melting channel is opened.

[0010] In an optional implementation, the low-temperature capillary migration coefficient is constructed in the following manner: The soil temperature gradient, soil moisture content, proportion of unfrozen water, and freezing depth are fused according to their corresponding weights and mapped to a set range to construct the low-temperature capillary migration coefficient. The low-temperature capillary migration coefficient is used to characterize the degree of redistribution of moisture along the temperature gradient and capillary action under low-temperature conditions.

[0011] In an optional implementation, the freeze-time monotonic constraint is constructed in the following manner: Determine the time set of the freezing phase, obtain the groundwater level prediction results for future times within the time set of the freezing phase based on the dual-branch deep network model, and obtain the groundwater level prediction result difference based on the groundwater level prediction results; The monotonic constraint for the freezing period is constructed by combining the difference in the predicted groundwater level with the set allowable fluctuation threshold.

[0012] In an optional implementation, the initial melting period infiltration hysteresis constraint is constructed in the following manner: Determine the time set of the initial melting stage and the length of the lag window, and obtain the effective replenishment sequence composed of precipitation and snowmelt signals; Within the initial melting stage time set, the infiltration lag constraint for the initial melting period is constructed based on the difference in the groundwater level prediction results, the lag window length, and the effective recharge sequence.

[0013] In an optional implementation, the facilitation threshold constraint is constructed in the following manner: Set a fusion threshold and obtain the fusion connectivity coefficients at future time points output by the dual-branch deep network model; Based on the fusion threshold, the fusion connectivity coefficient at future time, the difference in the groundwater level prediction results, and the groundwater level reference recovery response, the fusion threshold constraint is constructed.

[0014] In an optional implementation, the method further includes: The dual-branch deep network model is used to output the inflection point of groundwater level rise, the threshold crossing time when the groundwater level exceeds the threshold, and the stage risk level.

[0015] Compared to existing technologies, this invention provides a groundwater level prediction method for cold regions based on freeze-thaw phase constraints and a water-blocking layer index. It uses multi-source data from monitoring wells in cold regions to label freeze-thaw stages and constructs a freeze-thaw water-blocking layer index, a melting connectivity coefficient, and a low-temperature capillary migration coefficient. The freeze-thaw water-blocking layer index, melting connectivity coefficient, low-temperature capillary migration coefficient, historical groundwater level sequences, exogenous driving sequences, and freeze-thaw stage labels are used as input data to construct a bi-branch deep network model. This model is pre-trained under the guidance of a constructed loss function, which includes monotonic constraints for the freezing period, infiltration lag constraints for the initial thawing period, and melting threshold constraints. The bi-branch deep network model is used to extract features from the input data. Based on the extracted features, it outputs groundwater level prediction results for multiple future time periods, as well as predicted values ​​for at least one of the freeze-thaw water-blocking layer index, melting connectivity coefficient, and low-temperature capillary migration coefficient for these multiple future time periods.

[0016] This scheme explicitly incorporates freezing-water blocking, thawing connectivity, and low-temperature capillary migration into the prediction framework, which enhances the ability to express mechanisms in cold regions. Furthermore, by imposing monotonic constraints on the freezing period, infiltration lag constraints on the initial thawing period, and thawing threshold constraints, the prediction accuracy of the freeze-thaw transition stage is improved. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments of the present invention will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart of a groundwater level prediction method for cold regions based on freeze-thaw phase constraints and water-blocking layer index provided in an embodiment of the present invention; Figure 2 A schematic diagram of the overall logic of the cold region groundwater level prediction method based on freeze-thaw phase constraints and water-blocking layer index provided in this embodiment of the invention; Figure 3 for Figure 1 A flowchart of the sub-steps included in S11; Figure 4 for Figure 1 The flowchart of the sub-steps included in S13. Detailed Implementation

[0019] The technical solutions of the present invention will now be described with reference to the accompanying drawings in the embodiments of the present invention.

[0020] Please see Figure 1 This is a flowchart of a groundwater level prediction method for cold regions based on freeze-thaw phase constraints and aquitard index provided in an embodiment of the present invention. The groundwater level prediction method for cold regions specifically includes the following steps: S11, Collect multi-source data from monitoring wells in cold regions, divide the freeze-thaw stage labels based on the multi-source data, and construct the freeze water barrier index, melting connectivity coefficient and low temperature capillary migration coefficient. S12, the frozen water barrier index, melting connectivity coefficient, low temperature capillary migration coefficient, historical groundwater level sequence, external driving sequence and freeze-thaw stage label are used as input data to construct a dual-branch deep network model. The dual-branch deep network model is pre-trained under the guidance of a constructed loss function, which includes a monotonic constraint during the freezing period, a hysteresis constraint during the initial thawing period, and a thawing threshold constraint. S13, the dual-branch deep network model is used to extract features from the input data, and based on the extracted features, the predicted results of groundwater level at multiple future times are output, as well as the predicted values ​​of at least one of the following at multiple future times: frozen water barrier index, melting connectivity coefficient and low temperature capillary migration coefficient.

[0021] To address the shortcomings of existing technologies, this embodiment aims to solve the problems of large prediction errors and inaccurate identification of the inflection point of groundwater level rise during the freeze-thaw transition period in cold regions. It also addresses the difficulty of using general deep learning models to explicitly describe the processes of freezing-induced water blockage, thawing connectivity, and low-temperature capillary migration. Furthermore, it resolves the issue of a significant lag between precipitation or snowmelt input during the initial thaw period and the rise in groundwater level, making learning difficult.

[0022] Based on this, a groundwater level prediction method that combines the ability to express cold-region mechanisms with the predictive capabilities of deep learning is proposed.

[0023] The groundwater level prediction method for cold regions provided in this embodiment explicitly incorporates freezing-induced water blockage, thawing connectivity, and low-temperature capillary migration into the prediction framework, thereby enhancing the ability to express the mechanisms in cold regions. Furthermore, by imposing monotonic constraints during the freezing period, infiltration lag constraints during the initial thawing period, and thawing threshold constraints, the prediction accuracy of the freeze-thaw transition stage is improved.

[0024] The specific implementation methods of each of the above steps will be explained in detail below.

[0025] Please refer to the following: Figure 2 In this embodiment, multi-source data from monitoring wells in cold regions are collected, including but not limited to the following data: (1) Historical groundwater level sequence H(t); (2) Temperature sequence Ta(t); (3) Soil temperature profile Ts(z,t); (4) Precipitation sequence Pr(t); (5) Snow depth or snow water equivalent sequence Sn(t); (6) Soil moisture content sequence θ(t); (7) Evapotranspiration sequence ET(t); (8) Mining volume sequence Qp(t); (9) River and lake water level sequence R(t).

[0026] For the collected multi-source data, preprocessing is first performed, including resampling, missing value imputation, outlier removal, normalization, and uniform time step processing, in order to obtain a unified and standardized dataset.

[0027] The freeze-thaw stage labels are divided based on the preprocessed multi-source data. For details, please refer to [link to relevant documentation]. Figure 3This step can be achieved in the following ways: S111, construct freezing degree days and thawing degree days based on the temperature sequence in the multi-source data; S112, Based on the freezing days, thawing days, temperature crossing the zero point, soil temperature profile, and snow cover duration, the freeze-thaw stage labels are defined; The freeze-thaw stage labels include three or more of the following: initial freezing stage, stable freezing stage, initial thawing stage, thawing stage, and complete thawing stage.

[0028] In this embodiment, based on the collected temperature series, the frozen day is constructed in the following manner. FDD ( t ) and melting away TDD ( t ):

[0029]

[0030] in, ( ) indicates the first The temperature value at that moment.

[0031] Based on this, the freezing water barrier index, melting connectivity coefficient, and low-temperature capillary migration coefficient are constructed by combining multi-source data, freezing time, and thawing time.

[0032] The frozen water-blocking index is constructed in the following way: The freezing depth, ice content of frozen soil, freezing days, snow cover correction amount, and unfrozen water ratio are fused according to their corresponding weights and mapped to a set range to construct the frozen water barrier index.

[0033] The frozen water barrier index is used to characterize the ability of the frozen layer to impede the downward infiltration of precipitation and snowmelt.

[0034] Specifically, the freezing water barrier index is calculated using the following formula:

[0035] Among them: Z f (t) represents the freeze depth; ALT ref For reference active layer thickness; I f(t) S is an indicator of the ice content of permafrost. d(t) For snow cover duration or snow depth correction; U w(t) The proportion of unfrozen water; σ(·) is the Sigmoid mapping function; a 0 toa 5 represents the weight parameter.

[0036] Frozen water barrier index B t The larger the value, the stronger the water-blocking effect of freezing.

[0037] Furthermore, the connectivity coefficients are constructed in the following way: The melting depth, geothermal rise rate, local melting connectivity index, and frozen water-blocking layer index are fused according to their corresponding weights and mapped to a set range to construct the melting connectivity coefficient.

[0038] The connectivity coefficient is used to characterize the degree of connectivity between precipitation or snowmelt and the underground aquifer after the melting channel is opened.

[0039] Specifically, the connectivity coefficient is calculated using the following formula:

[0040] in: Z m ( t () represents the melting depth; G t The rate of geothermal rise; L t This is an indicator of localized melting connectivity. B t The freezing water barrier index; b 0 to b 4 represents the weight parameter.

[0041] Fusion connectivity coefficient C t The larger the value, the smoother the underground supply channels.

[0042] The low-temperature capillary migration coefficient was constructed in the following manner: The soil temperature gradient, soil moisture content, proportion of unfrozen water, and freezing depth are fused according to their corresponding weights and mapped to a set range to construct the low-temperature capillary migration coefficient. The low-temperature capillary migration coefficient is used to characterize the degree of redistribution of moisture along the temperature gradient and capillary action under low-temperature conditions.

[0043] Specifically, the low-temperature capillary migration coefficient is calculated using the following formula:

[0044] Among them: | Ts ( t | represents the absolute value of the soil temperature gradient; θ (t () represents soil moisture content; U w ( t () represents the proportion of unfrozen water; Z f ( t () represents the freezing depth; c 0 to c 4 represents the weight parameter.

[0045] Furthermore, this embodiment constructs a bi-branch deep network model. The frozen water-blocking layer index, melting connectivity coefficient, and low-temperature capillary migration coefficient obtained through the above methods, combined with historical groundwater level sequences, exogenous driving sequences, and freeze-thaw stage labels, are used as input data for the bi-branch deep network model, which can be characterized as follows:

[0046] in: The length of the history window; X e It is an externally driven sequence; S t w:t This is a tag sequence for the freeze-thaw phase. The exogenous driving sequence includes at least three of the following: precipitation, snow depth or snow water equivalent, soil moisture content, evapotranspiration, extraction, and river and lake water levels, to characterize the recharge and discharge processes of the groundwater system.

[0047] In this embodiment, the dual-branch deep network model includes a freeze-thaw phase branch and a hydrodynamic branch. After inputting the above-mentioned input data into the dual-branch deep network model, the model is used to extract features from the input data. For details, please refer to [link to relevant documentation]. Figure 4 This step can be achieved in the following ways: S131, phase change control features are extracted based on the input data using the freeze-thaw phase branch, and replenishment and discharge features are extracted based on the input data using the hydrodynamic branch; S132, the phase change control features and the replenishment and discharge features are fused to obtain the fused features.

[0048] In this embodiment, the freeze-thaw phase branch can extract phase change control features, including features related to freeze-thaw stages, water-blocking states, thawing states, and capillary migration. The hydrodynamic branch can extract recharge and discharge features, including features related to groundwater level changes, precipitation recharge, evapotranspiration, extraction discharge, and disturbance features at river and lake boundaries.

[0049] In addition, the dual-branch deep network model also includes a cross-attention fusion module and a dual-task output module.

[0050] The cross-attention fusion module is used to establish dynamic information interaction between the freeze-thaw phase branch and the hydrodynamic branch. The dual-task output module is used to simultaneously output the groundwater level prediction results for the next n time steps. And the predicted sequence of the frozen water barrier index for the next n time periods. , Connectivity Coefficient Prediction Sequence Low-temperature capillary migration coefficient prediction sequence At least one of them.

[0051] In this embodiment, the dual-branch deep network model is pre-trained under the guidance of a constructed loss function, which includes a monotonic constraint during the freezing period, a hysteresis constraint during the initial thawing period, and a thawing threshold constraint.

[0052] The overall loss function can be characterized as follows:

[0053] in, Indicates predicted losses due to groundwater level. This indicates the predicted loss under intermediate conditions (including the frozen water barrier index, melting connectivity coefficient, and low-temperature capillary migration coefficient). This indicates a monotonic constraint during the freeze period. Indicates the infiltration lag constraint during the initial melting period. This indicates a threshold constraint for facilitation. to This represents the weighting coefficient.

[0054] The predicted loss from groundwater level can be characterized as follows:

[0055] in: ω τ For the future τ The prediction weight at each time point.

[0056] The intermediate state prediction loss is represented as follows:

[0057] Furthermore, the monotonic constraint for the freeze period is constructed in the following way: A time set for the freezing phase is determined, and the groundwater level prediction results for future moments within the time set based on the dual-branch deep network model are obtained. The difference between the groundwater level prediction results is obtained based on the groundwater level prediction results. The monotonic constraint for the freezing period is constructed by combining the difference between the groundwater level prediction results and the set allowable fluctuation threshold.

[0058] The monotonic constraint during the freezing period is used to ensure that the predicted groundwater level sequence during the initial freezing period or the stable freezing period meets at least one of the preset monotonic decreasing, monotonic stable, or phased gradual decreasing patterns. Specifically, the monotonic constraint during the freezing period is characterized as follows:

[0059] Among them, Ω f For the frozen phase time set, ε f The allowable fluctuation threshold.

[0060] The infiltration hysteresis constraint during the initial melting period was constructed in the following way: The initial melting stage time set and lag window length are determined, and an effective recharge sequence consisting of precipitation and snowmelt signals is obtained. Within the initial melting stage time set, based on the groundwater level prediction result difference, lag window length, and effective recharge sequence, the infiltration lag constraint for the initial melting period is constructed.

[0061] The initial melting period infiltration lag constraint is used to constrain the rise of groundwater level after precipitation or snowmelt events from lagging behind a preset time window of effective surface recharge signal. Specifically, the initial melting period infiltration lag constraint is characterized as follows:

[0062] Among them, Ω m This refers to the time frame of the initial fusion phase. U r This is an effective replenishment sequence consisting of precipitation and snowmelt signals. δ The length of the lag window. g (·) represents the response function with a hysteresis window.

[0063] The facilitation threshold constraint is constructed in the following way: Set a convergence threshold and obtain the convergence connectivity coefficient at future time moments output by the dual-branch deep network model; based on the convergence threshold, the convergence connectivity coefficient at future time moments, the difference in groundwater level prediction results, and the groundwater level reference recovery response, construct the convergence threshold constraint.

[0064] The connectivity threshold constraint is used to suppress premature and rapid rise in groundwater level when the connectivity coefficient has not reached the connectivity threshold, and / or to enhance the groundwater level rise response after the connectivity coefficient has reached the connectivity threshold. Specifically, the connectivity threshold constraint is characterized as follows:

[0065] in, η As the threshold for facilitation, Δ H ref For reference, the recovery response amount.

[0066] The above formula represents the... If the aforementioned connectivity threshold is not reached, the groundwater level is prevented from rising too early and rapidly through the above constraints; once the connectivity coefficient reaches the connectivity threshold, the groundwater level rise response is enhanced.

[0067] In this embodiment, the dual-branch deep network model is pre-trained under the guidance of the above-mentioned loss function. The dual-branch deep network model adopts a multi-well joint training method, and different monitoring wells represent the differences in local landforms, lithology or site attributes through well point embedding vectors.

[0068] In practical applications, the aforementioned input data from the target cold-region monitoring wells are fed into a dual-branch deep network model for groundwater prediction in cold regions, and intermediate state variables can be output. Thus, by outputting intermediate state variables through dual tasks, the model's dependence on purely black-box historical data fitting is reduced.

[0069] In this embodiment, new monitoring data can also be received to continuously update the freeze-thaw stage label, the frozen water barrier index, the melting connectivity coefficient, the low-temperature capillary migration coefficient, and the parameters of the dual-branch depth network model, so as to achieve online prediction.

[0070] Furthermore, the groundwater level prediction method for cold regions provided in this embodiment can also utilize a dual-branch deep network model to output the inflection point of groundwater level rise, the threshold crossing time when the groundwater level exceeds the threshold, and the stage risk level. This solution is particularly suitable for scenarios where groundwater level rise in cold regions is lagging and difficult to learn, and inflection point identification is inaccurate.

[0071] In summary, the groundwater level prediction method for cold regions based on freeze-thaw phase constraints and water-blocking layer index provided by this invention obtains historical groundwater level sequences, air temperature sequences, soil temperature profile sequences, precipitation sequences, snow depth or snow water equivalent sequences, soil moisture content sequences, evapotranspiration sequences, extraction volume sequences, and river and lake water level sequences from monitoring wells in cold regions. It constructs freeze-thaw stage labels based on the zero-crossing point of air temperature, freezing degree-days, thawing degree-days, soil temperature profiles, and snow cover duration. Furthermore, it uses freezing depth, frozen soil ice content, unfrozen water content, thawing depth, ground temperature recovery rate, and soil temperature gradient... Information such as temperature is used to construct the frozen water barrier index, melting connectivity coefficient, and low-temperature capillary migration coefficient. Historical groundwater level sequences, external driving sequences, freeze-thaw stage labels, and the frozen water barrier index, melting connectivity coefficient, and low-temperature capillary migration coefficient are input into a dual-branch deep network model. Features are extracted using the freeze-thaw phase branch and the hydrodynamic branch, and fused through a cross-attention module. The model jointly outputs the groundwater level prediction results for multiple future time points and the prediction results of the intermediate state variables. At the same time, the network is trained by the monotonic constraint of the freezing period, the infiltration lag constraint of the initial thawing period, and the melting threshold constraint.

[0072] This solution addresses the challenges of complex response mechanisms, distinct stage transitions, and the difficulty of accurately characterizing the nonlinear changes during the freeze-thaw transition period in cold regions due to the combined effects of freezing and water blockage, delayed infiltration during initial thawing, thawing recovery, and low-temperature capillary migration. It improves the accuracy of groundwater level prediction during the freeze-thaw transition stage, enhances the ability to identify the inflection point of groundwater level rise during the initial thawing period, improves the ability to characterize delayed responses, and enhances the authenticity and engineering application value of groundwater level prediction results in cold regions.

[0073] In the embodiments provided by this invention, it should be understood that the disclosed apparatus and method can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the displayed or discussed mutual couplings, direct couplings, or communication connections may be through some communication interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0074] Furthermore, the units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0075] Furthermore, the functional modules in the various embodiments of the present invention can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0076] It should be noted that if the functionality is implemented as a software module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0077] The above description is merely an embodiment of the present invention and is not intended to limit the scope of protection of the present invention. For those skilled in the art, the present invention can have various modifications and variations. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for predicting groundwater levels in cold regions based on freeze-thaw phase constraints and aquitard index, characterized in that, The method includes: Multi-source data from monitoring wells in cold regions were collected, and freeze-thaw stage labels were defined based on the multi-source data. The freezing water barrier index, melting connectivity coefficient, and low-temperature capillary migration coefficient were constructed. The frozen water-blocking layer index, melting connectivity coefficient, low-temperature capillary migration coefficient, historical groundwater level sequence, exogenous driving sequence, and freeze-thaw stage labels are used as input data to construct a bi-branch deep network model. The bi-branch deep network model is pre-trained under the guidance of a constructed loss function, which includes a monotonic constraint during the freezing period, an infiltration lag constraint during the initial thawing period, and a melting threshold constraint. The dual-branch deep network model is used to extract features from the input data. Based on the extracted features, the predicted groundwater level for multiple future time periods is output, along with the predicted values ​​of at least one of the following: the frozen water barrier index, the melting connectivity coefficient, and the low-temperature capillary migration coefficient for multiple future time periods.

2. The method for predicting groundwater levels in cold regions based on freeze-thaw phase constraints and water-blocking layer index as described in claim 1, characterized in that, The dual-branch deep network model includes a freeze-thaw phase branch and a hydrodynamic branch; The step of extracting features from the input data using the dual-branch deep network model includes: Phase change control features are extracted based on the input data using the freeze-thaw phase branch, and replenishment and discharge features are extracted based on the input data using the hydrodynamic branch; The phase change control features and the replenishment and discharge features are fused to obtain the fused features.

3. The method for predicting groundwater levels in cold regions based on freeze-thaw phase constraints and aquitard index as described in claim 1, characterized in that, The step of dividing freeze-thaw stage labels based on the multi-source data includes: Freezing degree days and thawing degree days are constructed based on the temperature sequences in the multi-source data; Based on the freezing days, thawing days, temperature below zero, soil temperature profile, and snow cover duration, freeze-thaw stage labels are defined. The freeze-thaw stage labels include three or more of the following: initial freezing stage, stable freezing stage, initial thawing stage, thawing stage, and complete thawing stage.

4. The method for predicting groundwater levels in cold regions based on freeze-thaw phase constraints and water-blocking layer index according to claim 3, characterized in that, The frozen water-blocking layer index is constructed in the following way: The freezing depth, ice content of frozen soil, freezing degree days, snow cover correction amount and unfrozen water ratio are fused according to their corresponding weights and mapped to a set range to construct the frozen water barrier layer index. The frozen water barrier index is used to characterize the ability of the frozen layer to impede the downward infiltration of precipitation and snowmelt.

5. The method for predicting groundwater levels in cold regions based on freeze-thaw phase constraints and water-blocking layer index according to claim 3, characterized in that, The fusion connectivity coefficient is constructed in the following way: The melting depth, geothermal rise rate, local melting connectivity index, and frozen water-blocking layer index are fused according to their corresponding weights and mapped to a set range to construct the melting connectivity coefficient. The connectivity coefficient is used to characterize the degree of connectivity between precipitation or snowmelt and the underground aquifer after the melting channel is opened.

6. The method for predicting groundwater levels in cold regions based on freeze-thaw phase constraints and water-blocking layer index according to claim 3, characterized in that, The low-temperature capillary migration coefficient was constructed in the following manner: The soil temperature gradient, soil moisture content, proportion of unfrozen water, and freezing depth are fused according to their corresponding weights and mapped to a set range to construct the low-temperature capillary migration coefficient. The low-temperature capillary migration coefficient is used to characterize the degree of redistribution of moisture along the temperature gradient and capillary action under low-temperature conditions.

7. The method for predicting groundwater levels in cold regions based on freeze-thaw phase constraints and aquitard index according to claim 3, characterized in that, The monotonic constraint during the freeze period is constructed in the following way: Determine the time set of the freezing phase, obtain the groundwater level prediction results for future times within the time set of the freezing phase based on the dual-branch deep network model, and obtain the groundwater level prediction result difference based on the groundwater level prediction results; The monotonic constraint for the freezing period is constructed by combining the difference in the predicted groundwater level with the set allowable fluctuation threshold.

8. The method for predicting groundwater levels in cold regions based on freeze-thaw phase constraints and aquitard index according to claim 7, characterized in that, The initial melting period infiltration hysteresis constraint is constructed in the following way: Determine the time set of the initial melting stage and the length of the lag window, and obtain the effective replenishment sequence composed of precipitation and snowmelt signals; Within the initial melting stage time set, the infiltration lag constraint for the initial melting period is constructed based on the difference in the groundwater level prediction results, the lag window length, and the effective recharge sequence.

9. The method for predicting groundwater levels in cold regions based on freeze-thaw phase constraints and aquitard index according to claim 7, characterized in that, The facilitation threshold constraint is constructed in the following way: Set a fusion threshold and obtain the fusion connectivity coefficients at future moments output by the dual-branch deep network model; Based on the fusion threshold, the fusion connectivity coefficient at future time, the difference in the groundwater level prediction results, and the groundwater level reference recovery response, the fusion threshold constraint is constructed.

10. The method for predicting groundwater levels in cold regions based on freeze-thaw phase constraints and aquitard index according to claim 9, characterized in that, The method further includes: The dual-branch deep network model is used to output the inflection point of groundwater level rise, the threshold crossing time when the groundwater level exceeds the threshold, and the stage risk level.

Images