A hydrological model parameter recommendation method based on literature mining and watershed similarity matching

Abstract

The application discloses a hydrological model parameter recommendation method based on literature mining and basin similarity matching, and the method comprises the following steps: step 1, multi-source data acquisition and preprocessing; step 2, construction of a literature basin hydrological model parameter knowledge base; step 3, similarity calculation between a target basin and a literature research basin; and step 4, target basin hydrological model parameter recommendation. The application converts fragmented parameter knowledge into structured and reusable knowledge resources by mining existing parameter calibration results in massive literature, solves the problem of parameter acquisition difficulty in basins without data, realizes intelligent parameter recommendation based on basin similarity matching, avoids repeated calibration work, significantly improves the efficiency and accuracy of parameter regionalization, and provides transferable and verifiable parameter support for hydrological modeling.

Description

Technical Field

[0001] This invention belongs to the field of big data mining and hydrological simulation technology, and in particular relates to a method for recommending hydrological model parameters based on literature mining and watershed similarity matching. Background Technology

[0002] Hydrological models are the core tool for simulating watershed hydrological processes, and their simulation accuracy and forecasting capability largely depend on the reliability of model parameters. Hydrological model parameter calibration relies on long-term hydrological and meteorological observation data. However, due to constraints such as uneven spatial distribution and limited coverage of observation stations, practical modeling work often faces the challenge of data scarcity.

[0003] Over decades of development, hydrological research has accumulated a vast number of parameter calibration cases globally, covering diverse hydro-climatic zones and model structures, widely distributed across a wealth of academic literature. The parameter values, watershed attributes, and model performance information contained within this literature far exceed the scope of any single research team, constituting an invaluable knowledge resource. However, due to the lack of efficient knowledge extraction and integration methods, this valuable parameter knowledge remains highly fragmented and has not yet been integrated into a computable and reusable structured knowledge system. This leads to two problems: first, the rich parameter information in the literature cannot effectively support parameter regionalization practices; second, different research teams conduct duplicate studies on similar hydrological zones due to information asymmetry, resulting in a waste of research resources to some extent. If we can systematically mine and effectively utilize the existing parameter knowledge in the literature and establish an intelligent reuse mechanism for parameter knowledge, it is expected to significantly improve the accuracy and efficiency of parameter regionalization in data-scarce areas, while simultaneously promoting the optimal allocation of hydrological research resources. Therefore, constructing a parameter knowledge discovery and intelligent reuse system based on big data mining of literature, and forming an intelligent recommendation mechanism for parameters of similar watersheds, is an important technical path to solve the problem of watershed modeling without data.

[0004] Against this backdrop, there is an urgent need in this field for a method to recommend hydrological model parameters that can systematically mine, structure, store, and intelligently reuse parameter knowledge, providing reliable parameter support for hydrological simulation in data-free watersheds. Summary of the Invention

[0005] The purpose of this invention is to provide a method for recommending hydrological model parameters based on literature mining and watershed similarity matching, so as to solve the above-mentioned technical problems.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] This invention discloses a method for recommending hydrological model parameters based on literature mining and watershed similarity matching. The method includes the following steps:

[0008] Step 1: Multi-source data acquisition and preprocessing: Systematically retrieve hydrological model research literature from academic databases, institutional open platforms, and hydrological science community channels; for the acquired literature, identify the macro structure of the literature based on a deep learning document layout analysis model, and combine natural language processing technology to perform functional classification of the main text content based on semantic features, extracting the research area, research methods, results, and discussion sections; at the same time, use table detection and structure recognition algorithms to extract parameter table information from the literature;

[0009] Step 2, Construction of the Knowledge Base of Hydrological Model Parameters in the Literature: First, the characteristics of the research watershed are extracted, that is, the named entity recognition model is used to automatically identify and extract the characteristic information of the research watershed from the literature, including geographical location characteristics, climate characteristics, topographic and geomorphological characteristics, soil texture characteristics, and land use characteristics.

[0010] By comprehensively considering geographical location, climate, topography, soil texture, and land use characteristics, a watershed feature vector F is established for the study watershed:

[0011] (1)

[0012] In the formula, α represents geographical location characteristics; β represents climate characteristics; η represents topographic features; δ represents soil texture characteristics; and μ represents land use characteristics.

[0013] Secondly, hydrological model parameters and simulation performance indicators are extracted. Specifically, the named entity recognition model is used to extract the values ​​of hydrological model parameters used in the research watershed from the literature, as well as the performance indicators for model calibration and validation.

[0014] Finally, the watershed characteristic information, hydrological model parameters and simulation performance index information extracted from the literature will be stored in a structured manner, and a four-tuple knowledge structure relationship will be established, namely <watershed name, watershed characteristics, model parameter values, model performance index>, to construct a knowledge base of watershed hydrological model parameters from the literature.

[0015] Step 3: Calculate the similarity between the target watershed and the watershed studied in the literature: For the target watershed W to be recommended parameters 目标 Similarly, a feature vector F is formed. Using a multi-dimensional weighted similarity measure combined with a climate zoning constraint method, the similarity between the target watershed and the watersheds in the parameter knowledge base, i.e., the candidate watersheds, is calculated. The specific steps include:

[0016] (1) Feature normalization: Min-Max standardization is performed on various indicators of watershed climate characteristics, topographic features, soil texture characteristics and land use characteristics to eliminate the influence of dimensions;

[0017] (2) Multidimensional weighted similarity calculation: The target watershed W is defined based on weighted Euclidean distance. 目标and the i-th candidate watershed W i Similarity S i , 0≤S i ≤1, S i The larger the value, the higher the similarity; S i Defined as:

[0018] (3)

[0019] In the formula, C1, C2, C3, and C4 are weighting coefficients; , , , W respectively 目标 and W i The Euclidean distance between climate characteristics, topographic features, soil texture characteristics, and land use characteristics, i.e.:

[0020] (4)

[0021] (5)

[0022] (6)

[0023] (7)

[0024] In the formula, ω, σ, ψ, represents the weight coefficients for each feature sub-dimension; j is the summation index variable;

[0025] (3) Final watershed similarity calculation: The Köppen climate zone to which the watershed belongs is determined based on the geographical location feature vector α of the watershed. When the target watershed and the candidate watershed are in different Köppen climate zones, a penalty mechanism is introduced to correct S. i The final watershed similarity is calculated as follows:

[0026] (8)

[0027] In the formula, S final,i For the target watershed W 目标 With the i-th candidate watershed W i The final watershed similarity; λ is the penalty coefficient; I i For climate zone indicator functions, when W 目标 With W i When in different climate zones I i =1, if they are the same, then I i =0;

[0028] Step 4: Recommendation of hydrological model parameters for the target watershed: For candidate watersheds, according to S... final,i>Set a threshold and filter models whose accuracy meets the set criteria, then sort the filtered results by S. final,i Arrange the candidate watersheds in descending order and select the top K watersheds as the reference watersheds for the target watershed parameter recommendation;

[0029] For the hydrological model parameters to be recommended in the target watershed, the recommended values ​​are obtained by weighted calculation based on the results of the reference watershed parameters:

[0030] (9)

[0031] In the formula, Recommended values ​​for the parameters; This represents the parameter value of the i-th reference watershed. The weight coefficient for the i-th reference watershed is determined by the similarity between the target watershed and the i-th reference watershed, the model performance index of the reference watershed, and the literature quality factor.

[0032] (10)

[0033] In the formula, S final,i Let NSE be the similarity between the i-th reference watershed and the target watershed; i REQ i The Nash efficiency coefficient and relative error of peak flow in the simulation of the i-th reference watershed parameters; φ i f is the journal quality factor of the i-th reference watershed from the literature; The functional relationship between the indicators satisfies the condition of S. final,i NSE i and φ i Positively correlated with REQ i Negative correlation;

[0034] Based on the parameter results from K reference watersheds, the 5% and 95% quantiles, namely P5 and P6, were calculated respectively. 95 This forms the parameter recommendation interval [P] 5， P 95 Meanwhile, the uncertainty index U is defined as:

[0035] (11)

[0036] Recommended values ​​based on U-value evaluation parameters Confidence level:

[0037] (12)

[0038] In the formula, ξ1 and ξ2 are the confidence level classification thresholds, and ξ1 < ξ2;

[0039] Finally, the recommended parameter values ​​for the target watershed are provided. With confidence level.

[0040] Furthermore, the academic databases mentioned in step 1 include Web of Science and CNKI; the institutional open platforms include the U.S. Geological Survey (USGS) and water resources departments of various countries; and the hydrological science communities include CUAHSI and HydroShare.

[0041] Furthermore, the system retrieval of hydrological model research literature mentioned in step 1 specifically involves using an API interface or web crawler technology to perform a search using Boolean logic combined keyword search.

[0042] Furthermore, the geographical location features mentioned in step 2 include the basin name, the country and river system, and latitude and longitude coordinates; the climate features include climate zones, average annual rainfall, average annual temperature, seasonal rainfall characteristics, and humidity; the topographic features include basin area, average slope, river network density, main channel gradient, and basin shape coefficient; the soil texture features include the proportion of different soil texture types, soil permeability coefficient, and soil thickness; and the land use features include the proportion of different land use types, including forest coverage, impervious surface ratio, and arable land ratio.

[0043] Furthermore, the formula for Min-Max standardization of the various indicators of watershed climate characteristics, topographic features, soil texture characteristics, and land use characteristics mentioned in step 3 is as follows:

[0044] (2)

[0045] In the formula, x represents the standardized feature values; x represents the original feature values. and These are the minimum and maximum values ​​within the range of values ​​that the feature itself can take.

[0046] Furthermore, the S mentioned in step 4 final,i >Setting a threshold and ensuring the model accuracy meets the set standard specifically means: S final,i >0.75 and during the model validation period, NSE ≥ 0.6 and REQ ≤ 30%.

[0047] The beneficial effects of this invention are as follows: By mining existing parameter calibration results in a large amount of literature, this invention transforms fragmented parameter knowledge into structured and reusable knowledge resources, solving the problem of difficulty in obtaining watershed parameters without data; at the same time, it realizes intelligent parameter recommendation based on watershed similarity matching, avoiding repetitive calibration work, significantly improving the efficiency and accuracy of parameter regionalization, and providing transferable and verifiable parameter support for hydrological modeling.

[0048] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments. Attached Figure Description

[0049] Figure 1 This is a schematic diagram of the method flow described in this invention. Detailed Implementation

[0050] This invention discloses a method for recommending hydrological model parameters based on literature mining and watershed similarity matching, such as... Figure 1 As shown, the method includes the following steps:

[0051] Step 1: Multi-source data acquisition and preprocessing:

[0052] By using API interfaces or web crawling technology, we can systematically search for hydrological model research literature from academic databases (including Web of Science, CNKI, etc.), institutional open platforms (including the USGS, water resources departments of various countries, etc.), and hydrological science communities (including CUAHSI, HydroShare, etc.) using Boolean logic combined keyword search (such as "<model name> AND <parameter calibration>").

[0053] For the acquired academic literature, a deep learning document layout analysis model (such as LayoutLM) is used to identify the macro-structure of the literature (title, abstract, main text, figures, etc.). Combined with natural language processing technology, the main text content is functionally classified based on semantic features to accurately extract core sections such as research area, research methods, results, and discussion. Simultaneously, table detection and structure recognition algorithms are used to assist in extracting parametric table information from the literature.

[0054] Step 2: Construction of a knowledge base for watershed hydrological model parameters from literature:

[0055] For the massive amount of academic literature obtained in step 1, the first step is to extract the characteristic information of the research watershed. Using a Named Entity Recognition (NER) model, the characteristic information of the research watershed is automatically identified and extracted from the research area description sections of the literature, including:

[0056] 1) Geographical location characteristics This includes the name of the river basin, the country and river system, and latitude and longitude coordinates;

[0057] 2) Climate characteristics This includes climate zones, average annual rainfall, average annual temperature, seasonal characteristics of rainfall, and humidity levels.

[0058] 3) Topographic and geomorphological features This includes factors such as drainage area, average slope, river network density, main channel gradient, and drainage shape coefficient.

[0059] 4) Soil texture characteristics This includes the proportion of different soil texture types, soil permeability coefficient, soil thickness, etc.

[0060] 5) Land use characteristics This includes the proportion of different land use types, such as forest coverage, impervious surface ratio, and arable land ratio.

[0061] By comprehensively considering geographical location, climate, topography, soil texture, and land use characteristics, a watershed feature vector F is established for the study watershed:

[0062] (1)

[0063] Secondly, hydrological model parameters and simulation performance indicators were extracted. The NER model was used to extract the hydrological model parameter values ​​for the research watershed, as well as the performance indicators for model calibration and validation (Nash efficiency coefficient NSE, relative error of peak flow REQ, etc.) from the research methods, results, and discussion sections of the literature.

[0064] Finally, the watershed hydrological model parameter information extracted from the literature, such as watershed characteristics, hydrological model parameters, and simulation performance indicators, will be stored in a structured manner. A four-tuple knowledge structure relationship will be established, namely <watershed name, watershed characteristics, model parameter values, model performance indicators>, to construct a literature watershed hydrological model parameter knowledge base (hereinafter referred to as the parameter knowledge base).

[0065] Step 3: Calculate the similarity between the target watershed and the watershed studied in the literature:

[0066] For the target watershed W of the parameters to be recommended 目标 Similarly, a feature vector F is formed. Using a multidimensional weighted similarity measure combined with a climate zoning constraint method, the similarity between the target watershed and the watersheds in the parameter knowledge base, i.e., the candidate watersheds, is calculated. This includes the following steps:

[0067] (1) Feature normalization:

[0068] Min-Max standardization was performed on various indicators of watershed climate characteristics, topographic features, soil texture characteristics, and land use characteristics to eliminate the influence of dimensions.

[0069] (2)

[0070] In the formula, x represents the standardized feature values; x represents the original feature values. and These are the minimum and maximum values ​​within the range of values ​​that the feature itself can take.

[0071] (2) Multidimensional weighted similarity calculation:

[0072] The target watershed W is defined based on weighted Euclidean distance. 目标 and the i-th candidate watershed W i Similarity S i , 0≤S i ≤1, S i The larger the value, the higher the similarity. S i Defined as:

[0073] (3)

[0074] In the formula, C1, C2, C3, and C4 are weighting coefficients, determined based on expert experience; , , , W respectively 目标 and W i The Euclidean distance between climate characteristics, topographic features, soil texture characteristics, and land use characteristics, i.e.:

[0075] (4)

[0076] (5)

[0077] (6)

[0078] (7)

[0079] In the formula, ω, σ, ψ, is the weight coefficient for each feature sub-dimension, determined based on expert experience; j is the summation index variable, indicating that the j-th sub-dimension under each feature category is summed during the calculation of Euclidean distance.

[0080] (3) Final watershed similarity calculation:

[0081] The Köppen climate zone to which a watershed belongs can be determined based on the geographical location feature vector α. When the target watershed and the candidate watershed are located in different Köppen climate zones, a penalty mechanism is introduced to correct S. i The final watershed similarity is calculated as follows:

[0082] (8)

[0083] In the formula, S final,i For the target watershed W 目标 With the i-th candidate watershed W i The final watershed similarity; λ is the penalty coefficient; I i For climate zone indicator functions, when W 目标 With W iWhen in different climate zones I i =1, if they are the same, then I i =0.

[0084] Step 4: Recommended parameters for the target watershed hydrological model:

[0085] For candidate watersheds, according to S final,i >Set a threshold (e.g., 0.75) and filter models based on their accuracy meeting set criteria (e.g., NSE ≥ 0.6 and REQ ≤ 30% during validation period), and then sort the filtered results by S. final,i Arrange the candidate watersheds in descending order and select the top K watersheds as the reference watersheds for the target watershed parameter recommendation.

[0086] For the hydrological model parameters to be recommended in the target watershed, the recommended values ​​are obtained by weighted calculation based on the results of the reference watershed parameters:

[0087] (9)

[0088] In the formula, Recommended values ​​for the parameters; This represents the parameter value of the i-th reference watershed. The weight coefficient for the i-th reference watershed is determined by the similarity between the target watershed and the i-th reference watershed, the model performance index of the reference watershed, and the literature quality factor.

[0089] (10)

[0090] In the formula, S final,i Let NSE be the similarity between the i-th reference watershed and the target watershed; i REQ i The Nash efficiency coefficient and relative error of peak flow in the simulation of the i-th reference watershed parameters; φ i f is the journal quality factor of the literature from the i-th reference watershed (which can be determined based on journal impact factor, citation frequency, etc.); f is The functional relationship between the indicators satisfies the condition of S. final,i NSE i and φ i Positively correlated with REQ i It is negatively correlated.

[0091] In addition, the 5% and 95% quantiles (i.e., P5 and P6) were calculated based on the parameter results of the K reference watersheds. 95 ), forming the parameter recommendation interval [P] 5， P 95 Meanwhile, the uncertainty index U is defined as:

[0092] (11)

[0093] Recommended values ​​based on U-value evaluation parameters Confidence level:

[0094] (12)

[0095] In the formula, ξ1 and ξ2 are the confidence level classification thresholds, and ξ1 < ξ2.

[0096] Finally, the recommended parameter values ​​for the target watershed are provided. With confidence level, we can fully understand the uncertainty of recommendation results and assist in decision-making and parameter selection.

[0097] Finally, it should be noted that the above description is only used to illustrate the technical solution of the present invention and not to limit it. Although the present invention has been described in detail with reference to the preferred arrangement, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solution of the present invention without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. A method for recommending hydrological model parameters based on literature mining and watershed similarity matching, characterized in that, The method includes the following steps: Step 1: Multi-source data acquisition and preprocessing: Systematically retrieve hydrological model research literature from academic databases, institutional open platforms, and hydrological science community channels; for the acquired literature, identify the macro structure of the literature based on a deep learning document layout analysis model, and combine natural language processing technology to perform functional classification of the main text content based on semantic features, extracting the research area, research methods, results, and discussion sections; at the same time, use table detection and structure recognition algorithms to extract parameter table information from the literature; Step 2, Construction of the Knowledge Base of Hydrological Model Parameters in the Literature: First, the characteristics of the research watershed are extracted, that is, the named entity recognition model is used to automatically identify and extract the characteristic information of the research watershed from the literature, including geographical location characteristics, climate characteristics, topographic and geomorphological characteristics, soil texture characteristics, and land use characteristics. By comprehensively considering geographical location, climate, topography, soil texture, and land use characteristics, a watershed feature vector F is established for the study watershed: (1) In the formula, α represents geographical location characteristics; β represents climate characteristics; η represents topographic features; δ represents soil texture characteristics; and μ represents land use characteristics. Secondly, hydrological model parameters and simulation performance indicators are extracted. Specifically, the named entity recognition model is used to extract the values ​​of hydrological model parameters used in the research watershed from the literature, as well as the performance indicators for model calibration and validation. Finally, the watershed characteristic information, hydrological model parameters and simulation performance index information extracted from the literature will be stored in a structured manner, and a four-tuple knowledge structure relationship will be established, namely <watershed name, watershed characteristics, model parameter values, model performance index>, to construct a knowledge base of watershed hydrological model parameters from the literature. Step 3: Calculate the similarity between the target watershed and the watershed studied in the literature: For the target watershed W to be recommended parameters 目标 Similarly, a feature vector F is formed. Using a multi-dimensional weighted similarity measure combined with a climate zoning constraint method, the similarity between the target watershed and the watersheds in the parameter knowledge base, i.e., the candidate watersheds, is calculated. The specific steps include: (1) Feature normalization: Min-Max standardization is performed on various indicators of watershed climate characteristics, topographic features, soil texture characteristics and land use characteristics to eliminate the influence of dimensions; (2) Multidimensional weighted similarity calculation: The target watershed W is defined based on weighted Euclidean distance. 目标 and the i-th candidate watershed W i Similarity S i , 0≤S i ≤1, S i The larger the value, the higher the similarity; S i Defined as: (3) In the formula, C1, C2, C3, and C4 are weighting coefficients; , , , W respectively 目标 and W i The Euclidean distance between climate characteristics, topographic features, soil texture characteristics, and land use characteristics, i.e.: (4) (5) (6) (7) In the formula, ω, σ, ψ, represents the weight coefficients for each feature sub-dimension; j is the summation index variable; (3) Final watershed similarity calculation: The Köppen climate zone to which the watershed belongs is determined based on the geographical location feature vector α of the watershed. When the target watershed and the candidate watershed are in different Köppen climate zones, a penalty mechanism is introduced to correct S. i The final watershed similarity is calculated as follows: (8) In the formula, S final,i For the target watershed W 目标 With the i-th candidate watershed W i The final watershed similarity; λ is the penalty coefficient; I i For climate zone indicator functions, when W 目标 With W i When in different climate zones I i =1, if they are the same, then I i =0; Step 4: Recommendation of hydrological model parameters for the target watershed: For candidate watersheds, according to S... final,i >Set a threshold and filter models whose accuracy meets the set criteria, then sort the filtered results by S. final,i Arrange the candidate watersheds in descending order and select the top K watersheds as the reference watersheds for the target watershed parameter recommendation; For the hydrological model parameters to be recommended in the target watershed, the recommended values ​​are obtained by weighted calculation based on the results of the reference watershed parameters: (9) In the formula, Recommended values ​​for the parameters; This represents the parameter value of the i-th reference watershed. The weight coefficient for the i-th reference watershed is determined by the similarity between the target watershed and the i-th reference watershed, the model performance index of the reference watershed, and the literature quality factor. (10) In the formula, S final,i Let NSE be the similarity between the i-th reference watershed and the target watershed; i REQ i The Nash efficiency coefficient and relative error of peak flow in the simulation of the i-th reference watershed parameters; φ i f is the journal quality factor of the literature from the i-th reference watershed; The functional relationship between the indicators satisfies the condition of S. final,i NSE i and φ i Positively correlated with REQ i Negative correlation; Based on the parameter results from K reference watersheds, the 5% and 95% quantiles, namely P5 and P6, were calculated respectively. 95 This forms the parameter recommendation interval [P] 5， P 95 Meanwhile, the uncertainty index U is defined as: (11) Recommended values ​​based on U-value evaluation parameters Confidence level: (12) In the formula, ξ1 and ξ2 are the confidence level classification thresholds, and ξ1 < ξ2; Finally, the recommended parameter values ​​for the target watershed are provided. With confidence level.

2. The method for recommending hydrological model parameters based on literature mining and watershed similarity matching according to claim 1, characterized in that, The academic databases mentioned in Step 1 include Web of Science and CNKI; the institutional open platforms include the U.S. Geological Survey (USGS) and water resources departments of various countries; and the hydrological science communities include CUAHSI and HydroShare.

3. The method for recommending hydrological model parameters based on literature mining and watershed similarity matching according to claim 1, characterized in that, The system retrieval of hydrological model research literature in step 1 specifically involves using an API interface or web crawler technology to perform a search using Boolean logic combined keyword search.

4. The method for recommending hydrological model parameters based on literature mining and watershed similarity matching according to claim 1, characterized in that, The geographical location features mentioned in step 2 include the basin name, the country and river system, and latitude and longitude coordinates; the climate features include climate zones, average annual rainfall, average annual temperature, seasonal rainfall characteristics, and humidity; the topographic features include basin area, average slope, river network density, main channel gradient, and basin shape coefficient; the soil texture features include the proportion of different soil texture types, soil permeability coefficient, and soil thickness; and the land use features include the proportion of different land use types, including forest coverage, impervious surface ratio, and arable land ratio.

5. The method for recommending hydrological model parameters based on literature mining and watershed similarity matching according to claim 1, characterized in that, The formula for Min-Max standardization of the various indicators of watershed climate characteristics, topographic features, soil texture characteristics, and land use characteristics mentioned in step 3 is as follows: (2) In the formula, x represents the standardized feature values; x represents the original feature values. and These are the minimum and maximum values ​​within the range of values ​​that the feature itself can take.

6. The method for recommending hydrological model parameters based on literature mining and watershed similarity matching according to claim 1, characterized in that, The S mentioned in step 4 final,i >Setting a threshold and ensuring the model accuracy meets the set standard specifically means: S final,i >0.75 and during the model validation period, NSE ≥ 0.6 and REQ ≤ 30%.

Images