Aquifer vulnerability assessment method based on interpretable multi-integration strategy optimization framework
Through an interpretable multi-integration strategy optimization framework, the problem of low data quality and model interpretability in aquifer vulnerability assessment is solved, the accuracy and credibility of the assessment is improved, and the scientific and scientific scale is provided for large regional scales. Systematic evaluation method.
Patent Information
- Application Number
- CN202410990236.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-23
- Publication Date
- 2024-11-12
AI Technical Summary
The prior art has problems in the assessment of aquifer vulnerability, low data quality, low method standardization, low model interpretability and lack of a unified evaluation framework, which affects the credibility and stability of the evaluation results.
Using an interpretable multi-integration strategy optimization framework, by collecting and preprocessing feature data, calculating feature importance and bidirectional random effect model to calculate the correlation coefficient within the group, selecting multiple basis learners and integration strategies, and performing Model combination and accuracy comparison, the optimal integration model was selected, and verified and adjusted by the trend and slope of long-term groundwater nitrate concentration data.
Improve the accuracy, reliability and interpretability of aquifer vulnerability assessment, provide scientific, systematic and efficient evaluation methods, suitable for aquifer vulnerability assessment at large-area scale, and enhance the objectivity and applicability of the assessment results. .
Smart Images

Figure CN118941147A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of aquifer vulnerability prediction, and in particular to an aquifer vulnerability assessment method based on an interpretable multi-integration strategy optimization framework. Background Art
[0002] Aquifers are an important part of groundwater resources, but their vulnerability is becoming increasingly prominent due to climate change, human activities and natural factors. Assessing aquifer vulnerability can help us gain a deeper understanding of the stability and vulnerability of groundwater systems, help us take targeted water resource management measures, protect the ecological environment, and prevent overexploitation and pollution of groundwater resources. At present, aquifer vulnerability assessment faces challenges and problems in data quality, method standardization, model interpretability, uncertainty processing and long-term data monitoring. In addition, the standards and methods for aquifer vulnerability assessment have not yet been unified. The groundwater system involves many types of data, including geological, meteorological, hydrological and other data, but it is difficult to obtain and integrate these data. There are differences in the evaluation indicators and methods used in different studies, and there is a lack of a common evaluation framework and standardized operating procedures.
[0003] In recent years, the demand for vulnerability assessment of complex aquifers and large regional aquifers has been increasing in hydrogeological research, and traditional methods have some limitations. With the enhancement of computer computing power and the continuous optimization of big data algorithms, data-driven machine learning methods are becoming a powerful tool to solve these problems. From traditional single learner models such as logistic regression, support vector machines and neural network models to the application of more powerful integrated learning models, more flexible and effective modeling tools are provided for aquifer vulnerability assessment.
[0004] These new model structures and algorithms can better capture the complex relationships between data, improving the accuracy of evaluation and predictive ability. However, the application of machine learning methods also faces problems such as low data quality, low method standardization, low model interpretability, and lack of a unified evaluation framework, which greatly affect the credibility and stability of the evaluation results. Summary of the invention
[0005] In order to overcome the shortcomings of the prior art, the purpose of the present invention is to provide an aquifer vulnerability assessment method based on an interpretable multi-integrated strategy optimization framework, which solves problems such as data incompleteness, model complexity and feature selection, effectively improves the accuracy, reliability and interpretability of the assessment, and provides a scientific, systematic and efficient method for large-scale aquifer vulnerability assessment.
[0006] To achieve the above object, the present invention provides the following scheme: an aquifer vulnerability assessment method based on an interpretable multi-integrated strategy optimization framework, comprising the following steps:
[0007] Collecting feature data of a small area, performing trend processing and preprocessing on the feature data, and obtaining a feature data set; wherein the feature data set is composed of features of different categories, and each category of features constitutes a set of data sets;
[0008] Based on the feature data set, the feature importance is calculated using a random forest model and the intra-group correlation coefficient is calculated using a two-way random effect model to identify and obtain feature indicators;
[0009] Select multiple base learners for classification, combine multiple integration strategies and use the itertools.combinations function to permutate and combine the base learners to obtain multiple combination models, compare the accuracy of the multiple combination models, and select and obtain the optimal integration model;
[0010] The optimal integrated model and the characteristic index are used to evaluate the vulnerability of aquifers in large-scale areas lacking nitrate data, and the trend and slope of long-term groundwater nitrate concentration data are used to verify the accuracy, and the optimal integrated model is adjusted according to the verification results.
[0011] Optionally, the characteristic data include hydrogeological conditions, meteorological elements, soil conditions, land cover and human activities.
[0012] Optionally, the trend processing calculates and obtains the nitrate concentration change trend of the characteristic data by adopting the non-parametric Mann-Kendall trend test method; the preprocessing obtains the processed characteristic data by data cleaning, standardization processing and label encoding method; wherein, the data cleaning is achieved by processing missing values, outliers, duplicate values and format conversion; the standardization processing is achieved by the minimum-maximum normalization method; the label encoding method is used to perform classification encoding on category features.
[0013] Optionally, the process of calculating feature importance using the random forest model is:
[0014] Calculating the accuracy or mean square error of the feature data set to obtain an original value;
[0015] Randomly shuffle each feature in the feature data set, calculate the accuracy or mean square error of the shuffled feature data set and obtain a shuffled value;
[0016] The original value and the shuffled value and the difference are calculated to obtain the feature importance.
[0017] Optionally, the characteristic indicators include water layer thickness, soil medium, terrain slope, lithology of the aeration zone, permeability coefficient of the aquifer, distance from the river, vertical recharge, groundwater runoff, vegetation index, groundwater exploitation, groundwater table depth, population density, distance from pollution sources, land use, soil moisture and evaporation intensity.
[0018] Optionally, the base learners include random forest, AdaBoost, gradient boosting, support vector machine, decision tree, naive Bayes and multilayer perceptron, and the integration strategy includes Bagging, Boosting and Stacking.
[0019] Optionally, the process of comparing the accuracy of multiple combination models is: comparing the accuracy, precision, recall rate and F1 value of each combination model on the validation set to obtain the optimal integrated model.
[0020] Optionally, the process of verifying the accuracy of the evaluation result of the optimal integrated model is:
[0021] The long-term trend of groundwater nitrate concentration data was calculated using the nonparametric Mann-Kendall trend test method, and the slope of the long-term groundwater nitrate concentration data was calculated using the least squares method.
[0022] The positive and negative values and the size of the slope are used to explain the trend of groundwater nitrate concentration changes, and the vulnerability characteristics of the aquifer and the relationship between the vulnerability characteristics of the aquifer and environmental changes are obtained.
[0023] Optionally, the adjustment process of the optimal integrated model is:
[0024] Tuning the hyperparameters of the optimal ensemble model based on the accuracy verification result; the hyperparameters include learning rate, tree depth, and the number of base learners;
[0025] Using a cross-validation method to evaluate the optimal integrated model, identify overfitting or underfitting problems, and take corresponding measures to adjust;
[0026] New base learners are introduced into the optimal ensemble model or poorly performing learners are removed to improve the performance and stability of the model.
[0027] The present invention discloses the following technical effects by providing an aquifer vulnerability assessment method based on an interpretable multi-integrated strategy optimization framework:
[0028] 1. This invention is based on an interpretable multi-integrated strategy optimization framework for aquifer vulnerability assessment. The framework comprehensively considers various types of data, including hydrogeological conditions, meteorological elements, soil conditions, land cover and human activities, providing a comprehensive information basis that can more accurately assess the vulnerability of aquifers.
[0029] 2. The present invention can objectively evaluate the evolution trend and change characteristics of the aquifer by trending the nitrate concentration, avoiding the influence of subjective factors on the evaluation results and improving the objectivity and credibility of the evaluation results.
[0030] 3. In the selection of base learners and integration strategies, the present invention takes into account a variety of machine learning models and integration methods, which can adapt to different data characteristics and problem requirements, and improve the applicability and generalization ability of the evaluation model.
[0031] 4. The present invention selects the optimal integrated model through model accuracy comparison and verification, avoiding the problem of improper model selection caused by relying solely on a single model or integrated method, effectively assessing the vulnerability of aquifers, and providing scientific support and reference basis for decision-making in the fields of groundwater resource management, water source protection, land use planning, etc.; it is suitable for aquifer vulnerability assessment research in large areas where global hydrochemistry data is scarce.
[0032] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS
[0033] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.
[0034] Figure 1 A flow chart of a method provided by an embodiment of the present invention;
[0035] Figure 2 A schematic diagram of an aquifer vulnerability assessment process provided by an embodiment of the present invention;
[0036] Figure 3 A result diagram of feature importance and intra-group correlation coefficient provided by an embodiment of the present invention;
[0037] Figure 4 A schematic diagram of the structure of a random forest learner used in an embodiment of the present invention;
[0038] Figure 5 A schematic diagram of the structure of the Adaboost learner provided in an embodiment of the present invention;
[0039] Figure 6 A schematic diagram of the structure of a support vector machine learner provided in an embodiment of the present invention;
[0040] Figure 7 A schematic diagram of the structure of the integration model (Stacking integration strategy) of the present invention provided for an embodiment of the present invention;
[0041] Figure 8 A graph of aquifer vulnerability assessment results provided by an embodiment of the present invention;
[0042] Fig. 9 A graph of aquifer vulnerability assessment results verified by nitrate concentration change trend and slope provided in an embodiment of the present invention. DETAILED DESCRIPTION
[0043] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0044] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.
[0045] like Figure 1-2 As shown, the present invention provides an aquifer vulnerability assessment method based on an interpretable multi-integrated strategy optimization framework, comprising the following steps:
[0046] 1. Select a small area with rich data and collect feature data, perform trend processing and preprocessing on the feature data to obtain a feature data set. The feature data set consists of features of different categories, and each category of features constitutes a set of data sets.
[0047] 1.1 Characteristic data include various types of data including hydrogeological conditions, meteorological elements, soil conditions, land cover, human activities, etc.;
[0048] 1.2 Data trend processing The trend of nitrate concentration was calculated by using the non-parametric Mann-Kendall trend test method; the calculation steps were: first calculate the S statistic, then calculate Var(s) and the standardized statistic Z, and finally determine the significance level and test results.
[0049] 1.3 Data preprocessing: The processed feature data is obtained through data cleaning, standardization and label encoding methods;
[0050] Data cleaning includes processing missing values, outlier detection and processing, data format conversion, and deduplication.
[0051] The standardization process uses the minimum-maximum normalization method to standardize the filtered data. The expression is: Among them, X i is a characteristic variable used for groundwater level prediction, is the characteristic variable after standardization, E(X) is the expectation of the characteristic variable, and SD(X) is the standard deviation of the characteristic variable.
[0052] Classification coding uses label encoding method for category features.
[0053] 2. Based on the feature data set, the feature importance is calculated using a random forest model and the intra-group correlation coefficient is calculated using a two-way random effects model to identify and obtain feature indicators.
[0054] 2.1 The process of calculating the intraclass correlation coefficient is:
[0055] like Figure 3 As shown in Figure 2, the two-way random effects model (absolute consistency) evaluates the consistency between measurements within the same group. The specific calculation formula is ICC(A,1)=(σ 2 _between-σ 2 _error) / (σ 2 _between+(k-1)*σ 2 _error), where σ 2 _between is the between-group variance, σ 2 _error is the error variance and k is the number of raters.
[0056] Depend on Figure 3It can be seen that from the perspective of the impact of a single indicator, the thickness of the aquifer mainly controls the limit of change of the aquifer vulnerability on the spatial scale, accounting for 65.91% of the impact, which is negatively correlated with the vulnerability of the aquifer. In areas with higher aquifer thickness, the aquifer vulnerability presents a lower value, because the thicker the aquifer, the stronger its self-purification ability, and the more it can filter and dilute pollutants. Secondly, the vertical recharge accounts for 7.47% of the impact on the aquifer vulnerability, showing a positive correlation, because it is easy to carry pollutants into the aquifer during the process of precipitation recharging groundwater. In addition, groundwater exploitation, aquifer permeability coefficient and vadose zone lithology are all positively correlated with aquifer vulnerability, because excessive exploitation will lead to lower groundwater levels, reduced groundwater resources and changes in groundwater quality, which in turn affects the stability and sustainability of the aquifer; the larger the aquifer permeability coefficient, the easier it is for pollutants (nitrates) to enter the aquifer with the recharge water; the stronger the permeability of the vadose zone lithology, the greater the risk of aquifer contamination. Among them, soil media, land use and evaporation intensity have no obvious positive or negative correlation with aquifer vulnerability, and each accounted for a very low proportion, less than 1%.
[0057] After calculating the Pearson correlation coefficient of all characteristic variables, most of the indicators are independent of each other, and there is no indicator with a significant high correlation, avoiding the problem of multicollinearity. Among them, the correlation between aquifer thickness and NDVI is 37%, and the correlation strength of the remaining variables is less than 35%. At the same time, combined with the ranking of single-factor feature importance, land use, soil medium and evaporation intensity have little effect on aquifer vulnerability, and there is no correlation with other indicators, so they are removed from the data set.
[0058] 2.2 The process of calculating feature importance is:
[0059] like Figure 3 As shown in the figure, feature importance is an important output in the random forest algorithm, which is used to measure the contribution of each feature to the model prediction performance. First, the model performance (such as accuracy or mean square error) on the original feature data set is calculated. For each feature, the value of the feature is randomly shuffled, and then the model performance on the feature data set is calculated again. Feature importance is measured by the difference between the original model performance and the shuffled model performance. The larger the difference, the more important the feature.
[0060] The random forest model is used to calculate the feature importance. During the construction of each decision tree, the random forest measures the feature importance by calculating the reduction in impurity of each feature when the node is split. Information gain is usually used to measure the reduction in impurity.
[0061] The feature importance in random forest is calculated based on the average importance of all decision trees. For each feature, the importance of the feature in all trees is calculated and the average is taken as the final feature importance;
[0062] After calculating the feature importance, it is normalized so that the sum of all feature importances equals 1 or 100%. This allows for a more intuitive comparison of the relative importance of different features;
[0063] Once the Shapley value of each feature is calculated, a SHAP graph can be generated. A SHAP graph is usually a number of horizontal scatter plots, with each feature corresponding to a horizontal scatter point. The length of the scatter point represents the degree of influence of the feature on the model prediction, and the positive and negative values represent the positive and negative influence on the model output.
[0064] The calculation formula of Shapley value is as follows:
[0065]
[0066] in represents the Shapley value of participant i, v(S) represents the predicted output (the prediction result of the model) on a specific feature subset S, N represents the set of all features, |N| represents the number of features, represents the feature subset excluding feature i, |S| represents the size of feature subset S, and v(S∪{i}) represents the predicted output when the feature subset S includes feature i and feature i.
[0067] 2.3 Characteristic indicators include: aquifer thickness, soil medium, terrain slope, lithology of the aeration zone, aquifer permeability, distance from the river, vertical recharge, groundwater runoff, vegetation index (NDVI), groundwater extraction, groundwater table depth, population density, distance from pollution sources, land use, soil moisture and evaporation intensity.
[0068] 3. Select multiple base learners for classification, combine multiple integration strategies and use the itertools.combinations function to permutate and combine the base learners to obtain multiple combination models, compare the accuracy of the multiple combination models, and select and obtain the optimal integration model.
[0069] 3.1 Base learners include Random Forest, AdaBoost, Gradient Boosting, Support Vector Machine (SVC), Decision Tree, Naive Bayes and Multilayer Perceptron. Ensemble strategies include Bagging, Boosting and Stacking.
[0070] 3.11 Figure 4 is a schematic diagram of the structure of the random forest based learner, such as Figure 4 As shown in Figure 1, in a random forest, each decision tree is built by training on a random subset of the dataset. These random subsets are obtained through bootstrap sampling, with the same number of samples, but there may be duplicate samples. The decision tree construction process is based on the splitting of features to maximize the information gain indicator.
[0071] Random forest introduces two kinds of randomness to increase the diversity of the model. The training samples of each decision tree are obtained through bootstrap sampling, which means that each tree is trained on a different subset of data. At each node, when selecting the best split feature, a part of the features is randomly selected for consideration, rather than selecting the best split feature among all the features. This parameter is called max_features.
[0072] The prediction of random forest is obtained by integrating the prediction results of all decision trees. For classification problems, majority voting is usually used to determine the final prediction category.
[0073] 3.12 Figure 5 is a schematic diagram of the structure of the AdaBoost base learner, as shown in Figure 5 As shown, the principle of AdaBoost can be described by the following steps:
[0074] Step 1: For a training data set containing N samples, initialize the weight of each sample to Where i=1,2,…,N.
[0075] Step 2: For T rounds of iterations, train a weak classifier (e.g., decision stump) under the current weights, and select the best classifier by minimizing the weighted error. The weighted error is usually expressed as a weighted error rate.
[0076] Calculate the weight of the classifier (also called "weak classifier weight"), which depends on the performance of the classifier in the current round. The better the classifier performance, the higher the weight, and vice versa;
[0077] Update sample weights based on the classifier’s performance: the weights of samples that are misclassified are increased, while the weights of samples that are correctly classified are decreased.
[0078] Step 3: Perform weighted combination of all weak classifiers to obtain the final strong classifier. The weighted combination is usually completed by weighted summing the output of each weak classifier.
[0079] 3.13 Figure 6 It is a structural diagram of the SVC learner, as shown in Figure 6As shown, the principle of SVC can be described by the following steps:
[0080] Step 1: Assume that the training data is linearly separable, that is, there is a hyperplane that can completely separate data of different categories.
[0081] Step 2: The goal is to find a hyperplane that maximizes the distance (interval) between samples of different categories and the hyperplane. This hyperplane is called the optimal segmentation hyperplane.
[0082] Step 3: Some sample points near the optimal segmentation hyperplane are called support vectors. These support vectors are the key factors in determining the optimal segmentation hyperplane.
[0083] Step 4: For the nonlinearly separable case, SVM uses a kernel function to map the input features to a high-dimensional feature space, so that the samples are linearly separable in the high-dimensional space.
[0084] The Sigmoid kernel function is one of the commonly used kernel functions in support vector machines (SVMs). It can be used to map data to a high-dimensional feature space to handle nonlinear classification problems. The mathematical expression of the Sigmoid kernel function is as follows:
[0085]
[0086] where x i and x j is the feature vector of the input sample, α and θ are the parameters of the Sigmoid kernel function, which need to be determined by methods such as cross-validation.
[0087] 3.14 Figure 7 It is a structural diagram of the integrated model (Stacking integration strategy), such as Figure 7 As shown in the figure, the specific steps to establish the Stacking integration strategy are:
[0088] Step 1: Train the base learners (first layer models), divide the training set into k mutually exclusive subsets (through k-fold cross validation), train each base learner (e.g., decision tree, random forest, support vector machine, etc.) on k-1 subsets, and make predictions on the remaining subset. Repeat this process k times, rotating the validation set each time, and eventually each base learner will have a prediction result for each sample.
[0089] Step 2: Generate a new training set. Use the prediction results of the base learner (the prediction of the validation set) to construct a new training set. The features of this training set are the prediction results of the base learner, and the labels are the labels of the original data. The size of this new training set is the same as the original training set, but the feature dimension becomes the number of base learners.
[0090] Step 3: Train the meta-learner (second-layer model). Use the new training set to train a meta-learner. The goal of the meta-learner is to make a more accurate final prediction by learning the prediction results of the base learner.
[0091] Step 4: Prediction: Each base learner predicts the test sample to obtain a new feature set, and the meta-learner predicts this new feature set to obtain the final prediction result.
[0092] 3.2 The base learner generates combinations of specified lengths using the itertools.combinations function to achieve permutations and combinations. The calculation steps are: first initialize, then generate sequence indexes, use the initial index sequence to extract elements from the input iterable object, form a combination and output it, then iterate to generate the next combination, and finally repeat the above steps until all combinations are generated.
[0093] 3.3 The process of comparing the accuracy of multiple combined models is as follows: evaluate the accuracy, precision, recall and F1 value of the model based on the validation set; compare and find the optimal model to evaluate the vulnerability of aquifers in large areas lacking nitrate data.
[0094] Figure 8 The results of aquifer vulnerability assessment based on an interpretable multi-integrated strategy optimization framework are shown in Figure 2. Figure 8 As shown in the figure, the 13 selected indicators are input into the selected Stacking integrated model, and the aquifer vulnerability distribution results are obtained accordingly. The overall aquifer vulnerability is high, and the aquifer vulnerability gradually decreases from the western piedmont to the eastern plain. The extremely high vulnerability area and the high vulnerability area account for 20% and 23.5% respectively, and are mainly distributed in the piedmont area. This is mainly because the groundwater level is small, which reduces the distance between surface pollutants and groundwater, making it easier to enter the aquifer; and the piedmont area has a large vertical recharge, and the possibility of pollutants entering the aquifer with the recharge water increases. The medium vulnerability area and the low vulnerability area account for 26.8% and 29.7% respectively, and are mainly distributed in the eastern plain. This is mainly because the aquifer is thicker than the piedmont area and the vertical recharge is less than the piedmont area, making it difficult for pollutants to enter the aquifer.
[0095] 4. Use the optimal integrated model and the characteristic indicators to evaluate the aquifer vulnerability in large-scale areas that lack nitrate data, and use the trend and slope of long-term groundwater nitrate concentration data to verify the accuracy, and adjust the optimal integrated model based on the verification results.
[0096] 4.1 The process of accuracy verification is as follows:
[0097] Long-term trends were calculated using the nonparametric Mann-Kendall trend test method;
[0098] Then use the least squares method to calculate the slope; the least squares method first finds the best fit line of the data points, and then calculates the slope. The calculation formula for the slope is in, and They are x i and i The average value of .
[0099] By analyzing the positive and negative values of the slope and its magnitude, the changing trend of groundwater nitrate concentration is explained to further understand the vulnerability characteristics of the aquifer and its response to environmental changes.
[0100] Fig. 9 The aquifer vulnerability assessment result diagram verified for the mean, variance, trend and slope of nitrate concentration is shown in Figure 2. Figure 8 As shown in the figure, the calculation of average concentration and variance helps to understand the statistical characteristics of nitrate nitrogen concentration, including the degree of variation, volatility and distribution pattern, and provides an overall summary of the data. The average concentration and variance of nitrate nitrogen in five periods (2018-June 2022) were calculated to represent the nitrate pollution in the aquifer. After normalization analysis, it was found that the average concentration, variance and aquifer vulnerability prediction results had the same trend (Figure (c)). With the increase of average concentration and variance, the vulnerability of the aquifer gradually increased. By comparing the box plots of different vulnerability areas, the differences in the average concentration and variance in the vulnerability of different regions were observed (Figure (d)). It was found that the overall box and average values increased with the increase of vulnerability level.
[0101] The average concentration and variance alone are not enough to reflect the vulnerability of aquifers, so the trend of nitrate nitrogen concentration over time (2018-2022) in 94 wells was analyzed using the non-parametric Mann-Kendall trend test combined with the slope of the least squares method. The results show that the observed trend is considered statistically significant at a threshold of significance level (p) less than or equal to 0.1. The analysis results show that the proportion of wells with unchanged nitrate nitrogen concentration reached 58.5%, about one-third of the wells showed a downward trend (24.4%), and only 17% of the wells showed an upward trend. Further research found that among the 24 wells with a significantly negative trend slope of nitrate nitrogen concentration, the decline rate ranged from -0.00028 to -1.62 mg / L / year, with an average of -0.19 mg / L / year. Among the 70 wells with a significantly positive trend slope for nitrate nitrogen concentration, the growth rate ranged from 0.02 to 10.35 mg / L / year, with an average of 0.98 mg / L / year.
[0102] Through trend analysis and the spatial relationship between slope and aquifer vulnerability prediction results, it was found that although the trend of individual sampling points in the eastern part of the study area was inconsistent with the vulnerability zoning, most of the overall concentration points at all levels were in the corresponding vulnerability zoning, with a strong spatial relationship. By comparing the box plots of different regions, the differences in nitrate concentration trends and slopes in the vulnerability of different regions were observed (Figure (e)). It was found that the overall vulnerability level increased with the increase in the trend of nitrate nitrogen concentration, and the slope and predicted value also increased with the increase in the intensity of the vulnerability zoning.
[0103] 4.2 The adjustment process of the optimal integrated model is as follows:
[0104] Based on the verification results, the hyperparameters of the ensemble model are tuned, including the learning rate, tree depth, number of base learners, etc., to ensure that the model's performance is optimal;
[0105] Use methods such as cross-validation to evaluate the model, identify possible overfitting or underfitting problems, and take appropriate measures to adjust them;
[0106] Introduce new base learners or remove poorly performing learners in the ensemble model to further improve the performance and stability of the model.
[0107] Therefore, the present invention solves the problems of data incompleteness, model complexity and feature selection by providing an aquifer vulnerability assessment method based on an interpretable multi-integrated strategy optimization framework, effectively improves the accuracy, reliability and interpretability of the assessment, and provides a scientific, systematic and efficient method for large-scale aquifer vulnerability assessment.
[0108] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.
[0109] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of the present invention. At the same time, for those skilled in the art, according to the idea of the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.
Claims
1. Aquifer vulnerability assessment method based on an interpretable multi-integrated strategy optimization framework, characterized by: The following steps are involved: Collecting feature data of a small area, performing trend processing and preprocessing on the feature data, and obtaining a feature data set; wherein the feature data set is composed of features of different categories, and each category of features constitutes a set of data sets; Based on the feature data set, the feature importance is calculated using a random forest model and the intra-group correlation coefficient is calculated using a two-way random effect model to identify and obtain feature indicators; Select multiple base learners for classification, combine multiple integration strategies and use the itertools.combi nations function to permutate and combine the base learners to obtain multiple combination models, compare the accuracy of the multiple combination models, and select and obtain the optimal integration model; The optimal integrated model and the characteristic index are used to evaluate the vulnerability of aquifers in large-scale areas lacking nitrate data, and the trend and slope of long-term groundwater nitrate concentration data are used to verify the accuracy, and the optimal integrated model is adjusted according to the verification results.
2. The aquifer vulnerability assessment method based on an interpretable multi-integrated strategy optimization framework according to claim 1 is characterized in that: The characteristic data include hydrogeological conditions, meteorological elements, soil conditions, land cover and human activities.
3. The aquifer vulnerability assessment method based on an interpretable multi-integrated strategy optimization framework according to claim 2 is characterized in that: The trend processing calculates and obtains the nitrate concentration change trend of the characteristic data by adopting the non-parametric Mann-Kendall trend test method; the preprocessing obtains the processed characteristic data by data cleaning, standardization processing and label encoding method; wherein, the data cleaning is achieved by processing missing values, outliers, duplicate values and format conversion; the standardization processing is achieved by the minimum-maximum normalization method; the label encoding method is used to perform classification encoding on category features.
4. The aquifer vulnerability assessment method based on an interpretable multi-integrated strategy optimization framework according to claim 3 is characterized in that: The process of calculating feature importance using the random forest model is: Calculating the accuracy or mean square error of the feature data set to obtain an original value; Randomly shuffle each feature in the feature data set, calculate the accuracy or mean square error of the shuffled feature data set and obtain a shuffled value; The original value and the shuffled value and the difference are calculated to obtain the feature importance.
5. The aquifer vulnerability assessment method based on an interpretable multi-integrated strategy optimization framework according to claim 4 is characterized in that: The characteristic indicators include water layer thickness, soil medium, terrain slope, lithology of the aeration zone, permeability coefficient of the aquifer, distance from the river, vertical recharge, groundwater runoff, vegetation index, groundwater exploitation, groundwater table depth, population density, distance from pollution sources, land use, soil moisture and evaporation intensity.
6. The aquifer vulnerability assessment method based on an interpretable multi-integrated strategy optimization framework according to claim 5 is characterized in that: The base learners include random forest, AdaBoost, gradient boosting, support vector machine, decision tree, naive Bayes and multilayer perceptron, and the integration strategies include bagging, boosting and stacking.
7. The aquifer vulnerability assessment method based on an interpretable multi-integrated strategy optimization framework according to claim 6 is characterized in that: The process of comparing the accuracy of multiple combined models is: comparing the accuracy, precision, recall rate and F1 value of each combined model on the validation set to obtain the optimal integrated model.
8. The aquifer vulnerability assessment method based on an interpretable multi-integrated strategy optimization framework according to claim 7 is characterized in that: The process of verifying the accuracy of the evaluation results of the optimal integrated model is as follows: The long-term trend of the long-term groundwater nitrate concentration data was calculated using the nonparametric Mann-Kendall trend test method, and the slope of the long-term groundwater nitrate concentration data was calculated using the least squares method; The positive and negative values and the size of the slope are used to explain the trend of groundwater nitrate concentration changes, and the vulnerability characteristics of the aquifer and the relationship between the vulnerability characteristics of the aquifer and environmental changes are obtained.
9. The aquifer vulnerability assessment method based on an interpretable multi-integrated strategy optimization framework according to claim 8 is characterized in that: The adjustment process of the optimal integrated model is: Tuning the hyperparameters of the optimal ensemble model based on the accuracy verification result; the hyperparameters include learning rate, tree depth, and the number of base learners; Using a cross-validation method to evaluate the optimal integrated model, identify overfitting or underfitting problems, and take corresponding measures to adjust; New base learners are introduced into the optimal ensemble model or poorly performing learners are removed to improve the performance and stability of the model.
Citation Information
Cited By
Underground water hardness influence factor analysis method based on random forest and SHAP
CN119202951A
Regional lake and reservoir nutritive salt reference setting method based on Bayesian hierarchical model
CN119848026A
