Underground water level dynamic prediction method for excavation of overtopping ground layer foundation pit

Abstract

The invention discloses a method for dynamically predicting the underground water level during excavation of an overtopping ground layer foundation pit, and relates to the technical field of dynamic prediction of the underground water level, which comprises the following steps of: acquiring multi-source data related to the underground water level in an engineering range; comprising information of soil moisture content, stratum permeability coefficient, rainfall capacity, foundation pit excavation depth, foundation pit dewatering scheme and foundation pit waterproof and drainage measures; historical underground water level, underground aquifer data and Yangtze River water level change information are obtained. According to the method, data quality is optimized through multi-source data acquisition, correlation analysis and z-score standardization, nonlinear modeling is realized in combination with a support vector machine (SVM) and Bayesian optimization, and model precision and generalization ability are improved. Model parameters are dynamically adjusted through a real-time monitoring and feedback optimization mechanism, the adaptability to sudden hydrological changes is enhanced, engineering risks are effectively reduced, and construction safety and economical efficiency are guaranteed.

Description

Technical Field

[0001] The invention relates to the technical field of groundwater level dynamic prediction, and in particular to a method for dynamically predicting groundwater level in a floodplain foundation pit excavation. Background Art

[0002] In the process of underground space development in cities along the Yangtze River, foundation pit engineering, as its main component, faces complex groundwater problems, especially in the Yangtze River floodplain strata area. Due to the complexity of geological and hydrological conditions, its engineering risks are significantly increased. The Yangtze River floodplain is generally distributed with uneven thickness of weak silty soft soil layers, which have the characteristics of low shear strength, high compressibility, high water content, and low permeability. There is a direct hydraulic connection between the underlying thick sand layer and the Yangtze River water body. This special geological structure makes the dynamic change of groundwater level an important factor affecting the safety and construction economy of foundation pit excavation projects. Too high groundwater level may lead to disasters such as instability of foundation pit slopes and water gushing, while too low water level may have adverse effects on the surrounding environment and geological stability. Therefore, accurate prediction of the dynamic change of groundwater level is crucial for the successful implementation of foundation pit engineering in floodplain areas.

[0003] Existing technologies face significant limitations in the dynamic prediction of groundwater levels. On the one hand, groundwater levels are affected by the nonlinear coupling of multiple factors such as rainfall, temperature, surface seepage, geological characteristics and construction process. Traditional linear models or methods based on single factor analysis are difficult to effectively capture these complex nonlinear relationships; on the other hand, in some scenarios, historical data on groundwater levels are relatively scarce, especially in new construction areas or when seasonal changes are significant. Traditional statistical methods often have poor prediction results due to insufficient samples. In addition, existing models lack dynamic adjustment capabilities and are difficult to adapt to real-time changing environmental data during construction, resulting in prediction accuracy that is difficult to meet actual needs, which in turn poses a threat to construction safety.

[0004] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not constitute the prior art that is already known to one of ordinary skill in the art. Summary of the invention

[0005] The purpose of the present invention is to provide a method for dynamically predicting the groundwater level in the excavation of a foundation pit in a floodplain stratum. Through multi-source data collection and correlation analysis, a comprehensive description of the complex geological structure and dynamic hydrological conditions of the floodplain stratum is achieved, and the quality of input data is optimized in combination with z-score standardization, laying the foundation for high-precision modeling. The nonlinear changes of the groundwater level are modeled using a support vector machine (SVM) in combination with a radial basis kernel function (RBF kernel), and the generalization ability and prediction accuracy of the model are improved through Bayesian optimization, cross-validation, and L2 regularization, effectively solving the problems of complex nonlinear relationships and data scarcity. In addition, the real-time monitoring and feedback optimization mechanism, by dynamically adjusting the model input and parameter configuration, ensures the adaptability of the model to sudden hydrological changes, reduces engineering risks such as water gushing and slope instability, and improves construction safety and economy to solve the problems in the above-mentioned background technology.

[0006] In order to achieve the above object, the present invention provides the following technical solution: a method for dynamically predicting groundwater level in a floodplain foundation pit excavation, comprising the following steps:

[0007] Collect multi-source data related to groundwater level within the scope of the project, including soil moisture content, stratum permeability coefficient, rainfall, foundation pit excavation depth, foundation pit dewatering plan, foundation pit drainage measures; as well as historical groundwater level, underground aquifer data and Yangtze River water level change information. Through the collection of the above multi-dimensional data, the dynamic characteristics of geological structure, hydrological conditions and construction environment within the scope of the project are fully reflected, providing a complete input data set for subsequent groundwater level prediction.

[0008] The hydraulic connection between the water level of the Yangtze River and the underground aquifer was analyzed, and the correlation coefficients at different lag times were calculated based on the time series data to determine the main influencing factors of the groundwater level; data with strong correlation were used as input features, and weak correlation or redundant features were eliminated. Subsequently, different data features were z-score standardized to ensure the consistency of data features at different scales and optimize the input quality of subsequent models.

[0009] The radial basis kernel function (RBF kernel) was used to construct the support vector machine (SVM) model, and the Bayesian optimization algorithm was used to search and determine the optimal combination of the model's penalty parameter and kernel parameter γ to enhance the model's learning ability. The normalized data was divided into a training set and a test set, and the SVM model was trained based on the training set. The model performance was evaluated by cross-validation method, and the model was repeatedly optimized to improve the prediction accuracy.

[0010] The test set data is input into the optimized SVM model to predict the dynamic changes of the groundwater level during the foundation pit excavation process; the prediction results include the height values ​​of the groundwater level at different time periods. The error calculation between the predicted value output by the model and the actual monitoring value is used to evaluate the accuracy and reliability of the model to ensure that the prediction results can meet the on-site construction needs.

[0011] During the foundation pit excavation construction process, the groundwater level is monitored in real time, and the input data of soil, hydrological and climatic conditions within the project scope are continuously updated; the latest data is input into the SVM model, and when there is a significant error between the monitoring data and the predicted value, the real-time monitoring data is used to feedback and optimize the SVM model, and the adaptability of the model is improved by adjusting the input features or model parameters to ensure that the prediction results are dynamically adjusted to fit the actual situation.

[0012] Preferably, in the data collection step, the hydraulic connection between the water level of the Yangtze River and the water level of the underground aquifer is studied through time series analysis, and the lag time of the change of the underground aquifer behind the change of the water level of the Yangtze River is analyzed by correlation calculation, and the correlation coefficients of different lag times are calculated respectively; if the lag time correlation is strong, the underground aquifer data is used as the main input, otherwise a comprehensive analysis is carried out in combination with the Yangtze River water level data.

[0013] Preferably, in the data normalization processing step, a z-score standardization method is used to perform scale uniformity processing on characteristic values ​​of different types of data, and the standardization formula is:

[0014] Among them, x is the original data value, μ is the mean of the feature, and σ is the standard deviation of the feature. This standardization process eliminates the differences in scales between data features and improves the prediction accuracy of the model.

[0015] Preferably, in the step of constructing the support vector machine model (SVM), a radial basis kernel function (RBF kernel) is used as a kernel function, and its form is defined as:

[0016] K(x i , x j ) = exp(-γ·‖x i -x j ‖ 2 ), where x i and x j are two sample points in the input space, ‖x i -x j ‖ 2 represents the Euclidean distance between two points, γ is the kernel parameter, and the kernel parameter γ and penalty parameter are adjusted through the Bayesian optimization algorithm to achieve the optimal parameter selection.

[0017] Preferably, in the model optimization step, when optimizing the support vector machine model, the robustness of the model is improved by using the L2 regularization method, and the optimization target formula is: Among them, ‖w‖ 2 is the model complexity, C is the regularization parameter, ξ i is a slack variable, i is the index of the data, which is used to represent the i-th sample, and n is the total number of samples, indicating that there are a total of n samples in the data set. By gradually adjusting the parameters using small steps within the optimal parameter range, the generalization ability of the model can be further improved.

[0018] Preferably, in the step of predicting the dynamic change of groundwater level, an optimized support vector machine model is used, and the soil moisture content, stratum permeability, rainfall data, foundation pit excavation depth, foundation pit dewatering plan, drainage measures, historical groundwater level and underground aquifer data within the newly collected project scope are taken as input, and the predicted value is used to represent the dynamic change trend of the groundwater level. The prediction accuracy of the model output result is evaluated by the mean square error (MSE).

[0019] Preferably, in the real-time monitoring and feedback optimization step, when the error between the real-time monitored groundwater level data and the model prediction value exceeds a preset range, updated soil moisture content, formation permeability, rainfall, foundation pit depth and precipitation scheme data are re-collected to form a new test set and optimize the support vector machine model to reduce the prediction error.

[0020] Preferably, real-time monitoring data is automatically collected through a sensor network, including groundwater level monitoring wells, rain gauges and soil moisture monitoring equipment arranged within the foundation pit. The collected data is transmitted to the prediction system in real time as input to the feedback optimization model to improve the dynamic adaptability of groundwater level prediction.

[0021] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0022] The present invention achieves a comprehensive description of the complex geological structure and dynamic hydrological conditions of the floodplain strata by collecting multi-source data and performing correlation analysis and normalization processing on the data. This multidimensional data fusion strategy can capture the main driving factors of groundwater level changes, while avoiding the interference of redundant features on the model and improving the input data quality of the prediction model. In addition, the z-score standardization method eliminates the deviation of various eigenvalues ​​caused by scale differences, ensuring the effective learning of data features by the model. This comprehensive and accurate data acquisition and processing method lays a solid foundation for high-precision modeling of groundwater level prediction in foundation pit construction, breaking through the limitations of traditional methods in data collection and feature selection.

[0023] The present invention adopts the support vector machine (SVM) model combined with the radial basis kernel function (RBF kernel) to model the multi-source nonlinear relationship, which can effectively capture the complex coupling relationship between groundwater level and factors such as rainfall, soil permeability, and the water level of the Yangtze River. The penalty parameters and kernel parameters are automatically tuned by the Bayesian optimization algorithm. Combined with the cross-validation technology and the L2 regularization method, the model achieves high-precision prediction of groundwater level changes under finite sample conditions. In addition, the design of the objective function comprehensively considers the balance between model complexity and misclassification, so that the model has a strong generalization ability while dealing with nonlinear problems. Especially in the complex geological conditions and construction environment of the floodplain area, the accurate prediction ability of the SVM model can provide a reliable basis for risk prediction in foundation pit excavation. Compared with traditional statistical methods, this modeling method not only improves the adaptability to nonlinear multivariate problems, but also significantly improves the prediction performance under data scarcity.

[0024] The real-time monitoring and feedback optimization mechanism of the present invention realizes the adaptive adjustment of the groundwater level prediction model in a dynamic construction environment. By real-time monitoring of data such as soil moisture content, rainfall, foundation pit depth, and error comparison with the model prediction value, when the error exceeds the preset range, the new data is promptly incorporated into the model training set, and the parameter configuration of the support vector machine is re-optimized to adapt the model to changes in field conditions. This circular feedback mechanism ensures that the prediction model can still maintain high-precision prediction capabilities in the face of emergencies (such as extreme rainfall or drastic fluctuations in the water level of the Yangtze River). In addition, the flexible arrangement of the monitoring system (such as groundwater level monitoring wells, rain gauges, and soil moisture sensors, etc.) enhances the timeliness and coverage of data collection, and provides high-quality data support for feedback optimization. This real-time adjustment capability not only ensures construction safety, but also significantly reduces the risks of engineering disasters such as water gushing and slope instability caused by inaccurate groundwater level prediction, reflecting the economy and reliability of this method in practical applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the embodiments of the present application 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 recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0026] Figure 1 The present invention is a method flow chart of the method for dynamically predicting groundwater level in floodplain stratum foundation pit excavation.

[0027] Figure 2 This is a prediction principle diagram of the SVM model for dynamic prediction of groundwater level in floodplain foundation pit excavation of the present invention. DETAILED DESCRIPTION

[0028] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these example embodiments are provided so that the description of the present disclosure will be more comprehensive and complete, and the concept of the example embodiments will be fully conveyed to those skilled in the art.

[0029] The present invention provides Figure 1 to Figure 2 The method for dynamically predicting groundwater level in floodplain foundation pit excavation shown in the figure comprises the following steps:

[0030] Collect multi-source data related to groundwater level within the scope of the project, including soil moisture content, stratum permeability coefficient, rainfall, foundation pit excavation depth, foundation pit dewatering plan, foundation pit drainage measures; as well as historical groundwater level, underground aquifer data and Yangtze River water level change information. Through the collection of the above multi-dimensional data, the dynamic characteristics of geological structure, hydrological conditions and construction environment within the scope of the project are fully reflected, providing a complete input data set for subsequent groundwater level prediction.

[0031] The hydraulic connection between the water level of the Yangtze River and the underground aquifer was analyzed, and the correlation coefficients at different lag times were calculated based on the time series data to determine the main influencing factors of the groundwater level; data with strong correlation were used as input features, and weak correlation or redundant features were eliminated. Subsequently, different data features were z-score standardized to ensure the consistency of data features at different scales and optimize the input quality of subsequent models.

[0032] The radial basis kernel function (RBF kernel) was used to construct the support vector machine (SVM) model, and the Bayesian optimization algorithm was used to search and determine the optimal combination of the model's penalty parameter and kernel parameter γ to enhance the model's learning ability. The normalized data was divided into a training set and a test set, and the SVM model was trained based on the training set. The model performance was evaluated by cross-validation method, and the model was repeatedly optimized to improve the prediction accuracy.

[0033] The test set data is input into the optimized SVM model to predict the dynamic changes of the groundwater level during the foundation pit excavation process; the prediction results include the height values ​​of the groundwater level at different time periods. The error calculation between the predicted value output by the model and the actual monitoring value is used to evaluate the accuracy and reliability of the model to ensure that the prediction results can meet the on-site construction needs.

[0034] During the foundation pit excavation construction process, the groundwater level is monitored in real time, and the input data of soil, hydrological and climatic conditions within the project scope are continuously updated; the latest data is input into the SVM model, and when there is a significant error between the monitoring data and the predicted value, the real-time monitoring data is used to feedback and optimize the SVM model, and the adaptability of the model is improved by adjusting the input features or model parameters to ensure that the prediction results are dynamically adjusted to fit the actual situation.

[0035] The method for dynamically predicting groundwater level in floodplain foundation pit excavation of the present invention first deploys sensor networks and monitoring equipment in the construction area to collect multi-source data to fully reflect the characteristics of dynamic changes in geological structure, hydrological conditions and construction environment. The collected data mainly include the following categories:

[0036] (1) Soil and stratum property data: Soil moisture content is measured by soil moisture monitoring equipment, and stratum permeability coefficient is obtained through geotechnical tests. These data reflect the water absorption capacity and permeability characteristics of the soil and are key factors affecting groundwater level changes;

[0037] (2) Meteorological data: Use rain gauges to monitor rainfall in real time and obtain information on changes in climate conditions in the project area;

[0038] (3) Construction-related data: record the excavation depth of the foundation pit, dewatering plan (such as the layout of dewatering wells and the amount of water pumped), drainage measures, etc. These data can directly reflect the impact of foundation pit construction on the groundwater level;

[0039] (4) Historical hydrological data: collect historical groundwater level data and underground aquifer characteristics data, including aquifer distribution, water storage capacity, etc.;

[0040] (5) Yangtze River water level change information: Monitor the Yangtze River water level changes through hydrological stations and study its coupling relationship with the groundwater level of the foundation pit.

[0041] After completing the data collection, the hydraulic connection between the Yangtze River water level and the underground aquifer water level is analyzed through time series. During the analysis process, the lag response of the Yangtze River water level to the underground aquifer water level is first calculated, that is, the degree of influence of the Yangtze River water level change on the groundwater level under different lag times (such as 1 day, 2 days, 3 days, etc.), which is specifically quantified by calculating the correlation coefficient. If the lag time correlation is strong, it means that the groundwater level change is mainly driven by the Yangtze River water level, and the underground aquifer and Yangtze River water level data should be used as model input features as a priority; if the correlation is weak, a comprehensive analysis is performed in combination with other features such as soil permeability and rainfall. In order to improve the consistency of the model input data, z-score standardization is used to process various types of data, and the standardization formula is:

[0042] Among them, x is the original data value, μ is the mean of the feature, and σ is the standard deviation of the feature. Standardization eliminates the differences in data feature scales, optimizes the quality of input data, and lays the foundation for the accurate prediction of subsequent models. This data collection and correlation analysis process can comprehensively obtain the main factors affecting the dynamic changes of groundwater levels, effectively simplify the model input features, and improve prediction efficiency.

[0043] After completing multi-source data collection and processing, this method uses support vector machine (SVM) to build a prediction model to deal with the nonlinear problem of groundwater level change. The SVM model selects the radial basis kernel function (RBF kernel) as the core algorithm, and its kernel function is defined as:

[0044] K(x i , x j )=exp(-γ·‖x i -x j ‖ 2 ), where x i and x j are two sample points in the input space, ‖x i -x j ‖ 2 Represents the Euclidean distance between two points, γ is the kernel parameter, and the kernel parameter γ and the penalty parameter are adjusted by the Bayesian optimization algorithm to determine the mapping ability of the model in high-dimensional space. The RBF kernel function is suitable for processing nonlinear coupling relationships between features such as formation permeability and rainfall. To ensure that the model has optimal performance, the penalty parameter and kernel parameter γ are adjusted using the Bayesian optimization algorithm, and the generalization ability of the model is evaluated through cross-validation (such as 5-fold or 10-fold). The specific process is as follows:

[0045] Parameter optimization: Set the parameter range, such as the penalty parameter between 0.1 and 100, the kernel parameter γ between 0.001 and 1, and search for the best combination through Bayesian optimization to minimize the mean square error (MSE) of the model on the training set and the validation set.

[0046] Model evaluation: The collected standardized data is divided into training sets and test sets. The training set is used for model training, and the test set is used to evaluate model performance. The cross-validation method is used, with K-1K-1K-1 subsets as training sets and the remaining subset as validation sets. The performance of all subsets is tested cyclically, thereby reducing the model's dependence on a single data partition.

[0047] Regularization optimization: Using L2 regularization strategy, optimize the objective function:

[0048] Among them, ‖w‖ 2 is the model complexity, C is the regularization parameter, ξ i is a slack variable, i is the index of the data, which is used to represent the i-th sample, and n is the total number of samples, which means that there are n samples in the data set. By adjusting the parameters, the model complexity and misclassification rate are balanced to prevent the model from overfitting or underfitting. After the optimization is completed, the test data is input into the model to further verify its prediction performance and provide a reliable basis for the dynamic prediction of groundwater level.

[0049] During the excavation of the foundation pit, this method dynamically tracks the groundwater level through a real-time monitoring system, including setting up groundwater level monitoring wells, rain gauges and soil moisture sensors to collect monitoring data in real time. These data are transmitted to the prediction system wirelessly and input into the optimized SVM model for dynamic prediction of groundwater levels. The prediction results include the dynamic change trend of the groundwater level in different time periods. The predicted value output by the model is compared with the actual monitoring value to evaluate the accuracy of the model. When the error between the predicted value and the actual monitoring value exceeds the preset range (for example, within 5%), it means that the current model has not fully adapted to the changes on site and feedback optimization is required.

[0050] Feedback optimization specifically includes the following steps: first, the latest collected data is integrated with historical data to form a new training set; second, the SVM model parameters are readjusted based on the new training set, especially the kernel parameter γ and penalty parameter are optimized to adapt to the new construction environment changes; finally, the optimized model is used to predict the trend of groundwater level changes again and conduct real-time monitoring verification. This cyclic process ensures that the model can be dynamically adjusted, always maintains high-precision prediction capabilities, and provides real-time support for construction decisions.

[0051] In addition, this method can adjust the monitoring frequency and data collection strategy according to the needs of the construction site. For example, when rainfall is frequent or the water level of the Yangtze River changes dramatically, the monitoring frequency and collection density are increased to obtain the latest hydrological change information in a timely manner. Under this dynamic feedback optimization mechanism, the present invention can effectively respond to changes in complex construction environments, provide scientific groundwater level dynamic prediction support for foundation pit excavation, help reduce the risk of water gushing and slope instability, and ensure the construction safety and economy of the project.

[0052] The present invention realizes a comprehensive description of the complex geological structure and dynamic hydrological conditions of the floodplain stratum by collecting multi-source data (such as soil moisture content, stratum permeability coefficient, rainfall, foundation pit excavation depth, Yangtze River water level, etc.) and performing correlation analysis and normalization on the data. This multidimensional data fusion strategy can capture the main driving factors of groundwater level changes, while avoiding the interference of redundant features on the model, and improving the input data quality of the prediction model. For example, the time series analysis of the lagged correlation between the Yangtze River water level and the underground aquifer water level can be used to accurately judge the changing law of the groundwater level, which significantly improves the scientificity and pertinence of the data. In addition, the z-score standardization method eliminates the deviation of various eigenvalues ​​caused by scale differences, ensuring the effective learning of data features by the model. This comprehensive and accurate data acquisition and processing method has laid a solid foundation for high-precision modeling of groundwater level prediction in foundation pit construction, and has broken through the limitations of traditional methods in data collection and feature selection.

[0053] The present invention adopts the support vector machine (SVM) model combined with the radial basis kernel function (RBF kernel) to model the multi-source nonlinear relationship, which can effectively capture the complex coupling relationship between groundwater level and factors such as rainfall, soil permeability, and the water level of the Yangtze River. The penalty parameter CC and the kernel parameter γ\gammaγ are automatically tuned by the Bayesian optimization algorithm. Combined with the cross-validation technology and the L2 regularization method, the model achieves high-precision prediction of groundwater level changes under finite sample conditions. In addition, the design of the objective function comprehensively considers the balance between model complexity and misclassification, so that the model has a strong generalization ability while dealing with nonlinear problems. Especially in the complex geological conditions and construction environment of the floodplain area, the accurate prediction ability of the SVM model can provide a reliable basis for risk prediction in foundation pit excavation. Compared with traditional statistical methods, this modeling method not only improves the adaptability to nonlinear multivariate problems, but also significantly improves the prediction performance under data scarcity.

[0054] The real-time monitoring and feedback optimization mechanism of the present invention realizes the adaptive adjustment of the groundwater level prediction model in a dynamic construction environment. By real-time monitoring of data such as soil moisture content, rainfall, foundation pit depth, and error comparison with the model prediction value, when the error exceeds the preset range, the new data is promptly incorporated into the model training set, and the parameter configuration of the support vector machine is re-optimized to adapt the model to changes in field conditions. This circular feedback mechanism ensures that the prediction model can still maintain high-precision prediction capabilities in the face of emergencies (such as extreme rainfall or drastic fluctuations in the water level of the Yangtze River). In addition, the flexible arrangement of the monitoring system (such as groundwater level monitoring wells, rain gauges, and soil moisture sensors, etc.) enhances the timeliness and coverage of data collection, and provides high-quality data support for feedback optimization. This real-time adjustment capability not only ensures construction safety, but also significantly reduces the risks of engineering disasters such as water gushing and slope instability caused by inaccurate groundwater level prediction, reflecting the economy and reliability of this method in practical applications.

[0055] The above description is only by way of illustration of certain exemplary embodiments of the present invention. It is undoubted that those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A method for dynamically predicting groundwater level in floodplain foundation pit excavation, characterized in that: The following steps are involved: Collect multi-source data related to groundwater levels within the scope of the project, including soil moisture content, stratum permeability coefficient, rainfall, foundation pit excavation depth, foundation pit dewatering plan, foundation pit drainage measures; as well as historical groundwater levels, underground aquifer data and Yangtze River water level change information; Analyze the hydraulic connection between water level and underground aquifer, calculate the correlation coefficient at different lag times based on time series data, and determine the influencing factors of groundwater level; use data with strong correlation as input features, and remove weak correlation or redundant features. Then, perform z-score standardization on different data features to ensure the consistency of data features at different scales and optimize the input quality of subsequent models; The radial basis kernel function is used to construct a support vector machine model (SVM model), and the Bayesian optimization algorithm is used to search and determine the optimal combination of the model's penalty parameters and kernel parameters to enhance the model's learning ability. The normalized data is divided into a training set and a test set, and the SVM model is trained based on the training set. The model performance is evaluated through the cross-validation method, and the model is repeatedly optimized to improve the prediction accuracy. The test set data is input into the optimized SVM model to predict the dynamic changes of the groundwater level during the foundation pit excavation process; the prediction results include the height values ​​of the groundwater level at different time periods. The accuracy and reliability of the model are evaluated by calculating the error between the predicted value output by the model and the actual monitoring value to ensure that the prediction results can meet the on-site construction needs; During the excavation process, the groundwater level is monitored in real time, and the input data of soil, hydrological and climatic conditions within the project scope are continuously updated; The latest data is input into the SVM model. When there is a significant error between the monitoring data and the predicted value, the real-time monitoring data is used to perform feedback optimization on the SVM model. The adaptability of the model is improved by adjusting the input features or model parameters to ensure that the prediction results are dynamically adjusted to fit the actual situation.

2. The method for dynamic prediction of groundwater level in floodplain foundation pit excavation according to claim 1 is characterized in that: In the data collection step, the hydraulic connection between the Yangtze River water level and the underground aquifer water level was studied through time series analysis, and the lag time of the underground aquifer change lagging behind the Yangtze River water level change was analyzed by correlation calculation, and the correlation coefficients of different lag times were calculated respectively; if the lag time correlation is strong, the underground aquifer data is used as input, otherwise a comprehensive analysis is carried out in combination with the Yangtze River water level data.

3. The method for dynamic prediction of groundwater level in floodplain foundation pit excavation according to claim 1, characterized in that: In the data normalization step, the z-score standardization method is used to scale the eigenvalues ​​of different types of data. The standardization formula is: Among them, x is the original data value, μ is the mean of the feature, and σ is the standard deviation of the feature. The standardization process eliminates the scale differences between the data features and improves the prediction accuracy of the model.

4. The method for dynamic prediction of groundwater level in floodplain foundation pit excavation according to claim 1, characterized in that: In the construction step of the support vector machine model, the radial basis kernel function is used as the kernel function, and its form is defined as: K(x i , x j )=exp(-γ·‖x i -x j ‖ 2 ), where x i and x j are two sample points in the input space, ‖x i -x j ‖ 2 represents the Euclidean distance between two points, γ is the kernel parameter, and the kernel parameter γ and penalty parameter are adjusted through the Bayesian optimization algorithm to achieve the optimal parameter selection.

5. The method for dynamic prediction of groundwater level in floodplain foundation pit excavation according to claim 1, characterized in that: In the model optimization step, when optimizing the support vector machine model, the L2 regularization method is used to improve the robustness of the model. The optimization target formula is: Among them, ‖w‖ 2 is the model complexity, C is the regularization parameter, ξ i is a slack variable, i is the index of the data, used to represent the i-th sample, and n is the total number of samples, indicating that there are a total of n samples in the data set.

6. The method for dynamic prediction of groundwater level in floodplain foundation pit excavation according to claim 1, characterized in that: In the step of predicting the dynamic changes of groundwater levels, the optimized support vector machine model is used, and the soil moisture content, stratum permeability coefficient, rainfall data, foundation pit excavation depth, foundation pit dewatering plan, drainage measures, historical groundwater level and underground aquifer data within the newly collected project scope are taken as input. The predicted value is used to represent the dynamic change trend of the groundwater level, and the prediction accuracy of the model output result is evaluated by the mean square error.

7. The method for dynamic prediction of groundwater level in floodplain foundation pit excavation according to claim 1, characterized in that: In the real-time monitoring and feedback optimization step, when the error between the real-time monitored groundwater level data and the model prediction value exceeds the preset range, the updated soil moisture content, formation permeability coefficient, rainfall, foundation pit depth and precipitation scheme data are re-collected to form a new test set and optimize the support vector machine model to reduce the prediction error.

8. The method for dynamic prediction of groundwater level in floodplain foundation pit excavation according to claim 1, characterized in that: Real-time monitoring data is automatically collected through a sensor network, including groundwater level monitoring wells, rain gauges and soil moisture monitoring equipment installed within the foundation pit. The collected data is transmitted to the prediction system in real time as input to the feedback optimization model to improve the dynamic adaptability of groundwater level prediction.

Images