A method, device, equipment and medium for predicting earth-rock dam construction material parameters

Abstract

The application discloses a kind of earth-rock dam dam building material parameter prediction method, device, equipment and medium, it is related to water conservancy and hydropower engineering and artificial intelligence cross field, method includes: obtaining original rock mechanics parameters and dam building material physical characteristic parameters;Abnormal value detection, missing value filling and normalization preprocessing are carried out, standardization data are generated and fixed statistical value is solidified;Standardization data are input into preset parameter prediction model, and the normalized prediction value of 12 Duncan-Zhang constitutive model core parameters is output, and fixed statistical value is called to obtain original engineering value by de-normalization;Based on original engineering value, perform differentiated explainability analysis and performance evaluation.The application realizes the efficient, high-precision mapping of dam building material parameters through double-path parallel architecture and full-link standardization data flow, significantly reduces the dependence on traditional indoor test, and improves the explainability and reliability of engineering decision.

Description

Technical Field

[0001] This application relates to the interdisciplinary field of water conservancy and hydropower engineering and artificial intelligence, and in particular to a method, device, equipment and medium for predicting the parameters of earth-rock dam construction materials. Background Technology

[0002] Earth-rock dams are one of the most widely used dam types in water conservancy and hydropower projects. The mechanical performance parameters of their dam construction materials are key bases for dam structural design, construction control, and operational safety assessment. Among them, the Duncan-Chang constitutive model has been widely used in the nonlinear finite element analysis of earth-rock dams because it can well characterize the nonlinear stress-strain characteristics of soil and rock materials. The accuracy of its parameter values ​​directly affects the reliability of the calculation results of dam deformation, stress, and stability.

[0003] Currently, the acquisition of constitutive parameters for dam construction materials mainly relies on two traditional methods: one is the empirical formula estimation method, which estimates macroscopic mechanical parameters through original rock mechanical indices or material physical indices; the other is the laboratory test method, which directly obtains material strength and deformation parameters through triaxial compression tests, direct shear tests, etc. In recent years, with the development of artificial intelligence technology, machine learning methods and large language model technology have been gradually applied to the field of geotechnical engineering parameter prediction, providing new technical approaches to improve prediction efficiency.

[0004] However, existing technologies still fall short of meeting the demands for rapid, precise, and intelligent design in high earth-rock dam projects, mainly in the following aspects: First, indoor testing is inefficient and costly. Obtaining the complete set of parameters for the Duncan-Chang constitutive model requires multiple sets of triaxial parallel tests. Each test has a long cycle and high cost, making it difficult to achieve full coverage of a wide range of material sources and limiting the engineering representativeness of the parameters. Second, existing machine learning methods have shortcomings in engineering applications. Most solutions use a single model, which has insufficient generalization ability under small sample and highly discrete data in geotechnical engineering. Moreover, they mostly only predict a few parameters and lack effective interpretability analysis, making it difficult to directly support engineering decisions. Third, the engineering adaptability of local knowledge bases for large language models is insufficient. Existing local knowledge base solutions based on large models such as DeepSeek do not perform standardized preprocessing for the characteristics of geotechnical engineering parameters, which easily leads to problems such as inaccurate matching of similar cases and insufficient prediction accuracy. Furthermore, they lack standardized performance evaluation and result traceability mechanisms, which cannot meet the requirements of engineering design.

[0005] To address the problems existing in the prior art, this application proposes an intelligent prediction method for earth-rock dam construction material parameters based on multiple machine learning algorithms and the DeepSeek local knowledge base. It sets up two parallel prediction paths adapted to different engineering scenarios. The method achieves high-precision mapping from the original rock and dam construction material basic parameters to the constitutive parameters used in the design through a multi-machine learning algorithm model pool, and also achieves lightweight and rapid prediction through the DeepSeek local knowledge base. This improves parameter prediction efficiency and engineering adaptability, reduces reliance on traditional indoor tests, and provides comprehensive quantitative support for engineering decision-making through differentiated interpretable analysis methods. Summary of the Invention

[0006] The purpose of this application is to provide a method, device, equipment and medium for predicting the parameters of earth-rock dam construction materials, which can achieve high-precision prediction from basic parameters to 12 core parameters of the Duncan-Chang constitutive model, supporting the efficient and high-quality design of earth-rock dams.

[0007] To achieve the above objectives, this application provides the following solution: Firstly, this application provides a method for predicting the parameters of earth-rock dam construction materials, including: The original rock mechanical parameters and dam material physical property parameters of the earth-rock dam to be predicted are obtained, an initial input dataset is constructed, and the core parameters of the Duncan-Chang constitutive model of the earth-rock dam material are used as the target to be predicted. The initial input dataset is preprocessed to generate standardized preprocessed data, and the fixed statistical values ​​of each parameter generated during the preprocessing process are recorded and solidified simultaneously. The standardized preprocessed data is input into a preset parameter prediction model. The parameter prediction model performs forward inference calculation or vector similarity matching inference based on the input dataset, outputs the normalized prediction value of the target to be predicted, and calls the fixed statistical value to perform denormalization processing on the normalized prediction value to obtain the original engineering value prediction result. The parameter prediction model includes at least one of the optimal prediction model of multiple machine learning algorithms and the prediction model of DeepSeek local knowledge base. Based on the original engineering value prediction results, differentiated interpretability analysis and performance quantification evaluation are performed according to the type of the selected parameter prediction model, and a standardized analysis report is generated.

[0008] Optionally, the original rock mechanical parameters include saturated compressive strength, cohesion, and internal friction angle; the dam construction material physical property parameters include characteristic particle size D60, characteristic particle size D30, characteristic particle size D10, and dry density; the target to be predicted specifically includes: tangent modulus coefficient, tangent modulus exponent, failure ratio, bulk modulus coefficient, bulk modulus exponent, initial internal friction angle in saturated state, change of internal friction angle in saturated state with confining pressure, initial internal friction angle in air-dried state, change of internal friction angle in air-dried state with confining pressure, unloading-reloading modulus coefficient, shear modulus coefficient, and shear modulus exponent.

[0009] Optionally, preprocessing the initial input dataset includes: sequentially performing outlier detection and correction, missing value imputation, and maximum-minimum normalization, specifically including: Outlier data is identified using the Z-score standard score method. When |Z|>3, it is judged as outlier data and corrected using the median of the corresponding parameters during the training phase. Missing data were filled using the mean imputation method; The corrected and padded data are standardized to the [0,1] interval using the maximum-minimum normalization formula to obtain the standardized preprocessed data; the fixed statistical values ​​include the mean, standard deviation, maximum value, and minimum value of each input parameter and 12 prediction labels determined during the model training phase, which are used to ensure that the data distribution in the prediction phase is completely consistent with that in the training phase.

[0010] Optionally, when the parameter prediction model is an optimal prediction model using multiple machine learning algorithms, the optimal prediction model using multiple machine learning algorithms is obtained by pre-training and solidifying it through the following steps: Build a basic model pool containing various machine learning algorithms; Preprocessing is performed on the input features and predicted labels of the engineering sample dataset to obtain a standardized training dataset; The standardized training dataset is split into a training set, a validation set, and a test set using a hierarchical random partitioning method. The training set is used to complete the independent training of each algorithm model in the model pool, and the validation set is used to optimize the hyperparameters of each model. Using the coefficient of determination R 2 The mean squared error (MSE) is used as a quantitative evaluation index. The model pool is evaluated through cross-validation, and the single model with the highest comprehensive prediction accuracy of 12 parameters is selected as the optimal prediction model of the multi-machine learning algorithm. Its model structure and weights are then fixed.

[0011] Optionally, when the parameter prediction model is a DeepSeek local knowledge base prediction model, the DeepSeek local knowledge base prediction model is pre-built and configured through the following steps: Set up a local knowledge base runtime environment for the large language model and complete the API interface access and connectivity verification of the DeepSeek-R1 large language model; We collected sample data of earth-rock dam projects, and performed structured text conversion on the original rock mechanical parameters, physical property parameters of dam construction materials, and the measured values ​​of 12 corresponding Duncan-Chang constitutive model parameters of the samples to build a standardized engineering sample corpus. Vectorization embedding and index construction are performed on the standardized engineering sample corpus to generate a local knowledge base for earth-rock dam construction material parameters. The standardized preprocessed data is converted into standardized inference instructions that conform to the input specifications of large language models and input into the DeepSeek-R1 model. The model retrieves similar engineering samples from the specialized local knowledge base based on vector similarity to complete matching inference and outputs the normalized prediction value.

[0012] Optionally, the machine learning algorithms included in the base model pool include: Artificial Neural Network (ANN), K-Nearest Neighbors Regression (KNN), Gradient Boosting Tree (GBDT), Random Forest (RF), XGBoost, and Support Vector Regression (SVR).

[0013] Optionally, based on the original engineering value prediction results, differentiated interpretability analysis and performance quantification evaluation are performed according to the type of the selected parameter prediction model, generating a standardized analysis report, specifically including: When the parameter prediction model is the optimal prediction model of multiple machine learning algorithms, the SHAP value is used to analyze the contribution of each input feature to the prediction result of the original engineering value, quantify the contribution of each input feature and generate a feature contribution visualization chart. When the parameter prediction model is a DeepSeek local knowledge base prediction model, if there are measured values ​​for the corresponding parameters, then the coefficient of determination R is used. 2 The accuracy of the prediction results is quantitatively evaluated using the mean squared error (MSE). If no measured value exists, the source of the similar engineering sample reference, the sample matching degree value, and the dispersion of the matching sample parameters corresponding to the prediction result are output, so as to realize the full traceability of the prediction process.

[0014] Secondly, this application provides a device for predicting the parameters of earth-rock dam construction materials, comprising: The dataset construction module is used to obtain the original rock mechanical parameters and dam material physical property parameters of the earth-rock dam to be predicted, construct the initial input dataset, and use the core parameters of the Duncan-Chang constitutive model of the earth-rock dam construction material as the target to be predicted. The preprocessing module is used to preprocess the initial input dataset, generate standardized preprocessed data, and synchronously record and solidify the fixed statistical values ​​of each parameter generated during the preprocessing process. The original engineering value prediction result determination module is used to input the standardized preprocessed data into a preset parameter prediction model. The parameter prediction model performs forward inference calculation or vector similarity matching inference based on the input dataset, outputs the normalized prediction value of the target to be predicted, and calls the fixed statistical value to perform inverse normalization processing on the normalized prediction value to obtain the original engineering value prediction result. The parameter prediction model includes at least one of the optimal prediction model of multiple machine learning algorithms and the prediction model of DeepSeek local knowledge base. The standardized analysis report generation module is used to perform differentiated interpretability analysis and performance quantification evaluation based on the original engineering value prediction results and the type of the selected parameter prediction model, and generate a standardized analysis report.

[0015] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method for predicting the parameters of earth-rock dam construction materials as described above.

[0016] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method for predicting the parameters of earth-rock dam construction materials as described above.

[0017] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the method for predicting the parameters of earth-rock dam construction materials as described above.

[0018] According to the specific embodiments provided in this application, this application has the following technical effects: This application provides a method, device, equipment, and medium for predicting the parameters of earth-rock dam construction materials. By setting up two parallel parameter prediction paths—one using multiple machine learning algorithms and the other using a DeepSeek local knowledge base—an intelligent prediction system for earth-rock dam construction material parameters adapted to the design process of hydraulic engineering is constructed. This solves the problems of low efficiency, insufficient accuracy, limited scenario adaptability, and weak engineering application support in existing technologies for acquiring dam construction material parameters. A model pool is constructed using multiple machine learning algorithms adapted to the small sample characteristics of geotechnical engineering. The optimal single model is selected through cross-validation using multi-dimensional quantitative evaluation indicators, addressing the issues of insufficient generalization ability of traditional single models and large prediction deviations of traditional empirical formulas. The prediction accuracy meets the design requirements of earth-rock dam engineering. The end-to-end parameter prediction process can be completed in minutes, improving prediction efficiency by over 100 times compared to the 2-3 week cycle of traditional indoor triaxial tests. This significantly reduces the workload of indoor tests, lowering the time and economic costs of engineering design. Simultaneously, it can output 12 strength and variation parameters of the Duncan-Chang constitutive model. The core parameters cover the parameter input requirements for nonlinear finite element simulation analysis of dam bodies, eliminating the need for independent step-by-step predictions and adapting to the conventional design process of water conservancy projects. For the DeepSeek local knowledge base prediction scheme, a standardized preprocessing system fully unified with the machine learning model has been established. Normalization processing eliminates the dimensional differences and magnitude spans of geotechnical engineering parameters, improving the matching accuracy of similar engineering samples. At the same time, a standardized performance quantification evaluation system and a prediction process traceability mechanism have been established, making the prediction results of the large model have engineering application value and fully adapting to the needs of lightweight scenarios such as rapid pre-research and small-sample reconnaissance. In addition, differentiated interpretability analysis methods are adopted for the two types of prediction models. The machine learning scheme uses SHAP values ​​to quantify the contribution of each input feature to the prediction results, while the DeepSeek scheme achieves the traceability of the prediction process through sample reference tracing. This provides full-dimensional quantitative support for engineering technicians to evaluate the reliability of prediction results, conduct material source selection and parameter sensitivity analysis, and significantly enhances the practical application value of the scheme in engineering design decisions. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 A flowchart illustrating a method for predicting the parameters of earth-rock dam construction materials, provided as an embodiment of this application; Figure 2(a) is a scatter plot of the actual value – predicted value of the SVR model for the prediction of the initial internal friction angle Φ0 (saturation) in saturation state provided by an embodiment of this application; Figure 2(b) shows the coefficient of determination R of six types of models for predicting the initial internal friction angle Φ0 (saturation) in a saturated state according to an embodiment of this application. 2 Compare the bar charts; Figure 2(c) is a bar chart comparing the mean square error (MSE) of six types of models for predicting the initial internal friction angle Φ0 (saturation) in a saturated state according to an embodiment of this application. Figure 2(d) is a residual diagram of the SVR model for predicting the initial internal friction angle Φ0 (saturation) in a saturated state according to an embodiment of this application; Figure 3 A schematic diagram of the functional modules of a device for predicting the parameters of earth-rock dam construction materials, provided in another embodiment of this application; Figure 4 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0021] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0022] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0023] In one exemplary embodiment, such as Figure 1 As shown, a method for predicting the parameters of earth-rock dam construction materials is provided. This method is executed by computer equipment, specifically by a terminal or server alone, or by both a terminal and a server. In this embodiment, the method includes the following steps: Step 101: Obtain the original rock mechanical parameters and dam material physical property parameters of the earth-rock dam to be predicted, construct the initial input dataset, and use the core parameters of the Duncan-Chang constitutive model of the earth-rock dam material as the target to be predicted.

[0024] The original rock mechanical parameters include saturated uniaxial compressive strength, cohesion c, and internal friction angle. There are a total of 3 items; the physical property parameters of the dam construction material include characteristic particle size D60, characteristic particle size D30, characteristic particle size D10 and dry density, totaling 4 items.

[0025] The cohesive force c and the internal friction angle These are shear strength parameters determined based on the Mohr–Coulomb strength criterion for soil and rock masses, where: cohesion *c* characterizes the cohesive force between material particles; and internal friction angle... The shear resistance is characterized by interparticle friction and interlocking; these parameters can be obtained through indoor triaxial compression tests or direct shear tests.

[0026] The dam material parameters to be predicted specifically include: tangent modulus coefficient K, tangent modulus exponent n, failure ratio Rf, bulk modulus coefficient Kb, bulk modulus exponent m, initial internal friction angle Φ0 (saturated), change in internal friction angle with confining pressure ΔΦ (saturated), initial internal friction angle Φ0 (air-dried), change in internal friction angle with confining pressure ΔΦ (air-dried), unloading-reloading modulus coefficient D, shear modulus coefficient G, and shear modulus exponent F, totaling 12 core parameters of the Duncan-Chang constitutive model, covering the parameter input requirements for nonlinear finite element simulation analysis of earth-rock dams.

[0027] Step 102: Preprocess the initial input dataset to generate standardized preprocessed data, and simultaneously record and solidify the fixed statistical values ​​of each parameter generated during the preprocessing process.

[0028] The preprocessing includes outlier detection and correction, missing value imputation, and data normalization. The specific process is as follows: 1. Outlier Detection and Correction: Outlier data is identified using the Z-score standard score method. The Z-score calculation formula is as follows: ,in x For input sample values, μ This represents the mean of the parameters stored during the pre-stored training phase. σ The standard deviation of the parameters corresponding to the pre-stored training phase is used. When |Z|>3, it is judged as outlier data. Outlier data is corrected using the median of the parameters corresponding to the training phase. 2. Missing value imputation: The mean imputation method is used to fill in the missing values ​​in the input data with the fixed mean of the input parameters corresponding to the model training stage, so as to ensure the integrity of the input data. 3. Data Normalization: The corrected input data is standardized to the [0,1] interval using the maximum-minimum normalization formula to eliminate differences in parameter dimensions. The maximum-minimum normalization formula is as follows: ,in For standardized data, x The original data, , These represent the maximum and minimum values ​​of the corresponding parameters; the prediction phase uses the inverse normalization formula. The predicted values ​​are restored to the original engineering values, with the signs and normalization formulas being completely consistent. The entire preprocessing process uses rules and fixed statistical values ​​identical to those used in the model training / knowledge base construction phases to ensure that the input data distribution is consistent with the training set, guaranteeing the accuracy of the model's prediction results.

[0029] Step 103: Input the standardized preprocessed data into a preset parameter prediction model. The parameter prediction model performs forward inference calculation or vector similarity matching inference based on the input dataset, outputs the normalized prediction value of the target to be predicted, and calls the fixed statistical value to perform inverse normalization processing on the normalized prediction value to obtain the original engineering value prediction result. The parameter prediction model includes at least one of the optimal prediction model of multiple machine learning algorithms and the prediction model of DeepSeek local knowledge base.

[0030] In practice, users can choose the appropriate path to perform prediction based on the application scenario: Path a: For high-precision scenarios such as construction drawing design and dam stability analysis, the optimal prediction model using multiple machine learning algorithms is selected to perform predictions. The specific process is as follows: 1. Forward inference calculation: The standardized preprocessed data obtained in step 102 is input into the pre-stored optimal prediction model after training. The model performs forward inference calculation through the weight parameters after training and outputs the normalized prediction values ​​of the 12 core parameters of the Duncan-Zhang constitutive model. 2. Inverse Normalization Restoration: The system automatically calls the maximum and minimum fixed statistical values ​​of the 12 predicted labels pre-stored during the training phase, and restores the normalized predicted values ​​to the original engineering values ​​with the same distribution as the training set through the inverse normalization formula; 3. Results Output: The final output directly provides the original engineering values ​​of 12 core parameters of the Duncan-Chang constitutive model that can be used for nonlinear finite element simulation analysis of earth-rock dams.

[0031] Path b: For lightweight scenarios such as project pre-feasibility study, material source survey, and rapid scheme comparison, the DeepSeek local knowledge base prediction model is selected. The standardized preprocessed data is input into the DeepSeek local knowledge base prediction model to complete similar sample matching reasoning and inverse normalization, and output the original engineering values ​​of dam construction material parameters. For the core dam section of important projects, two paths can be selected simultaneously to perform prediction, and the reliability of parameter values ​​can be further improved through cross-validation of the results.

[0032] The detailed training method for the optimal prediction model using multiple machine learning algorithms described in this application includes the following steps 301 to 305: Step 301: Construction of the Basic Model Pool. Based on the Python open-source machine learning ecosystem, and using mainstream frameworks such as Scikit-learn, XGBoost, and Keras, a basic model pool adapted to the small sample size and high dispersion characteristics of geotechnical engineering is built. The model pool includes six types of machine learning algorithms: Artificial Neural Network (ANN), K-Nearest Neighbor Regression (KNN), Gradient Boosting Tree (GBDT), Random Forest (RF), XGBoost, and Support Vector Regression (SVR). These algorithms are all mainstream adaptation algorithms for geotechnical engineering parameter regression prediction scenarios and can cover the fitting needs of various data characteristics such as linearity, nonlinearity, high dimensionality, and small sample size.

[0033] Step 302: Dataset Preprocessing and Splitting. Valid engineering sample data from existing and under-construction earth-rock dam projects in China are collected to construct an original dataset. The dataset input features include three original rock mechanical parameters and four dam construction material physical property parameters. The dataset labels are the measured true values ​​of the 12 core parameters of the Duncan-Chang constitutive model corresponding to the samples. For the input features and predicted labels of the original dataset, the preprocessing operations are performed exactly the same as in Step 202 of Example 1, sequentially completing outlier detection and correction, missing value imputation, and max-min normalization to obtain a standardized training dataset. A hierarchical random partitioning method is used to divide the standardized training dataset into a training set, a validation set, and a test set in a 7:2:1 ratio. The training set is used for model fitting training, the validation set is used for hyperparameter optimization and overfitting suppression, and the test set is used for independent verification of the model's generalization performance, ensuring that the trained model possesses excellent engineering generalization ability.

[0034] Step 303: Independent Training and Hyperparameter Optimization of Single Models. This step is the core of model training. Independent training is performed on the six basic models in the model pool. Considering the characteristics of small sample sizes and high dispersion in geotechnical engineering datasets, suitable training strategies and hyperparameter optimization schemes are set for each model type: 1. Training of tree models (GBDT, Random Forest, XGBoost): Using mean squared error (MSE) as the loss function, Bayesian optimization (TPE algorithm) is used to optimize core hyperparameters such as maximum depth of decision tree, learning rate, minimum number of samples in leaf nodes, and subsampling ratio. Five-fold cross-validation is used during training to reduce the risk of overfitting in small sample scenarios. 2. Support Vector Regression (SVR) Model Training: The Radial Basis Function (RBF) is selected to adapt to nonlinear fitting. The mean squared error (MSE) is used as the loss function. The grid search algorithm is used to globally optimize the three core hyperparameters: the penalty coefficient C, the kernel function bandwidth γ, and the insensitive loss coefficient ε, to ensure that the model balances fitting accuracy and generalization ability on small sample datasets. 3. K-Nearest Neighbors (KNN) Model Training: The two core hyperparameters, K-nearest neighbors and distance metric, are optimized. The Manhattan distance, which is adapted to the dimensional characteristics of multi-dimensional soil and rock parameters, is selected as the distance metric function. The optimization range of the number of nearest neighbors K is set to 3-15 to ensure the prediction stability of the model in small sample scenarios. 4. Artificial Neural Network (ANN) Model Training: A 3-layer fully connected forward propagation network structure was constructed, with 7 neurons in the input layer (corresponding to 7 input features), 14-28 neurons in the hidden layer, and 12 neurons in the output layer (corresponding to 12 parameters to be predicted). The hidden layer used the ReLU activation function, and the output layer used the linear activation function. The Adam optimizer was selected, with the mean squared error (MSE) as the loss function. The batch size was set to 8, and the maximum number of training epochs was set to 300. During training, an early stopping mechanism (training was terminated when the validation set loss did not decrease for 15 consecutive iterations) and L2 regularization were added to effectively suppress the risk of model overfitting in small sample scenarios.

[0035] During all model training processes, the changes in validation set loss are monitored in real time to ensure model convergence and stability. After training is completed, the optimal structure and training weights of each model class are saved.

[0036] Step 304: Optimal Model Selection and Performance Validation. The coefficient of determination R0 is used. 2 Ri and mean squared error (MSE), two industry-standard core metrics, are used to quantitatively evaluate the generalization performance of six types of trained models. 2 The closer the value is to 1, the smaller the MSE value, indicating better prediction accuracy and generalization ability of the model; the average R-value of the 12 Duncan-Chang constitutive model parameters is used as the metric. 2 The core selection criteria are the highest and the lowest average MSE. The single model with the best comprehensive prediction accuracy among the six types of models is selected as the optimal prediction model for the multi-machine learning algorithm in this application.

[0037] Taking the prediction effect of the initial internal friction angle Φ0 (saturation) in the saturated state as an example, the performance of each model on the independent test set is shown in Figures 2(a)-2(d): Among the six types of models, the test set determination coefficient R of the Support Vector Regression (SVR) model is... 2 The mean squared error (MSE) reached 0.8471, the highest among all models (corresponding to Figure 2(b)); the mean squared error (MSE) was only 1.0553, the lowest among all models (corresponding to Figure 2(c)), indicating the best overall prediction performance. From the scatter plot of the true-to-predicted values ​​of the SVR model (Figure 2(a)), it can be seen that the sample points are basically distributed along the 1:1 fitting line, indicating a high degree of fit between the predicted and actual values. From the residual plot of the SVR model (Figure 2(d)), it can be seen that the sample residuals are basically randomly distributed near the 0 line, with no obvious systematic bias, verifying the model's fitting effectiveness and generalization stability.

[0038] Step 305: Model and Statistical Consolidation. Generate and save a complete model performance evaluation report, including the R values ​​for each of the 12 dam construction material parameters. 2 The system includes the MSE metric, performance comparison results of different models, and prediction effect diagrams of core parameters. It also preserves the complete structure and training weights of the optimal prediction model, as well as the fixed statistical values ​​of the mean, standard deviation, maximum value, and minimum value corresponding to the 7 input features and 12 prediction labels determined during the training phase. This ensures that the prediction phase can use the same preprocessing rules and inverse normalization parameters as the training phase, guaranteeing the accuracy of the prediction results.

[0039] The method for building and predicting the DeepSeek local knowledge base prediction model described in this application is executed by a computer device and includes the following steps 401 to 404: Step 401: Prediction Environment Configuration: Set up a local knowledge base runtime environment for the large language model, and complete the API interface access and connectivity verification for the DeepSeek-R1 large language model. Build the runtime environment using an open-source local knowledge base framework, configure the API interface for the DeepSeek-R1 large language model, connect the corresponding embedded model, and perform connectivity tests to ensure the model can respond normally to inference requests, providing a stable operating foundation for subsequent corpus construction and parameter prediction.

[0040] Step 402: Construction of a Standardized Engineering Sample Corpus: Collect sample data of earth-rock dam projects, complete the structured text conversion of the samples, and construct a standardized engineering sample corpus. Using the same engineering sample data as in Example 2, perform structured text conversion on the original rock mechanics parameters, dam material physical property parameters, and the corresponding measured values ​​of the 12 core parameters of the Duncan-Chang constitutive model of the training set samples. Organize the data using a unified standardized format to ensure a one-to-one correspondence between input features and output labels, thus constructing a standardized engineering sample corpus.

[0041] Step 403: Construction of a Specialized Local Knowledge Base: Perform vectorized embedding processing on the sample corpus to complete the index construction and storage of a specialized knowledge base for earth-rock dam construction material parameters. Select an appropriate embedding model, perform vectorized embedding processing, vector slicing, and index construction on the standardized engineering sample corpus to generate a vector database of the specialized knowledge base for earth-rock dam construction material parameters; set the retrieval mode to precise vector matching to ensure that the engineering samples with the highest similarity can be quickly retrieved during the inference stage.

[0042] Step 404: Parameter Prediction Inference and Result Output: Input standardized preprocessed data, complete similar engineering sample matching inference and denormalization, and output the original engineering values ​​of 12 Duncan-Chang constitutive model parameters. Convert the standardized preprocessed data to be predicted into standardized inference instructions that conform to the input specifications of the large language model, and input them into the DeepSeek-R1 large language model; the model uses vector similarity retrieval, calls the engineering sample with the highest similarity in the local knowledge base as a reference, completes parameter matching inference, and outputs the normalized predicted values ​​of 12 core parameters of the Duncan-Chang constitutive model; then, through the denormalization formula, the predicted values ​​are restored to the original engineering values ​​that can be directly used for engineering design, and the final prediction result is output. At the same time, the source of the similar engineering sample reference corresponding to the prediction result can be output synchronously to realize the traceability of the prediction process.

[0043] Step 104: Based on the original engineering value prediction results, perform differentiated interpretability analysis and performance quantification evaluation according to the type of the selected parameter prediction model, and generate a standardized analysis report.

[0044] For the prediction results of the optimal prediction model of multiple machine learning algorithms, the contribution characteristics of each input feature to the prediction results are analyzed by SHAP (SHapley Additive ex Planations) value, the degree of contribution of each input feature to the prediction results is quantified, and visual analysis reports such as feature contribution ranking bar charts and bee colony diagrams are generated, providing quantitative reference for engineering technicians to evaluate the reliability of prediction results and carry out material source selection and parameter sensitivity analysis. For the prediction results of the DeepSeek local knowledge base prediction model, a differentiated evaluation method is adopted: when users obtain the measured values ​​of the Duncan-Chang model parameters corresponding to the prediction samples through indoor experiments, the measured values ​​can be imported into the system, and the coefficient of determination R is used. 2 The mean square error (MSE) is used to quantitatively evaluate the accuracy of the prediction results for this batch. When there are no measured values, the source of similar engineering samples corresponding to the prediction results, the sample matching degree value, and the dispersion of matching sample parameters are output to realize full traceability of the prediction process and provide a quantitative reference for engineering technicians to evaluate the reliability of the results. Among them, the coefficient of determination R 2 The calculation formula is: The formula for calculating the mean squared error (MSE) is: In the above formula, n Indicates the number of samples. For the first i The true value of each sample For the first i The predicted value for each sample, This is the mean of the true values ​​of the sample.

[0045] Finally, a standardized interpretability analysis report of the prediction results is generated and output simultaneously with the parameter prediction results.

[0046] Based on the same inventive concept, this application also provides an earth-rock dam construction material parameter prediction device for implementing the above-mentioned method for predicting earth-rock dam construction material parameters. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of the earth-rock dam construction material parameter prediction device provided below can be found in the limitations of the earth-rock dam construction material parameter prediction method described above, and will not be repeated here.

[0047] In one exemplary embodiment, such as Figure 3 As shown, a device for predicting the parameters of earth-rock dam construction materials is provided, comprising: The dataset construction module is used to obtain the original rock mechanical parameters and dam material physical property parameters of the earth-rock dam to be predicted, construct the initial input dataset, and use the core parameters of the Duncan-Chang constitutive model of the earth-rock dam construction material as the target to be predicted. The preprocessing module is used to preprocess the initial input dataset, generate standardized preprocessed data, and synchronously record and solidify the fixed statistical values ​​of each parameter generated during the preprocessing process. The original engineering value prediction result determination module is used to input the standardized preprocessed data into a preset parameter prediction model. The parameter prediction model performs forward inference calculation or vector similarity matching inference based on the input dataset, outputs the normalized prediction value of the target to be predicted, and calls the fixed statistical value to perform inverse normalization processing on the normalized prediction value to obtain the original engineering value prediction result. The parameter prediction model includes at least one of the optimal prediction model of multiple machine learning algorithms and the prediction model of DeepSeek local knowledge base. The standardized analysis report generation module is used to perform differentiated interpretability analysis and performance quantification evaluation based on the original engineering value prediction results and the type of the selected parameter prediction model, and generate a standardized analysis report.

[0048] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 4As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores predicted data for earth-rock dam construction material parameters. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a method for predicting earth-rock dam construction material parameters.

[0049] Those skilled in the art will understand that Figure 4 The structures shown are merely block diagrams of some structures related to the present application and do not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than shown in the figures, or combine certain components, or have different component arrangements. In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0050] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0051] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0052] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0053] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0054] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0055] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for predicting the parameters of earth-rock dam construction materials, characterized in that, The method for predicting the parameters of earth-rock dam construction materials includes: The original rock mechanical parameters and dam material physical property parameters of the earth-rock dam to be predicted are obtained, an initial input dataset is constructed, and the core parameters of the Duncan-Chang constitutive model of the earth-rock dam material are used as the target to be predicted. The initial input dataset is preprocessed to generate standardized preprocessed data, and the fixed statistical values ​​of each parameter generated during the preprocessing process are recorded and solidified simultaneously. The standardized preprocessed data is input into a preset parameter prediction model. The parameter prediction model performs forward inference calculation or vector similarity matching inference based on the input dataset, outputs the normalized prediction value of the target to be predicted, and calls the fixed statistical value to perform denormalization processing on the normalized prediction value to obtain the original engineering value prediction result. The parameter prediction model includes at least one of the optimal prediction model of multiple machine learning algorithms and the prediction model of DeepSeek local knowledge base. Based on the original engineering value prediction results, differentiated interpretability analysis and performance quantification evaluation are performed according to the type of the selected parameter prediction model, and a standardized analysis report is generated.

2. The method for predicting the parameters of earth-rock dam construction materials according to claim 1, characterized in that, The original rock mechanical parameters include saturated compressive strength, cohesion, and internal friction angle; the dam construction material physical properties include characteristic particle size D60, characteristic particle size D30, characteristic particle size D10, and dry density. The targets to be predicted specifically include: tangent modulus coefficient, tangent modulus exponent, failure ratio, bulk modulus coefficient, bulk modulus exponent, initial internal friction angle in saturated state, change of internal friction angle in saturated state with confining pressure, initial internal friction angle in air-dried state, change of internal friction angle in air-dried state with confining pressure, unloading-reloading modulus coefficient, shear modulus coefficient, and shear modulus exponent.

3. The method for predicting the parameters of earth-rock dam construction materials according to claim 1, characterized in that, Preprocessing the initial input dataset includes: sequentially performing outlier detection and correction, missing value imputation, and max-min normalization, specifically including: Outlier data is identified using the Z-score standard score method. When |Z|>3, it is judged as outlier data and corrected using the median of the corresponding parameters during the training phase. Missing data were filled using the mean imputation method; The corrected and padded data are standardized to the [0,1] interval using the maximum-minimum normalization formula to obtain the standardized preprocessed data; the fixed statistical values ​​include the mean, standard deviation, maximum value, and minimum value of each input parameter and 12 prediction labels determined during the model training phase, which are used to ensure that the data distribution in the prediction phase is completely consistent with that in the training phase.

4. The method for predicting the parameters of earth-rock dam construction materials according to claim 1, characterized in that, When the parameter prediction model is the optimal prediction model of multiple machine learning algorithms, the optimal prediction model of multiple machine learning algorithms is obtained by pre-training and solidifying through the following steps: Build a basic model pool containing various machine learning algorithms; Preprocessing is performed on the input features and predicted labels of the engineering sample dataset to obtain a standardized training dataset; The standardized training dataset is split into a training set, a validation set, and a test set using a hierarchical random partitioning method. The training set is used to complete the independent training of each algorithm model in the model pool, and the validation set is used to optimize the hyperparameters of each model. Using the coefficient of determination R 2 The mean squared error (MSE) is used as a quantitative evaluation index. The model pool is evaluated through cross-validation, and the single model with the highest comprehensive prediction accuracy of 12 parameters is selected as the optimal prediction model of the multi-machine learning algorithm. Its model structure and weights are then fixed.

5. The method for predicting the parameters of earth-rock dam construction materials according to claim 1, characterized in that, When the parameter prediction model is a DeepSeek local knowledge base prediction model, the DeepSeek local knowledge base prediction model is pre-built and configured through the following steps: Set up a local knowledge base runtime environment for the large language model and complete the API interface access and connectivity verification of the DeepSeek-R1 large language model; We collected sample data of earth-rock dam projects, and performed structured text conversion on the original rock mechanical parameters, physical property parameters of dam construction materials, and the measured values ​​of 12 corresponding Duncan-Chang constitutive model parameters of the samples to build a standardized engineering sample corpus. Vectorization embedding and index construction are performed on the standardized engineering sample corpus to generate a local knowledge base for earth-rock dam construction material parameters. The standardized preprocessed data is converted into standardized inference instructions that conform to the input specifications of large language models and input into the DeepSeek-R1 model. The model retrieves similar engineering samples from the specialized local knowledge base based on vector similarity to complete matching inference and outputs the normalized prediction value.

6. The method for predicting the parameters of earth-rock dam construction materials according to claim 1, characterized in that, The basic model pool includes machine learning algorithms such as Artificial Neural Network (ANN), K-Nearest Neighbors Regression (KNN), Gradient Boosting Tree (GBDT), Random Forest (RF), XGBoost, and Support Vector Regression (SVR).

7. The method for predicting the parameters of earth-rock dam construction materials according to claim 1, characterized in that, Based on the predicted results of the original engineering values, differentiated interpretability analysis and performance quantification evaluation are performed according to the type of the selected parameter prediction model, generating a standardized analysis report, specifically including: When the parameter prediction model is the optimal prediction model of multiple machine learning algorithms, the SHAP value is used to analyze the contribution of each input feature to the prediction result of the original engineering value, quantify the contribution of each input feature and generate a feature contribution visualization chart. When the parameter prediction model is a DeepSeek local knowledge base prediction model, if there are measured values ​​for the corresponding parameters, then the coefficient of determination R is used. 2 The accuracy of the prediction results is quantitatively evaluated using the mean square error (MSE). If no measured value exists, the source of the similar engineering sample reference, the sample matching degree value, and the dispersion of the matching sample parameters corresponding to the prediction result are output, so as to realize the full traceability of the prediction process.

8. A device for predicting the parameters of earth-rock dam construction materials, characterized in that, The earth-rock dam construction material parameter prediction device includes: The dataset construction module is used to obtain the original rock mechanical parameters and dam material physical property parameters of the earth-rock dam to be predicted, construct the initial input dataset, and use the core parameters of the Duncan-Chang constitutive model of the earth-rock dam construction material as the target to be predicted. The preprocessing module is used to preprocess the initial input dataset, generate standardized preprocessed data, and synchronously record and solidify the fixed statistical values ​​of each parameter generated during the preprocessing process. The original engineering value prediction result determination module is used to input the standardized preprocessed data into a preset parameter prediction model. The parameter prediction model performs forward inference calculation or vector similarity matching inference based on the input dataset, outputs the normalized prediction value of the target to be predicted, and calls the fixed statistical value to perform inverse normalization processing on the normalized prediction value to obtain the original engineering value prediction result. The parameter prediction model includes at least one of the optimal prediction model of multiple machine learning algorithms and the prediction model of DeepSeek local knowledge base. The standardized analysis report generation module is used to perform differentiated interpretability analysis and performance quantification evaluation based on the original engineering value prediction results and the type of the selected parameter prediction model, and generate a standardized analysis report.

9. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the method for predicting the parameters of earth-rock dam construction materials according to any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the method for predicting the dam construction material parameters of any one of claims 1-7.

Images