A method for coupling derivation of dam construction mechanical performance parameters of a rock-fill dam
By constructing a method for deriving the performance of rockfill dams using a hybrid machine learning model, the problems of high testing costs, long cycles, and poor parameter adaptability in rockfill dam construction have been solved. This method enables efficient and accurate acquisition of mechanical performance parameters, adapts to complex climatic environments, and improves engineering safety and design efficiency.
Patent Information
- Application Number
- CN202610318092.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-03-16
- Publication Date
- 2026-07-21
AI Technical Summary
The construction of rockfill dams faces problems such as high testing costs, long cycles, insufficient simulation of coupling relationships, and poor parameter adaptability, which makes it impossible to meet the design and construction time requirements. Furthermore, the prediction accuracy decreases under complex climatic environments, affecting safety and reliability.
A hybrid machine learning model is adopted, including a local feature extraction layer, a global coupling modeling layer, a physical constraint embedding layer, and a scenario adaptation and optimization layer. By collecting and preprocessing raw engineering data, a dam construction performance derivation model of rockfill dam is constructed to capture the coupling relationship of multiple factors and adapt parameters to the scenario.
It significantly shortens the parameter acquisition cycle, reduces costs, improves prediction accuracy, ensures engineering safety and design efficiency, adapts to complex climatic environments, and reduces experimental workload and material consumption.
Smart Images
Figure CN122433155A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of engineering big data and machine learning, and more specifically, to a coupled derivation method for the mechanical performance parameters of rockfill dams. Background Technology
[0002] Rockfill dams are a common type of dam in high dam construction. They can make full use of local materials, reduce transportation costs, adapt well to complex geological conditions, have simple construction methods and good seismic performance. They are one of the core dam types in water conservancy projects, hydropower projects and pumped storage power station projects.
[0003] However, the construction of rockfill dams faces numerous challenges. First, compared to homogeneous dams such as conventional concrete dams and roller-compacted concrete dams, the physical and mechanical properties of rockfill are more discrete and non-uniform, and are susceptible to deterioration due to various factors, directly affecting the overall safety of the rockfill dam. Second, rockfill dams are also known as "local material dams," heavily reliant on the material properties of the dam site. Large quantities of dam-building materials need to be transported from the reservoir area to laboratories for a series of performance tests. In remote areas with poor transportation and harsh site conditions, sampling and transporting tens or even hundreds of tons of soil and rock within the dam area is extremely difficult and costly. Meanwhile, conducting a series of large-scale coarse-grained soil tests is extremely time-consuming and costly, especially large-scale triaxial shear tests, which take 3 to 7 days per test. Obtaining the results of a complete parameter system, including permeability tests, compression tests, and triaxial shear tests, requires at least 40 to 60 sets of tests, taking more than 2 months, which cannot meet the time requirements for rapid design and construction of rockfill dams. The mechanical properties of coarse-grained soil are strongly coupled with factors such as climate, stone characteristics, and design parameters. If traditional methods are used, relying only on empirical formulas or linear fitting such as the Terzaghi formula and the Hazen-Williams formula, these can only describe the unidirectional influence of a single factor on a single parameter and cannot capture the nonlinear coupling law of multiple factors, resulting in generally high errors in parameter derivation. During the test, it is also impossible to specifically adapt to complex climatic environments such as high altitude and low oxygen, severe cold and freeze-thaw, and design parameter differences such as stress levels corresponding to different dam heights and gradation differences in dam construction zones. When the test results are used in complex scenarios such as high altitude and severe cold, the prediction accuracy drops significantly, resulting in significant deviations between the dam stability calculation results and the actual working conditions, which will affect its safety and reliability.
[0004] Therefore, there is an urgent need for a solution that can comprehensively address the problems of high testing costs, long testing cycles, insufficient simulation of coupling relationships, and poor parameter adaptability, in order to improve the design efficiency of rockfill dams, ensure project safety, and reduce upfront costs. Summary of the Invention
[0005] To achieve the above objectives, this application provides a coupled derivation method for the mechanical performance parameters of rockfill dams, comprising the following steps: The raw engineering data is collected, preprocessed, and a dataset for rockfill dam construction is generated. The data dimensions of the raw engineering data include: climate data, lithological data, dam height data, dam design parameters, dam construction material test data, and target mechanical performance parameters. Define the structure, optimization objective, and training rules of the hybrid machine learning model; train the hybrid machine learning model to construct a derivation model of the dam construction performance of the rockfill dam; the structure of the hybrid machine learning model includes a local feature extraction layer, a global coupling modeling layer, a physical constraint embedding layer, and a scene adaptation optimization layer; the optimization objective is to minimize the derivation error of the target mechanical performance parameters; Load the dam construction performance derivation model of the rockfill dam to achieve coupled derivation of the mechanical performance parameters of the target project.
[0006] The preprocessing of the original engineering data includes: defining constraint criteria and removing abnormal data based on these criteria; identifying missing data and supplementing it; unifying the processing of parameters from different units to achieve data standardization; data standardization and coding; extracting key coupling relationships of engineering features and generating coupling item data to be added to the dataset. The coupling term data is used to improve the model's ability to capture the coupling relationships of multiple factors. The coupling term data includes: a coupling feature of "lithology + gradation + relative density", which represents the influence of rock lithology on the particle gradation and relative density properties in the natural or fractured state; a coupling feature of "freeze-thaw cycle count + fracture rate", which reflects the influence of the number of freeze-thaw cycles on the physical and mechanical properties of the rock and the fracture rate after crushing; and a coupling feature of "dam height + dam construction zone + stress level", which reflects the influence of the designed dam height on the range of dam construction zones and the stress level of each zone of the entire dam.
[0007] Furthermore, the local feature extraction layer is implemented by constructing multiple parallel decision trees based on the RF model; the parallel decision trees include environmental-stone feature branches and design-basic experiment feature branches; The input data for the environment-stone characteristics branch are climate data, lithological data, and coupling term data, which are used to extract local correlation features such as "freeze-thaw cycle-lithology-weathering resistance" and "chemical composition-particle strength". The input data for the design-basic test feature branch are dam design parameters, dam material test data, and coupling term data, which are used to extract local coupling features such as "relative density-compaction degree-deformation characteristics" and "fracture rate-gradation-strength parameters".
[0008] The global coupling modeling layer is a global nonlinear coupling model constructed using the GBR algorithm based on the feature vectors output by the local feature extraction layer. It is used to capture the multiple coupling effects of climate data, lithological data, dam design parameters, and dam construction material test data on the target mechanical parameters, and output preliminary permeability and mechanical performance parameters. The physical constraint embedding layer is used to embed the basic laws of geotechnical mechanics and engineering experience to realize logical constraints; The scene adaptation and optimization layer is used to introduce an attention mechanism and build multiple engineering scene libraries.
[0009] Furthermore, during the training of the hybrid machine learning model, scenario-specific sub-model optimization is implemented; scenario-specific sub-model optimization refers to: constructing dedicated sub-models based on transfer learning; the dedicated sub-models can be directly called for actual engineering scenarios; The exclusive sub-model is constructed by combining 3 types of climate environment, 5 types of lithology and 4 types of dam height.
[0010] Furthermore, the coupled derivation of the mechanical performance parameters of the target project includes the following steps: Input the basic data of the target project into the derivation model of the performance of rockfill dams; The system identifies engineering scenarios, derives models and matches sub-models, and outputs preliminary permeability and mechanical performance parameters. These preliminary permeability and mechanical performance parameters include: permeability coefficient. k Compression modulus Es Duncan-Chang model parameters K, Kb, c, 、 n, m, , Δ 、R f ; Constraint optimization of preliminary permeability and mechanical performance parameters.
[0011] Constraint optimization includes: Permeability coefficient k The seepage control requirements of the dam body must be met; compression modulus Es It can be dynamically adjusted according to dam height, compaction standards, and rock properties; Ensure that the parameters of the Duncan-Chang model satisfy the properties of the rock.
[0012] It also includes analyzing errors through actual target experiments, including the following steps: Conduct 3-5 sets of key target tests to obtain test data. Key target tests include permeation tests, compression tests, and triaxial shear tests. The test data items are consistent with the preliminary permeation and mechanical performance parameters. The error of the preliminary permeability and mechanical performance parameters is calculated based on the experimental data. If the error does not meet the requirements, the experimental data is added to the training set and the corresponding sub-model is retrained.
[0013] Furthermore, after achieving the coupled derivation of the mechanical performance parameters of the target project, it also includes incorporating the basic data of the target project and the preliminary penetration and mechanical performance parameters that meet the error requirements into the project dataset to achieve continuous iterative optimization of the model.
[0014] According to the present invention, the problems of high sampling and transportation difficulty and high cost, long test cycle and high cost, insufficient simulation of coupling relationship and poor parameter adaptability in the acquisition of traditional rockfill material performance parameters can be solved, and technical breakthroughs can be achieved in engineering big data reuse, machine learning empowerment and parameter scenario adaptation. Attached Figure Description
[0015] Figure 1 This is a flowchart illustrating the coupled derivation method of mechanical performance parameters of rockfill dams according to an embodiment of the present invention. Detailed Implementation
[0016] The specific implementation of the present invention will now be described in detail with reference to the accompanying drawings.
[0017] The coupled derivation method for the mechanical performance parameters of rockfill dams provided by this invention is as follows: Figure 1 As shown, it includes the following steps: Step S100: Collect raw engineering data, preprocess the raw engineering data, and generate a rockfill dam construction engineering dataset; This step specifically includes the following: Step S101: Collect raw engineering data from multiple data sources and determine the data dimensions; In this step, the goal of collecting raw engineering data is to obtain a full-element big data set containing 5 dimensions and 40-50 core parameters, and to construct a dimensional mapping relationship between environment, parent rock, height, design parameters, and dam construction material tests to ensure the comprehensiveness and relevance of the data.
[0018] Based on this, the original engineering data were constructed from the engineering data of nearly 200 typical rockfill dams at home and abroad, covering different climate zones such as severe cold and high altitude, high temperature and rain, and plains; different lithologies such as basalt, granite, and limestone; and different levels such as low, medium, high, and ultra-high dams. The data dimensions include: climate data, lithology data, dam height data, dam design parameters, dam construction material test data, and target mechanical performance parameters. The data sources include: archived materials from design institutes, third-party testing reports, on-site test records, scientific research test data, and in-situ monitoring data.
[0019] The structure of the original engineering data is shown in Table 1: Table 1. Structure of Original Engineering Data
[0020] Step S102: Preprocess the original engineering data to optimize the data for engineering purposes, solve the problem of insufficient data directionality, and generate a dataset for rockfill dam construction engineering. The preprocessing process includes the following: 1) Define constraint criteria and remove outlier data based on these criteria. The constraint criteria are determined based on physical property values and engineering experience thresholds. Data to be removed includes those with a permeability coefficient between 1×10⁻⁶ and 10⁻⁶. -6 Data that do not conform to the basic laws of geotechnical mechanics include being outside the range of cm / s to 1 cm / s, having an internal friction angle outside the range of 25° to 55°, and having no obvious peak value in the triaxial test stress-strain curve.
[0021] 2) A joint KNN-engineering experience threshold algorithm is employed to identify missing data. This missing data is then supplemented using scenario similarity interpolation. This includes: deriving the missing breakage rate from parameters such as lithology, gradation, and relative density of dam materials from similar projects to identify discontinuous or missing data; and deriving the design stress level from design parameters such as dam height and compaction degree from similar projects to supplement missing data. By supplementing missing data, a complete set of data is ensured, avoiding deviations from actual conditions by pure algorithmic interpolation and effectively improving the rationality and completeness of the data.
[0022] 3) Standardize the parameters of different units to achieve data standardization, and use conversion formulas to correct the differences in parameter definitions between the old and new standards to ensure data consistency; for example, standardize the unit of permeability coefficient to cm / s and the unit of compression modulus to MPa.
[0023] 4) Data standardization and coding, including: using Z-score standardization for continuous parameters, such as temperature, altitude and gradation parameters, to eliminate the influence of dimensions; using unique thermal coding for categorical parameters, such as lithology, dam construction zoning and particle characteristics; and converting textual descriptions into feature values that can be recognized by machine learning.
[0024] 5) By using PPMCC and XGBoost to filter the core engineering features of the data, extract the key coupling relationships of the engineering features, generate coupling item data, and add it to the dataset. The coupling item data includes: The coupling feature of "lithology + gradation + relative density" is used to represent the influence of rock lithology on the particle gradation and relative density properties in the natural or fractured state. The coupling feature of "number of freeze-thaw cycles + breakage rate" is used to reflect the influence of the number of freeze-thaw cycles on the physical and mechanical properties of rocks and the breakage rate after crushing. The coupling characteristic of "dam height + dam construction zone + stress level" is used to reflect the influence of the design dam height on the range of dam construction zones and the stress level of each zone of the entire dam.
[0025] In summary, coupling term data can improve the model's ability to capture the coupling relationships of multiple factors.
[0026] The preprocessed dataset is divided into training, validation, and test sets. The training set is used for model training, the validation set for parameter optimization, and the test set for accuracy verification. The division ensures that the training set covers three different climate environments (severe cold, mild, and hot) and three rock types (basalt, granite, and limestone), and matches three dam height levels (low, medium, and high) to improve the model's generalization ability. Specifically, low dams are <30m, medium dams are 30-70m, and high dams are >70m.
[0027] This step in constructing the dataset for rockfill dam construction projects breaks through the limitations of traditional methods, such as reliance on single data and data fragmentation, and achieves systematic coupling of multiple types of heterogeneous data. Furthermore, by establishing a standardized data integration framework, it solves the industry pain points of low reuse rate of historical engineering data and difficulty in coordinating multi-source data, providing comprehensive data support for parameter derivation, rather than the experience or single-set experiment support of traditional methods.
[0028] Step S110: Define the structure, optimization objective, and training rules of the hybrid machine learning model; train the hybrid machine learning model to construct a derivation model for the performance of rockfill dams.
[0029] Specifically, the following steps are included: Step S111: Define the structure of the hybrid machine learning model, which includes a local feature extraction layer, a global coupling modeling layer, a physical constraint embedding layer, and a scene adaptation optimization layer; 1) The local feature extraction layer is implemented by constructing multiple parallel decision trees based on the RF model; different dimensions of local coupling features are extracted from the parallel decision trees respectively, and feature vectors are output to adapt to the independent analysis logic of environment and materials, design and experiment in actual engineering.
[0030] The parallel decision tree includes an environment-stone feature branch and a design-basic test feature branch. The input data for the environment-stone characteristics branch are climate data, lithological data, and coupling term data, which are used to extract local correlation features such as "freeze-thaw cycle-lithology-weathering resistance" and "chemical composition-particle strength". The input data for the design-basic test feature branch are dam design parameters, dam material test data, and coupling term data, which are used to extract local coupling features such as "relative density-compaction degree-deformation characteristics" and "fracture rate-gradation-strength parameters".
[0031] 2) The global coupling modeling layer is a global nonlinear coupling model constructed using the GBR algorithm based on the feature vectors output by the local feature extraction layer. This layer captures the multiple coupling effects of climate data, lithological data, dam design parameters, and dam material test data on the target mechanical parameters, and initially outputs the permeability coefficient, compression modulus, and other parameters from the Duncan-Chang model. K, Kb, c, , n, m, , Δ 、R f Predicted values of parameters such as...
[0032] 3) The physical constraint embedding layer is used to embed fundamental laws of geotechnical mechanics and engineering experience to implement logical constraints, thereby improving the model's realism; specific logical constraints include: Numerical constraints: The numerical range of the permeability coefficient satisfies the rule, including 1×10⁻⁶. -6 cm / s~1×10cm / s, internal friction angle 25°~55°, compression modulus 10MPa~300MPa; if it exceeds the range, correction will be triggered. Logical constraints: The parameter changes follow a pattern: increasing the number of freeze-thaw cycles leads to a higher breakage rate and a lower internal friction angle; if the output shows the opposite result, the feature weights will be automatically adjusted. Empirical constraints: The derivation results reflect patterns, for example: the internal friction angle of basalt is 5°~8° higher than that of limestone; if the derivation value deviates beyond this range, historical data from the same scenario will be called to correct it.
[0033] 4) The scenario adaptation and optimization layer is used to introduce an attention mechanism and build multiple engineering scenario libraries. The attention mechanism assigns dynamic weights to key influencing factors to improve the model's relevance. The engineering scenario library covers at least 3 types of climate data, 5 types of lithological data, and 4 types of dam height data. The definition of dynamic weights includes the following: For different dam construction zones: the dam shell zone is given a high weight for "gradation-permeability coefficient", the core wall zone is given a high weight for "relative density-compression modulus", and the transition zone is given a high weight for "fracture rate-strength parameter". For different climatic environments: "freeze-thaw cycle number - triaxial shear parameter" is given a high weight for frigid regions, and "altitude - compressive modulus" is given a high weight for high-altitude regions; For different dam heights: high dams are given a high weight for "stress level - nonlinear strength parameter Rf", and low dams are given a high weight for "gradation - permeability coefficient".
[0034] In this step, key engineering design variables are used as core adaptation factors to achieve scenario-based customization of parameters: for example, the derivation weight of the permeability coefficient is automatically adjusted for the dam shell region and the core wall region; for high dams and low dams, the derivation logic of the Duncan-Chang model parameters is optimized. This ensures that the output parameters are not generic values, but appropriate values that are highly matched to specific engineering scenarios, solving the core pain point of traditional parameters being out of touch with actual working conditions.
[0035] Step S112: Determine the optimization objective and train the hybrid machine learning model; Specifically, a multi-objective optimization function is adopted, with the goal of minimizing the derivation error of the target mechanical performance parameters (including: permeability coefficient, compressibility modulus, and parameters of the Duncan-Chang model), and optimizing the core parameters of the Duncan-Chang model. k, kb, , Rf Assign higher weights to match the priority of engineering calculations; The training process uses the Adam optimizer with an initial learning rate of 0.001. After every 100 iterations, the learning rate is dynamically adjusted based on the validation set error. If the error decreases, the rate is maintained; if the error increases, the rate is reduced by 50%. For regularization, an L2 regularization term with a coefficient λ=0.001 is introduced to suppress model overfitting and ensure that the model has generalization ability on the new project data.
[0036] The training process also includes scenario-specific sub-model optimization: based on transfer learning, 60 exclusive sub-models are constructed by combining 3 types of climate environment, 5 types of lithology and 4 types of dam height; the 3 types of climate environment include severe cold, temperate and hot, the 5 types of lithology include basalt, granite, limestone, sandstone and dolomite, and the 4 types of dam height include low, medium, high and ultra-high dams; the exclusive sub-models can be directly called for actual engineering scenarios to improve the derivation accuracy.
[0037] After training the hybrid machine learning model using training and test sets based on data from rockfill dam construction projects, a performance derivation model for rockfill dam construction is constructed.
[0038] Furthermore, validation is performed using a validation set. If the parameter error does not meet the requirements, the corresponding sub-model is retrained.
[0039] This step abandons the inefficient traditional model of fitting empirical formulas and derivation through single experiments. Instead, it uses machine learning as the core to achieve simultaneous coupled derivation of multiple objective parameters: simultaneously outputting the permeability coefficient, compression modulus, and the complete set of parameters from the Duncan-Chang model. Compared to the shortcomings of traditional methods, such as independent derivation of single parameters and the amplification of accumulated errors, this invention uses machine learning to uncover the nonlinear coupling patterns between multiple data points, achieving accurate derivation with one input and multiple outputs. Furthermore, it is specifically adapted to the mechanical properties of coarse-grained rockfill soil, avoiding the problem that traditional empirical formulas are highly general but lack specificity.
[0040] Step S120: Load the derivation model of the rockfill dam construction performance to achieve coupled derivation of the mechanical performance parameters of the target project; specifically including the following steps: Step S121: Input the basic data of the target project. The basic data includes: climate and environmental parameters, stone material characteristic parameters, design parameters, and basic test data. Among them, the climate and environmental parameters include information such as average annual temperature, extreme high / low temperatures, average annual precipitation, number of freeze-thaw cycles, altitude, and average annual sunshine duration. The stone material characteristic parameters include strength (uniaxial saturated compressive strength, softening coefficient, elastic modulus), lithology, chemical composition, and particle size distribution (D). max D 50 D 10 Information such as the coefficient of uniformity Cu, the coefficient of curvature Cc, particle shape (angular / round) and roundness; design parameters include dam height, dam construction zones, compaction standards and design stress; basic test data include relative density test results (maximum and minimum dry density), particle specific gravity test results (coarse and fine particles), breakage rate test results and compaction test results.
[0041] Step S122: Identify the engineering scenario and achieve model derivation and sub-model matching; In this step, based on the dimensions of climate environment, lithology and dam height, a corresponding dedicated sub-model is matched from 60 sub-models; After inputting basic data, the dedicated sub-model performs local feature extraction, global coupling modeling, physical constraint and scenario adaptation optimization through the rockfill dam construction performance derivation model to output preliminary permeability and mechanical performance parameters; Relevant parameters include: permeability coefficient k Compression modulus Es Duncan-Chang model parameters K, Kb, c, , n, m, , Δ 、R f Content such as...
[0042] Step S123: Combining engineering experience thresholds with specifications / standards requirements, optimize the preliminary permeability and mechanical performance parameters under constraints; Constraint optimization includes: 1) Permeability coefficient k The seepage control requirements of the dam body must be met, such as in the rockfill area and secondary rockfill area. k ≥1×10 -2 cm / s, transition zone and infiltration zone k ≥1×10 -3 cm / s, reverse filter area k ≥1×10 -4 cm / s, heart wall region k ≥1×10 -5 cm / s; 2) Compression modulus Es It can be dynamically adjusted according to dam height, compaction standards, rock properties, etc., such as for high dams. Es 1-2 ≥60MPa, medium dam Es 1-2 ≥40MPa, etc.; 3) Ensure that the parameters of the Duncan-Chang model satisfy the properties of the rock, such as the internal friction angle of the basalt riprap. ≥40°, internal friction angle of limestone riprap ≥35°, destructive ratio R f ≥0.7, etc.
[0043] Furthermore, constrained optimization also includes error analysis through actual target experiments: conducting 3-5 sets of key target experiments, such as 1 set of permeation experiments, 1-2 sets of compression experiments, and 1-2 sets of triaxial shear experiments, to verify the accuracy of the derived parameters. If the parameter errors do not meet the requirements, the experimental data are added to the training set, and the corresponding sub-model is retrained. Parameter error requirements include: permeability coefficient k prediction error ≤ 5%, compression modulus Es prediction error ≤ 6%, and core parameters K, Kb, c of the Duncan-Chang model... n, m 0, Δ The prediction error for parameters such as Rf is ≤8%, and the prediction error for other parameters is ≤10%.
[0044] Traditional methods for large-scale testing, especially large-scale compression and triaxial shear tests, require specialized equipment, a large number of samples, and professional personnel. The testing cost for a single medium-to-large rockfill dam can reach 700,000 to 1 million yuan. This method can reduce the workload of complex testing by more than 80%, requiring only investment in basic and verification tests. The testing cost for a single project can be reduced by 500,000 to 800,000 yuan. At the same time, it reduces the consumption of raw materials for samples, equipment wear and tear, and manpower input, which is in line with the development concept of green engineering.
[0045] Step S130: Output the parameters whose errors have been verified as acceptable for dam stability and seepage calculations. Simultaneously, incorporate the complete data of the project (input parameters + verified parameters) into the rockfill dam construction project dataset to achieve continuous iterative optimization of the model and improve the derivation accuracy of subsequent projects. If the parameter errors do not meet the requirements, supplement the original project data with experimental data, retrain the rockfill dam construction performance derivation model, and perform parameter optimization and experimental verification again. Use the parameter error requirement as the iteration threshold, continuously optimize until the error requirement is finally met.
[0046] This step allows for the continuous addition of actual data and validation parameters from the target project to the historical project database. Through iterative training of the machine learning model, the accuracy of parameter derivation for subsequent projects is continuously improved, forming a closed-loop system from data accumulation to model optimization and accuracy improvement.
[0047] The rockfill dam construction performance derivation model provided by this invention is applicable to various rockfill dams with a dam height of <200m, covering climatic environments including frigid, temperate, and hot regions, and lithology including common rockfill dam construction materials such as basalt, granite, limestone, sandstone, and dolomite. It supports parameter input and output in multiple formats such as Excel, CSV, and CAD, and can be directly connected to mainstream dam stability calculation software and seepage calculation software. In the application of the rockfill dam construction performance derivation model, after a single set of engineering parameters is input, the entire process from model inference to parameter optimization can be completed within 24 hours, with model inference time ≤1 hour.
[0048] This invention provides specific implementation examples for clay-core rockfill dam projects in high-altitude and cold regions. The designed dam height is 184m, the dam type is a clay-core rockfill dam, and the dam construction materials are local basalt rockfill and clay with a D-grade aggregate. max =800mm, D 50 =80mm, Cu=11.2, Cc=1.7; the climate conditions are an average annual temperature of -2.2℃, an extreme minimum temperature of -30.8℃, 64 freeze-thaw cycles per year, and an altitude of 3350m; the compaction standard is Dr=0.92~0.94 for the dam shell area and Dr=0.95~0.97 for the core wall area.
[0049] Climate, stone, and design parameters were input into the performance derivation model of the rockfill dam, and basic tests were conducted, including relative density, specific gravity, and breakage rate tests. The performance derivation model of the rockfill dam was identified as a scenario of "high altitude and cold climate + basalt + high dam," and a specific sub-model was matched among 60 sub-models. After the initial parameters of the performance derivation model of the rockfill dam were output, they were optimized according to the rules of dam zoning, including: the permeability coefficient of the dam shell zone was taken as the upper limit of the derived value, the permeability coefficient of the core zone was taken as the lower limit of the derived value, and the compression modulus and Duncan-Chang model parameters were optimized according to the characteristic rules of dam zoning to avoid outliers that do not conform to physical laws and engineering experience. Finally, five sets of parameters were selected and large-scale indoor permeability tests, large-scale compression tests, and large-scale three-week shear tests were carried out for verification. The error between the derived values and the actual test values is shown in Table 2.
[0050] Table 2 Comparison of Model Derived Values and Actual Experimental Values
[0051] This engineering example uses the method of this invention to derive mechanical performance parameters, requiring only 5 sets of verification tests. Compared with the traditional method of 40 to 60 sets of tests, this reduces the workload by more than 87.5% and saves more than 700,000 yuan in testing costs. The parameter acquisition cycle does not exceed 24 hours, and the error between the derived parameters and the actual test results does not exceed 8%, verifying the reliability of the derived parameters.
[0052] Traditional methods require 40 to 60 complex tests to obtain a complete set of permeability and mechanical parameters, taking no less than two months. This method only requires 3 to 5 basic tests. Combined with data entry and model derivation, a complete set of suitable parameters can be output within 24 hours, shortening the parameter acquisition cycle by more than 95%. This saves critical time for dam design and construction plan formulation, and can shorten the overall engineering design cycle by 30 to 50 days.
[0053] This invention solves the problems of high sampling and transportation difficulty and high cost, long test cycle and high cost, insufficient simulation of coupling relationship and poor parameter adaptability in the traditional acquisition of performance parameters of riprap materials by creating a three-in-one closed-loop system of "engineering dataset - dedicated model training - performance verification feedback optimization". It achieves technical breakthroughs in engineering big data reuse, machine learning empowerment and parameter scenario adaptation.
[0054] The above-disclosed embodiments are merely a few specific examples of the present invention. However, the present invention is not limited thereto, and any variations that can be conceived by those skilled in the art should fall within the protection scope of the present invention.
Claims
1. A method for coupling derivation of damming mechanical performance parameters of a rockfill dam, characterized in that, Includes the following steps: Collect raw engineering data, preprocess the raw engineering data, and generate a rockfill dam construction engineering dataset; The data dimensions of the original engineering data include: climate data, lithological data, dam height data, dam design parameters, dam construction material test data, and target mechanical performance parameters; Define the structure, optimization objective, and training rules of the hybrid machine learning model; train the hybrid machine learning model to construct a dam construction performance derivation model for rockfill dams; the structure of the hybrid machine learning model includes a local feature extraction layer, a global coupling modeling layer, a physical constraint embedding layer, and a scene adaptation optimization layer; the optimization objective is to minimize the derivation error of the target mechanical performance parameters; Load the dam construction performance derivation model of the rockfill dam to achieve coupled derivation of the mechanical performance parameters of the target project.
2. The method for deriving the coupling of dam mechanics performance parameters of a rockfill dam according to claim 1, characterized in that, Preprocessing the raw engineering data includes: defining constraint criteria and removing abnormal data based on these criteria; identifying missing data and supplementing it; standardizing parameters from different units to achieve data uniformity; data standardization and coding; extracting key coupling relationships of engineering features and generating coupling item data to be added to the dataset. The coupling term data is used to improve the model's ability to capture the coupling relationships of multiple factors. The coupling term data includes: a coupling feature of "lithology + gradation + relative density", which represents the influence of rock lithology on the particle gradation and relative density properties in the natural or fractured state; a coupling feature of "freeze-thaw cycle count + fracture rate", which reflects the influence of the number of freeze-thaw cycles on the physical and mechanical properties of the rock and the fracture rate after crushing; and a coupling feature of "dam height + dam construction zone + stress level", which reflects the influence of the designed dam height on the range of dam construction zones and the stress level of each zone of the entire dam.
3. The method for deriving the coupling of dam mechanics performance parameters of a rockfill dam according to claim 2, characterized in that, The local feature extraction layer is implemented by constructing multiple parallel decision trees based on the RF model; the parallel decision trees include environmental-stone feature branches and design-basic test feature branches. The input data for the environment-stone feature branch are climate data, lithological data, and coupling term data, which are used to extract local correlation features; The local correlation features include: "freeze-thaw cycle-lithology-weathering resistance" and "chemical composition-particle strength"; The input data for the design-basic test feature branch are dam design parameters, dam construction material test data, and coupling term data, which are used to extract local coupling features. The local coupling features include "relative density-compaction degree-deformation characteristics" and "fracture rate-gradation-strength parameters".
4. The coupled derivation method for the mechanical performance parameters of rockfill dams according to claim 1, characterized in that, The global coupling modeling layer is a global nonlinear coupling model constructed using the GBR algorithm based on the feature vectors output by the local feature extraction layer. It is used to capture the multiple coupling effects of climate data, lithological data, dam design parameters, and dam construction material test data on the target mechanical parameters, and output preliminary permeability and mechanical performance parameters. The physical constraint embedding layer is used to embed the basic laws of geotechnical mechanics and engineering experience to realize logical constraints; The scenario adaptation and optimization layer is used to introduce an attention mechanism and build multiple engineering scenario libraries.
5. The coupled derivation method for the mechanical performance parameters of rockfill dams according to claim 1, characterized in that, During the training of the hybrid machine learning model, scenario-specific sub-model optimization is implemented; the scenario-specific sub-model optimization refers to: constructing a dedicated sub-model based on transfer learning; the dedicated sub-model can be directly called for actual engineering scenarios; The exclusive sub-model is constructed by combining 3 types of climate environment, 5 types of lithology and 4 types of dam height.
6. The coupled derivation method for the mechanical performance parameters of rockfill dams according to claim 1, characterized in that, The coupled derivation of the mechanical performance parameters for achieving the target project includes the following steps: Input the basic data of the target project into the derivation model of the rockfill dam construction performance; The system identifies engineering scenarios, derives models and matches sub-models, and outputs preliminary permeability and mechanical performance parameters. These preliminary permeability and mechanical performance parameters include: permeability coefficient. k Compression modulus Es Duncan-Chang model parameters K, Kb, c, n, m Δ 、R f ; Constraint optimization of preliminary permeability and mechanical performance parameters.
7. The coupled derivation method for the mechanical performance parameters of rockfill dams according to claim 6, characterized in that, The constraint optimization includes: Permeability coefficient k The seepage control requirements of the dam body must be met; compression modulus Es It can be dynamically adjusted according to dam height, compaction standards, and rock properties; Ensure that the parameters of the Duncan-Chang model satisfy the properties of the rock.
8. The coupled derivation method for the mechanical performance parameters of rockfill dams according to claim 7, characterized in that, The constraint optimization also includes analyzing errors through actual target experiments, including the following steps: Conduct 3-5 sets of key target tests to obtain test data. The key target tests include permeation tests, compression tests, and triaxial shear tests. The test data items are consistent with the preliminary permeation and mechanical performance parameters. The error of the preliminary permeability and mechanical performance parameters is calculated based on the experimental data. If the error does not meet the requirements, the experimental data is added to the training set and the corresponding sub-model is retrained.
9. The coupled derivation method for the mechanical performance parameters of rockfill dams according to claim 8, characterized in that, After the coupled derivation of the mechanical performance parameters of the target project, the method further includes incorporating the basic data of the target project and the preliminary penetration and mechanical performance parameters that meet the error requirements into the project dataset to achieve continuous iterative optimization of the model.