Rock-soil body parameter intelligent test method and system based on artificial intelligence

Abstract

The invention belongs to the field of geotechnical engineering, and provides a rock-soil body parameter intelligent test method and system based on artificial intelligence, and the method comprises the steps: obtaining the shear experiment data of a rock-soil body under different environment conditions, and calculating an equivalent shear weakening coefficient according to the experiment data; the shear modulus of the material is adjusted by weakening the shear strength of the rock-soil body through the water weakening factor; deriving a shear damage constitutive model under the action of the dynamic water head, and calculating the shear stiffness of the rock-soil body; establishing a Gaussian process regression model through experimental data and the shear stiffness obtained through calculation; carrying out global optimization on the Gaussian process regression model by adopting a self-adaptive stochastic optimization algorithm; dynamically adjusting model parameters; and outputting a final optimized rock-soil body parameter prediction result.

Description

Technical Field

[0001] The present invention belongs to the field of geotechnical engineering, and in particular relates to an artificial intelligence-based intelligent testing method and system for geotechnical parameters. Background Art

[0002] As an important foundation material in engineering construction, rock and soil's mechanical parameters, such as shear strength, shear modulus, and shear stiffness, directly impact the design and performance evaluation of geotechnical engineering structures, including foundation bearing capacity, slope stability, foundation pit support safety, and tunnel surrounding rock deformation control. Therefore, accurately obtaining the mechanical parameters of rock and soil under different environmental conditions, especially shear stiffness and shear strength, is a crucial prerequisite for ensuring engineering safety and economic efficiency.

[0003] Currently, the acquisition of geotechnical parameters mainly relies on indoor triaxial shear tests, direct shear tests, or on-site in-situ testing methods. Although these methods have strong engineering applicability and data intuitiveness, they have significant shortcomings in the following aspects:

[0004] Samples must be prepared separately and repeated tests must be carried out under each working condition, resulting in a long overall testing cycle and large investment in manpower and material resources; traditional methods often find it difficult to fully consider the impact of multiple coupled environmental factors such as moisture content, groundwater level, and water pressure disturbance on rock and soil parameters; test data are usually difficult to generalize to untested working conditions or complex nonlinear loading paths, and cannot meet the needs of actual engineering for intelligent prediction; most existing empirical models or constitutive relationships lack adjustability and it is difficult to maintain prediction accuracy when environmental variables fluctuate significantly. Summary of the Invention

[0005] In order to solve the problems in the prior art, the present invention provides an intelligent testing method for rock and soil parameters based on artificial intelligence, comprising the following steps:

[0006] Obtain shear test data of rock and soil under different environmental conditions and calculate the equivalent shear weakening coefficient based on the experimental data;

[0007] Introducing groundwater weakening factors to consider the impact of groundwater on rock and soil materials, and adjusting the shear modulus of the material by weakening the shear strength of the rock and soil through the water weakening factor;

[0008] Based on the equivalent shear weakening coefficient and groundwater weakening factor, the shear damage constitutive model under dynamic water head is derived. Based on the shear damage constitutive model, the shear stiffness of the rock and soil is calculated.

[0009] A Gaussian process regression model was established using experimental data and calculated shear stiffness.

[0010] Adopting adaptive stochastic optimization algorithm to perform global optimization on Gaussian process regression model, optimize model parameters and minimize the error between prediction results and experimental data;

[0011] According to the error feedback of the model during the optimization process, the model parameters are dynamically adjusted, and the prediction results are corrected for errors based on actual experimental data;

[0012] In each iteration, it is determined whether the model has reached the convergence condition. If so, the optimization process is stopped; otherwise, the iteration is continued until the convergence requirement is met.

[0013] Output the final optimized geotechnical parameter prediction results.

[0014] Furthermore, the calculation of the equivalent shear weakening coefficient includes the following steps:

[0015] Carry out shear experiments under different moisture content conditions to obtain the relationship curve between shear stress and shear displacement;

[0016] Based on the trend of shear stiffness changing with shear displacement, the initial value of shear stiffness and the stable value at the end of the shear process are extracted;

[0017] The change in shear stiffness is calculated by the ratio between the initial value and the stable value of shear stiffness, the change in the slope of the curve, and the difference in stiffness before and after shear failure.

[0018] The change in shear stiffness is divided by the change in moisture content to obtain the equivalent shear weakening coefficient, and a shear weakening factor matrix is constructed with different moisture contents as dimensions to form data support for subsequent modeling.

[0019] Furthermore, the construction of the groundwater weakening factor includes the following steps:

[0020] Collect hydrological and geological parameters such as groundwater level, water saturation, pore water pressure and hydration reaction degree in the area where the rock and soil are located;

[0021] Establishing a functional relationship model between groundwater parameters and shear modulus based on empirical formulas, statistical regression or experimental data, wherein the model is one of a linear function, an exponential decay function or a piecewise broken line function;

[0022] Substituting the collected hydrogeological data into the model, a groundwater weakening factor is obtained with a value ranging from 0 to 1, where a smaller weakening factor value indicates a more significant weakening of the shear modulus;

[0023] The initial shear modulus is multiplied by the groundwater weakening factor to obtain the corrected shear modulus, which is used as the input parameter of the shear damage constitutive model.

[0024] Furthermore, the shear damage constitutive model includes the following forms:

[0025] When the shear displacement is less than or equal to the damage initial displacement, the shear stress is proportional to the shear displacement and shows a linear growth;

[0026] When the shear displacement is greater than the damage initiation displacement, the shear stress shows softening behavior in the form of exponential decay. Its mathematical form includes shear stiffness, unit volume failure energy, power term of shear displacement and exponential parameter, and residual strength term.

[0027] The dynamic calculation formula for the evolution of shear stiffness with shear displacement is obtained by derivation. The formula contains the shear damage variable G(d), where G(d) is defined as the ratio of shear stiffness to unit volume failure energy multiplied by the exponential power of the difference between the shear displacement and the damage onset displacement.

[0028] Furthermore, the execution process of the adaptive stochastic optimization algorithm includes the following steps:

[0029] Initializing multiple model parameter combinations, the parameters including kernel function length scale, noise coefficient, and model output variance;

[0030] The fitness function is defined as the weighted mean square error or negative log marginal likelihood between the predicted shear stiffness and the experimental shear stiffness;

[0031] Adopting an adaptive perturbation strategy, a jump search is performed in the parameter space of the Gaussian process regression model, and the parameter group with the minimum current error is selected as the candidate solution.

[0032] After each round of optimization, the current optimal result is recorded and compared with the historical optimal solution, and the optimal parameters are retained;

[0033] If the prediction error is significant, the model training samples are expanded based on the prediction error feedback, and the model is dynamically updated and the residual is corrected;

[0034] When the convergence conditions are met, the optimization process is terminated and the final prediction results are output.

[0035] Another aspect of the present invention provides an artificial intelligence-based intelligent testing system for rock and soil parameters, comprising the following modules:

[0036] Shear test data acquisition module, used to obtain shear stress and shear displacement data of rock and soil under different environmental conditions, and calculate the equivalent shear weakening coefficient;

[0037] Weakening factor modeling module, which is used to introduce groundwater weakening factors, consider the influence of groundwater on the mechanical properties of rock and soil materials, and modify the shear modulus accordingly;

[0038] The shear damage modeling module is used to construct a shear damage constitutive model based on the equivalent shear weakening coefficient and the groundwater weakening factor, and calculate the shear stiffness accordingly;

[0039] Regression modeling module, used to build a Gaussian process regression model based on experimental data and shear stiffness;

[0040] The optimization control module is used to optimize the parameters of the Gaussian process regression model using an adaptive stochastic optimization algorithm to minimize the prediction error;

[0041] Dynamic feedback correction module, used to dynamically adjust model parameters and correct prediction results according to prediction errors;

[0042] The convergence judgment and result output module is used to judge whether the optimization process meets the convergence conditions and output the final optimized rock and soil shear parameter prediction results after convergence.

[0043] Furthermore, the shearing experiment data acquisition module further includes:

[0044] Curve analysis submodule, used to obtain shear stress-displacement curves under different moisture content conditions;

[0045] Initial stiffness and stable stiffness extraction submodule, used to extract the initial value and later stable value based on the evolution trend of shear stiffness;

[0046] The weakening coefficient calculation submodule is used to calculate the equivalent shear weakening coefficient according to the change in shear stiffness and moisture content, and to construct a shear weakening factor matrix.

[0047] Furthermore, the weakening factor modeling module includes:

[0048] Groundwater parameter acquisition unit, used to collect hydrological and geological information such as groundwater level, water saturation, pore pressure and hydration reaction intensity;

[0049] Weakening function establishment unit, used to construct the mapping relationship between groundwater parameters and shear modulus based on linear, exponential or piecewise models;

[0050] a weakening factor generating unit, configured to generate a groundwater weakening factor ranging from 0 to 1 according to the mapping relationship;

[0051] The shear modulus correction unit is used to multiply the groundwater weakening factor by the initial shear modulus to obtain the corrected shear modulus and input it into the constitutive model.

[0052] Furthermore, the shear damage modeling module includes:

[0053] Threshold recognition unit, used to identify the initial displacement of shear damage;

[0054] The segmented model construction unit is used to establish the segmented relationship between shear displacement and stress. When the shear displacement is less than the starting value, it increases linearly, and when it is greater than the starting value, it decays exponentially.

[0055] Shear stiffness differential calculation unit, used to derive the shear stress-displacement relationship to obtain the dynamic expression of shear stiffness;

[0056] The damage variable calculation unit is used to calculate the shear damage variable G(d), which is an important input parameter of the stiffness evolution function.

[0057] Furthermore, the optimization control module includes:

[0058] Parameter initialization unit, used to initialize multiple parameter combinations of the Gaussian process regression model, including kernel function length scale, noise coefficient and output variance;

[0059] The fitness evaluation unit is used to calculate the weighted mean square error or negative log marginal likelihood value corresponding to each set of parameter combinations;

[0060] The jump search submodule is used to perform global jump search in the parameter space based on the adaptive random perturbation strategy;

[0061] The optimal tracking unit is used to record the current optimal parameters after each round of optimization and compare them with the historical optimal solution;

[0062] Error feedback and data expansion unit, used to re-incorporate samples with large errors into the training set and dynamically update the model;

[0063] The convergence judgment module is used to terminate the optimization when the preset error change threshold is met and output the final prediction result.

[0064] Compared with the existing technology, the artificial intelligence-based intelligent testing method and system for rock and soil parameters proposed in this invention has the following beneficial effects:

[0065] The present invention integrates the equivalent shear weakening coefficient, groundwater weakening factor and shear damage constitutive model, combines Gaussian process regression with adaptive stochastic optimization algorithm, and constructs a complete shear stiffness prediction system that combines data-driven and physical constraints. It can significantly improve the prediction accuracy and degree of automation of shear modulus and shear strength parameters, and reduce dependence on a large number of physical experiments.

[0066] By introducing hydrological and geological parameters such as groundwater saturation, water level, and pore pressure to construct a weakening factor model, and combining the shear stiffness evolution characteristics under water content changes, a shear damage constitutive model under dynamic water head was constructed. This model can comprehensively reflect the coupled influence of environmental factors on the mechanical behavior of rock and soil, and is suitable for shear response prediction under complex working conditions.

[0067] Using a Gaussian process regression model as the prediction core, the system not only outputs shear stiffness predictions but also provides corresponding uncertainty confidence intervals. Combined with a residual feedback mechanism and a dynamic data correction strategy, the system possesses continuous learning and iterative optimization capabilities, enhancing the model's stability and engineering usability.

[0068] Compared with the data acquisition method that relies on testing multiple groups of working conditions one by one, the present invention constructs an agent prediction model, which can realize the intelligent prediction of geotechnical parameters under untested conditions based on a small amount of existing experimental data, thereby effectively saving experimental resources and time costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0070] Figure 1 is the shear force-shear displacement curve;

[0071] Figure 2 is the shear stiffness-shear displacement curve;

[0072] Figure 3 This is the shear stiffness derivation process;

[0073] Figure 4 It is the model training process. DETAILED DESCRIPTION

[0074] Below, the invention is preferably described with reference to the accompanying drawings and specific embodiments.

[0075] This embodiment solves the above problem through the following steps:

[0076] In one embodiment, the present invention provides an artificial intelligence-based intelligent testing method for geotechnical parameters. By introducing artificial intelligence modeling and optimization techniques, the method performs data-driven modeling and intelligent prediction of the shear mechanical behavior of geotechnical materials under the influence of various environmental factors. Based on the stress-displacement relationship data collected in shear experiments, the method combines the physical significance of the equivalent shear weakening coefficient and the groundwater weakening factor to establish a constitutive model that can dynamically reflect the shear damage process. Using shear stiffness as a characterization parameter, a Gaussian process regression model is introduced to intelligently learn and predict shear parameters. Furthermore, an adaptive stochastic optimization algorithm is used to iteratively optimize and dynamically correct the model, thereby achieving high-precision, automated, and highly interpretable intelligent testing of key mechanical parameters of geotechnical materials.

[0077] Specifically, the method comprises the following steps:

[0078] Step S10: obtaining shear test data of the rock and soil mass under different environmental conditions, and calculating the equivalent shear weakening coefficient based on the test data.

[0079] In this step, representative rock and soil samples are selected and a standardized shear test device is used to conduct shear loading tests on the samples under different moisture content conditions. Moisture content is the main environmental variable affecting shear strength and stiffness. In this embodiment, four gradient conditions are set: 5%, 10%, 15% and 20%, corresponding to Figure 1 The shear stress-shear displacement curve is shown in .

[0080] Specifically, the shear displacement is applied by the loading device, and the shear stress generated during the shearing process is recorded in real time, and the following is obtained: Figure 1 The response curves of shear stress versus shear displacement are shown. The different curves in the figure represent the shear behavior under different moisture contents (w = 5%, 10%, 15%, 20%). It can be observed that:

[0081] The shear strength is highest when the moisture content is 10%.

[0082] The most obvious strength weakening phenomenon is shown when the moisture content is 20%.

[0083] Based on the above experimental data, the shear stiffness is further calculated. The shear stiffness is defined as the incremental ratio of shear stress to shear displacement and is updated in real time during the shearing process. Figure 2 It can be seen that the shear stiffness drops sharply in the initial shear stage (less than 0.3 mm), then decays rapidly and tends to be stable. The initial values of the shear stiffness are significantly different under different moisture contents, among which the stiffness is highest under 5% moisture content and lowest under 20% moisture content.

[0084] Combined shear stiffness-shear displacement curve ( Figure 2 ) is used to define the equivalent shear weakening coefficient, which reflects the strength degradation of geomaterials under specific moisture content and shear displacement conditions. Its calculation is based on the ratio between the initial and stable shear stiffness values, the slope of the shear curve, and the difference in stiffness before and after shear failure.

[0085] In practical applications, the following processing methods can be used to generate an equivalent shear weakening coefficient database:

[0086] Numerical fitting is performed on the shear stiffness variation curves under various moisture content conditions;

[0087] Extract the critical point of stiffness drop and the mean value of the stable segment;

[0088] The stiffness reduction is calculated and normalized to form the equivalent shear weakening factor matrix.

[0089] The obtained equivalent shear weakening coefficient serves as a key parameter reflecting the damage evolution of rock and soil under the combined influence of hydrological environment and shear history, providing data support for the subsequent construction of shear damage constitutive model.

[0090] Step S20 , introducing a groundwater weakening factor, taking into account the influence of groundwater on the rock and soil material, and adjusting the shear modulus of the material by weakening the shear strength of the rock and soil through the water weakening factor.

[0091] In this step, based on the substantial impact of groundwater on the microstructure and interfacial mechanical properties of geotechnical materials, a groundwater weakening factor (GDF) is introduced to reflect the weakening effect of water on the internal structure and mechanical response characteristics of geotechnical materials. The GDF is a normalized parameter whose physical significance lies in quantitatively modifying the shear modulus of geotechnical materials, simulating the actual reduction in material strength caused by a water-containing environment.

[0092] The specific implementation steps are as follows:

[0093] Obtain groundwater parameters. Collect hydrogeological data such as groundwater level, water saturation, pore water pressure, and hydration reaction level in the rock and soil area as input for influencing factors. This data can be obtained through on-site surveys, geological profiles, hydrological drilling, or water content testing.

[0094] Establish a groundwater weakening function model. Based on empirical formulas, statistical analysis, or experimental regression, construct a functional relationship between groundwater parameters and shear modulus to form a quantitative calculation model for groundwater weakening factors. The weakening model can be a linear function, an exponential decay function, or a piecewise broken line function. Examples are as follows:

[0095] The groundwater weakening factor and water content show a monotonically decreasing relationship;

[0096] The groundwater weakening factor varies inversely with the effective stress;

[0097] When the saturation reaches a certain threshold, the shear modulus decreases significantly.

[0098] To calculate the groundwater weakening factor, substitute the actual water content, water saturation, or pore pressure parameters into the weakening function model to obtain the corresponding groundwater weakening factor value. This factor value is generally between 0 and 1, with smaller values indicating a higher degree of weakening and a more significant reduction in the shear modulus.

[0099] Adjust the shear modulus and correct the initial shear modulus of the rock and soil based on the groundwater weakening factor to obtain the equivalent shear modulus under environmental conditions. The specific method is to multiply the original shear modulus by the groundwater weakening factor to obtain a corrected modulus value, thereby accounting for the actual mechanical softening effect of the rock and soil under the influence of water.

[0100] Generate a shear modulus input data set, and use the shear modulus adjusted by the weakening factor and other physical and mechanical parameters to form a characteristic data set, which serves as one of the important inputs for subsequent shear damage constitutive model and artificial intelligence model training.

[0101] Boundary control and adaptive adjustment: Under some extreme groundwater conditions (such as sudden water level rise and strong infiltration field), a dynamic correction mechanism is adopted to implement boundary restrictions on the weakening factor and modulus adjustment process to prevent non-physical model deviations caused by extreme input values, thereby ensuring prediction stability and reliability.

[0102] Step S30, deriving a shear damage constitutive model under the action of dynamic water head based on the equivalent shear weakening coefficient and the groundwater weakening factor, and calculating the shear stiffness of the rock and soil mass based on the shear damage constitutive model;

[0103] The specific implementation process includes the following:

[0104] like Figure 3 As shown in Figure 1, first, based on the experimental data of shear stiffness variation with displacement under different moisture content conditions, the equivalent shear weakening coefficient of the rock mass is calculated. The equivalent shear weakening coefficient is used to characterize the degree of influence of moisture content change on shear stiffness, and is defined as:

[0105] Equivalent shear weakening coefficient = ratio of shear stiffness change to moisture content change, that is:

[0106]

[0107] Among them, W s represents the equivalent shear weakening coefficient, Δk represents the change in shear stiffness under different moisture contents, and Δw represents the corresponding change in moisture content.

[0108] On the basis of equivalent shear weakening, the groundwater weakening factor μ is introduced as the control coefficient of the water environment affecting the evolution of shear damage. The weakening factor μ is obtained by the equivalent shear weakening coefficient W s It is constructed with the control parameter α and is defined as:

[0109] μ=1-αW s

[0110] Among them, α is an empirical coefficient that can be calibrated according to actual experiments; the value of μ is between 0 and 1, which is used to control the shear damage growth rate and the degree of stiffness degradation.

[0111] Based on the above weakening coefficient and weakening factor, a shear damage constitutive model under dynamic water head is established to describe the stress response relationship of the rock mass at different shear displacement stages. The stress-displacement relationship σ(d) is defined as follows:

[0112]

[0113] That is, when the shear displacement is less than or equal to the damage initiation threshold, the stress growth is linear. When the shear displacement is greater than, the material enters the damage evolution stage, and the exponential decay term is used to describe the shear softening behavior.

[0114] in:

[0115] σ(d) represents shear stress;

[0116] d represents shear displacement;

[0117] d0 represents the initial displacement of shear damage evolution;

[0118] μ1, μ2 represent the groundwater weakening factors in the early shearing period and the softening period;

[0119] k s0 represents the initial shear stiffness;

[0120] τ r represents the residual shear strength;

[0121] a represents the stress equilibrium parameter;

[0122] GF0 represents the destruction energy per unit volume;

[0123] m represents the exponential parameter that controls the shear damage evolution rate.

[0124] To realize the dynamic calculation of shear stiffness, the above constitutive relation is differentiated to obtain the expression of shear stiffness k(d), which is used to reflect the stiffness evolution under displacement increment:

[0125] The shear stiffness k(d) is expressed as the derivative of the shear stress with respect to the displacement

[0126] When d≤d0, the shear stiffness is a constant. When d>d0, the shear stiffness includes damage expansion and modulus attenuation terms, which can be expressed as:

[0127]

[0128] Where G(d) is a dimensionless damage variable that increases with shear displacement and is defined as:

[0129]

[0130] The shear damage constitutive model and its shear stiffness calculation formula can be used to respond in real time to different shear paths, groundwater disturbances, and stress state changes. During model training and prediction, the dynamic evolution mechanism of the shear modulus with time and environment is realized, providing physical driving constraints for artificial intelligence models.

[0131] Step S40: Establishing a Gaussian process regression model based on the experimental data and the calculated shear stiffness.

[0132] The purpose of this step is to construct a proxy model that can predict the shear stiffness response of the rock mass. Figure 4 As shown, the specific process is as follows:

[0133] Data preparation and formatting: the shear test data (shear stress, shear displacement), the shear damage model output results (shear stiffness), and the geological-environmental characteristic parameters (such as water content, pore pressure, saturation) are uniformly encoded into a multidimensional input feature vector X = {x1, x2, ..., x n}, where each eigenvalue represents a factor affecting the shear stiffness.

[0134] Output variable definition, the shear stiffness k corresponding to each set of input is used as the output variable y for supervised modeling.

[0135] Kernel function selection and structure design,In the Gaussian process regression (GPR) framework, Gaussian radial basis kernel (RBF kernel) or Mahalanobis distance kernel is selected, and the covariance function is defined to control the sensitivity of the model to the similarity between different features.

[0136] Based on the prior assumptions, the GPR model assumes that the output variables follow a joint Gaussian distribution with a mean of zero, and the covariance matrix is determined by the kernel function between the input features. Noise terms can be added to simulate actual observation errors.

[0137] Training data input and model fitting, input training set (X, y) into GPR framework, get prediction function And obtain the predicted mean and confidence interval.

[0138] Feature weight adjustment, combined with the shear mechanics parameter set and the feature parameter probability distribution model (see the auxiliary box on the right side of the figure), can weight some features or perform principal component analysis (PCA) dimensionality reduction to reduce redundant input dimensions.

[0139] This step outputs a GPR shear stiffness proxy model with prediction and uncertainty assessment capabilities, which serves as a basic tool for subsequent optimization and correction.

[0140] In step S50 , an adaptive stochastic optimization algorithm is used to perform global optimization on the Gaussian process regression model, optimize the model parameters, and minimize the error between the prediction results and the experimental data.

[0141] The core of this step is to find the GPR model parameter combination that can minimize the prediction error. The process is as follows:

[0142] Model parameter initialization, including kernel function parameters (such as length scale, output variance), noise terms, model structure hyperparameters, etc., generates a set of candidate solutions with multiple sets of different initial values to form an optimized population.

[0143] The fitness function is constructed, and the objective function is defined as the weighted mean square error (MSE) or negative log marginal likelihood (NLML) between the true shear stiffness and the model predicted value, which can be expressed as:

[0144]

[0145] The ARO search mechanism uses the Adaptive Random Optimization (ARO) method to perform a perturbative jump search in parameter space. This method does not rely on gradient information and is suitable for global optimization in non-convex target spaces. Strategies such as search radius, perturbation probability, and direction weighting can be configured.

[0146] Model evaluation and ranking, after each set of candidate parameters is predicted by the GPR model and evaluated by the fitness function, they are sorted from best to worst according to performance, and the optimal parameters of the current round are selected as the current candidate solution.

[0147] The global optimal tracking mechanism saves the current optimal solution and the historical optimal solution in each round of iteration to ensure that the global minimum error solution is finally output.

[0148] This step ensures that the GPR proxy model not only has strong fitting ability but also has good generalization performance under specific shear conditions and moisture content environments.

[0149] Step S60: dynamically adjust the model parameters according to the error feedback of the model during the optimization process, and perform error correction on the prediction results in combination with actual experimental data.

[0150] This step introduces an error-driven mechanism based on the optimization of the GPR proxy model to implement dynamic adjustment of the model structure and correction of the prediction value error. Specifically, it includes:

[0151] Error residual extraction: for each prediction sample, calculate the deviation ε=k between the predicted value and the true shear stiffness true -k pred .

[0152] Error significance detection, setting the error threshold. If the error of some samples significantly exceeds the confidence interval (such as outside the prediction interval), they are marked as "abnormal samples" and enter the experience correction set.

[0153] Build a dynamic experience data set, add "abnormal samples" to the experience data buffer pool, and combine them with the existing sample set to form a new training set to enhance the model's learning ability in the "weak fitting" area.

[0154] Retrain the GPR model, use the updated data set, reconstruct the kernel function matrix and prediction distribution, and obtain the improved model parameters and prediction results.

[0155] Residual correction mechanism, introducing residual prediction correction terms in the prediction output, such as constructing an auxiliary residual regression model to model ε, to offset systematic bias.

[0156] Update the convergence evaluation benchmark and use the corrected model output for the next round of convergence judgment to improve the robustness of the model in practical applications.

[0157] This step ensures that the GPR model has the ability of adaptive learning and self-correction, and is more consistent with the actual experimental response in the shear stiffness prediction.

[0158] Step S70: During each iteration, determine whether the model has reached the convergence condition. If the convergence condition is met, stop the optimization process; otherwise, continue iterating until the convergence requirement is met.

[0159] To ensure the convergence and computational resource efficiency of the entire optimization process, the following judgment is made after each model update:

[0160] Convergence criteria design, defining convergence criteria, including but not limited to:

[0161] The change ratio between the current round's optimal fitness function value and the previous round's value is less than the preset threshold ε;

[0162] No better solution was obtained after several consecutive rounds;

[0163] The model error variance is less than the given stability index;

[0164] The credible intervals of the model predictions are stable within the set intervals.

[0165] Convergence judgment implementation:

[0166]

[0167] Among them, F (t) It represents the fitness value of the optimization result of the tth round, and δ is the allowable error ratio (for example, 0.01).

[0168] Set the maximum number of iterations to prevent the optimization from getting into an infinite loop.

[0169] If convergence is determined, the final prediction model structure, the global optimal parameter combination and the shear stiffness prediction result are output; otherwise, the process returns to step S50 to continue the optimization.

[0170] Step S80: Output the final optimized rock and soil parameter prediction results.

[0171] This step is used to output the final converged prediction results in a parameterized form after the Gaussian process regression model training is completed and adaptive stochastic optimization and error feedback correction are performed. This is used for design calculations, stability assessments, and construction decision support in geotechnical engineering.

[0172] The predicted key shear mechanical parameters of the rock and soil are output as a structured data table or multi-dimensional tensor. The output results include but are not limited to:

[0173] Shear stiffness k: reflects the soil's ability to resist shear deformation, unit is kPa / mm;

[0174] Shear strength index τ p : predicted peak shear stress, in kPa;

[0175] Shear failure displacement d f : predicted shear displacement at failure, in mm;

[0176] Model credibility intervals (e.g., 95% confidence intervals), used to quantify prediction uncertainty;

[0177] Residual strength τ r and the shear stiffness evolution function k(d).

[0178] Specifically, for example, a soil sample from a river in Sichuan, under the following characteristic conditions:

[0179] Moisture content: 15%

[0180] Shearing path: displacement controlled loading, maximum displacement 10mm

[0181] Groundwater weakening factor μ: 0.85 (derived from α and Ws)

[0182] Initial shear modulus: 70kPa / mm

[0183] The predicted output results of this method are as follows:

[0184]

[0185] This table illustrates the predicted trend of the shear hardening-peak-softening process of the rock mass under given conditions, and quantifies the corresponding stiffness degradation process and the range of model uncertainty.

[0186] In another embodiment, the present invention further provides an artificial intelligence-based intelligent testing system for rock and soil parameters, comprising:

[0187] Shear test data acquisition module, used to obtain shear stress and shear displacement data of rock and soil under different environmental conditions, and calculate the equivalent shear weakening coefficient;

[0188] Weakening factor modeling module, which is used to introduce groundwater weakening factors, consider the influence of groundwater on the mechanical properties of rock and soil materials, and modify the shear modulus accordingly;

[0189] The shear damage modeling module is used to construct a shear damage constitutive model based on the equivalent shear weakening coefficient and the groundwater weakening factor, and calculate the shear stiffness accordingly;

[0190] Regression modeling module, used to build a Gaussian process regression model based on experimental data and shear stiffness;

[0191] The optimization control module is used to optimize the parameters of the Gaussian process regression model using an adaptive stochastic optimization algorithm to minimize the prediction error;

[0192] Dynamic feedback correction module, used to dynamically adjust model parameters and correct prediction results according to prediction errors;

[0193] The convergence judgment and result output module is used to judge whether the optimization process meets the convergence conditions and output the final optimized rock and soil shear parameter prediction results after convergence.

[0194] In one embodiment, the shearing experiment data acquisition module further includes:

[0195] Curve analysis submodule, used to obtain shear stress-displacement curves under different moisture content conditions;

[0196] Initial stiffness and stable stiffness extraction submodule, used to extract the initial value and later stable value based on the evolution trend of shear stiffness;

[0197] The weakening coefficient calculation submodule is used to calculate the equivalent shear weakening coefficient according to the change in shear stiffness and moisture content, and to construct a shear weakening factor matrix.

[0198] In one embodiment, the weakening factor modeling module includes:

[0199] Groundwater parameter acquisition unit, used to collect hydrological and geological information such as groundwater level, water saturation, pore pressure and hydration reaction intensity;

[0200] Weakening function establishment unit, used to construct the mapping relationship between groundwater parameters and shear modulus based on linear, exponential or piecewise models;

[0201] a weakening factor generating unit, configured to generate a groundwater weakening factor ranging from 0 to 1 according to the mapping relationship;

[0202] The shear modulus correction unit is used to multiply the groundwater weakening factor by the initial shear modulus to obtain the corrected shear modulus and input it into the constitutive model.

[0203] In one embodiment, the shear damage modeling module includes:

[0204] Threshold recognition unit, used to identify the initial displacement of shear damage;

[0205] The segmented model construction unit is used to establish the segmented relationship between shear displacement and stress. When the shear displacement is less than the starting value, it increases linearly, and when it is greater than the starting value, it decays exponentially.

[0206] Shear stiffness differential calculation unit, used to derive the shear stress-displacement relationship to obtain the dynamic expression of shear stiffness;

[0207] The damage variable calculation unit is used to calculate the shear damage variable G(d), which is an important input parameter of the stiffness evolution function.

[0208] In one embodiment, the optimization control module includes:

[0209] Parameter initialization unit, used to initialize multiple parameter combinations of the Gaussian process regression model, including kernel function length scale, noise coefficient and output variance;

[0210] The fitness evaluation unit is used to calculate the weighted mean square error or negative log marginal likelihood value corresponding to each set of parameter combinations;

[0211] The jump search submodule is used to perform global jump search in the parameter space based on the adaptive random perturbation strategy;

[0212] The optimal tracking unit is used to record the current optimal parameters after each round of optimization and compare them with the historical optimal solution;

[0213] Error feedback and data expansion unit, used to re-incorporate samples with large errors into the training set and dynamically update the model;

[0214] Convergence judgment module, used to terminate the optimization when the preset error change threshold is met and output the final prediction result

[0215] It should be noted that the explanation of the embodiment of the above-mentioned intelligent testing method for rock and soil parameters based on artificial intelligence is also applicable to the device of the embodiment of the present application and will not be repeated here.

[0216] Those skilled in the art will appreciate that the various units and algorithm steps described in the embodiments disclosed herein can be implemented using a combination of electronic hardware, computer software, and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0217] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0218] In the several embodiments provided in this application, if any function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of this application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory; hereinafter referred to as: ROM), random access memory (Random Access Memory; hereinafter referred to as: RAM), magnetic disk or optical disk, and other media that can store program code.

[0219] The above is only a specific embodiment of the present application. Any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in this application, which should be included in the scope of protection of this application. For some module structures that are not particularly clear in the present invention, the content recorded in the prior art shall prevail. The prior art mentioned in the above background technology section and the specific embodiment section of the present invention can be regarded as part of the present invention and is used to understand the meaning of some technical features or parameters.

Claims

1. An intelligent testing method for rock and soil parameters based on artificial intelligence, characterized in that: The method comprises the following steps: Obtain shear test data of rock and soil under different environmental conditions and calculate the equivalent shear weakening coefficient based on the experimental data; Introducing groundwater weakening factors to consider the impact of groundwater on rock and soil materials, and adjusting the shear modulus of the material by weakening the shear strength of the rock and soil through the water weakening factor; Based on the equivalent shear weakening coefficient and groundwater weakening factor, the shear damage constitutive model under dynamic water head is derived. Based on the shear damage constitutive model, the shear stiffness of the rock and soil is calculated. A Gaussian process regression model was established using experimental data and calculated shear stiffness. Adopting adaptive stochastic optimization algorithm to perform global optimization on Gaussian process regression model, optimize model parameters and minimize the error between prediction results and experimental data; According to the error feedback of the model during the optimization process, the model parameters are dynamically adjusted, and the prediction results are corrected for errors based on actual experimental data; In each iteration, it is determined whether the model has reached the convergence condition. If so, the optimization process is stopped; otherwise, the iteration is continued until the convergence requirement is met. Output the final optimized geotechnical parameter prediction results.

2. The method for intelligent testing of rock and soil parameters based on artificial intelligence according to claim 1, characterized in that: The calculation of the equivalent shear weakening coefficient comprises the following steps: Carry out shear experiments under different moisture content conditions to obtain the relationship curve between shear stress and shear displacement; Based on the trend of shear stiffness changing with shear displacement, the initial value of shear stiffness and the stable value at the end of the shear process are extracted; The change in shear stiffness is calculated by the ratio between the initial value and the stable value of shear stiffness, the change in the slope of the curve, and the difference in stiffness before and after shear failure. The change in shear stiffness is divided by the change in moisture content to obtain the equivalent shear weakening coefficient, and a shear weakening factor matrix is constructed with different moisture contents as dimensions to form data support for subsequent modeling.

3. The method for intelligent testing of rock and soil parameters based on artificial intelligence according to claim 1, characterized in that: The construction of the groundwater weakening factor comprises the following steps: Collect hydrological and geological parameters such as groundwater level, water saturation, pore water pressure and hydration reaction degree in the area where the rock and soil are located; Establishing a functional relationship model between groundwater parameters and shear modulus based on empirical formulas, statistical regression or experimental data, wherein the model is one of a linear function, an exponential decay function or a piecewise broken line function; Substituting the collected hydrogeological data into the model, a groundwater weakening factor is obtained with a value ranging from 0 to 1, where a smaller weakening factor value indicates a more significant weakening of the shear modulus; The initial shear modulus is multiplied by the groundwater weakening factor to obtain the corrected shear modulus, which is used as the input parameter of the shear damage constitutive model.

4. The method for intelligent testing of rock and soil parameters based on artificial intelligence according to claim 1, characterized in that: The shear damage constitutive model includes the following forms: When the shear displacement is less than or equal to the damage initial displacement, the shear stress is proportional to the shear displacement and shows a linear growth; When the shear displacement is greater than the damage initiation displacement, the shear stress shows softening behavior in the form of exponential decay. Its mathematical form includes shear stiffness, unit volume failure energy, power term of shear displacement and exponential parameter, and residual strength term. The dynamic calculation formula for the evolution of shear stiffness with shear displacement is obtained by derivation. The formula contains the shear damage variable G(d), where G(d) is defined as the ratio of shear stiffness to unit volume failure energy multiplied by the exponential power of the difference between the shear displacement and the damage onset displacement.

5. The method for intelligent testing of rock and soil parameters based on artificial intelligence according to claim 1, characterized in that: The execution process of the adaptive random optimization algorithm includes the following steps: Initializing multiple model parameter combinations, the parameters including kernel function length scale, noise coefficient, and model output variance; The fitness function is defined as the weighted mean square error or negative log marginal likelihood between the predicted shear stiffness and the experimental shear stiffness; Adopting an adaptive perturbation strategy, a jump search is performed in the parameter space of the Gaussian process regression model, and the parameter group with the minimum current error is selected as the candidate solution. After each round of optimization, the current optimal result is recorded and compared with the historical optimal solution, and the optimal parameters are retained; If the prediction error is significant, the model training samples are expanded based on the prediction error feedback, and the model is dynamically updated and the residual is corrected; When the convergence conditions are met, the optimization process is terminated and the final prediction results are output.

6. An artificial intelligence-based intelligent testing system for rock and soil parameters, characterized in that: The system includes the following modules: Shear test data acquisition module, used to obtain shear stress and shear displacement data of rock and soil under different environmental conditions, and calculate the equivalent shear weakening coefficient; Weakening factor modeling module, which is used to introduce groundwater weakening factors, consider the influence of groundwater on the mechanical properties of rock and soil materials, and modify the shear modulus accordingly; The shear damage modeling module is used to construct a shear damage constitutive model based on the equivalent shear weakening coefficient and the groundwater weakening factor, and calculate the shear stiffness accordingly; Regression modeling module, used to build a Gaussian process regression model based on experimental data and shear stiffness; The optimization control module is used to optimize the parameters of the Gaussian process regression model using an adaptive stochastic optimization algorithm to minimize the prediction error; Dynamic feedback correction module, used to dynamically adjust model parameters and correct prediction results according to prediction errors; The convergence judgment and result output module is used to judge whether the optimization process meets the convergence conditions and output the final optimized rock and soil shear parameter prediction results after convergence.

7. The artificial intelligence-based intelligent testing system for rock and soil parameters according to claim 6 is characterized in that: The shearing experiment data acquisition module further includes: Curve analysis submodule, used to obtain shear stress-displacement curves under different moisture content conditions; Initial stiffness and stable stiffness extraction submodule, used to extract the initial value and later stable value based on the evolution trend of shear stiffness; The weakening coefficient calculation submodule is used to calculate the equivalent shear weakening coefficient according to the change in shear stiffness and moisture content, and to construct a shear weakening factor matrix.

8. The artificial intelligence-based intelligent testing system for rock and soil parameters according to claim 6 is characterized in that: The weakening factor modeling module includes: Groundwater parameter acquisition unit, used to collect hydrological and geological information such as groundwater level, water saturation, pore pressure and hydration reaction intensity; Weakening function establishment unit, used to construct the mapping relationship between groundwater parameters and shear modulus based on linear, exponential or piecewise models; a weakening factor generating unit, configured to generate a groundwater weakening factor ranging from 0 to 1 according to the mapping relationship; The shear modulus correction unit is used to multiply the groundwater weakening factor by the initial shear modulus to obtain the corrected shear modulus and input it into the constitutive model.

9. The artificial intelligence-based intelligent testing system for rock and soil parameters according to claim 6 is characterized in that: The shear damage modeling module includes: Threshold recognition unit, used to identify the initial displacement of shear damage; The segmented model construction unit is used to establish the segmented relationship between shear displacement and stress. When the shear displacement is less than the starting value, it increases linearly, and when it is greater than the starting value, it decays exponentially. Shear stiffness differential calculation unit, used to derive the shear stress-displacement relationship to obtain the dynamic expression of shear stiffness; The damage variable calculation unit is used to calculate the shear damage variable G(d), which is an important input parameter of the stiffness evolution function.

10. The artificial intelligence-based intelligent testing system for rock and soil parameters according to claim 6, characterized in that: The optimization control module includes: Parameter initialization unit, used to initialize multiple parameter combinations of the Gaussian process regression model, including kernel function length scale, noise coefficient and output variance; The fitness evaluation unit is used to calculate the weighted mean square error or negative log marginal likelihood value corresponding to each set of parameter combinations; The jump search submodule is used to perform global jump search in the parameter space based on the adaptive random perturbation strategy; The optimal tracking unit is used to record the current optimal parameters after each round of optimization and compare them with the historical optimal solution; Error feedback and data expansion unit, used to re-incorporate samples with large errors into the training set and dynamically update the model; The convergence judgment module is used to terminate the optimization when the preset error change threshold is met and output the final prediction result.

Images