CN117252063A - Machine learning-based prediction method and system for anchorage deformation in rocky high slope excavation - Google Patents

Abstract

The invention discloses a machine learning-based rock high slope excavation anchoring deformation prediction method and system, wherein the method comprises the following steps: obtaining geological survey reports, slope excavation progress and monitoring information of a high arch dam area, and constructing a rock high slope multistage excavation anchoring model; based on a machine learning algorithm, importance ranking is carried out on the physical and mechanical parameters of the rock mass of the constructed model, and the key physical and mechanical parameters of the rock mass are identified as random variables to be updated; constructing a proxy model between random variables and slope response in each excavation anchoring stage based on a machine learning algorithm; constructing a Bayesian framework, and incorporating the monitoring information into posterior distribution of physical and mechanical parameters of the inverted rock mass; inputting posterior distribution of physical and mechanical parameters of the inverted rock mass into a proxy model of each excavation and anchoring stage to predict side slope deformation of the subsequent excavation and anchoring stage. The method and the device realize the prediction of rock mass high slope excavation anchoring deformation by utilizing the monitoring information to make up for the defect of rock mass physical and mechanical parameter test data.

Description

Machine learning-based prediction method and system for anchorage deformation in rocky high slope excavation

Technical field

The invention belongs to the technical field of slope deformation prediction, and specifically relates to a machine learning-based method and system for predicting the anchoring deformation of high rock slope excavation.

Background technique

In recent years, high arch dams with good economic performance and safety have been widely favored by engineers. A large number of high arch dams have been completed and put into operation in southwest China, such as Jinping I Hydropower Station, Xiluodu Hydropower Station, Baihetan Hydropower Station and Wudongde Hydropower Station. . Due to the huge scale of excavation during the construction of high arch dams, the problems of excavation deformation and stability of rocky high slopes in the dam area are particularly prominent, which is one of the main engineering technical issues in hydropower construction.

Generally speaking, slope deformation is usually predicted based on finite element models in engineering. However, since the slope rock mass is a natural material with the characteristics of concealment and spatiotemporal variability, and is affected by limited test data, the physical and mechanical parameters of the rock mass input into the finite element model to calculate the slope deformation cannot represent the real The physical and mechanical properties of the rock mass lead to obvious errors in predicting slope deformation based only on the finite element model. In recent years, with the development of monitoring technology, a large number of monitoring instruments are usually installed during the construction process of high arch dams to obtain monitoring data such as slope displacement, stress, and seepage pressure in real time. These data reflect the real status of the slope in real time. Compared with those based on The physical and mechanical parameters of rock mass obtained from survey data and tests are more reliable. How to use monitoring information to invert the physical and mechanical parameters of rock mass and combine finite element models and machine learning algorithms to predict slope deformation in real time needs further development.

Contents of the invention

The purpose of the present invention is to provide a method for predicting anchorage deformation in high rock slope excavation based on machine learning in view of the shortcomings of the existing technology. This method can use monitoring information to make up for the lack of experimental data on the physical and mechanical parameters of the rock mass. Predict the anchorage deformation of excavation of high rocky slopes under certain conditions.

In order to solve the above technical problems, the present invention adopts the following technical solutions:

A machine learning-based prediction method for anchorage deformation in excavation of high rock slopes, which is characterized by including the following steps:

Step 1: Obtain the geological survey report, slope excavation progress and monitoring information of the high arch dam area, and use this to build a multi-stage excavation and anchoring model for the rocky high slope;

Step 2: Rank the importance of the rock mass physical and mechanical parameters used to construct the model based on the machine learning algorithm, and identify the key rock mass physical and mechanical parameters as random variables to be updated;

Step 3: Build a surrogate model between random variables and slope response in each excavation and anchoring stage based on a machine learning algorithm;

Step 4: Establish the functional relationship between the monitoring information and the proxy model, construct a Bayesian framework, and incorporate this functional relationship into it to invert the posterior distribution of the physical and mechanical parameters of the rock mass;

Step 5: Input the inverted posterior distribution of rock mass physical and mechanical parameters into the proxy model of each excavation and anchoring stage to predict the slope deformation in subsequent excavation and anchoring stages.

Furthermore, the method for constructing the multi-stage excavation and anchoring model of the rocky high slope in step 1 is: based on the dam area geological survey report, slope excavation progress and monitoring information, generate the grid of the finite element model in ANSYS, and pass The plug-in is imported into FLAC3D to build a multi-stage excavation and anchoring model for rocky high slopes.

Furthermore, for the multi-stage excavation and anchorage model of rocky high slopes, the Mohr-Coulomb constitutive model is used, with normal constraints imposed around the model and the bottom side fixed; different groups are used to distinguish various types of rock mass and structural surface materials in the model , and use the null model to simulate excavation materials and the cable model to simulate prestressed anchor cables.

Further, the implementation method of step 2 is:

Generate a rock mass physical and mechanical parameter data set based on Latin hypercube sampling, generate a corresponding command stream based on the rock mass physical and mechanical parameter data set, and input the command stream into the FLAC3D model to calculate the corresponding slope response data set;

The random forest algorithm is used for analysis. The rock mass physical and mechanical parameter data set is defined as the input and the slope response data set is defined as the output. The importance of the rock mass physical and mechanical parameters is ranked based on the importance measurement method in the random forest algorithm. The physical and mechanical parameters of the rock mass that have a key impact on the slope response are identified as random variables to be updated.

Further, the physical and mechanical parameters of the rock mass include elastic modulus E, Poisson's ratio μ, gravity γ, cohesion c and internal friction angle .

Further, step 3 includes:

Define the key rock mass physical and mechanical parameters identified in step 2 as random variables x = [x ₁ , x ₂ ,..., x _d ]; use the support vector machine algorithm to train the relationship between random variables and slope response in each excavation and anchoring stage The proxy model SVM(x) is used to replace the FLAC3D model.

Further, in step 4, the functional relationship between the established monitoring information Y = [y ₁ , y ₂ ,..., y _n ] and the agent model SVM (x) is:

Y＝SVM(x)+ε

In the formula, ε represents the measurement error.

Furthermore, based on Bayesian theory, the functional relationship of monitoring information is incorporated into the following formula, and the adaptive differential evolution Metropolis algorithm DREAM _(ZS) with multi-chain parallel computing is used to invert the posterior distribution of the physical and mechanical parameters of the rock mass:

p(x|Y)=cL(x|Y)p(x)

In the formula, p(x|Y) represents the parameter posterior probability density function; p(x) represents the parameter prior probability density function, and the parameters obey lognormal distribution and are independent of each other; c is a normalization constant; L(x |Y) represents the likelihood function; the subscript j represents the jth rock mass physical and mechanical parameters; μ _N and σ _N represent the logarithmic mean and standard deviation of the parameters; the subscript n represents the nth excavation and anchoring stage; μ _ε and σ _ε represent the mean and standard deviation of the measurement error ε.

Further, step 5 specifically includes:

The posterior distribution of the inverted rock mass physical and mechanical parameters is input into the proxy model SVM(x) of each excavation stage to predict the slope deformation in the subsequent excavation and anchoring stages. According to the calculation results, the average value is taken as the predicted value of the slope deformation in each subsequent excavation and anchoring stage and compared with the actual monitoring value of the slope.

Another object of the present invention is to provide a system based on the above machine learning-based anchorage deformation prediction method for high rock slope excavation, including:

The data acquisition module is used to obtain the geological survey report, slope excavation progress and monitoring information of the high arch dam area;

The model building module is used to build a multi-stage excavation and anchoring model for rocky high slopes based on the data obtained by the data acquisition module;

The random variable screening module is used to rank the importance of the rock mass physical and mechanical parameters used in building the model based on machine learning algorithms, and identify key rock mass physical and mechanical parameters as random variables to be updated;

The surrogate model training module is used to train the surrogate model between random variables and slope response in each excavation and anchoring stage based on machine learning algorithms;

The posterior distribution acquisition module of the physical and mechanical parameters is used to establish the functional relationship between the monitoring information and the agent model, and build a Bayesian framework to incorporate this functional relationship into it to invert the posterior distribution of the physical and mechanical parameters of the rock mass;

The slope deformation prediction module is used to input the posterior distribution of the rock mass physical and mechanical parameters inverted by the physical and mechanical parameters acquisition module into the agent model of each excavation and anchoring stage to predict the slope in the subsequent excavation and anchoring stages. Deformation.

Compared with the existing technology, the beneficial effects of the present invention are: the invention provides a machine learning-based method for predicting the deformation prediction of rock high slope excavation anchorage, taking into account the physical and mechanical parameters of the rock mass of high rock slopes. Under the premise of certainty, the Bayesian method is used combined with actual engineering monitoring information to gradually invert the posterior distribution of the physical and mechanical parameters of the rock mass and predict the slope deformation in the subsequent excavation and anchoring stages, solving the problem based only on limited survey data and It is difficult to reliably invert the physical and mechanical parameters of rock mass from test data; the method of the present invention ranks the physical and mechanical parameters of the rock mass input to the FLAC3D model based on the machine learning algorithm, and identifies the key physical and mechanical parameters of the rock mass as those to be updated. Random variables are used to reduce the number of inversion parameters. At the same time, this method trains the agent model to reduce the number of calls to complex models, which greatly improves the computational efficiency of Bayesian update; The excavation and anchorage deformation prediction method provides a basis and has good engineering application value.

Description of drawings

Figure 1 is a flow chart of a method for predicting anchorage deformation in high rock slope excavation based on machine learning according to an embodiment of the present invention;

Figure 2 is a diagram of a multi-stage excavation and anchoring model of a high rocky slope related to an embodiment of the present invention.

Detailed ways

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without any creative work fall within the scope of protection of the present invention.

It should be noted that, as long as there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

The present invention will be further described below with reference to specific embodiments, but shall not be used as a limitation of the present invention.

As shown in Figure 1, an embodiment of the present invention provides a method for predicting anchoring deformation in rocky high slope excavation based on machine learning, which includes the following steps:

Step 1: Obtain the geological survey report, slope excavation progress and monitoring information of the high arch dam area, and use this to build a multi-stage excavation and anchoring model for the rocky high slope;

Obtain the geological survey report of the high arch dam area, slope excavation progress and monitoring information based on the actual engineering conditions, generate the grid of the finite element model in ANSYS, and import it into FLAC3D through the plug-in to construct the multi-stage excavation and anchoring of the rocky high slope. model, as shown in Figure 2. The multi-stage excavation and anchoring model of the rocky high slope adopts the Mohr-Coulomb constitutive model, with normal constraints imposed around the model and the bottom side fixed. The simulated rock mass and structural surface materials in the multi-stage excavation and anchorage model of rocky high slopes are distinguished by different groups, and the null model is used to simulate the excavation materials, and the cable model is used to simulate the prestressed anchor cables.

Step 2: Rank the importance of the rock mass physical and mechanical parameters used to construct the model based on the machine learning algorithm, and identify the key rock mass physical and mechanical parameters as random variables to be updated;

Based on the dam area geological survey report, engineering experience and literature, determine the physical and mechanical parameters of the rock mass. Among them, the physical and mechanical parameters of the rock mass include elastic modulus E, Poisson's ratio μ, gravity γ, cohesion c and internal friction angle. , generate the corresponding command stream based on the rock mass physical and mechanical parameter data set, input the command stream into the FLAC3D model to calculate the corresponding slope response data set; use the random forest algorithm for analysis, and define the rock mass physical and mechanical parameter data set as the input, edge The slope response data set is defined as the output; the importance of the physical and mechanical parameters of the rock mass is ranked based on the importance measurement method in the random forest algorithm, and the physical and mechanical parameters of the rock mass that have a key impact on the slope response are identified as random variables to be updated. .

In this embodiment, the physical and mechanical parameters obtained are specifically shown in Table 1. Among them, the elastic modulus E and cohesion c with high variability are considered as uncertain parameters, and random parameters are constructed through Latin hypercube sampling. The input set of the forest algorithm. The importance ranking of rock mass physical and mechanical parameters is based on the importance measurement method in the random forest algorithm. The physical and mechanical parameters of the rock mass that have a key impact on the displacement of the slope monitoring point are identified as random variables to be updated. The prior distribution of the selected random variables is shown in Table 2.

Table 1 Physical and mechanical parameters of slope rock mass

Table 2 Prior distribution table of random variables to be updated

Step 3: Build a surrogate model between random variables and slope response in each excavation and anchoring stage based on a machine learning algorithm;

According to the order of parameter importance, the five identified random variables x = [x ₁ , x ₂ ,..., x _d ] identified to be updated are constructed as the input set, and the corresponding FLAC3D model calculation results are used as the output set, which is constructed using the support vector machine algorithm. For the surrogate model between random variables and slope monitoring point displacement in each excavation and anchoring stage, the input set and output set are constructed as data sets, and the data set is divided into a training set and a prediction set, and the training set is used to pair the surrogate model Perform training and use the prediction set for verification to obtain each agent model. The Bayesian update method based on the response of geotechnical structures usually consists of two parts, namely obtaining the posterior distribution of uncertain model parameters and updating the structural response based on the posterior distribution. In this process, a large number of numerical model simulations need to be performed. Therefore, in the subsequent In Bayesian updating, the surrogate model is used instead of the FLAC3D model to achieve high computing efficiency while ensuring calculation accuracy.

Step 4: Establish the functional relationship between the monitoring information and the proxy model, construct a Bayesian framework, and incorporate this functional relationship into it to invert the posterior distribution of the physical and mechanical parameters of the rock mass;

The functional relationship between the established monitoring information Y = [y ₁ , y ₂ ,..., y _n ] and the surrogate model SVM (x) is:

Y＝SVM(x)+ε

In the formula, ε represents the measurement error;

Among them, the measurement error ε obeys the standard normal distribution with a mean of 1mm and a variance of 0. Based on Bayesian theory p(x|Y)=cL(x|Y)p(x), combined with prior knowledge and actual engineering monitoring information, the parameter prior probability density function p(x) and likelihood function L( x|Y), thereby calculating the parameter posterior probability density function p(x|Y). Since the random variable x is high-dimensional and nonlinear, the posterior distribution p(x|Y) usually does not have an analytical solution. This embodiment uses the adaptive differential evolution Metropolis algorithm DREAM _(ZS) with multi-chain parallel computing to invert the random variable x the posterior distribution of. Specifically, the adaptive differential evolution Metropolis algorithm DREAM _(ZS) with multi-chain parallel computing is used to invert the posterior distribution of the physical and mechanical parameters of the rock mass as:

p(x|Y)=cL(x|Y)p(x)

In the formula, p(x|Y) represents the parameter posterior probability density function; p(x) represents the parameter prior probability density function, and the parameters obey lognormal distribution and are independent of each other; c is a normalization constant; L(x |Y) represents the likelihood function; the subscript j represents the jth rock mass physical and mechanical parameters; μ _N and σ _N represent the logarithmic mean and standard deviation of the parameters; the subscript n represents the nth excavation and anchoring stage; μ _ε and σ _ε represent the mean and standard deviation of the measurement error ε.

The DREAM _(ZS) algorithm calculates multiple different Markov chains in parallel. If the parameter state of the i-th chain is x ⁱ , the new suggestion point x ^i,prop = ^xi +Δ ⁱ , where the increment Δ ⁱ is to be updated. The incremental part of , the acceptance rate of the proposed sample is determined by the Metropolis ratio α. In this embodiment, after the Markov chain converges, the next 25% of the samples are intercepted as the posterior distribution samples of the random variable play.

Step 5: Input the inverted posterior distribution of rock mass physical and mechanical parameters into the proxy model of each excavation and anchoring stage to predict the slope deformation in subsequent excavation and anchoring stages;

The posterior distribution of the inverted rock mass physical and mechanical parameters is input into the proxy model SVM(x) of each excavation and anchoring stage to predict the slope deformation in the subsequent excavation and anchoring stage. According to the calculation results, the average value is taken as the predicted value of the slope deformation in each subsequent excavation and anchoring stage and compared with the actual monitoring value of the slope, as shown in Table 3.

Table 3 Comparison table between slope deformation prediction values and monitoring values

The above embodiments are only illustrations of the technical solutions of the present invention. The machine learning-based anchorage deformation prediction method for high rock slope excavation involved in this embodiment is not limited to what is described in the above embodiments, but is subject to the scope defined by the claims. This embodiment uses a machine learning algorithm to identify key rock mass physical and mechanical parameters and build a proxy model. By incorporating on-site monitoring information into the Bayesian framework, it is possible to predict the anchoring deformation of high rock slope excavation, thereby providing engineering safety supervision in the dam area. in accordance with.

The present invention also provides a system based on the above machine learning-based anchoring deformation prediction method for high rock slope excavation, including:

The data acquisition module is used to obtain the geological survey report, slope excavation progress and monitoring information of the high arch dam area;

The model building module is used to build a multi-stage excavation and anchoring model for rocky high slopes based on the data obtained by the data acquisition module;

The random variable screening module is used to rank the importance of the rock mass physical and mechanical parameters used in building the model based on machine learning algorithms, and identify key rock mass physical and mechanical parameters as random variables to be updated;

The surrogate model training module is used to train the surrogate model between random variables and slope response in each excavation and anchoring stage based on machine learning algorithms;

The posterior distribution acquisition module of the physical and mechanical parameters is used to establish the functional relationship between the monitoring information and the agent model, and build a Bayesian framework to incorporate this functional relationship into it to invert the posterior distribution of the physical and mechanical parameters of the rock mass;

The slope deformation prediction module is used to input the posterior distribution of the rock mass physical and mechanical parameters inverted by the physical and mechanical parameters acquisition module into the agent model of each excavation and anchoring stage to predict the slope in the subsequent excavation and anchoring stages. Deformation.

The above are only preferred embodiments of the present invention, and are not intended to limit the implementation and protection scope of the present invention. Those skilled in the art should be able to realize equivalent substitutions and obvious changes made by applying the contents of the description of the present invention. The solutions obtained should be included in the protection scope of the present invention.

Claims (10)

1. A machine learning-based prediction method for anchorage deformation in excavation of high rock slopes, which is characterized by including the following steps: Step 1: Obtain the geological survey report, slope excavation progress and monitoring information of the high arch dam area, and use this to build a multi-stage excavation and anchoring model for the rocky high slope; Step 2: Rank the importance of the rock mass physical and mechanical parameters used to construct the model based on the machine learning algorithm, and identify the key rock mass physical and mechanical parameters as random variables to be updated; Step 3: Build a surrogate model between random variables and slope response in each excavation and anchoring stage based on a machine learning algorithm; Step 4: Establish the functional relationship between the monitoring information and the proxy model, construct a Bayesian framework, and incorporate this functional relationship into it to invert the posterior distribution of the physical and mechanical parameters of the rock mass; Step 5: Input the inverted posterior distribution of rock mass physical and mechanical parameters into the proxy model of each excavation and anchoring stage to predict the slope deformation in subsequent excavation and anchoring stages. 2. The machine learning-based method for predicting deformation of rocky high slope excavation anchorage according to claim 1, characterized in that step 1 is to construct a multi-stage rocky high slope excavation anchorage model as follows: according to the dam area The geological survey report, slope excavation progress and monitoring information were generated in ANSYS to generate the grid of the finite element model, and imported into FLAC3D through the plug-in to build a multi-stage excavation and anchoring model for the rocky high slope. 3. The machine learning-based method for predicting the deformation of rocky high slope excavation and anchorage according to claim 2, characterized in that, for the multi-stage rocky high slope excavation and anchorage model, the Mohr-Coulomb constitutive model is used. , normal constraints are applied around the model, and the bottom side is fixed; different groups are used to distinguish various types of rock mass and structural surface materials in the model, and the null model is used to simulate excavation materials, and the cable model is used to simulate prestressed anchor cables. 4. The machine learning-based anchoring deformation prediction method for high rock slope excavation according to claim 1, characterized in that the implementation method of step 2 is: Generate a rock mass physical and mechanical parameter data set based on Latin hypercube sampling, generate a corresponding command stream based on the rock mass physical and mechanical parameter data set, and input the command stream into the FLAC3D model to calculate the corresponding slope response data set; The random forest algorithm is used for analysis. The rock mass physical and mechanical parameter data set is defined as the input and the slope response data set is defined as the output. The importance of the rock mass physical and mechanical parameters is ranked based on the importance measurement method in the random forest algorithm. The physical and mechanical parameters of the rock mass that have a key impact on the slope response are identified as random variables to be updated. 5. The machine learning-based method for predicting anchorage deformation in excavation of high rock slopes according to claim 4, characterized in that the rock mass physical and mechanical parameters include elastic modulus E, Poisson's ratio μ, severity γ, cohesion Force c and internal friction angle 6. The machine learning-based anchoring deformation prediction method for high rock slope excavation according to claim 1, characterized in that step 3 includes: Define the key rock mass physical and mechanical parameters identified in step 2 as random variables x = [x ₁ , x ₂ ,..., x _d ]; use the support vector machine algorithm to train the relationship between random variables and slope response in each excavation and anchoring stage The proxy model SVM(x) is used to replace the FLAC3D model. 7. The machine learning-based anchoring deformation prediction method for high rock slope excavation according to claim 1, characterized in that in step 4, the established monitoring information Y=[y ₁ , y ₂ ,..., y _n The functional relationship between ] and the surrogate model SVM(x) is: Y＝SVM(x)+ε In the formula, ε represents the measurement error. 8. The method for predicting anchorage deformation in high rock slope excavation based on machine learning according to claim 7, characterized in that, based on Bayesian theory, the functional relationship of monitoring information is incorporated into the following formula, using multi-chain parallelism. The adaptive differential evolution Metropolis algorithm DREAM _(ZS) is calculated to invert the posterior distribution of the physical and mechanical parameters of the rock mass: p(x|Y)=cL(x|Y)p(x) In the formula, p(x|Y) represents the parameter posterior probability density function; p(x) represents the parameter prior probability density function, and the parameters obey lognormal distribution and are independent of each other; c is a normalization constant; L(x |Y) represents the likelihood function; the subscript j represents the jth rock mass physical and mechanical parameters; μ _N and σ _N represent the logarithmic mean and standard deviation of the parameters; the subscript n represents the nth excavation and anchoring stage; μ _ε and σ _ε represent the mean and standard deviation of the measurement error ε. 9. The machine learning-based anchoring deformation prediction method for high rock slope excavation according to claim 1, characterized in that step 5 specifically includes: The posterior distribution of the inverted rock mass physical and mechanical parameters is input into the proxy model SVM(x) of each excavation stage to predict the slope deformation in the subsequent excavation and anchoring stages. According to the calculation results, the average value is taken as the predicted value of the slope deformation in each subsequent excavation and anchoring stage and compared with the actual monitoring value of the slope. 10. A system based on a machine learning-based anchorage deformation prediction method for high rock slope excavation according to any one of claims 1 to 9, characterized in that it includes: The data acquisition module is used to obtain the geological survey report, slope excavation progress and monitoring information of the high arch dam area; The model building module is used to build a multi-stage excavation and anchoring model for rocky high slopes based on the data obtained by the data acquisition module; The random variable screening module is used to rank the importance of the rock mass physical and mechanical parameters used in building the model based on machine learning algorithms, and identify key rock mass physical and mechanical parameters as random variables to be updated; The surrogate model training module is used to train the surrogate model between random variables and slope response in each excavation and anchoring stage based on machine learning algorithms; The posterior distribution acquisition module of the physical and mechanical parameters is used to establish the functional relationship between the monitoring information and the agent model, and build a Bayesian framework to incorporate this functional relationship into it to invert the posterior distribution of the physical and mechanical parameters of the rock mass; The slope deformation prediction module is used to input the posterior distribution of the rock mass physical and mechanical parameters inverted by the physical and mechanical parameters acquisition module into the agent model of each excavation and anchoring stage to predict the slope in the subsequent excavation and anchoring stages. Deformation.