A method for predicting and analyzing the permeability of rock masses in hydropower projects

Abstract

The present invention provides a method for predicting and analyzing the permeability of rock masses in hydropower projects, comprising the following steps: determining four rock mass permeability analysis indicators whose correlation with the rock mass permeability coefficient K exceeds a threshold, namely: rock mass burial depth RD, rock quality designation RQD, fracture surface density feature FSD of rock cores, and rock mass integrity index RID; constructing a plurality of training samples to train a pre-constructed permeability prediction and analysis model to obtain a trained permeability prediction and analysis model; and using the trained permeability prediction and analysis model to predict the rock mass permeability coefficient K. The present invention provides a method for predicting and analyzing the permeability of rock masses in hydropower projects, which can accurately and completely predict and analyze the permeability of various rock masses, and effectively improve the design reference value of hydropower projects.

Description

Technical Field

[0001] The present invention belongs to the technical field of prediction and analysis of rock mass permeability, and particularly relates to a method for predicting and analyzing rock mass permeability in a hydropower project. Background Art

[0002] During the survey and design stage of a hydropower project, it is necessary to obtain the permeability performance of geological rock masses at the locations of buildings such as dam sites, water conveyance tunnels, powerhouse parts, surge chambers, shafts, and traffic tunnels, so as to obtain the groundwater permeability of the underground rock masses and provide basic data for subsequent design work such as anti-seepage and structural stability. Therefore, it is necessary to carry out geological exploration and test work to obtain the permeability performance of geological rock masses.

[0003] The permeability performance of geological rock masses is affected by various factors such as burial depth and rock mass fracture structure, and often exhibits significant anisotropy and spatial variability characteristics, with great uncertainty, making it difficult to accurately obtain the permeability performance parameter values. At present, the geological survey profession generally arranges exploration boreholes and water pressure tests, and calculates and statistically recommends the permeability values of rock masses. However, due to the limitations of complex geological conditions, funds, and construction period, the relevant data obtained from water pressure tests may be scattered and incomplete. Especially, there is a risk of contract performance in the completion of deep-hole water pressure tests. Moreover, currently, the permeability of various underground rock masses can generally only give recommended values of rock mass permeability by combining the recommended values in the specifications, expert experience, and partial water pressure test values. Therefore, it is currently difficult to accurately and completely obtain the permeability performance parameters of geological rock masses. Summary of the Invention

[0004] In view of the defects existing in the prior art, the present invention provides a method for predicting and analyzing rock mass permeability in a hydropower project, which can effectively solve the above problems.

[0005] The technical solution adopted by the present invention is as follows:

[0006] The present invention provides a method for predicting and analyzing rock mass permeability in a hydropower project, including the following steps:

[0007] Step S1, preliminarily determine various rock mass permeability analysis indexes related to the rock mass permeability coefficient K;

[0008] Step S2, conduct a correlation analysis on the preliminarily determined various rock mass permeability analysis indexes and the rock mass permeability coefficient K, and obtain four rock mass permeability analysis indexes whose correlation with the rock mass permeability coefficient K exceeds the threshold, namely: rock mass burial depth RD, rock quality designation RQD, fracture surface density feature FSD of rock core, and rock mass integrity index RID;

[0009] Step S3, construct multiple training samples, and perform normalization processing on each training sample to obtain multiple normalized training samples, forming a training sample set;

[0010] Each training sample in the training sample set is represented as s=(X, Y); where X is the vector of rock mass permeability analysis indexes, X=(x 1 , x 2 , x 3 , x 4 ), x 1 , x 2 , x 3 , x 4 , representing the sample values of the rock mass buried depth RD, the rock core quality index RQD, the rock core structural plane density characteristic FSD, and the rock mass integrity index RID respectively; Y represents the corresponding value of the rock mass permeability coefficient K;

[0011] Step S4: Use the training sample set to train the pre-constructed permeability prediction and analysis model to obtain a trained permeability prediction and analysis model;

[0012] Step S5: For the rock mass that needs to be subjected to permeability prediction and analysis currently, obtain the measured values of the rock mass buried depth RD, the rock core quality index RQD, the rock core structural plane density characteristic FSD, and the rock mass integrity index RID that need to be interpolated and calculated, and input them into the trained permeability prediction and analysis model. The trained permeability prediction and analysis model outputs the predicted value of the corresponding rock mass permeability coefficient K.

[0013] Preferably, in step S2, perform a correlation analysis on the initially determined various rock mass permeability analysis indexes and the rock mass permeability coefficient K. Specifically:

[0014] Step S2.1: Represent any one of the rock mass permeability analysis indexes as C, and obtain the measured values of the M rock mass permeability analysis indexes C, which are represented as: C 1 , C 2 , …, C M ; Obtain the rock mass permeability coefficients K corresponding to the measured values of each rock mass permeability analysis index C, and represent them in sequence as: K 1 , K 2 , …, K M ;

[0015] For the measured value sequence C 1 , C 2 , …, C M of the rock mass permeability analysis index C, the descending order serial numbers of the measured values of each rock mass permeability analysis index C in the measured value sequence of the rock mass permeability analysis index C are respectively represented as: Q 1 , Q 2 , …, Q M ;

[0016] For the sequence K 1 , K2 ,…,K M , the values of the permeability coefficient K of each rock mass are in the sequence K of the permeability coefficient K of the rock mass 1 ,K 2 ,…,K M , and the descending order numbers in it are respectively represented as: P 1 ,P 2 ,…,P M ;

[0017] Calculate the serial number Q 1 and the serial number P 1 , and get d 1 ;

[0018] Calculate the serial number Q 2 and the serial number P 2 , and get d 2 ;

[0019] And so on

[0020] Calculate the serial number Q M and the serial number P M , and get d M ;

[0021] Step S2.2, use the following formula to obtain the correlation coefficient r between the rock mass permeability analysis index C and the rock mass permeability coefficient K s :

[0022]

[0023] where: k = 1, 2, …, M;

[0024] Step S2.3, judge whether the correlation coefficient r s is greater than the threshold ε; if it is greater, then select this rock mass permeability analysis index C

[0025] Preferably, in step S2, the method for obtaining the core structural plane density characteristic FSD is:

[0026] Step S2-1, the drilling footage depth is L; the core structural plane dip angle trend density is divided into three dip angle sections according to the dip angle range, which are: (0, 30], (30, 60], (60, 90];

[0027] Step S2-2, count the number of structural planes in each dip angle section, which are: N 0～30 、N 30～60 and N 60～90 ;

[0028] Step S2-3, according to the rock mass lithology factors currently analyzed, set the weights corresponding to each dip angle section respectively, which are: W1 , W 2 and W 3 ;

[0029] Step S2-4, use the following formula to obtain the core structural plane density feature FSD:

[0030]

[0031] Thus, the value of the core structural plane density feature FSD is obtained.

[0032] Preferably, for metamorphic rocks, their weights are respectively: W 1 = 0.1, W 2 = 0.3, W 3 = 0.6.

[0033] Preferably, step S4 is specifically:

[0034] Step S4.1, the constructed permeability prediction analysis model is the DELM permeability prediction analysis model, including 1 input layer, N cascaded hidden layers before and after, and 1 output layer; among them, the input layer has 4 neurons, each hidden layer has 4 neurons, and the output layer has 4 neurons;

[0035] Step S4.2, randomly generate the weights of each neuron in the input layer through an orthogonal matrix;

[0036] Step S4.3, for any training sample s = (X, Y), X = (x 1 , x 2 , x 3 , x 4 ), Y represents the corresponding value of the rock mass permeability coefficient K. First, input x 1 , x 2 , x 3 , x 4 into the 4 neurons in the input layer respectively. Each neuron in the input layer calculates through the weights and outputs x 11 , x 21 , x 31 , x 41 respectively, forming a matrix X 1 = (x 11 , x 21 , x 31 , x 41 );

[0037] Step S4.4, the matrix X 1 = (x 11 , x 21 , x 31 , x 41)Input into the hidden layer of the first layer, and set the output matrix X of the hidden layer of the first layer 2 =(x 12 ,x 22 ,x 32 ,x 42 ) is equal to the input matrix X 1 . Therefore, through the following formula, the weight vector β of the hidden layer of the second layer is obtained 2 :

[0038]

[0039] β 2 β 2 T =I

[0040] where: C is the regularization parameter; I is the identity matrix;

[0041] Step S4.5, input the matrix X 2 =(x 12 ,x 22 ,x 32 ,x 42 ) into the hidden layer of the second layer, and adopt the following formula to obtain the output matrix X of the hidden layer of the second layer 3 :

[0042]

[0043] β 2 β 2 T =I

[0044] At the same time, determine the weight vector β of the hidden layer of the third layer 3 :

[0045] β 3 =X 3 Y

[0046] Step S4.6, input the matrix X 3 and the weight vector β 3 into the hidden layer of the third layer, and adopt the same method as in Step S4.5 to obtain the output matrix X of the hidden layer of the third layer 4 and the weight vector β of the hidden layer of the fourth layer 4 ; and so on, until the output matrix X of the Nth hidden layer N+1 and the weight vector β N+1 ;

[0047] Step S4.7, input the matrix X N+1 and the weight vector β N+1 into the output layer, and the output layer outputs the final output vector Y';

[0048] Step S4.8: By comparing the output vector Y′ with the marked value of the rock mass permeability coefficient K, that is, Y; if the two are less than the threshold value, then the weight vectors of each layer at this time form the final trained permeability prediction analysis model; if not less than the threshold value, aiming at Y′ being equal to Y, the least squares method is used to solve, and the solved weight vectors of each layer are obtained to form the final trained permeability prediction analysis model.

[0049] The method for predicting and analyzing the permeability of rock masses in hydropower projects provided by the present invention has the following advantages:

[0050] The present invention provides a method for predicting and analyzing the permeability of rock masses in hydropower projects, especially relating to methods for predicting and analyzing the permeability of underlying rock masses in pumped-storage power stations, conventional hydropower stations, etc., which can accurately and completely predict and analyze the permeability of various rock masses and effectively improve the reference value for hydropower project design. Description of the Drawings

[0051] Figure 1 It is a schematic flow chart of a method for predicting and analyzing the permeability of rock masses in hydropower projects provided by the present invention.

[0052] Figure 2 It is a flow chart for training a pre-constructed permeability prediction analysis model provided by the present invention;

[0053] Figure 3 It is a relationship diagram between the number of iterations and the convergence curve provided by the present invention;

[0054] Figure 4 It is a relationship diagram between the true value and the predicted value when training with a training set at different numbers of iterations provided by the present invention;

[0055] Figure 5 It is a relationship diagram between the true value and the predicted value when training with a test set at different numbers of iterations provided by the present invention. Detailed Embodiments

[0056] In order to make the technical problems, technical solutions and beneficial effects solved by the present invention clearer, the following further details the present invention with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0057] The present invention provides a method for predicting and analyzing the permeability of rock masses in hydropower projects, especially relating to methods for predicting and analyzing the permeability of underlying rock masses in pumped-storage power stations, conventional hydropower stations, etc., which can accurately and completely predict and analyze the permeability of various rock masses and effectively improve the reference value for hydropower project design.

[0058] Refer to Figure 1, the present invention provides a method for predicting and analyzing the permeability of rock masses in hydropower projects, comprising the following steps:

[0059] Step S1, preliminarily determine various rock mass permeability analysis indexes related to the rock mass permeability coefficient K;

[0060] Step S2, conduct a correlation analysis on the various preliminarily determined rock mass permeability analysis indexes and the rock mass permeability coefficient K, and obtain four rock mass permeability analysis indexes whose correlation with the rock mass permeability coefficient K exceeds the threshold, namely: rock mass burial depth RD, rock quality designation RQD, fracture density feature of core FSD, and rock mass integrity index RID;

[0061] In step S2, the correlation analysis on the various preliminarily determined rock mass permeability analysis indexes and the rock mass permeability coefficient K is specifically as follows:

[0062] Step S2.1, represent any one of the rock mass permeability analysis indexes as C, and obtain the measured values of M rock mass permeability analysis indexes C, expressed as: C 1 , C 2 , …, C M ; obtain the rock mass permeability coefficient K corresponding to the measured values of each rock mass permeability analysis index C, and represent them in sequence as: K 1 , K 2 , …, K M ;

[0063] For the measured value sequence C 1 , C 2 , …, C M of the rock mass permeability analysis index C, the descending order serial numbers of the measured values of each rock mass permeability analysis index C in the measured value sequence of the rock mass permeability analysis index C are respectively represented as: Q 1 , Q 2 , …, Q M ;

[0064] For the sequence K 1 , K 2 , …, K M of the rock mass permeability coefficient K, the descending order serial numbers of the values of each rock mass permeability coefficient K in the sequence K 1 , K 2 , …, K M of the rock mass permeability coefficient K are respectively represented as: P 1 , P 2 , …, P M ;

[0065] Calculate the difference between the serial number Q 1 and the serial number P 1 to obtain d 1 ;

[0066] Calculate serial number Q 2 and serial number P 2 to obtain the difference d 2 ;

[0067] And so on

[0068] Calculate serial number Q M and serial number P M to obtain the difference d M ;

[0069] Step S2.2, use the following formula to obtain the correlation coefficient r between the rock mass permeability analysis index C and the rock mass permeability coefficient K s :

[0070]

[0071] where: k = 1, 2, …, M;

[0072] Step S2.3, determine whether the correlation coefficient r s is greater than the threshold ε; if it is greater, then select this rock mass permeability analysis index C

[0073] In step S2, the acquisition method of the core structural plane density feature FSD is as follows

[0074] Step S2-1, the drilling footage depth is L; divide the core structural plane dip angle trend density into three dip angle sections according to the dip angle range, which are: (0, 30], (30, 60], (60, 90]

[0075] Step S2-2, count the number of structural planes in each dip angle section, which are: N 0～30 、N 30～60 and N 60～90 ;

[0076] Step S2-3, according to the current analyzed rock mass lithology factors, set the corresponding weights for each dip angle section, which are: W 1 、W 2 and W 3 ;

[0077] Step S2-4, use the following formula to obtain the core structural plane density feature FSD

[0078]

[0079] Thus, obtain the value of the core structural plane density feature FSD

[0080] As an example, for metamorphic rocks, the preferred weights are respectively: W 1 = 0.1, W 2= 0.3, W 3 = 0.6.

[0081] Step S3, construct multiple training samples, and perform normalization processing on each training sample to obtain multiple normalized training samples, forming a training sample set;

[0082] Each training sample in the training sample set is represented as s = (X, Y); where X is the vector of rock mass permeability analysis indicators, X = (x 1 , x 2 , x 3 , x 4 ), x 1 , x 2 , x 3 , x 4 , respectively representing the sample values of the rock mass burial depth RD, the rock core quality index RQD, the rock core structural plane density characteristic FSD, and the rock mass integrity index RID; Y represents the corresponding value of the rock mass permeability coefficient K;

[0083] Step S4, use the training sample set to train the pre-constructed permeability prediction analysis model to obtain a trained permeability prediction analysis model;

[0084] Step S4 is specifically as follows:

[0085] Step S4.1, the constructed permeability prediction analysis model is the DELM permeability prediction analysis model, including 1 input layer, N cascaded hidden layers, and 1 output layer; among them, the input layer has 4 neurons, each hidden layer has 4 neurons, and the output layer has 4 neurons;

[0086] Step S4.2, randomly generate the weights of each neuron in the input layer through an orthogonal matrix;

[0087] Step S4.3, for any training sample s = (X, Y), X = (x 1 , x 2 , x 3 , x 4 ), Y represents the corresponding value of the rock mass permeability coefficient K. First, input x 1 , x 2 , x 3 , x 4 into the 4 neurons of the input layer respectively. Each neuron in the input layer calculates through the weights and outputs x 11 , x 21 , x 31 , x 41 , forming a matrix X 1 = (x 11 , x 21 , x 31,x 41 );

[0088] Step S4.4, input the matrix X 1 =(x 11 ,x 21 ,x 31 ,x 41 ) into the hidden layer of the first layer, and set the output matrix X 2 =(x 12 ,x 22 ,x 32 ,x 42 ) of the hidden layer of the first layer to be equal to the input matrix X 1 . Therefore, through the following formula, the weight vector β 2 of the hidden layer of the second layer is obtained:

[0089]

[0090] β 2 β 2 2 =I

[0091] where: C is the regularization parameter; I is the identity matrix;

[0092] Step S4.5, input the matrix X 2 =(x 12 ,x 22 ,x 32 ,x 42 ) into the hidden layer of the second layer, and adopt the following formula to obtain the output matrix X 3 of the hidden layer of the second layer:

[0093]

[0094] β 2 β 2 T =I

[0095] Meanwhile, determine the weight vector β 3 of the hidden layer of the third layer:

[0096] β 3 =X 3 Y

[0097] Step S4.6, input the matrix X 3 and the weight vector β 3 into the hidden layer of the third layer, and adopt the same method as in Step S4.5 to obtain the output matrix X 4 of the hidden layer of the third layer and the weight vector β 4 of the hidden layer of the fourth layer; and so on until the output matrix X of the Nth hidden layerN+1 and weight vector β N+1 ;

[0098] Step S4.7: Input matrix X N+1 and weight vector β N+1 into the output layer, and the output layer outputs the final output vector Y'.

[0099] Step S4.8: By comparing the output vector Y' with the marked value of the rock mass permeability coefficient K, i.e., Y; if the two are less than the threshold, then the weight vectors of each layer at this time form the final trained permeability prediction analysis model; if not less than the threshold, aiming at Y' being equal to Y, the least squares method is used to solve, and the weight vectors of each layer after solution are obtained to form the final trained permeability prediction analysis model.

[0100] Step S5: For the rock mass that needs to conduct permeability prediction analysis currently, obtain the measured values of the rock mass buried depth RD, rock quality designation RQD, fracture density feature FSD of the core, and rock mass integrity index RID that need to be interpolated and calculated, and input them into the trained permeability prediction analysis model, and the trained permeability prediction analysis model outputs the predicted value of the corresponding rock mass permeability coefficient K.

[0101] The following introduces an embodiment:

[0102] In this embodiment, prediction analysis is carried out based on the drilling, water pressure test, and geophysical exploration test result data completed during the preliminary feasibility study and feasibility study stages of a certain pumped storage project area.

[0103] Step S1: Initially determine various rock mass permeability analysis indicators related to the rock mass permeability coefficient K

[0104] Specifically, according to the drilling test results of hydropower projects and the influencing factors of permeability indicators, initially set and select multiple rock mass permeability analysis indicators. More specifically, according to the exploration results arranged in the design stage of hydropower projects, combined with professional specifications and industry experience, and according to the lithology factors of hydropower projects, select multiple rock mass permeability analysis indicators.

[0105] Step S2: Conduct a correlation analysis on the initially determined various rock mass permeability analysis indicators and the rock mass permeability coefficient K, and obtain four rock mass permeability analysis indicators whose correlation with the rock mass permeability coefficient K exceeds the threshold, namely: rock mass buried depth RD, rock quality designation RQD, fracture density feature FSD of the core, and rock mass integrity index RID

[0106] In this process, set and select the rock mass buried depth RD, rock quality designation RQD, fracture density feature FSD of the core, and rock mass integrity index RID of each borehole as optional parameters for permeability prediction analysis, and use the rock mass permeability coefficient K obtained from the water pressure test as the prediction analysis indicator.

[0107] Among them, the lithology of the project area in this example is gneiss. The dip angle trend density of the core structural plane is divided into three dip angle sections of (0, 30], (30, 60], and (60, 90] according to the dip angle degrees. Considering the lithology factors of metamorphic rocks, the weight of the 0 - 30° section is defined as 0.1, the weight of the 30 - 60° section is 0.3, and the weight of the 60 - 90° section is 0.6. The statistical formula for the density characteristics of the core structural plane is as follows:

[0108]

[0109] In the formula, N 0～30 represents the number of structural planes within the dip angle section of (0, 30], and W 1 = 0.1; N 30～60 represents the number of structural planes within the dip angle section of (30, 60], and W 2 = 0.3; N 60～90 represents the number of structural planes within the dip angle section of (60, 90], and W 3 = 0.6; L is the footage depth.

[0110] The schematic representation of this process index is as follows:

[0111]

[0112] In step S2, a correlation analysis is performed on various initially determined rock mass permeability analysis indicators and the rock mass permeability coefficient K, and four rock mass permeability analysis indicators with a correlation with the rock mass permeability coefficient K exceeding the threshold are obtained, namely: rock mass burial depth RD, rock quality designation RQD, density characteristics of core structural planes FSD, and rock mass integrity index RID;

[0113] First, according to the influencing factors of permeability prediction analysis, the parameter values of the rock mass burial depth RD, rock quality designation RQD, density characteristics of core structural planes FSD, and rock mass integrity index RID are pre - processed using the linear normalization method. In this example, the exploration results in the pre - feasibility study to feasibility study stage have 51 boreholes and 531 groups of prediction analysis indicators for permeability prediction analysis.

[0114] Based on the correlation analysis among the permeability prediction analysis indicators, according to the predicted value of the rock mass permeability coefficient K and by superimposing the correlation degree of the permeability indicators, the industry experience can be further improved, and the recommended values of the permeability of various rock masses can be perfected. In this example, the Spearman correlation coefficient method is used to evaluate and analyze the non - parametric statistical index of the dependence between the rock mass permeability coefficient K and the four rock mass permeability analysis index variables. Its value is independent of the specific values of the variables, but is related to the magnitude and order of the variables. Moreover, this method does not require assuming that the variables follow a specific distribution law and has strong robustness to outliers in the data, so its application range is wider.

[0115] Using the correlation analysis method in the aforementioned step S2, the results are shown in the following table:

[0116]

[0117] It can be seen from this that the correlation coefficients between the rock mass burial depth RD, the rock core quality index RQD, the rock mass integrity index RID and the rock mass permeability coefficient K parameter are all greater than 0.6, showing a strong negative correlation. This indicates that as the rock mass burial depth RD increases and the RQD integrity degree increases in the engineering area, the rock mass permeability coefficient K decreases accordingly. There is a certain medium positive correlation between the density characteristics of the rock core structural plane, indicating that the steep dip angle has a certain influence on the permeability of the rock mass in this engineering area. For the parts where the water pressure test is not carried out, appropriate permeability recommended values need to be made in combination with the integrity degree of the rock core.

[0118] Of course, in practical applications, other correlation coefficient analysis methods can also be used, including the Pearson method, the Spearman method, and the Kendall rank correlation method.

[0119] Step S3: Construct multiple training samples and perform normalization processing on each training sample to obtain multiple normalized training samples, forming a training sample set;

[0120] Step S4: Use the training sample set to train the pre-constructed permeability prediction analysis model to obtain a trained permeability prediction analysis model;

[0121] In this process, based on the aforementioned selected, defined and sorted rock mass permeability analysis indexes and the rock mass permeability coefficient K of the existing exploration results, in order to solve the problem of scarce and scattered exploration data, this example uses the ISSA-optimized DELM deep learning algorithm to construct a permeability prediction analysis model.

[0122] DELM is composed of multiple ELM-AE basic units, performs hierarchical unsupervised training and does not require a reverse fine-tuning process, and can minimize the reconstruction error. The input layer weights of the DELM algorithm used in this example are randomly generated by an orthogonal matrix. This orthogonal design can remove the noise other than the features, make the features uniform, and be more linearly independent, thereby enhancing the generalization ability of the system. To further reduce the volatility of the prediction results and optimize and improve the accuracy of the prediction results.

[0123] Furthermore, the ISSA algorithm is used to optimize the convergence accuracy and speed of the DELM algorithm, and a rock mass permeability prediction analysis model based on the deep learning algorithm is constructed for rock mass permeability prediction analysis. The specific flow of the ISSA-DELM optimization algorithm is shown in Figure 2 as follows.

[0124] Specifically, in this example, for the verification test model, 430 sets of predictive analysis indicators are used as training data, and the remaining 101 sets of data are used as predictive verification data. Using the constructed ISSA-DELM rock mass permeability prediction analysis model, after 30 iterations, it basically converges. The goodness of fit R 2 between the predicted value and the actual value of the ISSA-DELM model reaches 0.83. The predicted value basically meets the permeability index under the specified conditions, and generally has a certain predictive effect on the rock mass permeability. See Figure 3 , Figure 4 and Figure 5 . They are respectively: the relationship diagram between the number of iterations and the convergence curve; the relationship diagram between the true value and the predicted value when training with the training set at different numbers of iterations; and the relationship diagram between the true value and the predicted value when training with the test set at different numbers of iterations. It can be seen from the results that the model in this example can predict the permeability of the rock mass without conducting the water pressure test.

[0125] Step S5: For the rock mass that needs to conduct permeability prediction analysis currently, obtain the measured values of the rock mass burial depth RD, rock quality designation RQD, fracture density feature FSD of the core, and rock mass integrity index RID that need to be interpolated and calculated, and input them into the trained permeability prediction analysis model. The trained permeability prediction analysis model outputs the predicted value of the corresponding rock mass permeability coefficient K.

[0126] Specifically, according to the constructed deep learning algorithm permeability index prediction model, the data that has not conducted the water pressure test and needs to conduct permeability analysis is used as the permeability prediction parameter value, the corresponding rock mass permeability index is predicted, and the sensitivity of the rock mass permeability analysis index and the rock mass permeability coefficient K is analyzed, so as to combine the parameter value, predicted value, and industry experience to give reasonable recommended parameters and improve the reference value for hydropower engineering design.

[0127] The following introduces Example 2:

[0128] This example discloses a method for predicting and analyzing the permeability coefficient of rock masses in hydropower engineering, including the following steps:

[0129] A. Define the permeability prediction analysis indicators, which are used to set and select the permeability prediction analysis indicators according to the drilling test results and the influencing factors of the permeability indicators in hydropower engineering.

[0130] Define the permeability prediction analysis indicator A as follows: According to the exploration results arranged in the design stage of the hydropower project, combined with professional specifications and industry experience, and according to the lithology factors of the hydropower project, select the rock mass burial depth, core RQD, fracture density feature of the core, and rock mass integrity index RID as the permeability prediction analysis parameters, and use the permeability coefficient value obtained from the water pressure test as the prediction analysis indicator.

[0131] The core structural plane density characteristics described in Process A are based on the results of drilled cores. The core structural plane data of the existing water pressure test sections and the planned prediction and analysis sections are cataloged. Among them, the structural plane data are independently divided into sections according to the dip angle in combination with the dip angle trend density. Then, according to the core lithology factors and the rock mass integrity factors, the weight ratio of the structural plane dip angle in each section is custom-set, and the proportion density value of the number of developed fracture structural planes within the corresponding footage depth is delimited and statistically analyzed. The formula is as follows:

[0132]

[0133] In the formula, FSD represents the core structural plane density characteristic value; N i represents the number of structural planes in the i-th dip angle section; W i is the weight ratio of the i-th dip angle section, and ∑W i = 1; L is the footage depth.

[0134] Generally, the permeability of the rock mass is related to factors such as lithology and fracture connectivity. For example, sedimentary rocks are mostly due to bedding fracture factors, and gently dipping fractures have a greater influence on permeability. Metamorphic rocks are mostly due to tectonic fracture factors, and steeply dipping fractures have a greater influence on the permeability coefficient.

[0135] B. Sort out the permeability prediction and analysis parameters, and use them to normalize and sort out the permeability prediction and analysis parameter values according to the set parameter indicators.

[0136] Sorting out the permeability prediction and analysis parameter B is based on the permeability prediction and analysis influencing factors defined in Process A, and uses data normalization processing to ensure the equivalence of each factor in the subsequent prediction and analysis model, and eliminate the deviation caused by different dimensions and orders of magnitude.

[0137] The data normalization processing methods described in Process B include linear normalization, standard deviation standardization, non-linear normalization, and decimal scaling standardization, etc. Each method helps to convert the data to a unified dimension, which is a key step in the subsequent deep learning data preprocessing.

[0138] C. Establish a permeability index prediction model, which is used to construct a prediction permeability index model based on the deep learning algorithm with the permeability prediction and analysis parameter values.

[0139] Establishing the permeability prediction and analysis model C selects, defines, and sorts out the permeability prediction and analysis influencing factor parameters and index values of the existing exploration results according to Process A and B. The normalized data of the influencing factors are used as sample parameters, and the water permeability index is used as the sample target. The deep learning algorithm is independently selected, and the sample parameters and sample target data are gradually trained to construct a rock mass permeability index prediction model based on the deep learning algorithm for rock mass permeability prediction and analysis.

[0140] The deep learning algorithms described in Process C mainly include algorithms such as multi-layer perceptron (MLP), convolutional neural network (CNN), recurrent neural network (RNN), long short-term memory network (LSTM), deep extreme learning machine (DELM), etc., as well as further improved and optimized versions based on these basic algorithms.

[0141] D. Predict and analyze the rock mass permeability index, and use the established prediction permeability index model to input the permeability prediction parameter values of the parts where the water pressure test has not been carried out for rock mass permeability prediction and analysis.

[0142] Based on the permeability index prediction model of the deep learning algorithm constructed in Process C, the rock mass permeability index prediction analysis method predicts the corresponding rock mass permeability index by using the data that has not undergone the water pressure test and requires permeability analysis as the permeability prediction parameter values. And analyze the sensitivity of the permeability prediction parameters and the rock mass permeability, so as to combine the parameter values, prediction values and industry experience to give reasonable recommended parameters and improve the reference value of hydropower project design.

[0143] The prediction of the corresponding rock mass permeability index described in Process D is a rock mass permeability index model constructed based on the existing results, which trains the distribution characteristics and laws of the permeability of various underground rock masses, and predicts the permeability index of the underlying rock mass for the parts where the water pressure test results are incomplete.

[0144] A method for predicting and analyzing the rock mass permeability of a hydropower project provided by the present invention can, for pumped storage and water conservancy and hydropower projects with a large project area, select the influencing factor parameters suitable for rock mass permeability prediction and analysis based on the completed drilling and test result data, and combine the water pressure test permeability index of the corresponding factor parameters to construct a rock mass permeability prediction and analysis model based on the deep learning algorithm. It can predict the rock mass permeability of the parts where the water pressure test results are incomplete, and can also analyze the interaction relationship of each factor of the rock mass in the project area on the permeability, which can improve the empirical judgment ability, design quality and efficiency.

[0145] The above are only the preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention.

Claims

1. A method for predicting and analyzing rock mass permeability in hydropower engineering, characterized in that: The following steps are involved: Step S1, preliminarily determining a plurality of rock mass permeability analysis indices related to the rock mass permeability coefficient K; Step S2, performing correlation analysis on various rock mass permeability analysis indicators and rock mass permeability coefficient K determined initially, and obtaining four rock mass permeability analysis indicators having correlation with rock mass permeability coefficient K exceeding a threshold, namely: rock mass burial depth RD, core quality index RQD, core structural surface density feature FSD and rock mass integrity index RID; In step S2, the method for obtaining the core structure surface density feature FSD is: Step S2-1, the drilling footage depth is L; the core structural surface inclination trend density is divided into three inclination sections according to the inclination degree interval, namely: (0, 30], (30, 60], (60, 90]; Step S2-2, counting the number of structural surfaces in each inclination section, respectively: N 0～30 、N 30～60 and N 60～90 ; Step S2-3, according to the rock mass lithology factors currently analyzed, the weights corresponding to the inclination sections are set respectively, namely: W1, W2 and W3; Step S2-4, using the following formula, obtain the core structure surface density feature FSD: Thus, the value of the core structure surface density characteristic FSD is obtained; Step S3, constructing multiple training samples, and normalizing each training sample to obtain multiple normalized training samples to form a training sample set; Each training sample in the training sample set is represented by s = (X, Y); where X is the rock mass permeability analysis index vector, X = (x1, x2, x3, x4), x1, x2, x3, x4, respectively represent the sample values ​​of rock mass burial depth RD, core quality index RQD, core structure surface density feature FSD and rock mass integrity index RID; Y represents the corresponding value of rock mass permeability coefficient K; Step S4, using the training sample set to train the pre-built permeability prediction and analysis model to obtain a trained permeability prediction and analysis model; Step S5, for the rock mass that currently needs to be predicted and analyzed for permeability, obtain the measured values ​​of the rock mass burial depth RD, the core quality index RQD, the core structure surface density characteristic FSD and the rock mass integrity index RID that need to be interpolated and calculated, and input them into the trained permeability prediction and analysis model, and the trained permeability prediction and analysis model outputs the corresponding predicted value of the rock mass permeability coefficient K.

2. A method for predicting and analyzing rock mass permeability of a hydropower project according to claim 1, characterized in that: In step S2, correlation analysis is performed on various rock mass permeability analysis indicators and rock mass permeability coefficient K that are initially determined, specifically: Step S2.1, denote any rock mass permeability analysis index as C, obtain the measured values ​​of M rock mass permeability analysis indexes C, expressed as: C1, C2, ..., C M ; Obtain the rock mass permeability coefficient K corresponding to the measured value of each rock mass permeability analysis index C, which is expressed as: K1, K2, …, K M ; For the measured value sequence of rock permeability analysis index C, C1, C2, ..., C M The measured values ​​of each rock mass permeability analysis index C are expressed in descending order in the measured value sequence of the rock mass permeability analysis index C: Q1, Q2, …, Q M ; For the rock mass permeability coefficient K sequence K1, K2, ..., K M The value of the permeability coefficient K of each rock mass is in the sequence of the permeability coefficient K of the rock mass K1, K2, …, K M The descending numbers in are represented as: P1, P2, ..., P M ; Calculate the difference between sequence number Q1 and sequence number P1 to obtain d1; Calculate the difference between sequence number Q2 and sequence number P2 to obtain d2; And so on Calculate the sequence number Q M and serial number P M The difference between M ; Step S2.2, using the following formula, obtain the correlation coefficient r between the rock mass permeability analysis index C and the rock mass permeability coefficient K s : Where: k = 1, 2, ..., M; Step S2.3, determine the correlation coefficient r s Is it greater than the threshold ε? If so, select the rock permeability analysis index C.

3. A method for predicting and analyzing rock mass permeability of a hydropower project according to claim 1, characterized in that: For metamorphic rocks, the weights are: W1=0.1, W2=0.3, W3=0.

6.

4. A method for predicting and analyzing rock mass permeability of a hydropower project according to claim 1, characterized in that: Step S4 is specifically as follows: Step S4.1, the constructed permeability prediction and analysis model is a DELM permeability prediction and analysis model, including 1 input layer, N layers of hidden layers cascaded forward and backward, and 1 output layer; wherein the input layer has 4 neurons, each hidden layer has 4 neurons, and the output layer has 4 neurons; Step S4.2, randomly generate the weight of each neuron in the input layer through an orthogonal matrix; Step S4.3, for any training sample s = (X, Y), X = (x1, x2, x3, x4), Y represents the corresponding value of the rock mass permeability coefficient K, firstly input x1, x2, x3, x4 into the 4 neurons of the input layer respectively, and each neuron of the input layer outputs x through weight calculation respectively. 11 ,x 21 ,x 31 ,x 41 , forming a matrix X1=(x 11 ,x 21 ,x 31 ,x 41 ); Step S4.4, transform the matrix X1=(x 11 ,x 21 ,x 31 ,x 41 ) is input to the hidden layer of the first layer, and the output matrix of the hidden layer of the first layer is set to X2 = (x 12 ,x 22 ,x 32 ,x 42 ) is equal to the input matrix X1, so the weight vector β2 of the hidden layer of the second layer is obtained by the following formula: β2β2 T ＝I Where: C is the regularization parameter; I is the identity matrix; Step S4.5, transform the matrix X2=(x 12 ,x 22 ,x 32 ,x 42 ) is input to the hidden layer of the second layer, and the output matrix X3 of the hidden layer of the second layer is obtained using the following formula: β2β2 T ＝I At the same time, determine the weight vector β3 of the hidden layer of the third layer: β3=X3Y Step S4.6, input the matrix X3 and the weight vector β3 to the hidden layer of the third layer, and use the same method as step S4.5 to obtain the output matrix X4 of the hidden layer of the third layer and the weight vector β4 of the hidden layer of the fourth layer; and so on, until the output matrix X of the hidden layer of the Nth layer is obtained. N+1 and the weight vector β N+1 ; Step S4.7, transform the matrix X N+1 and the weight vector β N+1 Input to the output layer, the output layer outputs the final output vector Y′; Step S4.8, by comparing the output vector Y′ and the sign value of the rock permeability coefficient K, that is, Y; if both are less than the threshold, then the weight vectors of each layer at this time form the final trained permeability prediction and analysis model; if it is not less than the threshold, then with Y′ and Y being equal as the goal, the least squares method is used to solve, and the weight vectors of each layer after the solution are obtained to form the final trained permeability prediction and analysis model.

Images