A method for predicting rock mass roughness coefficient JRC based on an Internet platform neural network model

Abstract

The present invention discloses a method for predicting the rock roughness coefficient JRC based on an Internet platform neural network model, which is applicable to the occasion of predicting the rock roughness coefficient JRC of a structural surface by the surface morphology characteristics of the structural surface, and belongs to the field of rock and soil mechanics technology. The present invention considers the relationship between the five parameters of angle average value, angle expectation value, height average value, height variation coefficient and back slope characteristic value K and the shear behavior of the structural surface, and uses the results of quantitative characterization of the five parameters and the rock roughness coefficient JRC of the structural surface as the input and output parameters of the model respectively. Finally, the GA-BP neural network model is used to train and predict the samples, and it is verified that the method has higher accuracy.

Description

Technical Field

[0001] The invention belongs to the field of rock and soil mechanics, and specifically relates to a method for predicting a roughness coefficient (JRC) of a neural network on an Internet platform. Background Art

[0002] A large number of engineering practices have confirmed that the instability problems of many engineering rock masses are related to the existence of structural surfaces. Shear failure of structural surfaces is the main failure mode of engineering rock masses. The stability of rock masses depends to a large extent on the shear strength of structural surfaces, and the roughness of structural surfaces is one of the important indicators affecting the shear strength characteristics. However, at present, some two-dimensional statistical parameter and three-dimensional statistical parameter methods for calculating the roughness of structural surfaces proposed by domestic and foreign scholars have certain shortcomings.

[0003] BP neural network is a commonly used artificial neural network model with strong learning ability and adaptability. The GA genetic algorithm can be used to globally search for the best initial weights and thresholds of the BP neural network, so that the BP neural network can converge to the global optimal solution more quickly and improve the prediction accuracy of the model. The GA genetic algorithm has a mutation operation that can effectively prevent the BP neural network from falling into the local optimal situation.

[0004] In view of this, the present invention is proposed. Summary of the invention

[0005] The technical problem to be solved by the present invention is to overcome the deficiencies of the prior art and provide a method for predicting the roughness coefficient JRC of an Internet platform neural network, thereby solving the problems raised in the above-mentioned background technology.

[0006] In order to solve the above technical problems, the basic concept of the technical solution adopted by the present invention is:

[0007] A method for predicting roughness coefficient (JRC) of a neural network on an Internet platform, comprising:

[0008] Step 1: Use a 3D laser scanner to extract the point cloud data of the structural surface, use 3D printing technology to print the 3D structural surface and cast the experimental sample, conduct a shear test on the structural surface sample, and analyze the angle, height, and back slope characteristics;

[0009] Step 2: Use MATLAB programming software to quantitatively characterize the angle, height, and back slope characteristics;

[0010] Quantify the angle features as shown in formula (1) (2):

[0011]

[0012]

[0013] Where: is the apparent inclination angle corresponding to the two-dimensional curve; i is the number of points, which gradually increases along the shear direction; x i is the position of the i-th point in the structural surface reference plane; y i For x i The height of the point structure surface relative to the reference surface; The corresponding apparent inclination angle of the three-dimensional structure surface; Figure 1 Medium θ i is the effective surface apparent inclination, is the azimuth, β i is the inclination angle of the convex element of the structural surface; s is the shear direction, n0 is the normal vector of the shear plane, n is the external normal of the convex element of the structural surface, and n1 is the projection vector of n on the shear plane.

[0014] The height feature is quantified as shown in formula (3):

[0015] h=y i -y min (3)

[0016] Where: i is the number of points, which gradually increases along the shear direction; y min is the minimum height of the structural surface relative to the reference surface; i For x i The height of the point structure surface relative to the reference surface, the height of the two-dimensional curve is y i The relative height of the point, the height y of the micro-convex body on the surface of the three-dimensional structure surface i That is its center height.

[0017] The back slope characteristics are quantified as shown in formula (4) (5):

[0018]

[0019]

[0020] Where: K 2D is the back slope characteristic value corresponding to the two-dimensional structural surface; K 3D is the back slope characteristic value corresponding to the three-dimensional structural surface; is the apparent inclination angle θ i When it is less than zero, it corresponds to the area of ​​the two-dimensional structural surface; is the apparent inclination angle θ i When it is less than zero, it corresponds to the volume of the three-dimensional structural surface; S t V is the total area of ​​the two-dimensional structure surface relative to the lowest point of its surface; t It is the total volume of the structural surface relative to the lowest point of the structural surface.

[0021] Step 3: Determine the average inclination angle, expected inclination angle, average height, height variation coefficient and back slope characteristic value after quantifying the angle, height and back slope characteristics as the input of the GA-BP neural network, and the structural surface roughness coefficient JRC as the output of the neural network model.

[0022] Construct a BP neural network based on MATLAB programming software, set the classification method of neural network samples, the number of neurons, the number of hidden layers, the accuracy of the network and the algorithm used.

[0023] The activation function f(x) formula of the BP neural network is as follows:

[0024]

[0025] The number of nodes in the hidden layer is determined by empirical formula and multiple experiments. The expression of the empirical formula is shown in formula (7):

[0026]

[0027] The output formula of the hidden layer is as follows:

[0028]

[0029] The output formula of the output layer is as follows:

[0030]

[0031] The network error formula is as follows:

[0032] e k =Y k -O k (10)

[0033] The update formula from the input layer to the hidden layer weights is as follows:

[0034]

[0035] The update formula from the hidden layer to the output layer weights is as follows:

[0036] ω jk =ω jk +νH j e k (12)

[0037] The update formula of the hidden layer node threshold is as follows:

[0038]

[0039] The update formula of the output layer node threshold is as follows:

[0040] b k =b k +ve k (14)

[0041] Where n is the number of input layer nodes, l is the number of hidden layer nodes, and m is the number of output layer nodes. ij is the weight from the input layer to the hidden layer, ω ij is the weight from the hidden layer to the output layer, c j is the threshold from the input layer to the hidden layer, c k is the threshold from the hidden layer to the output layer, and v is the learning rate.

[0042] Construct BP neural network after GA optimization:

[0043] Population initialization: randomly generate a set of initial solutions as the population, where each individual represents a set of weights and thresholds of the BP neural network;

[0044] Fitness calculation: The inverse of the mean square error obtained during the BP neural network training process is used as the fitness value. The formula is as follows:

[0045]

[0046] Where: k is the coefficient, n is the number of neural network output nodes, yi is the expected output of the ith node of the neural network, o i is the predicted output of the i-th node of the neural network.

[0047] Selection: Select a certain number of individuals to enter the next generation based on the fitness value. The probability P of each individual i being selected i The formula is as follows:

[0048]

[0049] Where: N is the population size, F i is the fitness value of the ith individual.

[0050] Crossover: Randomly select two individuals from the selected set for crossover to produce new offspring, the jth individual N j and k individuals N k The crossover process at position i is as follows:

[0051]

[0052] Where a is a random number in the interval [0,1];

[0053] Mutation: Perform mutation operations on newly generated offspring to increase population diversity.

[0054] Update the population: replace some individuals in the original population with the newly generated offspring to form a new generation of population, and select the i-th gene a of the i-th individual ij To perform mutation, the operation process is as follows:

[0055]

[0056]

[0057] Where a max and a min are the maximum and minimum values ​​of individual i, s is the number of iterations, G max is the maximum number of evolutions, and r is a random number in the interval [0,1].

[0058] After adopting the above technical solution, the present invention has the following beneficial effects compared with the prior art. Of course, any product implementing the present invention does not necessarily need to achieve all the advantages described below at the same time:

[0059] The present invention can establish a highly nonlinear relationship between the influencing factors in the shearing process of the structural surface and the roughness coefficient of the structural surface, and accurately predict the roughness of the structural surface.

[0060] The specific implementation modes of the present invention are further described in detail below in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] The drawings described below are only some embodiments. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0062] In the figure:

[0063] Figure 1 This is the relationship between the apparent inclination angle of the structural surface unit and the shear direction;

[0064] Figure 2 A schematic diagram of the process of inventing a GA-BP neural network prediction model for structural surface roughness;

[0065] Figure 3 It is a schematic diagram of the structure of the GA-BP neural network;

[0066] Figure 4 It is a schematic diagram of the GA-BP neural network training process;

[0067] Figure 5 This is the result diagram of neural network training;

[0068] Figure 6 This is a summary graph of the neural network error.

[0069] It should be noted that these drawings and textual descriptions are not intended to limit the conceptual scope of the present invention in any way, but are intended to illustrate the concept of the present invention for those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0070] The present invention will now be described in further detail with reference to the accompanying drawings.

[0071] See also Figure 1-6 As shown, in this embodiment, a method for predicting the roughness coefficient JRC of an Internet platform neural network is provided, including:

[0072] Step 1: Use a 3D laser scanner to extract the point cloud data of the structural surface, use 3D printing technology to print the 3D structural surface and cast the experimental sample, conduct a shear test on the structural surface sample, and analyze the angle, height, and back slope characteristics;

[0073] Step 2: Use MATLAB programming software to quantitatively characterize the angle, height, and back slope characteristics;

[0074] Quantify the angle features as shown in formula (1) (2):

[0075]

[0076]

[0077] Where: is the apparent inclination angle corresponding to the two-dimensional curve; i is the number of points, which gradually increases along the shear direction; x i is the position of the i-th point in the structural surface reference plane; y i For x i The height of the point structure surface relative to the reference surface; The corresponding apparent inclination angle of the three-dimensional structure surface; Figure 1 Medium θ i is the effective surface apparent inclination, is the azimuth, β i is the inclination angle of the convex element of the structural surface; s is the shear direction, n0 is the normal vector of the shear plane, n is the external normal of the convex element of the structural surface, and n1 is the projection vector of n on the shear plane.

[0078] The height feature is quantified as shown in formula (3):

[0079] h=y i -y min (3)

[0080] Where: i is the number of points, which gradually increases along the shear direction; y min is the minimum height of the structural surface relative to the reference surface; i For xi The height of the point structure surface relative to the reference surface, the height of the two-dimensional curve is y i The relative height of the point, the height y of the micro-convex body on the surface of the three-dimensional structure surface i That is its center height.

[0081] The back slope characteristics are quantified as shown in formula (4) (5):

[0082]

[0083]

[0084] Where: K 2D is the back slope characteristic value corresponding to the two-dimensional structural surface; K 3D is the back slope characteristic value corresponding to the three-dimensional structural surface; is the apparent inclination angle θ i When it is less than zero, it corresponds to the area of ​​the two-dimensional structural surface; is the apparent inclination angle θ i When it is less than zero, it corresponds to the volume of the three-dimensional structural surface; S t V is the total area of ​​the two-dimensional structure surface relative to the lowest point of its surface; t It is the total volume of the structural surface relative to the lowest point of the structural surface.

[0085] Step 3: Determine the average inclination angle, expected inclination angle, average height, height variation coefficient and back slope characteristic value after quantifying the angle, height and back slope characteristics as the input of the GA-BP neural network, and the structural surface roughness coefficient JRC as the output of the neural network model.

[0086] Construct a BP neural network based on MATLAB programming software, set the classification method of neural network samples, the number of neurons, the number of hidden layers, the accuracy of the network and the algorithm used.

[0087] The activation function f(x) formula of the BP neural network is as follows:

[0088]

[0089] The number of nodes in the hidden layer is determined by empirical formula and multiple experiments. The expression of the empirical formula is shown in formula (7):

[0090]

[0091] The output formula of the hidden layer is as follows:

[0092]

[0093] The output formula of the output layer is as follows:

[0094]

[0095] The network error formula is as follows:

[0096] e k =Y k -O k (10)

[0097] The update formula from the input layer to the hidden layer weights is as follows:

[0098]

[0099] The update formula from the hidden layer to the output layer weights is as follows:

[0100]

[0101] The update formula of the hidden layer node threshold is as follows:

[0102]

[0103] The update formula of the output layer node threshold is as follows:

[0104] b k =b k +ve k (14)

[0105] Where n is the number of input layer nodes, l is the number of hidden layer nodes, and m is the number of output layer nodes. ij is the weight from the input layer to the hidden layer, ω ij is the weight from the hidden layer to the output layer, c j is the threshold from the input layer to the hidden layer, c k is the threshold from the hidden layer to the output layer, and v is the learning rate.

[0106] Construct BP neural network after GA optimization:

[0107] Population initialization: randomly generate a set of initial solutions as the population, where each individual represents a set of weights and thresholds of the BP neural network;

[0108] Fitness calculation: The inverse of the mean square error obtained during the BP neural network training process is used as the fitness value. The formula is as follows:

[0109]

[0110] Where: k is the coefficient, n is the number of output nodes of the neural network, y i is the expected output of the ith node of the neural network, o i is the predicted output of the i-th node of the neural network.

[0111] Selection: Select a certain number of individuals to enter the next generation based on the fitness value. The probability P of each individual i being selected i The formula is as follows:

[0112]

[0113] Where: N is the population size, F i is the fitness value of the ith individual.

[0114] Crossover: Randomly select two individuals from the selected set for crossover to produce new offspring, the jth individual N j and k individuals N k The crossover process at position i is as follows:

[0115]

[0116] Where a is a random number in the interval [0,1];

[0117] Mutation: Perform mutation operations on newly generated offspring to increase population diversity.

[0118] Update the population: replace some individuals in the original population with the newly generated offspring to form a new generation of population, and select the i-th gene a of the i-th individual ij To perform mutation, the operation process is as follows:

[0119]

[0120]

[0121] Where a max and a min are the maximum and minimum values ​​of individual i, s is the number of iterations, G max is the maximum number of evolutions, and r is a random number in the interval [0,1].

[0122] Example 1: Sample data comes from 105 sets of structural surfaces collected, and specific parameters and test results are shown in Table 1. The shear strength is back-calculated from the predicted value of structural surface roughness obtained by the GA-BP neural network model and compared with the original value to verify the accuracy of the trained GA-BP neural network.

[0123] The neural network is set up, the number of nodes in the input layer is set to 5, and the average inclination angle, expected inclination angle, average height, coefficient of variation of height and back slope characteristic value K are used as input parameters of the model. The number of nodes in the output layer is set to 1, and the structural surface roughness coefficient JRC is used as the model output. The number of nodes in the hidden layer is set to 6. During training, the input parameters are divided into three groups, which are used as training data, verification data and test data respectively. Only the training data participates in the training, and the other two parts of the data do not participate in the training and are used for testing. The error tolerance is set to 1e-6, and the number of iterations and learning rate are set to 1000 and 0.01 respectively. In the GA genetic algorithm parameters, the population size is set to 61 and the evolutionary generation is set to 200.

[0124] Table 1: Specific parameters and test results of 105 groups of structural surfaces

[0125]

[0126]

[0127]

[0128]

[0129]

[0130]

[0131]

[0132]

[0133]

[0134]

[0135]

[0136]

[0137]

[0138] When training the GA-BP neural network, the program randomly divides 100 groups of sample data into training samples and test samples. Figure 5 The comparison chart of the training output value of the structural surface roughness coefficient corresponding to each group of samples and the actual output value of the model is shown. The maximum error of the test data is basically controlled within 3, the RMSE is also fluctuating around 0.3, and the MAPE of the test data is within 5%, and the error is within an acceptable range. Five groups of data were further selected for verification, such as Figure 6The maximum error of the shear strength estimated by the JRC value obtained by the GA-BP model is 15.91% and the average error is 7.25%, which is smaller than the results calculated by other methods, indicating that the training results of the GA-BP neural network are relatively ideal.

[0139] The present invention is not limited to the above-mentioned embodiments. Anyone should be aware that any structural changes made under the enlightenment of the present invention, and any technical solutions that are the same or similar to the present invention, fall within the protection scope of the present invention. The technology, shape, and structural parts not described in detail in the present invention are all well-known technologies.

Claims

1. A method for predicting rock mass roughness coefficient JRC based on an Internet platform neural network model, characterized in that: include: Step 1: Use a 3D laser scanner to extract the point cloud data of the structural surface, use 3D printing technology to print the 3D structural surface and cast the experimental sample, conduct a shear test on the structural surface sample, and analyze the angle, height, and back slope characteristics; Step 2: Use MATLAB programming software to quantitatively characterize the angle, height, and back slope characteristics; Step 3: Determine the angle average value, angle expectation value, height average value, height variation coefficient and back slope characteristic value after quantifying the angle, height and back slope characteristics as the input of the GA-BP neural network, and the structural surface rock mass roughness coefficient JRC as the output of the neural network model; In step 2, the angle feature is quantified as shown in formula (1) (2): Where: is the apparent inclination angle corresponding to the two-dimensional curve; i is the number of points, which gradually increases along the shear direction; x i is the position of the i-th point in the structural surface reference plane; y i For x i The height of the point structure surface relative to the reference surface; is the apparent inclination angle of the three-dimensional structure surface; θ i is the effective surface apparent inclination, is the azimuth, β i is the inclination angle of the microelement of the convex body of the structural surface; In step 2, the height feature is quantified as shown in formula (3): h=y i -y min (3) Where: i is the number of points, which gradually increases along the shear direction; y min is the minimum height of the structural surface relative to the reference surface; i For x i The height of the point structure surface relative to the reference surface, the height of the two-dimensional curve is y i The relative height of the point, the height y of the micro-convex body on the surface of the three-dimensional structure surface i That is, its center height; In step 2, the back slope characteristics are quantified, as shown in formulas (4) and (5): Where: K 2D is the back slope characteristic value corresponding to the two-dimensional structural surface; K 3D is the characteristic value of the back slope surface corresponding to the three-dimensional structural surface; is the apparent inclination angle θ i When it is less than zero, it corresponds to the area of ​​the two-dimensional structural surface; is the apparent inclination angle θ i When it is less than zero, it corresponds to the volume of the three-dimensional structural surface; S t V is the total area of ​​the two-dimensional structure surface relative to the lowest point of its surface; t It is the total volume of the structural surface relative to the lowest point of the structural surface.

2. According to claim 1, a method for predicting rock mass roughness coefficient JRC based on an Internet platform neural network model is characterized in that: In step three, a BP neural network is constructed based on MATLAB programming software, and the classification method of the neural network samples, the number of neurons, the number of hidden layers, the accuracy of the network, and the algorithm used are set.

3. The method for predicting rock mass roughness coefficient JRC based on an Internet platform neural network model according to claim 1 is characterized in that: In step three, construct the BP neural network after GA optimization: Population initialization: randomly generate a set of initial solutions as the population, where each individual represents a set of weights and thresholds of the BP neural network; Fitness calculation: The inverse of the mean square error obtained during the BP neural network training process is used as the fitness value. The formula is as follows: Where: k is the coefficient, n is the number of output nodes of the neural network, y i is the expected output of the ith node of the neural network, o i is the predicted output of the i-th node of the neural network.

4. The method for predicting rock mass roughness coefficient JRC based on an Internet platform neural network model according to claim 3 is characterized in that: Selection: A certain number of individuals are selected to enter the next generation according to the fitness value, and the probability P of each individual i being selected i The formula is as follows: Where: N is the population size, F i is the fitness value of the ith individual.

Images