Sandstone uniaxial compressive strength prediction method based on SSA-BP neural network

Abstract

The invention discloses a sandstone uniaxial compressive strength prediction method based on an SSA-BP neural network. The sandstone uniaxial compressive strength prediction method comprises the following steps: S1, establishing a basic database; s2, performing data processing on the data obtained in the S1; and S3, a BP neural network is established through the data processed in the S2, and prediction is completed. According to the sandstone uniaxial compressive strength prediction method based on the SSA-BP neural network provided by the invention, the problem that the measured rock mechanical parameter error is relatively large in the prior art is solved.

Description

Technical Field

[0001] The invention belongs to the technical field of rock mechanics and relates to a method for predicting the uniaxial compressive strength of sandstone based on an SSA-BP neural network. Background Art

[0002] Affected by the early sedimentary environment, later tectonic changes and natural weathering, the overlying rock and soil of the coal seam is in a complex geological stress environment with significant characteristics such as discontinuity, heterogeneity and anisotropy. The engineering mechanical properties of the overburden depend on its macroscopic sedimentary environment and microscopic structural characteristics. From the perspective of the macroscopic sedimentary environment, the hydrodynamic conditions of different sedimentary environments are different, resulting in different microstructures such as the composition, particle size, degree of sorting, cementation type and pore filling method of the sedimentary debris. The sedimentary environment includes continental environment, mixed marine and terrestrial environment and marine environment. The specific continental environment also includes terrestrial environment, glacier and desert environment, river environment, lake and swamp environment, and cave environment.

[0003] The engineering geological characteristics of the overburden restrict its deformation and failure behavior under mining conditions, and have a broad fundamental significance for revealing the movement laws of the overburden. During the exploration and supplementary exploration of the mine, a large number of rock samples are taken through the construction of a large number of boreholes and the rock mechanics strength table of the mine is formed. However, due to the limitations of rock sample quality or laboratory errors, the errors of the measured rock mechanics parameters are large under the same depth and lithology conditions. For indoor similar material simulation and numerical simulation, how to select reasonable rock parameters has a great influence on the expected results of the model. Summary of the invention

[0004] The purpose of the present invention is to provide a sandstone compressive strength prediction method based on SSA-BP neural network, which solves the problem of large errors in the rock mechanics parameters measured in the prior art.

[0005] The technical solution adopted by the present invention is a sandstone compressive strength prediction method based on the SSA-BP neural network, which is specifically implemented according to the following steps: S1, establish basic database; S2, performs data processing on the data obtained in S1; S3, establishes a BP neural network through the data processed by S2 to complete the prediction.

[0006] The technical solution of the present invention is also characterized in that: S1 is as follows: by collecting rock mechanics test data from a mine and its surrounding mines, four parameter indicators, namely lithology, burial depth X2, porosity X3 and uniaxial compressive strength Y, are screened out. At the same time, thin sections of cast sandstones of different types in the mine are collected. By measuring the apparent long axis Dmm of the particle size of the sample, the particle size φ value of each sandstone is obtained based on the Wooden-Windward φ value particle size standard. The average particle size of the same type of sandstone is called the lithology characteristic value X n1 The sandstone lithology is replaced by the average value of the lithology characteristic value. n1 , burial depth X n2 , porosity X n3 and uniaxial compressive strength Y n It is called a data group n, where n=1, 2, 3,…, n.

[0007] Lithology characteristic value X in S1 n1 The specific formula is: (1); Where D mm is the apparent long axis of the mineral grains in the sample thin section in the rock sedimentary characteristics identification table; m is the number of the same type of sandstone.

[0008] S2 data processing includes the following steps: S201: remove incomplete data and exclude abnormal data values ​​according to the deviation rate and production practice experience value; S202: Remove the dimension effect, improve the model convergence speed and ensure the fairness of the model, and n1 , X n2 , X n3 and Y n Standardize them separately; S203: Divide the data in the database into a training set, a test set and a validation set according to certain intervals.

[0009] The specific standardization in S202 is: (2); (3); In the formula, x, y are the standardized data; X is the processed database X n1 , X n2 and X n3 ; Y is the uniaxial compressive strength Y in the processed database n ;X max With X min are the maximum and minimum values ​​of the corresponding parameters respectively; Y max With Y min are the maximum and minimum values ​​of the uniaxial compressive parameters, respectively.

[0010] S203 is specifically: Arrange the data sets from small to large in terms of burial depth and mark them with numbers 1, 2, 3, ..., n. Divide the database into training set, validation set and test set in a ratio of 8:1:1. Test set: extract data sets from the database at intervals of 10, and the extracted data sets form the test set; validation set: extract data sets from the remaining data sets at intervals of 10, and the extracted data sets form the test set; Training set: Take the remaining data set after the test set and validation set as the training set.

[0011] S3 specifically includes the following steps: S301, determining the extended structure of the BP neural network, the extended structure includes the number of input layer nodes, the number of hidden layers and their nodes, the output layer nodes, the learning rate and the number of iterations, and at the same time, initializing the threshold and weight value of the BP neural network; S302, optimizing the threshold and weight value of the BP neural network by using the SSA sparrow search algorithm, including determining the proportion of discoverers of the population, the number of population evolutions, and the population size; S303, by analyzing the mean absolute error, mean absolute percentage error, mean square error, root mean square error and determination coefficient R of the BP neural network test set. 2 Metrics measure the accuracy of the model and the predictions made.

[0012] The extended structure of the BP neural network in S301 is as follows: an input layer with 3 nodes, 2 hidden layers, each with h nodes, 1 output layer, one output node, a learning rate of 0.0001 and a maximum iteration of 70 times, where the input layer has 3 nodes, 2 hidden layers, each with h nodes, 1 output layer, one output node, a learning rate of 0.0001 and a maximum iteration of 70 times, n1 , X n2 and Xn3 , the number of nodes in different layers is determined by the following formula: (4); Where h is the number of hidden layer nodes, b is the number of input layer nodes, c is the number of output layer nodes, and a is an adjustment value between 1 and 10.

[0013] S303 is specifically: (5); (6); (7); |(8); (9); In the formula, y i is the original data, yi' is the predicted value of the BP neural network for the i-th data point, and n is the number of predicted points.

[0014] Compared with the prior art, the present invention has the following beneficial effects: The sandstone compressive strength prediction method based on SSA-BP neural network provided by the present invention establishes a basic database by analyzing and processing the statistical table of rock mechanical properties and the rock casting thin section data made since the mine was built, and optimizes the BP neural network by using the SSA sparrow search algorithm to calculate the compressive strength of sandstone based on the buried depth x. n1 、Lite X n2 and porosity X n3 is the independent variable, uniaxial compressive strength Y n The neural network model for automatic prediction of rock strength as the dependent variable fully and effectively utilizes the original basic data of the mine, solving the problems of high cost, great difficulty and large error in the previous process of obtaining rock mechanics parameters. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a flow chart of the sandstone compressive strength prediction method based on the SSA-BP neural network of the present invention. Figure 2 This is a schematic diagram of the effect of the gray fine-grained medium-grained lithic feldspar sandstone casting thin slice in Example 4 of the present invention. Figure 3 This is the grain size characteristic diagram of the gray fine-grained medium-grained lithic feldspar sandstone in Example 4 of the present invention. Figure 4 It is a fitting effect diagram of the BP neural network model in Example 4 of the present invention. DETAILED DESCRIPTION

[0016] The present invention will be further explained below in conjunction with the accompanying drawings and specific embodiments.

[0017] The present invention provides a method for predicting the compressive strength of sandstone based on the SSA-BP neural network. Figure 1 As shown, the specific implementation steps are as follows: S1, establish basic database; S1 is as follows: by collecting rock mechanics test data from a mine and its surrounding mines, four parameter indicators are selected: lithology, burial depth X2, porosity X3 and uniaxial compressive strength Y. At the same time, cast thin sections of different types of sandstone in the mine are collected, and the apparent long axis D of the particle size of the sample is measured. mm Based on the Wooden-Windward φ value particle size standard, the particle size φ value of each sandstone is obtained. The average particle size of the same sandstone is called the lithology characteristic value X n1 , sandstone lithology uses lithology characteristic value X n1 The average value of the lithology characteristic value X n1, burial depth X n2 , porosity X n3 and uniaxial compressive strength Y n , is called a data group n, n=1, 2, 3,…, n.

[0018] Among them, the lithology characteristic value X n1 The specific formula is: (1); Where D mm is the apparent long axis of the mineral grains in the sample thin section in the rock sedimentary characteristics identification table; m is the number of the same type of sandstone.

[0019] S2, performs data processing on the data obtained in S1; S2 is specifically: the lithology characteristic value X obtained by S1 n1 , burial depth X n2 , porosity X n3 and uniaxial compressive strength Y n The four parameter indicators are arranged from small to large according to the burial depth, and the data processing can be divided into the following three steps: S201: De-noising, removing incomplete data; at the same time, excluding abnormal data values ​​according to the deviation rate and production practice experience values; Among them, incomplete data specifically refers to when any parameter in a data set is missing, the data set is deleted from the data set. At the same time, in the structurally undeveloped area, under the same lithology, the burial depth X n2 With porosity X n3 , burial depth X n2 And uniaxial compressive strength Y n It conforms to the logarithmic function relationship and eliminates outliers by logarithmic fitting and setting the threshold of the deviation rate.

[0020] Secondly, the mine has carried out a lot of physical and mechanical tests on the roof and floor of the coal seam during daily production. The average value of the rock mechanical parameters of the coal seam roof is used as the standard value, and the burial depth and uniaxial compressive strength of the coal seam roof are used as the standard data set (X 标1 ,X 标2 ,X 标3 ,Y 标 ).

[0021] S202: Standardization: In order to remove the dimension effect, improve the model convergence speed and ensure the fairness of the model, the X n1 , X n2 , X n3 and Y n Standardize them respectively, the formula is as follows: (2); (3); In the formula, x, y are the standardized data; X is the processed database X n1 , X n2 and X n3 ; Y is the uniaxial compressive strength Y in the processed database n ;X max With X min are the maximum and minimum values ​​of the corresponding parameters respectively; Y max With Y min are the maximum and minimum values ​​of the uniaxial compressive parameters, respectively.

[0022] S203: Divide the data in the database into a training set, a test set and a validation set.

[0023] Arrange the data sets from smallest to largest depth and mark them with numbers 1, 2, 3, ..., n. Divide the database into training set, validation set and test set in a ratio of 8:1:1, including: Test set: extract data groups from the database at intervals of 10, and the extracted data groups constitute the test set; Validation set: extract data groups from the remaining data set at intervals of 10, and the extracted data groups constitute the test set; Training set: Take the remaining data set after the test set and validation set as the training set; S3, establishes a BP neural network through the data processed by S2 to make predictions.

[0024] S3 is specifically: S301, determining the extended structure of the BP neural network, the extended structure includes the number of input layer nodes, the number of hidden layers and their nodes, the output layer nodes, the learning rate and the number of iterations, and at the same time, initializing the threshold and weight value of the BP neural network; The extended structure of the BP neural network in S301 is as follows: an input layer with 3 nodes, 2 hidden layers, each with h nodes, 1 output layer, one output node, a learning rate of 0.0001 and a maximum iteration of 70 times, where the input layer has 3 nodes, 2 hidden layers, each with h nodes, 1 output layer, one output node, a learning rate of 0.0001 and a maximum iteration of 70 times, n1 , X n2 and Xn3 , the number of nodes in different layers is determined by the following formula: (4); Where h is the number of hidden layer nodes, b is the number of input layer nodes, c is the number of output layer nodes, and a is an adjustment value between 1 and 10.

[0025] S302, using the SSA sparrow search algorithm to optimize the threshold and weight value of the BP neural network, first initialize the SSA sparrow population algorithm parameters, including population size, maximum number of iterations, bird flock warning value, discoverer ratio and scout ratio. Then iterate, calculate the individual fitness and get the current optimal solution, and update the positions of the discoverer, joiner and scout. Iterate until the end condition is met, and output the optimal position and the best fitness value; S303, by analyzing the mean absolute error (MAE), mean absolute percentage error (MAPE), mean square error (MSE), root mean square error (RMSE) and determination coefficient (R 2 ) indicators measure the accuracy of the neural network, complete the construction of the BP neural network, and make predictions. Mean absolute percentage error (MAPE), mean square error (MSE), root mean square error (RMSE) and determination coefficient (R 2 ) Specifically: (5); (6); (7); (8); (9); In formulas (5), (6), (7), (8) and (9), y i is the original data, y i' is the predicted value of the BP neural network for the i-th data point, and n is the number of predicted points.

[0026] Example 1 This implementation provides a sandstone compressive strength prediction method based on SSA-BP neural network, such as Figure 1 As shown, the specific implementation steps are as follows: S1, establish basic database; S2, performs data processing on the data obtained in S1; S3, establishes a BP neural network through the data processed by S2 to make predictions.

[0027] S1 is specifically: by collecting rock mechanics test data in a mine and its surrounding mines, the rock type, burial depth X n2 , porosity X n3 and uniaxial compressive strength Y n Four parameter indicators, at the same time, the casting thin sections of different types of sandstone in the mine are collected, and the particle size φ value of each sandstone is obtained by measuring the apparent long axis Dmm of the sample particle size based on the Wooden-Windward φ value particle size standard. The average particle size of the same type of sandstone is called the lithology characteristic value Xn1 The sandstone lithology is replaced by the average value of the lithology characteristic value. n1 , burial depth X n2 , porosity X n3 and uniaxial compressive strength Y n is called a data set n, where n = 1, 2, 3, ..., n. The lithology characteristic value X n1 The specific formula is: (1); Where D mm is the apparent long axis of the mineral grains in the sample thin section in the rock sedimentary characteristics identification table; m is the number of the same type of sandstone.

[0028] Example 2 On the basis of Example 1, S2 comprises the following steps: S201: remove incomplete data and exclude abnormal data values ​​according to the deviation rate and production practice experience value; S202: Remove the dimension effect, improve the model convergence speed and ensure the fairness of the model, and n1 , X n2 , X n3 and Y n Standardize them separately; S203: Divide the data in the database into a training set, a test set and a validation set according to certain intervals.

[0029] Among them, the standardization in S202 is specifically: (2); (3); In the formula, x, y are the standardized data; X is the processed database X n1 , X n2 and X n3 ; Y is the uniaxial compressive strength Y in the processed database n ;X max With X min are the maximum and minimum values ​​of the corresponding parameters respectively; Y max With Y min are the maximum and minimum values ​​of the uniaxial compressive parameters, respectively.

[0030] S203 is specifically: The data sets are arranged from small to large according to the burial depth, and are marked with serial numbers 1, 2, 3, ..., n. The database is divided into a training set, a validation set, and a test set in a ratio of 8:1:1. The test set: data sets are taken from the database at intervals of 10, and the taken data sets constitute the test set; the validation set: data sets are taken from the remaining data sets at intervals of 10, and the taken data sets constitute the test set; Training set: Take the remaining data set after the test set and validation set as the training set.

[0031] Example 3 On the basis of Example 2, S3 specifically includes the following steps: S301, determining the extended structure of the BP neural network, the extended structure includes the number of input layer nodes, the number of hidden layers and their nodes, the output layer nodes, the learning rate and the number of iterations, and at the same time, initializing the threshold and weight value of the BP neural network; S302, using the SSA sparrow search algorithm to optimize the threshold and weight value of the BP neural network, first initialize the SSA sparrow population algorithm parameters, including population size, maximum number of iterations, bird flock warning value, discoverer ratio and scout ratio. Then iterate, calculate the individual fitness and get the current optimal solution, and update the positions of the discoverer, joiner and scout. Iterate until the end condition is met, and output the optimal position and the best fitness value; S303, by analyzing the mean absolute error, mean absolute percentage error, mean square error, root mean square error and determination coefficient R of the BP neural network test set. 2 The indicator measures the accuracy of the BP neural network, completes the construction of the BP neural network, and makes predictions.

[0032] The extended structure of the BP neural network in S301 is specifically as follows: an input layer including 3 nodes, 2 hidden layers, each of which has h nodes, an output layer, an output node, a learning rate set to 0.0001 and a maximum iteration of 70 times, wherein the nodes of the input layer are X n1 , X n2 and Xn3 , the number of nodes in different layers is determined by the following formula: (4); Where h is the number of hidden layer nodes, b is the number of input layer nodes, c is the number of output layer nodes, and a is an adjustment value between 1 and 10.

[0033] S303 is specifically: (5); (6); (7); (8); (9); In formulas (5), (6), (7), (8) and (9), y i is the original data, y i' is the predicted value of the neural network for the i-th data point, and n is the number of predicted points.

[0034] Example 4 This implementation selected the stratigraphic and mechanical parameter data of a mine in Mongolia and Shaanxi for modeling and calculation, such as Figure 1 As shown, specifically: S1, establish basic database: establish a database containing burial depth X n1 , lithology characteristic value X n2 , porosity X n3 and uniaxial compressive strength Y n Basic database of four parameter indicators.

[0035] By collecting rock mechanics test data from the mine and its surrounding mines, four indicators were screened out: lithology, burial depth, porosity, and uniaxial compressive strength. At the same time, cast thin sections of different types of sandstone in the mine were collected, and the apparent long axis D of the particle size of the samples was measured. mm Based on the Wooden-Windward φ value particle size standard, the particle size φ value of each sandstone is obtained, as shown in Table 1. The average particle size of the same sandstone is called the lithology characteristic value X n1 , the sandstone lithology is replaced by the average value of the lithology characteristic value, as shown in Table 2, Figure 2 and Figure 3 As shown, Figure 2 It is a thin section of gray fine-grained feldspar sandstone casting. Figure 3 is the particle size characteristic of gray fine-grained lithic feldspar sandstone, and the lithology characteristic value X n1 , burial depth X n2 、PorosityX n3 and uniaxial compressive strength Y n is called a data set n, n = 1, 2, 3, ..., n, where the lithology characteristic value X n1 The specific formula is: (1); Where D mm is the apparent long axis of the mineral grains in the sample thin section in the rock sedimentary characteristics identification table; m is the number of the same type of sandstone.

[0036] Table 1 Characteristic values ​​of different lithologies (X n1 ) List

[0037] Table 2 Original database

[0038] S2, data processing, the above-obtained lithology characteristic values ​​Xn1, burial depth Xn2, porosity Xn3 and uniaxial compressive strength Y n The four parameter indicators are arranged from small to large according to the burial depth, and the data processing can be divided into the following three steps: S201: De-noising, removing incomplete data; at the same time, excluding abnormal data values ​​according to the deviation rate and production practice experience values; Among them, incomplete data specifically refers to when any parameter in a data set is missing, the data set is deleted from the data set. At the same time, in the structurally undeveloped area, under the same lithology, the burial depth X n2 With porosity X n3 , burial depth X n2 And uniaxial compressive strength Y n It conforms to the logarithmic function relationship and eliminates outliers by logarithmic fitting and setting the threshold of the deviation rate.

[0039] Secondly, the mine has carried out a lot of physical and mechanical tests on the coal seam roof and floor in the daily production process, taking the average value of the rock mechanical parameters of the coal seam roof as the standard value, as shown in Table 3, and taking the burial depth of the coal seam roof and the uniaxial compressive strength of the coal seam roof as the standard data set X 标1 ,X 标2 ,X 标3 With Y 标 In this embodiment, the deviation rate threshold is set to 50%.

[0040] Table 3 Standard values ​​of rock mechanical strength of coal mine roof lithology

[0041] S202: Standardization: In order to remove the dimension effect, improve the model convergence speed and ensure the fairness of the model, the X n1 , X2, X3, and Y n Each column is standardized separately, the formula is as follows: (2); (3); In the formula, x, y are the standardized data; X is the processed database X n1 , X n2 and X n3 ; Y is the uniaxial compressive strength Y in the processed database n ;X max With X min are the maximum and minimum values ​​of the corresponding parameters respectively; Y max With Ymin are the maximum and minimum values ​​of the uniaxial compressive parameters, respectively.

[0042] S203: As shown in Table 4, the data in the database is divided into a training set, a test set and a validation set.

[0043] Arrange the data sets from smallest to largest in terms of burial depth and mark them with serial numbers 1, 2, 3, ..., n. Divide the database into training set, validation set, and test set in a ratio of 8:1:1, including: Test set: extract data groups from the database at intervals of 10, and the extracted data groups constitute the test set; Validation set: extract data groups from the remaining data set at intervals of 10, and the extracted data groups constitute the test set; Training set: Take the remaining data set after the test set and validation set as the training set; Table 4 Database after data processing

[0044] S3, establishes a BP neural network through the data processed by S2 to make predictions.

[0045] S3 is specifically: S301, determine the extended structure of the BP neural network: including the input layer including 3 nodes (X n1 , X n2 , X n3 ), 2 hidden layers (10 nodes per layer), 1 output layer, one output node, the learning rate was set to 0.0001 and the maximum iteration was 70.

[0046] S302, SSA sparrow search algorithm optimization: SSA sparrow population algorithm population size 10, maximum iteration number 30 times, bird flock warning value 0.8, discoverer ratio 20%. Then iterate, calculate individual fitness and get the current optimal solution, and update the positions of discoverers, joiners, and scouts. Iterate until the end condition is met, and output the optimal position and the best fitness value; S303, Error test: By analyzing the mean absolute error (MAE), mean absolute percentage error (MAPE), mean square error (MSE), root mean square error (RMSE) and determination coefficient R of the BP neural network model test set 2 The coefficient index measures the accuracy of the BP neural network model. Among them: Table 5 Error table of sandstone uniaxial compressive strength prediction model based on SSA-BP neural network

[0047] S304: Figure 4As shown, it can be clearly seen that the BP neural network model built based on the SSA-BP neural network sandstone uniaxial compressive strength prediction method can accurately predict future trends based on learning historical data, and can be used as a prediction model for the subsequent uniaxial compressive strength of sandstone in the mine.

Claims

1. The sandstone compressive strength prediction method based on SSA-BP neural network is characterized by: Follow the steps below to implement it: S1, establish basic database; S2, performs data processing on the data obtained in S1; S3, establishes a BP neural network through the data processed by S2 to complete the prediction.

2. The method for predicting the compressive strength of sandstone based on the SSA-BP neural network according to claim 1 is characterized in that: S1 specifically includes: collecting rock mechanics test data in a mine and its surrounding mines, screening out rock properties, burial depth x n2 , porosity X n3 and uniaxial compressive strength Y n Four parameter indicators, at the same time, the casting thin sections of different types of sandstone in the mine are collected, and the particle size φ value of each sandstone is obtained by measuring the apparent long axis Dmm of the sample particle size based on the Wooden-Windward φ value particle size standard. The average particle size of the same type of sandstone is called the lithology characteristic value X n1 The sandstone lithology is replaced by the average value of the lithology characteristic value. n1 , burial depth X n2 , porosity X n3 and uniaxial compressive strength Y n It is called a data group n, where n=1, 2, 3,…, n.

3. The method for predicting the compressive strength of sandstone based on the SSA-BP neural network according to claim 2 is characterized in that: The lithology characteristic value X described in S1 n1 The specific formula is: （1）； Where D mm is the apparent long axis of the mineral grains in the sample thin section in the rock sedimentary characteristics identification table; m is the number of the same type of sandstone.

4. The method for predicting the compressive strength of sandstone based on the SSA-BP neural network according to claim 1 is characterized in that: The S2 comprises the following steps: S201: remove incomplete data and exclude abnormal data values ​​according to the deviation rate and production practice experience value; S202: Remove the dimension effect, improve the model convergence speed and ensure the fairness of the model, and n1 , X n2 , X n3 and Y n Standardize them separately; S203: Divide the data in the database into a training set, a test set and a validation set according to certain intervals.

5. The method for predicting the compressive strength of sandstone based on the SSA-BP neural network according to claim 4 is characterized in that: The standardization described in S202 is specifically: （2）； （3）； In the formula, x, y are the standardized data; X is the processed database X n1 , X n2 and X n3 ; Y is the uniaxial compressive strength Y in the processed database n ;X max With X min are the maximum and minimum values ​​of the corresponding parameters respectively; Y max With Y min are the maximum and minimum values ​​of the uniaxial compressive parameters, respectively.

6. The method for predicting the compressive strength of sandstone based on the SSA-BP neural network according to claim 4 is characterized in that: The S203 is specifically as follows: The data sets are arranged from small to large according to the burial depth, and are marked with serial numbers 1, 2, 3, ..., n. The database is divided into a training set, a validation set, and a test set in a ratio of 8:1:

1. The test set: data sets are taken from the database at intervals of 10, and the taken data sets constitute the test set; the validation set: data sets are taken from the remaining data sets at intervals of 10, and the taken data sets constitute the test set; The training set: the remaining data set after taking the test set and the validation set is used as the training set.

7. The method for predicting sandstone compressive strength based on SSA-BP neural network according to claim 1, characterized in that: The S3 specifically includes the following steps: S301, determining the extended structure of the BP neural network, the extended structure includes the number of input layer nodes, the number of hidden layers and their nodes, the output layer nodes, the learning rate and the number of iterations, and at the same time, initializing the threshold and weight value of the BP neural network; S302, optimizing the threshold and weight value of the BP neural network by using the SSA sparrow search algorithm, including determining the proportion of discoverers of the population, the number of population evolutions, and the population size; S303, by analyzing the mean absolute error, mean absolute percentage error, mean square error, root mean square error and determination coefficient R of the BP neural network test set. 2 The indicator measures the accuracy of the BP neural network and completes the construction and result prediction of the BP neural network.

8. The method for predicting the compressive strength of sandstone based on the SSA-BP neural network according to claim 7, characterized in that: The extended structure of the BP neural network in S301 is specifically as follows: an input layer including 3 nodes, 2 hidden layers, each of which has h nodes, an output layer, one output node, a learning rate set to 0.0001 and a maximum iteration of 70 times, wherein the nodes of the input layer are X n1 , X n2 and Xn3 , the number of nodes in different layers is determined by the following formula: （4）； Where h is the number of hidden layer nodes, b is the number of input layer nodes, c is the number of output layer nodes, and a is an adjustment value between 1 and 10.

9. The method for predicting the compressive strength of sandstone based on SSA-BP neural network according to claim 7, characterized in that: The S303 is specifically as follows: （5）； （6）； （7）； |（8）； （9）； In formulas (5), (6), (7), (8) and (9), y i is the original data, y i' is the predicted value of the BP neural network for the i-th data point, and n is the number of predicted points.

Images