Coal reservoir ground stress prediction method based on ANFIS-PSO model

Abstract

The invention discloses a coal reservoir ground stress prediction method based on an ANFIS-PSO model, and the method comprises the steps: collecting a data sample through on-site logging information, and forming a logging data set; a coal reservoir rock mechanical parameter set is obtained through a uniaxial compression mechanical experiment, a triaxial compression mechanical experiment, a Brazilian splitting mechanical experiment, an on-site hydraulic fracturing experiment and the like; measuring the correlation between the logging data set and the coal reservoir rock mechanical parameter set, determining the input characteristics for predicting the horizontal ground stress, and constructing a total sample set; an ANFIS adaptive neural fuzzy inference system is constructed; and improving the ANFIS model by using a Fletcher-Reeves conjugate gradient method and a PSO (particle swarm optimization) algorithm, so as to obtain an optimized ANFIS-PSO model. After the trained model is obtained, the prediction precision of the model is tested by using the test set according to the evaluation index; and a support is provided for research work related to subsequent coal reservoir reformability evaluation.

Description

Technical Field

[0001] The present invention relates to a coal reservoir geostress prediction method based on the ANFIS-PSO model, which is capable of solving the actual horizontal geostress prediction problem and providing support for the subsequent coal reservoir transformability evaluation. Background Art

[0002] Geostress, that is, the internal stress existing in the earth's crust, is the main driving force of crustal movement and geological tectonic activities. As the natural stress inherent in the rock mass, it is formed by the combined action of multiple forces such as gravity, tectonic force, fluid dynamics and crystallization force. It is usually characterized by three parameters: vertical geostress, maximum horizontal geostress and minimum horizontal geostress. In the exploration and development of coalbed methane, by accurately predicting the state of geostress, it can guide fracturing transformation, affect the permeability of coal seams, evaluate development prospects and promote technological innovation, providing strong support for the sustainable development of the coalbed methane industry. As my country's coalbed methane exploration and development continues to expand to deeper depths, the geological environment it faces is becoming increasingly complex, and the study of geostress has therefore received high attention from all parties.

[0003] So far, the main methods for obtaining geostress include indoor core testing technology, field measurement analysis, logging curve interpretation, seismic prediction method, numerical simulation based on geomechanics, calculation using logging data, and prediction based on prestack seismic inversion. In general, there are many geostress calculation methods based on logging interpretation and numerical simulation. In actual application, they rely on logging acoustic wave data and geostress measured data. However, in the actual field operation process, there are often insufficient experimental test samples, logging, and well test data, which limits the acquisition of geostress. Therefore, in order to calculate the geostress more simply and accurately and obtain the geostress of coal reservoirs more accurately, a coal reservoir geostress prediction method based on the ANFIS-PSO model was invented by taking advantage of the advantages of machine learning and deep learning algorithms, thereby providing a basic basis for geostress prediction and coal reservoir reformability evaluation. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a coal reservoir geostress prediction method based on the ANFIS-PSO model, which can provide a relatively simple and accurate prediction of geostress when test samples and well test data are insufficient, and provide a reference basis for subsequent coal reservoir compressibility evaluation.

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

[0006] A method for predicting coal reservoir ground stress based on ANFIS-PSO model, the method comprising:

[0007] Step 1: Collect data samples through on-site logging data to form a logging data set;

[0008] Step 2, obtaining a set of coal reservoir rock mechanics parameters through uniaxial compression mechanics experiments, triaxial compression mechanics experiments, Brazilian splitting mechanics experiments, field hydraulic fracturing experiments, etc.;

[0009] Step 3: By measuring the correlation between the well logging data set and the coal reservoir rock mechanics parameter set, the input characteristics for predicting horizontal ground stress are determined and the total sample set is constructed;

[0010] Step 4, constructing an ANFIS adaptive neural fuzzy inference system, wherein the ANFIS adaptive neural fuzzy inference system includes a fuzzification layer, a rule layer, a normalization layer, a defuzzification layer, and a summation layer;

[0011] Step 5, using the Fletcher-Reeves conjugate gradient method to improve the ANFIS model and obtain the optimized ANFIS adaptive neuro-fuzzy inference model parameters;

[0012] Step 6: Use the particle swarm optimization algorithm to update the particle position, and continuously update the network parameters of the ANFIS algorithm through cyclic iteration to obtain the optimized ANFIS-PSO model;

[0013] Step 7: Train the ANFIS-PSO model according to the training set. After obtaining the trained model, use the test set to test the prediction accuracy of the model according to the evaluation index.

[0014] Step 8. Finally, the ground stress is effectively predicted based on the conventional logging curve and the trained ANFIS-PSO model.

[0015] A further technical solution is that the logging data include natural gamma GR, deep directional resistivity LLD, shallow directional resistivity LLS, compensated neutron CNL, compensated acoustic wave AC, compensated density DEN, wellbore CAL, and vertical depth TVD; indoor experiments and field data collection obtain coal reservoir rock mechanical parameters, including Young's modulus, Poisson's ratio, horizontal ground stress and tensile strength.

[0016] A further technical solution of the previous step is that the coal reservoir rock mechanics parameter set includes Young's modulus, Poisson's ratio, horizontal ground stress and tensile strength.

[0017] A further technical solution is that the specific process of step 3 is:

[0018] Step 3.1, using the Pearson correlation coefficient to measure the correlation between logging parameters and mechanical parameters, and determine the input characteristics for predicting horizontal geostress;

[0019] Correlation analysis formula:

[0020] Where: R is the correlation coefficient; xi ,y i are two sets of sample data; μ x , μ y The average value of the two sample data in group 2; σ x , σ y is the standard deviation of the two sets of sample data;

[0021] Step 3.2: Normalize the dataset before training the model.

[0022] Normalization formula:

[0023] Where: x i is the value after data normalization; x is the original logging data; x min is the minimum value of the sample data; x max is the maximum value of the sample data;

[0024] Step 3.3: Each set of input feature values ​​constitutes a complete sample, and the sum constitutes the total sample set. The total sample set is randomly divided into training set and test set in a ratio of 8:2.

[0025] A further technical solution is that in step 4, ANFIS realizes fuzzification processing through a neural network, converts it into fuzzy membership, then generates fuzzy reasoning based on the fuzzy membership, and finally obtains the final output result through fuzzy interface processing. The specific steps are as follows:

[0026] Step 4.1, calculate the result of each layer node i:

[0027] O 1,i =μA i (x) (Formula 3);

[0028]

[0029] O 2,i =w i =μA i (x).μB i (y) i=1,2 (Formula 5);

[0030]

[0031] Step 4.2. Calculate the equation used for the ANFIS layer output:

[0032]

[0033] Where: μA i (x) is the MF from x to fuzzy set A; c i and σ i is the premise parameter; μBi (y) is the MF from y to fuzzy set B; w i is the result of the normalization layer; (p i ,q i , r i ) is the result parameter set; 1,i , O 2,i , O 3,i and O 4,i are the results of nodes i from the first to the fourth layer respectively; 5,i It is the result calculated by ANFIS algorithm.

[0034] A further technical solution is to improve the ANFIS model in step 5 using the Fletcher-Reeves conjugate gradient method and obtain optimized ANFIS adaptive neuro-fuzzy inference model parameters. The specific steps include:

[0035] Step 5.1. The first step is to search along the negative gradient direction of the target error function:

[0036] D0=G0 (Formula 9);

[0037] Where: G0 is the direction vector of the negative gradient in the first step; D0 is the search direction in the first step;

[0038] Step 5.2: Perform a linear search along the current direction to determine the optimal distance to move in that direction:

[0039] W n+1 =W n +α n D n (Formula 10);

[0040] Where: W n is the current weight vector; D n is the current search direction; α n is the learning rate;

[0041] Step 5.3: Search in the direction of the negative gradient search in the previous step:

[0042] D n =G n +β n D n-1 (Formula 11);

[0043] Where: G n is the direction vector of the current negative gradient; β n is the search step length;

[0044] Step 5.4: Search step length β n The Fletcher-Reeves conjugate gradient method used:

[0045]

[0046] In order to ensure the convergence of the algorithm and make the value of the objective error function continue to decrease, the direction should be corrected when the number of training times is an integer multiple of the number of weights, that is, β n = 0, and at the same time, judgment and correction must be made in each training process, that is, G n T D n When ≥0, the negative gradient direction is used as the new search direction, and the search continues in the descending direction. Finally, when the minimum value of the set target error function or the maximum number of training times is reached, the search ends.

[0047] A further technical solution is that the step 6 uses a particle swarm optimization algorithm to update the particle position, and continuously updates the network parameters of the ANFIS algorithm through cyclic iterations. The formula is as follows:

[0048]

[0049] Where: x i k+1 To update the position; x i k is the current position; v i k+1 is the update speed; v i k is the current speed; c1 and c2 are learning coefficients; rand1 and rand2 are random numbers distributed in the interval [0,1]; pbest i k is the best position for an individual; gbest i k is the optimal position of the group; i is the number of particles; k is the number of iterations.

[0050] A further technical solution is that the evaluation indicators of step 7 are root mean square error RMSE, mean absolute error MAE, and determination coefficient R 2 , which is used as an evaluation index to assess the accuracy and reliability of the model;

[0051] The evaluation index formula is as follows:

[0052]

[0053]

[0054] Beneficial effects of the present invention:

[0055] The invention comprises the following steps: firstly, selecting a mine in the study area for conventional logging interpretation of geostress to form a sample data set, performing feature selection and normalization on the logging data, establishing an ANFIS horizontal geostress prediction model based on particle swarm optimization, applying the ANFIS-PSO model to the horizontal geostress prediction of the mine, and performing root mean square error (RMSE), mean absolute error (MAE), and determination coefficient (R) to obtain the horizontal geostress prediction model. 2 It is used as an evaluation indicator to assess the accuracy and reliability of the model and provide support for the subsequent transformability evaluation. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 A schematic diagram of the flow of the geostress prediction method based on the ANFIS-PSO model provided by the present invention;

[0057] Figure 2 A heat map showing the correlation between the original logging data of Well W15 and the mechanical parameters according to an embodiment of the present invention;

[0058] Figure 3 This is an interpretation diagram of the W15 well in-situ stress well according to an embodiment of the present invention;

[0059] Figure 4 ANFIS model structure diagram of an embodiment of the present invention;

[0060] Figure 5 The figure is an evolution diagram of the RMSE value of the ANFIS model and the ANFIS-PSO model according to the embodiment of the present invention with the number of iterations;

[0061] Figure 6 is a W1 geostress profile diagram of an embodiment of the present invention;

[0062] Figure 7 is a W6 geostress profile diagram of an embodiment of the present invention; DETAILED DESCRIPTION

[0063] The technical solution of the present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. Obviously, the specific implementation of the present invention is not limited to the specific embodiments provided.

[0064] like Figure 1 As shown, a method for predicting coal reservoir ground stress based on ANFIS-PSO model is characterized by comprising the following steps:

[0065] Step 1, collecting data samples through on-site logging data, wherein the logging data includes natural gamma ray GR, deep direction resistivity LLD, shallow direction resistivity LLS, compensated neutron CNL, compensated acoustic wave AC, compensated density DEN, wellbore CAL, and vertical depth TVD;

[0066] Step 2: Indoor experiments and field data collection are used to obtain coal reservoir rock mechanical parameters, including Young's modulus, Poisson's ratio, horizontal ground stress and tensile strength.

[0067] Step 3: By measuring the correlation between logging parameters and mechanical parameters, the input characteristics of predicting horizontal geostress are determined and the total sample set is constructed;

[0068] Step 3.1, using the Pearson correlation coefficient to measure the correlation between logging parameters and mechanical parameters, and determine the input characteristics for predicting horizontal geostress;

[0069] Correlation analysis formula:

[0070] Where: R is the correlation coefficient; x i ,y i are two sets of sample data; μ x , μ y The average value of the two sample data in group 2; σ x , σ y is the standard deviation of the two sets of sample data;

[0071] Step 3.2: Normalize the dataset before training the model.

[0072] Normalization formula:

[0073] Where: x i is the value after data normalization; x is the original logging data; x min is the minimum value of the sample data; x max is the maximum value of the sample data;

[0074] Step 3.3, each set of input feature values ​​constitutes a complete sample, and the sum constitutes the total sample set. The total sample set is randomly divided into training set and test set in a ratio of 8:2;

[0075] Step 4, constructing an ANFIS adaptive neural fuzzy inference system, wherein the ANFIS adaptive neural fuzzy inference system includes a fuzzification layer, a rule layer, a normalization layer, a defuzzification layer, and a summation layer;

[0076] Step 4.1, calculate the result of each layer node i:

[0077] O 1,i =μA i (x) (Formula 3);

[0078]

[0079] O 2,i =w i =μA i(x).μB i (y) i=1,2 (Formula 5);

[0080]

[0081] Step 4.2. Calculate the equation used for the ANFIS layer output:

[0082]

[0083] Where: μA i (x) is the MF from x to fuzzy set A; c i and σ i is the premise parameter; μB i (y) is the MF from y to fuzzy set B; w i is the result of the normalization layer; (p i ,q i , r i ) is the result parameter set; 1,i , O 2,i , O 3,i and O 4,i are the results of nodes i from the first to the fourth layer respectively; 5,i It is the result calculated by ANFIS algorithm;

[0084] Step 5, using the Fletcher-Reeves conjugate gradient method to improve the ANFIS model and obtain the optimized ANFIS adaptive neuro-fuzzy inference model parameters;

[0085] Step 5.1. The first step is to search along the negative gradient direction of the target error function:

[0086] D0=G0 (Formula 9);

[0087] Where: G0 is the direction vector of the negative gradient in the first step; D0 is the search direction in the first step;

[0088] Step 5.2: Perform a linear search along the current direction to determine the optimal distance to move in that direction:

[0089] W n+1 =W n +α n D n (Formula 10);

[0090] Where: W n is the current weight vector; D n is the current search direction; α n is the learning rate;

[0091] Step 5.3: Search in the direction of the negative gradient search in the previous step:

[0092] D n =G n +β n D n-1 (Formula 11);

[0093] Where: G n is the direction vector of the current negative gradient; β n is the search step length;

[0094] Step 5.4: Search step length β n The Fletcher-Reeves conjugate gradient method used:

[0095]

[0096] In order to ensure the convergence of the algorithm and make the value of the objective error function continue to decrease, the direction should be corrected when the number of training times is an integer multiple of the number of weights, that is, β n = 0, and at the same time, judgment and correction must be made in each training process, that is, G n T D n ≥0, use the negative gradient direction as the new search direction and search along the descending direction.

[0097] Finally, the search ends when the minimum value of the set objective error function or the maximum number of training times is reached.

[0098] Step 6: Use the particle swarm optimization algorithm to update the particle position and continuously update ANFIS through loop iteration

[0099] The network parameters of the algorithm are as follows

[0100]

[0101] Where: x i k+1 To update the position; x i k is the current position; v i k+1 is the update speed; v i k is the current speed; c1 and c2 are learning coefficients; rand1 and rand2 are random numbers distributed in the interval [0,1]; pbest i k is the best position for an individual; gbest i k is the optimal position of the group; i is the number of particles; k is the number of iterations.

[0102] Step 7: Train the ANFIS-PSO model according to the training set. After obtaining the trained model, use the test set to test the prediction accuracy of the model according to the evaluation index.

[0103]

[0104] Where: m is the total number of evaluation samples; y i is the actual sample value of the i-th sample; is the predicted value of the i-th sample; R 2 is the coefficient of determination; RMSE is the root mean square error; MAE is the mean absolute error;

[0105] Step 8. Finally, the ground stress is effectively predicted based on the conventional logging curve and the trained ANFIS-PSO model.

[0106] Taking the section from 286m to 500m of the coal reservoir in Well W15 in the Wenjiaba block of Zhina Coalfield as an example,

[0107] A total of 216 sample points were collected at 0.4m intervals through field logging data. Hotspots of correlation between original logging data and mechanical parameters of Well W15 in the study area Figure 2 The input characteristics for predicting horizontal geostress are determined to be natural gamma GR, shallow direction resistivity LLS, compensated acoustic wave AC, compensated density DEN, and vertical depth TVD. The horizontal geostress profile of W15 in the study area is obtained through well logging interpretation, including the horizontal minimum principal stress and the horizontal maximum principal stress. The results are as follows: Figure 2 Table 1 shows some of the original sample data. Before training the model, the data set needs to be normalized;

[0108] Table 1 Part of the original sample data

[0109] TVD AC LLS CNL GR CAL DEN 286.51 420.9167 440.3452 21.407127 81 222.92 1.563834 287.31 435.0208 496.0865 25.934365 67 230.775 1.487096 288.11 374.0833 29.1639 30.859041 204 228.25 2.300525 288.91 253.2292 37.7827 19.628339 72 214.625 2.583605 289.71 204.3333 343.3616 21.172381 46 215 2.551205 290.51 196.2917 525.6284 21.518102 43 214.325 2.566211 291.31 228.5417 306.6904 17.470171 46 213.96001 2.552569 292.11 267.3958 260.9323 16.778379 57 214.4 2.570304 292.91 315.7188 74.696 19.931438 128 215.825 2.516416 293.71 330.8958 38.0715 15.721202 119 223.68 2.478558 294.51 608.8438 69.1962 33.42544 89 216.825 2.471055

[0110] For the process of predicting Young's modulus, the initial ANFIS model is first constructed. Figure 3 As shown in the figure, the membership function is set to 3, and the Fletcher-Reeves conjugate gradient method is used to improve it, reduce its structural complexity, and increase the running rate. Then, the PSO particle swarm algorithm is used to optimize the effect of the ANFIS model. During the PSO search process, the optimal structural parameters are determined through sensitivity analysis, the population size is set to 40, the learning factors C1 and C2 are set to 1 and 2, and the inertia weight is set to 0.7. During the ANFIS-PSO model training process, the RMSE value of each iteration is recorded.

[0111] In order to verify the reliability of the geostress prediction method based on the ANFIS-PSO model proposed in this paper, the ANFIS model and the ANFIS-PSO model are selected for comparison. Taking the prediction of the minimum horizontal principal stress (σh) as an example, the RMSE value evolution curves of the ANFIS model and the ANFIS-PSO model are compared ( Figure 4 ), it can be found that the ANFIS-PSO model converges faster and the error decreases in a curve. After 228 iterations, the RMSE value drops from 0.60 to about 0.28 and stabilizes, while the RMSE value of the ANFIS model only drops from 0.60 to 0.31.

[0112] The above two models are used to predict reservoir in-situ stress respectively, and the predicted data are compared with the actual data. The root mean square error RMSE, mean absolute error MAE, and determination coefficient R are used to calculate the predicted data. 2 The evaluation indicators are analyzed and the results are shown in Table 2.

[0113] Table 2 Comparison of prediction results before and after model optimization

[0114]

[0115] From the comparison results, it can be seen that compared with the ANFIS model, the RMSE and MAE of the ANFIS-PSO model in the experiment of predicting the minimum horizontal principal stress were reduced by 14.8% and 11.7% respectively; in the experiment of predicting the maximum horizontal principal stress, the RMSE and MAE of the ANFIS-PSO model were reduced by 27.2% and 23.1% respectively. In the prediction of horizontal geostress, the ANFIS-PSO model was 96% and 97% respectively, and the performance was better than the ANFIS model. It can be seen that the use of PSO to optimize the ANFIS model can significantly improve the accuracy of the prediction, and has the ability to solve the actual horizontal geostress prediction problem, providing support for the subsequent transformation evaluation.

[0116] Based on the ANFIS-PSO horizontal geostress model established above and combined with the conventional logging data in the study area, the geostress logging interpretation was carried out for the W1 and W6 wells in the study area, and the corresponding single-well geostress profile was established. Figure 5 and Figure 6 ; The maximum horizontal principal stress in the 100.0m-350.0m section of Well W1 in the study area is 2.166-8.36MPa, with an average of 5.32MPa; the minimum horizontal principal stress is 1.47-5.92MPa, with an average of 3.75MPa; the maximum horizontal principal stress in the 300.0m-500.0m section of Well W6 in the study area is 5.06-11.05MPa, with an average of 8.49MPa; the minimum horizontal principal stress is 3.66-7.57MPa, with an average of 6.04MPa.

Claims

1. A method for predicting coal reservoir ground stress based on ANFIS-PSO model, characterized by: The method comprises the following steps: Step 1: Collect data samples through on-site logging data to form a logging data set; Step 2, obtaining a set of coal reservoir rock mechanics parameters through uniaxial compression mechanics experiments, triaxial compression mechanics experiments, Brazilian splitting mechanics experiments, field hydraulic fracturing experiments, etc.; Step 3: By measuring the correlation between logging parameters and mechanical parameters, the input characteristics of predicting horizontal geostress are determined and the total sample set is constructed; Step 4, constructing an ANFIS adaptive neural fuzzy inference system, wherein the ANFIS adaptive neural fuzzy inference system includes a fuzzification layer, a rule layer, a normalization layer, a defuzzification layer, and a summation layer; Step 5, using the Fletcher-Reeves conjugate gradient method to improve the ANFIS model and obtain the optimized ANFIS adaptive neuro-fuzzy inference model parameters; Step 6: Use the particle swarm optimization algorithm to update the particle position, and continuously update the network parameters of the ANFIS algorithm through cyclic iteration to obtain the optimized ANFIS-PSO model; Step 7: Train the ANFIS-PSO model according to the training set. After obtaining the trained model, use the test set to test the prediction accuracy of the model according to the evaluation index. Step 8. Finally, the ground stress is effectively predicted based on the conventional logging curve and the trained ANFIS-PSO model.

2. The method for predicting ground stress based on the ANFIS-PSO model according to claim 1, characterized in that: The logging data set includes natural gamma ray GR, deep direction resistivity LLD, shallow direction resistivity LLS, compensated neutron CNL, compensated acoustic wave AC, compensated density DEN, wellbore CAL, and vertical depth TVD; The rock mechanical parameters of coal reservoirs obtained through indoor experiments and field data collection include Young's modulus, Poisson's ratio, horizontal ground stress and tensile strength.

3. The method for predicting ground stress based on the ANFIS-PSO model according to claim 1, characterized in that: The coal reservoir rock mechanics parameter set includes Young's modulus, Poisson's ratio, horizontal ground stress and tensile strength.

4. The method for predicting ground stress based on the ANFIS-PSO model according to claim 1, characterized in that: The specific process of step 3 is as follows.

5. Step 3.1, use the Pearson correlation coefficient to measure the correlation between the well logging data set and the coal reservoir rock mechanics parameter set to determine the input characteristics for predicting horizontal ground stress; Correlation analysis formula: Where: R is the correlation coefficient; x i ,y i are two sets of sample data; μ x , μ y The average value of the two sample data in group 2; σ x , σ y is the standard deviation of the two sets of sample data; Step 3.2: Normalize the dataset before training the model. Normalization formula: Where: x i is the value after data normalization; x is the original logging data; x min is the minimum value of the sample data; x max is the maximum value of the sample data; Step 3.3: Each set of input feature values ​​constitutes a complete sample, and the sum constitutes the total sample set. The total sample set is randomly divided into training set and test set in a ratio of 8:

2.

6. The method for predicting geostress based on the ANFIS-PSO model according to claim 1, characterized in that: In step 4, ANFIS realizes fuzzification processing through neural network, converts it into fuzzy membership, then generates fuzzy reasoning according to the fuzzy membership, and finally obtains the final output result through fuzzification interface processing. The specific steps are as follows: Step 4.1, calculate the result of each layer node i: O 1,i = μA i (x) (Equation 3); O 2,i = w i = μA i (x).μB i (y) i = 1, 2 (Equation 5); Step 4.

2. Calculate the equation used for the ANFIS layer output: Where: μA i (x) is the MF from x to fuzzy set A; c i and σ i is the premise parameter; μB i (y) is the MF from y to fuzzy set B; w i is the result of the normalization layer; (p i ,q i , r i ) is the result parameter set; 1,i , O 2,i , O 3,i and O 4,i are the results of nodes i from the first to the fourth layer respectively; 5,i It is the result calculated by ANFIS algorithm.

7. The method for predicting geostress based on the ANFIS-PSO model according to claim 1, characterized in that: In step 5, the ANFIS model is improved by using the Fletcher-Reeves conjugate gradient method, and the optimized ANFIS adaptive neural fuzzy inference model parameters are obtained. The specific steps include: Step 5.

1. The first step is to search along the negative gradient direction of the target error function: D0=G0 (Formula 9); Where: G0 is the direction vector of the negative gradient in the first step; D0 is the search direction in the first step; Step 5.2: Perform a linear search along the current direction to determine the optimal distance to move in that direction: W n+1 = W n + α n D n (Equation 10); Where: W n is the current weight vector; D n is the current search direction; α n is the learning rate, Step 5.3: Search in the direction of the negative gradient search in the previous step: D n =G n +b n D n-1 (expression 11); Where: G n is the direction vector of the current negative gradient; β n is the search step length; Step 5.4: Search step length β n The Fletcher-Reeves conjugate gradient method used: In order to ensure the convergence of the algorithm and make the value of the objective error function continue to decrease, the direction should be corrected when the number of training times is an integer multiple of the number of weights, that is, β n = 0, and at the same time, judgment and correction must be made in each training process, that is, G n T D n When ≥0, the negative gradient direction is used as the new search direction, and the search continues in the descending direction. Finally, when the minimum value of the set target error function or the maximum number of training times is reached, the search ends.

8. The method for predicting geostress based on the ANFIS-PSO model according to claim 1, characterized in that: The step 6 uses the particle swarm optimization algorithm to update the particle position, and continuously updates the network parameters of the ANFIS algorithm through cyclic iteration. The formula is as follows: Where: x i k+1 To update the position; x i k is the current position; v i k+1 is the update speed; v i k is the current speed; c1 and c2 are learning coefficients; rand1 and rand2 are random numbers distributed in the interval [0,1]; pbest i k is the best position for an individual; gbest i k is the optimal position of the group; i is the number of particles; k is the number of iterations.

9. The method for predicting geostress based on the ANFIS-PSO model according to claim 1, characterized in that: The evaluation indicators of step 7 are root mean square error RMSE, mean absolute error MAE, and determination coefficient R 2 , which is used as an evaluation indicator to assess the accuracy and reliability of the model.

Images