A fault geologic cavern surrounding rock deformation prediction method based on excavation sequence optimization and NPDOA-LSTM model

Abstract

The application provides a fault geologic cavern surrounding rock deformation prediction method based on excavation sequence optimization and an NPDOA-LSTM model, step 1, a fault cavern excavation model is established. Step 2, different excavation sequence working conditions are simulated. Step 3, a real-time surrounding rock monitoring system is arranged, and the cavern is excavated according to the optimal excavation sequence. Step 4, the monitoring data are pretreated, and the correlation of the deformation decomposition term and the temperature and stress data is calculated. Step 5, the model is initialized and trained. The data set is divided, and the data are normalized. Step 6, the deformation prediction result is reconstructed, and the prediction effect is evaluated. The fault geologic cavern excavation sequence optimization can be realized, and the surrounding rock deformation is reduced.

Description

Technical Field

[0001] This invention relates to the field of underground engineering safety control technology, and more specifically to a method for predicting the deformation of surrounding rock in fault geological caverns based on excavation sequence optimization and NPDOA-LSTM model. Background Technology

[0002] In recent years, with the development of underground engineering, these projects have increasingly faced complex geological conditions such as deep burial, high ground stress, and traversing adverse geological structures. Among these, faults have a particularly significant impact on the stability of the surrounding rock of tunnels. Fault zones exhibit well-developed joints and poor rock integrity, easily forming stress concentration zones under excavation disturbance, leading to significant surrounding rock deformation and a high risk of geological problems such as spalling, collapse, and rockbursts, posing a serious threat to the safety of construction personnel and the progress of the project. The deformation of the surrounding rock caused by excavation disturbance varies significantly under different excavation sequences. Optimizing the excavation sequence for tunnels in fault geology can greatly reduce the deformation caused by excavation disturbance. Furthermore, the deformation of the surrounding rock after excavation is uncertain, especially in fault-related tunnels. Therefore, predicting the deformation of the surrounding rock can guide timely on-site support and prevent construction disasters.

[0003] Currently, research on cavern excavation mainly focuses on monitoring and assessing the stability of surrounding rock. Some scholars have optimized the excavation sequence for twin-tunnel and multi-chamber excavations; however, optimization for cavern excavation in fault geology requires further in-depth research. In terms of surrounding rock deformation prediction, numerical simulation and prediction models are primarily used, but these methods have certain limitations. Numerical simulation, for example, operates on a time-step basis, making it difficult to reflect the true changes in surrounding rock deformation over time. With the development of artificial intelligence, prediction models are increasingly being applied in deformation prediction. Patents CN113408190A and CN117113842A respectively use the Bayes algorithm and particle swarm optimization algorithm to optimize LSTM models for predicting surrounding rock deformation. The Bayes algorithm requires high-quality sample sets and is not suitable for complex geological surrounding rock deformation data. The particle swarm optimization algorithm is prone to getting trapped in local optima and its computational efficiency is significantly reduced for multi-dimensional input problems. Patent CN116817777A predicts rock deformation based on time-series data. However, due to the complex influence of excavation methods and geological conditions on rock deformation, this method has poor accuracy in predicting long-term rock deformation. Furthermore, these methods lack a real-time monitoring and management system for rock deformation, failing to guarantee the timeliness of the predictions. Therefore, there is an urgent need to propose a method for controlling and predicting rock deformation in fault-resistant caverns based on excavation sequence optimization and prediction models. This method would have practical value and significant importance in guiding safe construction of fault-resistant caverns. Summary of the Invention

[0004] The technical problem this invention aims to solve is to provide a method for predicting the deformation of surrounding rock in fault geological caverns based on excavation sequence optimization and an NPDOA-LSTM model. This method can optimize the excavation sequence of fault geological caverns and reduce surrounding rock deformation. Furthermore, based on real-time monitoring results of surrounding rock deformation, temperature, and stress, the hyperparameters of the LSTM model are optimized using the Neural Population Dynamics Optimization Algorithm (NPDOA) to predict surrounding rock deformation, guiding reasonable support during construction and reducing the occurrence of construction disasters.

[0005] To achieve the above-mentioned technical features, the objective of this invention is as follows: a method for predicting the deformation of surrounding rock in fault geological caverns based on excavation sequence optimization and an NPDOA-LSTM model, the method comprising the following steps:

[0006] S1, Establish a fault cavern excavation model:

[0007] Obtain relevant data on fault caverns and establish a numerical model of fault geological caverns based on the relevant data;

[0008] S2 simulates different excavation sequence conditions:

[0009] Based on the numerical model of the geological cavern of the S1 fault layer, different excavation sequence conditions were formulated, and numerical simulations were performed on the different sequence conditions to obtain the optimal excavation sequence condition.

[0010] S3, deploy a real-time surrounding rock monitoring system and excavate the tunnel according to the optimal excavation sequence:

[0011] During the excavation of the tunnel in accordance with the optimal excavation sequence in S2, the surrounding rock deformation, stress and temperature monitoring data are monitored in real time by deploying a surrounding rock monitoring system and connecting it to the monitoring information management system.

[0012] S4, preprocess the monitoring data and calculate the correlation between the deformation decomposition terms and the monitoring data:

[0013] The monitoring data obtained in S3 is preprocessed by the monitoring information management system and then decomposed into trend items and periodic items. The correlation between the decomposed items and the monitoring data is calculated, and the monitoring data with high correlation are selected to build a prediction model.

[0014] S5, initialize the model and train the model;

[0015] Using the monitoring data selected in S4, the hyperparameters in the LSTM model were optimized using the NPDOA optimization algorithm, and the LSTM model was trained. Then, the prediction results of the deformation trend term and the periodic term were obtained through the NPDOA-LSTM prediction model.

[0016] S6, Reconstruct the deformation prediction results and evaluate the prediction effect.

[0017] Preferably, S1 specifically includes:

[0018] Through preliminary field investigation and physical model tests, parameters of fault, surrounding rock, geostress, and cavern size were obtained, and a numerical model of the fault geological cavern was established using 3DEC discrete element software.

[0019] Preferably, S2 specifically includes:

[0020] The excavation section is divided into excavation zones to generate different excavation sequence conditions. In the numerical model of the fault geological cavern, monitoring points for the displacement of the top arch and sidewalls are set up, and numerical simulations are performed for different excavation sequence conditions.

[0021] By comparing the displacement of monitoring points under each working condition, the excavation sequence with the smallest average displacement of the surrounding rock was selected as the optimal excavation sequence.

[0022] Preferably, S3 specifically includes:

[0023] Integrated fiber optic grating sensors are installed at the top arch and left and right sidewalls of the excavated tunnel section and connected to the monitoring information management system to form a surrounding rock monitoring system;

[0024] The fault cavern was excavated according to the optimal excavation sequence, and the deformation, stress, and temperature data of the surrounding rock were monitored in real time.

[0025] Preferably, S4 specifically includes:

[0026] The monitoring information management system acquires the surrounding rock monitoring data of each monitoring point and preprocesses the surrounding rock monitoring data. The preprocessing methods include outlier removal, missing data filling, and data denoising.

[0027] The deformation monitoring data was decomposed into trend and periodic components using the moving average method.

[0028] The correlation between the deformation decomposition term and the monitoring data is calculated using the maximum information coefficient method, and the monitoring data with high correlation are selected to construct the prediction model.

[0029] Preferably, S5 specifically includes:

[0030] The monitoring dataset selected in S4 is divided and the data is normalized.

[0031] Initialize the LSTM model parameters, and set the initial population size and number of iterations for NPDOA;

[0032] The NPDOA optimization algorithm is used to optimize the hyperparameters in the LSTM model.

[0033] The optimal hyperparameters were fed into the LSTM model for training, and the prediction results of the deformation trend term and the periodic term were obtained respectively.

[0034] Preferably, S6 specifically includes:

[0035] The surrounding rock deformation trend term and periodic term predicted by the NPDOA-LSTM model are added together to obtain the surrounding rock deformation prediction result.

[0036] By calculating the coefficient of determination R 2 Mean absolute error MAE and root mean square error RMSE The predictive performance of the prediction model is evaluated.

[0037] Preferably, the normalization process in S4 is as follows:

[0038] ;

[0039] In the formula, The data is after normalization; Represents the original data; This represents the maximum value in the dataset; This represents the minimum value in the dataset;

[0040] The decomposition method for the trend term of deformation monitoring data in S4 is as follows, assuming the original displacement time series is:

[0041] ;

[0042] Trend Item The calculation formula is:

[0043] ;

[0044] In the formula, The total number of data sets; for t The cumulative displacement at any given time; For the movement cycle.

[0045] Preferably, the NPDOA optimization algorithm specifically includes:

[0046] 1) Initialize the neural population;

[0047] By randomly generating several neural populations in the neural state space The neural state, among which It refers to the number of neural populations;

[0048] 2) Attractor trend strategy;

[0049] The neural states of a neural population tend to be multi-attractor in the neural state space. By using the optimal neural state as an attractor, the neural states of the remaining neural population randomly converge to one of these attractors.

[0050] 3) Coupling interference strategy;

[0051] The neural state of a specific neural population can be affected by the neural state of other neural populations. The influence of other neural populations can be simulated by using a combination of two couplings.

[0052] 4) Information projection strategy;

[0053] Information exchange and interaction between neural populations are carried out through a communication subspace. In the NPDOA algorithm, a matrix representing adjacency and communication strength is used to represent the communication subspace of information projection between neural populations.

[0054] Ultimately, the neural state of each neural population is influenced by the attractor and the neural states of other neural populations.

[0055] Preferably, the determination coefficient in step 6 R 2 Mean absolute error MAE and root mean square error RMSE The formula is as follows:

[0056] ;

[0057] ;

[0058] ;

[0059] In the formula, For the sample size, This represents the actual value of the surrounding rock deformation. This represents the predicted value of surrounding rock deformation. It is the average value of the actual deformation of the surrounding rock.

[0060] The present invention has the following beneficial effects:

[0061] 1. This invention has good applicability to fault cavern excavation simulation using 3DEC discrete element software. By dividing the cavern into excavation zones, different excavation sequence conditions are generated, and the different excavation sequence conditions are optimized by combining 3DEC numerical simulation. This effectively controls the deformation of the surrounding rock caused by fault geological cavern excavation and avoids construction disasters caused by excavation.

[0062] 2. This invention utilizes an integrated fiber optic grating sensor with modular features, enabling simultaneous monitoring of deformation, temperature, and stress data. This reduces on-site construction difficulty and improves the efficiency of measuring point installation. The monitoring data is uploaded to a cloud server in real-time via a wireless base station, and can be processed in real-time within a monitoring information management system to obtain real-time, high-precision surrounding rock monitoring data. This provides a high-quality data foundation for model training, reduces errors from manual processing, and improves data processing efficiency.

[0063] 3. This invention uses a monitoring information management system to process monitoring data, which can fully utilize server resources, save manpower, and effectively improve data processing efficiency. The moving average method can fully explore the trend and periodic characteristics of deformation monitoring data, increasing the number of data features. The maximum information coefficient method can handle complex relationships between variables by calculating the correlation between deformation trend terms and periodic terms and temperature and stress monitoring data, avoiding errors caused by feature data with low correlation in model training.

[0064] 4. The LSTM model of this invention effectively solves the gradient explosion and gradient vanishing problems that occur in traditional recurrent neural network prediction by incorporating three gating mechanisms. This model has a good ability to capture long-term dependencies in time series data and can predict the trend of surrounding rock deformation over time, making it suitable for the field of surrounding rock deformation prediction.

[0065] 5. The NPDOA optimization algorithm of this invention, by simulating the dynamic process of a neural population, demonstrates strong adaptability in solving high-dimensional nonlinear problems. Specifically, the attractor trend strategy drives the neural population towards optimal decision-making, ensuring exploitation capability; the coupling interference strategy, by coupling with other neural populations, deviates the neural population from the attractor, improving exploration capability; and the information projection strategy controls communication between neural populations, realizing the transition from exploration to exploitation. The optimal hyperparameters of the LSTM model are obtained through the NPDOA optimization algorithm, improving the model's prediction accuracy and efficiency. Attached Figure Description

[0066] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0067] Figure 1 A flowchart of the method provided by the present invention.

[0068] Figure 2 This is a diagram showing the zoning of the tunnel excavation and the layout of displacement measuring points provided by the present invention.

[0069] Figure 3 This is a schematic diagram of the surrounding rock monitoring system provided by the present invention.

[0070] Figure 4 The flowchart of the NPDOA-LSTM model provided by this invention.

[0071] Figure 5 The flowchart of the NPDOA optimization algorithm provided by this invention.

[0072] Figure 6 The prediction results and prediction effect diagrams provided by this invention.

[0073] Figure 7 The prediction effect diagram provided by the present invention. Detailed Implementation

[0074] The embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0075] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0076] Example 1:

[0077] See Figure 1 This invention provides a method for predicting the deformation of surrounding rock in fault geological caverns based on excavation sequence optimization and an NPDOA-LSTM model, comprising the following steps:

[0078] Step 1: Establish a fault cavern excavation model. Through preliminary field investigation and physical model tests, obtain parameters of the fault, surrounding rock, geostress, and cavern dimensions. Use 3DEC discrete element software to establish a numerical model of the fault geological cavern.

[0079] Step 2: Simulate different excavation sequence conditions. Divide the excavation section into excavation zones to generate different excavation sequence conditions. Set up displacement monitoring points for the crown and sidewalls in the model, and perform numerical simulations for each excavation sequence condition. By comparing the displacement magnitudes of the monitoring points for each condition, select the excavation sequence condition with the smallest average displacement of the surrounding rock as the optimal excavation sequence condition.

[0080] Step 3: Deploy a real-time surrounding rock monitoring system and excavate the tunnel according to the optimal excavation sequence. Integrated fiber optic grating sensors are installed at the arch and sidewalls of the excavated section and connected to the monitoring information management system. The fault tunnel is excavated according to the preferred excavation sequence, and the deformation, stress, and temperature data of the surrounding rock are monitored in real time.

[0081] Step 4: Preprocess the monitoring data and calculate the correlation between the deformation decomposition term and the temperature and stress data. Obtain the surrounding rock monitoring data from each monitoring point through the monitoring information management system and preprocess the data. Processing methods include outlier removal, missing data imputation, and data denoising. The moving average method is used to decompose the deformation monitoring data into trend and periodic terms. The maximum information coefficient (MIC) method is used to calculate the correlation between the deformation decomposition term and the monitoring data, and monitoring data with high correlation are selected to construct a prediction model.

[0082] Step 5: Initialize and train the model. Divide the dataset and normalize the data. Initialize the LSTM model parameters, set the initial population size and number of iterations for NPDOA, use the NPDOA optimization algorithm to optimize the hyperparameters in the LSTM model, and train the model using the optimal hyperparameters to obtain the prediction results for the deformation trend term and the periodic term.

[0083] Step 6: Reconstruct the deformation prediction results and evaluate the prediction effect. Add the surrounding rock deformation trend term and periodic term predicted by the NPDOA-LSTM model to obtain the surrounding rock deformation prediction result. Calculate the coefficient of determination. R 2 Mean absolute error MAE and root mean square error RMSE The predictive performance of the prediction model is evaluated.

[0084] Furthermore, the method for decomposing the trend term of the deformation monitoring data is as follows: assuming the original displacement time series is:

[0085] ;

[0086] Trend Item The calculation formula is:

[0087] ;

[0088] In the formula, The total number of data sets; for t The cumulative displacement at any given time; For the movement cycle.

[0089] The normalization process is as follows:

[0090] ;

[0091] In the formula, The data is after normalization; Represents the original data; This represents the maximum value in the dataset; This represents the minimum value in the dataset.

[0092] Furthermore, the basic steps of the NPDOA optimization algorithm are as follows:

[0093] (1) Initialize neural populations. The optimization problem is solved by simulating the process of multiple neural populations with the same number of neurons performing cognitive activities. For a given set of neural populations, the initialization process is as follows: The problem of dimensional optimization, the solution to the problem For having The neural states of a population of neurons. Each decision variable in the solution. This represents a neuron in a neural population, and its value indicates the firing frequency of that neuron. Target value This corresponds to the decisions or actions made by the brain. Several neural populations are randomly generated in the neural state space. The neural state, among which It refers to the number of neural populations.

[0094] (2) Attractor trend strategy. The neural states of the neural population tend to converge to multiple attractors in the neural state space. This algorithm uses the optimal neural state as the attractor, and the neural states of the remaining neural population randomly converge to one of these attractors.

[0095] ;

[0096] In the formula, It is the scaling factor of the attractor trend. It is a random number in [0,1]. Is The Middle A random value, yes random values ​​in Is The Middle The neural state of individual neural populations It is the first Covariance of individual neural states The zero-mean Gaussian noise vector For the peak values ​​of all neural states, It is the smallest boundary of all neural states.

[0097] (3) Coupling Interference Strategy. The neural states of a specific neural population can also be affected by the neural states of other neural populations. It is usually assumed that the interaction between neural populations occurs through additive or diffusion coupling. This algorithm considers the existence of these two couplings in neural population dynamics. In additive coupling, the input of a neural population is a function of the sum of the neural states of other neural populations. To simplify the calculation, the algorithm represents the neural states of other neural populations as the average of the sum of all neural states, as shown in the following formula:

[0098] ;

[0099] In the formula, A random number in the range [0, 0.5]. The number of neural populations.

[0100] In diffuse coupling, the input of a neural population is a function of the sum of the differences between its neural state and the neural states of other neural populations, as shown in the following formula:

[0101] ;

[0102] In the formula, A random number in the range [0, 0.5]. It is a neural population The Middle The neural state of a neural population.

[0103] The proposed algorithm uses a combination of two couplings to simulate the effects of other neural populations, as shown in the following formula:

[0104] ;

[0105] In the formula, This is the scaling factor for the coupling perturbation.

[0106] (4) Information projection strategy. Information exchange and interaction between neural populations are carried out through a communication subspace. In the NPDOA algorithm, a matrix representing adjacency and communication strength is used to represent the communication subspace of information projection between neural populations.

[0107] ;

[0108] In the formula, It is a 1×D matrix with binary random values; A 1×D matrix with randomly generated values ​​in [0,1]. It is the dynamic vector obtained by the attractor trend strategy.

[0109] ;

[0110] In the formula, It is a 1×D matrix with binary random values; A 1×D matrix with randomly generated values ​​in [0,1]. The dynamic vector obtained by the coupling interference strategy; A random number in the range [0,1]. The number of evaluation criteria for the function; This represents the maximum function evaluation number. As the iteration process progresses, The effect of coupling disturbances gradually increases, while the effect of coupling disturbances gradually decreases.

[0111] Ultimately, the neural state of each neural population is influenced by the attractor and the neural states of other neural populations. The neural states of the current updated neural populations are as follows:

[0112]

[0113] Furthermore, the determination coefficient in step 6 R 2 Mean absolute error MAE and root mean square error RMSE The formula is as follows:

[0114] ;

[0115] ;

[0116] ;

[0117] In the formula, For the sample size, This represents the actual value of the surrounding rock deformation. This represents the predicted value of surrounding rock deformation. It is the average value of the actual deformation of the surrounding rock.

[0118] Example 2:

[0119] See Figure 1-7 This embodiment provides a method for predicting the deformation of surrounding rock in fault geological caverns based on excavation sequence optimization and NPDOA-LSTM model, including the following steps:

[0120] 1) Through preliminary field investigation and physical model tests, the fault, surrounding rock, geostress parameters, and cavern dimensions for this embodiment were obtained. This embodiment is a horseshoe-shaped cavern section with a height of 20m, a width of 20m, and a length of 10m. The width of the central guide tunnel is 8m. The fault dip angle is 10°, the width is 0.1m, and it passes through the cavern's arch, 5m away from the arch. A numerical model of the fault geological cavern was established based on 3DEC discrete element software to avoid the influence of boundary effects. The model size is 150m × 150m × 10m (height × width × length). Boundary conditions were set using the tectonic geostress field. The yielding failure of the surrounding rock was simulated using the Mohr-Coulomb criterion, and the yielding of the fault was simulated using the Coulomb slip criterion.

[0121] 2) The excavation section was divided into three layers: I, II, and III. Each layer was further divided into a central area and a sidewall area, numbered 1 for the central area and 2 for the sidewall area. There were two excavation sequences for each layer: excavating the central area first or excavating the sidewall area first. A total of eight excavation sequence scenarios were obtained by excavating layer by layer from the top arch to the bottom. Displacement monitoring points for the top arch and sidewalls were set in the middle of the model tunnel section, at distances of 1.5m, 3.5m, and 6.5m from the excavation surface. The monitoring points for the left and right sidewalls were 5m from the bottom slab. Each scenario was simulated sequentially using a high-performance host computer, with an excavation advance of 5m. Figure 2 As shown. Due to the prominent deformation of the shallow surrounding rock, this embodiment uses the average displacement value of the 1.5m measuring point under each working condition as an indicator for comparison. It is found that the displacement value of each measuring point is the smallest when the excavation sequence is Ⅰ1-Ⅰ2-Ⅱ1-Ⅱ2-Ⅲ1-Ⅲ2. This excavation sequence is selected as the actual excavation sequence in this embodiment.

[0122] 3) Deploy a real-time surrounding rock monitoring system and excavate the cavern according to the preferred excavation sequence. Drill monitoring holes at the top arch and left and right sidewalls. The monitoring holes at the top arch include those at the left and right arch shoulders and the main arch, symmetrically arranged at the arch shoulders. The drilling angle is 45°. The monitoring holes at the left and right sidewalls are 10m above the bottom slab. Each monitoring hole is drilled through pre-excavated anchor observation holes. The diameter of the monitoring hole is larger than the outer diameter of the monitoring module. In this embodiment, the diameter of the monitoring hole is 0.5m. An integrated fiber optic grating sensor is used, which can monitor the surrounding rock deformation, stress, and temperature data in real time. The integrated fiber optic grating sensor is placed in the monitoring hole, with the monitoring module 1.5m, 3.5m, and 6.5m away from the excavation surface of the cavern. Micro-expansion cement grouting is used for backfilling. The fault cavern is excavated according to the preferred excavation sequence, with an excavation advance of 5m. All monitoring data is uploaded to the cloud server in real time via a wireless base station and viewed and processed through the monitoring information management system, such as... Figure 3 As shown.

[0123] 4) Acquire surrounding rock monitoring data for each measuring point through the monitoring information management system, and preprocess the surrounding rock monitoring data, including removing outliers, filling in missing data using piecewise cubic Hermite interpolation polynomials, and denoising the data using methods such as filtering and smoothing to improve the quality of the monitoring data. Decompose the deformation monitoring data for each measuring point using the system's embedded moving average method. In this embodiment, a k value of 7 is selected to obtain the deformation trend term, and the deformation period term is obtained by subtracting the trend term from the deformation monitoring data. Calculate the correlation between the deformation trend term and period term and the stress and temperature monitoring data for each measuring point using the system's embedded maximum information coefficient (MIC) method. In this embodiment, stress and temperature monitoring data with a correlation of 0.8 or higher with the trend term are selected as input feature values ​​for the trend term prediction model, and stress and temperature monitoring data with a correlation of 0.4 or higher with the period term are selected as input feature values ​​for the period term prediction model. The data is then normalized and scaled to [0,1].

[0124] 5) Based on the NPDOA-LSTM prediction model embedded in the monitoring information management system, predictions were made for deformation trend and periodicity terms. The dataset was divided into two parts: the first 80% was the training set, and the last 20% was the prediction set. The LSTM model parameters were initialized, with 1000 training iterations and a gradient threshold of 1. The initial NPDOA population size was set to 10, and the number of iterations was set to 20. The NPDOA optimization algorithm was used to optimize the hyperparameters of the LSTM model, including the learning rate, the number of hidden layer neurons, and the L2 regularization coefficient. The optimal hyperparameters were then fed into the LSTM model for training, yielding prediction results for deformation trend and periodicity terms at each measurement point. The flowchart of the NPDOA-LSTM model is shown below. Figure 4 As shown; NPDOA algorithm flowchart, as follows Figure 5 As shown.

[0125] 6) The surrounding rock deformation prediction result is obtained by adding the surrounding rock deformation trend term and periodic term predicted by the NPDOA-LSTM model. The determination coefficient is then calculated. R 2 Mean absolute error MAE and root mean square error RMSE The prediction performance of the prediction model is evaluated. Taking the 1.5m measuring point at the top of the arch as an example, the deformation prediction results and prediction performance are as follows: Figure 6 , 7 As shown. Under the same conditions, compared with the prediction results of the traditional prediction model and the unoptimized LSTM model, the deformation results of the NPDOA-LSTM prediction model match the actual deformation well. R 2 The accuracy reached 0.995, an improvement of approximately 60% compared to traditional prediction models. Furthermore, its... MAE as well as RMSE The value of this method is the smallest among all models. The deformation prediction curves show that this method demonstrates accuracy and stability in long-term prediction of surrounding rock deformation, while the unoptimized LSTM model and traditional prediction models exhibit significant errors in predicting long-term surrounding rock deformation, with the maximum error approaching 5 mm. This method shows significant improvements in data preprocessing and model computation efficiency, with an overall prediction efficiency increase of 80% compared to traditional prediction models. In summary, this method has the highest accuracy and best prediction effect for predicting surrounding rock deformation in fault-resistant cavern excavation.

[0126] The specific embodiments of the present invention have been described in detail above, but these are merely one example, and the present invention is not limited to the specific embodiments described above. For those skilled in the art, any equivalent modifications and substitutions to the present invention are also within the scope of the present invention. Therefore, all equivalent changes and modifications made without departing from the spirit and scope of the present invention should be covered within the scope of the present invention.

Claims

1. A method for predicting the deformation of surrounding rock in fault geological caverns based on excavation sequence optimization and an NPDOA-LSTM model, characterized in that, The method includes the following steps: S1, Establish a fault cavern excavation model: Obtain relevant data on fault caverns and establish a numerical model of fault geological caverns based on the relevant data; S2 simulates different excavation sequence conditions: Based on the numerical model of the geological cavern of the S1 fault layer, different excavation sequence conditions were formulated, and numerical simulations were performed on the different sequence conditions to obtain the optimal excavation sequence condition. S3, deploy a real-time surrounding rock monitoring system and excavate the tunnel according to the optimal excavation sequence: During the excavation of the tunnel in accordance with the optimal excavation sequence in S2, the surrounding rock deformation, stress and temperature monitoring data are monitored in real time by deploying a surrounding rock monitoring system and connecting it to the monitoring information management system. S4, preprocess the monitoring data and calculate the correlation between the deformation decomposition terms and the monitoring data: The monitoring data obtained in S3 is preprocessed by the monitoring information management system and then decomposed into trend items and periodic items. The correlation between the decomposed items and the monitoring data is calculated, and the monitoring data with high correlation are selected to build a prediction model. S5, initialize the model and train the model; Using the monitoring data selected in S4, the hyperparameters in the LSTM model were optimized using the NPDOA optimization algorithm, and the LSTM model was trained. Then, the prediction results of the deformation trend term and the periodic term were obtained through the NPDOA-LSTM prediction model. S6, Reconstruct the deformation prediction results and evaluate the prediction effect.

2. The method for predicting the deformation of surrounding rock in fault geological caverns based on excavation sequence optimization and NPDOA-LSTM model as described in claim 1, characterized in that, S1 specifically includes: Through preliminary field investigation and physical model tests, parameters of fault, surrounding rock, geostress, and cavern size were obtained, and a numerical model of the fault geological cavern was established using 3DEC discrete element software.

3. The method for predicting the deformation of surrounding rock in fault geological caverns based on excavation sequence optimization and NPDOA-LSTM model according to claim 1, characterized in that, S2 specifically includes: The excavation section is divided into excavation zones to generate different excavation sequence conditions. In the numerical model of the fault geological cavern, monitoring points for the displacement of the top arch and sidewalls are set up, and numerical simulations are performed for different excavation sequence conditions. By comparing the displacement of monitoring points under each working condition, the excavation sequence with the smallest average displacement of the surrounding rock was selected as the optimal excavation sequence.

4. The method for predicting the deformation of surrounding rock in fault geological caverns based on excavation sequence optimization and NPDOA-LSTM model according to claim 1, characterized in that, S3 specifically includes: Integrated fiber optic grating sensors are installed at the top arch and left and right sidewalls of the excavated tunnel section and connected to the monitoring information management system to form a surrounding rock monitoring system; The fault cavern was excavated according to the optimal excavation sequence, and the deformation, stress, and temperature data of the surrounding rock were monitored in real time.

5. The method for predicting the deformation of surrounding rock in fault geological caverns based on excavation sequence optimization and NPDOA-LSTM model according to claim 1, characterized in that, S4 specifically includes: The monitoring information management system acquires the surrounding rock monitoring data of each monitoring point and preprocesses the surrounding rock monitoring data. The preprocessing methods include outlier removal, missing data filling, and data denoising. The deformation monitoring data was decomposed into trend and periodic components using the moving average method. The correlation between the deformation decomposition term and the monitoring data is calculated using the maximum information coefficient method, and the monitoring data with high correlation are selected to construct the prediction model.

6. The method for predicting the deformation of surrounding rock in fault geological caverns based on excavation sequence optimization and NPDOA-LSTM model according to claim 1, characterized in that, S5 specifically includes: The monitoring dataset selected in S4 is divided and the data is normalized. Initialize the LSTM model parameters, and set the initial population size and number of iterations for NPDOA; The NPDOA optimization algorithm is used to optimize the hyperparameters in the LSTM model. The optimal hyperparameters were fed into the LSTM model for training, and the prediction results of the deformation trend term and the periodic term were obtained respectively.

7. The method for predicting the deformation of surrounding rock in fault geological caverns based on excavation sequence optimization and NPDOA-LSTM model according to claim 1, characterized in that, S6 specifically includes: The surrounding rock deformation trend term and periodic term predicted by the NPDOA-LSTM model are added together to obtain the surrounding rock deformation prediction result. By calculating the coefficient of determination R 2 Mean absolute error MAE and root mean square error RMSE The predictive performance of the prediction model is evaluated.

8. The method for predicting the deformation of surrounding rock in fault geological caverns based on excavation sequence optimization and NPDOA-LSTM model according to claim 5, characterized in that, The normalization process in S4 is as follows: ； In the formula, The data is after normalization; Represents the original data; This represents the maximum value in the dataset; This represents the minimum value in the dataset; The decomposition method for the trend term of deformation monitoring data in S4 is as follows, assuming the original displacement time series is: ； Trend Item The calculation formula is: ； In the formula, The total number of data sets; for t The cumulative displacement at any given time; For the movement cycle.

9. The method for predicting the deformation of surrounding rock in fault geological caverns based on excavation sequence optimization and NPDOA-LSTM model according to claim 6, characterized in that, The NPDOA optimization algorithm specifically includes: 1) Initialize the neural population; By randomly generating several neural populations in the neural state space The neural state, among which It refers to the number of neural populations; 2) Attractor trend strategy; The neural states of a neural population tend to be multiple attractors in the neural state space. By using the optimal neural state as an attractor, the neural states of the remaining neural population randomly converge to one of these attractors. 3) Coupling interference strategy; The neural state of a specific neural population can be affected by the neural state of other neural populations. The influence of other neural populations can be simulated by using a combination of two couplings. 4) Information projection strategy; Information exchange and interaction between neural populations are carried out through a communication subspace. In the NPDOA algorithm, a matrix representing adjacency and communication strength is used to represent the communication subspace of information projection between neural populations. Ultimately, the neural state of each neural population is influenced by the attractor and the neural states of other neural populations.

10. The method for predicting the deformation of surrounding rock in fault geological caverns based on excavation sequence optimization and NPDOA-LSTM model according to claim 7, characterized in that, The determination coefficient in step 6 R 2 Mean absolute error MAE and root mean square error RMSE The formula is as follows: ； ； ； In the formula, For the sample size, This represents the actual value of the surrounding rock deformation. This represents the predicted value of surrounding rock deformation. It is the average value of the actual deformation of the surrounding rock.

Images