A Deep Learning-Based Method for Inverting Rock Mechanics Parameters and Predicting Subsequent Excavation Response in Underground Powerhouses
By applying deep learning algorithms and numerical simulation technology in underground engineering, an analysis model of underground powerhouse cavern groups is established, which can quickly obtain rock mechanics parameters and predict excavation response. This solves the timeliness and accuracy problems of inversion analysis in traditional methods and achieves more efficient support design guidance.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- POWERCHINA HUADONG ENG CORP LTD
- Filing Date
- 2023-02-14
- Publication Date
- 2026-05-26
AI Technical Summary
Traditional rock mechanics parameter inversion in underground engineering relies on manual comparison of monitoring data, which makes it difficult to achieve rapid and accurate inversion analysis. This results in long construction cycles, large workloads, and difficulty in effectively guiding subsequent excavation and support design.
By combining deep learning algorithms with numerical simulation technology, an analysis model of underground powerhouse cavern group is established. The deep learning network is trained by monitoring data to quickly obtain rock mechanics parameters and predict excavation response, thereby guiding the design of support schemes.
This improved the timeliness and accuracy of rock mass mechanics parameter inversion, reduced the need for human resources, increased the efficiency of inversion analysis, and ensured the safety and efficiency of underground powerhouse construction.
Smart Images
Figure CN116341052B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of underground engineering technology, particularly underground powerhouses, and provides a method for inverting rock mechanics parameters and predicting subsequent excavation response of underground powerhouses based on deep learning. Background Technology
[0002] For underground engineering, especially underground powerhouses, due to the uncertainty of geological conditions, complex construction organization, prominent engineering problems, long construction period, and large amount of monitoring data, traditional rock mechanics parameter inversion relies on manual comparison of monitoring data, which is difficult to compare globally, and the repeated debugging of rock mechanics parameters is a large workload and time-consuming.
[0003] Deep learning is a new research direction in the field of machine learning. Applying deep learning to underground engineering construction allows for the pre-establishment of an underground engineering analysis sample database using numerical simulation technology. During the construction process, based on the established learning sample database, rock mechanics parameters that reflect the response characteristics of the surrounding rock in the underground engineering can be quickly obtained to conduct rapid stability assessment of the underground engineering. Furthermore, it can quickly update the excavation increment prediction information based on the response and inversion parameters, which can save a lot of human resources, improve the efficiency and accuracy of inversion analysis, and better guide the subsequent excavation and support design. Summary of the Invention
[0004] The purpose of this invention is to provide a deep learning-based method for inverting rock mechanics parameters of underground powerhouses and predicting subsequent excavation responses. This method addresses the problems of poor timeliness, difficulty in accurate inversion, and large workload in conventional inversion analysis. Furthermore, it predicts subsequent excavation responses to guide the design of excavation support schemes and better leverages the role of inversion analysis.
[0005] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:
[0006] A deep learning-based method for inverting rock mass mechanical parameters and predicting subsequent excavation response of underground powerhouses, wherein the response includes the influence range and changes of surrounding rock deformation, stress, plastic zone, and support structure stress during layered excavation, and its features include the following steps:
[0007] S1: Excavation of the i-th layer of the underground powerhouse, 2≤i≤n, where n is the total number of layers to be excavated in the underground powerhouse. Based on the geological information of the (i-1)-th layer of the underground powerhouse, a numerical analysis model of the underground powerhouse cavern group is established.
[0008] S2: Based on geological information and the suggested range of mechanical parameters for each stratum and structural surface, parameter combinations are made. For each parameter combination, numerical simulation of the construction process is carried out according to the actual excavation and support process of the underground powerhouse cavern group, and the excavation response of the i-th and i+1-th layers of the underground powerhouse is calculated.
[0009] S3: Based on the monitoring design and instrument layout of the underground powerhouse, compile the numerical simulation results of the corresponding monitoring variables of the monitoring instrument location for the excavation of the i-th layer of the powerhouse for each combination of mechanical parameters of strata, faults and structural surfaces in S2;
[0010] S4: For the numerical simulation results of the excavation of the i-th layer of the underground powerhouse, a deep learning algorithm is used to train the network to obtain a trained deep learning network.
[0011] S5: Input the actual monitoring data of the construction site after the excavation of the i-th layer of the underground powerhouse is completed. Apply the deep learning network trained in step S4 to quickly predict the rock mechanics parameters and the excavation response of the next layer, and judge the rationality of the subsequent excavation and support scheme.
[0012] S6: Update the geological model and parameter range based on the newly revealed information from the excavation of the i-th layer of the underground powerhouse, and repeat steps S1-S5 for the (i+1)-th layer until the excavation of the underground powerhouse is completed.
[0013] The aforementioned method for inverting rock mechanics parameters and predicting subsequent excavation response of underground powerhouses based on deep learning is applicable to the 2nd to nth layers of an underground powerhouse excavated in n layers.
[0014] The aforementioned method for inverting rock mechanics parameters and predicting subsequent excavation responses in underground powerhouses based on deep learning is performed when the excavation of the i-th layer of the underground powerhouse is completed. After accessing monitoring data, it can quickly realize the inversion of rock mechanics parameters and the rapid prediction of the surrounding rock deformation and support structure stress responses in the next layer.
[0015] Furthermore, in S2, the rock mechanics parameters of each stratum and fault are considered according to the Hoek-Brown criterion, and the varying parameter is the geological strength factor GSI; the main structural planes that play a controlling role in the excavation response of the tunnel are considered to have three main parameters: comprehensive deformation modulus, shear resistance parameters (cohesion and internal friction angle), which are divided into m equal parts (5≤m≤10) according to the suggested parameter value range. The values include two cases: the upper and lower limits of the value range, and the middle (m-1) cases, that is, each stratum and structural plane considers (m+1) cases;
[0016] In S2, the combination of parameter values of each stratum and structural plane is sampled using the Pairwise algorithm, which effectively reduces the number of parameter combinations and improves analysis efficiency while ensuring sample coverage.
[0017] In S2, the excavation of the underground powerhouse cavern group is accurate to the layer level in the numerical simulation. In order to improve the simulation efficiency, the block and zone excavation process within each layer is no longer considered. The layer excavation of each cavern and the installation timing of support such as anchor bolts, anchor cables, and shotcrete are considered according to stress release and are consistent with the actual construction plan.
[0018] Furthermore, in S3, the monitoring instruments mainly include multi-point displacement gauges, anchor bolt force gauges, anchor cable force gauges, convergence monitoring, etc. In most cases, these instruments are immediately buried. Therefore, in the method of this invention, the monitoring instruments are all considered to be immediately buried. In S3, the numerical simulation results of the monitoring instrument position are obtained by setting the same measuring points in the numerical simulation model as the actual monitoring instrument position and the content being measured.
[0019] The results of the measuring points need to take into account the timing of the installation of the monitoring instruments. Since the numerical simulation model is only accurate to the layer level, the monitoring instruments buried in the current excavation layer can only be installed after the excavation location is exposed. The monitoring data for the current layer is incomplete, but it can obtain the response of the subsequent layers. For example, the monitoring data of the monitoring instruments installed during the excavation of the first layer of the plant can be compared with the incremental results of the numerical simulation of the second layer excavation from the start t1 to the end t2 of the second layer construction. Therefore, the monitoring results of the measuring points mainly use the incremental results of the excavation of the subsequent layers.
[0020] Furthermore, in S3, the numerical simulation results are organized into two types:
[0021] Q1. For the i-th layer (2≤i≤(n-1)), once the excavation of the i-th layer is completed, the increments of the instruments installed from the 1st to the (i-1th)th layer can be obtained. Therefore, it is necessary to organize the numerical simulation results of the incremental measurement points of the instrument positions installed from the 1st to the (i-1th)th layer. At the same time, in order to predict the response of the next layer, it is also necessary to organize the numerical simulation results of the incremental measurement points of the instrument positions installed from the 1st to the i-th layer after the excavation of the (i+1th)th layer is completed.
[0022] Q2. For the last floor of the underground plant, the nth floor, there is no further excavation. It is only necessary to organize the incremental results of the instrument position measurement points installed from the 1st to the (n-1th)th floor after the excavation of the nth floor is completed by numerical simulation.
[0023] Furthermore, in S4, the preparation of deep learning samples is also divided into two cases:
[0024] Q1: For the i-th layer (2≤i≤(n-1)), take the numerical simulation incremental results of the instrument locations installed in the 1st to (i-1)th layers as input, and the corresponding rock mechanics parameters and the numerical simulation incremental results of the instruments installed in the 1st to i-th layers of the next excavation as output, and train the deep learning network.
[0025] Q2: For the last floor of the underground powerhouse, the incremental results of numerical simulation of the instrument locations installed on floors 1 to (n-1) are used as inputs, and the corresponding rock mechanics parameters are used as outputs to train a deep learning network.
[0026] Furthermore, in S4, 80% of the parameter combinations are used as training samples, and the other 20% are used as test samples. When the mean square error of the deep learning network test samples is less than 0.001, the deep learning network is considered to have completed training and met the accuracy requirements.
[0027] In S4, the deep learning network models that have been trained from the 2nd to the nth layer and meet the accuracy requirements are saved respectively.
[0028] Furthermore, in S5, the actual incremental monitoring data of the instruments installed on the (i-1)th layer after the excavation of the i-th layer of the underground powerhouse is completed is used as input. The deep learning network model trained and meeting the accuracy requirements for the corresponding excavation layer in step S4 is applied to obtain the corresponding rock mechanics parameters and the monitoring instrument increment caused by the excavation of the next layer, such as the deformation increment of the multi-point displacement gauge and the anchor stress increment of the (i+1)th layer excavation. The rationality of the excavation support scheme can be further judged.
[0029] Furthermore, in S6, considering the changes in geological information, models, and parameter ranges revealed by the excavation of the i-th layer, newly revealed strata, faults, structural planes, etc. need to be taken into account in the numerical analysis model. The parameter ranges of each stratum, fault, and structural plane need to be re-evaluated, and steps S1 to S5 are repeated to ensure the accuracy of the analysis and prediction of the (i+1)-th layer.
[0030] This invention presents a deep learning-based method for inverting rock mechanics parameters and predicting subsequent excavation responses in underground powerhouses. It establishes a numerical analysis model based on the current geological information of the underground powerhouse, performs numerical simulations of various combinations based on the distribution of strata, faults, and structural plane parameters, and the excavation and support design of the underground powerhouse. The numerical simulation results are then compiled based on the monitoring design of the underground powerhouse. A deep learning model is trained, and monitoring data is integrated to quickly complete the inversion of rock mechanics parameters and the prediction of subsequent excavation responses. This improves the timeliness and accuracy of rock mechanics parameter inversion and provides more accurate prediction data for subsequent excavation responses, guiding the design of subsequent excavation and support for the underground powerhouse and ensuring safe construction. Attached Figure Description
[0031] Figure 1 This is a flowchart of a method for inverting rock mass mechanical parameters and predicting subsequent excavation response of an underground powerhouse based on deep learning, according to the present invention.
[0032] Figure 2 This is a flowchart illustrating the process of retrieving rock mass mechanical parameters and predicting the response of the third layer based on monitoring data after the completion of the second layer excavation in this embodiment of the invention.
[0033] Figure 3 This is a geological model diagram of the first level of the underground powerhouse as shown in this embodiment of the invention.
[0034] Figure 4 This is a numerical analysis model diagram established based on the geological model exposed during the excavation of the first layer of the underground powerhouse, as an embodiment of the present invention.
[0035] Figure 5 This is a diagram showing the layout of monitoring instruments in the underground powerhouse according to an embodiment of the present invention;
[0036] Figure 6 This is a schematic diagram of the deep learning network model for the second layer numerical analysis results in an embodiment of the present invention;
[0037] Figure 7 This illustrates how the error of the deep learning model changes with the number of training iterations in an embodiment of the present invention. Detailed Implementation
[0038] This embodiment utilizes the deep learning-based method for inverting rock mechanics parameters and predicting subsequent excavation response in underground powerhouses. By employing deep learning algorithms, it achieves rapid inversion of the rock mechanics parameters of the second layer of an underground powerhouse and predicts the monitoring instrument data for the third layer excavation, thereby determining the rationality of the third layer excavation support design scheme.
[0039] To better understand the above technical solutions, the following will describe the technical solutions in detail with reference to the accompanying drawings and specific embodiments. It should be understood that the specific features in the implementation of the present invention are detailed descriptions of the technical solutions of the present invention, rather than limitations thereof. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.
[0040] according to Figure 1 This invention provides a method for inverting rock mass mechanical parameters and predicting subsequent excavation response of underground powerhouses based on deep learning, comprising the following steps:
[0041] S1: Excavation of the i-th layer of the underground powerhouse (2≤i≤n, n is the total number of excavation layers of the underground powerhouse). Based on the geological information of the (i-1)-th layer of the underground powerhouse, establish a numerical analysis model of the underground powerhouse cavern group.
[0042] S2: Based on geological information, the recommended range of mechanical parameters for each stratum and structural surface, parameter combinations are made, and numerical simulation of the construction process is carried out for each parameter combination according to the actual excavation and support process of the underground powerhouse cavern group, and the excavation response of the i-th and (i+1)-th layers of the underground powerhouse is calculated.
[0043] S3: Based on the monitoring design and instrument layout of the underground powerhouse, compile the numerical simulation results of the corresponding monitoring variables of the monitoring instrument location for the excavation of the i-th layer of the powerhouse for each combination of mechanical parameters of strata, faults and structural surfaces in S2;
[0044] S4: For the numerical analysis results of the excavation of the i-th layer of the underground powerhouse, a deep learning algorithm is used to train the network to obtain a well-trained deep learning network.
[0045] S5: Input the actual monitoring data of the construction site after the excavation of the i-th layer of the underground powerhouse is completed. Apply the deep learning network trained in step S4 to quickly predict the rock mechanics parameters and the excavation response of the next layer, and judge the rationality of the subsequent excavation and support scheme.
[0046] S6: Update the geological model and parameter range based on the newly revealed information from the excavation of the i-th layer of the underground powerhouse, and repeat steps S1-S6 for the (i+1) layer until the excavation of the underground powerhouse is completed.
[0047] Reference Figure 2 In this embodiment, the process of inverting rock mass mechanical parameters and predicting the response of the third layer based on monitoring data after the completion of the second layer excavation specifically includes the following steps:
[0048] (1) Establish a numerical analysis model for the second layer based on the geological model information of the first layer;
[0049] (2) Based on the range of GSI values of each stratum fault in the first geological model, the values are divided into 10 equal parts, totaling 11 values. The Pairwise algorithm is used to combine the GSI and structural parameters of each stratum. In this example, there are 10 strata and structural surfaces, resulting in 209 combinations.
[0050] (3) Based on the excavation and support design scheme of the underground powerhouse, simulate the excavation and support process under the above 209 parameter combinations;
[0051] (4) Organize the numerical simulation results, mainly the results of the monitoring instruments corresponding to the completion of the second floor excavation and the results of the monitoring instruments corresponding to the completion of the third floor excavation. The results of the monitoring instruments corresponding to the completion of the second floor excavation are used to compare with the actual monitoring data, and the results of the monitoring instruments corresponding to the completion of the third floor excavation are used as the prediction results of the excavation response of the next floor.
[0052] (5) For 209 parameter combinations, the increment of the instrument installation position after the completion of the second layer excavation is used as the input, and the corresponding rock mechanics parameters and the increment of the instrument installation position after the completion of the third layer excavation are used as the output;
[0053] (6) For 209 parameter combinations, 80% are selected as training samples for deep learning training, and the remaining 20% are used as test samples. When the mean square error of the test samples is less than 0.001, the network training accuracy is considered to meet the prediction requirements, and the trained deep learning network model is saved.
[0054] (7) Input the incremental data of the instrument of the first floor after the completion of the second floor construction of the underground powerhouse at the engineering site. Apply the trained deep learning model to quickly feed back the rock mechanics parameters that can reflect the current status of the powerhouse and predict the excavation response of the next floor.
[0055] The geological model obtained after the excavation of the first floor of the underground powerhouse in this embodiment is as follows: Figure 3 As shown, the model includes 10 stratigraphic faults: Bs-strong III, Bs-weak III, Bs-weak IV, Bs-weak V, Pyr IV, and faults C1, f22, f31, f44, and f46. The geologically recommended GSI value range for Bs-strong III is 45-55, divided into 10 equal parts, resulting in 11 values. Other strata and faults are treated similarly. A pairwise algorithm is applied to generate 209 parameter combinations.
[0056] The numerical analysis model established based on the geological model revealed during the excavation of the first level of the underground powerhouse in this embodiment is as follows: Figure 4 As shown, the model is excavated in four layers.
[0057] according to Figure 5 In this embodiment, 8 multi-point displacement gauges, 6 anchor bolt force gauges, and 4 sets of convergence monitoring are set up. All of them are immediately buried monitoring instruments. Before the excavation of the second layer, the monitoring instruments for the first layer, multi-point displacement gauges Mph-R0+047-1, Mph-R0+047-2, and Mph-R0+047-3, anchor bolt force gauge Rph-R0+047-1, and convergence monitoring points ABC, have been installed.
[0058] After the second layer of construction is completed, multi-point displacement gauges Mph-R0+047-4 and Mph-R0+047-5, anchor bolt force gauges Rph-R0+047-2 and Rph-R0+047-3, and convergence monitoring point DE will be installed.
[0059] according to Figure 6 The deep learning network model based on the numerical analysis results of the second layer uses the incremental results of the multi-point displacement gauges Mph-R0+047-1~3 and the anchor force gauge Rph-R0+047-1, as well as the convergence deformations AB, AC, and BC, from the monitoring instruments used for the second layer excavation of the plant as inputs. The GSI values of 10 stratigraphic faults and the incremental results of the convergence deformations AB, AC, BC, AD, AE, and DE from the multi-point displacement gauges Mph-R0+047-1~5 and the anchor force gauge Rph-R0+047-1~3 from the monitoring instruments used for the third layer excavation as outputs for network training.
[0060] Figure 7 The example demonstrates the mean squared error (MSE) and its variation with the number of training iterations of the deep learning model. After 300 training iterations, the mean squared error (MSE) is less than 0.001.
[0061] Finally, it should be noted that the above specific embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to examples, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.
Claims
1. A deep learning-based method for inverting rock mechanics parameters and predicting subsequent excavation response of underground powerhouses, wherein the response includes the changes in surrounding rock deformation and stress on the support structure as a result of layered excavation. Its features include the following steps: S1: Excavation of the i-th layer of the underground powerhouse, 2≤i≤n, where n is the total number of layers to be excavated in the underground powerhouse. Based on the geological information of the (i-1)-th layer of the underground powerhouse, a numerical analysis model of the underground powerhouse cavern group is established. S2: Based on geological information, the recommended range of mechanical parameters for each stratum and structural surface, parameter combinations are made, and numerical simulation of the construction process is carried out for each parameter combination according to the actual excavation and support process of the underground powerhouse cavern group. The excavation response of the i-th and (i+1)-th layers of the underground powerhouse is calculated. S3: Based on the monitoring design and instrument layout of the underground powerhouse, compile the numerical simulation results of the corresponding monitoring variables of the monitoring instrument location for the excavation of the i-th layer of the powerhouse for each combination of mechanical parameters of strata, faults and structural surfaces in S2; S4: For the numerical simulation results of the excavation of the i-th layer of the underground powerhouse, a deep learning algorithm is used to train the network to obtain a trained deep learning network. S5: Input the actual monitoring data of the construction site after the excavation of the i-th layer of the underground powerhouse is completed. Apply the deep learning network trained in step S4 to quickly predict the rock mechanics parameters and the excavation response of the next layer, and judge the rationality of the subsequent excavation and support scheme. S6: Update the geological model and parameter range based on the newly revealed information from the excavation of the i-th layer of the underground powerhouse, and repeat steps S1-S5 for the (i+1)-th layer until the excavation of the underground powerhouse is completed; In S2, the rock mechanics parameters of each stratum and fault are considered according to the Hoek-Brown criterion, and the variable parameter is the geological strength factor (GSI). The main structural planes that play a controlling role in the cavern excavation response consider three main parameters: comprehensive deformation modulus, shear resistance parameter, etc. The parameter values are divided into m equal parts according to the suggested range, 5≤m≤10. The values include two cases: the upper and lower limits of the range, and the middle m-1 cases. That is, each stratum and structural plane considers m+1 cases. The combination of GSI values of each stratum and structural plane parameters is sampled using the Pairwise algorithm to effectively reduce the number of parameter combinations and improve analysis efficiency while ensuring sample coverage. In the numerical simulation, the excavation of the underground powerhouse cavern group is accurate to the layer level. To improve the simulation efficiency, the block and zone excavation process within each layer is no longer considered. The excavation layering of each cavern and the timing of support installation are consistent with the actual construction plan.
2. The method for inverting rock mass mechanical parameters and predicting subsequent excavation response of underground powerhouses based on deep learning as described in claim 1, characterized in that... The process is conducted when the excavation of the i-th layer of the underground powerhouse is completed. After the monitoring data is connected, the rock mechanics parameters can be quickly inverted and the deformation of the surrounding rock and the stress response of the support structure in the next layer can be quickly predicted.
3. The method for inverting rock mass mechanical parameters and predicting subsequent excavation response of underground powerhouses based on deep learning as described in claim 1, characterized in that... In S3, the monitoring instruments include multi-point displacement gauges, anchor bolt force gauges, anchor cable force gauges, and convergence monitoring. All monitoring instruments are considered to be immediately buried. The numerical simulation results of the monitoring instrument locations are obtained by setting measuring points in the numerical simulation model that are identical to the actual monitoring instrument locations and the content measured. The results of the measuring points need to take into account the timing of the installation of the monitoring instruments. Since the numerical simulation model is only accurate to the layer level, for monitoring instruments immediately buried in the current excavation layer, they can only be installed after the excavation location is exposed. Their monitoring data for the current layer is incomplete, but they can obtain the response of subsequent layers. The monitoring results of the measuring points mainly use the incremental results of subsequent layers.
4. The method for inverting rock mass mechanical parameters and predicting subsequent excavation response of underground powerhouses based on deep learning as described in claim 1, characterized in that... In S3, the numerical simulation results are organized in two ways: Q1. For the i-th layer, 2≤i≤(n-1), when the excavation of the i-th layer is completed, the increments of the instruments installed from the 1st to the (i-1th)th layer can be obtained. Therefore, it is necessary to organize the numerical simulation results of the incremental results of the instrument position measurement points installed from the 1st to the (i-1th)th layer. At the same time, in order to predict the response of the next layer, it is also necessary to organize the numerical simulation results of the incremental results of the instrument position measurement points installed from the 1st to the i-th layer after the excavation of the (i+1th)th layer is completed. Q2. For the last floor of the underground plant, the nth floor, there is no further excavation. It is only necessary to compile the incremental results of the instrument position measurement points installed on the 1st to the (n-1th)th floors after the excavation of the nth floor is completed.
5. The method for inverting rock mass mechanical parameters and predicting subsequent excavation response of underground powerhouses based on deep learning as described in claim 1, characterized in that... In S4, the organization of deep learning samples also falls into two categories: Q1: For the i-th layer, 2≤i≤(n-1), take the numerical simulation incremental results of the instrument locations installed from the 1st to the (i-1th)th layer as input, and the corresponding rock mechanics parameters and the numerical simulation incremental results of the instruments installed from the 1st to the i-th layer in the next excavation as output, and train the deep learning network. Q2: For the last floor and the nth floor of the underground powerhouse, use the numerical simulation incremental results of the instrument locations installed on floors 1 to n-1 as input and the corresponding rock mechanics parameters as output to train a deep learning network.
6. The method for inverting rock mass mechanical parameters and predicting subsequent excavation response of underground powerhouses based on deep learning as described in claim 1, characterized in that... In S4, 80% of the parameter combinations are used as training samples and 20% are used as test samples. When the mean square error of the deep learning model test samples is less than 0.001, the deep learning network is considered to have been trained and the accuracy meets the requirements. The deep learning network models that have been trained from the 2nd to the nth layers and meet the accuracy requirements are saved respectively.
7. The method for inverting rock mass mechanical parameters and predicting subsequent excavation response of underground powerhouses based on deep learning as described in claim 1, characterized in that... In S5, the actual incremental monitoring data of the instrument installed on the (i-1)th layer after the excavation of the i-th layer of the underground powerhouse is completed is used as input. The deep learning network model trained in step S4 for the corresponding excavation layer and meeting the accuracy requirements is applied to obtain the corresponding rock mechanics parameters and the monitoring instrument increment caused by the excavation of the next layer. Furthermore, the rationality of the excavation support scheme can be judged.
8. The method for inverting rock mass mechanical parameters and predicting subsequent excavation response of underground powerhouses based on deep learning as described in claim 1, characterized in that... In S6, the changes in geological information, model, and parameter ranges revealed by the excavation of the i-th layer are considered. For newly revealed strata, faults, and structural planes, they need to be taken into account in the numerical analysis model. The parameter ranges of each stratum, fault, and structural plane are re-evaluated, and steps S1 to S5 are repeated to ensure the accuracy of the analysis and prediction of the i+1 layer.