Rock mass mechanical parameter prediction method based on pumped storage power station underground powerhouse

Abstract

A rock mass mechanical parameter prediction method based on an underground powerhouse of a pumped storage power station relates to the technical field of mechanical parameter prediction, and comprises the following steps: identifying an embedding position and distribution characteristics of a sulfur-containing shale interlayer by constructing a three-dimensional geologic structure model and lithologic information of a construction area, and establishing a structural domain partition parameter prediction area system; constructing a nonlinear regression function model taking infrared spectrum characteristic factors and environment control parameters as input and taking elasticity modulus and the like as output by combining infrared spectrum characteristic data and a historical evolution track; the model is embedded into a regional system for dynamic assignment, and a parameter space-time change atlas is constructed; the laser point cloud data and the micro-seismic monitoring data are fused to invert the actual response of the surrounding rock; when the monitoring value deviates from the prediction map, triggering parameter degradation function re-calibration to complete parameter dynamic updating and prediction closed loop; according to the method, the accuracy of rock mass mechanical parameter prediction of the pumped storage power station underground powerhouse can be improved.

Description

Technical Field

[0001] This application relates to the technical field of mechanical parameter prediction, specifically to a method for predicting rock mass mechanical parameters based on the underground powerhouse of a pumped storage power station. Background Technology

[0002] The rock mechanics parameters of the underground powerhouse of a pumped storage power station refer to a quantitative index system describing the mechanical response behavior of the surrounding rock under the coupled effects of multiple physical fields such as static load, dynamic load, seepage, and temperature. These parameters mainly include elastic modulus, shear modulus, and cohesion, used to characterize the rigidity, strength, deformability, permeability, and damage evolution capacity of the rock mass. In actual engineering, these rock mechanics parameters are not only affected by lithology, joint and fracture distribution, and structural integrity, but also controlled by the state of the geostress field, the groundwater chemical environment, and the history of excavation disturbance, exhibiting significant spatial heterogeneity, anisotropy, and temporal evolution. Accurately obtaining and dynamically updating these rock mechanics parameters is a prerequisite for ensuring the structural stability of the underground powerhouse, developing a reasonable support system, conducting numerical analysis and simulation, and implementing disaster early warning and control.

[0003] In high-sulfur shale sections, especially when interlayered within metamorphic sandstone structures, the rock mass is susceptible to stress corrosion, sulfidation expansion, and redox reactions during construction drainage. This leads to gradual microstructural disintegration, a decrease in the stress threshold for fracture closure, and rapid degradation of joint shear strength. This results in irreversible evolution characterized by weakened elastic modulus, decreased cohesion, and a sudden drop in shear strength. The evolution path is typically accompanied by functional group reconstruction in infrared spectroscopy, mineral oxidation state migration, and hysteresis in mechanical response. This degradation process is no longer controlled by traditional tectonic stress and joint geometry, but rather by dynamic chemical environment-driven processes that nonlinearly intensify with drainage time. However, existing rock mass mechanical parameter prediction techniques generally rely on static experimental data and structural surface geometric models, lacking a systematic consideration of chemical stress field coupling, spectroscopic response characteristics, and parameter hysteresis weakening behavior in reactive interlayer sections. Consequently, they cannot accurately reflect the true mechanical evolution mechanism of sulfur-bearing shale interlayers during underground powerhouse excavation disturbance. Summary of the Invention

[0004] This application provides a method for predicting rock mechanics parameters of underground powerhouses in pumped storage power stations. This method can reflect the actual mechanical evolution mechanism of sulfur-bearing shale interlayers during the excavation and disturbance process of underground powerhouses, thereby improving the accuracy of rock mechanics parameter prediction for underground powerhouses in pumped storage power stations.

[0005] The first aspect of this application provides a method for predicting rock mechanics parameters based on the underground powerhouse of a pumped storage power station, the method comprising:

[0006] Obtain a three-dimensional geological structure model and lithological information of the construction area, identify the embedding location and spatial distribution characteristics of sulfur-bearing shale interlayers in the metamorphic sandstone structure, and construct a parameter prediction regional system based on tectonic domain partitioning.

[0007] The historical evolution trajectory of elastic modulus, cohesion and shear strength, as well as infrared spectral characteristic data of the sulfur-bearing shale interlayer region were obtained;

[0008] Based on the historical evolution trajectory and the infrared spectral feature data, a nonlinear regression function model is constructed. The nonlinear regression function model takes infrared spectral feature factors and environmental control parameters as input variables and the predicted evolution trajectory of elastic modulus, cohesion and shear strength as target output quantities.

[0009] The nonlinear regression function model is embedded into the parameter prediction region system, and dynamic parameter assignment is performed on the sulfur-bearing shale interlayer region to construct a spatiotemporal variation map of parameters that evolves with the construction drainage process.

[0010] Acquire laser point cloud data and microseismic monitoring data, invert the actual deformation trend and microfracture density change of the surrounding rock during construction, and output the actual monitoring results;

[0011] When the actual monitoring result deviates from the predicted envelope of the spatiotemporal variation map of the parameters, the recalibration process of the parameter degradation function is triggered until the error between the actual monitoring result and the predicted result of the spatiotemporal variation map of the parameters meets the convergence condition, and the parameter update stops.

[0012] Based on the above technical solutions, preferably, the acquisition of a three-dimensional geological structure model and lithological information of the construction area, identification of the embedding location and spatial distribution characteristics of sulfur-bearing shale interlayers in the metamorphic sandstone structure, and construction of a parameter prediction regional system based on tectonic domain zoning specifically include:

[0013] By fusing borehole logging data, core scanning data, ground-penetrating radar profile data, and 3D seismic reflection data, a spatial dataset under a unified coordinate system is established.

[0014] A spatial modeling method was used to perform three-dimensional fitting on the structural boundaries, lithological interfaces and discontinuities in the spatial dataset to construct a three-dimensional geological structure model of the construction area.

[0015] A grid of the surrounding rock surface of the factory is generated based on laser scanning point cloud data, and the three-dimensional geological structure model is divided into multiple block units using voxel mesh generation technology, with each block unit assigned a lithological attribute label.

[0016] By extracting lithological segments with sulfide enrichment responses from borehole logging data and core elemental analysis results, identifying the combined characteristics of resistivity, gamma ray spectrum and density, and constructing a spatial probability distribution map of sulfur-bearing shale interlayers;

[0017] Combining geological profiles and stratigraphic information, the embedding angle, extension direction, thickness parameters, and contact surface attributes of the sulfur-bearing shale interlayer in the metamorphic sandstone structure are determined based on the lithological attribute tags, and an embedding feature parameter set is constructed.

[0018] Based on the distribution of fracture structures, lithological variation interfaces, stress concentration zones and groundwater zoning information, tectonic domain division rules are established, and a parameter weighting mechanism is used to complete the tectonic domain division of the three-dimensional geological structure model.

[0019] In the results of the structural domain partitioning, the structural perturbation factor and chemical reaction sensitivity factor of each of the embedded feature parameter sets are defined, and the parameter dynamic evolution boundary conditions of the embedded feature parameter sets are set to construct a parameter prediction region system for dynamic prediction of rock mass mechanics parameters.

[0020] Based on the above technical solutions, preferably, before constructing the nonlinear regression function model based on the historical evolution trajectory and the infrared spectral feature data, the method further includes:

[0021] Extract the infrared spectral features corresponding to the elastic modulus, cohesion and shear strength from the infrared spectral feature data, including the position of the main absorption peak, the area of ​​the absorption peak, the ratio of the spectral band intensity and the change of the characteristic peak of the functional group, and construct the infrared spectral feature factor;

[0022] The infrared spectral characteristic factors are combined and encoded with environmental control parameters, including the pH value of the wetting solution, dissolved oxygen concentration, mass ratio of sulfate to chloride ions, and loading rate, to construct an input variable set;

[0023] Principal component analysis was performed on the infrared spectral characteristic factors to remove redundant dimensions and retain the principal component factors that have a preset strength of correlation with elastic modulus, cohesion and shear strength.

[0024] Using the principal component factors and the set of input variables as input variables, and the predicted values ​​of the elastic modulus, the cohesion, and the shear strength at different time steps as target outputs, the nonlinear regression function model is constructed.

[0025] Based on the above technical solutions, preferably, the step of constructing a nonlinear regression function model based on the historical evolution trajectory and the infrared spectral feature data specifically includes:

[0026] The principal component factors and the input variable set are subjected to zero-mean normalization to form an input variable matrix under a uniform scale.

[0027] The input variable matrix is ​​used as the input vector of the nonlinear regression function model, and the predicted values ​​of the elastic modulus, the cohesion and the shear strength at each time step are used as the target output quantities to construct a regression modeling structure with multiple output mapping relationships.

[0028] Long short-term memory neural networks are selected as the modeling structure for the nonlinear regression function model. The input variable matrix is ​​received in the input layer, the variable evolution features are extracted in the hidden layer through gated recursive units, and the causal relationship between the changes in historical spectrum and the evolution of rock mass mechanical parameters is captured through the time state transition matrix.

[0029] Predictive output nodes for the elastic modulus, cohesion, and shear strength are set in the output layer respectively, and a parallel structure is used to achieve multi-objective output;

[0030] When constructing the loss function, a multi-objective error function is introduced to constrain the prediction accuracy of multiple target outputs with weighted mean square error, and a time smoothing regularization term is introduced to suppress non-physical jumps.

[0031] The nonlinear regression function model is trained using the Adam optimizer. The training process is completed under the condition of meeting the convergence criterion, and the model accuracy and generalization ability are evaluated by cross-validation.

[0032] Based on the above technical solutions, preferably, the step of embedding the nonlinear regression function model into the parameter prediction region system to dynamically assign parameters to the sulfur-bearing shale interlayer region and constructing a spatiotemporal variation map of parameters evolving with the construction drainage process specifically includes:

[0033] In the parameter prediction region system, each spatial unit is associated with its spatial coordinates, lithological label, tectonic domain number and interlayer identification status, and a spatial unit index matrix belonging to the sulfur-bearing shale interlayer region is constructed.

[0034] Extract the construction time step, drainage status, initial rock mass properties and geochemical conditions of each spatial unit in the spatial unit index matrix to form a dataset for inputting the nonlinear regression function model;

[0035] The environmental control parameters and infrared spectral feature factors of each spatial unit in the dataset at a specific time step are used as inputs and imported into the nonlinear regression function model to output the predicted values ​​of elastic modulus, cohesion and shear strength, thereby realizing the dynamic parameter assignment of the spatial unit.

[0036] During the assignment of the dynamic parameters, a time filtering algorithm is used to constrain the rate of change of the prediction results of adjacent time steps, and a smoothing process is performed based on the local weighted regression method to ensure the physical continuity of the dynamic parameters in the time series.

[0037] The predicted elastic modulus, predicted cohesion and predicted shear strength corresponding to each time step are summarized in the form of three-dimensional tensors to construct a spatiotemporal variation map of parameters covering the sulfur-bearing shale interlayer region.

[0038] The spatiotemporal variation map of the parameters is embedded in the parameter prediction region system to support rock mass response simulation, support structure design evolution calculation and automatic identification and analysis of abnormal trends, so as to realize the dynamic prediction of the rock mass mechanical parameters throughout the whole process and model feedback correction.

[0039] Based on the above technical solutions, preferably, the acquisition of laser point cloud data and microseismic monitoring data, the inversion of the actual deformation trend and microfracture density change of the surrounding rock during construction, and the output of actual monitoring results specifically include:

[0040] By deploying a high-frequency laser scanner and a three-dimensional laser ranging system in the construction area, point cloud data of the construction area at different time periods are collected, and the point cloud data is subjected to attitude registration, noise filtering and mesh reconstruction to form a time series point cloud model of the deformation of the surrounding rock surface.

[0041] Based on the point cloud model, the displacement vector, normal convergence value and profile deformation rate of each monitoring block in the construction area at different time steps are calculated to construct a surrounding rock deformation trend map.

[0042] By deploying a microseismic monitoring array system around the construction area, microseismic event waveform data is collected, including event trigger time, amplitude, spectral characteristics, and initial motion direction;

[0043] Based on waveform analysis and travel time inversion methods, the waveform data of the microseismic event are spatially located to construct a three-dimensional spatiotemporal evolution map of the microseismic event;

[0044] Based on the aforementioned three-dimensional spatiotemporal evolution map, a micro-fracture density distribution map is generated using a fracture cluster identification algorithm;

[0045] The deformation trend map of the surrounding rock and the microfracture density distribution map are spatiotemporally registered and spatially superimposed to identify structural disturbance areas with coupling characteristics in the surrounding rock mass, and output actual monitoring results including deformation rate field, microfracture density field and structural disturbance level.

[0046] The actual monitoring results are embedded in the constructed domain coordinate system for deviation comparison with the spatiotemporal variation map of the parameters and for recalibration judgment of the parameter degradation function.

[0047] Based on the above technical solution, preferably, when the actual monitoring result deviates from the predicted envelope of the spatiotemporal variation map of the parameters, the recalibration process of the parameter degradation function is triggered, until the error between the actual monitoring result and the predicted result of the spatiotemporal variation map of the parameters meets the convergence condition, and the parameter update is stopped. Specifically, this includes:

[0048] The deformation rate field and micro-fracture density field in the actual monitoring results are spatially aligned with the spatiotemporal variation map of the parameters in the structural domain coordinate system, and the predicted value and monitored value of each spatial unit at the same time step are extracted. The predicted deviations of the elastic modulus, the cohesion and the shear strength are calculated, and the deviation magnitude matrix is ​​constructed.

[0049] Determine whether there is a target spatial unit in the deviation magnitude matrix that exceeds the preset error tolerance threshold. If the target spatial unit exists, define the monitoring block where the target spatial unit is located as the target recalibration area.

[0050] Extract the infrared spectral feature factors and environmental control parameters corresponding to the target recalibration region at the current time step, and construct a local training dataset by combining the actual monitoring values.

[0051] The transfer learning mechanism is adopted. Based on the original nonlinear regression function model, the local training dataset is retrained or the parameters are fine-tuned while keeping the irrelevant parameters frozen. The weight path related to the prediction bias is adaptively adjusted through the error backpropagation algorithm.

[0052] The updated nonlinear regression function model is used to re-predict the parameters of the spatial cells in the target recalibration region, and the predicted values ​​at the corresponding positions in the spatiotemporal variation map of the parameters are replaced to obtain the corrected spatiotemporal variation map of the parameters.

[0053] A transition region is constructed between the target recalibrated region and the uncorrected region. A weighted fusion strategy is used to continuously correct the output of the nonlinear regression function model and the updated nonlinear regression function model, maintaining the spatial smoothness of the parameter field.

[0054] The corrected spatiotemporal variation map of the parameters is re-embedded into the parameter prediction region system to realize the dynamic evolution and closed-loop update of rock mechanics parameters under construction disturbance conditions.

[0055] A second aspect of this application provides a rock mechanics parameter prediction device based on an underground powerhouse of a pumped storage power station. The device is used to execute a rock mechanics parameter prediction method based on an underground powerhouse of a pumped storage power station as described above. The device includes an acquisition module, a processing module, and an output module, wherein:

[0056] The acquisition module is used to acquire a three-dimensional geological structure model and lithological information of the construction area, identify the embedding location and spatial distribution characteristics of sulfur-bearing shale interlayers in the metamorphic sandstone structure, and construct a parameter prediction regional system based on tectonic domain partitioning.

[0057] The acquisition module is used to acquire the historical evolution trajectory of the elastic modulus, cohesion and shear strength of the sulfur-bearing shale interlayer region, as well as infrared spectral characteristic data.

[0058] The processing module is used to construct a nonlinear regression function model based on the historical evolution trajectory and the infrared spectral feature data. The nonlinear regression function model takes infrared spectral feature factors and environmental control parameters as input variables and the predicted evolution trajectory of elastic modulus, cohesion and shear strength as target output quantities.

[0059] The processing module is used to embed the nonlinear regression function model into the parameter prediction region system, assign dynamic parameters to the sulfur-bearing shale interlayer region, and construct a spatiotemporal variation map of parameters that evolves with the construction drainage process.

[0060] The acquisition module is used to acquire laser point cloud data and microseismic monitoring data, invert the actual deformation trend and microfracture density change of the surrounding rock during construction, and output the actual monitoring results.

[0061] The output module is used to trigger the recalibration process of the parameter degradation function when the actual monitoring result deviates from the predicted envelope of the parameter spatiotemporal variation map, until the error between the actual monitoring result and the predicted result of the parameter spatiotemporal variation map meets the convergence condition, and then stop parameter updating.

[0062] Based on the above technical solutions, the steps for implementing the specific functions of the acquisition module, processing module, and output module are described in the embodiments, and will not be repeated here.

[0063] A third aspect of this application provides an electronic device including a processor, a memory, a user interface, and a network interface, wherein the memory is used to store instructions, the user interface and the network interface are both used to communicate with other devices, and the processor is used to execute the instructions stored in the memory to cause the electronic device to perform the method as described in any of the foregoing.

[0064] A fourth aspect of this application provides a computer-readable storage medium storing instructions that, when executed, perform the method described in any of the preceding descriptions.

[0065] In summary, one or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages:

[0066] 1. A method for predicting rock mechanics parameters based on the underground powerhouse of a pumped storage power station is proposed. By integrating three-dimensional geological modeling and tectonic domain zoning technology, the embedding characteristics of sulfur-bearing shale interlayers in metamorphic sandstone are accurately identified. A nonlinear regression function model driven by infrared spectral characteristic factors and environmental control parameters is introduced to establish a time evolution prediction mechanism for elastic modulus, cohesion, and shear strength. This model is embedded into the parameter prediction regional system to achieve dynamic assignment of mechanical parameters with drainage disturbance. Furthermore, by combining laser point cloud and microseismic monitoring data, a spatiotemporal comparison system of the actual response of the surrounding rock is constructed. When the monitoring results deviate from the predicted envelope of the model, the local recalibration of the parameter degradation function is triggered, effectively feeding back and correcting the prediction results. This enables dynamic modeling and updating of the actual mechanical evolution path of sulfur-bearing shale interlayers under the combined driving forces of chemical coupling, excavation disturbance, and time-dependent changes, improving the environmental adaptability, response accuracy, and evolution consistency of rock mechanics parameter prediction.

[0067] 2. In a method for predicting rock mechanics parameters based on the underground powerhouse of a pumped storage power station, a three-dimensional geological structure model with high spatial resolution and attribute accuracy is established by fusing multi-source geological data and voxel meshing. Based on tectonic domain zoning, the embedding state and tectonic disturbance sensitivity of sulfur-bearing shale interlayers are effectively identified, significantly improving the response capability of the parameter prediction regional system to tectonic complexity and lithological heterogeneity, and providing spatial accuracy assurance for subsequent mechanical parameter prediction.

[0068] 3. In a method for predicting rock mechanics parameters based on the underground powerhouse of a pumped storage power station, spectral feature factors closely related to mechanical properties are extracted from the infrared spectrum and combined with environmental control parameters to construct an input variable set. This ensures that the constructed nonlinear regression function model can capture the combined driving effect of changes in the rock mass chemical environment on mechanical parameters, providing a more physically based input source for modeling the evolution of mechanical parameters.

[0069] 4. In a method for predicting rock mechanics parameters based on the underground powerhouse of a pumped storage power station, a long short-term memory neural network is used to model the input variable matrix. While preserving the changes in time state and causal relationships, the method achieves multi-objective dynamic prediction of elastic modulus, cohesion and shear strength. Regularization terms and weighted error functions are used to constrain the physical rationality and accuracy of the prediction. A nonlinear prediction model with adaptive learning capability is constructed, which significantly improves the model's ability to fit complex evolution processes.

[0070] 5. In a method for predicting rock mechanics parameters based on the underground powerhouse of a pumped storage power station, a nonlinear regression function model is embedded into the parameter prediction region system after the tectonic domain is partitioned, and each sulfur-bearing shale interlayer spatial unit is dynamically assigned time-by-time. The spatiotemporal variation map of parameters is constructed by combining time filtering and regression smoothing methods, so as to realize the continuous prediction of rock mechanics parameters under multi-physics field disturbance conditions and provide a stable input boundary for support structure design and response simulation.

[0071] 6. In a method for predicting rock mechanics parameters based on the underground powerhouse of a pumped storage power station, the real response data of the surrounding rock is obtained by high-frequency laser point cloud and microseismic monitoring, and deformation trend map and microfracture density map are constructed. A spatial registration and disturbance level identification mechanism is adopted to output the quantitative monitoring results of structural disturbance, realize the high spatiotemporal accuracy of field feedback data, and enhance the ability of the prediction model to perceive the actual rock mass response.

[0072] 7. In a method for predicting rock mechanics parameters based on the underground powerhouse of a pumped storage power station, the target area that needs to be updated is automatically identified by comparing the deviation between the monitoring results and the predicted map. The nonlinear regression function model is retrained using transfer learning and local fine-tuning mechanisms to achieve targeted model enhancement and parameter correction. Combined with spatial smoothing processing to maintain the continuity of the parameter field, an adaptive closed-loop update mechanism between model prediction and measured response is formed, which significantly improves the dynamic adaptability and long-term accuracy of rock mass parameter evolution prediction. Attached Figure Description

[0073] Figure 1 This is a flowchart illustrating a method for predicting rock mechanics parameters based on an underground powerhouse of a pumped storage power station, as disclosed in an embodiment of this application.

[0074] Figure 2 This is a schematic diagram of a rock mechanics parameter prediction device based on an underground powerhouse of a pumped storage power station, as disclosed in an embodiment of this application.

[0075] Figure 3 This is a schematic diagram of the structure of an electronic device disclosed in an embodiment of this application.

[0076] Explanation of reference numerals in the attached drawings: 201, acquisition module; 202, processing module; 203, output module; 301, processor; 302, communication bus; 303, user interface; 304, network interface; 305, memory. Detailed Implementation

[0077] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0078] Example 1:

[0079] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0080] The rock mechanics parameters of the underground powerhouse of a pumped storage power station are core quantitative indicators describing the mechanical response behavior of the surrounding rock under the coupling of multiple physical fields. They exhibit significant spatial heterogeneity, anisotropy, and temporal evolution. Accurately acquiring and dynamically updating this parameter system is fundamental to ensuring the structural safety of the powerhouse. However, in the high-sulfur shale interlayer embedded metamorphic sandstone structure system, the rock mass is subjected to the coupled driving force of stress corrosion, sulfidation expansion, and redox reactions during construction and drainage. This results in an irreversible deterioration process characterized by weakening of elastic modulus, decrease in cohesion, and a sudden drop in shear strength. This evolution path is accompanied by infrared spectral characteristics and mineral oxidation state migration, which has broken through the traditional static structural model and geometric joint analysis paradigm. Existing prediction methods are unable to accurately capture its dynamic evolution mechanism, and there is an urgent need to introduce parameter dynamic modeling technology based on chemical environment driving and spectroscopic response mapping.

[0081] This embodiment discloses a method for predicting rock mechanics parameters based on the underground powerhouse of a pumped storage power station, referring to... Figure 1 This includes the following steps S110-S160:

[0082] S110. Obtain a three-dimensional geological structure model and lithological information of the construction area, identify the embedding location and spatial distribution characteristics of sulfur-bearing shale interlayers in the metamorphic sandstone structure, and construct a parameter prediction regional system based on tectonic domain partitioning.

[0083] The rock mechanics parameter prediction method for underground powerhouses of pumped storage power stations disclosed in this application is applied to a server. The server includes, but is not limited to, electronic devices such as mobile phones, tablets, wearable devices, and PCs (Personal Computers), and can also be a backend server running the rock mechanics parameter prediction method for underground powerhouses of pumped storage power stations. The server can be implemented using a standalone server or a server cluster composed of multiple servers.

[0084] In one possible implementation, a three-dimensional geological structure model and lithological information of the construction area are acquired, the embedding location and spatial distribution characteristics of sulfur-bearing shale interlayers in the metamorphic sandstone structure are identified, and a parameter prediction regional system based on tectonic domain partitioning is constructed. Specifically, this includes: fusing borehole logging data, core scanning data, ground-penetrating radar profile data, and three-dimensional seismic reflection data to establish a spatial dataset in a unified coordinate system; using spatial modeling methods to perform three-dimensional fitting of tectonic boundaries, lithological interfaces, and discontinuities in the spatial dataset to construct a three-dimensional geological structure model of the construction area; generating a surface mesh of the surrounding rock of the plant based on laser scanning point cloud data, and dividing the three-dimensional geological structure model into multiple volumetric units using voxel mesh generation technology, assigning lithological attribute labels to each volumetric unit; and extracting borehole logging data and core element analysis. The results revealed lithological segments with sulfide enrichment responses. Resistivity, gamma ray spectroscopy, and density combination characteristics were identified, and a spatial probability distribution map of sulfur-bearing shale interlayers was constructed. Combining geological profiles and stratigraphic information, the embedding angle, extension direction, thickness parameters, and contact surface attributes of the sulfur-bearing shale interlayers within the metamorphic sandstone structure were determined based on lithological attribute labels, constructing an embedding feature parameter set. Based on fault structure distribution, lithological variation interfaces, stress concentration zones, and groundwater zoning information, tectonic domain division rules were established, and a parameter weighting mechanism was used to complete the tectonic domain partitioning of the three-dimensional geological structure model. In the tectonic domain partitioning results, structural perturbation factors and chemical reaction sensitivity factors were defined for each embedding feature parameter set, and dynamic evolution boundary conditions for the embedded feature parameter sets were set, constructing a parameter prediction regional system for dynamic prediction of rock mass mechanical parameters.

[0085] Specifically, firstly, borehole logging data, core scanning data, ground-penetrating radar (GPR) profile data, and 3D seismic reflection data are integrated to construct a spatial dataset under a unified coordinate system. The borehole logging data refers to stratigraphic records and lithological information obtained through manual drilling. The core scanning data refers to data obtained by high-resolution imaging and physical property identification of core samples extracted from underground. The GPR profile data is geological structural images obtained based on GPR technology. The 3D seismic reflection data obtains 3D image information of underground structural boundaries through artificial seismic wave reflection. These data, after spatial registration and standardization, constitute a continuous and analyzable 3D geological input source.

[0086] Then, spatial modeling methods are used to perform three-dimensional fitting on the structural boundaries (such as faults, interlayer contact surfaces), lithological interfaces, and geological discontinuities in the above data to establish a complete three-dimensional geological structure model. The spatial modeling is the process of transforming discrete geological data into a continuously analyzable three-dimensional structure; the structural boundary refers to the interface with differential deformation characteristics within the rock mass (such as fracture surfaces or fold axes).

[0087] Subsequently, point cloud data of the surrounding rock surface of the plant was acquired using laser scanning technology to establish an accurate surface mesh model. Based on voxel mesh generation technology, the entire three-dimensional structural model was divided into small volume units of equal size with spatial indexes. The voxel mesh is a cubic structure divided at equal intervals in three-dimensional space, used to represent the spatial distribution of various physical properties in space. Each volume unit is assigned a lithological attribute label to identify the geological material type, diagenetic characteristics, and geometric structural features of that location.

[0088] Next, sections with sulfide enrichment were extracted from borehole logging data and core elemental analysis results. Based on resistivity (rock conductivity), gamma spectrum (natural radioactivity intensity), and density characteristics, the location of sulfur-bearing shale interlayers was identified. A probabilistic model was established by combining the sulfur content in the elemental analysis, and a spatial distribution map of sulfur-bearing interlayers was generated in the model.

[0089] By combining geological profile and stratigraphic information, the structural geometry of sulfur-bearing interlayer blocks is described, and their embedding angle (angle with the surrounding rock interface), extension direction (spatial strike), thickness parameters (interlayer thickness), and contact surface properties (interlayer and surrounding rock boundary characteristics) are determined, forming a set of embedding feature parameters for structural analysis.

[0090] Based on this, tectonic domain division rules are established according to the distribution of fracture structures, lithological variation interfaces, stress concentration zones, and groundwater zoning information. A tectonic domain refers to a spatial region with consistency in mechanical response, hydrological characteristics, and structural composition. By assigning weighted parameters based on fracture strength, stress gradient, and hydrological activity to each volumetric unit, the tectonic domain division of the three-dimensional model is completed.

[0091] Finally, within the framework of the tectonic domain, a structural disturbance factor (measuring the responsiveness of joints and bedding planes to excavation disturbance) and a chemical reaction sensitivity factor (measuring the tendency of rock mass to undergo chemical changes such as sulfidation, dissolution, and expansion under contact with groundwater and gas) are defined for each embedded feature parameter set. At the same time, dynamic evolution boundary conditions for parameters are set for each tectonic domain, including the initial disturbance time, environmental change threshold, and spectral response change rate, forming a parameter prediction region system with the ability to express and model reaction mechanisms.

[0092] Overall, through high-resolution geological modeling, chemical response identification, and regional structural feature embedding, a foundation for dynamic modeling of rock mass mechanical parameters in complex interlayer systems has been established, significantly improving the physical accuracy and regional adaptability of rock mass response prediction during unsteady construction processes.

[0093] S120. Obtain the historical evolution trajectory of elastic modulus, cohesion, and shear strength, as well as infrared spectral characteristic data, of the sulfur-bearing shale interlayer region.

[0094] First, representative rock samples were selected within the identified sulfur-bearing shale interlayers, ensuring that the samples covered typical embedment angles, extension directions, thickness parameters, and different contact surface properties. Standard-sized test samples were prepared for simultaneous mechanical and spectroscopic testing. To simulate the disturbance environment induced by drainage during underground powerhouse construction, various stress loading paths and chemical erosion scenarios were set up, including static load-unloading cycles, immersion in solutions with different pH values, high-sulfur environment simulation, and redox cycle loading. Elastic modulus, cohesion, and shear strength were tested at each stage.

[0095] The elastic modulus was calculated using the slope of the linear segment of the stress-strain curve under axial loading; cohesion and shear strength were obtained through indoor direct shear tests and triaxial shear tests, and envelope fitting was performed using the Mohr-Coulomb criterion. Each set of tests was repeated after a set time step, and the values ​​of material parameters changing with time and environment were recorded to form a data sequence of elastic modulus, cohesion, and shear strength at multiple time points, thus constructing the historical evolution trajectory of rock mechanics parameters in this region.

[0096] Immediately after each mechanical test, infrared spectroscopy was performed on the resulting rock samples. Fourier transform infrared spectroscopy was used to acquire reflectance spectra in the 0.5 to 25 micrometer wavelength range. The focus was on extracting changes in the position and area of ​​the main absorption peak, band intensity ratios, and functional group peak values, such as hydroxyl, carboxyl, and sulfur-oxygen functional groups. Spectral features coupling with the evolution of mechanical parameters were identified. To improve the temporal resolution of the spectral analysis, the sampling frequency was increased during key change stages to ensure the continuous expression of the spectral response throughout the evolution process.

[0097] Finally, the elastic modulus, cohesion, and shear strength of multiple samples at different time points were matched with their corresponding infrared spectral features, and a parameter-spectral mapping database was established with a unified numbering structure. This formed a set of data on the historical evolution trajectory of rock mass parameters and infrared spectral features with three-dimensional mechanical-chemical-time expression capabilities, providing complete data support for subsequent regression model construction, spectral response factor extraction, and dynamic prediction model training.

[0098] S130. Based on historical evolutionary trajectories and infrared spectral characteristic data, a nonlinear regression function model is constructed.

[0099] In one possible implementation, before constructing a nonlinear regression function model based on historical evolution trajectories and infrared spectral feature data, the method further includes: extracting infrared spectral features corresponding to elastic modulus, cohesion, and shear strength from the infrared spectral feature data, including the position of the main absorption peak, the area of ​​the absorption peak, the intensity ratio of the spectral bands, and the changes in the characteristic peaks of functional groups, to construct infrared spectral feature factors; combining and encoding the infrared spectral feature factors with environmental control parameters, including the pH value of the wetting solution, dissolved oxygen concentration, the mass ratio of sulfate to chloride ions, and the loading rate, to construct an input variable set; performing principal component analysis on the infrared spectral feature factors, eliminating redundant dimensions, and retaining principal component factors that have a preset correlation with elastic modulus, cohesion, and shear strength; using the principal component factors and the input variable set as input variables, and using the predicted values ​​of elastic modulus, cohesion, and shear strength at different time steps as target output quantities, to construct a nonlinear regression function model.

[0100] Specifically, firstly, the infrared spectral characteristic data is analyzed to extract infrared spectral features highly correlated with the evolution of elastic modulus, cohesion, and shear strength, including the position of the main absorption peak, the absorption peak area, the band intensity ratio, and the changes in functional group characteristic peaks. The position of the main absorption peak refers to the wavenumber (in units of...) corresponding to the maximum absorption intensity in the infrared spectrum. The shift of the absorption peak reflects changes in bond energy within the mineral; the absorption peak area is the integral value of the main peak in the spectrum, representing the product of absorption intensity and functional group concentration; the band intensity ratio represents the ratio of absorption intensity of two characteristic bands, such as the ratio of hydroxyl peak to sulfur-oxygen peak, which can be used to quantify the evolution of the ratio of structural water to sulfides; the change in characteristic peak value of functional groups refers to the change in intensity of the absorption characteristics of specific functional groups in the infrared band, such as the carboxyl group (COOH) in... Sulfur and oxygen (S=O) in This corresponds to the stress corrosion and oxidation reaction behavior. Assume infrared spectra are collected at n time points, and the spectrum of each sample is a function... ,in For wave number, If the time is constant, then the position of the main absorption peak is:

[0101] ;

[0102] The absorption peak area is:

[0103] ;

[0104] The band intensity ratio is:

[0105] ;

[0106] Where: λ1 to λ2 is the main peak integration interval, λ a With λb The two functional groups represent the peak.

[0107] Then, the extracted infrared spectral feature factors are combined and encoded with environmental control parameters to construct an input variable set. The environmental control parameters include: the pH value of the wetting solution (representing corrosive acidity or alkalinity), dissolved oxygen concentration (affecting the oxidation reaction rate), the mass ratio of sulfate to chloride ions (regulating the reaction pathway between sulfides and halogens), and the loading rate (reflecting the influence of the mechanical excitation rate on the response behavior). When constructing the input variable set, the infrared spectral feature factor vector of each sample is concatenated with the environmental control parameter vector to form the total input vector X.

[0108] ;

[0109] in: For infrared spectral characteristic factors, For environmental control parameters, Input feature representation used to describe the mechanical state of rock mass under current conditions.

[0110] Next, principal component analysis is performed on the infrared spectral characteristic factors in the high-dimensional input variable X to remove redundant dimensions and retain principal component factors that are statistically significantly correlated with elastic modulus, cohesion, and shear strength. Principal component analysis is performed by solving the principal component matrix through covariance matrix decomposition. After sorting the eigenvalues ​​in descending order, the top n principal components with a cumulative variance contribution rate greater than 90% are retained:

[0111] ; ;

[0112] in: Let be the covariance matrix of the infrared spectral characteristic factors. Its corresponding eigenvector matrix, These are the principal component factors after dimensionality reduction.

[0113] Finally, the dimensionality-reduced principal component factor was used. With environmental control parameters Concatenate into the final set of input variables The predicted values ​​of elastic modulus, cohesion, and shear strength at different time steps are used as the target output quantities. Construct a nonlinear regression function model:

[0114] ;

[0115] Where f represents the nonlinear function mapping relationship, and ε represents the residual term. This nonlinear regression model can be implemented using a long short-term memory neural network, a radial basis function neural network, or a gradient boosting regression tree to fit the complex coupling relationship between the input variables and the target output, and to provide mapping capabilities for subsequent dynamic prediction. The training process uses the weighted mean squared error loss function:

[0116] ;

[0117] Where: N is the total number of samples. These are the error weighting coefficients for elastic modulus, cohesion, and shear strength, used to adjust the model's prediction accuracy for each physical parameter. After training, the model can be used to predict the dynamic evolution trajectory of rock mass mechanical parameters in real time under new infrared spectral and environmental variable conditions.

[0118] In one possible implementation, a nonlinear regression function model is constructed based on historical evolutionary trajectories and infrared spectral feature data. Specifically, this includes: standardizing the principal component factors and input variable set using zero mean normalization to form an input variable matrix at a uniform scale; using the input variable matrix as the input vector of the nonlinear regression function model, and using the predicted values ​​of elastic modulus, cohesion, and shear strength at each time step as the target outputs to construct a regression modeling structure with multiple output mappings; and selecting a long short-term memory neural network as the modeling structure for the nonlinear regression function model, receiving the input variable matrix at the input layer and using gating at the hidden layer. The recursive unit extracts the evolution features of variables and captures the causal relationship between changes in historical spectra and the evolution of rock mechanics parameters through the time state transition matrix. Predictive output nodes for elastic modulus, cohesion, and shear strength are set in the output layer, and a parallel structure is used to achieve multi-objective output. When constructing the loss function, a multi-objective error function is introduced to constrain the prediction accuracy of multiple objective outputs with weighted mean square error, and a time smoothing regularization term is introduced to suppress non-physical jumps. The Adam optimizer is used to train the nonlinear regression function model. The training process is completed under the condition of meeting the convergence criterion, and the model accuracy and generalization ability are evaluated through cross-validation.

[0119] Specifically, firstly, the set of input variables consisting of principal component factors and environmental control parameters is standardized using zero-mean normalization. For each input variable, the following formula is used for standardization:

[0120] ;

[0121] in, The variables are after normalization. For the original input variables, This represents the mean of the variable in the sample set. The standard deviation is denoted as . This treatment can unify the numerical scale of different variables, eliminate the influence of units, and improve the numerical stability of the model during training.

[0122] Then, the normalized input variable matrix is ​​used as the input vector of the nonlinear regression function model, with the output target being the elastic modulus. Cohesion With shear strength The predicted values ​​at multiple time steps are used to construct a regression model with a multi-output mapping relationship:

[0123] ;

[0124] in: It is a nonlinear mapping function. For time steps The input variable vector below, This is the corresponding target output vector.

[0125] Next, a long short-term memory neural network structure is adopted as the framework for constructing the regression model. This structure receives the input variable matrix in time-series form at the input layer, and extracts the features of the variables evolving over time in the hidden layer through gated recursive units (including input gate, forget gate, and output gate). A state transition mechanism is used to model the temporal causal relationship between spectral features and rock mass mechanical parameters. The output layer adopts a parallel structure, setting independent output channels for the three mechanical parameters to achieve decoupling of the three target outputs.

[0126] During the model training phase, a multi-objective loss function is defined, primarily in the form of weighted mean squared error, while a time smoothing regularization term is introduced to suppress non-physical abrupt changes in the predicted sequence. The total loss function is:

[0127] ;

[0128] in: These are the error weighting coefficients for elastic modulus, cohesion, and shear strength, respectively. This is the adjustment parameter for the time smoothing regularization term, where T is the total number of time steps.

[0129] Model training employs the Adam optimizer for parameter updates. The Adam optimizer is an adaptive gradient optimization method that comprehensively considers both the first moment (mean) and the second moment (variance), enabling fast and robust convergence in a non-convex loss function space. During training, convergence criteria are set, such as terminating training when the loss function decreases to a preset threshold or the validation set error shows no improvement for N consecutive rounds. After training, the model's generalization ability is evaluated using cross-validation, where the data is divided into multiple mutually exclusive subsets that are used alternately as the validation set. This ensures the model maintains stable predictive performance under different data distributions, ultimately yielding a nonlinear regression function model that reliably expresses the coupling relationship between infrared spectral response and the evolution of rock mass mechanical parameters.

[0130] S140. Embed the nonlinear regression function model into the parameter prediction region system, implement dynamic parameter assignment for the sulfur-bearing shale interlayer region, and construct a spatiotemporal variation map of parameters evolving with the construction drainage process.

[0131] In one possible implementation, the above steps specifically include: establishing a correspondence between each spatial unit and its spatial coordinates, lithological label, tectonic domain number, and interlayer identification status in the parameter prediction region system, and constructing a spatial unit index matrix belonging to the sulfur-bearing shale interlayer region; extracting the construction time step, drainage status, initial rock mass properties, and geochemical conditions of each spatial unit in the spatial unit index matrix to form a dataset for inputting into the nonlinear regression function model; using the environmental control parameters and infrared spectral characteristic factors of each spatial unit in the dataset at a specific time step as input, importing them into the nonlinear regression function model, and outputting the predicted values ​​of elastic modulus, cohesion, and shear strength. The system achieves dynamic parameter assignment for spatial units. During the dynamic parameter assignment process, a time filtering algorithm is used to constrain the rate of change of the prediction results of adjacent time steps, and a local weighted regression method is used for smoothing to ensure the physical continuity of dynamic parameters in the time series. The predicted elastic modulus, predicted cohesion, and predicted shear strength corresponding to each time step are summarized in the form of three-dimensional tensors to construct a spatiotemporal variation map of parameters covering the sulfur-bearing shale interlayer region. The spatiotemporal variation map of parameters is embedded into the parameter prediction region system to support rock mass response simulation, support structure design evolution calculation, and automatic identification and analysis of abnormal trends, realizing full-process dynamic prediction and model feedback correction of rock mass mechanical parameters.

[0132] Specifically, firstly, a complete attribute identifier is established for each spatial unit in the parameter prediction regional system. The spatial coordinates, lithological labels, tectonic domain numbers, and interlayer identification status of each spatial unit are bound one by one, forming a multi-dimensional identifier system with geographical location, physical properties, and structural attribution information. Based on this identifier system, all spatial units identified as sulfur-bearing shale interlayers are extracted from the three-dimensional geological structure model, and a spatial unit index matrix is ​​constructed to identify all sets of computational units that require dynamic parameter prediction.

[0133] Secondly, the local construction progress, drainage status, initial rock mass property parameters (such as initial elastic modulus, initial porosity, etc.) and geochemical conditions (such as pore water pH value, etc.) of each spatial unit at the current construction time step are extracted from the spatial unit index matrix. / Ratios, etc., serve as core variables for constructing the input dataset. This dataset is used to build the input tensor required for the current prediction point in time, ensuring that the model input has consistent contextual information in both the temporal and spatial dimensions.

[0134] Next, the environmental control parameters and infrared spectral characteristic factors of each spatial unit in the above input dataset at a specific time step are used as input variables and imported into the trained nonlinear regression model. This outputs the predicted values ​​of the elastic modulus, cohesion, and shear strength of that spatial unit at the current time step, thus achieving dynamic parameter assignment for that spatial unit. Each prediction result here has a spatial dimension and a time step label, achieving accurate mapping of the prediction results to the parameter prediction region system.

[0135] In the process of assigning dynamic parameters, considering the potential for drastic fluctuations in parameters during construction disturbances, a time filtering algorithm is used to limit the rate of change of the prediction results at adjacent time steps, constraining the range of change of their time derivatives, and ensuring that the trend of the predicted values ​​conforms to the physical response law of the rock mass. At the same time, a local weighted regression method is introduced to smooth the prediction results at each time point, using the weighted average of the predicted values ​​before and after within the time window as the final output of the current time step, thereby enhancing the temporal continuity and physical consistency of the predicted values.

[0136] Subsequently, the predicted elastic modulus, predicted cohesion, and predicted shear strength of each spatial unit at each time step are combined into a three-dimensional tensor form according to spatial location and time sequence, and parameter tensors E(x,y,z,t), C(x,y,z,t), and S(x,y,z,t) are constructed respectively to form a complete spatiotemporal variation map of parameters, covering the entire sulfur-bearing shale interlayer region.

[0137] Finally, the spatiotemporal variation map of the above parameters is embedded into the parameter prediction region system as a dynamic input of the rock mass mechanical state, driving the operation of the surrounding rock response prediction model, the support structure design evolution model, and the risk trend identification model during construction. This supports the full-process safety assessment, structural response control, and model closed-loop correction of the underground powerhouse under complex interlayer conditions, thereby realizing a dynamic prediction system for rock mass mechanical parameters based on environment-driven, spectroscopic response, and time evolution.

[0138] S150: Acquire laser point cloud data and microseismic monitoring data, invert the actual deformation trend and microfracture density change of the surrounding rock during construction, and output the actual monitoring results.

[0139] In one possible implementation, the above steps specifically include: collecting point cloud data of the construction area at different time periods by deploying a high-frequency laser scanner and a three-dimensional laser ranging system in the construction area, and performing attitude registration, noise filtering, and mesh reconstruction on the point cloud data to form a time-series point cloud model of the deformation of the surrounding rock surface; based on the point cloud model, calculating the displacement vector, normal convergence value, and profile deformation rate of each monitoring block in the construction area at different time steps to construct a deformation trend map of the surrounding rock; and collecting microseismic event waveform data, including event trigger time, amplitude, spectral characteristics, and initial motion, by deploying a microseismic monitoring array system around the construction area. The research focuses on: spatially locating microseismic event waveform data based on waveform analysis and travel-time inversion methods to construct a three-dimensional spatiotemporal evolution map of the microseismic event; generating a microfracture density distribution map using a fracture cluster identification algorithm based on the three-dimensional spatiotemporal evolution map; performing spatiotemporal registration and spatial overlay analysis on the surrounding rock deformation trend map and the microfracture density distribution map to identify structural disturbance regions with coupling characteristics in the surrounding rock mass, and outputting actual monitoring results including deformation rate field, microfracture density field, and structural disturbance level; embedding the actual monitoring results into the tectonic domain coordinate system for deviation comparison with the parameter spatiotemporal change map and parameter degradation function recalibration judgment.

[0140] Specifically, by deploying high-frequency laser scanners and 3D laser ranging systems in the construction area, point cloud data of the construction area at different time periods are collected. The point cloud data is then subjected to attitude registration, noise filtering, and mesh reconstruction to form a time-series point cloud model of the deformation of the surrounding rock surface. Specifically, multiple fixed high-frequency laser scanners and 3D laser ranging devices are set up in the excavation area of ​​the underground powerhouse surrounding rock, and a synchronous trigger control module is used to acquire point cloud data at regular intervals. Each frame of point cloud data is first aligned to a unified coordinate system through an attitude registration algorithm (such as the ICP algorithm), and then isolated points and noise caused by equipment errors are removed using a statistical filter. Finally, the multiple frames of point cloud data in the time series are reconstructed into triangular meshes according to the equal spacing rule to construct a time-series point cloud model that can reflect the deformation evolution of the surrounding rock surface, which serves as the geometric basis for subsequent displacement and deformation calculations.

[0141] Based on the point cloud model, the displacement vector, normal convergence value, and profile deformation rate of each monitoring block in the construction area at different time steps are calculated to construct a surrounding rock deformation trend map. Specifically, the point cloud model is first divided into several spatial monitoring blocks. Based on the spatial correspondence between the initial point cloud frame and the point cloud frames at each time step, the three-dimensional displacement vector of each point in the time series is calculated. Then, the unit normal displacement change rate is calculated in the normal direction of the monitoring block and defined as the normal convergence value. At the same time, the profile deformation change trend of the characteristic cross section of the surrounding rock in the time series is extracted, and the profile curvature and shear strain rate are calculated by surface fitting. Finally, the above displacement vector field, convergence rate field, and deformation field are combined to form the surrounding rock deformation trend map.

[0142] By deploying a microseismic monitoring array system around the construction area, waveform data of microseismic events are collected, including event trigger time, amplitude, spectral characteristics, and initial motion direction. Specifically, a microseismic monitoring array consisting of multiple three-component short-period seismic sensors is deployed in the underground powerhouse and surrounding rock mass. Waveform data is continuously collected at high temporal resolution (e.g., 1 kHz) through a multi-channel data acquisition system. The STA / LTA short-time energy ratio detection algorithm is applied to perform real-time trigger analysis on the waveform data, extracting the trigger time, maximum amplitude, dominant frequency, and spectral energy distribution of each microseismic event. The initial motion direction of the event source is obtained by determining the initial motion polarity, forming an event waveform feature dataset.

[0143] Based on waveform analysis and travel time inversion methods, spatial positioning of microseismic event waveform data is performed to construct a three-dimensional spatiotemporal evolution map of the microseismic event. Specifically, the travel time inversion algorithm is combined with a three-dimensional velocity model to perform spatial inversion of the arrival time difference of P-waves and S-waves of the microseismic event received by each sensor. Nonlinear positioning methods (such as SimulPS) are used to obtain the precise location of the seismic source in three-dimensional spatial coordinates. The event trigger time is combined with the spatial location to generate a three-dimensional event point set with timestamp attributes, which constitutes the three-dimensional spatiotemporal evolution map of the microseismic event.

[0144] Based on the three-dimensional spatiotemporal evolution map, a micro-fracture density distribution map is generated using a fracture cluster identification algorithm. Specifically, a density clustering algorithm (such as DBSCAN) is used to cluster the microseismic event point set to identify temporally and spatially continuous fracture activity areas. Within each cluster subset, the number of events per unit volume is counted to obtain the micro-fracture event density within that area. Furthermore, density field interpolation is performed on the entire underground powerhouse area according to the tectonic domain division rules to form a micro-fracture density distribution map covering the entire area, reflecting the spatial activity pattern of micro-scale fracture activity.

[0145] The deformation trend map of the surrounding rock and the microfracture density distribution map are spatiotemporally registered and spatially overlaid to identify structural disturbance regions with coupling characteristics in the surrounding rock mass. The actual monitoring results, including the deformation rate field, microfracture density field, and structural disturbance level, are output. Specifically, spatial interpolation is used to unify the two types of maps to the same grid resolution and time step. Coupling analysis is performed based on regional spatial units. A joint threshold criterion is set for regions that simultaneously show abnormal deformation rate and high microfracture density values. The disturbance level is divided by fuzzy logic or empirical rules, and the coupling response degree layer is output. Finally, a dataset of actual monitoring results containing the deformation rate field, microfracture density field, and disturbance level is generated.

[0146] The actual monitoring results are embedded into the structural domain coordinate system for deviation comparison with the spatiotemporal variation map of parameters and for recalibration judgment of parameter degradation functions. Specifically, the actual monitoring results are made consistent with the structural domain coordinates in the parameter prediction area system, and the monitoring values ​​are mapped to the corresponding prediction units according to the spatial unit numbers. The deviation between the monitoring values ​​and the predicted values ​​at the same time step is calculated to construct a deviation matrix. Based on the spatial region exceeding the set threshold, the recalibration mechanism of the parameter degradation function is triggered to realize the dynamic correction and parameter update of the prediction model, providing highly reliable monitoring support for subsequent mechanical state feedback and support optimization of the structural domain.

[0147] S160. When the actual monitoring result deviates from the predicted envelope of the spatiotemporal variation map of the parameters, the recalibration process of the parameter degradation function is triggered until the error between the actual monitoring result and the predicted result of the spatiotemporal variation map of the parameters meets the convergence condition, and the parameter update is stopped.

[0148] In one possible implementation, when the actual monitoring results deviate from the predicted envelope of the parameter spatiotemporal variation map, a recalibration process of the parameter degradation function is triggered to achieve dynamic updating and prediction output of rock mechanics parameters. Specifically, this includes: spatially aligning the deformation rate field and microfracture density field from the actual monitoring results with the parameter spatiotemporal variation map in the tectonic domain coordinate system; extracting the predicted and monitored values ​​of each spatial unit at the same time step; calculating the prediction deviations of elastic modulus, cohesion, and shear strength; constructing a deviation magnitude matrix; determining whether there are target spatial units in the deviation magnitude matrix that exceed a preset error tolerance threshold; if there are target spatial units, defining the monitoring block where the target spatial unit is located as the target recalibration region; extracting the infrared spectral characteristic factors and environmental control parameters corresponding to the target recalibration region at the current time step; and constructing a local training model based on the actual monitoring values. Based on the data set, a transfer learning mechanism is adopted. Using the original nonlinear regression function model as a foundation, and while keeping irrelevant parameters frozen, the local training dataset is retrained or the parameters are fine-tuned. The weight paths related to prediction bias are adaptively adjusted using the error backpropagation algorithm. The updated nonlinear regression function model is used to re-predict the parameters of spatial units in the target recalibration region, replacing the predicted values ​​at corresponding positions in the parameter spatiotemporal variation map, resulting in a corrected parameter spatiotemporal variation map. A transition region is constructed between the target recalibration region and the uncorrected region. A weighted fusion strategy is used to continuously correct the outputs of the nonlinear regression function model and the updated nonlinear regression function model, maintaining the spatial smoothness of the parameter field. The corrected parameter spatiotemporal variation map is re-embedded into the parameter prediction region system, realizing the dynamic evolution and closed-loop update of rock mechanics parameters under construction disturbance conditions.

[0149] Specifically, the deformation rate field and microfracture density field from the actual monitoring results are spatially aligned with the spatiotemporal variation map of parameters in the tectonic domain coordinate system. The predicted and monitored values ​​for each spatial unit at the same time step are extracted, and the predicted deviations of elastic modulus, cohesion, and shear strength are calculated to construct a deviation magnitude matrix. Based on the spatial unit number in the tectonic domain coordinate system, the monitoring data and predicted data are resampled and synchronized with the time step within the same spatial grid. The deviations of each spatial unit at the time step are then calculated. Predicted value Compared with monitoring inversion values Based on the differences, construct a prediction bias vector:

[0150] ;

[0151] The prediction biases of all spatial units at the same time step are combined into a deviation magnitude matrix. It is used to determine the effectiveness of local predictions and identify mismatched segments.

[0152] Determine whether there are target spatial cells in the deviation amplitude matrix that exceed the preset error tolerance threshold. If a target spatial cell exists, define the monitoring block where the target spatial cell is located as the target recalibration area, and specifically set a physically meaningful error tolerance threshold. For each spatial unit, determine whether its three types of parameter deviations satisfy the following:

[0153] ;

[0154] If any condition is met, the spatial unit is classified as a prediction mismatch unit, and a monitoring block containing multiple units is constructed based on its spatial neighborhood radius as the target recalibration region for the correction and reconstruction of the local regression model.

[0155] Infrared spectral feature factors and environmental control parameters corresponding to the target recalibration region at the current time step are extracted. A local training dataset is constructed by combining this dataset with actual monitoring values. Specifically, the original input vectors of each spatial unit in the target recalibration region are extracted, including the infrared spectral feature factors after principal component dimensionality reduction and the standardized environmental control parameters, to construct a local input matrix. Simultaneously, the monitoring inversion values ​​are extracted to form the target output matrix. This forms a training dataset that can be used for local model correction. .

[0156] A transfer learning mechanism is employed, based on the original nonlinear regression model. While keeping irrelevant parameters frozen, retraining or parameter fine-tuning is performed on a local training dataset. The backpropagation algorithm adaptively adjusts weight paths related to prediction bias. Specifically, the general weight layers in the original model (such as the input embedding layer and some hidden layers) are frozen, while only weight paths related to the target output are unfrozen. Parameters are updated using local training data with a small learning rate, following the following update rules:

[0157] ;

[0158] in: These are the model weight parameters. For learning rate, This is the local error loss function. This strategy improves the response to local prediction biases while retaining the original model's generalization ability to non-recalibrated regions.

[0159] The updated nonlinear regression model is used to re-predict the parameters of spatial cells in the target recalibration region, and the predicted values ​​at the corresponding positions in the spatiotemporal variation map of the parameters are replaced to obtain the corrected spatiotemporal variation map of the parameters. Specifically, the corrected model is applied to the input variables... The forward inference is performed to obtain the corrected prediction value:

[0160] ;

[0161] Then, it is embedded into the corresponding spatial unit and time step position in the original parametric map, replacing the original predicted value, and constructing a parametric tensor containing correction information. .

[0162] A transition region is constructed between the target recalibrated region and the uncorrected region. A weighted fusion strategy is used to continuously correct the outputs of the nonlinear regression function model and the updated nonlinear regression function model, maintaining the spatial smoothness of the parameter field. Specifically, the transition boundary bandwidth is defined. Weighted fusion of boundary units:

[0163] ;

[0164] in: For fusion weights based on spatial distance, This is the minimum distance from the current cell center to the boundary of the recalibration region.

[0165] The revised spatiotemporal variation map of the parameters is re-embedded into the parameter prediction regional system to achieve dynamic evolution and closed-loop updating of rock mass mechanics parameters under construction disturbance conditions. Specifically, the completed system will be... The data structure is re-imported into the parameter prediction region system to update the physical parameter status of all relevant spatial units in the corresponding tectonic domain. This serves as the data input for subsequent iterative prediction and support strategy correction, realizing a closed loop of adaptive update of parameter modeling driven by monitoring data.

[0166] Furthermore, at each construction time step Next, extract the predicted elastic modulus values ​​from all structural domain spatial elements. Cohesion prediction value Predicted shear strength Simultaneous monitoring and inversion values , , Perform point-by-point difference calculations to form a prediction error vector:

[0167] ;

[0168] ;

[0169] ;

[0170] The differences of each type of parameter are used to form three sets of error matrices, covering all target spatial units and the current time step.

[0171] For each type of rock mass mechanical parameter, an acceptable error tolerance threshold is set, denoted as:

[0172] ;

[0173] The threshold is used to determine whether the prediction of each spatial unit at the current time step meets the convergence requirement. If all three types of parameters simultaneously meet the condition of not exceeding the limit, then the unit is regarded as a "locally convergent unit" at that time step.

[0174] To avoid misjudgments caused by random disturbances, a minimum continuous time window is defined. (e.g., 3 time steps) Within this time window, if a spatial unit satisfies the above three types of error threshold conditions at all time steps, then the unit is determined to have reached a stable convergence state, denoted as:

[0175] ;

[0176] Then the spatial unit is considered to meet the prediction error convergence threshold condition.

[0177] Statistical analysis is performed on all target spatial units. If more than a set proportion (e.g., 90%) of the spatial units in the recalibration area meet the above continuous convergence condition, then the entire recalibration area is determined to have reached the error convergence state, and the system automatically terminates the parameter update process of that area.

[0178] For spatial units that have converged, their predicted state is marked as "frozen". They will no longer participate in model correction and prediction update in subsequent iterations, but will still retain their role in the maintenance of full map continuity and neighborhood transition.

[0179] In summary, by setting absolute error thresholds for each type of physical parameter, introducing minimum time window constraints and spatial unit convergence ratio control, we have achieved strict control over the convergence of prediction errors, ensuring that parameter updates can be terminated in a timely manner after reaching a steady-state response, thus guaranteeing model stability and engineering safety.

[0180] Example 2:

[0181] This embodiment also discloses a rock mechanics parameter prediction device based on the underground powerhouse of a pumped storage power station, referring to... Figure 2 The device includes an acquisition module 201, a processing module 202, and an output module 203. It is used to execute any of the rock mechanics parameter prediction methods described above for underground powerhouses of pumped storage power stations, wherein:

[0182] The acquisition module 201 is used to acquire the three-dimensional geological structure model and lithological information of the construction area, identify the embedding location and spatial distribution characteristics of sulfur-bearing shale interlayers in the metamorphic sandstone structure, and construct a parameter prediction regional system based on tectonic domain partitioning.

[0183] The acquisition module 201 is used to acquire the historical evolution trajectory of elastic modulus, cohesion and shear strength, and infrared spectral characteristic data of sulfur-bearing shale interlayer regions.

[0184] The processing module 202 is used to construct a nonlinear regression function model based on historical evolution trajectory and infrared spectral feature data. The nonlinear regression function model takes infrared spectral feature factors and environmental control parameters as input variables and the predicted evolution trajectory of elastic modulus, cohesion and shear strength as target output quantities.

[0185] The processing module 202 is used to embed the nonlinear regression function model into the parameter prediction region system, implement dynamic parameter assignment for the sulfur-bearing shale interlayer region, and construct a spatiotemporal variation map of parameters that evolves with the construction drainage process.

[0186] The acquisition module 201 is used to acquire laser point cloud data and microseismic monitoring data, invert the actual deformation trend and microfracture density change of the surrounding rock during construction, and output the actual monitoring results.

[0187] The output module 203 is used to trigger the recalibration process of the parameter degradation function when the actual monitoring result deviates from the predicted envelope of the parameter spatiotemporal variation map, until the error between the actual monitoring result and the predicted result of the parameter spatiotemporal variation map meets the convergence condition, and then stop the parameter update.

[0188] In one possible implementation, the output module 203 is used to fuse borehole logging data, core scanning data, ground-penetrating radar profile data, and three-dimensional seismic reflection data to establish a spatial dataset in a unified coordinate system.

[0189] The processing module 202 is used to perform three-dimensional fitting of the structural boundaries, lithological interfaces and discontinuities in the spatial dataset using spatial modeling methods, and to construct a three-dimensional geological structure model of the construction area.

[0190] The processing module 202 is used to generate a grid of the surrounding rock surface of the plant based on laser scanning point cloud data, and to divide the three-dimensional geological structure model into multiple block units through voxel mesh generation technology, and to assign lithological attribute labels to each block unit.

[0191] The acquisition module 201 is used to extract lithological segments with sulfide enrichment responses from borehole logging data and core elemental analysis results, identify the combined characteristics of resistivity, gamma spectrum and density, and construct a spatial probability distribution map of sulfur-bearing shale interlayers.

[0192] The acquisition module 201 is used to combine geological profiles and stratigraphic information, and based on lithological attribute tags, to determine the embedding angle, extension direction, thickness parameters and contact surface attributes of sulfur-bearing shale interlayers in metamorphic sandstone structures, and to construct an embedding feature parameter set.

[0193] The acquisition module 201 is used to establish tectonic domain division rules based on the distribution of fracture structures, lithological variation interfaces, stress concentration zones and groundwater zoning information, and to complete the tectonic domain division of the three-dimensional geological structure model by using a parameter weighting mechanism.

[0194] The acquisition module 201 is used to define the structural perturbation factor and chemical reaction sensitivity factor of each embedded feature parameter set in the result of structural domain partitioning, and to set the parameter dynamic evolution boundary conditions of the embedded feature parameter set to construct a parameter prediction region system for dynamic prediction of rock mass mechanics parameters.

[0195] In one possible implementation, the acquisition module 201 is used to extract the infrared spectral features corresponding to the elastic modulus, cohesion and shear strength from the infrared spectral feature data, including the position of the main absorption peak, the area of ​​the absorption peak, the ratio of the spectral band intensity and the change of the characteristic peak of the functional group, and to construct the infrared spectral feature factor.

[0196] The acquisition module 201 is used to combine and encode infrared spectral characteristic factors with environmental control parameters, including the pH value of the wetting solution, dissolved oxygen concentration, mass ratio of sulfate to chloride ions and loading rate, to construct an input variable set.

[0197] The acquisition module 201 is used to perform principal component analysis on the infrared spectral feature factors, remove redundant dimensions, and retain principal component factors that have a preset strength of correlation with elastic modulus, cohesion and shear strength.

[0198] Output module 203 is used to construct a nonlinear regression function model by using principal component factors and the set of input variables as input variables, and using the predicted values ​​of elastic modulus, cohesion and shear strength at different time steps as target output quantities.

[0199] In one possible implementation, the processing module 202 is used to perform zero-mean normalization on the principal component factors and the set of input variables to form an input variable matrix under a uniform scale.

[0200] The processing module 202 is used to take the input variable matrix as the input vector of the nonlinear regression function model, and take the predicted values ​​of elastic modulus, cohesion and shear strength at each time step as the target output, and construct a regression modeling structure with multiple output mapping relationships.

[0201] The processing module 202 is used to select a long short-term memory neural network as the modeling structure for the nonlinear regression function model. It receives the input variable matrix in the input layer, extracts the variable evolution features in the hidden layer through gated recursive units, and captures the causal relationship between the changes in the historical spectrum and the evolution of rock mass mechanical parameters through the time state transition matrix.

[0202] The output module 203 is used to set prediction output nodes for elastic modulus, cohesion and shear strength in the output layer respectively, and adopts a parallel structure to realize multi-objective output.

[0203] The processing module 202 is used to introduce a multi-objective error function when constructing the loss function, to constrain the prediction accuracy of multiple target outputs with weighted mean square error, and to introduce a time smoothing regularization term to suppress non-physical jumps.

[0204] The processing module 202 is used to train the nonlinear regression function model using the Adam optimizer, complete the training process under the condition of meeting the convergence criterion, and evaluate the model accuracy and generalization ability through cross-validation.

[0205] In one possible implementation, the processing module 202 is used to establish a correspondence between each spatial unit and its spatial coordinates, lithological label, tectonic domain number and interlayer identification status in the parameter prediction region system, and to construct a spatial unit index matrix belonging to the sulfur-bearing shale interlayer region.

[0206] The acquisition module 201 is used to extract the construction time step, drainage status, initial rock mass properties and geochemical conditions of each spatial unit in the spatial unit index matrix, forming a dataset for inputting into the nonlinear regression function model.

[0207] The processing module 202 is used to take the environmental control parameters and infrared spectral characteristic factors of each spatial unit in the dataset at a specific time step as input, import them into a nonlinear regression function model, and output the predicted values ​​of elastic modulus, cohesion and shear strength, thereby realizing the dynamic parameter assignment of the spatial unit.

[0208] The processing module 202 is used to constrain the rate of change of the prediction results of adjacent time steps by using a time filtering algorithm during the dynamic parameter assignment process, and to perform smoothing based on the local weighted regression method to ensure the physical continuity of the dynamic parameters in the time series.

[0209] The processing module 202 is used to summarize the predicted elastic modulus, predicted cohesion and predicted shear strength corresponding to each time step in the form of three-dimensional tensors to construct a spatiotemporal variation map of parameters covering the sulfur-bearing shale interlayer region.

[0210] The processing module 202 is used to embed the spatiotemporal variation map of parameters into the parameter prediction area system to support rock mass response simulation, support structure design evolution calculation and automatic identification and analysis of abnormal trends, so as to realize the dynamic prediction of rock mass mechanical parameters throughout the whole process and model feedback correction.

[0211] In one possible implementation, the processing module 202 is used to collect point cloud data of the construction area at different times based on the deployment of a high-frequency laser scanner and a three-dimensional laser ranging system in the construction area, and to perform attitude registration, noise filtering and mesh reconstruction on the point cloud data to form a time series point cloud model of the deformation of the surrounding rock surface.

[0212] The processing module 202 is used to calculate the displacement vector, normal convergence value and profile deformation rate of each monitoring block in the construction area at different time steps based on the point cloud model, and to construct the surrounding rock deformation trend map.

[0213] The processing module 202 is used to collect microseismic event waveform data, including event trigger time, amplitude, spectral characteristics and initial motion direction, based on the microseismic monitoring array system deployed around the construction area.

[0214] The processing module 202 is used to spatially locate the waveform data of microseismic events based on waveform analysis and travel time inversion methods, and to construct a three-dimensional spatiotemporal evolution map of the microseismic events.

[0215] Based on the three-dimensional spatiotemporal evolution map, a crack cluster identification algorithm is used to generate a micro-fracture density distribution map.

[0216] By performing spatiotemporal registration and spatial overlay analysis on the surrounding rock deformation trend map and the microfracture density distribution map, structural disturbance regions with coupling characteristics in the surrounding rock mass are identified, and actual monitoring results including deformation rate field, microfracture density field and structural disturbance level are output.

[0217] The processing module 202 is used to embed the actual monitoring results into the coordinate system of the constructed domain, and to compare the deviation with the spatiotemporal variation map of the parameters and to determine the recalibration of the parameter degradation function.

[0218] In one possible implementation, the processing module 202 is used to spatially align the deformation rate field and micro-fracture density field in the actual monitoring results with the spatiotemporal variation map of parameters in the structural domain coordinate system, extract the predicted value and the monitored value of each spatial unit at the same time step, calculate the prediction deviation of elastic modulus, cohesion and shear strength, and construct the deviation magnitude matrix.

[0219] The processing module 202 is used to determine whether there is a target spatial unit in the deviation magnitude matrix that exceeds the preset error tolerance threshold. If there is a target spatial unit, the monitoring block where the target spatial unit is located is defined as the target recalibration area.

[0220] The acquisition module 201 is used to extract the infrared spectral feature factors and environmental control parameters of the target recalibration region at the current time step, and construct a local training dataset by combining the actual monitoring values.

[0221] The processing module 202 is used to retrain or fine-tune the parameters of the local training dataset based on the original nonlinear regression function model using a transfer learning mechanism, while keeping the irrelevant parameters frozen, and adaptively adjust the weight path related to the prediction bias through the error backpropagation algorithm.

[0222] The processing module 202 is used to re-predict the parameters of the spatial units in the target recalibration region using the updated nonlinear regression function model, and replace the predicted values ​​at the corresponding positions in the parameter spatiotemporal variation map to obtain the corrected parameter spatiotemporal variation map.

[0223] The processing module 202 is used to construct a transition region between the target recalibrated region and the uncorrected region. It adopts a weighted fusion strategy to continuously correct the output of the nonlinear regression function model and the updated nonlinear regression function model, thereby maintaining the spatial smoothness of the parameter field.

[0224] The processing module 202 is used to re-embed the corrected spatiotemporal variation map of parameters into the parameter prediction region system, so as to realize the dynamic evolution and closed-loop update of rock mass mechanics parameters under construction disturbance conditions.

[0225] It should be noted that the above embodiments of the apparatus are only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the apparatus and method embodiments provided in the above embodiments belong to the same concept, and the specific implementation process can be found in the method embodiments, which will not be repeated here.

[0226] Example 3:

[0227] This embodiment also discloses an electronic device, as shown in the reference. Figure 3 The electronic device may include: at least one processor 301, at least one communication bus 302, user interface 303, network interface 304, and at least one memory 305.

[0228] The communication bus 302 is used to enable communication between these components.

[0229] The user interface 303 may include a display screen and a camera. Optionally, the user interface 303 may also include a standard wired interface and a wireless interface.

[0230] The network interface 304 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface).

[0231] The processor 301 may include one or more processing cores. The processor 301 connects to various parts of the server using various interfaces and lines, and performs various functions of the server and processes data by running or executing instructions, programs, code sets or instruction sets stored in the memory 305, and by calling data stored in the memory 305.

[0232] exist Figure 3 In the illustrated electronic device, the user interface 303 is mainly used to provide an input interface for the user and to obtain the user's input data. The processor 301 can be used to call an application program stored in the memory 305 that provides a method for predicting rock mechanics parameters based on the underground powerhouse of a pumped storage power station. When executed by one or more processors 301, the electronic device performs one or more methods as described in the above embodiments.

[0233] The aforementioned memory 305 includes various media capable of storing program code, such as USB flash drives, portable hard drives, magnetic disks, or optical disks.

[0234] This application also discloses a computer-readable storage medium storing instructions. When executed by one or more processors 301, these instructions cause an electronic device to perform one or more methods as described in the above embodiments.

[0235] The above are merely exemplary embodiments of this application and should not be construed as limiting the scope of this application. Any equivalent changes and modifications made in accordance with the teachings of this application shall still fall within the scope of this application. Those skilled in the art will readily conceive of other embodiments of this application upon considering the disclosure of the specification and practical truths. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary technical means in the art not described in this application. The specification and embodiments are considered exemplary only, and the scope and spirit of this application are defined by the claims.

Claims

1. A method for predicting rock mechanics parameters based on the underground powerhouse of a pumped storage power station, characterized in that, The method includes: Obtain a three-dimensional geological structure model and lithological information of the construction area, identify the embedding location and spatial distribution characteristics of sulfur-bearing shale interlayers in the metamorphic sandstone structure, and construct a parameter prediction regional system based on tectonic domain partitioning. The historical evolution trajectory of elastic modulus, cohesion and shear strength, as well as infrared spectral characteristic data of the sulfur-bearing shale interlayer region were obtained; Based on the historical evolution trajectory and the infrared spectral feature data, a nonlinear regression function model is constructed. The nonlinear regression function model takes infrared spectral feature factors and environmental control parameters as input variables and the predicted evolution trajectory of elastic modulus, cohesion and shear strength as target output quantities. The nonlinear regression function model is embedded into the parameter prediction region system, and dynamic parameter assignment is performed on the sulfur-bearing shale interlayer region to construct a spatiotemporal variation map of parameters that evolves with the construction drainage process. Acquire laser point cloud data and microseismic monitoring data, invert the actual deformation trend and microfracture density change of the surrounding rock during construction, and output the actual monitoring results; When the actual monitoring result deviates from the predicted envelope of the spatiotemporal variation map of the parameters, the recalibration process of the parameter degradation function is triggered until the error between the actual monitoring result and the predicted result of the spatiotemporal variation map of the parameters meets the convergence condition, and the parameter update stops.

2. The method for predicting rock mechanics parameters based on the underground powerhouse of a pumped storage power station according to claim 1, characterized in that, The acquisition of a three-dimensional geological structure model and lithological information of the construction area, identification of the embedding location and spatial distribution characteristics of sulfur-bearing shale interlayers in the metamorphic sandstone structure, and construction of a parameter prediction regional system based on tectonic domain zoning specifically include: By fusing borehole logging data, core scanning data, ground-penetrating radar profile data, and 3D seismic reflection data, a spatial dataset under a unified coordinate system is established. A spatial modeling method was used to perform three-dimensional fitting on the structural boundaries, lithological interfaces and discontinuities in the spatial dataset to construct a three-dimensional geological structure model of the construction area. A grid of the surrounding rock surface of the factory is generated based on laser scanning point cloud data, and the three-dimensional geological structure model is divided into multiple block units by voxel mesh generation technology, and each block unit is assigned a lithological attribute label. By acquiring lithological sections with sulfide enrichment responses from borehole logging data and core elemental analysis results, identifying the combined characteristics of resistivity, gamma ray spectrum and density, and constructing a spatial probability distribution map of sulfur-bearing shale interlayers; Combining geological profiles and stratigraphic information, the embedding angle, extension direction, thickness parameters, and contact surface attributes of the sulfur-bearing shale interlayer in the metamorphic sandstone structure are determined based on the lithological attribute tags, and an embedding feature parameter set is constructed. Based on the distribution of fracture structures, lithological variation interfaces, stress concentration zones and groundwater zoning information, tectonic domain division rules are established, and a parameter weighting mechanism is used to complete the tectonic domain division of the three-dimensional geological structure model. In the results of the structural domain partitioning, the structural perturbation factor and chemical reaction sensitivity factor of each of the embedded feature parameter sets are defined, and the parameter dynamic evolution boundary conditions of the embedded feature parameter sets are set to construct a parameter prediction region system for dynamic prediction of rock mass mechanics parameters.

3. The method for predicting rock mechanics parameters based on the underground powerhouse of a pumped storage power station according to claim 1, characterized in that, Before constructing the nonlinear regression function model based on the historical evolution trajectory and the infrared spectral feature data, the method further includes: Extract the infrared spectral features corresponding to the elastic modulus, cohesion and shear strength from the infrared spectral feature data, including the position of the main absorption peak, the area of ​​the absorption peak, the ratio of the spectral band intensity and the change of the characteristic peak of the functional group, and construct the infrared spectral feature factor; The infrared spectral characteristic factors are combined and encoded with environmental control parameters, including the pH value of the wetting solution, dissolved oxygen concentration, mass ratio of sulfate to chloride ions, and loading rate, to construct an input variable set; Principal component analysis was performed on the infrared spectral characteristic factors to remove redundant dimensions and retain the principal component factors that have a preset strength of correlation with elastic modulus, cohesion and shear strength. Using the principal component factors and the set of input variables as input variables, and the predicted values ​​of the elastic modulus, the cohesion, and the shear strength at different time steps as target outputs, the nonlinear regression function model is constructed.

4. The method for predicting rock mechanics parameters based on the underground powerhouse of a pumped storage power station according to claim 3, characterized in that, The construction of a nonlinear regression function model based on the historical evolution trajectory and the infrared spectral feature data specifically includes: The principal component factors and the input variable set are subjected to zero-mean normalization to form an input variable matrix under a uniform scale. The input variable matrix is ​​used as the input vector of the nonlinear regression function model, and the predicted values ​​of the elastic modulus, the cohesion and the shear strength at each time step are used as the target output quantities to construct a regression modeling structure with multiple output mapping relationships. Long short-term memory neural networks are selected as the modeling structure for the nonlinear regression function model. The input variable matrix is ​​received in the input layer, the variable evolution features are extracted in the hidden layer through gated recursive units, and the causal relationship between the changes in historical spectrum and the evolution of rock mass mechanical parameters is captured through the time state transition matrix. Predictive output nodes for the elastic modulus, cohesion, and shear strength are set in the output layer respectively, and a parallel structure is used to achieve multi-objective output; When constructing the loss function, a multi-objective error function is introduced to constrain the prediction accuracy of multiple target outputs with weighted mean square error, and a time smoothing regularization term is introduced to suppress non-physical jumps. The nonlinear regression function model is trained using the Adam optimizer. The training process is completed under the condition of meeting the convergence criterion, and the model accuracy and generalization ability are evaluated by cross-validation.

5. The method for predicting rock mechanics parameters based on the underground powerhouse of a pumped storage power station according to claim 3, characterized in that, The step of embedding the nonlinear regression function model into the parameter prediction region system, dynamically assigning parameters to the sulfur-bearing shale interlayer region, and constructing a spatiotemporal variation map of parameters evolving with the construction drainage process specifically includes: In the parameter prediction region system, each spatial unit is associated with its spatial coordinates, lithological label, tectonic domain number and interlayer identification status, and a spatial unit index matrix belonging to the sulfur-bearing shale interlayer region is constructed. Extract the construction time step, drainage status, initial rock mass properties and geochemical conditions of each spatial unit in the spatial unit index matrix to form a dataset for inputting the nonlinear regression function model; The environmental control parameters and infrared spectral feature factors of each spatial unit in the dataset at a specific time step are used as inputs and imported into the nonlinear regression function model to output the predicted values ​​of elastic modulus, cohesion and shear strength, thereby realizing the dynamic parameter assignment of the spatial unit. During the assignment of the dynamic parameters, a time filtering algorithm is used to constrain the rate of change of the prediction results of adjacent time steps, and a smoothing process is performed based on the local weighted regression method to ensure the physical continuity of the dynamic parameters in the time series. The predicted elastic modulus, predicted cohesion and predicted shear strength corresponding to each time step are summarized in the form of three-dimensional tensors to construct a spatiotemporal variation map of parameters covering the sulfur-bearing shale interlayer region. The spatiotemporal variation map of the parameters is embedded in the parameter prediction region system to support rock mass response simulation, support structure design evolution calculation and automatic identification and analysis of abnormal trends, so as to realize the dynamic prediction of the rock mass mechanical parameters throughout the whole process and model feedback correction.

6. The method for predicting rock mechanics parameters based on the underground powerhouse of a pumped storage power station according to claim 1, characterized in that, The acquisition of laser point cloud data and microseismic monitoring data, inversion of the actual deformation trend and microfracture density changes of the surrounding rock during construction, and output of actual monitoring results specifically include: By deploying a high-frequency laser scanner and a three-dimensional laser ranging system in the construction area, point cloud data of the construction area at different time periods are collected, and the point cloud data is subjected to attitude registration, noise filtering and mesh reconstruction to form a time series point cloud model of the deformation of the surrounding rock surface. Based on the time series point cloud model, the displacement vector, normal convergence value and profile deformation rate of each monitoring block in the construction area at different time steps are calculated to construct a surrounding rock deformation trend map. By deploying a microseismic monitoring array system around the construction area, microseismic event waveform data is collected, including event trigger time, amplitude, spectral characteristics, and initial motion direction; Based on waveform analysis and travel time inversion methods, the waveform data of the microseismic event are spatially located to construct a three-dimensional spatiotemporal evolution map of the microseismic event; Based on the aforementioned three-dimensional spatiotemporal evolution map, a micro-fracture density distribution map is generated using a fracture cluster identification algorithm; The deformation trend map of the surrounding rock and the microfracture density distribution map are spatiotemporally registered and spatially superimposed to identify structural disturbance areas with coupling characteristics in the surrounding rock mass, and output actual monitoring results including deformation rate field, microfracture density field and structural disturbance level. The actual monitoring results are embedded in the constructed domain coordinate system for deviation comparison with the spatiotemporal variation map of the parameters and for recalibration judgment of the parameter degradation function.

7. The method for predicting rock mechanics parameters based on the underground powerhouse of a pumped storage power station according to claim 1, characterized in that, When the actual monitoring result deviates from the predicted envelope of the spatiotemporal variation map of the parameters, a recalibration process of the parameter degradation function is triggered until the error between the actual monitoring result and the predicted result of the spatiotemporal variation map of the parameters meets the convergence condition, at which point parameter updates are stopped. Specifically, this includes: The deformation rate field and micro-fracture density field in the actual monitoring results are spatially aligned with the spatiotemporal variation map of the parameters in the structural domain coordinate system, and the predicted value and monitored value of each spatial unit at the same time step are extracted. The predicted deviations of the elastic modulus, the cohesion and the shear strength are calculated, and the deviation magnitude matrix is ​​constructed. Determine whether there is a target spatial unit in the deviation magnitude matrix that exceeds the preset error tolerance threshold. If the target spatial unit exists, define the monitoring block where the target spatial unit is located as the target recalibration area. Extract the infrared spectral feature factors and environmental control parameters corresponding to the target recalibration region at the current time step, and construct a local training dataset by combining the actual monitoring values. The transfer learning mechanism is adopted. Based on the original nonlinear regression function model, the local training dataset is retrained or the parameters are fine-tuned while keeping the irrelevant parameters frozen. The weight path related to the prediction bias is adaptively adjusted through the error backpropagation algorithm. The updated nonlinear regression function model is used to re-predict the parameters of the spatial cells in the target recalibration region, and the predicted values ​​at the corresponding positions in the spatiotemporal variation map of the parameters are replaced to obtain the corrected spatiotemporal variation map of the parameters. A transition region is constructed between the target recalibrated region and the uncorrected region. A weighted fusion strategy is used to continuously correct the output of the nonlinear regression function model and the updated nonlinear regression function model, maintaining the spatial smoothness of the parameter field. The corrected spatiotemporal variation map of the parameters is re-embedded into the parameter prediction region system to realize the dynamic evolution and closed-loop update of rock mechanics parameters under construction disturbance conditions.

8. A rock mechanics parameter prediction device based on the underground powerhouse of a pumped storage power station, characterized in that, The device is used to execute a method for predicting rock mechanics parameters based on an underground powerhouse of a pumped storage power station as described in any one of claims 1-7. The device includes an acquisition module (201), a processing module (202), and an output module (203), wherein: The acquisition module (201) is used to acquire the three-dimensional geological structure model and lithological information of the construction area, identify the embedding position and spatial distribution characteristics of sulfur-bearing shale interlayers in the metamorphic sandstone structure, and construct a parameter prediction regional system based on tectonic domain partitioning. The acquisition module (201) is used to acquire the historical evolution trajectory of the elastic modulus, cohesion and shear strength of the sulfur-bearing shale interlayer region, as well as infrared spectral characteristic data. The processing module (202) is used to construct a nonlinear regression function model based on the historical evolution trajectory and the infrared spectral feature data. The nonlinear regression function model takes the infrared spectral feature factors and environmental control parameters as input variables and the predicted evolution trajectory of elastic modulus, cohesion and shear strength as the target output. The processing module (202) is used to embed the nonlinear regression function model into the parameter prediction region system, implement dynamic parameter assignment for the sulfur-bearing shale interlayer region, and construct a parameter spatiotemporal variation map that evolves with the construction drainage process. The acquisition module (201) is used to acquire laser point cloud data and microseismic monitoring data, invert the actual deformation trend and microfracture density change of the surrounding rock during construction, and output the actual monitoring results. The output module (203) is used to trigger the recalibration process of the parameter degradation function when the actual monitoring result deviates from the predicted envelope of the parameter spatiotemporal variation map, until the error between the actual monitoring result and the predicted result of the parameter spatiotemporal variation map meets the convergence condition, and then stop the parameter update.

9. An electronic device, characterized in that, The device includes a processor (301), a communication bus (302), a user interface (303), a network interface (304), and a memory (305). The memory (305) is used to store instructions. The user interface (303) and the network interface (304) are both used to communicate with other devices. The communication bus (302) is used to realize the connection and communication between the components within the electronic device. The processor (301) is used to execute the instructions stored in the memory (305) so that the electronic device performs the method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions that, when executed, perform the method as described in any one of claims 1-7.

Images