A three-dimensional stress field nonlinear intelligent inversion method based on prototype learning

Abstract

The present application relates to the technical field of ground stress field inversion, and discloses a three-dimensional stress field nonlinear intelligent inversion method based on prototype learning, which solves the problems of difficult identification of local stress transmission relationship, insufficient unit boundary traction closure constraint in three-dimensional ground stress field inversion of deep underground engineering, and low inversion accuracy and poor reliability. In the present application, information such as in-situ measuring points, tectonic boundaries, rock mass parameters, and self-weight load is obtained first to form spatial unit state data; then the load and boundary features are extracted, and a stress transmission prototype library is constructed through prototype learning combined with measured constraints; the stress tensor and boundary traction closure amount are solved, and the prototype similarity is corrected using mechanical constraints; the unit stress transmission law is determined based on the similarity weight, the global three-dimensional ground stress field is generated, the principal stress is solved, and the stress concentration area is demarcated. The present application takes into account both measured data and rock mass stress law, and improves the prediction accuracy.

Description

Technical Field

[0001] This invention relates to the field of geostress field inversion technology, specifically to a three-dimensional stress field nonlinear intelligent inversion method based on prototype learning. Background Technology

[0002] Excavation, support, and disaster prevention in deep underground engineering require a thorough understanding of the three-dimensional geostress distribution within the computational domain. Geostress is determined by regional tectonic activity, the weight of the rock mass, its heterogeneity, and the engineering boundaries. Different spatial units transmit normal and shear forces through shared boundaries. In fault zones, caverns, goaf areas, or locations of abrupt parameter changes, the source and direction of local stress are prone to alteration, thus affecting the determination of principal stress directions and stress concentration zones.

[0003] The number of in-situ stress measurement points that can be obtained at the engineering site is limited, and the locations of the measurement points are often discretely distributed. The structural boundary conditions, random fields of rock mechanics parameters, and self-weight loads also need to correspond with the computational grid. Therefore, it is necessary to use the inversion method to unify multi-source information into spatial units and generate a three-dimensional geostress field containing six stress components.

[0004] Existing 3D geostress inversion techniques typically begin by establishing a computational domain grid or finite element model. Measurement point stress, topography, structural boundary conditions, rock mass density, and elastic parameters are input into the model. Then, boundary load adjustments, initial stress field superposition, finite element equilibrium solutions, or statistical interpolation are used to generate the stress field. Some schemes employ multiple regression, Kriging interpolation, genetic algorithms, neural networks, or deep learning models to establish mapping relationships between measurement point stress and features such as spatial coordinates, burial depth, geological zoning, lithological parameters, and boundary loads. The trained model then predicts the stress components in the target area. Other schemes incorporate force balance equations, boundary condition residuals, or smoothing constraints during neural network training to ensure the model output is consistent with existing measurement points and engineering constraints. For discrete spatial elements, a common approach is to first read rock mass parameters and load information based on the element center location, then arrange the stress results according to element adjacency relationships, and finally calculate the magnitude, direction, and potential stress concentration range of the principal stresses through post-processing.

[0005] However, in the aforementioned existing technical solutions, stress inversion is often treated as a global mapping from measuring points to spatial locations, failing to organize local stress sources, rock mass parameter variations, and element boundary transfer relationships as discriminable objects. When boundary conditions and self-weight loads are transferred to adjacent spatial elements through heterogeneous rock masses, the closure relationship between the normal traction, tangential traction, and the resultant traction within the control body on shared boundaries and the self-weight load is usually only checked after the solution is obtained or expressed through the overall residuals, without participating in similarity discrimination before the selection of local transfer relationships. Consequently, in computational domains with sparse measuring points and complex boundary effects, the model lacks a basis for element boundary traction closure when selecting local stress transfer modes, making it difficult to maintain consistency between the transfer interpretation between adjacent elements and the engineering load relationship. Summary of the Invention

[0006] The technical problem to be solved by this invention is to provide a three-dimensional stress field nonlinear intelligent inversion method based on prototype learning, which solves the problems of difficulty in identifying local stress transmission relationships and insufficient traction closure constraints of unit boundaries in the three-dimensional inversion of deep underground engineering stress fields, resulting in low inversion accuracy and poor reliability.

[0007] The technical solution adopted by the present invention to solve the above-mentioned technical problems is as follows:

[0008] A prototype-based nonlinear intelligent inversion method for three-dimensional stress fields includes the following steps:

[0009] S1. Obtain in-situ stress measurement data, structural boundary conditions, random field of rock mechanics parameters, spatial unit boundaries, spatial unit adjacency relationships, and rock mass self-weight load in the deep underground engineering calculation domain to form spatial unit state data;

[0010] S2. Based on the spatial unit state data, establish a prototype library of local stress transfer relationships and obtain prototypes of local stress transfer relationships;

[0011] S3. In the process of similarity calculation between spatial unit and local stress transfer relationship prototype, the local stress transfer relationship prototype is taken as the local stress transfer relationship to be judged, and the stress tensor to be judged is generated according to the local stress transfer relationship to be judged and the spatial unit state data, forming the stress tensor state to be judged;

[0012] S4. Based on the stress tensor state to be determined, the spatial unit boundary, the spatial unit adjacency relationship, and the rock mass self-weight load, calculate the surface traction of the spatial unit control body boundary, and obtain the unit boundary traction closure amount based on the surface traction. Embed the unit boundary traction closure amount into the similarity calculation process to determine the prototype similarity under traction closure constraint.

[0013] S5. Determine the assignment weight of the prototype of the local stress transfer relationship based on the prototype similarity under the traction closure constraint, and form the local stress transfer relationship of the spatial unit;

[0014] S6. Based on the local stress transmission relationship and the spatial unit state data, generate a three-dimensional geostress field containing six stress components;

[0015] S7. Based on the three-dimensional geostress field containing six stress components, output the magnitude of the principal stress, the direction of the principal stress, and the stress concentration area.

[0016] In this scheme, in-situ stress measurement points, structural boundaries, random fields of rock mass mechanical parameters, spatial unit geometry and adjacency relationships, and self-weight loads are uniformly organized into spatial unit state data. Based on prototype learning, local stress transfer patterns are extracted. At the same time, unit boundary traction closure constraints are introduced in advance during the prototype similarity discrimination stage. This makes the inversion process no longer a simple overall mapping from spatial location to stress components, but is based on an interpretable local stress transfer mechanism and strict mechanical equilibrium constraints. This solves the problems of difficulty in identifying local stress transfer, non-closure of traction between adjacent units, and poor physical consistency of stress field caused by sparse measurement points, heterogeneous rock mass, and complex boundary effects in deep underground engineering.

[0017] Furthermore, step S1 specifically includes:

[0018] S11. Obtain the spatial unit boundary of the deep underground engineering calculation domain, determine the spatial unit adjacency relationship based on the shared boundary between adjacent spatial units, and form the spatial unit geometric relationship by the spatial unit boundary and the spatial unit adjacency relationship;

[0019] S12. Based on the geometric relationship of the spatial unit, the in-situ stress measurement point data is mapped to the spatial unit containing the measurement point location to form a spatial unit stress constraint;

[0020] S13. Based on the geometric relationship of the spatial units, the construction boundary conditions are mapped to the boundary spatial units, and the loading direction and magnitude of each spatial unit are determined according to the self-weight load of the rock mass, thus forming the loading relationship of the spatial units;

[0021] S14. Based on the spatial unit geometric relationship, the spatial unit stress constraint, the spatial unit loading relationship, and the random field of rock mass mechanical parameters, form spatial unit state data including in-situ stress measurement point data, structural boundary conditions, random field of rock mass mechanical parameters, spatial unit boundary, spatial unit adjacency relationship, and rock mass self-weight load.

[0022] Furthermore, step S2 specifically includes:

[0023] S21. Extract structural boundary conditions, random field of rock mass mechanical parameters and rock mass self-weight load from the spatial unit state data, and map the structural boundary conditions and rock mass self-weight load to the spatial unit position in the random field of rock mass mechanical parameters to form the load transfer characteristics of the spatial unit.

[0024] S22. Extract the spatial unit boundary and spatial unit adjacency relationship from the spatial unit state data, calculate the boundary normal transmission direction and boundary action area of ​​each spatial unit relative to the adjacent spatial units, and form the spatial unit boundary transmission characteristics;

[0025] S23. Based on the load transfer characteristics of the spatial unit and the boundary transfer characteristics of the spatial unit, construct a spatial unit transfer representation that represents the transfer path of structural boundary conditions and rock mass self-weight load between adjacent spatial units;

[0026] S24. Based on the in-situ stress measurement point data in the spatial unit state data, match the stress constraint of the spatial unit corresponding to the measurement point with the spatial unit transfer representation to form a measurement point constraint transfer representation;

[0027] S25. Based on the measured point constraint transfer representation, perform prototype learning and metric learning on the spatial unit, and form prototype partitioning results according to the consistency of local stress source and boundary transfer direction of the spatial unit;

[0028] S26. Aggregate the corresponding measurement point constraint transfer representations based on the prototype partitioning results, and configure the transfer mapping relationship of the stress tensor to be judged for the aggregation results to form a local stress transfer relationship prototype.

[0029] S27. Establish a local stress transfer relationship prototype library based on the aforementioned local stress transfer relationship prototype, and use the local stress transfer relationship prototype library as input for the similarity calculation process between spatial elements and local stress transfer relationship prototypes.

[0030] Furthermore, in step S25, forming the prototype partitioning result based on the consistency of the local stress source and boundary transmission direction of the spatial unit includes:

[0031] Based on the load transfer characteristics of the spatial unit, the spatial unit positions of the structural boundary conditions and rock mass self-weight load in the random field of rock mass mechanical parameters are determined. Based on the boundary transfer characteristics of the spatial unit, the spatial unit boundary direction corresponding to the transfer of structural boundary conditions and rock mass self-weight load along the adjacency relationship of spatial units is determined. The change direction of the random field of rock mass mechanical parameters, the spatial unit boundary direction, and the measurement point constraint transfer representation are used together as the basis for the division between prototype learning and metric learning.

[0032] Furthermore, step S3 specifically includes:

[0033] S31. Receive the prototype of the local stress transfer relationship and the state data of the spatial unit. In the process of similarity calculation between the spatial unit and the prototype of the local stress transfer relationship, extract the structural boundary conditions, random field of rock mechanics parameters and rock mass self-weight load from the state data of the spatial unit to form the load representation of the spatial unit.

[0034] S32. Based on the spatial unit boundaries and spatial unit adjacency relationships in the spatial unit state data, determine the boundary transfer direction and boundary connection state between spatial units to form a spatial unit boundary transfer representation;

[0035] S33. Based on the load representation of the spatial unit and the boundary transfer representation of the spatial unit, the action position of the prototype of the local stress transfer relationship is matched to form the prototype action representation of the local stress transfer relationship in the spatial unit;

[0036] S34. Based on the prototype action, the prototype of the local stress transfer relationship is defined as the local stress transfer relationship to be determined, which corresponds to the spatial unit local stress transfer relationship formed by the structural boundary conditions and the rock mass self-weight load through the random field action of the rock mass mechanical parameters.

[0037] S35. Based on the local stress transfer relationship to be determined, the load representation of the spatial element, and the boundary transfer representation of the spatial element, generate a stress tensor to be determined that includes normal stress components and shear stress components;

[0038] S36. Based on the stress tensor to be determined and the spatial unit boundary, the normal stress component and shear stress component in the stress tensor to be determined are spatially aligned to form a stress tensor state to be determined for calculating the surface traction of the spatial unit control body boundary.

[0039] Furthermore, in step S35, generating the stress tensor to be determined, which includes normal stress components and shear stress components, includes:

[0040] Based on the local stress transfer relationship to be determined, the structural boundary conditions are matched to the boundary transfer direction in the spatial unit boundary transfer representation, the rock mass self-weight load is matched to the spatial direction in the spatial unit loading representation, and the rock mass mechanical parameters random field is matched to the boundary connection state in the spatial unit boundary transfer representation; and the normal stress component and shear stress component in the stress tensor to be determined are determined according to the boundary transfer direction, spatial direction, and boundary connection state.

[0041] Furthermore, step S4 specifically includes:

[0042] S41. Receive the stress tensor state to be determined, the spatial unit boundary, the spatial unit adjacency relationship and the rock mass self-weight load. In the similarity calculation process between the spatial unit and the prototype of the local stress transfer relationship, determine the surface normal relationship and surface area relationship of the spatial unit control body boundary according to the spatial unit boundary to form the traction calculation boundary state.

[0043] S42. Based on the stress tensor state to be determined and the traction calculation boundary state, calculate the normal traction and tangential traction of each element surface of the spatial element control body boundary to form surface traction;

[0044] S43. Based on the surface traction and the spatial unit adjacency relationship, match the corresponding surface traction of adjacent spatial units on the common boundary to form an adjacent surface traction correspondence relationship;

[0045] S44. Based on the traction correspondence of the adjacent surfaces and the self-weight load of the rock mass, calculate the closure deviation between the traction resultant within the spatial unit control body and the self-weight load of the rock mass, and form the traction closure amount of the unit boundary.

[0046] S45. Embed the unit boundary traction closure amount into the similarity calculation process, and update the constraint basis for the similarity calculation of the prototype of the local stress transmission relationship corresponding to the local stress transmission relationship to be judged, so as to form a similarity calculation object under traction closure constraint.

[0047] S46. Based on the similarity calculation object under the traction closure constraint, the stress tensor state to be determined, and the prototype of the local stress transfer relationship, calculate the matching degree of the local stress transfer relationship to be determined in the spatial unit, and determine the prototype similarity under the traction closure constraint.

[0048] Furthermore, in step S46, when determining the prototype similarity under the traction closure constraint, for each local stress transfer relationship prototype corresponding to the same spatial unit, the corresponding surface traction and unit boundary traction closure amount are calculated according to the stress tensor state to be determined, the spatial unit boundary, the spatial unit adjacency relationship and the rock mass self-weight load. The unit boundary traction closure amount corresponding to each local stress transfer relationship prototype is embedded into the similarity calculation process of the same local stress transfer relationship prototype. The basis for the similarity calculation of the same local stress transfer relationship prototype is updated with constraints, and the prototype similarity under the traction closure constraint is determined before the attribution weight is determined.

[0049] Furthermore, step S5 specifically includes:

[0050] S51. Receive the prototype similarity under the traction closure constraint and the prototype of the local stress transfer relationship, and establish a correspondence between the prototype similarity under the traction closure constraint corresponding to the same spatial unit and the prototype of the local stress transfer relationship respectively.

[0051] S52. Based on the correspondence, normalize the prototype similarity under the traction closure constraint of each local stress transfer relationship prototype to obtain the assignment weight of the local stress transfer relationship prototype.

[0052] S53. Compare the attribution weight with a preset threshold. If the comparison conditions are met, determine the corresponding local stress transfer relationship prototype as the local stress transfer relationship prototype that participates in the formation of the spatial unit.

[0053] S54. Based on the assigned weights, the prototypes of local stress transfer relationships that participate in forming the local stress transfer relationships of the spatial unit are weighted and assigned to form the local stress transfer relationships of the spatial unit.

[0054] Furthermore, step S6 specifically includes:

[0055] S61. Receive the local stress transfer relationship and the spatial unit state data, extract the structural boundary conditions, random field of rock mechanics parameters and rock mass self-weight load from the spatial unit state data, and determine the stress transfer generation relationship of the spatial unit according to the local stress transfer relationship;

[0056] S62. Based on the stress transfer generation relationship, the structural boundary conditions, the random field of rock mass mechanical parameters, and the rock mass self-weight load, generate six stress components of the spatial unit;

[0057] S63. Based on the spatial unit boundary and spatial unit adjacency relationship in the spatial unit state data, the six stress components of adjacent spatial units are matched according to the shared boundary to form a spatial unit stress arrangement relationship;

[0058] S64. Based on the stress arrangement relationship of the spatial unit and the six stress components of the spatial unit, a three-dimensional geostress field containing the six stress components is generated in the deep underground engineering calculation domain.

[0059] Furthermore, step S7 specifically includes:

[0060] S71. Receive the three-dimensional geostress field containing six stress components, and extract six stress components from the three-dimensional geostress field containing six stress components for each spatial unit to form a spatial unit stress tensor.

[0061] S72. Calculate the magnitude and direction of the principal stresses of the spatial element based on the spatial element stress tensor to form the principal stress state of the spatial element;

[0062] S73. Based on the principal stress state of the space unit and the three-dimensional geostress field containing six stress components, calculate the difference in principal stress magnitude and principal stress direction between adjacent space units to form a space unit stress concentration determination quantity;

[0063] S74. Compare the stress concentration determination quantity of the spatial unit with a preset threshold, determine the stress concentration area if the comparison conditions are met, and output the principal stress magnitude, principal stress direction and stress concentration area.

[0064] The beneficial effects of this invention are:

[0065] (1) By unifying and integrating multi-source engineering data, a solid foundation for inversion data is established:

[0066] This invention unifies in-situ stress measurement points, structural boundaries, random fields of rock mechanics parameters, spatial geometric topology, and self-weight load information into spatial units to form standardized state data. This achieves alignment and unification of multiple heterogeneous information in the same spatial dimension, avoids spatial misalignment and data loss caused by step-by-step processing of multi-source information, and ensures spatial consistency of geological, load, and measured constraint information. It provides a complete and accurate data source for subsequent prototype construction and stress inversion.

[0067] (2) Improve the geological adaptability and generalization performance of the scheme by constructing a prototype library through multi-element collaboration:

[0068] This invention establishes a load transfer characterization based on load transfer characteristics and boundary geometry features. It then uses measured stress constraints to divide the model into prototypes. The division process integrates multiple factors, including the direction of rock mass parameter variation, the direction of stress transfer at unit boundaries, and measurement point constraints. The resulting prototypes correspond to differentiated local stress transfer patterns. This prototype system, based on diversified constraints, is adaptable to various complex geological conditions such as abrupt lithological changes, caverns, and faults. Even in areas with sparse measurement point layouts, it can stably characterize the actual stress transmission patterns, effectively improving the generalization ability of the inversion model.

[0069] (3) By matching the relationship between load, parameters and spatial orientation, the physical logic of the stress tensor is guaranteed:

[0070] This invention constructs layered load characterization and boundary transfer characterization, and determines the stress transfer logic to be judged by combining the prototype spatial matching results. When generating the stress tensor, the structural boundary, self-weight load, and rock mass parameters are matched to the corresponding spatial orientations, and the normal and shear stress components are solved item by item based on the orientation information. The generation process of each stress component closely follows the load action form and the spatial distribution characteristics of the rock mass, so that the stress tensor has a clear mechanical meaning. Then, by matching the stress components with the unit boundary direction, a standard stress state is formed, providing reliable intermediate data for accurate calculation of surface traction and closure.

[0071] (4) Based on the traction closure constraint, participate in the prototype similarity evaluation and optimize the stress field force matching effect:

[0072] This invention solves for the boundary surface traction of elements based on standardized stress states, and calculates the traction closure by combining element adjacency relationships and self-weight loads. It introduces closure indices characterizing element force balance and adjacent boundary force continuity into prototype similarity calculations, optimizing prototype matching results using mechanical criteria. A three-dimensional geostress field is generated based on the combined transfer relationships obtained through force constraint screening. This stress field satisfies the mechanical laws of continuous boundary traction and closed element self-weight force, maintaining overall force coordination without additional post-equilibrium corrections.

[0073] (5) Based on similarity-weighted fusion of multiple prototypes, the nonlinear stress transfer characteristics are accurately characterized:

[0074] This invention calculates the attribution weight based on the constrained and optimized prototype similarity, selects effective prototypes, and weights and fuses them to form unit-specific stress transfer relationships. It adapts to geological environments with multiple stress transfer mechanisms through multi-prototype combinations. The proportion of different transfer modes can be adaptively adjusted according to the actual geological conditions of the spatial unit, conforming to the nonlinear stress transfer characteristics under the coupling effects of deep rock mass structure, self-weight, and lithology, effectively reducing principal stress inversion errors.

[0075] (6) Enhance the engineering application value of the results by integrating the calculation of principal stresses and stress concentration zones:

[0076] This invention obtains the magnitude and direction of principal stresses based on the six-component three-dimensional geostress field decomposition, quantifies the difference in principal stresses between adjacent units to construct a stress concentration judgment index, delineates stress concentration areas based on the index, and provides a one-stop output of key parameters required for engineering. The obtained principal stress parameters and high stress concentration area results can be directly used for support design, surrounding rock disaster prediction, and excavation scheme optimization. The stress concentration prediction results have a high degree of matching with the damaged areas of the surrounding rock on site, facilitating on-site risk management of underground engineering. Attached Figure Description

[0077] Figure 1 This is a flowchart of the nonlinear intelligent inversion method for three-dimensional stress field based on prototype learning in this invention.

[0078] Figure 2 This is a diagram illustrating the process of constructing spatial unit state data in this invention.

[0079] Figure 3 This is a diagram illustrating the process of constructing a prototype library for local stress transfer in this invention.

[0080] Figure 4 This is a process diagram of generating the stress tensor state to be determined in this invention.

[0081] Figure 5This is a diagram illustrating the process of determining prototype similarity under traction closure constraints in this invention.

[0082] Figure 6 This is a diagram illustrating the process of determining the local stress transfer relationship of a spatial unit in this invention.

[0083] Figure 7 This is a diagram illustrating the process of generating a three-dimensional geostress field in this invention.

[0084] Figure 8 This is a process diagram of outputting principal stress parameters and stress concentration areas in this invention. Detailed Implementation

[0085] This invention aims to provide a prototype-based nonlinear intelligent inversion method for three-dimensional stress fields, addressing the challenges of identifying local stress transfer relationships and insufficient traction closure constraints at unit boundaries in the three-dimensional inversion of stress fields in deep underground engineering, which leads to low inversion accuracy and poor reliability. The core idea is as follows: Addressing the challenges of stress inversion in deep underground engineering characterized by scattered measuring points, complex structural boundaries, and significant spatial variations in rock mass properties, this invention abandons the traditional single mapping mode that directly relies on spatial coordinate fitting of stress. Instead, it uses the identification of local stress transfer patterns as the core framework, sequentially employing a design approach that integrates multi-source engineering data collection, constructs prototypes through multiple element constraints, optimizes prototype matching using traction closure indices, generates stress fields through prototype weight coupling, and converts stress data into engineering indices, thus forming a closed-loop integrated inversion architecture.

[0086] More specifically, to achieve the above core idea, the present invention employs the following means:

[0087] (1) The six types of heterogeneous engineering data, namely, in-situ stress measurement point data, structural boundary conditions, random field of rock mechanics parameters, spatial unit boundary, unit adjacency relationship, and rock mass self-weight load, are uniformly collected and transformed into spatial unit state data. This breaks the drawbacks of independent and scattered data and inconsistent spatial benchmarks. It integrates discrete measurement points, geological boundaries, rock mass properties, geometric topology, and load information into standardized data of the same calculation object, providing a unified data base for subsequent feature decomposition and prototype learning, and realizing the structured implementation of multi-source engineering information.

[0088] (2) Two key information categories, load transfer characteristics and boundary transfer characteristics, are extracted from the unit state data and matched with measured point data to form a measurement point constraint transfer representation. In the clustering and partitioning stage of prototype learning and metric learning, multiple limiting conditions are introduced simultaneously, including the location of the structural boundary, the direction of the self-weight load, the trend of rock mass mechanical parameters, the normal orientation of the unit boundary, and the measured constraint characterization. Prototypes are divided based on multiple boundaries and physical property constraints, and a unique six-component stress transfer mapping relationship is provided for each type of prototype. Finally, a local stress transfer relationship prototype library is constructed. Each prototype in the library corresponds to a set of local stress transfer forms constrained by boundary loads, rock mass heterogeneity characteristics, and unit adjacency transfer laws. This transforms the originally abstract and scattered engineering stress laws into standardized transfer objects that can be quantified, called, and weighted.

[0089] (3) In the process of matching prototypes with each spatial unit, the transmission mode is no longer selected solely based on the similarity of data features. Instead, a separate stress tensor is generated for each pair of units and prototypes. The surface traction solution is carried out based on the normal and tangential orientation of the unit boundary and the surface area of ​​the unit. Then, combined with the adjacent arrangement of units and the self-weight load of the rock mass, the deviation between the resultant traction force and the self-weight of the control body and the traction difference on both sides of the shared boundary are calculated. The traction closure amount of the unit boundary is then obtained. The above-mentioned closure index, which includes the overall force deviation of the unit and the force difference of the adjacent boundary, is incorporated into the prototype similarity calculation basis. Before the prototype assignment weight calculation, the mechanical constraints are used to correct the matching results. The selection process of local stress transmission relationship is controlled by the actual force law of the rock mass. This avoids the matching distortion problem caused by relying solely on the fitting of measuring points and spatial features and ignoring the real traction transmission relationship between units.

[0090] (4) After the prototype similarity is optimized by the traction closure condition, the corresponding belonging weight of each prototype is obtained by normalization. Valid prototypes are selected according to the weight and combined to form the local stress transfer relationship of the spatial unit. By using the form of multi-prototype collaborative combination, the shortcomings of the single transfer model cannot be adapted to the stress change in complex structure and multi-load superposition area are overcome. The six stress components are solved unit by unit based on the combined transfer relationship. The complete three-dimensional geostress field of the whole domain is formed by splicing the adjacent arrangement of the unit space.

[0091] (5) After obtaining the three-dimensional geostress field of the whole domain, the stress tensor is assembled based on the six-component stress of each unit. The principal stress magnitude and principal stress direction of each spatial unit are calculated by tensor decomposition. Then, the stress mutation degree is quantified by comparing the difference in principal stress values ​​and the difference in the directional angle between adjacent units. Based on this, the stress concentration area is delineated. The closed-loop processing of local transmission identification, mechanical constraint verification, whole domain stress inversion and engineering risk assessment is fully realized. It is highly adaptable to the application scenarios of sparse measuring points, variable structural boundaries and prominent heterogeneity of rock mass parameters in deep underground engineering.

[0092] To facilitate understanding of the technical solution of this invention, some technical terms involved in this invention are explained below:

[0093] Spatial unit state data: This is a data set formed by mapping in-situ stress measurement point data, structural boundary conditions, random field of rock mechanics parameters, spatial unit boundaries, spatial unit adjacency relationships, and rock mass self-weight load to the same spatial unit.

[0094] Geometric relationships of spatial units: These are the spatial unit locations, shared boundaries, and adjacent connections determined by the spatial unit boundaries and adjacency relationships.

[0095] Spatial element stress constraint: The stress constraint formed by mapping the in-situ stress measurement data to a spatial element containing the measurement point location.

[0096] Spatial unit loading relationship: This refers to the relationship between the loading direction and magnitude formed after the structural boundary conditions and the rock mass self-weight load are mapped to the spatial unit.

[0097] Local stress transfer relationship prototype library: a collection consisting of multiple local stress transfer relationship prototypes for use in spatial element similarity calculation.

[0098] Prototype of local stress transfer relationship: A prototype object representing the local stress transfer mode formed between spatial units after the structural boundary conditions and the self-weight load of the rock mass are subjected to the random field of rock mass mechanical parameters.

[0099] Spatial unit load transfer characteristics: These are the characteristics formed by the spatial unit position and direction of action of structural boundary conditions and rock mass self-weight load in the random field of rock mass mechanical parameters.

[0100] Spatial unit boundary transfer characteristics: These are the characteristics formed by the boundary normal transfer direction and boundary action area of ​​a spatial unit relative to its adjacent spatial units.

[0101] Spatial element transfer representation: A representation constructed based on the load transfer characteristics and boundary transfer characteristics of spatial elements, representing the transfer path of boundary conditions and rock mass self-weight load between adjacent spatial elements.

[0102] Measurement point constraint transfer representation: This is the representation formed by matching the stress constraint of the spatial element corresponding to the measurement point with the transfer representation of the spatial element.

[0103] Prototype partitioning results: These are the partitioning results formed after performing prototype learning and metric learning on spatial elements based on the consistency of local stress sources and boundary transmission directions of the spatial elements.

[0104] Transfer mapping relationship: A mapping relationship configured on the aggregated result of the prototype partitioning and used to generate the stress tensor to be judged from the local stress transfer relationship prototype.

[0105] Stress tensor to be determined: This is a stress tensor containing normal stress components and shear stress components generated after the local stress transfer relationship to be determined is applied to the state data of the spatial element.

[0106] The stress tensor state to be determined is the stress state formed after the stress tensor to be determined corresponds to the spatial direction of the spatial element boundary, which is used to calculate the surface traction of the boundary of the spatial element control body.

[0107] Spatial element loading representation: This is a representation of the loading effect on a spatial element formed by extracting structural boundary conditions, random fields of rock mechanics parameters, and rock mass self-weight load from the spatial element state data.

[0108] Spatial cell boundary transfer representation: This is a representation formed after determining the boundary transfer direction and boundary connection state based on the spatial cell boundary and the spatial cell adjacency relationship.

[0109] Application location matching: This refers to the matching process for determining the application location of the prototype of the local stress transfer relationship in the spatial element based on the spatial element load representation and the spatial element boundary transfer representation.

[0110] Prototype action representation: This is the representation formed after matching the action position of the prototype of local stress transfer relationship with the load representation of the spatial element and the boundary transfer representation of the spatial element.

[0111] Spatial orientation correspondence: This refers to the process of mapping the normal stress components and shear stress components in the stress tensor to the normal and tangential directions of the spatial element boundary.

[0112] Surface traction: The traction vector obtained by the action of the stress tensor state to be determined along the surface normal direction of the boundary of the spatial element control volume.

[0113] Unit boundary traction closure quantity: This is a calculated quantity used to represent the degree of traction transfer closure of the spatial unit control body boundary, obtained from surface traction, spatial unit adjacency relationship, and rock mass self-weight load.

[0114] Prototype similarity under traction closure constraint: This is the similarity obtained in the process of embedding the traction closure amount of the element boundary into the similarity calculation of the prototype of the spatial element and the local stress transfer relationship.

[0115] Traction calculation boundary state: the surface normal relationship and surface area relationship of the spatial unit control body boundary determined based on the spatial unit boundary.

[0116] Adjacent surface traction correspondence: This is the relationship formed by corresponding the surfaces of adjacent spatial units on the shared boundary based on the adjacency relationship of the spatial units.

[0117] Traction Resultant: The effect obtained by combining the surface traction of each unit surface at the boundary of the spatial unit control body according to the spatial direction.

[0118] Closure deviation: The difference between the traction resultant and the rock mass self-weight load when no closure is formed.

[0119] Similarity calculation object under traction closure constraint: The calculation object is formed after embedding the traction closure amount of the unit boundary into the similarity calculation process and updating the constraint basis of the similarity calculation.

[0120] The similarity calculation basis serves as the input for calculating the degree of matching between spatial units and prototypes of local stress transfer relationships.

[0121] Matching degree: The degree of similarity between the local stress transfer relationship to be judged and the spatial element calculated in the similarity calculation object under traction closure constraint.

[0122] Assignment weight: This is the weight determined based on the prototype similarity under traction closure constraints, used to indicate whether a spatial element belongs to the corresponding local stress transfer relationship prototype.

[0123] Local stress transfer relationship of spatial unit: The corresponding transfer relationship of spatial unit is formed by combining the prototype of local stress transfer relationship according to the assignment weight.

[0124] Stress transfer generation relationship: This is the relationship between the six stress components of a spatial element, determined based on the local stress transfer relationship, and used to generate the element from the structural boundary conditions, the random field of rock mechanics parameters, and the rock mass self-weight load.

[0125] Stress arrangement relationship of spatial elements: This is the arrangement relationship formed by corresponding the six stress components of adjacent spatial elements according to the shared boundary based on the spatial element boundary and the spatial element adjacency relationship.

[0126] Principal stress state of a spatial element: This refers to the magnitude and direction of the principal stresses of the spatial element, calculated based on the spatial element stress tensor.

[0127] Spatial element stress concentration determination quantity: This quantity is used to determine stress concentration areas based on the differences in the magnitude and direction of principal stresses between adjacent spatial elements.

[0128] In practical implementation, the implementation process of the three-dimensional stress field nonlinear intelligent inversion method based on prototype learning provided by this invention is as follows: Figure 1 This includes the following steps:

[0129] S1. Obtain multi-source engineering data and construct spatial unit status data:

[0130] In this step, in-situ stress measurement data, structural boundary conditions, random field of rock mechanics parameters, spatial unit boundaries, spatial unit adjacency relationships, and rock mass self-weight load are obtained in the deep underground engineering calculation domain to form spatial unit state data.

[0131] In one exemplary implementation, see Figure 2 This step includes the following sub-steps:

[0132] S11. Obtain the spatial unit boundary of the deep underground engineering calculation domain, determine the spatial unit adjacency relationship based on the shared boundary between adjacent spatial units, and form the spatial unit geometric relationship by the spatial unit boundary and the spatial unit adjacency relationship.

[0133] S12. Based on the geometric relationship of the spatial unit, the in-situ stress measurement point data is mapped to the spatial unit containing the measurement point location to form a spatial unit stress constraint.

[0134] S13. Based on the geometric relationship of the spatial units, the construction boundary conditions are mapped to the boundary spatial units, and the loading direction and magnitude of each spatial unit are determined according to the self-weight load of the rock mass, thus forming the loading relationship of the spatial units.

[0135] S14. Based on the spatial unit geometric relationship, the spatial unit stress constraint, the spatial unit loading relationship, and the random field of rock mass mechanical parameters, form spatial unit state data including in-situ stress measurement point data, structural boundary conditions, random field of rock mass mechanical parameters, spatial unit boundary, spatial unit adjacency relationship, and rock mass self-weight load.

[0136] This step first establishes an overall spatial geometric benchmark by analyzing the adjacent topology based on the unit boundaries. This benchmark serves as a unified carrier for all subsequent data, preventing misalignment of different data types due to inconsistent spatial benchmarks. Then, based on the established geometric relationships, the measured point constraints are anchored to the corresponding units, and the structural boundaries and self-weight loads are matched to their corresponding spatial locations, ensuring accurate implementation of measured constraints and boundary loads within the geometric framework. Finally, the data is uniformly incorporated into the random field of rock mechanics parameters, completing the aggregation and encapsulation of multi-source heterogeneous data within the same spatial unit. This processing method overcomes the problem of independent and inconsistent representational dimensions of various original data, ensuring that rock mass properties, boundary constraints, external loads, measured stresses, and spatial topology are all coupled through a unified geometric system. This guarantees consistent spatial location matching of various information sources from the data source, providing a standardized and unified basic data source for subsequent calculations.

[0137] S2. Based on the spatial element state data, construct a local stress transfer prototype library:

[0138] In this step, a prototype library of local stress transfer relationships is established based on the spatial unit state data to obtain the prototype of local stress transfer relationships.

[0139] In one exemplary implementation, see Figure 3 This step includes the following sub-steps:

[0140] S21. Extract structural boundary conditions, rock mass mechanical parameters random field, and rock mass self-weight load from the spatial unit state data, and map the structural boundary conditions and rock mass self-weight load to the spatial unit position in the rock mass mechanical parameters random field to form the load transfer characteristics of the spatial unit.

[0141] S22. Extract the spatial unit boundary and spatial unit adjacency relationship from the spatial unit state data, calculate the boundary normal transmission direction and boundary action area of ​​each spatial unit relative to the adjacent spatial units, and form the spatial unit boundary transmission characteristics.

[0142] S23. Based on the load transfer characteristics of the spatial unit and the boundary transfer characteristics of the spatial unit, construct a spatial unit transfer representation that represents the transfer path of structural boundary conditions and rock mass self-weight load between adjacent spatial units.

[0143] S24. Based on the in-situ stress measurement point data in the spatial unit state data, match the stress constraint of the spatial unit corresponding to the measurement point with the spatial unit transfer representation to form a measurement point constraint transfer representation.

[0144] S25. Based on the measured point constraint transfer representation, perform prototype learning and metric learning on the spatial unit, and form prototype partitioning results according to the consistency of the local stress source and boundary transfer direction of the spatial unit.

[0145] S26. Aggregate the corresponding measurement point constraint transfer representations based on the prototype partitioning results, and configure the transfer mapping relationship of the stress tensor to be judged for the aggregation results to form a local stress transfer relationship prototype.

[0146] S27. Establish a local stress transfer relationship prototype library based on the aforementioned local stress transfer relationship prototype, and use the local stress transfer relationship prototype library as input for the similarity calculation process between spatial elements and local stress transfer relationship prototypes.

[0147] Specifically, when performing prototype learning and metric learning on spatial units based on the measured point constraint transfer representation to form prototype partitioning results, the spatial unit positions in the random field of rock mass mechanical parameters are determined according to the load transfer characteristics of the spatial units. The spatial unit boundary directions corresponding to the transfer of structural boundary conditions and rock mass self-weight load along the spatial unit adjacency relationship are determined according to the boundary transfer characteristics of the spatial units. The change direction of the random field of rock mass mechanical parameters, the spatial unit boundary direction, and the measured point constraint transfer representation are used together as the partitioning basis for prototype learning and metric learning, so that the prototype of the local stress transfer relationship corresponding to the same prototype partitioning result is simultaneously limited by the random field of rock mass mechanical parameters, structural boundary conditions, and spatial unit adjacency relationship.

[0148] This step extracts two core features—load and boundary—from standardized unit state data, and completes the entire process layer by layer: constructing load transfer characteristics, coupling measured constraints, collaborative clustering of multi-dimensional indicators, and encapsulating prototypes into a library. First, load transfer characteristics are constructed based on load information and rock mass parameter distribution. Boundary transfer characteristics are obtained by solving for boundary orientation and area of ​​action based on unit geometry and topology. These two types of features together build a standardized representation of the load transfer path across units. Then, in-situ measuring point stress constraints are introduced and bound to the corresponding transfer characteristics, anchoring the true value benchmark of the transfer law based on measured data. In the prototype clustering stage, classification is no longer solely based on measuring point constraints. The direction of change in rock mass mechanical parameters and the boundary direction corresponding to load transfer are simultaneously combined as the basis for division, allowing the clustered prototypes to be naturally controlled by multiple geological conditions including rock mass properties, structural loads, and unit topology. After grouping and aggregation, a dedicated stress mapping rule is provided, enabling each prototype to directly calculate stress components. Finally, a complete prototype library is compiled and provided for subsequent similarity matching.

[0149] This step, on the one hand, relies on the constraints of measured points to control the overall accuracy of the prototype, and on the constraints of parameters and boundary orientation to control the geological adaptability of the prototype. Even in areas with sparse measuring points, variable lithology, and complex structural boundaries, it can still effectively summarize the real stress laws. On the other hand, the prototype is pre-configured with stress generation mapping relationships, which saves the step of repeatedly building the calculation logic in the subsequent stress generation stage. It can directly participate in the solution of the stress tensor to be judged, effectively improving the calculation efficiency and reliability of subsequent prototype matching and stress calculation.

[0150] S3. Transform the prototype relation to generate the stress tensor state to be determined:

[0151] In this step, during the similarity calculation between the spatial unit and the prototype of the local stress transfer relationship, the prototype of the local stress transfer relationship is used as the local stress transfer relationship to be determined, and the stress tensor to be determined is generated based on the local stress transfer relationship to be determined and the state data of the spatial unit, thus forming the stress tensor state to be determined.

[0152] In one exemplary implementation, see Figure 4 This step includes the following sub-steps:

[0153] S31. Receive the prototype of the local stress transfer relationship and the state data of the spatial unit. During the similarity calculation process between the spatial unit and the prototype of the local stress transfer relationship, extract the structural boundary conditions, the random field of rock mechanics parameters and the self-weight load of the rock mass from the state data of the spatial unit to form the load representation of the spatial unit.

[0154] S32. Based on the spatial unit boundaries and spatial unit adjacency relationships in the spatial unit state data, determine the boundary transfer direction and boundary connection state between spatial units to form a spatial unit boundary transfer representation.

[0155] S33. Based on the load representation of the spatial unit and the boundary transfer representation of the spatial unit, the action position of the prototype of the local stress transfer relationship is matched to form the prototype action representation of the local stress transfer relationship prototype in the spatial unit.

[0156] S34. Based on the prototype action representation, the prototype of the local stress transfer relationship is defined as the local stress transfer relationship to be determined, which corresponds to the spatial unit local stress transfer relationship formed by the structural boundary conditions and the rock mass self-weight load through the random field action of the rock mass mechanical parameters.

[0157] S35. Based on the local stress transfer relationship to be determined, the load representation of the spatial element, and the boundary transfer representation of the spatial element, generate a stress tensor to be determined that includes normal stress components and shear stress components.

[0158] S36. Based on the stress tensor to be determined and the spatial unit boundary, the normal stress component and shear stress component in the stress tensor to be determined are spatially aligned to form a stress tensor state to be determined for calculating the surface traction of the spatial unit control body boundary.

[0159] Specifically, when generating the stress tensor to be determined, the boundary conditions are mapped to the boundary transfer direction in the spatial unit boundary transfer representation according to the local stress transfer relationship to be determined, the rock mass self-weight load is mapped to the spatial direction in the spatial unit loading representation, and the random field of rock mass mechanical parameters is mapped to the boundary connection state in the spatial unit boundary transfer representation. The normal stress component and shear stress component in the stress tensor to be determined are determined according to the boundary transfer direction, spatial direction, and boundary connection state, respectively, so that the stress tensor to be determined is generated by the effect of the local stress transfer relationship to be determined on the spatial unit boundary direction.

[0160] This step relies on an existing prototype library and element state data to sequentially build load representations and boundary transfer representations. Spatial location matching is used to lock the actual operational range of the prototype on the element, completing the transformation from a general prototype to a transfer relationship adapted to this element. During the construction of the stress tensor to be determined, the boundary conditions, rock mass self-weight load, and random field of rock mass mechanical parameters are anchored to the boundary transfer direction, spatial force direction, and boundary connection state, respectively. Based on these three types of spatial information, the normal stress and shear stress components are determined, ensuring that the values ​​of each stress component closely follow the load orientation and the spatial distribution characteristics of the rock mass. Finally, the spatial orientation of the stress components is calibrated using the element boundary, resulting in a standardized stress tensor state that can be directly used for surface traction calculations.

[0161] This step binds three types of constraints—load, physical properties, and spatial geometry—from the source of stress structure, giving the stress tensor to be determined a clear physical meaning. This achieves precise adaptation between the general prototype and the geological conditions of individual spatial units, and ensures that subsequent calculations of surface traction and traction closure are based on stress data that closely matches actual stress patterns. This input-level assurance guarantees the accuracy of the traction closure calculation in the next stage, laying a reliable data foundation for optimizing prototype similarity using mechanical constraints.

[0162] S4. Calculate the boundary traction closure quantity and introduce constraints to obtain the prototype similarity:

[0163] In this step, the surface traction of the spatial unit control body boundary is calculated based on the stress tensor state to be determined, the spatial unit boundary, the spatial unit adjacency relationship, and the rock mass self-weight load. The unit boundary traction closure amount is obtained based on the surface traction, and the unit boundary traction closure amount is embedded in the similarity calculation process to determine the prototype similarity under traction closure constraint.

[0164] In one exemplary implementation, see Figure 5 This step includes the following sub-steps:

[0165] S41. Receive the stress tensor state to be determined, the spatial unit boundary, the spatial unit adjacency relationship and the rock mass self-weight load. In the similarity calculation process between the spatial unit and the prototype of the local stress transfer relationship, determine the surface normal relationship and surface area relationship of the spatial unit control body boundary according to the spatial unit boundary to form the traction calculation boundary state.

[0166] S42. Based on the stress tensor state to be determined and the traction calculation boundary state, calculate the normal traction and tangential traction of each element surface of the spatial element control body boundary to form surface traction;

[0167] S43. Based on the surface traction and the spatial unit adjacency relationship, match the corresponding surface traction of adjacent spatial units on the common boundary to form an adjacent surface traction correspondence relationship;

[0168] S44. Based on the traction correspondence of the adjacent surfaces and the self-weight load of the rock mass, calculate the closure deviation between the traction resultant within the spatial unit control body and the self-weight load of the rock mass, and form the traction closure amount of the unit boundary.

[0169] S45. Embed the unit boundary traction closure amount into the similarity calculation process, and update the constraint basis for the similarity calculation of the prototype of the local stress transmission relationship corresponding to the local stress transmission relationship to be judged, so as to form a similarity calculation object under traction closure constraint.

[0170] S46. Based on the similarity calculation object under the traction closure constraint, the stress tensor state to be determined, and the prototype of the local stress transfer relationship, calculate the matching degree of the local stress transfer relationship to be determined in the spatial unit, and determine the prototype similarity under the traction closure constraint.

[0171] Specifically, when determining the prototype similarity under traction closure constraints, for each local stress transfer relationship prototype corresponding to the same spatial unit, the corresponding surface traction and unit boundary traction closure amount are calculated based on the stress tensor state to be determined, the spatial unit boundary, the spatial unit adjacency relationship, and the rock mass self-weight load. The unit boundary traction closure amount corresponding to each local stress transfer relationship prototype is embedded into the similarity calculation process of the same local stress transfer relationship prototype. The basis for similarity calculation of the same local stress transfer relationship prototype is updated with constraints, and the prototype similarity under traction closure constraints is determined before the attribution weight is determined.

[0172] This step first relies on the surface normal and area information extracted from the unit boundary to form the boundary calculation benchmark. Then, it combines the stress tensor to solve the normal and tangential traction of each boundary surface item by item. Next, it matches the corresponding traction on both sides of the shared boundary based on the unit adjacency relationship. Combined with the unit's self-weight, it calculates the deviation between the boundary traction resultant force and the self-weight load, and quantifies the traction closure quantity that can reflect the overall force coordination of the unit. Unlike the method of relying solely on spatial features to carry out similarity evaluation, this scheme calculates the closure index separately for the pairing result of each group of units and prototypes before assigning weights. It uses the closure deviation to synchronously correct the evaluation criteria of prototype similarity, realizing the pre-intervention of mechanical constraints in the prototype screening process.

[0173] This step effectively avoids prototype selection bias caused by simply relying on the fitting of measurement points and spatial features by transforming the laws of physical mechanics into quantitative constraints for similarity scoring. It ensures that the prototype matching results take into account both data fitting effects and the objective stress laws of the rock mass. The similarity score, corrected for closure, objectively reflects the degree of fit between the stress transmission mode corresponding to the prototype and the actual stress state of the unit. This provides a precise quantitative evaluation basis for the subsequent reasonable allocation of prototype combination weights, reducing the problem of incoordination of shared boundary forces between adjacent units from the source of prototype selection.

[0174] S5. Based on prototype similarity, determine the local stress transfer relationship of spatial elements using a weighted approach:

[0175] In this step, the assignment weight of the prototype of the local stress transmission relationship is determined according to the prototype similarity under the traction closure constraint, and the local stress transmission relationship of the spatial unit is formed.

[0176] In one exemplary implementation, see Figure 6 This step includes the following sub-steps:

[0177] S51. Receive the prototype similarity under the traction closure constraint and the prototype of the local stress transfer relationship, and establish a correspondence between the prototype similarity under the traction closure constraint corresponding to the same spatial unit and the prototype of the local stress transfer relationship respectively.

[0178] S52. Based on the correspondence, normalize the prototype similarity under the traction closure constraint of each local stress transfer relationship prototype to obtain the assignment weight of the local stress transfer relationship prototype.

[0179] S53. Compare the attribution weight with a preset threshold. If the comparison conditions are met, determine the corresponding local stress transfer relationship prototype as the local stress transfer relationship prototype that participates in the formation of the spatial unit.

[0180] S54. Based on the assigned weights, the prototypes of local stress transfer relationships that participate in forming the local stress transfer relationships of the spatial unit are weighted and assigned to form the local stress transfer relationships of the spatial unit.

[0181] This step uses the prototype similarity optimized by traction closure constraint as the quantitative basis to establish a one-to-one correspondence between prototypes and matching scores. The weight of each prototype is obtained through similarity normalization and conversion. Then, effective prototypes are screened based on preset thresholds. Finally, the unit-specific local stress transfer relationship is obtained by fusion according to the weight ratio.

[0182] This weighted combination construction method can adaptively adjust the proportion of different transmission laws according to the tectonic structure, lithology, and load environment of the spatial unit, adapting to the nonlinear stress transmission characteristics of deep rock masses caused by tectonic activity, self-weight load, and heterogeneous lithological coupling. The local transmission relationships obtained through screening and weighting not only take into account the fitting effect of prototype and measured data, but also follow the force law reflected by the previous traction closure, which can better fit the actual force transmission characteristics of the unit, laying a reliable calculation basis for the subsequent accurate generation of the six-component three-dimensional geostress field.

[0183] S6. Based on the local stress transfer relationship, a three-dimensional geostress field is generated:

[0184] In this step, a three-dimensional geostress field containing six stress components is generated based on the local stress transmission relationship and the spatial unit state data.

[0185] In one exemplary implementation, see Figure 7 This step includes the following sub-steps:

[0186] S61. Receive the local stress transfer relationship and the spatial unit state data, extract the structural boundary conditions, random field of rock mechanics parameters and rock mass self-weight load from the spatial unit state data, and determine the stress transfer generation relationship of the spatial unit according to the local stress transfer relationship;

[0187] S62. Based on the stress transfer generation relationship, the structural boundary conditions, the random field of rock mass mechanical parameters, and the rock mass self-weight load, generate six stress components of the spatial unit;

[0188] S63. Based on the spatial unit boundary and spatial unit adjacency relationship in the spatial unit state data, the six stress components of adjacent spatial units are matched according to the shared boundary to form a spatial unit stress arrangement relationship;

[0189] S64. Based on the stress arrangement relationship of the spatial unit and the six stress components of the spatial unit, a three-dimensional geostress field containing the six stress components is generated in the deep underground engineering calculation domain.

[0190] This step first extracts structural boundaries, rock mass parameters, and self-weight loads from the unit state data, and locks the stress generation rules specific to this unit based on the established local stress transfer relationships. Then, based on the generation rules, all six stress components of a single spatial unit are solved item by item to achieve accurate stress assignment at the unit scale. Next, using the unit boundary and adjacency arrangement relationship, the stress components at the shared boundary of adjacent units are connected and matched to standardize the stress arrangement logic of the entire domain. Finally, all unit stress data are integrated in the entire computational domain and stitched together to obtain the global six-component three-dimensional geostress field.

[0191] This step relies on the transfer relationship obtained by multi-prototype fusion to solve for stress, which can adapt to complex working conditions with multiple stress mechanisms superimposed in different regions. It effectively improves the shortcomings of a single calculation model that cannot take into account the coupling effects of structure, self-weight, and lithological anomalies. At the same time, by using spatial adjacency relationships to constrain the stress distribution of adjacent units, it ensures that the stress in the whole domain transitions smoothly at the geometric connection position, improves the spatial continuity and numerical rationality of the overall stress field, and provides complete and continuous basic stress data for subsequent decomposition of principal stresses and identification of stress concentration areas.

[0192] S7. Based on the three-dimensional geostress field, output the principal stress parameters and stress concentration areas:

[0193] In this step, based on the three-dimensional geostress field containing six stress components, the magnitude of the principal stress, the direction of the principal stress, and the stress concentration area are output.

[0194] In one exemplary implementation, see Figure 8 This step includes the following sub-steps:

[0195] S71. Receive the three-dimensional geostress field containing six stress components, and extract six stress components from the three-dimensional geostress field containing six stress components for each spatial unit to form a spatial unit stress tensor.

[0196] S72. Calculate the magnitude and direction of the principal stresses of the spatial element based on the spatial element stress tensor to form the principal stress state of the spatial element;

[0197] S73. Based on the principal stress state of the space unit and the three-dimensional geostress field containing six stress components, calculate the difference in principal stress magnitude and principal stress direction between adjacent space units to form a space unit stress concentration determination quantity;

[0198] S74. Compare the stress concentration determination quantity of the spatial unit with a preset threshold, determine the stress concentration area if the comparison conditions are met, and output the principal stress magnitude, principal stress direction and stress concentration area.

[0199] This step first extracts the six stress components of each spatial unit from the stress field and assembles them into a stress tensor. The magnitude and direction of the principal stresses for each unit are obtained through tensor eigenvalue decomposition. Then, the numerical difference and directional deviation of the principal stresses of adjacent units are compared to quantify and generate stress concentration judgment quantities. Finally, based on the preset judgment threshold, regions exceeding the range of abnormal changes are screened, the stress concentration range is delineated, and the principal stress parameters and stress concentration partitioning results required for the project are output.

[0200] Example:

[0201] Taking a deep underground tunnel project as an application example, the intelligent inversion method for three-dimensional geostress field based on prototype learning in this invention is used for calculation. The implementation process is as follows:

[0202] S1. Obtain multi-source engineering data and construct spatial unit status data:

[0203] In this embodiment, the process of constructing spatial unit state data is as follows:

[0204] The computational domain for deep underground engineering is denoted as... The set of spatial units obtained by dividing the computational domain of deep underground engineering is denoted as... ,in, Indicates the first Each spatial unit Indicates the total number of spatial units. For each spatial unit Obtain the boundary of the spatial unit ,in, Representing spatial units The Each unit surface Each unit face Includes surface normal direction Surface area and boundary location range normal direction Used to determine the direction of boundary action, surface area Used to determine the area of ​​boundary action and the range of boundary location. Used to determine the shared boundary between adjacent spatial units.

[0205] Based on the boundary location range of different spatial units Perform shared boundary determination: Determine the spatial unit unit surface With spatial units unit surface Perform location matching, where, , and Different When the boundary location range and boundary location range The spatial coincidence error meets the preset geometric tolerance, and the surface normal direction and the normal direction of the surface When the directions are opposite, the spatial unit With spatial units Determine them as adjacent spatial units, and then define the unit surfaces. With unit surface It is determined to be a shared boundary. When the element surface... When there is no shared boundary with any spatial cell surface, the cell surface will be... The spatial unit to which it belongs It is identified as a boundary space unit. The adjacency relationship of the space units is formed based on the shared boundary determination results. ,in, Representing a unit surface The adjacency state of the unit face; When there is a shared boundary, Write the adjacency tag and bind the adjacent spatial cell index and the corresponding cell face index; when the cell face... When there is no shared boundary, Write boundary markers. These are defined by the spatial unit boundaries. Spatial unit adjacency relationship The spatial unit's geometric relationship is formed by the control volume area enclosed by the spatial unit's boundary. Geometric relationships of spatial units It includes the location of spatial units, shared boundaries, and adjacent connection relationships.

[0206] In-situ stress measurement data are denoted as ,in, Indicates the first Location of each measuring point Indicates the first The six stress components at each measuring point location This indicates the number of in-situ stress measurement points. The number of data points is determined based on the location of the measurement points. Geometric relationship with spatial units The spatial unit locations are determined to determine inclusion relationships, and the in-situ stress measurement point data is written into the spatial unit containing the measurement point location. When the same measurement point location is contained by multiple spatial units, the spatial unit with the earliest index is selected according to the preset spatial unit index order. When the same spatial unit contains multiple measurement point locations, the in-situ stress measurement point data with the smallest distance is selected in ascending order of distance from the measurement point location to the center of the spatial unit; if the distances are the same, the in-situ stress measurement point data with the earliest number is selected in ascending order of measurement point number. For spatial units containing measurement point locations, an in-situ stress measurement point data field is formed. ,in, This indicates that the measurement point data is marked. For spatial elements that do not contain measurement point locations, the six stress components are written with preset empty values, and the measurement point data presence marker is written with a "no measurement point" marker. This is derived from the in-situ stress measurement point data field. Spatial element stress constraints are formed, which are the stress limits of the measuring points formed after the in-situ stress measuring point data are mapped to the spatial element containing the measuring point location.

[0207] Construct boundary conditions denoted as ,in, Indicates the first The boundary locations corresponding to the constructed boundary conditions. Indicates the first The six stress components corresponding to the structural boundary conditions This indicates the number of boundary conditions to be constructed. The boundary locations are... Boundary location range of the boundary space unit Perform matching at the boundary positions Falling into the boundary position range In the case of matching boundary conditions, the boundary conditions are written to the corresponding boundary space element. For space elements without matching boundary conditions, the boundary condition field is modified. Write a preset empty value. The rock mass self-weight load field is denoted as... ,in, , and These represent the rock mass self-weight load components in three spatial directions. (Based on the rock mass self-weight load field...) The spatial direction and sign of the non-empty load components determine the load direction of each spatial element, based on the rock mass self-weight load field. The magnitude of the load component determines the load magnitude of each spatial element. The boundary condition fields are then constructed. Rock mass self-weight load field When these conditions are mapped to the same spatial unit, a spatial unit loading relationship is formed. This spatial unit loading relationship is the relationship between the load direction and the load magnitude formed after the structural boundary conditions and the rock mass self-weight load are mapped to the spatial unit.

[0208] The random field of rock mass mechanics parameters is written into the spatial cell according to the spatial cell location. The random field field of rock mass mechanics parameters is denoted as... ,in, , and These represent the mean values ​​of elastic modulus, Poisson's ratio, and density, respectively. , and Let represent the variances of the elastic modulus, Poisson's ratio, and density, respectively. , and These represent the relevant lengths in the three spatial directions, respectively. This represents the gradient magnitude of the spatial unit parameters. If the random field of rock mechanics parameters is already given according to the spatial unit index, then the random field of rock mechanics parameters is written into the corresponding spatial unit according to the spatial unit index; if the random field of rock mechanics parameters is given according to the spatial location sampling points, then for each spatial unit center... Select the sampling point with the smallest distance and write the random field field of the rock mechanics parameters corresponding to the sampling point into the spatial cell; if the distances are the same, select the sampling point with the earliest number in ascending order of sampling point number.

[0209] For each spatial unit The in-situ stress measurement point data field Construct boundary condition fields Random field of rock mass mechanics parameters Spatial unit boundary field Spatial cell adjacency relationship field and rock mass self-weight load field Spatial unit status data is generated by binding spatial unit indexes. Among them, the spatial unit boundary field It consists of the normal directions and surface areas of six unit faces, with a dimension of twenty-four; spatial unit adjacency relationship field. It consists of the adjacency states of six unit surfaces, with a dimension of six; in-situ stress measurement point data field Construct boundary condition fields with seven dimensions. The dimension is six-dimensional, and the random field field of rock mass mechanics parameters is used. The dimension is ten-dimensional, and the rock mass self-weight load field is also included. The dimension is three-dimensional. Therefore, the state data of each spatial unit... It is a 56-dimensional vector. Spatial unit state data is generated by mapping in-situ stress measurement data, structural boundary conditions, random fields of rock mechanics parameters, spatial unit boundaries, spatial unit adjacency relationships, and rock mass self-weight load to the same spatial unit. The input layer receives vector sequences according to the spatial unit index. The loaded state portion read by the spatial unit state coding layer is The dimension is 26, and the boundary propagation part read by the spatial unit state encoding layer is... With a dimension of thirty, the process of establishing a prototype library of local stress transfer relationships can read loading information, rock mass heterogeneity information, and spatial adjacency transfer information from the state data of the same spatial unit.

[0210] S2. Based on the spatial element state data, construct a local stress transfer prototype library:

[0211] In this embodiment, the process of establishing the prototype library of local stress transfer relationships is as follows:

[0212] spatial unit state data For input, where, Indicates the first Each spatial unit Indicates the total number of spatial units. Status data for each spatial unit , This represents a field containing data from seven-dimensional in-situ stress measurement points. This indicates the six-dimensional construction boundary condition field. This represents the random field field of ten-dimensional rock mass mechanics parameters. Represents the boundary field of a 24-dimensional spatial unit. This field represents the adjacency relationship of six-dimensional spatial units. This field represents the self-weight load of the three-dimensional rock mass.

[0213] From spatial unit status data Extract the construction boundary condition field Random field of rock mass mechanics parameters and rock mass self-weight load field Construct boundary condition fields Includes six stress components, rock mass self-weight load field Includes self-weight load components in three spatial directions. Rock mass mechanics parameter random field. ,in, , , These represent the mean values ​​of elastic modulus, Poisson's ratio, and density, respectively. , , Let represent the variances of the elastic modulus, Poisson's ratio, and density, respectively. , , These represent the relevant lengths in the three spatial directions, respectively. This represents the gradient magnitude of the spatial unit parameters. It will then be used to construct the boundary condition field. and rock mass self-weight load field According to spatial unit index Corresponding to the random field field of rock mass mechanics parameters The location of the spatial unit.

[0214] For spatial units The six element faces are marked with the positions of the boundary conditions. ,in, Representing spatial units The Does each element surface bear the structural boundary conditions? For spatial units The six unit faces are respectively marked with the direction of random field variation of rock mass mechanical parameters. When the spatial cell adjacency field Indicates the first When a unit surface has adjacent spatial units, the random field field of rock mass mechanical parameters of the adjacent spatial units is read, and the differences of the mean elastic modulus, mean Poisson's ratio, and mean density are compared with the preset parameter change thresholds. Comparison; any difference in any item reaches the preset parameter change threshold. At that time, Write change flag, but the preset parameter change threshold has not been reached. At that time, Write the unchanged flag. , , , and Formation of load transfer characteristics of spatial units The load transfer characteristics of spatial units represent the position and direction of action of spatial units under structural boundary conditions and rock mass self-weight load in a random field of rock mass mechanical parameters.

[0215] From spatial unit status data Extracting spatial unit boundary fields Adjacency field of spatial units Spatial unit boundary field It is composed of the normal directions and surface areas of six unit faces, denoted as ,in, Indicates the first The normal direction of each unit surface. Indicates the first The surface area of ​​each unit face. Spatial unit adjacency field. ,in, Indicates the first The adjacency status of each unit face. As a spatial unit The direction of transmission of the boundary normal relative to the adjacent spatial unit will As the boundary area, As a boundary connection state, it forms the spatial unit boundary transfer characteristic. The boundary transfer characteristics of a spatial unit represent the boundary normal transfer direction and boundary area of ​​a spatial unit relative to its adjacent spatial units.

[0216] Based on the load transfer characteristics of spatial units and spatial unit boundary transfer characteristics Constructing spatial unit transfer representation The spatial unit state coding layer first reads the loaded state portion. Loaded state part It is a 26-dimensional vector. The first fully connected layer contains 96 neurons, each neuron and the loaded state part The system is fully connected, with each neuron including weights, biases, and linear rectified units. The first fully connected layer outputs a 96-dimensional intermediate representation. The second fully connected layer receives the 96-dimensional intermediate representation and the boundary propagation part. Boundary transfer part The first layer is a 30-dimensional vector. The second fully connected layer contains 96 neurons and outputs a 96-dimensional boundary-coupled representation. The metric representation layer contains 64 neurons and maps the 96-dimensional boundary-coupled representation to a spatial unit transitive representation. Spatial unit transmission representation It is a 64-dimensional vector used to represent the transmission path of structural boundary conditions and rock mass self-weight load between adjacent spatial units.

[0217] From spatial unit status data Read in-situ stress measurement point data fields ,in, This indicates that the measurement point data is marked. Marking the measurement point data... Six stress components and spatial element transfer representation Binding, forming a transitive representation of measurement point constraints When the measurement point data has markers When there is no in-situ stress measurement data, the six stress components are written to preset null values, which do not participate in the stress constraint error update.

[0218] Prototype learning and metric learning use measurement point constraint transitive representation Marking the direction of random field variation of rock mass mechanical parameters Marking the location of boundary conditions and spatial unit boundary transfer characteristics For input. Local stress transfer relationship prototype layer settings. A prototype of local stress transfer relationship Each local stress transfer relationship prototype contains a 64-dimensional prototype representation. The source of local stress in a spatial unit is jointly determined by the location of the tectonic boundary conditions, the direction of the rock mass's self-weight load, and the direction of the random field variation of the rock mass's mechanical parameters. For each spatial unit... Within a candidate range where the local stress sources of spatial elements are consistent and the boundary normal transmission directions are consistent, the spatial element transfer representation is... Represented by each of the sixty-four-dimensional prototypes Compare component differences along the same dimension, and write the prototype index of the local stress transfer relationship with the smallest sum of squared component differences into the prototype partitioning results. Prototype partitioning results Representing spatial units The corresponding local stress transfer relationship prototype index.

[0219] Based on prototype division results By aggregating the measurement point constraint transfer representations corresponding to the same local stress transfer relationship prototype, a form is formed. For each By averaging the values ​​of the same dimension in the 64-dimensional spatial unit representation, a 64-dimensional prototype representation is obtained. The measurement point data is marked. This represents the transfer of constraints based on in-situ stress measurement data, using six stress components as stress constraint inputs for a local stress transfer relationship prototype. A transfer mapping relationship is configured for each local stress transfer relationship prototype. ,in, Represents the weights and biases of the transitive mapping. Transitive mapping. The input is a 64-dimensional spatial unit transfer representation. With 64-dimensional prototype representation The 128-dimensional vector conveys the mapping relationship. It contains 48 hidden neurons and 6 output neurons, with the 6 output neurons corresponding to the 6 stress components of the stress tensor to be discriminated. The transfer mapping relationship is adopted using the 6 stress components in the measurement point constraint transfer representation. The weights and biases are updated via backpropagation. The training error is the sum of squared differences in each dimension between the six output stress components and the six stress components at the measurement points. The training stops when the preset number of training rounds is reached. Or the training error is not greater than a preset error threshold. .Depend on A prototype representation containing sixty-four dimensions and transitive mapping relationship The prototypes of local stress transfer relationships are used to form a prototype library of local stress transfer relationships, which serves as the input for the similarity calculation process between spatial elements and the prototypes of local stress transfer relationships.

[0220] S3. Transform the prototype relation to generate the stress tensor state to be determined:

[0221] In this embodiment, the process of generating the stress tensor state to be determined is as follows:

[0222] spatial unit state data Prototype of local stress transfer relationship For input, Indicates the first Each spatial unit Indicates the total number of spatial units. This indicates the number of prototypes representing local stress transfer relationships. Prototype of each local stress transfer relationship Including 64-dimensional prototype representation and transitive mapping relationship , This represents the weights and biases that propagate the mapping relationship.

[0223] From spatial unit status data Extract the construction boundary condition field Random field of rock mass mechanics parameters and rock mass self-weight load field This forms a spatial unit load representation. Spatial unit load representation This is a 19-dimensional vector, comprising six-dimensional structural boundary conditions, a 10-dimensional random field of rock mass mechanics parameters, and a 3-dimensional rock mass self-weight load. The structural boundary conditions represent the six stress components borne by the boundary spatial element. The random field of rock mass mechanics parameters represents the mean and variance of the elastic modulus, Poisson's ratio, and density, the correlation lengths in the three spatial directions, and the gradient magnitude of the spatial element parameters. The rock mass self-weight load represents the self-weight load components in the three spatial directions.

[0224] From spatial unit status data Extracting spatial unit boundary fields Adjacency field of spatial units This forms a spatial unit boundary transfer representation. Spatial cell boundary field , Representing spatial units The The normal direction of a unit surface of an element surface. Indicates the first The surface area of ​​each unit surface. Spatial cell adjacency field , Indicates the first Boundary connection status of each unit surface; Indicates the first Each unit surface has adjacent spatial units. Indicates the first Each element surface is the boundary surface of the computational domain. Spatial element boundary transfer representation. It is a 30-dimensional vector used to represent the boundary transfer direction and boundary connection state between spatial units.

[0225] The spatial unit state coding layer is based on the loaded state part and spatial unit boundary transfer representation Generate a 64-dimensional spatial unit transfer representation , This represents the seven-dimensional in-situ stress measurement data field. The spatial unit state coding layer includes a first fully connected layer, a second fully connected layer, and a metric representation layer. The first fully connected layer receives the twenty-six-dimensional loading state portion. The first layer outputs a 96-dimensional intermediate representation; the second fully connected layer receives the 96-dimensional intermediate representation and the 30-dimensional spatial unit boundary transfer representation. The output is a 96-dimensional boundary coupling representation; the metric representation layer maps the 96-dimensional boundary coupling representation to a 64-dimensional spatial unit transitive representation. .

[0226] For each spatial unit Prototype of each local stress transfer relationship Perform action location matching. Action location matching is represented by spatial element loading. Spatial unit boundary transfer representation 64-dimensional spatial unit transfer representation and 64-dimensional prototype representation Based on this, the boundary condition field will be constructed. For the element surface whose boundary connection state is the boundary surface of the computational domain, the rock mass self-weight load field is... This corresponds to an angle with the rock mass self-weight load direction that is not greater than a preset direction threshold. The unit surface normal direction will be used to determine the random field field of rock mass mechanical parameters. This corresponds to the boundary connection state being adjacent spatial cells and the gradient magnitude of the spatial cell parameters not being less than the preset parameter gradient threshold. For element surfaces that satisfy at least one of the following correspondences: structural boundary condition correspondence, rock mass self-weight load correspondence, or rock mass mechanical parameter random field correspondence, the action position matching result of the corresponding element surface is written into... For element surfaces that do not satisfy any of the correspondence relationships, the matching result of the corresponding element surface's position is written into... When the same spatial unit The matching results of the action positions of the six element faces are all At that time, the result of matching the action position of the element surface with the smallest angle to the direction of the rock mass self-weight load is written into the system. If multiple unit faces have the same included angle, then the position matching result of the unit face with the largest surface area is written into the system. If multiple element faces have the same included angle and surface area, then the matching result of the element face with the smallest element face index is written into the function position. The six-dimensional action position matching result is formed from the action position matching results of the six unit surfaces. , Indicates the first Each element surface represents a prototype of the local stress transfer relationship. In spatial units The position of function in Indicates the first The individual element surface is not considered as a prototype for local stress transfer relationships. In spatial units The position of its function within. According to... , and Forming a prototype represents The prototype action represents the position of the prototype in the spatial element, indicating the local stress transfer relationship.

[0227] Based on the prototype's function The prototype of the local stress transfer relationship Limited to the local stress transfer relationship to be determined The local stress transfer relationship to be determined. This indicates that after the structural boundary conditions and the self-weight load of the rock mass are subjected to the random field of rock mass mechanical parameters, the spatial unit... The transmission relationship of local stress is formed within the structure. The generation layer of the stress tensor to be determined is represented by a 64-dimensional spatial unit. and 64-dimensional prototype representation The 128-dimensional vector is used as the input. The stress tensor generation layer to be discriminated consists of 48 hidden neurons and 6 output neurons. The 48 hidden neurons are fully connected to the 128-dimensional input. Each hidden neuron includes weights, biases, and linear rectifier units. The 6 output neurons are fully connected to the 48 hidden neurons.

[0228] Transmit mapping relationship The stress tensor to be determined is output through the generation layer of the stress tensor to be determined. The first three components are normal stress components, and the last three components are shear stress components. During the input process, the boundary condition field is constructed. This has already been mapped to the boundary transfer direction in the spatial unit boundary transfer representation, and the rock mass self-weight load field. The spatial direction has been mapped to the spatial unit load representation, and the random field field of the rock mass mechanics parameters is now used. The boundary connection state, which has been mapped to the boundary transfer representation of the spatial unit, makes the stress tensor to be determined... Based on the local stress transfer relationship to be determined The effect is generated in the direction of the spatial unit boundary.

[0229] Based on the stress tensor to be determined and spatial unit boundary field The spatial orientation of the normal stress component and shear stress component in the stress tensor to be determined is mapped. The normal stress component is then classified according to... , , The three spatial directions correspond to the unit surface normal directions of each element surface. The shear stress components are arranged according to , , The shear plane corresponds to the tangential direction of each element surface. The tangential direction is derived from the normal direction of the unit surface. The orthogonality with the global coordinate axes is determined; when the normal direction of the unit surface is parallel to one global coordinate axis, the other two global coordinate axes are used as tangential directions; when the normal direction of the unit surface is not parallel to any global coordinate axis, the global coordinate axis with the smallest absolute inner product with the normal direction of the unit surface is selected as an auxiliary axis. The auxiliary axis is cross-multiplied with the normal direction of the unit surface and normalized to obtain the first unit tangential direction, and the normal direction of the unit surface is cross-multiplied with the first unit tangential direction and normalized to obtain the second unit tangential direction. In order to write the effect of the stress tensor to be discriminated on the spatial element boundary direction into the state of the stress tensor to be discriminated, the state record of the stress tensor to be discriminated facing the normal direction and the tangential direction is formed for each element surface:

[0230] ;

[0231] In the formula, Representing spatial units Relative to the prototype of local stress transfer relationship In the Records of the state of the undiscriminated stress tensor on each element surface; Indicates the spatial unit index; Indicates the prototype index of local stress transfer relationship; Indicates the cell face index; Indicates the first The normal direction of each unit surface; Indicates transpose; Represents the stress tensor to be determined The three-dimensional stress matrix, obtained by arranging the stress tensors in a symmetric manner, has its main diagonal lines sequentially written into... , and The non-principal diagonals of the three-dimensional stress matrix are written in symmetrical positions. , and ; Indicates the first The first unit tangential direction of each unit surface; Indicates the first The second unit tangential direction of each unit surface; Indicates the first The surface area of ​​each unit surface; Indicates the first Boundary connection status of each unit surface; Indicates the first Matching results of the effective positions of each unit surface.

[0232] The undiscriminated stress tensor states of the six element faces are recorded and written into the undiscriminated stress tensor state according to the element face index order. The stress tensor state to be determined Binding the stress tensor to be determined Unit surface normal direction First unit tangential direction Second unit tangential direction Surface area Boundary connection status and the results of the action location matching It is used to calculate the surface traction of the control volume boundary of the space unit.

[0233] S4. Calculate the boundary traction closure quantity and introduce constraints to obtain the prototype similarity:

[0234] In this embodiment, the similarity calculation process for the embedded unit boundary traction closure amount is as follows:

[0235] To determine the stress tensor state Spatial unit boundary field Adjacency Index Table in Spatial Unit Adjacency Relationship Rock mass self-weight load field 64-dimensional spatial unit transfer representation Prototype of local stress transfer relationship and the stress tensor to be determined For input, Indicates the first Each spatial unit , Indicates the total number of spatial units. , This indicates the number of prototypes for local stress transfer relationships. (Prototypes of local stress transfer relationships) Including 64-dimensional prototype representation The stress tensor state to be determined This includes the stress tensor to be determined, the unit surface normal direction of the six element faces, the first unit tangential direction, the second unit tangential direction, the surface area, the boundary connection status, and the matching result of the applied position. Rock mass self-weight load field. This represents the resultant force of the three-dimensional self-weight acting on the control volume of the spatial unit. , and These represent the self-weight load components in three spatial directions.

[0236] Based on spatial unit boundary field Determine the surface normal and surface area relationships of the boundary of the spatial unit control volume: Representing spatial units The The normal direction of a unit surface of an element surface. Indicates the first The surface area of ​​each unit surface. From the stress tensor state to be determined Read the first The first unit tangential direction of each unit surface Second unit tangential direction Boundary connection status Matching results with the location of action . Indicates the first Each unit surface has adjacent spatial units. Indicates the first Each element surface serves as the boundary surface of the computational domain; Indicates the first Each element surface represents a prototype of the local stress transfer relationship. In spatial units The position of function in Indicates the first The individual element surface is not considered as a prototype for local stress transfer relationships. In spatial units The position of action in the calculation is determined by the matching results of the normal direction of the unit surface, the first unit tangential direction, the second unit tangential direction, the surface area, the boundary connection state, and the position of action. The traction calculation boundary state is used to define the normal direction, tangential direction, surface area, and boundary connection state of each unit surface on the boundary of the spatial unit control volume.

[0237] Based on the stress tensor state to be determined And the traction calculation boundary state, calculate the surface traction of each element surface of the control volume boundary of the spatial element: for the first Each element surface, from the stress tensor state to be determined. Read the normal stress state along the unit surface normal direction, the first tangential stress state along the first unit tangential direction, and the second tangential stress state along the second unit tangential direction. Multiply the normal stress state by the surface area. Back along the unit surface normal direction Placement, multiplying the first tangential stress state by the surface area. Following the first unit tangential direction Placement, multiplying the second tangential stress state by the surface area. Following the second unit tangential direction Place it, then synthesize the force vectors in the three directions to form the first... Surface traction of each unit surface Surface traction This is a three-dimensional force vector, used to represent the traction vector obtained by the application of the stress tensor state to be determined along the surface normal and tangential directions of the boundary of the spatial element control volume. The surface traction of the six element surfaces is arranged in the order of the element surface indices to form an eighteen-dimensional surface traction representation. .

[0238] Adjacency Index Table in Spatial Unit Adjacency Relationship Determined by spatial cell boundaries and spatial cell adjacency relationships: Adjacency Index Table Records in Representing spatial units The Individual surface and spatial unit The Each element surface forms a shared boundary. Prototype of local stress transfer relationship. First, complete the surface traction calculation for all spatial units, then proceed according to the adjacency index table. Read space unit Surface traction and adjacent spatial units Surface traction Furthermore, the surfaces on the shared boundary are bound together to form an adjacent surface traction correspondence. The adjacent surface traction correspondence is used to represent the surface traction correspondence between adjacent spatial units on the shared boundary.

[0239] Based on the correspondence between surface traction and adjacent surface traction, and the rock mass self-weight load field The traction closure of the unit boundary is calculated. The traction closure of the unit boundary is represented by a six-dimensional vector. The first three dimensions represent the closure deviation between the resultant traction within the spatial unit control body and the rock mass self-weight load. The last three dimensions represent the traction difference between adjacent faces on the shared boundary. Face traction uses the external normal of the spatial unit control body as the positive direction, and the rock mass self-weight load field... The direction acting within the control volume of the spatial unit is taken as the positive direction. To ensure that the traction difference between adjacent faces is simultaneously constrained by the consistency of the shared boundary area and the matching result of the action position, a dimensionless weight determined by the surface area and the matching result of the action position is introduced into each adjacency index record, and the unit boundary traction closure quantity is formed as follows:

[0240] ;

[0241] In the formula, Representing spatial units Relative to the prototype of local stress transfer relationship The unit boundary traction closure amount; Indicates the spatial unit index; Indicates the prototype index of local stress transfer relationship; Representing spatial units cell surface index; Representation and spatial unit Adjacent spatial unit index; Representing spatial units cell surface index; Representing spatial units In the Prototype of local stress transfer relationship on each element surface Surface traction; Representing spatial units In the Prototype of local stress transfer relationship on each element surface Surface traction; Representing spatial units The three-dimensional rock mass self-weight load field; Representing spatial units The adjacency index table; Represents the adjacency index table The number of records in; Indicates in and Take the maximum value from the middle. Representing spatial units The Prototype of local stress transfer relationship for each element face The result of the position matching; Representing spatial units The Prototype of local stress transfer relationship for each element face The result of the position matching; Representing spatial units The The surface area of ​​each unit surface; Representing spatial units The The surface area of ​​each unit surface; Indicates in and The smaller value is taken; square brackets indicate that two three-dimensional force vectors are concatenated into a six-dimensional vector. Surface area weights are dimensionless values, application location matching weights are dimensionless values, and both surface traction and rock mass self-weight load fields are force vectors. The six components in the element boundary traction closure quantity retain the dimensions of the force. At that time, the unit boundary traction closure amount The three-dimensional vector is written into the zero vector.

[0242] Traction closure amount of unit boundary In the similarity calculation process between embedded spatial elements and the prototype of local stress transfer relationships, when the boundary traction closure of non-embedded elements is considered, the similarity calculation is based on the 64-dimensional spatial element transfer representation. 64-dimensional prototype representation and the six-dimensional undiscriminable stress tensor The closure amount of the six-dimensional element boundary pull. Added to the similarity calculation criteria, forming a similarity calculation object under traction closure constraints. ,in, This represents the identifier of the object used for similarity calculation. The object used for similarity calculation under traction closure constraints. It is a 140-dimensional vector, including a 64-dimensional spatial unit transfer representation, a 64-dimensional prototype representation, a 6-dimensional undiscriminated stress tensor, and a 6-dimensional unit boundary traction closure quantity.

[0243] The similarity calculation layer receives similarity calculation objects under traction closure constraints. The similarity calculation layer consists of 48 hidden neurons and one similarity output neuron. The 48 hidden neurons are fully connected to the similarity calculation objects under 140-dimensional traction closure constraints. Each hidden neuron includes weights, biases, and linear rectified units. The similarity output neuron is fully connected to the 48 hidden neurons, outputting a one-dimensional matching degree. Match degree This indicates the local stress transfer relationship to be determined in the spatial element. Prototype of the relationship between local stress transfer The degree of similarity. For the same spatial unit Prototype of each local stress transfer relationship Perform surface traction calculation, element boundary traction closure calculation, and similarity calculation respectively to obtain the prototype similarity under traction closure constraints. And before the attribution weights are determined, the prototype similarity calculation under the traction closure constraint is completed.

[0244] S5. Based on prototype similarity, determine the local transitive relationships of units using a weighted approach:

[0245] In this embodiment, the formation process of the local stress transfer relationship of the spatial unit is as follows:

[0246] Prototype similarity under traction closure constraints Prototype of local stress transfer relationship For input, Indicates the first Each spatial unit , Indicates the total number of spatial units. , This indicates the number of prototypes for local stress transfer relationships. (Prototypes of local stress transfer relationships) Including 64-dimensional prototype representation and transitive mapping relationship , Represents the weights and biases of the transitive mapping relationship. Prototype similarity under traction closure constraints. This indicates the local stress transfer relationship to be determined in the spatial element. Prototype of local stress transfer relationship The degree of matching.

[0247] For the same spatial unit According to the prototype index of local stress transfer relationship The similarity scores of the twelve prototypes under traction closure constraints are read in ascending order to form a twelve-dimensional similarity sequence. Simultaneously, twelve local stress transfer relationship prototypes are read in the same local stress transfer relationship prototype index order to form a local stress transfer relationship prototype sequence. The prototype similarity under each traction closure constraint is calculated. Prototype of local stress transfer relationship with the same index Binding, forming a corresponding relationship Correspondence The prototype similarity used to limit the traction closure constraint corresponding to the same spatial unit only applies to the local stress transfer relationship prototypes with the same index, and does not replace the local stress transfer relationship prototypes across indices.

[0248] The attribution weight layer receives a twelve-dimensional similarity sequence. The attribution weight layer contains twelve neurons, each associated with a twelve-dimensional similarity sequence. A fully connected read relationship is established, with each neuron corresponding to a prototype index of a local stress transfer relationship. Weight layers belong to the same spatial unit. Internal 12-dimensional similarity sequence After normalization, a twelve-dimensional attribution weight sequence is obtained. To ensure that the attribution weights simultaneously preserve the relative levels of prototype similarity under traction closure constraints and the differences in similarity distribution within the same spatial unit, the attribution weights... Determine as follows:

[0249] ;

[0250] In the formula, Representing spatial units Belongs to the prototype of local stress transfer relationship The attribution weight; Indicates the spatial unit index; Indicates the prototype index of the current local stress transfer relationship; The prototype index of the local stress transfer relationship traversed during normalized summation; This represents the prototype index of the local stress transfer relationship traversed when searching for the maximum or minimum prototype similarity under traction closure constraints within the same spatial unit. Indicates the number of prototypes representing local stress transfer relationships; Representing spatial units Corresponding local stress transfer relationship prototype Prototype similarity under traction closure constraints; Representing spatial units Corresponding local stress transfer relationship prototype index Prototype similarity under traction closure constraints; Representing spatial units Corresponding local stress transfer relationship prototype index Prototype similarity under traction closure constraints; This represents the minimum value among the prototype similarities under twelve traction closure constraints within the same spatial unit; This represents the maximum value among the twelve prototype similarities under traction closure constraints within the same spatial unit. This represents a preset positive smoothing constant. The value is dimensionless. This indicates that the difference between the prototype similarity and the minimum value under the traction closure constraint is limited to a non-negative number; the summation symbol indicates that the difference between the prototype similarity and the minimum value is limited to a non-negative number. to The prototype indices of the twelve local stress transfer relationships are summed item by item. The prototype similarity under traction closure constraint is a dimensionless value, the preset positive smoothing constant is a dimensionless value, and the assignment weight is a dimensionless value.

[0251] Set preset threshold Preset threshold A dimensionless value greater than zero and less than one. Assign weights. With preset threshold A comparison is made. The comparison condition is the attribution weight. Not less than the preset threshold Under the condition of satisfying the comparison criteria, the prototype of the local stress transfer relationship is then established. Write space unit Candidate prototype index sequence When all twelve attribution weights are less than the preset threshold... In this case, the prototype of the local stress transfer relationship with the highest attribution weight is written into the candidate prototype index sequence. If multiple attribution weights simultaneously reach their maximum values, the local stress transfer relationship prototype with the smallest local stress transfer relationship prototype index is written into the list. Candidate Prototype Index Sequence The prototype of the local stress transfer relationship is used to participate in the formation of the local stress transfer relationship of the spatial unit.

[0252] Based on the candidate prototype index sequence Assign weights to the attribution weights. First, sort the candidate prototype index sequence. The assignment weights corresponding to the prototypes of local stress transfer relationships are added together to obtain the total assignment weight; then the candidate prototype index sequence is used. The assignment weight is obtained by dividing the assignment weight corresponding to each local stress transfer relationship prototype by the sum of the selected assignment weights. Assign weights The prototype of the local stress transfer relationship that participates in the formation of the spatial unit is represented in the spatial unit. The proportion within. For the candidate prototype index sequence. The local stress transfer relationship prototype outside of this model will have its assigned weights written to zero.

[0253] Candidate prototype index sequence The 64-dimensional prototype representation of each local stress transfer relationship prototype. Transmitting mapping relationships and weight allocation Spatial units are formed by binding the prototype index according to the local stress transfer relationship. Local stress transfer relationship Local stress transfer relationship of spatial elements The transfer mapping relationship of the prototype of local stress transfer relationship is preserved, and the assigned weights are written in, so that the spatial element The local stress transfer relationship can be combined into multiple prototype local stress transfer relationships according to the attribution weight. The attribution weight layer performs the same processing on each spatial unit one by one, forming local stress transfer relationships arranged by spatial unit index.

[0254] S6. Based on the local stress transfer relationship, a three-dimensional geostress field is generated:

[0255] In this embodiment, the three-dimensional geostress field generation process is as follows:

[0256] Local stress transfer relationship of spatial unit and spatial unit state data For input, Indicates the first Each spatial unit , This represents the total number of spatial elements. The local stress transfer relationship within the spatial elements. Including candidate prototype index sequences Assigning weights 64-dimensional prototype representation and transitive mapping relationship . Indicates participation in the formation of spatial units The prototype index sequence of local stress transfer relationships. Indicates the prototype index of local stress transfer relationship; Prototype representing local stress transfer relationship In spatial units The allocation weights in; Prototype representing local stress transfer relationship The 64-dimensional prototype representation; Represents the prototype of local stress transfer relationship Generate the transfer mapping relationship of the six stress components; This indicates the weights and biases in the transitive mapping relationship.

[0257] From spatial unit status data Extract in-situ stress measurement point data fields Construct boundary condition fields Random field of rock mass mechanics parameters Spatial unit boundary field Spatial cell adjacency relationship field and rock mass self-weight load field In-situ stress measurement point data fields It is a seven-dimensional vector, with the first six components representing the six stress components at the measuring point. This indicates that the measured point data is marked. Construct the boundary condition field. This is a six-dimensional vector used to represent the six stress components at the boundary of the computational domain. (Random field of rock mechanics parameters) It is a ten-dimensional vector. , and These represent the mean values ​​of elastic modulus, Poisson's ratio, and density, respectively. , and Let represent the variances of the elastic modulus, Poisson's ratio, and density, respectively. , and These represent the relevant lengths in the three spatial directions, respectively. This indicates the gradient magnitude of spatial element parameters. Rock mass self-weight load field. It is a three-dimensional vector, and the three components represent the self-weight load components in three spatial directions.

[0258] Spatial cell boundary field It is a 24-dimensional field. Representing spatial units The The normal direction of a unit surface of an element surface. Indicates the first The surface area of ​​each unit surface. Spatial cell adjacency field It is a six-dimensional vector. Indicates the first The boundary connection state of each unit surface. The in-situ stress measurement data field, the structural boundary condition field, the rock mass mechanical parameter random field, and the rock mass self-weight load field are combined to form a 26-dimensional loading state section. The spatial cell boundary field and the spatial cell adjacency relationship field form a 30-dimensional boundary transfer part. .

[0259] Based on the local stress transfer relationship of the spatial unit Determine the stress transfer generation relationship of the spatial element. Stress transfer generation relationship Candidate prototype index sequence The assigned weights for each candidate prototype index 64-dimensional prototype representation Transmitting mapping relationships 26-dimensional loading state part and the 30-dimensional boundary transfer part Bonding is formed. Stress transfer relationship is generated. Used to define the construction boundary condition field Random field of rock mass mechanics parameters and rock mass self-weight load field Entering the transitive mapping relationship The input location is determined, and the weighting of the local stress transfer relationship prototype is defined.

[0260] A 64-dimensional spatial unit transfer representation is generated using a spatial unit state coding layer. The first fully connected layer of the spatial unit state coding layer receives the 26-dimensional loaded state portion. The first layer outputs a 96-dimensional intermediate representation; the second fully connected layer receives the 96-dimensional intermediate representation and the 30-dimensional boundary transfer portion. The output layer is a 96-dimensional boundary-coupled representation; the metric representation layer receives the 96-dimensional boundary-coupled representation and outputs a 64-dimensional spatial unit transfer representation. 64-dimensional spatial unit transfer representation It also includes the coding results of structural boundary conditions, random fields of rock mass mechanical parameters, rock mass self-weight load, spatial unit boundaries, and spatial unit adjacency relationships.

[0261] Candidate prototype index sequence Prototype of local stress transfer relationship Transmitting 64-dimensional spatial units With 64-dimensional prototype representation The 128-dimensional input is generated. The mapping relationship is then passed on. Receives 128-dimensional generated input. Transmits mapping relationships. It consists of 48 hidden neurons and 6 output neurons. The 48 hidden neurons are fully connected to the 128-dimensional generator input. Each hidden neuron includes weights, biases, and linear rectifier units. The 6 output neurons are fully connected to the 48 hidden neurons. The 6 output neurons sequentially generate candidate stress components. The first three components are normal stress components, and the last three components are shear stress components.

[0262] Candidate prototype index sequence Candidate stress components and assigned weights corresponding to each local stress transfer relationship prototype. Binding is performed according to the prototype index of the same local stress transfer relationship. Weights are assigned to the six stress component locations, and candidate stress components at the same stress component location are then bound according to the assigned weights. Component-by-component weighted synthesis yields spatial units. Six-dimensional stress components The prototype of the local stress transfer relationship with zero weight allocation does not participate in the generation of six-dimensional stress components.

[0263] Based on spatial unit boundary field Adjacency field of spatial units Create an adjacency index table Adjacency Index Table Records in Representing spatial units The Individual surface and spatial unit The Each unit surface forms a shared boundary. Indicates the index of adjacent spatial units. According to the adjacency index table , spatial units Six-dimensional stress components With adjacent spatial units Six-dimensional stress components Based on the shared boundary, a spatial element stress arrangement relationship is formed. Stress arrangement relationship of spatial elements Record the unified arrangement order of spatial element index, element surface index, adjacent spatial element index, adjacent element surface index, shared boundary marker, and six stress components.

[0264] Based on the spatial location of the spatial units within the calculation domain of deep underground engineering, the spatial units are... Six-dimensional stress components Writing three-dimensional geostress field In the middle and spatial units Corresponding location. Three-dimensional geostress field. Six stress components are stored at each spatial unit location, and the order of the six stress components is fixed. , , , , and For adjacent spatial elements with a shared boundary, the spatial element stress arrangement relationship is adopted. Maintain consistency in stress component indices on both sides of the shared boundary. Perform six-dimensional stress component generation and spatial location writing for each spatial unit to form a three-dimensional geostress field containing six stress components within the deep underground engineering calculation domain. .

[0265] S7. Based on the three-dimensional geostress field, output the principal stress parameters and stress concentration areas:

[0266] In this embodiment, the process for determining the stress concentration area is as follows:

[0267] Three-dimensional geostress field containing six stress components For input, This represents the stress field arranged according to spatial unit positions within the calculation domain of deep underground engineering. Three-dimensional geostress field. The spatial element index, six stress components, and common boundary correspondences are stored at each spatial element location. For each spatial element... , , Represents the total number of spatial units, from the three-dimensional geostress field Reading six-dimensional stress components . , and Representing spatial units Normal stress components in the three coordinate directions, , and Representing spatial units Shear stress components on the three shear planes.

[0268] The geostress output layer comprises a stress tensor assembly unit, a principal stress decomposition unit, an adjacency difference calculation unit, and a threshold determination unit. The stress tensor assembly unit receives six-dimensional stress components. Forming the spatial unit stress tensor Spatial element stress tensor It is a 3x3 symmetric matrix, with the main diagonals written sequentially. , and The non-main diagonal of the first Line 1 Column and number Line 1 Column write , No. Line 1 Column and number Line 1 Column write , No. Line 1 Column and number Line 1 Column write Spatial element stress tensor It consists directly of six stress components without introducing additional feature fields.

[0269] Principal stress decomposition element to spatial element stress tensor Perform eigenvalue decomposition on a 3x3 symmetric matrix to obtain three eigenvalues ​​and three unit eigenvectors. The three eigenvalues ​​represent the magnitudes of the three principal stresses, and the three unit eigenvectors represent the directions of the three principal stresses. Arrange the three principal stress magnitudes in descending order of value to form the three-dimensional principal stress magnitudes. , Representing spatial units The The magnitude of the principal stress, Arrange the unit eigenvectors corresponding to the magnitudes of the three principal stresses according to the same principal stress index to form a nine-dimensional principal stress direction. , Representing spatial units The One principal stress direction, , and They represent the first The unit vector components of each principal stress direction in the three coordinate directions. To avoid duplicate records due to sign reversal of the unit eigenvector for the same principal stress direction, the first non-zero coordinate component of the unit eigenvector is written as a positive value; if the first coordinate component is zero, the second coordinate component is checked; if the second coordinate component is zero, the third coordinate component is checked.

[0270] When the stress tensor of the space element When principal stresses of the same magnitude exist, the principal stress decomposition element adopts a preset stress equality threshold. Make a judgment. The stress dimension is greater than zero. The absolute value of the difference between the magnitudes of the two principal stresses is not greater than... When the principal stresses are equal, their magnitudes are determined to be the same. For the principal stress directions corresponding to the same principal stress magnitude, a global approach is first used. The projection direction of the axis onto the corresponding feature subspace is taken as the first unit direction; if global If the axis projection length is zero, then a global approach is used. The projection direction of the axis into the corresponding characteristic subspace is taken as the first unit direction; the remaining unit directions are determined by directions orthogonal to the already determined unit directions. This process ensures that the principal stress directions still have a definite arrangement under the condition of the same principal stress magnitude.

[0271] Adjacent difference calculation unit is based on three-dimensional geostress field The adjacency index table is determined by the correspondence between the spatial unit location and the shared boundary. Adjacency Index Table Records in Representing spatial units The Individual surface and spatial unit The Each unit surface forms a shared boundary. , Indicates the index of adjacent spatial units. For each adjacent index record Read space unit The magnitude of the three-dimensional principal stress and the nine principal stress directions and according to the same principal stress number Compare them.

[0272] For the same principal stress number ,Will and The absolute value of the difference is written as the difference in the magnitude of the principal stresses. .Will and Take the absolute value of the inner product, and then apply an inverse cosine function to the absolute value of the inner product to obtain the difference in principal stress directions. Difference in principal stress directions The value ranges from zero to a right angle, and is used to eliminate the influence of the choice of positive and negative directions of the unit eigenvector on directional differences. For the same adjacent spatial unit... The maximum value among the differences in the three principal stresses is selected, and spatial elements are used. and spatial units The maximum absolute value of the six principal stresses plus the preset stress smoothing amount As the denominator, we obtain the difference in the magnitude of the dimensionless principal stresses; Represents a stress dimensionless constant greater than zero. For the same adjacent spatial unit... The maximum value among the differences in the three principal stress directions is taken, and the radian value corresponding to the right angle is used as the denominator to obtain the dimensionless principal stress direction difference. The larger value between the difference in the magnitude of the dimensionless principal stress and the difference in the direction of the dimensionless principal stress is written as the candidate quantity of adjacent stress concentration.

[0273] For spatial units The maximum value is selected from all adjacent stress concentration candidates to form the stress concentration determination quantity of the spatial element. . Representing spatial units A comprehensive criterion for determining the magnitude and direction of principal stress changes relative to adjacent spatial elements. (When the adjacency index table...) When there is no record, the stress concentration determination quantity of the spatial element will be used. Write it as zero.

[0274] Threshold determination unit sets preset threshold , It is a dimensionless value greater than zero. The comparison condition is... Not less than Under the condition that the comparison is satisfied, the spatial unit is... Stress concentration zone marking Write it as one, and use the spatial unit. Spatial location written into stress concentration area If the comparison conditions are not met, the spatial unit will be... Stress concentration zone marking Write it as zero. The geostress output layer performs spatial cell stress tensor assembly, principal stress decomposition, adjacency difference calculation, and threshold determination for each spatial cell one by one, forming the three-dimensional principal stress magnitude of each spatial cell. Nine-dimensional principal stress directions Spatial element stress concentration determination quantity Stress concentration zone marking and stress concentration areas .

Claims

1. A three-dimensional stress field nonlinear intelligent inversion method based on prototype learning, characterized in that, Includes the following steps: S1. Obtain in-situ stress measurement data, structural boundary conditions, random field of rock mechanics parameters, spatial unit boundaries, spatial unit adjacency relationships, and rock mass self-weight load in the deep underground engineering calculation domain to form spatial unit state data; S2. Based on the spatial unit state data, establish a prototype library of local stress transfer relationships and obtain prototypes of local stress transfer relationships; S3. In the process of similarity calculation between spatial unit and local stress transfer relationship prototype, the local stress transfer relationship prototype is taken as the local stress transfer relationship to be judged, and the stress tensor to be judged is generated according to the local stress transfer relationship to be judged and the spatial unit state data, forming the stress tensor state to be judged; S4. Based on the stress tensor state to be determined, the spatial unit boundary, the spatial unit adjacency relationship, and the rock mass self-weight load, calculate the surface traction of the spatial unit control body boundary, and obtain the unit boundary traction closure amount based on the surface traction. Embed the unit boundary traction closure amount into the similarity calculation process to determine the prototype similarity under traction closure constraint. S5. Determine the assignment weight of the prototype of the local stress transfer relationship based on the prototype similarity under the traction closure constraint, and form the local stress transfer relationship of the spatial unit; S6. Based on the local stress transmission relationship and the spatial unit state data, generate a three-dimensional geostress field containing six stress components; S7. Based on the three-dimensional geostress field containing six stress components, output the magnitude of the principal stress, the direction of the principal stress, and the stress concentration area.

2. The three-dimensional stress field nonlinear intelligent inversion method based on prototype learning as described in claim 1, characterized in that, Step S1 specifically includes: S11. Obtain the spatial unit boundary of the deep underground engineering calculation domain, determine the spatial unit adjacency relationship based on the shared boundary between adjacent spatial units, and form the spatial unit geometric relationship by the spatial unit boundary and the spatial unit adjacency relationship; S12. Based on the geometric relationship of the spatial unit, the in-situ stress measurement point data is mapped to the spatial unit containing the measurement point location to form a spatial unit stress constraint; S13. Based on the geometric relationship of the spatial units, the construction boundary conditions are mapped to the boundary spatial units, and the loading direction and magnitude of each spatial unit are determined according to the self-weight load of the rock mass, thus forming the loading relationship of the spatial units; S14. Based on the spatial unit geometric relationship, the spatial unit stress constraint, the spatial unit loading relationship, and the random field of rock mass mechanical parameters, form spatial unit state data including in-situ stress measurement point data, structural boundary conditions, random field of rock mass mechanical parameters, spatial unit boundary, spatial unit adjacency relationship, and rock mass self-weight load.

3. The three-dimensional stress field nonlinear intelligent inversion method based on prototype learning as described in claim 1, characterized in that, Step S2 specifically includes: S21. Extract structural boundary conditions, random field of rock mass mechanical parameters and rock mass self-weight load from the spatial unit state data, and map the structural boundary conditions and rock mass self-weight load to the spatial unit position in the random field of rock mass mechanical parameters to form the load transfer characteristics of the spatial unit. S22. Extract the spatial unit boundary and spatial unit adjacency relationship from the spatial unit state data, calculate the boundary normal transmission direction and boundary action area of ​​each spatial unit relative to the adjacent spatial units, and form the spatial unit boundary transmission characteristics; S23. Based on the load transfer characteristics of the spatial unit and the boundary transfer characteristics of the spatial unit, construct a spatial unit transfer representation that represents the transfer path of structural boundary conditions and rock mass self-weight load between adjacent spatial units; S24. Based on the in-situ stress measurement point data in the spatial unit state data, match the stress constraint of the spatial unit corresponding to the measurement point with the spatial unit transfer representation to form a measurement point constraint transfer representation; S25. Based on the measured point constraint transfer representation, perform prototype learning and metric learning on the spatial unit, and form prototype partitioning results according to the consistency of local stress source and boundary transfer direction of the spatial unit; S26. Aggregate the corresponding measurement point constraint transfer representations based on the prototype partitioning results, and configure the transfer mapping relationship of the stress tensor to be judged for the aggregation results to form a local stress transfer relationship prototype. S27. Establish a local stress transfer relationship prototype library based on the aforementioned local stress transfer relationship prototype, and use the local stress transfer relationship prototype library as input for the similarity calculation process between spatial elements and local stress transfer relationship prototypes.

4. The three-dimensional stress field nonlinear intelligent inversion method based on prototype learning as described in claim 3, characterized in that, In step S25, forming the prototype partitioning result based on the consistency of the local stress source and boundary transmission direction of the spatial unit includes: Based on the load transfer characteristics of the spatial unit, the spatial unit positions of the structural boundary conditions and rock mass self-weight load in the random field of rock mass mechanical parameters are determined. Based on the boundary transfer characteristics of the spatial unit, the spatial unit boundary direction corresponding to the transfer of structural boundary conditions and rock mass self-weight load along the adjacency relationship of spatial units is determined. The change direction of the random field of rock mass mechanical parameters, the spatial unit boundary direction, and the measurement point constraint transfer representation are used together as the basis for the division between prototype learning and metric learning.

5. The three-dimensional stress field nonlinear intelligent inversion method based on prototype learning as described in claim 3, characterized in that, Step S3 specifically includes: S31. Receive the prototype of the local stress transfer relationship and the state data of the spatial unit. In the process of similarity calculation between the spatial unit and the prototype of the local stress transfer relationship, extract the structural boundary conditions, random field of rock mechanics parameters and rock mass self-weight load from the state data of the spatial unit to form the load representation of the spatial unit. S32. Based on the spatial unit boundaries and spatial unit adjacency relationships in the spatial unit state data, determine the boundary transfer direction and boundary connection state between spatial units to form a spatial unit boundary transfer representation; S33. Based on the load representation of the spatial unit and the boundary transfer representation of the spatial unit, the action position of the prototype of the local stress transfer relationship is matched to form the prototype action representation of the local stress transfer relationship in the spatial unit; S34. Based on the prototype action, the prototype of the local stress transfer relationship is defined as the local stress transfer relationship to be determined, which corresponds to the spatial unit local stress transfer relationship formed by the structural boundary conditions and the rock mass self-weight load through the random field action of the rock mass mechanical parameters. S35. Based on the local stress transfer relationship to be determined, the load representation of the spatial element, and the boundary transfer representation of the spatial element, generate a stress tensor to be determined that includes normal stress components and shear stress components; S36. Based on the stress tensor to be determined and the spatial unit boundary, the normal stress component and shear stress component in the stress tensor to be determined are spatially aligned to form a stress tensor state to be determined for calculating the surface traction of the spatial unit control body boundary.

6. The three-dimensional stress field nonlinear intelligent inversion method based on prototype learning as described in claim 5, characterized in that, In step S35, generating the stress tensor to be determined, which includes normal stress components and shear stress components, includes: Based on the local stress transfer relationship to be determined, the structural boundary conditions are matched to the boundary transfer direction in the spatial unit boundary transfer representation, the rock mass self-weight load is matched to the spatial direction in the spatial unit loading representation, and the rock mass mechanical parameters random field is matched to the boundary connection state in the spatial unit boundary transfer representation; and the normal stress component and shear stress component in the stress tensor to be determined are determined according to the boundary transfer direction, spatial direction, and boundary connection state.

7. The three-dimensional stress field nonlinear intelligent inversion method based on prototype learning as described in claim 5, characterized in that, Step S4 specifically includes: S41. Receive the stress tensor state to be determined, the spatial unit boundary, the spatial unit adjacency relationship and the rock mass self-weight load. In the similarity calculation process between the spatial unit and the prototype of the local stress transfer relationship, determine the surface normal relationship and surface area relationship of the spatial unit control body boundary according to the spatial unit boundary to form the traction calculation boundary state. S42. Based on the stress tensor state to be determined and the traction calculation boundary state, calculate the normal traction and tangential traction of each element surface of the spatial element control body boundary to form surface traction; S43. Based on the surface traction and the spatial unit adjacency relationship, match the corresponding surface traction of adjacent spatial units on the common boundary to form an adjacent surface traction correspondence relationship; S44. Based on the traction correspondence of the adjacent surfaces and the self-weight load of the rock mass, calculate the closure deviation between the traction resultant within the spatial unit control body and the self-weight load of the rock mass, and form the traction closure amount of the unit boundary. S45. Embed the unit boundary traction closure amount into the similarity calculation process, and update the constraint basis for the similarity calculation of the prototype of the local stress transmission relationship corresponding to the local stress transmission relationship to be judged, so as to form a similarity calculation object under traction closure constraint. S46. Based on the similarity calculation object under the traction closure constraint, the stress tensor state to be determined, and the prototype of the local stress transfer relationship, calculate the matching degree of the local stress transfer relationship to be determined in the spatial unit, and determine the prototype similarity under the traction closure constraint.

8. The three-dimensional stress field nonlinear intelligent inversion method based on prototype learning as described in claim 7, characterized in that, In step S46, when determining the prototype similarity under traction closure constraints, for each local stress transfer relationship prototype corresponding to the same spatial unit, the corresponding surface traction and unit boundary traction closure amount are calculated according to the stress tensor state to be determined, spatial unit boundary, spatial unit adjacency relationship and rock mass self-weight load. The unit boundary traction closure amount corresponding to each local stress transfer relationship prototype is embedded into the similarity calculation process of the same local stress transfer relationship prototype. The basis for similarity calculation of the same local stress transfer relationship prototype is updated with constraints, and the prototype similarity under traction closure constraints is determined before the assignment weight is determined.

9. The three-dimensional stress field nonlinear intelligent inversion method based on prototype learning as described in claim 7, characterized in that, Step S5 specifically includes: S51. Receive the prototype similarity under the traction closure constraint and the prototype of the local stress transfer relationship, and establish a correspondence between the prototype similarity under the traction closure constraint corresponding to the same spatial unit and the prototype of the local stress transfer relationship respectively. S52. Based on the correspondence, normalize the prototype similarity under the traction closure constraint of each local stress transfer relationship prototype to obtain the assignment weight of the local stress transfer relationship prototype. S53. Compare the attribution weight with a preset threshold. If the comparison conditions are met, determine the corresponding local stress transfer relationship prototype as the local stress transfer relationship prototype that participates in the formation of the spatial unit. S54. Based on the assigned weights, the prototypes of local stress transfer relationships that participate in forming the local stress transfer relationships of the spatial unit are weighted and assigned to form the local stress transfer relationships of the spatial unit.

10. The three-dimensional stress field nonlinear intelligent inversion method based on prototype learning as described in claim 9, characterized in that, Step S6 specifically includes: S61. Receive the local stress transfer relationship and the spatial unit state data, extract the structural boundary conditions, random field of rock mechanics parameters and rock mass self-weight load from the spatial unit state data, and determine the stress transfer generation relationship of the spatial unit according to the local stress transfer relationship; S62. Based on the stress transfer generation relationship, the structural boundary conditions, the random field of rock mass mechanical parameters, and the rock mass self-weight load, generate six stress components of the spatial unit; S63. Based on the spatial unit boundary and spatial unit adjacency relationship in the spatial unit state data, the six stress components of adjacent spatial units are matched according to the shared boundary to form a spatial unit stress arrangement relationship; S64. Based on the stress arrangement relationship of the spatial unit and the six stress components of the spatial unit, a three-dimensional geostress field containing the six stress components is generated in the deep underground engineering calculation domain.

11. The three-dimensional stress field nonlinear intelligent inversion method based on prototype learning as described in claim 10, characterized in that, Step S7 specifically includes: S71. Receive the three-dimensional geostress field containing six stress components, and extract six stress components from the three-dimensional geostress field containing six stress components for each spatial unit to form a spatial unit stress tensor. S72. Calculate the magnitude and direction of the principal stresses of the spatial element based on the spatial element stress tensor to form the principal stress state of the spatial element; S73. Based on the principal stress state of the space unit and the three-dimensional geostress field containing six stress components, calculate the difference in principal stress magnitude and principal stress direction between adjacent space units to form a space unit stress concentration determination quantity; S74. Compare the stress concentration determination quantity of the spatial unit with a preset threshold, determine the stress concentration area if the comparison conditions are met, and output the principal stress magnitude, principal stress direction and stress concentration area.

Images