Slope restoration scheme design method based on generative adversarial fusion algorithm

Abstract

The invention discloses a side slope restoration scheme design method based on a generative adversarial fusion algorithm, and aims to solve the problem that a generation scheme is difficult to embed into engineering specifications and physical constraints and joint design of a discrete construction method and continuous parameters. A hierarchical adversarial generator and a double-branch discriminator are adopted, a geometric mask-parameter projection two-stage corrector is arranged, multi-working-condition rapid checking and calculation are carried out in combination with a calibrated physical agent model, a compliance proof chain is generated, and a repair scheme arrangement diagram, a parameter list and an engineering quantity list are output. The technical effects that the scheme compliance and physical consistency are unified, the design efficiency is improved, and the interpretability is enhanced are achieved.

Description

Technical Field

[0001] This invention relates to the field of geotechnical engineering design, and in particular to a slope repair scheme design method based on a generative adversarial fusion algorithm. Background Technology

[0002] The design of slope restoration schemes requires comprehensive consideration of topography, geology and hydrology, constraints of surrounding infrastructure boundaries, and the impact of multiple conditions such as heavy rainfall, long-duration rainfall, and earthquakes, while also meeting the requirements of relevant engineering specifications. Current engineering practices typically employ limit equilibrium methods and numerical analysis methods (such as the finite element method and finite difference method) for stability, seepage, and deformation assessment, combined with geographic information systems and building information models for data integration and result presentation. Regarding intelligent solutions, there are existing multi-objective optimization methods based on heuristic and evolutionary algorithms, physical proxy models based on machine learning for rapid calculation, and explorations into scheme generation based on generative adversarial networks and variational autoencoders. Simultaneously, some research attempts to use knowledge graphs and constraint solving to perform machine-readable verification of specification clauses. Overall, the toolchain is continuously improving, but the linkage between generation, verification, and physical consistency remains relatively loose.

[0003] However, existing technologies still have the following shortcomings:

[0004] 1. It is difficult to embed the specification clauses and physical constraints simultaneously during the generation stage. The clauses are mostly verified by textual post-verification, resulting in repeated rework. High-cost simulation is difficult to support real-time feedback during the iteration period and is difficult to form effective constraints on the generation process.

[0005] 2. Discrete construction methods and continuous parameters lack a collaborative design mechanism. Common practices involve processing layout and parameters step by step. Geometric restricted areas, minimum spacing, boundary setbacks, drainage connectivity and construction passage continuity are difficult to meet in an integrated manner. After local adjustments, there is a lack of systematic correction methods to quickly project the solution back to the feasible region with the minimum modification criterion.

[0006] 3. The interpretability of rapid multi-condition calculation and compliance is insufficient. The proxy model has limited cross-scenario calibration and generalization capabilities, making it difficult to consistently assess heavy rainfall, long-duration and earthquake conditions. The generated results lack a clause-level, traceable evidence chain with spatial positioning and time series between the standard clauses and the standard clauses, which affects the engineering review and acceptance.

[0007] Therefore, a slope repair design method that can overcome the shortcomings of the existing technology is a problem that needs to be solved by those skilled in the art. Summary of the Invention

[0008] One objective of this invention is to propose a slope repair scheme design method based on a generative adversarial fusion algorithm. Addressing the problems in existing technologies such as the difficulty in simultaneously embedding engineering specifications and physical constraints during the scheme generation stage, the lack of collaborative design between discrete construction methods and continuous parameters, and insufficient multi-condition verification and compliance interpretability, this invention proposes a core technical solution. This solution includes structured alignment of data and specifications, modeling of geometric constraints and neural symbolic logic constraints, joint scoring by a hierarchical adversarial generator and a dual-branch discriminator, two-level correction using geometric masks and parameter projections, rapid verification using a multi-condition proxy model, and output of a compliance proof chain. This invention achieves unified compliance with specifications and physical consistency during the generation stage, rapidly converges to a feasible solution with minimal modifications, ensures drainage connectivity and construction channel continuity, improves design efficiency and first-time compliance rate, and enhances the interpretability and traceability of results.

[0009] A slope restoration scheme design method based on a generative adversarial fusion algorithm according to an embodiment of the present invention is characterized by comprising:

[0010] S1. Obtain scene data of the target slope and the red line of surrounding facilities, perform preprocessing, generate aligned scene data and red line layers, and perform structured parsing of engineering specification clauses to generate a standardized set of specification clauses.

[0011] S2. Construct a set of geometric constraints based on the red line layer and spatial setback rules, construct a neural symbolic logic constraint model based on the standardized set of normative clauses, extract scene features based on aligned scene data to generate scene feature tensors, and train and calibrate a physical proxy model for rapid calculation of stability, seepage and deformation.

[0012] S3. First, use the hierarchical adversarial generator to generate the initial construction method layout diagram, then generate the initial parameter matrix under conditions. Calculate the compliance score by combining the compliance branch of the adversarial discriminator with the neural symbolic logic constraint model, and calculate the physical score by combining the physical consistency branch of the adversarial discriminator with the calibrated physical proxy model. Then, score and select the candidate schemes.

[0013] S4. Generate a geometric mask based on the set of geometric constraints, perform restricted area masking, minimum spacing correction and boundary setback correction on the initial construction method layout drawing, while maintaining drainage connectivity and construction passage continuity, and output the construction method layout drawing and geometric correction vector for geometric correction.

[0014] S5. Based on the clause threshold of the neural symbolic logic constraint model and the calculation results of the calibrated physical agent model, construct a continuous parameter feasible region, perform a projection operation on the initial parameter matrix with the criterion of minimum modification, and output the parameter matrix and parameter correction vector after parameter correction.

[0015] S6. Perform rapid calculations under heavy rainfall, long-duration rainfall and earthquake conditions to generate a set of physical indicators including stability safety factor, pore water pressure, seepage flow and displacement, and calculate the set of clause satisfaction based on neural symbolic logic constraint model.

[0016] S7. Generate clause-level records containing clause number, clause satisfaction, corresponding physical index reference, spatial location label, geometric correction vector and parameter correction vector according to the standard clause number, and organize them into a compliance proof chain in chronological order and spatial index.

[0017] S8. Generate the repair plan layout diagram, parameter list and bill of quantities.

[0018] Optionally, step S1 specifically includes:

[0019] The scene data includes remote sensing images, lidar point clouds, geological exploration data, and hydrological monitoring data;

[0020] The coordinates of remote sensing images, lidar point clouds, geological exploration data and hydrological monitoring data are unified to the same plane coordinate system and the same elevation datum, the timestamps are aligned to the same time axis, the spatial resolution is coordinated and the units are standardized, and outlier removal and noise suppression are implemented to ensure data quality.

[0021] The surrounding facility boundary lines are standardized into vector layers, their coordinates are transformed to the aforementioned plane coordinate system and elevation datum, and topology repair is performed to eliminate geometric breaks and overlaps.

[0022] The engineering specification clauses are structured and parsed to extract clause numbers, parameter thresholds, logical relationships and scope of application, and then converted into machine-readable standard fields.

[0023] Definition of the noun:

[0024] The surrounding facilities red line is a set of boundary lines for roads, pipelines, buildings, and legally designated land use areas and safety control areas, used to define restricted areas and setback distances;

[0025] The red line layer is a layer that represents the red lines of surrounding facilities in vector data form and has undergone coordinate transformation and topology repair, and is used for geometric constraint construction;

[0026] The engineering specification clauses refer to a collection of clauses from engineering standards and technical specifications applicable to slope repair design and construction.

[0027] The remote sensing images are surface image data acquired through aerial or satellite sensors, used to reflect topographic and surface cover characteristics;

[0028] The lidar point cloud is spatial point data collected by airborne or ground lidar, used to reconstruct the three-dimensional morphology of terrain and structure.

[0029] The geological exploration data refers to the rock and soil parameters and stratigraphic information data formed by drilling, sampling, testing and geological mapping.

[0030] The hydrological monitoring data refers to the spatiotemporal monitoring data of hydrological elements such as rainfall, groundwater level, pore water pressure and flow rate;

[0031] The clause number is a unique identifier for the normative clause in the normative text or structured set;

[0032] The parameter threshold is the upper or lower limit or range boundary of the value specified in the standard clause;

[0033] The logical relationship refers to the logical combination relationship between the conditions in the normative clause, including forms such as AND, OR, NOT, and implication.

[0034] The scope of application refers to the definition of the applicable conditions of the standard provisions under specific scenarios, materials, construction methods and working conditions.

[0035] The machine-readable standard fields are the normative clause data represented by fields such as clause number, parameter threshold, logical relationship and scope of application after structured parsing.

[0036] Optionally, step S2 specifically includes:

[0037] Using aligned scene data, redline layers, and a standardized set of specifications as input, multi-source feature extraction is performed on the aligned scene data. The spatial morphological features of remote sensing images and lidar point clouds, along with the attribute features of geological exploration data and hydrological monitoring data, are encoded using a unified raster or grid to generate scene feature tensors that express topography, strata, and hydrological conditions.

[0038] Based on the red line layer, construct a set of geometric constraints. The set of geometric constraints includes at least the restricted area constraint, the minimum spacing constraint, and the boundary setback constraint, and limits the drainage connectivity and the continuity of the construction passage.

[0039] A neural symbolic logic constraint model is constructed based on a standardized set of normative clauses. Clause numbers, parameter thresholds, and logical relationships are mapped to computable logical constraints, enabling the logical constraints to output clause satisfaction when subsequent input of construction method layout diagrams and parameter matrices.

[0040] Based on the aligned scene data combined with historical numerical analysis or on-site calculation results, the physical surrogate model used for rapid calculation of stability, seepage and deformation is trained and calibrated to obtain the calibrated physical surrogate model.

[0041] Definition of the noun:

[0042] The aligned scene data is a data set that has been unified to the same plane coordinate system and the same elevation datum, timestamps aligned to a unified time axis, resolution coordinated and units standardized, and outlier removal and noise suppression completed.

[0043] The standardized set of normative clauses is a set of normative clause data formed after structured parsing, including standard fields such as clause number, parameter threshold, logical relationship and scope of application;

[0044] The space setback rule is a space restriction rule for the minimum setback distance or buffer range determined based on the safety control requirements of surrounding facilities;

[0045] The set of geometric constraints is a set of spatial restrictions for the construction method layout, including at least no-entry zone constraints, minimum spacing constraints, and boundary setback constraints, and includes requirements for drainage connectivity and construction passage continuity.

[0046] The restricted area constraint is a geometric restriction that prohibits the placement of construction method entities within a specific area;

[0047] The minimum spacing constraint is a geometric restriction that limits the minimum spatial distance between construction entities.

[0048] The boundary setback constraint is a geometric restriction on the minimum setback distance that the construction entity must maintain relative to the boundary.

[0049] The drainage connectivity refers to the requirement that drainage facilities form a continuous and unobstructed passage in space.

[0050] The continuity of the construction access means that the construction access path remains spatially connected and meets the requirements of access width and slope, etc.

[0051] The multi-source feature extraction is a process of jointly extracting and fusing features from data from different sources.

[0052] The spatial morphological features are geometric features such as terrain undulations and structure shapes reflected by remote sensing images and lidar point clouds.

[0053] The attribute features are non-geometric features such as soil and rock parameters, stratigraphic properties, and hydrological conditions provided by geological exploration data and hydrological monitoring data.

[0054] The unified grid or mesh is a grid discretization representation of the scene spatial domain with consistent resolution and coordinate alignment;

[0055] The encoding process is a procedure that converts multi-source features into numerical multi-channel representations using a unified raster or grid.

[0056] The scene feature tensor is a multi-channel numerical tensor that expresses the terrain, strata and hydrological conditions, and can be used as input for generation and discrimination models;

[0057] The neural symbolic logic constraint model is a model that maps normative clauses into computable logical constraints and combines them with neural network reasoning to output the clause satisfaction degree.

[0058] The computable logical constraints are combinations of predicates or inequalities that can be executed by the program, used to assess compliance;

[0059] The construction method layout diagram is a discrete layout diagram that spatially represents the type, location, and geometric shape of the construction method entities.

[0060] The parameter matrix is ​​a matrix representation of a continuous set of parameters corresponding to the construction method entity, including values ​​for dimensions, angles, spacing, and longitudinal slope.

[0061] The clause satisfaction level is an indicator of the degree to which the normative clauses are satisfied under the current arrangement and parameters, and can be in Boolean or numerical form.

[0062] The physical proxy model is a fast computational model that approximates stability, seepage and deformation response using machine learning methods;

[0063] The training and calibration process involves learning parameters and correcting errors in the surrogate model using historical numerical analysis and on-site verification results.

[0064] The historical numerical analysis is a collection of calculation results of mechanical and hydrodynamic responses under different working conditions obtained by numerical methods in the past.

[0065] The on-site verification results are calculated or inverted based on physical indicators obtained from monitoring and on-site tests.

[0066] The calibrated physical proxy model is a proxy model that has completed training and calibration and achieved a predetermined accuracy under the target scenario and working conditions;

[0067] The rapid calculation is a process of efficiently outputting physical indicators such as stability safety factor, pore water pressure, seepage flow and displacement using a calibrated physical surrogate model.

[0068] Optionally, step S3 specifically includes:

[0069] Using scene feature tensor, neural symbolic logic constraint model and calibrated physical proxy model as input, the discrete construction method generator of the hierarchical adversarial generator is invoked to generate a construction method layout probability map under the conditions of the scene feature tensor and the neural symbolic logic constraint model, and the construction method layout probability map is converted into an initial construction method layout map.

[0070] After obtaining the initial construction method layout diagram, the continuous parameter sub-generator of the hierarchical adversarial generator is called to generate the initial parameter matrix under the conditions of the scene feature tensor, the initial construction method layout diagram and the neural symbolic logic constraint model.

[0071] The initial construction layout diagram and the initial parameter matrix are input into the compliance branch of the adversarial discriminator, and a compliance score is calculated by combining the neural symbolic logic constraint model. The initial construction layout diagram and the initial parameter matrix are input into the physical consistency branch of the adversarial discriminator, and a physical score is calculated by combining the calibrated physical proxy model. One or more candidate initial schemes are selected based on a comprehensive score of compliance score and physical score.

[0072] Definition of the noun:

[0073] The hierarchical adversarial generator is a two-stage generation model that includes a discrete construction method sub-generator and a continuous parameter sub-generator. First, the construction method layout is generated, and then the parameters are generated under its conditions.

[0074] The discrete construction method sub-generator is a sub-model that outputs a construction method layout probability map under the constraints of scene feature tensor and neural symbolic logic, and obtains an initial construction method layout map accordingly.

[0075] The construction method layout probability map is a multi-channel numerical layer that gives the probability of construction method category or layout state at each location on a spatial grid.

[0076] The initial construction method layout diagram is the first version of the construction method layout result obtained by thresholding and discretizing the construction method layout probability diagram with conflict resolution rules.

[0077] The continuous parameter sub-generator is a sub-model that outputs continuous parameters corresponding to each construction entity under the conditions of scene feature tensor and initial construction method layout diagram;

[0078] The initial parameter matrix is ​​the output of the continuous parameter sub-generator and the initial construction method layout. Figure 1 A corresponding continuous parameter value matrix;

[0079] The adversarial discriminator is an evaluation model paired with the generator, used to determine the normative compliance and physical consistency of candidate solutions;

[0080] The compliance branch is an evaluation branch that calculates the degree to which candidate solutions meet the normative clauses by combining the neural symbolic logic constraint model;

[0081] The physical consistency branch is an evaluation branch that combines a calibrated physical surrogate model to calculate the rationality of candidate schemes in terms of stability, seepage, and deformation.

[0082] The compliance score is the score of the degree to which the measurement scheme output by the compliance branch meets the normative clauses; the physical score is the score of the physical reasonableness of the measurement scheme output by the physical consistency branch.

[0083] The candidate scheme is an example of a scheme to be evaluated, consisting of an initial construction method layout diagram and an initial parameter matrix;

[0084] The comprehensive score is an overall evaluation value obtained by fusing the compliance score and the physical score according to a set weight or the Pareto optimization principle.

[0085] The scoring selection is a process of ranking candidate solutions based on comprehensive scores and selecting the solution with the higher score to enter the subsequent process.

[0086] The conditional generation is a constrained generation method performed under given scene feature tensors, initial construction method layout diagrams, and logical constraints.

[0087] Optionally, step S4 specifically includes:

[0088] Using the initial construction layout diagram and geometric constraint set as input, and based on the forbidden zone boundary, minimum spacing threshold and boundary setback distance in the geometric constraint set, a geometric mask is generated to indicate the deployable and non-deployable areas through buffering, offsetting and set operations.

[0089] The geometric mask is applied to the initial construction method layout drawing, and the forbidden area masking, minimum spacing correction and boundary setback correction are performed in sequence according to the principle of minimum modification. Among them, the construction method entities that fall into the un-layout area are projected to the nearest feasible position or locally clipped. The construction method entities with insufficient spacing are adjusted with minimum displacement and low priority entities are deleted when the spacing cannot be met. The construction method entities with insufficient setback are offset or length clipped.

[0090] In the revised layout, drainage connectivity and construction passage continuity are checked. If they are not met, the relevant linear entities are finely adjusted or locally extended within the allowable range of the geometric mask according to the principle of minimum modification until the requirements are met.

[0091] Output the construction method layout diagram and geometric correction vector for geometric correction. The geometric correction vector records the spatial changes and operation types for each construction method entity.

[0092] Definition of the noun:

[0093] The geometric mask is a mask layer used to mark the deployable and non-deployable areas, generated by buffering, offsetting and set operations based on the no-entry zone boundary, minimum spacing threshold and boundary setback distance.

[0094] The restricted zone boundary is the boundary line of the non-arrangeable area specified in the geometric constraint set, used to limit the range in which the construction method entity is not allowed to enter;

[0095] The minimum spacing threshold is a numerical limit on the minimum spatial distance that must be met between construction entities.

[0096] The boundary setback distance is the minimum safe distance that the construction entity must maintain relative to the red line or important boundary;

[0097] The buffer is a geometric generation operation that expands or contracts a strip-shaped region around an existing boundary by a set distance.

[0098] The offset is a geometric operation that moves the entire geometric object in a specified direction and distance;

[0099] The set operation is a processing method that performs union, intersection, difference, and other operations on geometric regions to form a mask;

[0100] The arrangeable and non-arrangeable regions are sets of spatial units in the geometric mask that are respectively marked as allowed and prohibited from being arranged.

[0101] The restricted area masking is a process that eliminates the illegal occupation of construction entities that fall into the undesirable area by projecting them to the nearest feasible location or by partially cutting them off;

[0102] The minimum spacing correction is a process of performing minimum displacement adjustment on construction entities with insufficient spacing or deleting low-priority entities when the minimum spacing cannot be met.

[0103] The boundary setback correction involves performing offset or length trimming on construction entities with insufficient setback to meet the setback distance requirement;

[0104] The projection of the nearest feasible location is the operation of mapping the construction entity located in the undesirable area to the nearest location within the allowable range of the geometric mask;

[0105] The local trimming is an operation that partially cuts off linear or planar structural entities that exceed the allowable range;

[0106] The minimum displacement adjustment is the operation of changing the position of the construction entity with the minimum spatial movement to meet the spacing or setback constraints.

[0107] The low-priority entities are construction method entities that are ranked at a lower level according to a preset importance order.

[0108] The linear entity is a construction facility with linear geometric expression, including drainage ditches, pipelines, or retaining structures represented by axes, etc.

[0109] The drainage connectivity and construction passage continuity verification is a process of checking whether the drainage path and the construction passage path are continuous and unobstructed in the revised layout.

[0110] The position fine-tuning refers to the operation of making small-scale position adjustments to linear entities within the allowable range of the mask to satisfy connectivity or constraints.

[0111] The local extension is the operation of increasing the length of a linear entity within an allowable range to connect it to the target path or node;

[0112] The principle of minimum modification is to select the combination of displacement, clipping, and deletion based on the criterion of minimizing the total amount of modification during the masking and correction process;

[0113] The geometric correction method layout diagram is the layout result diagram after completing the no-entry zone masking, minimum spacing and boundary setback correction and passing the connectivity verification;

[0114] The geometric correction vector is a set of spatial changes and operation types recorded for each construction entity, used to describe the specific modifications for geometric correction.

[0115] Optionally, step S5 specifically includes:

[0116] Using the geometrically corrected construction layout diagram, initial parameter matrix, neural symbolic logic constraint model, and calibrated physical proxy model as input, while keeping the geometrically corrected construction layout diagram unchanged, the calibrated physical proxy model is called to quickly calculate the initial parameter matrix to obtain physical indicators. Based on the clause thresholds and applicable relationships interpreted by the neural symbolic logic constraint model, the continuous parameter feasible region composed of interval constraints such as parameter upper and lower limits, minimum or maximum inclination angle, minimum or maximum spacing, minimum longitudinal slope, and inequality constraints is constructed as a constraint set on the initial parameter matrix.

[0117] Under the continuous parameter feasible region, a projection model with minimum modification as the criterion is established. The parameter matrix with the minimum difference metric from the initial parameter matrix and simultaneously satisfying the constraint set is determined as the parameter matrix after parameter correction. For parameters with coupling relationship, simultaneous projection is used to satisfy the intra-group constraints.

[0118] Record the parameter correction vector, which is the element difference between the parameter matrix after parameter correction and the initial parameter matrix.

[0119] Definition of the noun:

[0120] The threshold for the clause is the numerical boundary set by the canonical clauses for the relevant continuous parameters, as interpreted by the neural symbolic logic constraint model.

[0121] The applicable relationship refers to the mapping and condition judgment of the applicable threshold of the clause under different construction method categories, material conditions, spatial locations and working conditions;

[0122] The upper and lower limits of the parameter are defined as the minimum and maximum values ​​that the continuous parameter is allowed to take.

[0123] The inclination angle is the angle between the construction entity and the horizontal plane, and the minimum or maximum inclination angle is the boundary of the inclination angle range specified in the standard clauses;

[0124] The spacing is the center distance or net distance between adjacent construction entities, and the minimum or maximum spacing is the boundary of the distance range specified in the standard clauses.

[0125] The longitudinal slope is the slope of linear drainage or passage facilities along the route, and the minimum longitudinal slope is the minimum slope requirement to ensure drainage and passage.

[0126] The interval constraint is a restriction that the parameter value must fall within a specified closed interval or open interval.

[0127] The inequality constraints are parameter relationship restrictions expressed in the form of greater than, less than, not less than, not greater than, etc.

[0128] The feasible region of the continuous parameters is the set of all parameter values ​​that simultaneously satisfy the clause threshold and the physical calculation requirements.

[0129] The constraint set is represented by a combination of all interval constraints and inequality constraints imposed on the initial parameter matrix;

[0130] The projection model is a mathematical model that maps the initial parameter matrix into a parameter matrix that satisfies the constraints and has the smallest difference from the original value within the feasible domain specified by the constraint set.

[0131] The difference metric is an index that measures the difference between the parameter matrix after correction and the initial parameter matrix, and can be vector norm, weighted absolute difference or weighted squared difference;

[0132] The minimum modification criterion is a criterion that minimizes the difference metric in the projection model as the optimization objective;

[0133] The coupling relationship refers to the interdependence and linkage between multiple parameters determined by norms or physical mechanisms;

[0134] Simultaneous projection is a processing method that involves jointly adjusting parameters with coupling relationships in the same optimization step to simultaneously satisfy intra-group constraints.

[0135] The intra-group constraints are the linkage restrictions and consistency requirements within the same construction method entity or the same parameter group;

[0136] The parameter matrix after parameter correction is a continuous parameter value matrix that has been adjusted by the projection model and satisfies the constraint set.

[0137] The parameter correction vector is a set of vectors composed of the differences between the corresponding elements of the parameter matrix after parameter correction and the initial parameter matrix.

[0138] The element difference is the numerical difference between the value obtained after correction at the same parameter index and the initial value.

[0139] The physical indicators are quantitative indicators such as stability safety factor, pore water pressure, seepage flow and displacement output by the calibrated physical surrogate model under a given arrangement and parameters.

[0140] Optionally, step S6 specifically includes:

[0141] Using the parameter matrix after parameter correction, the construction layout diagram after geometric correction, the calibrated physical surrogate model, and the neural symbolic logic constraint model as input, the calibrated physical surrogate model is invoked to quickly calculate the construction layout diagram after geometric correction and the parameter matrix after parameter correction under heavy rainfall, long-duration rainfall, and earthquake conditions, respectively, to obtain the stability safety factor, pore water pressure, seepage flow, and displacement for each condition. The calculation results of each condition are then summarized to form a set of physical indicators.

[0142] The physical index set, the geometric correction construction method layout diagram, and the parameter matrix after parameter correction are input into the neural symbolic logic constraint model. The clause satisfaction is calculated according to the standard clause number, and a clause satisfaction set is formed.

[0143] Definition of the noun:

[0144] The stability safety factor is a safety reserve index used to evaluate the anti-sliding stability of a slope, reflecting the ratio of anti-sliding force to sliding force;

[0145] The pore water pressure is the pressure generated by pore water in the soil and rock mass, reflecting the seepage process and changes in the effective stress state.

[0146] The seepage flow rate is the amount of water passing through a specified boundary or cross-section per unit time, reflecting the drainage capacity of the drainage system.

[0147] The displacement refers to the horizontal and vertical displacement responses of characteristic points of the slope or related structures under given working conditions.

[0148] The calculation results are the stability safety factor, pore water pressure, seepage flow and displacement values ​​obtained by the calibrated physical proxy model under each working condition.

[0149] The set of physical indicators is a collection that summarizes and organizes the calculation results under different working conditions according to working conditions and spatial objects;

[0150] The clause satisfaction set is a set of results including satisfaction status and margins output according to clause number after inputting the physical index set, the geometric correction construction method layout diagram, and the parameter matrix after parameter correction into the neural symbolic logic constraint model.

[0151] Optionally, step S7 specifically includes:

[0152] Using the set of physical indicators and the set of clause satisfaction, as well as the geometric correction vector and the parameter correction vector, as inputs, each clause of the specification is processed sequentially according to the clause number.

[0153] For each specification clause, the clause number and clause satisfaction degree are read from the clause satisfaction degree set. One or more physical indicators used to calculate the clause satisfaction degree are selected from the physical indicator set as the corresponding physical indicator reference. The spatial location label corresponding to the clause is determined based on the construction method entity index and spatial change amount recorded in the geometric correction vector and parameter correction vector. The clause number, clause satisfaction degree, corresponding physical indicator reference, spatial location label, geometric correction vector and parameter correction vector are combined to form a clause-level record.

[0154] All clause-level records are sorted in chronological order of their creation time, and a spatial index is created based on spatial location annotations to organize them into a chain of compliance proofs.

[0155] Definition of the noun:

[0156] The corresponding physical index reference refers to the reference relationship between the index and its value selected from the physical index set and associated with a specific standard clause, which is used to support the calculation and explanation of the clause satisfaction.

[0157] The construction method entity index is a number or identifier used to uniquely identify the construction method entity, and is used to locate the specific entity in records and annotations;

[0158] The spatial location annotation is the positioning information of the spatial range corresponding to the clause-level record, including the construction method entity index and positioning parameters such as coordinates, line segments or regions;

[0159] The spatial change amount is a quantitative change such as displacement, offset, cutting length, angle adjustment and size increase or decrease recorded in the geometric correction vector or parameter correction vector;

[0160] The clause-level record is a record object formed by the standard clause as the basic unit, which includes fields such as clause number, clause satisfaction, corresponding physical index reference, spatial location label, geometric correction vector and parameter correction vector;

[0161] The generation time is the timestamp when the clause-level record is generated, used to reflect the order in which the records are formed;

[0162] The chronological order is a rule that sorts all clause-level records from earliest to latest based on their creation time;

[0163] The spatial index is an index structure built based on spatial location annotations, used for quickly retrieving and organizing clause-level records by location or entity;

[0164] The compliance proof chain is a clause-level record sequence organized chronologically and combined with spatial indexing, used to demonstrate the formation process and spatial distribution of compliance evidence.

[0165] Optionally, step S8 specifically includes:

[0166] Using the compliance proof chain, the geometrically corrected construction method layout diagram, and the parameter matrix after parameter correction as input, a repair scheme layout diagram is generated based on the geometrically corrected construction method layout diagram. The location, size, spacing, and boundary setback are marked in the repair scheme layout diagram according to the construction method category and construction method entity index. Compliance markings are formed in the corresponding spatial positions based on the clause number and clause satisfaction in the compliance proof chain.

[0167] The parameter list is generated by taking the parameter matrix after parameter correction as input. The parameter matrix is ​​expanded according to the construction method entity index and the names of continuous parameters and design values ​​are listed. The correspondence with the clause number is recorded.

[0168] The bill of quantities is generated by taking the repair plan layout drawing and parameter list as input. Based on the geometric dimensions and quantities in the repair plan layout drawing and the design values ​​in the parameter list, the length, area, volume and piecework quantity are calculated using the engineering quantity calculation rules and then summarized.

[0169] Definition of the noun:

[0170] The repair scheme layout diagram is a spatial result diagram that expresses the type, location and geometric shape of the construction method entity under a unified coordinate system and elevation datum and is organized according to the layer structure.

[0171] The parameter list is a list-based data table of continuous parameters, units, values ​​and allowable ranges of each construction method entity corresponding to the layout diagram;

[0172] The bill of quantities is a list that summarizes the quantities of various construction methods and materials according to measurement rules, and includes fields such as quantity items, units and calculation formulas;

[0173] The layer structure is a hierarchy and naming rule used to classify and organize the elements of the layout map, ensuring that the elements are searchable, editable, and statistically analyzeable.

[0174] The beneficial effects of this invention are:

[0175] 1. Simultaneously embed engineering specifications and physical constraints during the solution generation stage: The compliance branch and physical consistency branch of the neural symbolic logic constraint model and the adversarial discriminator are used for joint scoring, and the compliance proof chain records the evidence at the clause level, spatial location and temporal order, thereby improving the first-time compliance rate, reducing rework in the later stage and enhancing interpretability and auditability;

[0176] 2. Achieve collaborative design and rapid correction of discrete construction methods and continuous parameters: A hierarchical adversarial generator is used to generate construction layout and parameter matrices respectively. Combined with two-level correction of geometric mask and parameter projection, the solution is projected to the feasible region under the principle of minimum modification, while ensuring drainage connectivity and construction channel continuity, improving solution stability and design efficiency, and reducing manual fine-tuning.

[0177] 3. Provides rapid calculation and closed-loop optimization capabilities under multiple working conditions: The calibrated physical surrogate model can quickly output stability safety factors, pore water pressure, seepage flow and displacement under heavy rainfall, long-duration rainfall and earthquake conditions, which significantly reduces simulation costs, accelerates iterative convergence and improves the physical credibility and generalization applicability of the scheme. Attached Figure Description

[0178] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0179] Figure 1 This is a flowchart of a slope repair scheme design method based on a generative adversarial fusion algorithm proposed in this invention. Detailed Implementation

[0180] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0181] refer to Figure 1 A slope restoration scheme design method based on generative adversarial fusion algorithm, characterized by comprising:

[0182] S1. Obtain scene data of the target slope and the red line of surrounding facilities, perform preprocessing, generate aligned scene data and red line layers, and perform structured parsing of engineering specification clauses to generate a standardized set of specification clauses.

[0183] S2. Construct a set of geometric constraints based on the red line layer and spatial setback rules, construct a neural symbolic logic constraint model based on the standardized set of normative clauses, extract scene features based on aligned scene data to generate scene feature tensors, and train and calibrate a physical proxy model for rapid calculation of stability, seepage and deformation.

[0184] S3. First, use the hierarchical adversarial generator to generate the initial construction method layout diagram, then generate the initial parameter matrix under conditions. Calculate the compliance score by combining the compliance branch of the adversarial discriminator with the neural symbolic logic constraint model, and calculate the physical score by combining the physical consistency branch of the adversarial discriminator with the calibrated physical proxy model. Then, score and select the candidate schemes.

[0185] S4. Generate a geometric mask based on the set of geometric constraints, perform restricted area masking, minimum spacing correction and boundary setback correction on the initial construction method layout drawing, while maintaining drainage connectivity and construction passage continuity, and output the construction method layout drawing and geometric correction vector for geometric correction.

[0186] S5. Based on the clause threshold of the neural symbolic logic constraint model and the calculation results of the calibrated physical agent model, construct a continuous parameter feasible region, perform a projection operation on the initial parameter matrix with the criterion of minimum modification, and output the parameter matrix and parameter correction vector after parameter correction.

[0187] S6. Perform rapid calculations under heavy rainfall, long-duration rainfall and earthquake conditions to generate a set of physical indicators including stability safety factor, pore water pressure, seepage flow and displacement, and calculate the set of clause satisfaction based on neural symbolic logic constraint model.

[0188] S7. Generate clause-level records containing clause number, clause satisfaction, corresponding physical index reference, spatial location label, geometric correction vector and parameter correction vector according to the standard clause number, and organize them into a compliance proof chain in chronological order and spatial index.

[0189] S8. Generate the repair plan layout diagram, parameter list and bill of quantities.

[0190] In this specific embodiment, S1 specifically refers to:

[0191] First, four types of source data were collected: remote sensing imagery, lidar point clouds, geological exploration data, and hydrological monitoring data. Coordinates were unified to the same plane coordinate system and elevation datum; timestamps were aligned to a unified time axis; spatial resolution was coordinated; and units were standardized. Outlier removal and noise suppression were performed during this process to create a consistent data foundation suitable for subsequent constraint modeling and data generation. To align the source data in spatial reference, a joint transformation mapping was constructed to form a unified dataset, represented as:

[0192] With type mapping ;

[0193] in This represents a set of scene data after coordinates and elevation have been unified. Indicates the first class original dataset and These correspond to remote sensing images, lidar point clouds, geological exploration data, and hydrological monitoring data, respectively. Indicates the first The source coordinate reference frame of the data type Indicates the first The source elevation datum for this type of data Indicates the target plane coordinate system. Indicates the target elevation datum. This represents the joint transformation mapping from the source coordinate reference system and source elevation datum to the target plane coordinate system and target elevation datum;

[0194] After achieving spatial reference unification, a composite operator relationship is adopted to enable coordinated preprocessing of time, resolution, and units, and to mitigate anomalies and noise within this process:

[0195] ;

[0196] in This represents the set of scene data after preprocessing. The time alignment function is used to map timestamps from different data sources to a unified timeline. The resolution coordination operator is used to resample spatial data to the target resolution parameters. , The unit standardization operator is used to map the units of various physical quantities to a set of standard units. This is an outlier mask set used to remove sample points that do not meet the normal statistical criteria. This is a noise suppression filter operator, and the filter strength is determined by parameters. control;

[0197] For the surrounding facility boundary lines, to ensure geometric semantic consistency and spatial reference uniformity, the format was first standardized to a vector layer, then coordinate transformation was performed and topology repair was executed to eliminate geometric breaks and overlaps. The formal processing is as follows:

[0198] ;

[0199] in This represents the processed red line layer. This represents the original red line data. The format unification operator is used to unify red line data into a vector layer format. This indicates the transformation of the red line coordinates and elevation to the target plane coordinate system. With the target elevation benchmark The mapping, This indicates that the topology repair operator is used to eliminate geometric breaks and overlaps and ensure topological consistency;

[0200] Finally, to ensure that engineering specification clauses can be machine-readable for evaluation and constraint propagation in subsequent processes, the specification text is structured and its fields are extracted, defining a parsing mapping:

[0201] ;

[0202] in This represents a standardized set of normative clauses. This represents the structured parsing function for the clauses, where Text represents the text corpus of the specification clauses. Indicates the total number of clauses. Indicates the clause number, This represents the set of parameter thresholds related to the terms. The logical relationships indicating the internal conditions of a clause include AND, OR, NOT, and implied conditions. It indicates the scope of application of the terms and conditions, including construction method categories, material conditions, spatial location, and working condition constraints;

[0203] The aligned scene data is obtained after the above processing. With the processed red line layer And form a standardized set of normative clauses. As input for subsequent steps.

[0204] In this specific embodiment, S2 specifically refers to:

[0205] With aligned scene data Red line layer With a standardized set of normative clauses As input, firstly, unified raster encoding and fusion of multi-source features are performed to form a scene feature tensor for subsequent generation and discrimination calls. Its concise expression is as follows:

[0206] ;

[0207] in The scene feature tensor is used to represent topographic strata and hydrological conditions. Indicates from A subset of remote sensing images extracted from Indicates from The extracted lidar point cloud subset, Indicates from A subset of geological exploration data extracted from it. Indicates from A subset of hydrological monitoring data extracted from These represent the use of a unified raster or grid encoding operator for each data source to convert spatial morphology and attribute features into multi-channel numerical representations. A spatial discrete structure representing a uniform raster or grid is used to ensure that coordinates and resolution are consistent. This indicates that the channel-level and spatial-level fusion operators enable the alignment and synthesis of multi-source features in the same coordinate domain;

[0208] Subsequently, based on the red line constraint and spatial setback rules, a set of geometric constraints is constructed to standardize the layout boundaries, spacing, and connectivity of candidate construction methods, formalized as follows:

[0209] ;

[0210] in Represents a set of geometric constraints. This represents the candidate location vector of the construction method entity. This represents the set of areas that cannot be placed, derived from the boundaries of the restricted area. Indicates the boundary of the non-deployable area. The signified distance from a point to the boundary, with a non-negative value indicating that it lies within the layable domain, represents the distance from the point to the boundary. Represents the construction method entity index. This indicates the center distance or net distance between the construction entities. This represents the minimum spacing threshold. Represents the red line or set of important boundaries. This indicates the setback distance from the construction entity to the boundary. Indicates the minimum setback distance. The entity set and spatial relationships shown in the construction method layout drawing. A value of 1 for the drainage connectivity determination function indicates that the drainage path is continuous and unobstructed. The continuity determination function for the construction access route is set to 1, indicating that the construction access route meets the requirements for continuity and width / slope.

[0211] Furthermore, the numbering thresholds and logical relationships of the normative clauses are mapped to a computable neural symbolic logic constraint model, and the clause satisfaction degree is output given the arrangement and parameters to support the generation of discriminative linkages, denoted as:

[0212] ;

[0213] in Indicates the first The satisfaction level of a standard clause can be in Boolean or numerical form. Indicates that for the first Neural symbolic evaluation mapping of the clauses This represents a continuous parameter matrix that corresponds one-to-one with each construction method entity. Indicates the clause number, This represents the set of parameter thresholds related to the terms. The logical relationships indicating the internal conditions of a clause include AND, OR, NOT, and implied conditions. The scope of application of this clause is limited to the type of construction method, material conditions, spatial location, and working conditions. The satisfaction vector, organized by clause number, is used during the generation and discrimination phases.

[0214] Finally, based on the aligned scene data and previous numerical analysis or field calculation results, a physical surrogate model for rapid calculation of stability seepage and deformation is trained and calibrated to provide an assessment of physical consistency within iterations. Its learning objective can be summarized as follows:

[0215] ;

[0216] in This represents the set of parameters for the proxy model. Indicates the first Physical index prediction mapping on a sample The loss function is used to measure the difference between the prediction and the true value. Indicates the first The scene feature tensor of each sample Indicates the first Construction layout diagram for one sample. Indicates the first The parameter matrix of each sample Indicates the first The true vector of physical indicators for each sample includes a stability safety factor. pore water pressure seepage flow With displacement Indicates the regularization weight. Regularization terms are used to improve generalization ability. This indicates the number of training samples and, after calibration, forms a rapid calculation capability that can be stably generalized under conditions of heavy rainfall, long-duration rainfall, and earthquakes.

[0217] In this specific embodiment, S3 specifically refers to:

[0218] Given the scene feature tensor, the neural symbolic logic constraint model, and the calibrated physical proxy model, the discrete construction method sub-generator of the hierarchical adversarial generator first generates a construction method layout probability map under the constraint conditions and discretizes it to obtain the initial construction method layout map. Then, the continuous parameter sub-generator is called to generate the initial parameter matrix under the layout and constraint conditions. Subsequently, the initial scheme is sent to the compliance branch and physical consistency branch of the adversarial discriminator to calculate the score and perform comprehensive selection.

[0219] Discrete generation can be simply referred to as... ,in The probability map of construction method layout represents the layout probability output on a multi-channel spatial grid. This indicates that the discrete-time method sub-generator is a generating sub-model that outputs a probability graph under given conditions. This indicates that the scene feature tensor is a multi-channel input after multi-source data fusion. The parameter set of the discrete subgenerator represents its trainable weights and biases. This indicates that the neural symbolic logic constraint model is a computable set of canonical clauses;

[0220] The initial layout is obtained by discretizing the probability graph and resolving conflicts. ,in This indicates that the initial construction method layout diagram is a discretized layout of the construction method entities. The discretization and conflict resolution mapping is used to select categories based on thresholds and spatial rules. The inter-value vector represents the selection threshold for each construction method category;

[0221] In obtaining Then, continuous generation can be represented as ,in This indicates that the initial parameter matrix is ​​a continuous set of parameters that corresponds one-to-one with the construction entity. This indicates a sub-model where the continuous parameter sub-generator generates continuous design variables under certain conditions. The parameter set of the continuum generator represents its trainable weights and biases;

[0222] The compliance branch first calculates the clause satisfaction vector using a neural symbolic model. The compliance score is then obtained by linear summarization. ,in The clause satisfaction vector represents the satisfaction results organized by clause number. This indicates that the neural symbolic evaluation mapping is a function that outputs the satisfaction of each clause given a layout and parameters. The compliance score indicates that the compliance score is a quantitative measure of regulatory compliance. This indicates that the weight vector consists of weighting coefficients set according to the importance of the clauses;

[0223] The physical consistency branch normalizes and summarizes the multi-condition physical indices output by the calibrated physical surrogate model to obtain the physical score. ,in The physical score represents a comprehensive measure of stability, seepage, and the reasonableness of deformation. This indicates that the calibrated physical proxy model is a mapping that quickly outputs physical indicators under the target scene and operating conditions. This indicates that the set of calibration parameters is designed to improve accuracy across different scenarios. This indicates that the scoring function is a function that normalizes and weights the physical indicators under multiple operating conditions. This indicates that the scoring weights and the set of normalized parameters represent the scale and weights within the physics branch;

[0224] Given a set of candidate initial schemes obtained from different random perturbations and decoding strategies, the optimal scheme can be selected based on a comprehensive score, as shown in the table below:

[0225] ;

[0226] in Indicates the first The overall score for each candidate solution is a weighted fusion of compliance score and physical score. This indicates that the fusion weight coefficient takes values ​​in the range [range]. Used to balance compliance and physical consistency Indicates the first The compliance scores of each candidate Indicates the first One candidate physics score, This indicates that the candidate solution index set is the set of candidate numbers obtained through sampling. The optimal construction layout and parameter matrix, based on the overall score, are used in the subsequent error correction process. This represents the operator that maximizes the objective function.

[0227] In this specific embodiment, S4 specifically refers to:

[0228] Based on the initial construction method layout diagram With the set of geometric constraints as input, a geometric mask is constructed based on the restricted area, minimum spacing, and boundary setback. The restricted area masking, minimum spacing correction, and boundary setback correction are performed sequentially according to the principle of minimum modification. For construction entities that fall into the undesirable area, position fine-tuning or local trimming is performed. For entities with insufficient spacing, minimum displacement adjustment is performed or low-priority entities are deleted when the requirements cannot be met. For entities with insufficient setback, offset or length trimming is performed. At the same time, the continuity of drainage facilities and construction access paths is checked, and if the requirements are not met, position fine-tuning or local extension is performed within the allowable range of the mask.

[0229] Global geometric correction is expressed as a compact representation of minimal changes back to the feasible region. ,in A diagram showing the construction layout after geometric correction. The projection operator to the geometrically feasible set maps the layout based on the criterion of minimizing the total change. The geometrically feasible set is defined by the forbidden zone constraint, the minimum spacing threshold and the boundary setback distance, and the connectivity requirement. This represents the initial construction layout diagram;

[0230] Connectivity determination uses ,in The function indicating drainage connectivity is 1 indicates that the drainage path is continuous and unobstructed. The function for determining the continuity of the construction access route is set to 1 when the construction access route meets the requirements for continuity and width / slope.

[0231] To support auditing and traceability, record geometric correction vectors. ,in Represents the set of geometric correction vectors. This represents the unique identifier of the construction method entity. This indicates that the position displacement vector contains horizontal and vertical components. Indicates the amount of length cut or extension. The operation type enumeration includes projection, offset, clipping, and deletion. This represents the entity index of the construction method, and the final output is... and As a result of geometric correction.

[0232] In this specific embodiment, S5 specifically includes:

[0233] Construction layout diagram after geometric correction Initial parameter matrix Neural symbolic logic constraint model With a calibrated physical surrogate model as input, while maintaining Under unchanged conditions, the physical proxy model is first invoked for rapid calculation to obtain physical indicators, which are then used for feasible domain construction. This is compactly represented as follows: ,in The physical index vector can include stability safety factor, pore water pressure, seepage flow rate, and displacement, etc. This represents the calibrated physical proxy model mapping. Represents the set of calibration parameters;

[0234] Subsequently, based on the neural symbolic logic constraint model The interpreted clause thresholds and applicable relationships, combined with the verification results, are used to uniformly represent the upper and lower limits, inclination angle, spacing, and longitudinal slope of continuous parameters, forming a linear-interval hybrid expression of the parameter feasible region. ,in Represents the feasible region of continuous parameters. Represents the candidate parameter matrix, This represents the constraint coefficient matrix extracted and linearized from the clause thresholds and applicability relationships. Indicates and The accompanying inequality constraint boundary vectors, and These represent the upper and lower bound matrices of the parameters, respectively.

[0235] In the feasible region The initial parameter matrix is ​​projected using the criterion of minimum modification, denoted as . ,in This represents the parameter matrix after parameter correction. This represents the operator that minimizes the objective function. The L2 norm is used to measure the magnitude of differences. The weighted vector is used to reflect the importance and dimensional scale of each parameter. This indicates that Hadamard element-wise multiplication is used to perform weighted difference;

[0236] For parameter groups with coupling relationships, simultaneous intra-group adjustments are made in the above projection to satisfy linkage constraints and suppress non-physical solutions, and finally, the parameter correction vector is recorded. ,in This represents the set of element differences between the corrected and initial parameter matrices, which is used for subsequent compliance proof chains and outcome compilation.

[0237] In this specific embodiment, S6 specifically refers to:

[0238] The parameter matrix after parameter correction Geometric correction method layout diagram Calibrated physical proxy model and neural symbolic logic constraint model As input, rapid calculations are performed under three types of working conditions: heavy rainfall, long-duration rainfall, and earthquakes, and a set of physical indicators is formed. At the same time, a set of clause satisfaction is calculated according to the clause number to support the subsequent compliance proof chain.

[0239] For ease of description, a set of working conditions is defined. ,in Indicates heavy rainfall conditions, Indicates long-duration rainfall conditions, Indicates earthquake conditions;

[0240] For each working condition The physical index vector for this operating condition is obtained by rapid calculation using a calibrated physical surrogate model. ,in Represents the calibrated physical proxy model mapping, Represents the set of calibration parameters, Indicates working conditions The load and boundary input description includes rainfall intensity-duration, groundwater level and seismic motion parameters, This represents a vector consisting of stability, seepage, and deformation response.

[0241] The results of the three types of working conditions are summarized into a set of physical indicators:

[0242] ;

[0243] in Represents a set of physical indicators, Indicates working conditions Stability and safety factor Indicates working conditions pore water pressure characteristic quantity Indicates working conditions seepage flow rate Indicates working conditions The displacement response;

[0244] Then Together and The satisfaction level of the input neural symbolic logic constraint model is calculated according to the clause number, denoted as . ,in Indicates the degree vector of clause satisfaction. Represents neural symbolic evaluation mapping, Represents the neural symbolic logic constraint model, Indicates the number of normative clauses involved in the assessment, Indicates the first The satisfaction level of a standard clause is used to characterize the degree of satisfaction and the extent of surplus or deviation.

[0245] In this specific embodiment, S7 specifically refers to:

[0246] Taking the set of physical indicators, the set of clause satisfaction, the geometric correction vector, and the parameter correction vector as input, for each regulatory clause, clause-level records are generated sequentially by number and organized into a compliance proof chain, where the set of physical indicators is denoted as the summary set from the previous step. The set of physical indicators used to support the calculation of the clauses may include stability safety factors, pore water pressure, seepage flow, and displacement, etc. The set of clause satisfaction is denoted as a vector. The geometric correction vector represents the degree of satisfaction of each specification clause. This represents the set of spatial changes and operation types for each construction method entity, with the parameter correction vector denoted as... The set of element differences representing the continuous parameters of each construction method entity is denoted as the geometric correction method layout diagram. This indicates the results of the corrective action.

[0247] To construct the record for each clause, physical indices relevant to the calculation of that clause are selected, and spatial location labels and corresponding correction vector subsets are determined, collectively expressed as follows:

[0248] ;

[0249] in Indicates the first Clause-level record of the terms, The record construction mapping is used to combine clause-related fields into a structured object. Indicates the clause number, This indicates that the satisfaction level of the clause is taken from the vector. This indicates that the corresponding physical index reference function is used to retrieve data from a set. One or more physical indicators are selected to participate in the calculation of satisfaction. Spatial location labeling functions are used based on layout diagrams. With entity index set Provide the coordinates or the positioning information of the line segment region. Expressions and Terms The relevant construction method entity index set can be obtained by searching the scope of application of the clauses and the layout diagram. Indicates from Chinese Press Selected subset of geometric correction vectors Indicates from Chinese Press Selected subset of parameter correction vectors The timestamp indicating the generation of the record is used to reflect the order in which they were created.

[0250] All clause-level records are organized chronologically into a chain of proof of compliance, which can be abbreviated as:

[0251] ;

[0252] in Indicates a compliance proof chain, This indicates an operator that sorts by timestamp. Indicates that it is composed of all The set of records generated by the standard clauses, Indicates the number of canonical clauses involved in generating the proof chain;

[0253] Spatial indexes are built based on spatial location annotations to support rapid retrieval and auditing of locations or entities, denoted as... ,in Represents a spatial index structure. The index building operator is used to extract spatial labels from the proof chain and establish a mapping from location to record.

[0254] In this specific embodiment, S8 specifically refers to:

[0255] Using the compliance proof chain, the geometrically corrected construction method layout diagram, and the parameter matrix after parameter correction as input, a repair scheme layout diagram with spatial and compliance annotations is first generated. The location, size, spacing, and boundary setback are marked on the layout diagram according to the construction method category and construction method entity index. At the same time, compliance annotations are formed in the corresponding spatial locations with clause numbers and clause satisfaction to support review and traceability.

[0256] The compliance proof chain is as follows: This represents a sequence of compliant evidence organized chronologically and spatially by clause-level records. The geometrically corrected layout diagram is denoted as follows: This represents the layout result that satisfies the constraints of restricted areas, minimum spacing, setbacks, and connectivity. The parameter matrix after parameter correction is denoted as... Representing a continuous set of parameters corresponding one-to-one with the construction method entity, the generation and compliance annotation of the layout diagram can be compactly represented as follows: ,in The repair plan layout diagram is a spatial result diagram that expresses the type, location, and geometric shape of the construction method entities and superimposes compliant annotations under a unified coordinate system and elevation datum. The generation and annotation mapping is used in Attached to spatial elements The clause number and satisfaction level are located using an entity index;

[0257] Subsequently, a parameter list is generated, which expands the continuous parameters, units, values, and clause mappings of each construction entity corresponding to the layout drawing into an item set, and is compactly represented as follows: ,in This represents a set of parameter list entries. This represents the unique identifier of the construction method entity. Indicates the first The first entity Continuous parameter names such as dimensions, angles, spacing, and longitudinal slope. This indicates that the design value of this parameter is taken from... The specification clause number corresponding to the parameter constraint is used to form a parameter-clause correspondence.

[0258] Finally, a bill of quantities is generated. Based on the geometric dimensions and quantities in the layout drawings and the design values ​​in the parameter list, length, area, volume, and piecework quantity are calculated according to measurement rules, and then categorized and summarized, presented compactly as follows: ,in This indicates that the bill of quantities is a summary result including quantity items, units, and calculation formulas. The quantity calculation and summary mapping is used to transform spatial geometry and parameter values ​​into measurement results and aggregate them according to construction methods and materials. The set of rules for calculating quantities is used to define the measurement standards and rounding principles for length, area, volume, and piecework. The above results are linked to output a repair plan layout diagram, parameter list, and quantity list, and compliance markings are retained in the layout diagram for review and use.

[0259] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

[0260] This invention addresses three key technical challenges: the difficulty of synchronously embedding specifications during the generation phase, insufficient coordination between discrete and continuous parameters, and a lack of physical consistency and interpretability across multiple scenarios including heavy rainfall, long duration, and earthquakes. It establishes an end-to-end closed loop, linking data and specifications through structured alignment, geometric and neural symbolic logic constraint modeling, hierarchical adversarial generation and bi-branch discrimination, two-level correction using geometric mask and parameter projection, rapid multi-scenario calculation using calibrated physical proxies, and a clause-level compliance proof chain output. Firstly, neural symbolic logic constraints take effect simultaneously during generation and discrimination, shifting "post-hoc text verification" to "feedforward computable constraints," significantly reducing rework and improving first-time compliance rates. Secondly, the calibrated physical proxy model constitutes a complete process. The system employs several key technologies: First, it establishes a consistent branch to provide low-latency, cross-condition stability / seepage / deformation feedback for candidate solutions, reducing the proportion of iterations required for high-cost simulations. Second, it utilizes a geometric mask to ensure that restricted areas, minimum spacing, setback boundaries, and connectivity are strictly satisfied spatially. Parameter projection uses a minimum modification criterion to quickly back-project continuous variables into the feasible region, thereby achieving rapid convergence while maintaining drainage connectivity and construction channel continuity. Third, it employs a clause-level compliance proof chain with spatial positioning and time series to make the causal relationship between the solution, clauses, and physical indicators transparent, enhancing the interpretability and traceability of reviews. The overall technical effect is reflected in improved compliance rate, reduced iteration count and simulation cost, and enhanced solution robustness and auditability.

[0261] At the structural level, this case addresses the aforementioned technical issues with targeted improvements to further amplify the technical effects: A hierarchical adversarial generator employing a "discrete method-continuous parameters" approach solves the problems of search oscillations and frequent infeasible solutions caused by the coupling of placement and parameters; a dual-branch discriminator using a "compliance branch + physical consistency branch" quantifies clause satisfaction and physical response in parallel and weighs them in the comprehensive score, avoiding pattern collapse caused by a single objective; and a two-level correction mechanism using a "geometric mask-parameter projection" approach is introduced. The former constrains the spatially deployable domain with topological and geometric rules while maintaining drainage / channel connectivity, while the latter processes parameters through simultaneous projection within the group. The system couples data and minimizes modifications, thus balancing constructability and design stability. Neural symbolic logic constraints are explicitly embedded in generation conditions and loss terms, transforming clause thresholds and logical relationships into computable / differentiable constraint signals. The physical proxy model improves cross-scenario generalization and consistency through multi-condition shared backbone and scenario-based calibration. The compliance proof chain solidifies evidence with a four-dimensional index of "clause-physical indicator-spatial location-timestamp," achieving an auditable closed loop from generation to drawing. These structural enhancements collectively lead to faster convergence of the feasible region, a lower proportion of infeasible candidates, and stronger cross-condition reliability, thereby better achieving the technical effects described in this case.

Claims

1. A slope restoration scheme design method based on a generative adversarial fusion algorithm, characterized in that, include: S1. Obtain scene data of the target slope and the red line of surrounding facilities, perform preprocessing, generate aligned scene data and red line layers, and perform structured parsing of engineering specification clauses to generate a standardized set of specification clauses. S2. Construct a set of geometric constraints based on the red line layer and spatial setback rules, construct a neural symbolic logic constraint model based on the standardized set of normative clauses, extract scene features based on aligned scene data to generate scene feature tensors, and train and calibrate a physical proxy model for rapid calculation of stability, seepage and deformation. S3. First, use the hierarchical adversarial generator to generate the initial construction method layout diagram, then generate the initial parameter matrix under conditions. Calculate the compliance score by combining the compliance branch of the adversarial discriminator with the neural symbolic logic constraint model, and calculate the physical score by combining the physical consistency branch of the adversarial discriminator with the calibrated physical proxy model. Then, score and select the candidate schemes. S4. Generate a geometric mask based on the set of geometric constraints, perform restricted area masking, minimum spacing correction and boundary setback correction on the initial construction method layout drawing, while maintaining drainage connectivity and construction passage continuity, and output the construction method layout drawing and geometric correction vector for geometric correction. S5. Based on the clause threshold of the neural symbolic logic constraint model and the calculation results of the calibrated physical agent model, construct a continuous parameter feasible region, perform a projection operation on the initial parameter matrix with the criterion of minimum modification, and output the parameter matrix and parameter correction vector after parameter correction. S6. Perform rapid calculations under heavy rainfall, long-duration rainfall and earthquake conditions to generate a set of physical indicators including stability safety factor, pore water pressure, seepage flow and displacement, and calculate the set of clause satisfaction based on neural symbolic logic constraint model. S7. Generate clause-level records containing clause number, clause satisfaction, corresponding physical index reference, spatial location label, geometric correction vector and parameter correction vector according to the standard clause number, and organize them into a compliance proof chain in chronological order and spatial index. S8. Generate the repair plan layout diagram, parameter list and bill of quantities.

2. The slope restoration scheme design method based on generative adversarial fusion algorithm according to claim 1, characterized in that, S1 specifically refers to: The scene data includes remote sensing images, lidar point clouds, geological exploration data, and hydrological monitoring data; The coordinates of remote sensing images, lidar point clouds, geological exploration data and hydrological monitoring data are unified to the same plane coordinate system and the same elevation datum, the timestamps are aligned to the same time axis, the spatial resolution is coordinated and the units are standardized, and outlier removal and noise suppression are implemented to ensure data quality. The surrounding facility boundary lines are standardized into vector layers, their coordinates are transformed to the aforementioned plane coordinate system and elevation datum, and topology repair is performed to eliminate geometric breaks and overlaps. The engineering specification clauses are structured and parsed to extract clause numbers, parameter thresholds, logical relationships and scope of application, and then converted into machine-readable standard fields.

3. The slope restoration scheme design method based on generative adversarial fusion algorithm according to claim 1, characterized in that, S2 specifically refers to: Using aligned scene data, redline layers, and a standardized set of specifications as input, multi-source feature extraction is performed on the aligned scene data. The spatial morphological features of remote sensing images and lidar point clouds, along with the attribute features of geological exploration data and hydrological monitoring data, are encoded using a unified raster or grid to generate scene feature tensors that express topography, strata, and hydrological conditions. Based on the red line layer, construct a set of geometric constraints. The set of geometric constraints includes at least the restricted area constraint, the minimum spacing constraint, and the boundary setback constraint, and limits the drainage connectivity and the continuity of the construction passage. A neural symbolic logic constraint model is constructed based on a standardized set of normative clauses. Clause numbers, parameter thresholds, and logical relationships are mapped to computable logical constraints, enabling the logical constraints to output clause satisfaction when subsequent input of construction method layout diagrams and parameter matrices. Based on the aligned scene data combined with historical numerical analysis or on-site calculation results, the physical surrogate model used for rapid calculation of stability, seepage and deformation is trained and calibrated to obtain the calibrated physical surrogate model.

4. The slope restoration scheme design method based on generative adversarial fusion algorithm according to claim 1, characterized in that, S3 specifically refers to: Using scene feature tensor, neural symbolic logic constraint model and calibrated physical proxy model as input, the discrete construction method generator of the hierarchical adversarial generator is invoked to generate a construction method layout probability map under the conditions of the scene feature tensor and the neural symbolic logic constraint model, and the construction method layout probability map is converted into an initial construction method layout map. After obtaining the initial construction method layout diagram, the continuous parameter sub-generator of the hierarchical adversarial generator is called to generate the initial parameter matrix under the conditions of the scene feature tensor, the initial construction method layout diagram and the neural symbolic logic constraint model. The initial construction layout diagram and the initial parameter matrix are input into the compliance branch of the adversarial discriminator, and a compliance score is calculated by combining the neural symbolic logic constraint model. The initial construction layout diagram and the initial parameter matrix are input into the physical consistency branch of the adversarial discriminator, and a physical score is calculated by combining the calibrated physical proxy model. One or more candidate initial schemes are selected based on a comprehensive score of compliance score and physical score.

5. The slope restoration scheme design method based on generative adversarial fusion algorithm according to claim 1, characterized in that, S4 specifically refers to: Using the initial construction layout diagram and geometric constraint set as input, and based on the forbidden zone boundary, minimum spacing threshold and boundary setback distance in the geometric constraint set, a geometric mask is generated to indicate the deployable and non-deployable areas through buffering, offsetting and set operations. The geometric mask is applied to the initial construction method layout drawing, and the forbidden area masking, minimum spacing correction and boundary setback correction are performed in sequence according to the principle of minimum modification. Among them, the construction method entities that fall into the un-layout area are projected to the nearest feasible position or locally clipped. The construction method entities with insufficient spacing are adjusted with minimum displacement and low priority entities are deleted when the spacing cannot be met. The construction method entities with insufficient setback are offset or length clipped. In the revised layout, drainage connectivity and construction passage continuity are checked. If they are not met, the relevant linear entities are finely adjusted or locally extended within the allowable range of the geometric mask according to the principle of minimum modification until the requirements are met. Output the construction method layout diagram and geometric correction vector for geometric correction. The geometric correction vector records the spatial changes and operation types for each construction method entity.

6. The slope restoration scheme design method based on generative adversarial fusion algorithm according to claim 1, characterized in that, S5 specifically refers to: Using the geometrically corrected construction layout diagram, initial parameter matrix, neural symbolic logic constraint model, and calibrated physical proxy model as input, while keeping the geometrically corrected construction layout diagram unchanged, the calibrated physical proxy model is called to quickly calculate the initial parameter matrix to obtain physical indicators. Based on the clause thresholds and applicable relationships interpreted by the neural symbolic logic constraint model, the continuous parameter feasible region composed of interval constraints such as parameter upper and lower limits, minimum or maximum inclination angle, minimum or maximum spacing, minimum longitudinal slope, and inequality constraints is constructed as a constraint set on the initial parameter matrix. Under the continuous parameter feasible region, a projection model with minimum modification as the criterion is established. The parameter matrix with the minimum difference metric from the initial parameter matrix and simultaneously satisfying the constraint set is determined as the parameter matrix after parameter correction. For parameters with coupling relationship, simultaneous projection is used to satisfy the intra-group constraints. Record the parameter correction vector, which is the element difference between the parameter matrix after parameter correction and the initial parameter matrix.

7. The slope restoration scheme design method based on generative adversarial fusion algorithm according to claim 1, characterized in that, S6 specifically refers to: Using the parameter matrix after parameter correction, the construction layout diagram after geometric correction, the calibrated physical surrogate model, and the neural symbolic logic constraint model as input, the calibrated physical surrogate model is invoked to quickly calculate the construction layout diagram after geometric correction and the parameter matrix after parameter correction under heavy rainfall, long-duration rainfall, and earthquake conditions, respectively, to obtain the stability safety factor, pore water pressure, seepage flow, and displacement for each condition. The calculation results of each condition are then summarized to form a set of physical indicators. The physical index set, the geometric correction construction method layout diagram, and the parameter matrix after parameter correction are input into the neural symbolic logic constraint model. The clause satisfaction is calculated according to the standard clause number, and a clause satisfaction set is formed.

8. The slope restoration scheme design method based on generative adversarial fusion algorithm according to claim 1, characterized in that, S7 specifically refers to: Using the set of physical indicators and the set of clause satisfaction, as well as the geometric correction vector and the parameter correction vector, as inputs, each clause of the specification is processed sequentially according to the clause number. For each specification clause, the clause number and clause satisfaction degree are read from the clause satisfaction degree set. One or more physical indicators used to calculate the clause satisfaction degree are selected from the physical indicator set as the corresponding physical indicator reference. The spatial location label corresponding to the clause is determined based on the construction method entity index and spatial change amount recorded in the geometric correction vector and parameter correction vector. The clause number, clause satisfaction degree, corresponding physical indicator reference, spatial location label, geometric correction vector and parameter correction vector are combined to form a clause-level record. All clause-level records are sorted in chronological order of their creation time, and a spatial index is created based on spatial location annotations to organize them into a chain of compliance proofs.

9. The slope restoration scheme design method based on generative adversarial fusion algorithm according to claim 1, characterized in that, S8 specifically refers to: Using the compliance proof chain, the geometrically corrected construction method layout diagram, and the parameter matrix after parameter correction as input, a repair scheme layout diagram is generated based on the geometrically corrected construction method layout diagram. The location, size, spacing, and boundary setback are marked in the repair scheme layout diagram according to the construction method category and construction method entity index. Compliance markings are formed in the corresponding spatial positions based on the clause number and clause satisfaction in the compliance proof chain. The parameter list is generated by taking the parameter matrix after parameter correction as input. The parameter matrix is ​​expanded according to the construction method entity index and the names of continuous parameters and design values ​​are listed. The correspondence with the clause number is recorded. The bill of quantities is generated by taking the repair plan layout drawing and parameter list as input. Based on the geometric dimensions and quantities in the repair plan layout drawing and the design values ​​in the parameter list, the length, area, volume and piecework quantity are calculated using the engineering quantity calculation rules and then summarized.

Images