Geological engineering and surveying and mapping engineering data integrated analysis method

Abstract

The invention provides a geological engineering and surveying and mapping engineering data integrated analysis method, and belongs to the technical field of geological prospecting. Multi-level decomposition expression of a geologic body is realized by constructing a multi-scale geological spatial data model, and a heterogeneous data reference system is unified by adopting a space-time registration algorithm; performing optimization dimension reduction processing by combining a minimum spanning tree algorithm to reduce calculation complexity to O (ElogV), establishing probability association between geological and surveying and mapping data based on a probability graph model, integrating key information by applying an adaptive weight fusion algorithm to form a fusion feature set, and inputting the fusion feature set into a geological surveying and mapping double-domain model to perform deep analysis, a physical error correction module is arranged to ensure result reasonability, a calculated amount precision balance function is introduced to dynamically adjust model parameters, optimal balance is achieved between analysis quality and calculation efficiency, and the problems of calculation resource consumption and real-time performance in massive geological surveying and mapping data processing are effectively solved.

Description

Technical Field

[0001] The present invention belongs to the field of geological exploration technology, and in particular, relates to an integrated analysis method for geological engineering and surveying and mapping engineering data. Background Art

[0002] The fusion analysis of geological engineering and surveying and mapping engineering data has important application value in the fields of mineral resource exploration, large-scale engineering construction, natural disaster monitoring and urban planning. Traditional geological engineering relies on geological drilling, geophysical survey and other means to collect the structural characteristics and physical parameters of underground geological bodies, while surveying and mapping engineering uses advanced technologies such as InSAR, GNSS, and 3D laser scanning to obtain monitoring data such as surface deformation and displacement. In their respective fields, these technologies are relatively mature. For example, geological analysis uses 3D modeling and finite element analysis methods, while the surveying and mapping field uses differential interferometry and time series analysis technology to monitor surface deformation.

[0003] However, with the advancement of data acquisition technology, the fields of geology and surveying and mapping are facing the challenge of processing massive multi-source heterogeneous data. Traditional fusion analysis methods such as simple overlay method and regression analysis method are inefficient in processing high-dimensional and multi-scale geological surveying and mapping big data. Especially in large-scale engineering scenarios, geological drilling points are dense, surveying and monitoring data are highly continuous, and the amount of data is growing exponentially, resulting in huge consumption of computing resources. Existing technologies usually use methods such as downsampling or regional division to reduce the amount of calculation, but these methods often sacrifice data integrity and affect the accuracy of analysis results.

[0004] Therefore, how to effectively process massive geological surveying and mapping data, reduce computing resource consumption, and achieve efficient or even real-time analysis and processing while ensuring the quality of fusion analysis has become a core issue that needs to be solved in current technology. This problem is particularly prominent in application scenarios with high timeliness requirements, such as real-time monitoring and early warning during engineering construction. Summary of the invention

[0005] In view of this, the present invention provides an integrated analysis method for geological engineering and surveying and mapping engineering data, which can solve the problem in the prior art that the amount of data in the fusion analysis of geological engineering and surveying and mapping engineering data is large, resulting in excessive consumption of computing resources and difficulty in real-time processing.

[0006] The present invention is implemented as follows: The present invention provides an integrated analysis method for geological engineering and surveying and mapping engineering data, including: constructing a multi-scale geological spatial data model to perform multi-level decomposition on geological bodies; using a spatio-temporal registration algorithm to perform standardization processing on surveying and mapping data; using a minimum spanning tree algorithm to perform optimization and dimensionality reduction processing on multi-source heterogeneous data, constructing a feature connection network for geological surveying and mapping data, extracting key nodes and paths, and generating a dimensionality reduction parameter set; establishing an association rule between geological body attributes and surface deformation parameters based on a probabilistic graphical model; applying an adaptive weight fusion algorithm to integrate geological internal structure information and surface monitoring results to form a geological surveying and mapping fusion feature set; inputting the geological surveying and mapping fusion feature set into a geological surveying and mapping dual-domain model, dynamically adjusting model parameters in combination with a computational accuracy balance function, performing iterative optimization until convergence, and outputting the integrated analysis result of geological engineering and surveying and mapping engineering.

[0007] Among them, the multi-scale geological spatial data model extracts features of geological bodies at different scales through wavelet transform, constructs a multi-level mathematical expression including geological structure, physical property parameters, and spatial distribution characteristics, and realizes a complete description of geological bodies from the regional to the local.

[0008] Among them, the spatio-temporal registration algorithm is a data alignment method based on control point matching and non-rigid transformation. By identifying common feature points in geological data and surveying and mapping data, establishing a spatial correspondence relationship between data, and realizing the unified expression of data collected at different times through time series analysis.

[0009] Among them, the minimum spanning tree algorithm is a graph theory optimization method. Regarding the feature points of geological surveying and mapping data as the nodes of the graph and the feature similarity as the edge weight, by constructing a subgraph that connects all nodes and has the minimum total weight, the key structure of the data is extracted. The complexity of the minimum spanning tree algorithm is O(ElogV), reducing the computational amount of multi-source heterogeneous data processing while retaining the core topological relationship between data.

[0010] Among them, the dimensionality reduction parameter set includes the feature dimension reduction ratio, node connection threshold, graph structure sparsity, and data compression rate, which are used to quantify the data loss and information retention degree in the dimensionality reduction process. The dimensionality reduction parameter set is used as the input parameter of the computational accuracy balance function.

[0011] Among them, the probabilistic graphical model is a statistical model that uses graph theory to represent the dependence relationship between random variables. By representing geological and surveying and mapping parameters with nodes and the relevance between parameters with edges, establishing the conditional probability distribution between parameters, and realizing the uncertainty reasoning of geological states and surface responses.

[0012] Among them, the adaptive weight fusion algorithm dynamically adjusts the weight coefficients of various types of data in the fusion process according to data quality, reliability, and relevance, ensuring that high-quality data makes a greater contribution to the final analysis result and improving the analysis accuracy.

[0013] Among them, the geological survey fusion feature set is a collection of geological internal structure information and surface monitoring results integrated by an adaptive weight fusion algorithm, including data indicators such as lithological distribution, structural characteristics, physical and mechanical parameters of geological bodies, and surface deformation amounts, displacement gradients, deformation rates, etc. in surveying data, which describe the core characteristics of geological engineering and surveying engineering.

[0014] Among them, the specific structure of the geological survey dual-domain model is an encoder-decoder network based on the Transformer architecture, including two parallel branches: a geological domain encoder and a surveying domain encoder. Each branch consists of multiple self-attention modules, and feature interaction is realized through a cross-domain attention mechanism. The decoder part uses a multi-head attention mechanism to fuse dual-domain features, and the output layer includes a physical error correction module as a sub-module.

[0015] Among them, the geological survey dual-domain model uses an adaptive gating unit to screen features. The total number of parameters of the geological survey dual-domain model is dynamically adjusted according to the complexity of geological data and the density of surveying data. The number of self-attention heads is directly proportional to the complexity of the geological body structure, and the attention dimension is directly proportional to the accuracy level of surveying data.

[0016] Among them, the physical error correction module is a sub-module of the output layer of the geological survey dual-domain model. It is a constraint mechanism constructed based on geological mechanics principles and basic surveying rules. By detecting whether the analysis results violate physical constraint conditions, it corrects data that does not conform to physical laws. The correction intensity of the physical error correction module is determined by the output value of the computational accuracy balance function.

[0017] Among them, the input of the computational accuracy balance function includes the change feature intensity of multi-source data, the confidence of the neural network output, the dimensionality reduction parameter set, and the computational resource constraint index. By comprehensively evaluating the computational requirements and accuracy requirements, it outputs a balance value. When the balance value is less than 0.3, the number of self-attention heads and the correction intensity of the physical error correction module are increased to improve accuracy. When the balance value is greater than 0.7, the number of self-attention layers and the correction intensity of the physical error correction module are reduced to reduce the computational amount.

[0018] Among them, when the balance value is between 0.3 and 0.7, the self-attention parameters and the correction intensity of the physical error correction module are kept unchanged to achieve the dynamic balance of system resources and analysis accuracy.

[0019] Among them, the steps for establishing the training data set of the geological survey dual-domain model include collecting historical data of geological exploration and surveying monitoring in multiple regions, classifying and organizing them according to geological structure types and surveying data types, using the sliding window method to segment time-series data, constructing spatio-temporal registration annotations, establishing the corresponding relationship between geological anomalies and surface deformations using manual annotation and semi-supervised methods, and using data augmentation techniques to expand the training samples.

[0020] The steps for training the geological mapping dual-domain model include first performing self-supervised pre-training on a large-scale geological data set to learn geological feature representations, then performing self-supervised pre-training on mapping data to learn mapping feature representations, followed by dual-domain joint fine-tuning to optimize the parameters of the cross-domain attention mechanism and the physical error correction module. The entire training process adopts a dynamic learning rate strategy to automatically adjust the optimizer parameters according to the performance of the validation set.

[0021] Among them, the data augmentation technology includes random rotation, scaling, and adding noise operations, and finally forms a large-scale training set containing mapping response features under standard geological scenarios.

[0022] The present invention realizes the efficient processing of massive heterogeneous data by constructing a multi-scale geological space data model and a minimum spanning tree data dimensionality reduction processing technology. This method uses a spatio-temporal registration algorithm to unify the data reference system, establishes a quantitative correlation between geological states and surface responses using a probabilistic graph model, integrates key information through an adaptive weight fusion algorithm, and performs in-depth analysis and prediction through a geological mapping dual-domain model. This method effectively solves the problem of low efficiency in processing massive data in traditional technologies. The computational complexity is reduced from O( ) to O(ElogV) through the minimum spanning tree algorithm, significantly reducing the consumption of computing resources. At the same time, through the optimized configuration of the dimensionality reduction parameter set, while compressing the data volume, the key topological structure and feature information are retained to ensure the accuracy of the analysis results. In particular, the introduction of a computational amount-accuracy balance function can dynamically adjust the model parameters according to the application scenario requirements and computing resource constraints, realizing a flexible switch from high-precision analysis to real-time processing. Thus, the present invention solves the technical problem of excessive consumption of computing resources and difficulty in real-time processing due to large amounts of data in the data fusion analysis of geological engineering and mapping engineering, making real-time monitoring, early warning, and decision-making in complex geological engineering environments possible, and greatly improving the efficiency of geological engineering safety management and risk control. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 is a flowchart of the method of the present invention.

[0024] Figure 2 is an effect diagram of spatio-temporal registration processing in Example 2.

[0025] Figure 3 is a comparison diagram of retaining key information in dimensionality reduction processing in Example 2.

[0026] Figure 4 is an effect diagram of adaptive weight fusion in Example 2. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0027] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0028] As Figure 1 shown, it is a flowchart of an integrated analysis method for geological engineering and surveying and mapping engineering data provided by the present invention. The method includes the following steps: S01. Construct a multi-scale geological spatial data model, perform multi-level decomposition on geological bodies, realize the expression of geological information from macro to micro, and establish a mathematical description of the relationship between geological feature attributes and distributions; S02. Use a spatio-temporal registration algorithm to standardize surveying and mapping data, determine a unified spatial coordinate system and time reference, and establish a position correspondence relationship between surveying and mapping data and geological data; S03. Use a minimum spanning tree algorithm to perform optimization and dimensionality reduction processing on multi-source heterogeneous data, construct a connection network for the characteristics of geological surveying and mapping data, extract key nodes and paths, generate a dimensionality reduction parameter set, and realize data compression and retention of key information; S04. Based on a probabilistic graph model, establish an association rule between geological body attributes and surface deformation parameters to realize the quantitative expression of geological states and surface responses; S05. Apply an adaptive weight fusion algorithm to integrate the internal geological structure information and surface monitoring results to form an integrated feature set of geological surveying and mapping, laying a foundation for subsequent analysis; S06. Input the integrated feature set of geological surveying and mapping into a geological surveying and mapping dual-domain model, dynamically adjust the model parameters in combination with a computational accuracy balance function, perform iterative optimization until convergence, and output the final integrated analysis result of geological engineering and surveying and mapping engineering.

[0029] Among them, the multi-scale geological spatial data model extracts features of geological bodies at different scales through wavelet transform, constructs a multi-level mathematical expression including geological structures, physical property parameters, and their spatial distribution characteristics, and realizes a complete description of geological bodies from regions to local areas.

[0030] Among them, the spatio-temporal registration algorithm is a data alignment method based on control point matching and non-rigid transformation. By identifying common feature points in geological data and surveying and mapping data, a spatial correspondence relationship between the data is established, and unified expression of data collected at different times is realized through time series analysis.

[0031] Among them, the minimum spanning tree algorithm is a graph theory optimization method that regards the characteristic points of geological survey data as the nodes of a graph and the feature similarity as the edge weights. By constructing a subgraph that connects all nodes and has the minimum total weight, the extraction of the key structure of the data is realized. The complexity of the minimum spanning tree algorithm is O(ElogV), which reduces the computational amount of multi-source heterogeneous data processing while retaining the core topological relationship between data.

[0032] Among them, the dimensionality reduction parameter set includes the feature dimension reduction ratio, node connection threshold, graph structure sparsity, and data compression rate, which are used to quantify the data loss and information retention degree in the dimensionality reduction process. The dimensionality reduction parameter set serves as the input parameter of the computational amount precision balance function.

[0033] Among them, the probabilistic graphical model is a statistical model that uses graph theory to represent the dependence relationship between random variables. By representing geological and survey parameters with nodes and the correlation between parameters with edges, the conditional probability distribution between parameters is established to realize the uncertainty inference of geological states and surface responses.

[0034] Among them, the adaptive weight fusion algorithm dynamically adjusts the weight coefficients of various types of data in the fusion process according to data quality, reliability, and relevance, ensuring that high-quality data makes a greater contribution to the final analysis result and improving the analysis accuracy.

[0035] Among them, the geological survey fusion feature set is a set integrating the internal geological structure information and surface monitoring results through the adaptive weight fusion algorithm, including data indicators such as the lithology distribution, structural characteristics, physical and mechanical parameters of geological bodies, and the surface deformation amount, displacement gradient, deformation rate, etc. in survey data, which describe the core characteristics of geological engineering and surveying engineering.

[0036] Among them, the specific structure of the geological survey dual-domain model is an encoder-decoder network based on the Transformer architecture, including two parallel branches, the geological domain encoder and the survey domain encoder. Each branch consists of multiple self-attention modules, and feature interaction is realized through the cross-domain attention mechanism. The decoder part uses the multi-head attention mechanism to fuse dual-domain features, and the output layer includes a physical error correction module as a sub-module, which uses an adaptive gating unit to screen features. The total number of parameters of the geological survey dual-domain model is dynamically adjusted according to the complexity of geological data and the density of survey data. The number of self-attention heads is directly proportional to the complexity of the geological body structure, and the attention dimension is directly proportional to the precision level of survey data.

[0037] Among them, the physical error correction module is a sub-module of the output layer of the geological mapping dual-domain model. Based on the constraint mechanism constructed by the principles of geomechanics and the basic laws of surveying, it corrects the data that does not conform to physical laws by detecting whether the analysis results violate the physical constraint conditions. The correction intensity of the physical error correction module is determined by the output value of the computational precision balance function. Strict correction is adopted in scenarios with high-precision requirements, and loose correction is adopted in scenarios with high-efficiency requirements.

[0038] Among them, the inputs of the computational precision balance function include the change feature intensity of multi-source data, the confidence of the neural network output, the dimensionality reduction parameter set, and the computational resource constraint index. By comprehensively evaluating the computational requirements and precision requirements, it outputs a balance value. When the balance value is less than 0.3, the number of self-attention heads and the correction intensity of the physical error correction module are increased to improve precision. When the balance value is greater than 0.7, the number of self-attention layers and the correction intensity of the physical error correction module are reduced to reduce the computational amount. When the balance value is between 0.3 and 0.7, the self-attention parameters and the correction intensity of the physical error correction module are kept unchanged to achieve the dynamic balance of system resources and analysis precision.

[0039] Among them, the steps for establishing the training data set of the geological mapping dual-domain model specifically include collecting historical data of multi-region geological exploration and mapping monitoring, classifying and sorting them according to geological structure types and mapping data types, using the sliding window method to segment time-series data, constructing spatio-temporal registration annotations, establishing the corresponding relationship between geological anomalies and surface deformations by using manual annotation and semi-supervised methods, and adopting data augmentation techniques to expand the training samples, including random rotation, scaling, and adding noise operations. Finally, a large-scale training set containing the mapping response characteristics under standard geological scenarios is formed.

[0040] Among them, the steps for training the geological mapping dual-domain model specifically include first performing self-supervised pre-training on a large-scale geological data set to learn geological feature representations, then performing self-supervised pre-training on mapping data to learn mapping feature representations, and then performing dual-domain joint fine-tuning to optimize the parameters of the cross-domain attention mechanism and the physical error correction module. The entire training process adopts a dynamic learning rate strategy to automatically adjust the optimizer parameters according to the performance of the validation set.

[0041] The following describes the specific implementation manners of the above steps in detail.

[0042] The specific implementation of step S01 is to perform multi-scale decomposition and feature extraction on the geological body through wavelet transform theory. First, a wavelet basis function suitable for the characteristics of the geological body is selected. Commonly used wavelets include Haar wavelet, Daubechies wavelet, or Meyer wavelet, and the selection is determined according to the complexity of the geological body structure. Generally, when the geological structure complexity is relatively high, the Daubechies wavelet is more suitable. Then, the three-dimensional data of the geological body is decomposed by wavelet at different resolutions. Usually, 4 to 5 decomposition levels are set. The first layer represents the macroscopic regional characteristics, and the last layer represents the microscopic local characteristics. Next, key features are extracted from the decomposition results of each layer, including geological structure boundaries, mutation points of physical property parameters, and spatial distribution trends, etc. Subsequently, a multi-level feature matrix is constructed. The matrix elements include spatial position coordinates, geological physical property parameters, and their change rates. The matrix dimension has a linear correspondence with the decomposition level. Finally, a mathematical expression of the spatial distribution of geological features is established. The radial basis function is used to perform continuous interpolation on discrete feature points. When the Gaussian kernel is selected as the interpolation kernel function, the smoothing parameter threshold is usually set between 0.2 and 0.5. A smaller threshold is suitable for geological bodies with strong mutability, and a larger threshold is suitable for geological bodies with strong gradual change. This step realizes the complete feature expression of the geological body from the macroscopic region to the microscopic local, laying a data foundation for subsequent analysis.

[0043] The specific implementation of step S02 is to perform standardization processing on the surveying and mapping data by using a spatio-temporal registration algorithm based on control point matching and non-rigid transformation. First, common feature points in the geological data and the surveying and mapping data are identified, including obvious surface markers, outcrops of geological structures, etc., as control points. The number of effective control points is not less than 3% of the total data points and not less than 12 points. Then, a unified spatial coordinate system is established. Usually, the National Geodetic Coordinate System 2000 or the WGS84 coordinate system is adopted, and all data is transformed to this coordinate system through coordinate transformation. Next, a unified time reference is determined. One of the time points is selected as the reference, and all time-series data is standardized relative to this time point. Subsequently, non-rigid transformation is applied to the surveying and mapping data for spatial registration. When there are more control points, the thin plate spline interpolation method is used to achieve smooth transformation. When there are fewer control points, the affine transformation is used to achieve basic registration. The registration accuracy is evaluated by the root mean square error, and the threshold is usually controlled within 1.5 times of the measurement accuracy. Finally, time registration is performed on the data collected at different times through time series analysis methods. Linear or spline interpolation is used to achieve the unified expression of data with different sampling periods. The time resolution usually remains consistent with the lowest monitoring frequency. This step establishes an accurate spatio-temporal correspondence relationship between the surveying and mapping data and the geological data, ensuring the consistency of data from different sources in subsequent analysis.

[0044] The specific implementation of step S03 is to optimize and reduce the dimension of multi-source heterogeneous data using the minimum spanning tree algorithm. First, each feature point of the geological survey data is regarded as a node of the graph, and the total number of nodes is usually 20% to 40% of the original data points. The sampling method is based on feature significance scoring. Then, the feature similarity between nodes is calculated as the edge weight. The similarity calculation uses cosine distance or Mahalanobis distance, and the weight threshold is set between 0.3 and 0.7. Next, the Kruskal or Prim algorithm is applied to construct the minimum spanning tree, and the subgraph that connects all nodes and has the minimum total weight is selected. The complexity of this algorithm is , where represents the number of edges,[[]] represents the number of nodes; Subsequently, the key nodes and paths in the minimum spanning tree are extracted. The key nodes are defined as the nodes whose connection degree is greater than 1.5 times the average connection degree, and the key paths are defined as the paths that connect two key nodes and whose weight is less than 0.8 times the average weight. Finally, a set of dimensionality reduction parameters is generated, including the feature dimension reduction ratio set to 0.4 to 0.6, the node connection threshold set to 0.35 to 0.65, the sparsity of the graph structure controlled between 0.1 and 0.3, and the data compression ratio set between 60% and 85% according to the application scenario. This step realizes the effective dimensionality reduction of multi-source heterogeneous data and the extraction of key information, significantly reducing the computational amount of subsequent processing while retaining the core topological relationship.

[0045] The specific implementation of step S04 is to establish the association rules between the geological body attributes and the surface deformation parameters based on the probabilistic graph model. First, the node variables of the probabilistic graph are determined, including the internal attributes such as the lithology category, structural characteristics, and physical and mechanical parameters of the geological body, as well as the survey and monitoring parameters such as the surface deformation amount, displacement gradient, and deformation rate. Then, the conditional dependence relationship between nodes is constructed. The Bayesian network or Markov random field is used to establish the probabilistic connection between variables, and the dependence strength threshold is set to 0.25. Next, the conditional probability distribution is learned. The conditional probability table is used for discrete variables, and the Gaussian mixture model is used for continuous variables. The parameter estimation uses the expectation maximization algorithm, and the iteration convergence threshold is set to ; Subsequently, the causal relationship inference between variables is carried out. The causal direction between variables is determined through conditional independence testing, and the significance level is set to 0.05. Finally, a quantitative expression of the geological state and the surface response is established, in the form of conditional probability , indicating the probability distribution of the surface response under the given geological state . This step realizes the quantitative description of the association rules between the internal attributes of the geological body and the surface deformation parameters, providing probabilistic model support for the evaluation and prediction of the geological state.

[0046] The specific implementation of step S05 is to apply the adaptive weight fusion algorithm to integrate the geological internal structure information and the surface monitoring results. First, the quality indicators of different data sources are evaluated, including data integrity, measurement accuracy, spatio-temporal resolution, etc., which are quantified as quality scores with a score range of 0 to 1. Then, the correlation between each data source and the target analysis task is calculated, quantified using mutual information or Pearson correlation coefficient, and the threshold is usually set above 0.4. Next, the reliability of each data source is evaluated, including factors such as equipment accuracy rating and the degree of influence of the measurement environment, which is quantified as a reliability coefficient with a value range of 0.6 to 1. Subsequently, the initial weights are calculated comprehensively based on the quality scores, correlation, and reliability, and the weight coefficients are dynamically adjusted according to the data fusion effect during the iteration process, with an adjustment step size of 0.05 to 0.1. Finally, a geological mapping fusion feature set is generated, including core indicators such as the lithology distribution of geological bodies, structural features, physical and mechanical parameters, and surface deformation amounts, displacement gradients, and deformation rates in the mapping data. This step realizes the adaptive fusion of multi-source data, ensures that high-quality and highly correlated data makes a greater contribution to the final analysis, and improves the representativeness and reliability of the fusion feature set.

[0047] The specific implementation of step S06 is to input the geological mapping fusion feature set into the geological mapping dual-domain model for integrated analysis. First, the fusion feature set is divided into two parts: geological domain data and mapping domain data. Then, the geological domain data is input into the geological domain encoder, and the mapping domain data is input into the mapping domain encoder. Both encoders are composed of multiple self-attention modules, and the number of attention heads is automatically set according to the complexity of the geological body structure, usually ranging from 4 to 12. Next, feature interaction between the two domains is achieved through a cross-domain attention mechanism to learn the association patterns between the domains. Subsequently, a multi-head attention mechanism is applied for feature fusion, and a preliminary analysis result is generated through a decoder. Then, a physical error correction module is used to correct the result. This module constructs constraint conditions based on geological mechanics and surveying principles to correct data that does not conform to physical laws, and the correction intensity is determined by the output value of the computational accuracy balance function. Finally, the model parameters are dynamically adjusted according to the output of the computational accuracy balance function. When the balance value is less than 0.3, the number of attention heads and the physical error correction intensity are increased to improve the accuracy. When the balance value is greater than 0.7, the number of attention layers and the physical error correction intensity are reduced to reduce the computational amount. When the balance value is between 0.3 and 0.7, the model parameters remain unchanged. This step realizes the in-depth fusion analysis of geological engineering and mapping engineering data through the dual-domain model, and outputs an integrated analysis result with physical significance while taking into account computational efficiency and analysis accuracy.

[0048] The detailed structure of the geological survey dual-domain model is designed based on the Transformer architecture, including two parallel branches: the geological domain encoder and the survey domain encoder. The geological domain encoder is responsible for processing the internal geological structure information. It consists of 6 layers of self-attention modules. Each layer contains a multi-head self-attention sub-layer and a feed-forward network sub-layer. A residual connection structure is used between layers, and the LayerNorm method is adopted for normalization. The number of self-attention heads is dynamically set according to the complexity of the geological body. When the complexity score is above 0.8, 12-head attention is used; when the complexity score is between 0.4 and 0.8, 8-head attention is used; when the complexity score is below 0.4, 4-head attention is used. The dimension of each attention head is set to 64, and the hidden layer dimension of the feed-forward network is set to 2048. The survey domain encoder is responsible for processing surface monitoring data. Its structure is similar to that of the geological domain encoder, but the number of layers is usually set to 4 layers. The number of attention heads is related to the density of survey data. When the data density score is greater than 0.7, 10-head attention is used; when the score is between 0.3 and 0.7, 6-head attention is used; when the score is less than 0.3, 3-head attention is used. The cross-domain attention mechanism is located between the two encoders and consists of a bidirectional cross-attention layer to achieve information interaction between domains. The weight matrix is initialized using the Xavier method, and the bias term is initialized as a zero vector. The decoder part consists of 4 layers of multi-head attention modules. Each layer contains a self-attention sub-layer, an encoder attention sub-layer, and a feed-forward network sub-layer. After fusing the dual-domain features, the analysis results are output through a fully connected layer. The output layer contains a physical error correction module as a sub-module. This module constructs a regularization term based on geological mechanics constraints, filters features through an adaptive gating unit, and the gating threshold is dynamically adjusted according to the output of the computational amount-precision balance function to correct prediction results that do not conform to physical laws. The total number of parameters of the entire model is dynamically adjusted. When the geometric mean of the geological data complexity score and the survey data density score is greater than 0.6, the number of parameters is order of magnitude; when the average value is between 0.3 and 0.6, the number of parameters is order of magnitude; when the average value is less than 0.3, the number of parameters is order of magnitude. The number of self-attention heads is directly proportional to the complexity of the geological body structure. For every 0.1 increase in complexity, the number of heads increases by 1; the attention dimension is directly proportional to the accuracy level of survey data. For every one-level increase in the accuracy level, the dimension increases by 16.

[0049] The detailed steps for establishing the training dataset of the geological survey dual-domain model are as follows. First, collect historical data of multi-region geological exploration and survey monitoring, covering different types of geological structures, including fold structures, fault structures, magmatic rock bodies, etc., and different types of survey data, including leveling surveys, GNSS surveys, InSAR monitoring, etc.; then classify and organize them according to the geological structure type and survey data type, and construct a hierarchical database structure. Each type of data should contain at least 50 typical cases; then use the sliding window method to segment the time-series data. The window length is set according to the characteristic time scale of the geological process. The window for the rapidly changing process is set to 7 to 15 days, and the window for the slowly changing process is set to 30 to 90 days. The window sliding step size is 25% to 50% of the window length; subsequently, construct spatio-temporal registration annotations, select surface feature points and underground reference points to establish corresponding relationships, and the density of registration control points should not be less than each point; then use the manual annotation by professionals and semi-supervised methods to establish the corresponding relationship between geological anomalies and surface deformations, and ensure that the annotation accuracy is not less than 90% through cross-validation; then use data augmentation techniques to expand the training samples, including random rotation operations, with the rotation angle range of , scaling operations, with the scaling factor range of 0.8 to 1.2, adding noise operations, with the signal-to-noise ratio controlled above 15 dB. Through these operations, the original samples are expanded to 3 to 5 times; finally, a large-scale training set containing the survey response characteristics under standard geological scenarios is formed, and the scale of the training set is not less than 10,000 sample pairs. The detailed steps for training the geological survey dual-domain model are as follows. First, perform self-supervised pre-training on a large-scale geological dataset, use the masked auto-encoding method to learn geological feature representations, and set the masking ratio to 15% to 25%. The pre-training lasts for 100 to 200 rounds; then perform self-supervised pre-training on the survey data, also using the masked auto-encoding method to learn survey feature representations, and set the masking ratio to 10% to 20%. The pre-training lasts for 80 to 150 rounds; then perform dual-domain joint fine-tuning, use the geological survey paired data as input, optimize the parameters of the cross-domain attention mechanism and the physical error correction module, and the fine-tuning lasts for 50 to 100 rounds; the entire training process adopts a dynamic learning rate strategy, set the initial learning rate to , decay according to the cosine annealing strategy, evaluate the performance of the validation set every 10 rounds. If the performance does not improve for 3 consecutive rounds, the learning rate is reduced to 0.5 times the original, and the minimum learning rate is not less than ; select the Adam algorithm as the optimizer, parameter is set to 0.9, parameter is set to 0.999, and the weight decay coefficient is set to ; adopt the early stopping strategy during the training process. If the performance of the validation set does not improve for 10 consecutive rounds, stop the training, and finally select the model parameters with the best performance on the validation set.

[0050] It should be noted that the present invention uses the minimum spanning tree algorithm to optimize and reduce the dimension of multi-source heterogeneous data in geological engineering and surveying and mapping engineering. The characteristic points of geological surveying and mapping data are regarded as the nodes of the graph, and the feature similarity is used as the edge weight. By constructing a subgraph that connects all nodes and has the minimum total weight, the extraction of the key structure of the data is realized. Compared with traditional data dimension reduction methods such as principal component analysis or linear dimension reduction techniques, the minimum spanning tree algorithm has the advantage of preserving the topological relationship of the data, and can maintain the core structural association between the data while reducing the data dimension, effectively avoiding information loss. The computational complexity of this algorithm is reduced from O( ) to O(ElogV), which is suitable for the efficient processing of massive geological surveying and mapping data. At the same time, by adjusting parameters such as the node connection threshold and the sparsity of the graph structure, the degree of dimension reduction and the information retention ratio can be flexibly controlled, achieving a good balance between data compression and feature retention.

[0051] The geological surveying and mapping dual-domain model designed by the present invention is based on the Transformer architecture, and includes two parallel branches, namely the geological domain encoder and the surveying and mapping domain encoder, and realizes feature interaction through the cross-domain attention mechanism. Compared with traditional data fusion models, this dual-domain model fully considers the heterogeneous characteristics of geological data and surveying and mapping data, retains the unique expressions of their respective domain knowledge through domain-specific processing, and the cross-domain attention mechanism can capture the deep associations between the two types of data. This model uses the self-attention mechanism to screen important features, avoiding indiscriminate processing of all data, and greatly improving the computational efficiency; at the same time, the physical error correction module in the output layer constructs a constraint mechanism based on the principles of geomechanics and the basic laws of surveying, ensuring that the analysis results conform to physical laws and avoiding additional correction calculations caused by unreasonable model outputs.

[0052] The computational amount and accuracy balance function introduced by the present invention is an innovative mechanism for dynamically adjusting the system performance. By comprehensively evaluating the change feature intensity of multi-source data, the confidence of the neural network output, the dimension reduction parameter set, and the computational resource constraint index, it outputs a balance value to guide the adjustment of the model parameters. Different from traditional fixed-parameter analysis methods, this function can be adaptively adjusted according to the application scenario requirements and resource conditions. When the accuracy requirement is high, it increases the number of self-attention heads and the correction intensity of the physical error correction module, and when the efficiency requirement is high, it reduces the computational parameters. This dynamic mechanism enables the system to flexibly switch between different working modes, realizing a smooth transition from high-precision analysis to real-time processing, and effectively solving the technical problem of difficult to balance accuracy and efficiency in the processing of massive geological surveying and mapping data.

[0053] The above three core technical ideas form a complete technical chain from data processing to model construction and then to system optimization, generating a comprehensive advantage that exceeds individual technologies through synergistic effects. The minimum spanning tree algorithm achieves preliminary dimensionality reduction and information screening at the data input level, providing more concise and high-quality data for subsequent processing; the geological mapping dual-domain model makes full use of the data characteristics after dimensionality reduction, extracting key information through domain-parallel processing and cross-domain interaction; and the adaptive computational load precision balance function runs through the entire analysis process, dynamically adjusting system parameters to adapt to different scenario requirements. This multi-level optimization strategy enables the system to significantly reduce computational resource consumption while maintaining analysis quality when faced with massive heterogeneous data, and at the same time flexibly adjust the working mode according to application requirements, achieving efficient processing and precise analysis of the integrated analysis of geological engineering and surveying engineering data, providing strong technical support for real-time monitoring, early warning, and decision-making in engineering practice.

[0054] Specifically, the principle of the present invention is as follows: The integrated analysis method for geological engineering and surveying engineering data proposed by the present invention has its core principle in achieving efficient processing and precise analysis of massive heterogeneous data through hierarchical data representation, optimized dimensionality reduction processing, and an adaptive computing framework.

[0055] First, at the data representation level, this method uses a multi-scale geological spatial data model to perform multi-level decomposition of geological bodies through wavelet transform, achieving hierarchical representation from macro to micro. This multi-scale representation method conforms to the internal structural characteristics of geological bodies, can flexibly select information at an appropriate scale according to analysis requirements, and avoid the computational burden caused by simultaneous processing of full-scale data. At the same time, a unified coordinate system and time reference are established through a spatio-temporal registration algorithm, solving the prerequisite for heterogeneous data fusion and laying a foundation for subsequent analysis.

[0056] Secondly, at the data processing level, this method uses the minimum spanning tree algorithm to perform optimized dimensionality reduction on multi-source heterogeneous data. By treating the characteristic points of geological mapping data as nodes of a graph and the feature similarity as the edge weight, a subgraph that connects all nodes and has the minimum total weight is constructed. This processing process effectively extracts the key topological relationships between data, greatly reduces the data dimension, and the computational complexity is reduced from O( ) to O(ElogV), significantly improving the processing efficiency. At the same time, the association rules between the attributes of geological bodies and the surface deformation parameters are established through a probabilistic graphical model, converting the massive data in the two fields into a probabilistic distribution representation, further compressing the data representation space.

[0057] At the model construction level, this method designs a geological mapping dual-domain model based on the Transformer architecture. The domain features are extracted by the parallel geological domain encoder and mapping domain encoder respectively, and then information interaction is realized through the cross-domain attention mechanism. This structural design enables the model to focus on the most relevant features, avoiding indiscriminate processing of all data and significantly improving the computational efficiency. In particular, the physical error correction module in the output layer ensures that the analysis results conform to the basic laws of geomechanics and surveying, effectively improving the reliability of the results while avoiding unnecessary iterative calculations.

[0058] Most innovatively, this method introduces a computational complexity-precision balance function. This function comprehensively considers data features, model confidence, dimensionality reduction parameters, and computational resource constraints, and dynamically adjusts the model complexity and physical error correction intensity. When the balance value is less than 0.3, the system gives priority to ensuring analysis accuracy; when the balance value is greater than 0.7, the system gives priority to ensuring computational efficiency; when the balance value is between 0.3 and 0.7, the system keeps the parameters unchanged. This adaptive mechanism enables the system to flexibly adjust according to the actual application scenario and resource conditions, achieving the best balance between high-precision analysis and real-time processing.

[0059] In addition, this method adopts a large-scale training dataset and a phased training strategy. By combining self-supervised pre-training and dual-domain joint fine-tuning, the inherent laws in geological mapping data are fully exploited. The dynamic learning rate strategy is adopted in the model training process, and the optimizer parameters are automatically adjusted according to the performance of the validation set, ensuring the robustness and generalization ability of the model when dealing with massive data, and further improving the performance of the system in practical applications.

[0060] A specific embodiment 1 of the present invention is provided below. The specific implementation of each step in this embodiment 1 is described in detail as follows.

[0061] The specific implementation of step S01 is to construct a multi-scale geological spatial data model. The geological body is decomposed at multiple levels through wavelet transform to achieve multi-scale expression of geological information. First, a wavelet basis function suitable for the characteristics of the geological body is selected. For example, the Haar wavelet is selected when the geological structure boundary is obvious, and the Daubechies wavelet is selected when the geological body structure has a smooth transition. Then, the geological data in three-dimensional space is subjected to multi-scale wavelet decomposition, which is specifically expressed as: ; where, is the geological body spatial distribution function; is the decomposition scale, is the maximum decomposition scale, usually taking 4 to 5; is the spatial position index; is the wavelet detail coefficient; is the approximation coefficient on the scale ; is the wavelet basis function; is the scaling function.

[0062] Based on the multi-scale decomposition results, construct the geological body feature vector , representing the geological features at different scales: ; In the formula, is the scale feature vector at; is the th feature component, including geological structure, physical property parameters, etc.; is the feature dimension, which decreases as the scale increases, , where is the original feature dimension.

[0063] To express the spatial distribution relationship of geological features, a continuous mathematical model is constructed using a radial basis function: ; In the formula, is the geological feature interpolation function; is the weight coefficient; is the spatial position vector; is the position of the known feature point; is the radial basis function, usually the Gaussian kernel function ; is the smoothing parameter, and its value range is from 0.2 to 0.5.

[0064] The weight coefficient is obtained by solving a system of linear equations: ; In the formula, is the coefficient matrix, ; is the weight vector; is the target value vector. This step realizes the complete description of the geological body from macro to micro through multi-scale decomposition and spatial interpolation, providing a data basis for subsequent analysis.

[0065] The specific implementation method of step S02 is to standardize the surveying and mapping data using a spatio-temporal registration algorithm. First, identify the common feature points in the geological data and the surveying and mapping data, and establish the control point sets and , where and are the numbers of control points in the geological data and the surveying and mapping data respectively. Then establish a spatial coordinate transformation model, using affine transformation or non-rigid transformation, specifically expressed as: ; In the formula, is the original coordinate; is the coordinate after transformation; is the rotation and scaling parameter; is the translation parameter.

[0066] When the number of control points is large and irregularly distributed, thin plate spline interpolation is used to achieve non-rigid transformation: ; In the formula, is the transformation function; is the global affine transformation parameter; is the weight coefficient; is the basis function; is the position of the point to be transformed; is the position of the control point.

[0067] For time series data, an expression under a unified time reference is established: ; In the formula, is the time surveying and mapping data at the moment; is the surveying and mapping data at the reference time moment; is the rate of change function.

[0068] For discrete sampling data, interpolation method is used to achieve unified time expression: ; In the formula, is the time interpolation coefficient, satisfying ; is the time point data collected. The registration quality is evaluated using root mean square error: ; In the formula, is the registration root mean square error; is the transformation function; is the source data point; is the target data point; is the number of evaluation points. The registration accuracy threshold is usually set within 1.5 times of the measurement accuracy. This step establishes the position correspondence between the surveying and mapping data and the geological data, laying a foundation for the subsequent fusion analysis.

[0069] The specific implementation of step S03 is to use the minimum spanning tree algorithm to optimize and reduce the dimension of multi-source heterogeneous data. First, the feature points of the geological surveying and mapping data are regarded as the graph The nodes, node set , edge set . Then calculate the feature similarity between nodes as the edge weight: ; In the formula, is the weight of edge ; is the similarity between nodes and ; and are the feature vectors of nodes and respectively; represents the vector norm.

[0070] Then apply the Kruskal algorithm to construct the minimum spanning tree : 1. Sort all edges in ascending order of weight; 2. Initialize an empty tree ; 3. Examine each edge in the sorted order . If adding does not form a cycle, then ; 4. When contains edges, the algorithm terminates.

[0071] The complexity of this algorithm is , where is the number of edges and is the number of nodes. Based on the minimum spanning tree, extract the key nodes and paths. The key nodes are defined as: ; In the formula, is the key node set; is the degree of node ; is the average node degree; is a coefficient, usually taking the value of 1.5.

[0072] The key path is defined as: ; In the formula, is the key path set; is the average edge weight; is a coefficient, usually taking the value of 0.8.

[0073] The finally generated dimensionality reduction parameter set includes: ; In the formula, is the feature dimension reduction ratio, and its value range is from 0.4 to 0.6; is the node connection threshold, and its value range is from 0.35 to 0.65; is the sparsity of the graph structure, and its value range is from 0.1 to 0.3; is the data compression ratio, and its value range is from 60% to 85%. This step realizes the dimensionality reduction of multi-source heterogeneous data through graph theory optimization, reduces the computational complexity while retaining key information.

[0074] The specific implementation of step S04 is to establish the association rules between the geological body attributes and the surface deformation parameters based on the probabilistic graph model. First, construct a probabilistic graph , node set , where represents geological parameters, represents surveying and mapping parameters. Then establish the conditional dependence relationship between nodes and construct a Bayesian network or a Markov random field. For the Bayesian network, the conditional probability is expressed as: ; In the formula, is the joint probability distribution; is the conditional probability of node with respect to its parent node .

[0075] For the Markov random field, the joint probability distribution is expressed as: ; In the formula, is the normalization factor; is the set of maximum cliques in the graph; is the clique on the potential function; is the clique in the node set.

[0076] For continuous variables, the Gaussian mixture model is used to represent the conditional probability: ; In the formula, is the conditional probability of the surface response under the given geological state ; is the weight of the th mixture component, satisfying ; is a Gaussian distribution with a mean of and a covariance matrix of ; is the number of mixture components.

[0077] Mean function Expressed as a function of geological parameters: ; In the formula, is the weight matrix; is the bias vector.

[0078] Parameter estimation uses the expectation - maximization algorithm, and the iterative process is as follows: 1. E - step: Calculate the posterior probability ; 2. M - step: Update the parameters , , and .

[0079] Iterate until convergence, and the convergence threshold is set to . This step establishes a probability correlation model between the geological body attributes and the surface deformation parameters, and realizes the quantitative expression of the geological state and the surface response.

[0080] The specific implementation of step S05 is to apply the adaptive weight fusion algorithm to integrate the geological internal structure information and the surface monitoring results. First, evaluate the quality indicators of different data sources and construct a quality score vector , . Then calculate the correlation between each data source and the target task: ; In the formula, is the correlation index of the th data source; is the mutual information between the data source and the target variable ; is the joint probability distribution; and are the marginal probability distributions respectively.

[0081] For reliability assessment, construct a reliability coefficient vector , . Based on the above three indicators, calculate the initial fusion weights: ; In the formula, is the initial weight of the th data source.

[0082] During the iterative process, adjust the weights dynamically according to the fusion effect: ; In the formula, is the weight of the th data source at the The weight of each data source; is the learning rate, usually taking values from 0.05 to 0.1; is the fusion effect evaluation function; is the weight of the gradient of the evaluation function.

[0083] The finally generated fused feature set is expressed as: ; In the formula, is the fused feature set; is the th feature set of the data source; is the finally determined weight. This step realizes the adaptive fusion of heterogeneous data and ensures the leading role of high-quality data in the analysis results.

[0084] The specific implementation of step S06 is to input the geological surveying and mapping fused feature set into the geological surveying and mapping dual-domain model for integrated analysis. First, the fused feature set is divided into geological domain data and surveying and mapping domain data . Then, they are respectively input into the corresponding encoders for feature extraction: ; ; In the formula, and are the encoded features of the geological domain and the surveying and mapping domain respectively; and are the geological domain encoder and the surveying and mapping domain encoder respectively.

[0085] The encoder is based on the Transformer architecture, and the self-attention mechanism is expressed as: ; In the formula, , , are the query matrix, key matrix, and value matrix respectively; is the dimension of the key vector.

[0086] The multi-head attention mechanism is expressed as: ; ; In the formula, is the number of attention heads; , , and are the parameter matrices.

[0087] Feature interaction between two domains is achieved through a cross-domain attention mechanism: ; ; In the formula, represents the attention feature of the geological domain to the surveying and mapping domain; represents the attention feature of the surveying and mapping domain to the geological domain.

[0088] Then, the dual-domain features are fused through a decoder: ; In the formula, is the preliminary analysis result output by the decoder.

[0089] Finally, the result is corrected through a physical error correction module: ; In the formula, is the corrected analysis result; is the correction intensity coefficient; is the correction function based on physical constraints.

[0090] Correction intensity coefficient is determined by a computational complexity-accuracy balance function: ; In the formula, is the intensity of multi-source data variation features; is the confidence of the neural network output; is the dimensionality reduction parameter set; is the computational resource constraint index.

[0091] The balance value output by the balance function determines the model parameter adjustment strategy: ; In the formula, is the balance value, and its value range is [0, 1]; is the sigmoid function; is the weight coefficient; is the bias term.

[0092] When , increase the number of attention heads: , where ; When , reduce the number of attention layers: , where ; When , keep the model parameters unchanged.

[0093] This step realizes the in-depth fusion analysis of geological engineering and surveying and mapping engineering data through a dual-domain model, taking into account both analysis accuracy and computational efficiency.

[0094] The detailed structure of the geological surveying and mapping dual-domain model is designed based on the Transformer architecture, including two parallel branches: the geological domain encoder and the surveying and mapping domain encoder. The geological domain encoder is responsible for processing the internal geological structure information and consists of multiple self-attention modules. Each layer contains a multi-head self-attention sub-layer and a feed-forward network sub-layer. The calculation process of the self-attention mechanism is as follows: ; ; In the formula, is the input feature of the geological domain; , , are parameter matrices; is the dimension of the key vector.

[0095] The multi-head attention mechanism is expressed as: ; ; In the formula, is the number of attention heads, which is related to the complexity of the geological body structure : , ; , , and are parameter matrices.

[0096] The feed-forward network sub-layer is expressed as: ; In the formula, , , , are parameter matrices and bias vectors.

[0097] The structure of the surveying and mapping domain encoder is similar to that of the geological domain encoder, but the parameter settings are different. The cross-domain attention mechanism realizes the feature interaction between the two domains: ; ; ; In the formula, and are the encoded features of the geological domain and the surveying and mapping domain, respectively.

[0098] The decoder part consists of multiple attention modules. After fusing the dual-domain features, the analysis results are output through a fully connected layer. The physical error correction module constructs constraint conditions based on the principles of geomechanics and corrects the prediction results that violate physical laws: ; where is the physical constraint loss; is the th physical constraint function; is the constraint weight; is the number of constraint conditions.

[0099] The correction process is expressed as: ; where is the corrected result; is the correction intensity, determined by the output value of the computational amount accuracy balance function; is the gradient of the physical constraint loss with respect to the output.

[0100] The detailed steps for establishing the training dataset of the geological mapping dual-domain model include collecting historical data on geological exploration and mapping monitoring in multiple regions and classifying and organizing them according to geological structure types and mapping data types. Then, the sliding window method is used to segment the time-series data: ; ; where is the input time window; is the output time window; is the window length; is the prediction length; is the starting time index.

[0101] By constructing spatio-temporal registration annotations, the corresponding relationship between geological anomalies and surface deformations is established. Finally, data augmentation techniques are used to expand the training samples, including operations such as random rotation, scaling, and adding noise: ; where is the augmented sample; is the rotation matrix, is the rotation angle; is the scaling matrix, is the scaling factor; is the Gaussian noise.

[0102] The entire training process includes two stages: self-supervised pre-training and supervised fine-tuning. Self-supervised pre-training uses the masked auto-encoding method to learn feature representations: ; ; Wherein, is the input after adding the mask; is the reconstructed output; is the reconstruction loss.

[0103] Optimize the parameters of the cross-domain attention mechanism and the physical error correction module during the supervised fine-tuning stage: ; Wherein, is the fine-tuning loss; is the prediction loss; is the physical constraint loss; is the balance coefficient.

[0104] The entire model training process adopts a dynamic learning rate strategy, and the learning rate update rule is: ; Wherein, is the learning rate of the th round of iteration; is the initial learning rate; is the total number of iteration rounds. Automatically adjust the optimizer parameters through the performance of the validation set, and finally select the model parameters with the best performance.

[0105] Next, some common formulas in the existing technologies involved in this embodiment will be explained.

[0106] Optionally, in step S02, the rate of change function : ; Wherein, is the rate of change at time ; is the exponential decay coefficient, and its value range is from 0.001 to 0.01; is the decay constant, and its value range is from 0.1 to 0.5; is the periodic change amplitude, and its value range is from 0.005 to 0.02; is the angular frequency; is the initial phase; is the reference time.

[0107] Among them, the parameter acquisition method is: Obtained by fitting historical data, including step 1: Collecting the change data of the same type of geological body in the past 10 years; step 2: Using the least squares method to fit the exponential trend term. Obtained by spectral analysis, including step 1: Performing Fourier transform on the time series data; step 2: Extracting the amplitude of the main frequency component as value.

[0108] Optionally, in step S03, the computational precision balance function : ; In the formula, is the output value of the balance function, and its value range is [0, 1]; , , , are the weight coefficients; is the bias term; is the intensity of multi-source data variation characteristics; is the confidence of the neural network output; is the comprehensive score of the dimensionality reduction parameter set; is the computational resource constraint index.

[0109] Among them, the method for obtaining parameters is as follows: Obtained by statistical analysis, including step 1: calculating the coefficient of variation of each data source within the time window; step 2: performing weighted average on the coefficient of variation to obtain the value. Obtained by system monitoring, including step 1: real-time monitoring of CPU and memory usage; step 2: calculating the constraint intensity according to the resource usage, .

[0110] Optionally, in step S04, the potential function : ;

[0111] In the formula, is the potential function on the clique ; is the interaction intensity between nodes and ; is the incompatibility measure between nodes, defined as , where is a small constant to prevent division by zero.

[0112] Normalization factor : ; In the formula, is the normalization factor to ensure the normalization condition of the probability distribution; the summation traverses all possible variable configurations ; is the set of maximum cliques in the graph.

[0113] Optionally, in step S05, the fusion effect evaluation function : ; In the formula, is the comprehensive evaluation function; is the mean square error; is the KL divergence; is the consistency index; , , is the weight coefficient; is the correlation of the th data source;

[0114] Gradient function : ; In the formula, ; ; .

[0115] Optionally, in step S06, the physical constraint correction function : ; In the formula, is the correction function based on physical constraints; is the number of constraint conditions; is the correction intensity of the th constraint, and the value range is from 0.1 to 0.5; is the sensitivity parameter of the th constraint, and the value range is from 1 to 5; is the th component of the output; is the reference value determined based on physical laws.

[0116] Physical constraint function : ; In the formula, is the th physical constraint function; is the mathematical expression of the th physical law, including the stress balance condition , the deformation continuity condition and so on; is the tolerance threshold of the th constraint; is the stress tensor; is the density; is the gravitational acceleration; is the displacement vector.

[0117] Among them, the parameter acquisition method is: Obtained by cross-validation, including step 1: Testing the correction effects of different values on the validation set; step 2: Selecting the one that makes the comprehensive index optimal Value Obtained by statistical analysis, including Step 1: Collect the distribution of the degree of violation of physical constraints in historical data; Step 2: Take the 95% quantile as the tolerance threshold . The value range of is from 1 to 5. To better understand and implement the present invention, Example 2 of a specific application scenario of the present invention is provided below: In a mountain tunnel project, researchers applied the integrated analysis method of geological engineering and surveying and mapping engineering data to evaluate and predict the geological risks during the tunnel construction process. The tunnel is 3.5 kilometers long and passes through various complex geological environments, including fault zones, karst areas, and high in-situ stress areas, with relatively high construction safety risks. To predict possible geological disasters in advance and take effective measures, researchers collected multi-source heterogeneous data, including geological drilling data, geophysical survey results, and surface deformation monitoring data, and conducted integrated analysis.

[0118] First, the researchers constructed a multi-scale geological spatial data model according to Step S01. Based on the geological data obtained by drilling, wavelet transform multi-scale decomposition was performed on the geological bodies along the tunnel. The Daubechies-4 wavelet basis function was used, and the decomposition level was set to 4 layers. Table 1 shows the expression accuracy of geological features at different decomposition levels.

[0119] Table 1 Wavelet transform multi-scale decomposition accuracy table

[0120] During the radial basis function interpolation process, the smoothing parameter β was set to 0.35. After interpolation calculation, a continuous geological feature distribution function was obtained, which can accurately express geological structure features such as fault zones and karst caves. The multi-scale geological spatial data model can express geological information from macro to micro levels, providing a data basis for subsequent risk analysis.

[0121] Subsequently, the researchers performed the spatio-temporal registration process of Step S02. 42 control points were arranged along the tunnel and its surroundings, including 18 surface markers and 24 underground reference points, as shown in Table 2: Table 2 Spatio-temporal registration control point distribution table

[0122] The thin plate spline interpolation method was used for spatial registration, and the registration accuracy RMSE value was 3.2 mm, which is lower than the 1.5-fold threshold of the measurement accuracy, 4.5 mm. In terms of time registration, a monitoring frequency with a period of 7 days was used as the standard time reference to unify the surveying and mapping data collected at different times. As Figure 2 shown, the spatio-temporal registration process significantly improved the consistency of multi-source data, laying a foundation for subsequent analysis.

[0123] In step S03, the researchers performed dimensionality reduction on the registered multi-source heterogeneous data using the minimum spanning tree algorithm. 30% of the feature points were extracted from the original data as graph nodes, and the total number of nodes was 1286. The feature similarity between nodes was calculated using cosine distance, and the similarity threshold was set to 0.45. The Kruskal algorithm was applied to construct the minimum spanning tree, as shown in Table 3: Table 3 Dimensionality reduction effect of the minimum spanning tree

[0124] 37 key nodes and 52 key paths were extracted from the minimum spanning tree as key indicators for geological risk assessment. The dimensionality reduction parameter set was {0.52, 0.45, 0.18, 78%}, that is, the feature dimension reduction ratio was 0.52, the node connection threshold was 0.45, the sparsity of the graph structure was 0.18, and the data compression rate was 78%. As Figure 3 shown, the dimensionality reduction process retained the key geological structure information while significantly reducing the computational amount.

[0125] In step S04, a probabilistic graph model was constructed based on the dimensionality-reduced data to establish the association rules between the geological body attributes and the surface deformation parameters. The selected key geological parameters included 15 variables such as rock mass strength, fault dip angle, and groundwater level; the surveying and monitoring parameters included 12 variables such as surface settlement, horizontal displacement, and tilt angle. A Bayesian network was used to construct a conditional probability model, and the dependence strength threshold was set to 0.28. The number of Gaussian mixture model components was set to 3, and the parameters were trained using the expectation maximization algorithm, and the iteration convergence threshold was , and finally the conditional probability distribution was obtained, as shown in Table 4: Table 4 Association probability table between geological state and surface response

[0126] In step S05, the researchers applied the adaptive weight fusion algorithm to integrate the geological structure information and the monitoring results. The quality scores, correlations, and reliabilities of different data sources were evaluated, and the initial weight distribution was as follows: geological drilling data 0.35, geophysical exploration data 0.28, InSAR monitoring data 0.22, GNSS monitoring data 0.15. During the iterative fusion process, the weights were dynamically adjusted according to the data performance, and finally the weights were adjusted to: geological drilling data 0.32, geophysical exploration data 0.23, InSAR monitoring data 0.28, GNSS monitoring data 0.17. The adjustment step size was set to 0.08, and the weights tended to be stable after 5 iterations. As Figure 4 shown, the adaptive weight fusion significantly improved the accuracy of risk prediction.

[0127] Finally, in step S06, the fused feature set is input into the geological mapping dual-domain model for integrated analysis. The model adopts the Transformer architecture. The geological domain encoder uses an 8-layer self-attention structure with 10 attention heads, and the mapping domain encoder uses a 6-layer structure with 8 attention heads. The total number of parameters of the model is . The physical error correction module is based on the constraints of geomechanics. The correction strength is determined by the computational accuracy balance function, and the balance value is 0.45, which belongs to the equilibrium state. Through the analysis of this model, 3 potential risk areas during tunnel construction are accurately predicted, including the fault fracture zone at K1+250, the karst cave group at K2+420, and the high in-situ stress area at K3+180. The predicted settlement amounts are 28.5mm, 35.7mm, and 21.3mm respectively.

[0128] Traditional tunnel geological risk assessment mainly relies on geological exploration and empirical judgment, suffering from problems such as isolated information, low data fusion degree, and insufficient prediction accuracy. Usually, single-domain models are used to process geological data and mapping data separately, and then the analysis results are manually integrated, resulting in information loss and subjective deviation. In this embodiment, the settlement amounts of the three risk areas predicted by the traditional method are 33.2mm, 41.5mm, and 16.8mm, with differences of 18.7%, 13.2%, and 25.7% from the actual monitoring values respectively, and the average error is 19.2%. While applying the integrated analysis method of geological engineering and mapping engineering data of the present invention, the errors between the predicted settlement amounts and the actual monitoring values (30.1mm, 36.8mm, and 22.5mm respectively) are only 5.3%, 3.0%, and 5.3%, and the average error is reduced to 4.5%. Generally speaking, the present invention significantly improves the integrated analysis ability of geological engineering and mapping engineering data through the organic combination of technologies such as multi-scale spatial expression, spatio-temporal registration, graph theory optimization, probability modeling, and dual-domain deep learning, providing more reliable technical support for engineering safety and decision-making.

[0129] It should be noted that the detailed explanations of the variables involved in the present invention are shown in Tables 5, 6, 7, and 8 below.

[0130] Table 5 Variable Explanation Table (Part 1)

[0131] Table 6 Variable Explanation Table (Part 2)

[0132] Table 7 Variable Explanation Table (Part 3)

[0133] Table 8 Variable Explanation Table (Part 4)

[0134] As described above, it is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered within the protection scope of the present invention.

Claims

1. An integrated analysis method for geological engineering and surveying and mapping engineering data, characterized in that, Including: Construct a multi-scale geological spatial data model to perform multi-level decomposition on geological bodies; use a spatio-temporal registration algorithm to standardize surveying and mapping data; Use the minimum spanning tree algorithm to perform optimization and dimensionality reduction on multi-source heterogeneous data, construct a characteristic connection network for geological surveying and mapping data, extract key nodes and paths, and generate a dimensionality reduction parameter set; establish an association rule between geological body attributes and surface deformation parameters based on a probabilistic graphical model; apply an adaptive weight fusion algorithm to integrate geological internal structure information and surface monitoring results to form a geological surveying and mapping fusion feature set; input the geological surveying and mapping fusion feature set into the geological surveying and mapping dual-domain model, dynamically adjust the model parameters in combination with the computational complexity-precision balance function, perform iterative optimization until convergence, and output the integrated analysis results of geological engineering and surveying and mapping engineering.

2. The method according to claim 1, characterized in that, The multi-scale geological spatial data model extracts features of geological bodies at different scales through wavelet transform, constructs a multi-level mathematical expression including geological structure, physical property parameters, and spatial distribution characteristics, and realizes a complete description of geological bodies from region to local.

3. The method according to claim 2, wherein The spatio-temporal registration algorithm is a data alignment method based on control point matching and non-rigid transformation. By identifying common feature points in geological data and surveying and mapping data, it establishes a spatial correspondence relationship between the data, and realizes the unified expression of data collected at different times through time series analysis.

4. The method according to claim 3, wherein The minimum spanning tree algorithm is a graph theory optimization method. It regards the characteristic points of geological surveying and mapping data as the nodes of the graph, and the feature similarity as the edge weight. By constructing a subgraph that connects all nodes and has the minimum total weight, it realizes the extraction of the key structure of the data.

5. The method according to claim 4, wherein The dimensionality reduction parameter set includes the feature dimension reduction ratio, node connection threshold, graph structure sparsity, and data compression rate, which are used to quantify the data loss and information retention degree in the dimensionality reduction process. The dimensionality reduction parameter set is used as the input parameter of the computational complexity-precision balance function.

6. The method according to claim 5, characterized in that The probabilistic graphical model is a statistical model that uses graph theory to represent the dependence relationship between random variables. By representing geological and surveying and mapping parameters with nodes and the correlation between parameters with edges, it establishes the conditional probability distribution between parameters and realizes the uncertainty reasoning of geological states and surface responses.

7. The method according to claim 6, wherein The adaptive weight fusion algorithm dynamically adjusts the weight coefficients of various types of data in the fusion process according to data quality, reliability, and correlation, ensuring that high-quality data makes a greater contribution to the final analysis results and improving the analysis accuracy.

8. The method according to claim 7, characterized in that, The geological surveying and mapping fusion feature set is a set of geological internal structure information and surface monitoring results integrated by the adaptive weight fusion algorithm, including lithology distribution, structural characteristics, physical and mechanical parameters of geological bodies, and data indicators such as surface deformation amount, displacement gradient, and deformation rate in surveying and mapping data, which describe the core characteristics of geological engineering and surveying and mapping engineering.

9. The method according to claim 8, wherein The specific structure of the geological surveying and mapping dual-domain model is an encoder-decoder network based on the Transformer architecture, including two parallel branches: a geological domain encoder and a surveying and mapping domain encoder. Each branch consists of multiple self-attention modules, and feature interaction is realized through a cross-domain attention mechanism. The decoder part uses a multi-head attention mechanism to fuse dual-domain features, and the output layer includes a physical error correction module as a sub-module.

10. The method according to claim 9, wherein The geological mapping dual-domain model uses an adaptive gating unit to screen features. The total number of parameters of the geological mapping dual-domain model is dynamically adjusted according to the complexity of geological data and the density of mapping data. The number of self-attention heads is directly proportional to the complexity of the geological body structure, and the attention dimension is directly proportional to the accuracy level of mapping data.

Images