Evaluation method of geological characteristics of soil-rock composite strata based on multi-source data fusion

Abstract

The present invention discloses a method for evaluating geological characteristics of soil-rock composite strata based on multi-source data fusion, which belongs to the field of geotechnical engineering data processing technology. The method comprises obtaining a multi-source data set of soil-rock composite strata, performing spatiotemporal alignment and denoising processing, and generating a pre-processed multi-source data set; extracting geotechnical layered structural characteristics based on the pre-processed multi-source data set, and generating an initial geotechnical characteristic parameter set; performing incremental correction on the initial geotechnical characteristic parameter set based on sensor feedback data collected in real time during the construction process, and generating a dynamically updated geotechnical characteristic parameter set; based on the dynamically updated geotechnical characteristic parameter set, outputting a geological characteristic evaluation report through a preset engineering evaluation standard library. The present invention adopts spatiotemporal alignment of multi-source data, dynamic fusion of physical models and deep learning, and real-time feedback incremental correction technology, which can improve the accuracy and dynamic adaptability of geological characteristic modeling and provide accurate decision support for complex stratum construction.

Description

Technical Field

[0001] The present invention relates to the technical field of geotechnical engineering data processing, and in particular to a method for evaluating geological characteristics of soil-rock composite strata based on multi-source data fusion. Background Art

[0002] The evaluation of geological characteristics of geotechnical composite strata is a crucial technical step in underground engineering construction. By analyzing diverse information, including physical exploration data, remote sensing imagery, and historical engineering data, the evaluation establishes stratum mechanical parameters and a hierarchical structural model to guide the optimization of construction parameters. Existing technologies typically rely on modeling and analysis based on a single data source or simple data overlay methods.

[0003] Existing technologies often rely on static data processing, such as mechanical modeling based on geophysical radar reflection velocity or analyzing soil boundary demarcations using remote sensing image texture analysis. Multi-source data is often fused using weighted averaging or fixed thresholds, with equipment operations driven solely by pre-set model parameters during construction. Some approaches attempt to use machine learning algorithms for feature extraction, but these approaches lack the constraints of physical laws.

[0004] However, the lack of unified spatiotemporal benchmarks for multi-source data leads to the accumulation of fusion errors; sensor feedback data is only used for result verification and does not participate in the dynamic correction of the model; the identification of geotechnical stratification interfaces overly relies on the confidence of a single type of data and fails to effectively deal with the complementarity and conflict of multi-source data; and the static model cannot adapt to the dynamic changes in stratum characteristics during construction. Summary of the Invention

[0005] To solve the above problems, the present invention provides a geological characteristic evaluation method for soil-rock composite strata based on multi-source data fusion. It adopts the spatiotemporal alignment of multi-source data, the dynamic fusion of physical models and deep learning, and the real-time feedback incremental correction technology, which can improve the accuracy and dynamic adaptability of geological characteristic modeling and provide accurate decision-making support for complex stratum construction.

[0006] The above objectives can be achieved through the following solutions:

[0007] A geological characteristic evaluation method for soil-rock composite strata based on multi-source data fusion includes obtaining a multi-source data set of soil-rock composite strata, wherein the multi-source data includes preset physical detection parameters, remote sensing image parameters and historical engineering parameters; performing spatiotemporal alignment and denoising on the multi-source data set to generate a pre-processed multi-source data set; extracting rock and soil layered structural characteristics based on the pre-processed multi-source data set to generate an initial rock and soil characteristic parameter set; performing incremental correction on the initial rock and soil characteristic parameter set based on sensor feedback data collected in real time during the construction process to generate a dynamically updated rock and soil characteristic parameter set; and outputting a geological characteristic evaluation report based on the dynamically updated rock and soil characteristic parameter set through a preset engineering evaluation standard library.

[0008] Optionally, the spatiotemporal alignment and denoising processing of the multi-source dataset to generate a preprocessed multi-source dataset includes: performing spatiotemporal benchmark matching on the physical detection parameters in the multi-source dataset to generate spatiotemporal normalized physical detection data; performing edge node filtering processing on the remote sensing image parameters in the multi-source dataset to generate denoised remote sensing image data; and merging the spatiotemporal normalized physical detection data, the denoised remote sensing image data and the historical engineering parameters in the multi-source dataset to obtain a preprocessed multi-source dataset.

[0009] Optionally, the extraction of geotechnical hierarchical structural features based on the preprocessed multi-source dataset and the generation of an initial geotechnical characteristic parameter set include: inputting the preprocessed multi-source dataset into a preset physical modeling layer, and calculating the initial geotechnical mechanical parameter distribution based on the geomechanical equation; synchronously inputting the preprocessed multi-source dataset into a preset deep learning prediction network to generate a geotechnical implicit feature vector; and dynamically fusing the initial geotechnical mechanical parameter distribution with the geotechnical implicit feature vector to generate an initial geotechnical characteristic parameter set.

[0010] Optionally, the initial geotechnical parameter distribution is dynamically fused with the geotechnical implicit characteristic vector to generate an initial geotechnical characteristic parameter set, including: generating physical constraints based on the initial geotechnical parameter distribution, and embedding the physical constraints as a loss function into the deep learning prediction network; intercepting sensor feedback data of continuous time periods through a preset sliding time window algorithm, and performing dynamic gradient correction on the geotechnical implicit characteristic vector; and weightedly superimposing the corrected geotechnical implicit characteristic vector and the initial geotechnical parameter distribution according to a preset fusion weight coefficient to generate an initial geotechnical characteristic parameter set.

[0011] Optionally, the initial geotechnical characteristic parameter set is incrementally corrected based on the sensor feedback data collected in real time during the construction process to generate a dynamically updated geotechnical characteristic parameter set, including: real-time collection of sensor feedback data during the construction process; comparing the sensor feedback data with the parameter data of the initial geotechnical characteristic parameter set to generate a parameter deviation matrix; and using the parameter deviation matrix to update the fusion weight coefficient to generate a dynamically updated geotechnical characteristic parameter set.

[0012] Optionally, the parameter deviation matrix is ​​used to update the fusion weight coefficient to generate a dynamically updated set of geotechnical characteristic parameters, including: extracting regional geostress distribution laws from a preset cross-engineering knowledge migration database as migration optimization parameters; performing joint optimization calculations on the migration optimization parameters and the parameter deviation matrix to generate a cross-domain corrected fusion weight coefficient; using the cross-domain corrected fusion weight coefficient, the corrected geotechnical implicit characteristic vector is weightedly superimposed on the initial geotechnical mechanics parameter distribution to generate a dynamically updated set of geotechnical characteristic parameters.

[0013] Optionally, the method further includes: performing boundary identification on the rock and soil stratification interface based on the dynamically updated set of rock and soil characteristic parameters to generate optimized stratification interface data; performing statistical distribution compensation on the clay-gravel interaction layer parameters in the optimized stratification interface data to generate a high-resolution rock and soil stratification map.

[0014] Optionally, based on the dynamically updated set of geotechnical characteristic parameters, the boundaries of the geotechnical stratification interface are identified to generate optimized stratification interface data, including: dynamically allocating weights to the confidence levels of physical detection parameters and remote sensing image parameters to generate a multi-source attention weight matrix; and weightedly fusing the dynamically updated set of geotechnical characteristic parameters based on the multi-source attention weight matrix to generate optimized stratification interface data.

[0015] Optionally, generating a multi-source attention weight matrix includes: generating a semantic attention score based on the rock and soil texture characteristics of the remote sensing image parameters; generating a statistical attention score based on the signal stability of the physical detection parameters; and multiplying and fusing the semantic attention score with the statistical attention score to generate a multi-source attention weight matrix.

[0016] Optionally, the method further includes: matching the geological feature evaluation report with a preset construction parameter recommendation library to generate shield advancement speed and grouting pressure parameters to guide construction; and transmitting the shield advancement speed and grouting pressure parameters to the BIM collaborative management platform in real time through a preset interface protocol.

[0017] Compared with the prior art, the present invention has the following advantages:

[0018] 1. This invention solves the problems of spatiotemporal reference differences and noise interference in multi-source data through spatiotemporal alignment and denoising. Combining the fusion mechanism of physical modeling and deep learning feature extraction, it ensures the physical rationality of geotechnical parameter distribution and deep mining of implicit features, significantly reducing the uncertainty brought by a single data source or model, and providing high-precision basic data for engineering decision-making.

[0019] 2. This invention uses real-time sensor feedback data during construction to incrementally correct initial parameters and dynamically adjusts fusion weight coefficients through cross-engineering knowledge migration. This enables the geotechnical characteristic parameter set to dynamically respond to stratum changes during construction, avoiding the prediction bias caused by static modeling in traditional methods and improving the safety and efficiency of construction in complex strata.

[0020] 3. This invention uses a multi-source attention weight allocation strategy based on physical detection and remote sensing data, combined with statistical distribution compensation technology, to effectively solve the boundary fuzziness problem caused by insufficient data resolution at complex interfaces such as clay-gravel interaction layers, generate high-resolution rock and soil layer maps, and provide refined support for shield construction parameter optimization.

[0021] 4. The present invention connects the geological feature evaluation results with the BIM platform in real time, automatically generates shield propulsion speed and grouting pressure parameter recommendations, opens up a collaborative link between geological data analysis and construction execution, and solves the problem of inefficient matching of geological data and construction machinery parameters under the traditional model.

[0022] Other features and advantages of the present invention will be described in the following description, and in part will become apparent from the description, or will be understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures pointed out in the description, claims and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0024] Figure 1 It is a flow chart of a method for evaluating geological characteristics of soil-rock composite formations based on multi-source data fusion according to an embodiment of the present invention.

[0025] Figure 2 Schematic diagram of the change of the fusion weight coefficient during the construction process of an embodiment of the present invention. DETAILED DESCRIPTION

[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0027] Reference Figure 1 One embodiment of the present invention proposes a geological characteristic evaluation method for soil-rock composite strata based on multi-source data fusion. It adopts the spatiotemporal alignment of multi-source data, the dynamic fusion of physical models and deep learning, and the real-time feedback incremental correction technology, which can improve the accuracy and dynamic adaptability of geological characteristic modeling and provide accurate decision-making support for complex stratum construction.

[0028] The method of this embodiment specifically includes:

[0029] Acquiring a multi-source data set of a soil-rock composite stratum, wherein the multi-source data includes preset physical detection parameters, remote sensing image parameters, and historical engineering parameters;

[0030] Performing spatiotemporal alignment and denoising processing on the multi-source dataset to generate a preprocessed multi-source dataset;

[0031] Extracting geotechnical hierarchical structure features based on the preprocessed multi-source data set to generate an initial geotechnical feature parameter set;

[0032] Incrementally correcting the initial geotechnical characteristic parameter set based on sensor feedback data collected in real time during the construction process to generate a dynamically updated geotechnical characteristic parameter set;

[0033] Based on the dynamically updated geotechnical characteristic parameter set, a geological characteristic evaluation report is output through a preset engineering evaluation standard library.

[0034] Specifically, the system first performs a multi-dimensional feature matching between the elastic modulus, permeability, and cohesion in the dynamically updated set of geotechnical characteristic parameters and a pre-set engineering evaluation standard library. This library stores engineering stability evaluation indicators and their threshold ranges for different geotechnical combinations. A fuzzy membership algorithm is used to calculate the degree of match between the current parameters and the standard indicators. Based on the matching score, the corresponding evaluation conclusion framework is selected, and a structured report is automatically generated, including stratum bearing capacity assessment, deformation risk prediction, and construction recommendations.

[0035] Among them, the dynamically updated geotechnical characteristic parameter set refers to the gridded data set of geotechnical mechanical parameters such as elastic modulus and cohesion corrected by real-time sensor data; the engineering evaluation standard library refers to a relational database that presets evaluation indicators and response measures such as engineering safety factor and allowable settlement under different geological conditions; multidimensional feature matching refers to the process of comparing the geotechnical parameter vector with each dimension recorded in the engineering evaluation standard library item by item and calculating the comprehensive similarity; the fuzzy membership algorithm is a calculation method that quantifies the degree of proximity between the actual parameters and the standard threshold through a preset membership function; and a structured report refers to a technical document that integrates the evaluation conclusions, risk levels and construction recommendations according to a preset template.

[0036] For example, in the dynamically updated geotechnical characteristic parameter set, the elastic modulus of a certain area is 25 MPa, the permeability coefficient is The engineering evaluation standard library retrieved the standard bearing value for the clay-sandstone interbedded layer as 200 kPa and the settlement warning threshold as 15 mm. The calculated matching score between the current parameters and the standard reached 90%, triggering the evaluation conclusion of "medium bearing capacity, low permeability risk." The resulting construction parameter combination recommended a shield tunneling speed of ≤40 mm / min and a synchronous grouting pressure of 0.45-0.5 MPa. The final output included a geological characteristic evaluation report containing a screenshot of the 3D geological model and a parameter comparison table.

[0037] Specifically, by acquiring three types of heterogeneous data—physical detection parameters, remote sensing image parameters, and historical engineering parameters—we first utilize spatiotemporal alignment to eliminate differences in coordinate systems and time bases. Edge node filtering and wavelet transforms are then employed for noise suppression to generate normalized preprocessed data. By combining the analysis of geomechanical equations at the physical modeling layer with the implicit feature extraction of deep learning networks, we generate an initial geotechnical property distribution under physical constraints. Furthermore, we incorporate sensor feedback data from the construction process to establish a parameter deviation-driven dynamic weight adjustment mechanism. Through incremental correction, geotechnical characteristic parameters are brought close to actual working conditions in real time. Ultimately, dynamically optimized geological evaluation results are output based on a library of engineering evaluation standards.

[0038] By unifying the spatiotemporal benchmarks of multi-source data and deeply exploring the dual-driven physics and data features, this model overcomes the one-sidedness of traditional single-source modeling and improves the generalization of geotechnical stratification parameters in complex strata. A dynamic adjustment mechanism for fusion weights based on real-time sensor feedback enables the model to self-correct online, reducing engineering parameter prediction errors by 35%-48% compared to static models. Its incremental data processing architecture reduces the delay in geological property analysis to minutes. Combined with intelligent matching of the engineering evaluation standard library, it shortens the construction plan optimization cycle by over 50%, significantly improving the safety of shield construction and its adaptability to strata.

[0039] Optionally, performing spatiotemporal alignment and denoising on the multi-source dataset to generate a preprocessed multi-source dataset includes:

[0040] Performing spatiotemporal benchmark matching on the physical detection parameters in the multi-source data set to generate spatiotemporal normalized physical detection data;

[0041] Performing edge node filtering on the remote sensing image parameters in the multi-source data set to generate noise-reduced remote sensing image data;

[0042] The spatiotemporally normalized physical detection data, the denoised remote sensing image data and the historical engineering parameters in the multi-source data set are combined to obtain a preprocessed multi-source data set.

[0043] Specifically, the physical detection parameters are first subjected to spatiotemporal benchmarking using a preset spatiotemporal alignment algorithm. Spatiotemporal benchmarking involves identifying differences in coordinate systems and timestamps between data collected by different sensors. This data is then unified into a preset engineering coordinate system using a homogeneous coordinate transformation matrix. Data gaps caused by sampling frequency differences are then filled using temporal interpolation to generate spatiotemporally normalized physical detection data. Next, remote sensing image parameters are subjected to edge node filtering using an edge-cloud collaborative computing architecture. Edge node filtering involves deploying a Gaussian-Markov filter algorithm at the data acquisition terminal, using a variance-weighted method to eliminate impulse noise during image transmission, and then performing a frequency-domain wavelet transform via a cloud service to remove periodic interference components, generating de-noised remote sensing image data. Finally, the spatiotemporally normalized physical detection data, de-noised remote sensing image data, and historical engineering parameters are merged into a preprocessed multi-source dataset using a preset data fusion format. This merging operation implements structured storage based on the engineering geographic grid, with each grid cell associated with the corresponding physical detection parameter value, remote sensing pixel matrix, and historical parameter records.

[0044] Among them, the spatiotemporal alignment algorithm is a spatial registration method that calculates coordinate compensation based on sensor calibration parameters and performs timestamp interpolation. The physical detection parameters include geological radar reflection wave velocity and resistivity detection sequence values. The spatiotemporal normalized physical detection data refers to a set of physical detection parameters that have passed through a unified coordinate system and time reference. The edge-cloud collaborative computing architecture refers to a distributed processing system composed of edge computing modules deployed on-site acquisition equipment and cloud computing resources on remote servers. The denoised remote sensing image data refers to a digital image matrix that has undergone edge noise filtering and cloud-based frequency domain processing. The historical engineering parameters include geological drilling data and construction log structured information of neighboring projects. The preprocessed multi-source dataset is a fused record set of each data source after spatiotemporal alignment.

[0045] Optionally, extracting geotechnical hierarchical structure features based on the preprocessed multi-source data set to generate an initial geotechnical feature parameter set includes:

[0046] Inputting the preprocessed multi-source data set into a preset physical modeling layer, and calculating initial geomechanical parameter distribution based on geomechanical equations;

[0047] Synchronously inputting the preprocessed multi-source data set into a preset deep learning prediction network to generate geotechnical implicit feature vectors;

[0048] The initial geotechnical mechanics parameter distribution is dynamically fused with the geotechnical implicit characteristic vector to generate an initial geotechnical characteristic parameter set.

[0049] Optionally, dynamically fusing the initial geotechnical parameter distribution with the geotechnical implicit characteristic vector to generate an initial geotechnical characteristic parameter set includes:

[0050] Generating physical constraints based on the initial geomechanical parameter distribution, and embedding the physical constraints as a loss function into the deep learning prediction network;

[0051] The sensor feedback data of continuous time periods are intercepted by a preset sliding time window algorithm to perform dynamic gradient correction on the rock and soil implicit characteristic vector;

[0052] The modified rock and soil implicit characteristic vector and the initial rock and soil mechanical parameter distribution are weightedly superimposed according to a preset fusion weight coefficient to generate an initial rock and soil characteristic parameter set.

[0053] Specifically, the preprocessed multi-source datasets are first synchronously input into the preset physical modeling layer and deep learning prediction network. The physical modeling layer numerically solves the constitutive relationship and energy conservation law of the rock and soil based on geomechanical equations. The geomechanical equations include the stress balance equation, the constitutive equation based on the generalized Hooke's law, and the Darcy seepage equation. For the stress balance equation, there are:

[0054] ,

[0055] Where, is the stress tensor, is the divergence of the stress tensor, is the formation density, is the acceleration due to gravity; for the constitutive equation:

[0056] ,

[0057] Where, is the stiffness tensor, including the elastic modulus and Poisson's ratio , is the strain tensor; for the Darcy flow equation:

[0058] ,

[0059] Where, is the seepage velocity, is the permeability, is the fluid viscosity, is the rate of change of pore water pressure gradient. Input physical detection parameters such as resistivity and wave velocity, invert to obtain formation density and Poisson's ratio, then solve the stress field through finite element method and derive the strain tensor , and finally through the stiffness tensor Inverse elastic modulus :

[0060] ,

[0061] in, 、 、 、 、 、 is the Kronecker symbol, the Lamé constant and Depend on and Decide:

[0062] ,

[0063] In the formula, the input physical detection parameters such as resistivity, wave velocity and strain tensor are , the strain energy density is calculated by the constitutive equation :

[0064] ,

[0065] Using the maximum strain energy density at critical failure Calculate the cohesion :

[0066] ,

[0067] Where, is the proportional coefficient, which is determined through calibration tests. Finally, the spatial distribution matrix of elastic modulus, cohesion, and permeability is obtained, and the initial geotechnical parameter distribution is output. The deep learning prediction network extracts implicit correlation features from multi-source data through multi-layer convolution to generate geotechnical implicit feature vectors. Subsequently, physical constraints are generated based on the stress-strain distribution results output by the physical modeling layer. The constraints are converted into regularization terms and superimposed on the loss function of the deep learning prediction network to adjust the spatial distribution of network parameters during backpropagation. At the same time, a sliding time window algorithm is used to intercept sensor feedback data in continuous time periods during the construction process, calculate the gradient deviation from the geotechnical implicit feature vector, and correct the feature vector using the dynamic gradient descent method. Finally, the corrected geotechnical implicit feature vector and the initial geotechnical parameter distribution are weighted and superimposed on the preset fusion weight coefficient to generate the initial geotechnical feature parameter set.

[0068] Among them, the physical modeling layer is composed of a computing module with a pre-set finite element solver, which is used to perform numerical analysis of geomechanical equations; the initial geotechnical parameter distribution refers to the spatial distribution matrix of elastic modulus, cohesion and permeability generated according to the calculation results of the equation; the deep learning prediction network adopts a temporal convolution architecture with residual connections, and the input layer receives time series and spatial grid data of multi-source data; the geotechnical implicit feature vector is the high-dimensional feature mapping result output by the network hidden layer; the sliding time window algorithm is a dynamic sampling method that intercepts sensor data with a fixed time span; the dynamic gradient correction updates the gradient value in the back propagation link by calculating the root mean square error between the sensor measured data and the feature vector; the fusion weight coefficient in the weighted superposition operation is dynamically adjusted according to the deviation between the value of each parameter in the initial geotechnical characteristic parameter set and the actual measured value.

[0069] For example, the physical modeling layer substitutes resistivity data from preprocessed multi-source datasets into the stress balance equations in the geomechanical equations to calculate the initial distribution of Young's modulus for each geotechnical layer. A deep learning prediction network simultaneously performs convolution operations on ground radar reflections and historical engineering parameters to extract implicit feature vectors that characterize the irregular morphology of soil interface. During the backpropagation phase, the stress field continuity constraint corresponding to the Young's modulus distribution is added as a regularization term to the loss function, forcing the implicit features output by the network to conform to physical laws. A sliding time window is used to capture the previous 30 minutes of shield machine torque sensor data, calculate its gradient correction to the implicit features, and update the network parameters using an adaptive optimizer. Finally, the corrected implicit feature vector is linearly superimposed with the initial Young's modulus at a weight of 0.7:0.3 to generate an initial feature parameter set containing microscopic geotechnical properties.

[0070] Optionally, performing incremental correction on the initial geotechnical characteristic parameter set based on sensor feedback data collected in real time during the construction process to generate a dynamically updated geotechnical characteristic parameter set includes:

[0071] Real-time collection of sensor feedback data during the construction process;

[0072] Comparing the sensor feedback data with the parameter data of the initial geotechnical characteristic parameter set to generate a parameter deviation matrix;

[0073] The parameter deviation matrix is ​​used to update the fusion weight coefficient to generate a dynamically updated geotechnical characteristic parameter set.

[0074] Specifically, the changes in the fusion weight coefficient during the construction process are as follows: Figure 2As shown in the figure, sensor feedback data is first collected in real time during the construction process. This data includes shield machine hydraulic pressure, geological radar reflection intensity, and soil displacement values. The sensor feedback data is then compared one by one with the corresponding parameter data in the initial geotechnical characteristic parameter set. The difference between the actual measured value and the predicted value of each parameter is calculated to generate a parameter deviation matrix. The parameter deviation matrix is ​​then used to dynamically adjust the fusion weight coefficients. The weights of the physical modeling and deep learning modules are redistributed in inverse proportion to the deviation. Finally, the adjusted weights are applied to a weighted superposition operation to generate a dynamically updated geotechnical characteristic parameter set.

[0075] Among them, sensor feedback data refers to the geotechnical response data collected in real time by sensors deployed during the construction process, including but not limited to shield machine thrust, cutterhead torque and pore water pressure measurement values; the initial geotechnical characteristic parameter set refers to the geotechnical characteristic parameter set generated by the fusion of physical modeling and deep learning; the parameter deviation matrix is ​​a two-dimensional difference matrix between actual measurement values ​​and predicted parameters, with rows representing parameter categories and columns corresponding to time series; the fusion weight coefficient refers to the weight ratio coefficient used in the weighted superposition of the initial geotechnical parameter distribution output by the physical modeling layer and the geotechnical implicit characteristic vector generated by the deep learning prediction network; the dynamically updated geotechnical characteristic parameter set refers to the geotechnical characteristic data set that reflects the latest working conditions after real-time adjustment of the weights.

[0076] For example, when the real-time feedback value of the cutterhead torque sensor is 3800kN·m, and the initial geotechnical characteristic parameter set prediction value is 3200kN·m, the absolute deviation of the parameter is calculated to be 600kN·m. The deviation value is input into the parameter deviation matrix, and the weight coefficient update mechanism is triggered based on the preset deviation threshold. For example, when the deviation exceeds 500kN·m, the weight of the physical modeling layer is reduced from 0.7 to 0.6, and the deep learning prediction weight is correspondingly increased to 0.4. The updated weight coefficient is used to perform a weighted superposition operation on the two, so that the updated geotechnical characteristic parameter set is closer to the actual construction conditions. If the subsequent deviation is reduced to within 300kN·m, the initial weight ratio is gradually restored to achieve dynamic balance.

[0077] Optionally, using the parameter deviation matrix to update the fusion weight coefficient to generate a dynamically updated geotechnical characteristic parameter set includes:

[0078] Extract regional geostress distribution patterns from the preset cross-engineering knowledge migration database as migration optimization parameters;

[0079] Performing joint optimization calculation on the migration optimization parameters and the parameter deviation matrix to generate a cross-domain corrected fusion weight coefficient;

[0080] The cross-domain corrected fusion weight coefficient is used to perform weighted superposition on the corrected geotechnical implicit characteristic vector and the initial geotechnical mechanics parameter distribution to generate a dynamically updated geotechnical characteristic parameter set.

[0081] Specifically, the in-situ stress distribution patterns of historical projects in the region are first extracted from a cross-project knowledge transfer database, which stores geological exploration records and construction feedback data from adjacent projects, as migration optimization parameters. The migration optimization parameters and the generated parameter deviation matrix are then input into the optimization algorithm. The cross-domain revised fusion weight coefficient is calculated by calculating the combined influence of the two on the fusion weight coefficient. Finally, the updated fusion weight coefficient is used to perform a weighted superposition of the revised geotechnical implicit characteristic vector and the initial geotechnical parameter distribution, forming a dynamically updated set of geotechnical characteristic parameters.

[0082] Among them, the cross-engineering knowledge migration database refers to a relational database that stores multiple historical engineering geological data in the same area, including the measured values ​​of borehole geostress and deformation monitoring data during the construction period; the migration optimization parameter refers to the correlation coefficient between the formation inclination and the maximum principal stress direction obtained based on the analysis of historical engineering data; the parameter deviation matrix refers to the two-dimensional difference matrix of the difference between the measured values ​​of the aforementioned sensors and the initial geotechnical characteristic parameters; the joint optimization calculation refers to the objective function solution process that integrates the historical geostress law and the real-time deviation data, specifically through the weighted least squares method to realize the linear combination of the migration parameters and the deviation matrix; the cross-domain corrected fusion weight coefficient refers to the weight ratio of the physical model and the deep learning model after adjustment by the regional geostress law; the weighted superposition operation refers to the operation of adding the geotechnical parameter distribution and the deep learning feature vector according to the new weight to generate a fusion result.

[0083] For example, when the shield tunneling machine reached the clay-conglomerate interface, the cross-project knowledge transfer database extracted the average geostress direction of the same stratum segment from three adjacent projects as N30°E, and the migration optimization parameter was set to an inclination matching coefficient of 0.85. Combined with the 400kN shield thrust deviation in the parameter deviation matrix for the current construction segment, a joint optimization calculation increased the weight of the physical modeling layer from 0.6 to 0.68, and the deep learning weight was adjusted accordingly to 0.32. Finally, 0.68 times the initial geomechanical parameters (such as shear strength of 26kPa) were superimposed with 0.32 times the corrected implicit eigenvector (representing interface roughness) to obtain a dynamic geotechnical parameter set reflecting the regional geostress characteristics. If the migration parameters indicate stress concentration in the layer at that location in the historical project, the physical model weight is further increased to strengthen the mechanical constraints.

[0084] Optionally, the method further includes:

[0085] Based on the dynamically updated set of geotechnical characteristic parameters, boundary identification is performed on the geotechnical layer interface to generate optimized layer interface data;

[0086] Statistical distribution compensation is performed on the clay-gravel interaction layer parameters in the optimized layered interface data to generate a high-resolution rock and soil layering map.

[0087] Specifically, based on the dynamically updated set of geotechnical characteristic parameters, the confidence levels of physical detection parameters and remote sensing image parameters are first dynamically weighted, and semantic attention scores are generated by calculating the geotechnical texture features of remote sensing images. Statistical attention scores are generated by combining the signal stability of physical detection parameters, and the two are multiplied together to obtain a multi-source attention weight matrix. This matrix is ​​then weightedly fused with the dynamically updated set of geotechnical characteristic parameters to generate optimized stratified interface data. Finally, the statistical distribution deviation of the clay-gravel interaction layer parameters in the optimized data is compensated and corrected, and the parameter probability distribution model of similar rock layers in historical projects is used for data filling to generate a high-resolution geotechnical stratification map with millimeter-level accuracy.

[0088] Among them, the dynamically updated set of geotechnical characteristic parameters refers to the set of geotechnical mechanical characteristic parameters corrected by real-time sensor data, including spatial distribution data such as elastic modulus and permeability; the semantic attention score refers to the numerical weight calculated based on the morphological complexity of different rock textures in the remote sensing image, reflecting the contribution of image features to the stratified interface; the statistical attention score refers to the reliability index calculated based on the inverse of the signal variance during continuous sampling of physical detection parameters, which characterizes the stability of sensor data; the multi-source attention weight matrix refers to the composite weight distribution map obtained by multiplying the semantic score and the statistical score, which is used to quantify the credibility of multi-source data in interface identification; the optimized stratified interface data refers to the set of geotechnical stratification boundary coordinates after weighted fusion through the multi-source attention weight matrix; the statistical distribution compensation refers to the Bayesian estimation filling operation of the current missing data using the parameter probability density function of the clay-gravel layer in the historical engineering database; the high-resolution geotechnical stratification map refers to the final generated three-dimensional stratigraphic structure visualization model with millimeter-level grid accuracy.

[0089] For example, when a remote sensing image shows that a certain area has discontinuous gravel texture features, a semantic attention score of 0.85 is generated for it. If the signal variance of the physical detection parameters at the same time is lower than the threshold, a statistical attention score of 0.92 is assigned. The two are multiplied together to obtain a multi-source attention weight of 0.782 for the area. In the weighted fusion stage, the weight coefficient is applied to the dynamically updated set of geotechnical characteristic parameters to highlight the influence of high-confidence data on the stratification interface. For the missing particle size parameters in the identified clay-gravel interaction layer, the particle size statistical models of 500 groups of similar strata in the historical database are called, and the maximum likelihood estimation is used to fill in the current data, finally generating a high-precision map that can distinguish 5mm thick interlayers.

[0090] Optionally, based on the dynamically updated set of geotechnical characteristic parameters, performing boundary identification on the geotechnical layer interface to generate optimized layer interface data includes:

[0091] Dynamically assign weights to the confidence levels of physical detection parameters and remote sensing image parameters to generate a multi-source attention weight matrix;

[0092] The dynamically updated geotechnical characteristic parameter set is weightedly fused based on the multi-source attention weight matrix to generate optimized layered interface data.

[0093] Specifically, the semantic attention score is first calculated according to the rock and soil texture characteristics of the remote sensing image parameters, and the texture roughness and directional consistency quantitative indicators of the rock and soil texture characteristics are extracted through the gray level co-occurrence matrix; at the same time, the statistical attention score is calculated according to the inverse of the variance of the signal sampling sequence in the physical detection parameters; the semantic attention score and the statistical attention score are multiplied point by point and then normalized to generate a multi-source attention weight matrix; this matrix is ​​used as the basis for weight distribution, and a weighted sum operation is performed on each parameter in the dynamically updated rock and soil characteristic parameter set to generate optimized layered interface data.

[0094] Among them, the dynamically updated set of geotechnical characteristic parameters refers to the spatial distribution set of geotechnical mechanical parameters corrected by real-time sensor feedback data, including three-dimensional grid data of elastic modulus, Poisson's ratio and permeability; the semantic attention score refers to the weight value calculated based on the morphological complexity of the geotechnical texture in the local area of ​​the remote sensing image, which is quantified by the contrast characteristics of the grayscale co-occurrence matrix combined with the fractal dimension algorithm; the statistical attention score refers to the reliability index obtained by the signal stability of the statistical physical detection parameters in a continuous time window, and the signal stability is jointly determined by the inverse of the variance and the kurtosis; the multi-source attention weight matrix refers to the weighted factor distribution map formed by the dot product of the semantic attention score and the statistical attention score and normalized; the weighted sum operation refers to the operation of multiplying the geotechnical characteristic parameters of each spatial grid point with the weight factor of its corresponding position and then accumulating them; the optimized stratification interface data refers to the set of geotechnical stratification boundary coordinates generated after multi-source credibility weighting.

[0095] For example, when a remote sensing image detects a mottled gravel texture with disordered bedding direction in a certain area, the grayscale co-occurrence matrix calculates a contrast index of 0.89 and a fractal dimension of 2.3, which are converted into a semantic attention score of 0.78 through a preset linear mapping formula; at the same time, the inverse variance of the resistivity series in the physical detection parameters of the area is 0.83, the kurtosis conforms to the normal distribution, and the statistical attention score is 0.91; the two are multiplied to obtain the initial weight of 0.71 for the grid point, which is then globally normalized to form the actual value of 0.68 in the multi-source attention weight matrix; when the weight matrix is ​​multiplied by the dynamically updated set of geotechnical characteristic parameters, the grid point data with an elastic modulus of 23.5 MPa is assigned a weight of 0.68, and the elastic modulus of 19.2 MPa in the adjacent area is assigned a weight of 0.53. After global weighted summation, the interface depth of the bedding area is determined to be 15.6 meters, and the accuracy is improved by 37% compared with the unweighted state.

[0096] Optionally, generating a multi-source attention weight matrix includes:

[0097] generating a semantic attention score according to the rock and soil texture features of the remote sensing image parameters;

[0098] generating a statistical attention score based on the signal stability of the physical detection parameter;

[0099] The semantic attention score is multiplied and fused with the statistical attention score to generate a multi-source attention weight matrix.

[0100] Specifically, when dynamically allocating weights to the confidence levels of physical detection parameters and remote sensing image parameters, we first perform gray-level co-occurrence matrix analysis on the rock and soil texture features in the remote sensing image parameters, calculate the texture contrast and fractal dimension of each region, and map them to semantic attention scores using a linear mapping formula. The linear mapping formula can be:

[0101] ,

[0102] Where, is the semantic attention score, is the texture contrast coefficient, is the texture contrast, is the fractal dimension coefficient, is the fractal dimension, 、 Optimize through historical data, The initial value is set to 0.05, The initial value is set to 0.32. At the same time, the inverse variance and kurtosis index of the continuous time window data of the physical detection parameters are calculated, and the statistical attention score is generated by the stability discriminant function. For the stability discriminant function, there is:

[0103] ,

[0104] Where, is the stability value, is the variance of the continuous time window data x, is the kurtosis parameter of the continuous time window data x, is the stability weight coefficient, which is calibrated by experiments; when the stability value is greater than the preset stability threshold, the statistical attention score ,have:

[0105] ,

[0106] When the stability value is less than the stability threshold, the statistical attention score takes the preset fixed value of 0.1, and the stability threshold is calibrated based on historical working condition data; then the obtained semantic attention score is multiplied by the statistical attention score item by item according to the grid point, and then global normalization is performed to finally obtain the multi-source attention weight matrix.

[0107] Among them, the semantic attention score refers to the confidence index calculated by the complexity of the local texture of the remote sensing image, that is, the confidence of the remote sensing image parameters. The gray-level co-occurrence matrix is ​​used to extract the contrast parameters of the image and the fractal dimension algorithm is coupled for quantification. The statistical attention score refers to the reliability assessment value calculated based on the stability index of the physical detection parameter sequence, that is, the confidence of the physical detection parameter. The inverse of the variance is used to reflect the degree of signal fluctuation, and the kurtosis parameter characterizes the steepness of the data distribution. The multi-source attention weight matrix is ​​a two-dimensional grid weight distribution map formed by normalizing the product results of the above two scores, realizing the fusion expression of the credibility of physical detection and remote sensing image information. Normalization processing refers to scaling all product results in the grid to between 0 and 1 according to the total ratio to ensure that the weight ratio of each area meets the probability distribution characteristics.

[0108] For example, when a certain area in a remote sensing image presents a high-contrast sand-soil interlaced texture, the grayscale co-occurrence matrix calculates a contrast value of 1.2 and a fractal dimension of 2.5, which are converted into a semantic attention score of 0.86 through a linear mapping formula; in the same area, the inverse of the variance of 10 sets of data continuously collected by the resistivity detector is 0.79, and the kurtosis value is 3.1, which meets the stability condition and is assigned a statistical attention score of 0.82; the two scores are multiplied to obtain an initial weight of 0.705, and after all grid points are calculated, they are normalized and the weight value of the point is adjusted to 0.68; finally, when the multi-source attention weight matrix is ​​applied to the dynamically updated set of geotechnical characteristic parameters, the elastic modulus parameter weight of the area is increased to 1.5 times the original, significantly optimizing the recognition accuracy of the clay-sandstone interface.

[0109] Optionally, the method further includes:

[0110] Matching the geological characteristic evaluation report with a preset construction parameter recommendation library to generate shield advancement speed and grouting pressure parameters to guide construction;

[0111] The shield advancement speed and grouting pressure parameters are transmitted to the BIM collaborative management platform in real time through a preset interface protocol.

[0112] Specifically, the rock and soil strength, stratum inclination and permeability parameters contained in the geological characteristic evaluation report are first matched with the characteristic dimensions of the preset construction parameter recommendation library, which stores successful shield operation parameter combinations under different geological conditions in historical projects; the matching scores of the current geological parameters and each record in the recommendation library are calculated using a preset fuzzy similarity algorithm, and the top five parameter groups with the highest scores are selected as candidate sets; feasibility filtering is then performed based on the equipment status constraints of the real-time construction environment, and the optimal shield propulsion speed and grouting pressure parameter combination is finally determined; the parameter data packet is encapsulated using the industrial bus protocol and transmitted in real time according to the IFC standard interface format preset by the BIM collaborative management platform.

[0113] Among them, the geological characteristic evaluation report refers to a geological characteristic analysis document generated based on a dynamically updated set of geotechnical characteristic parameters, which contains quantitative data such as elastic modulus, cohesion and interlayer friction angle; the construction parameter recommendation library refers to an associated database that stores verified shield tunneling speeds, grouting pressure values ​​and corresponding geological conditions in historical projects; the fuzzy similarity algorithm uses a membership function to calculate the degree of closeness between the current geological parameters and historical cases in each dimension, and the membership function is defined as the ratio of the inverse of the difference between the actual parameters and the recommended parameters to the maximum allowable deviation; the shield tunneling speed refers to the product of the number of rotations per minute of the shield machine cutter head and the forward distance, and the grouting pressure parameter refers to the cement slurry pumping pressure value required to maintain the stability of the tunnel lining; the industrial bus protocol refers to an industrial communication protocol that complies with the OPC UA specification, which is used to ensure the real-time and reliability of data transmission; the IFC standard interface refers to the international common format for building information model data exchange, which includes the data structure and semantic description rules of shield parameters.

[0114] For example, when a geological assessment report indicated a certain section's rock and soil strength of 28 MPa and a stratum dip of 12°, three sets of historical records were retrieved from the construction parameter recommendation database: Case 1, with a strength of 27.5 MPa and an inclination of 11°, corresponded to a thrust speed of 35 mm / min and a grouting pressure of 0.45 MPa; Case 2, with a strength of 29 MPa and an inclination of 13°, corresponded to parameters of 33 mm / min and 0.48 MPa; and Case 3, with a strength of 28.2 MPa and an inclination of 12.5°, corresponded to parameters of 34 mm / min and 0.47 MPa. Using a fuzzy similarity algorithm, the matching scores between the current parameters and each case were 92, 88, and 95, respectively. Combined with the current shield machine's maximum thrust limit of 28,000 kN, the parameter combination for Case 3 was selected. Finally, it was encapsulated into a JSON data packet containing timestamp, coordinate positioning and parameter values ​​through the OPC UA protocol, and transmitted to the BIM platform according to the IFC standard equipment operation template. The working condition change of the shield machine's travel speed from 32mm / min to 34mm / min was displayed in real time in the 3D model.

[0115] It should be noted that the above is only an exemplary embodiment of the present invention and cannot be used to limit the scope of the present invention. That is, any equivalent changes and modifications made according to the teachings of the present invention are still within the scope of the present invention. After considering the disclosure of the specification and the truth of practice, those skilled in the art will easily think of other embodiments of the present invention. This application is intended to cover any variation, use or adaptation of the present invention, which follows the general principles of the present invention and includes common knowledge or customary technical means in the art that are not described in the present invention.

Claims

1. A geological characteristics evaluation method for soil-rock composite strata based on multi-source data fusion, characterized by: The method comprises: Acquiring a multi-source data set of a soil-rock composite stratum, wherein the multi-source data includes preset physical detection parameters, remote sensing image parameters, and historical engineering parameters; Performing spatiotemporal alignment and denoising processing on the multi-source dataset to generate a preprocessed multi-source dataset; Based on the preprocessed multi-source data set, geotechnical hierarchical structural feature extraction is performed to generate an initial geotechnical feature parameter set; the method includes: inputting the preprocessed multi-source data set into a preset physical modeling layer, calculating the initial geotechnical parameter distribution based on the geomechanics equation; simultaneously inputting the preprocessed multi-source data set into a preset deep learning prediction network to generate a geotechnical implicit feature vector; and dynamically fusing the initial geotechnical parameter distribution with the geotechnical implicit feature vector to generate an initial geotechnical feature parameter set; Incrementally correcting the initial geotechnical characteristic parameter set based on sensor feedback data collected in real time during the construction process to generate a dynamically updated geotechnical characteristic parameter set; this includes: collecting sensor feedback data in real time during the construction process; comparing the sensor feedback data with parameter data of the initial geotechnical characteristic parameter set to generate a parameter deviation matrix; and using the parameter deviation matrix to update a fusion weight coefficient to generate a dynamically updated geotechnical characteristic parameter set; Based on the dynamically updated geotechnical characteristic parameter set, a geological characteristic evaluation report is outputted through a preset engineering evaluation standard library; Among them, the dynamic fusion of the initial geotechnical parameter distribution and the geotechnical implicit characteristic vector to generate an initial geotechnical characteristic parameter set includes: generating physical constraints based on the initial geotechnical parameter distribution, and embedding the physical constraints as a loss function into the deep learning prediction network; intercepting sensor feedback data of continuous time periods through a preset sliding time window algorithm, and performing dynamic gradient correction on the geotechnical implicit characteristic vector; and weighted superposition of the corrected geotechnical implicit characteristic vector and the initial geotechnical parameter distribution according to a preset fusion weight coefficient to generate an initial geotechnical characteristic parameter set.

2. The method for evaluating geological characteristics of soil-rock composite strata based on multi-source data fusion according to claim 1, characterized in that: The performing spatiotemporal alignment and denoising on the multi-source dataset to generate a pre-processed multi-source dataset includes: Performing spatiotemporal benchmark matching on the physical detection parameters in the multi-source data set to generate spatiotemporal normalized physical detection data; Performing edge node filtering on the remote sensing image parameters in the multi-source data set to generate noise-reduced remote sensing image data; The spatiotemporally normalized physical detection data, the denoised remote sensing image data and the historical engineering parameters in the multi-source data set are combined to obtain a preprocessed multi-source data set.

3. The method for evaluating geological characteristics of soil-rock composite strata based on multi-source data fusion according to claim 1, characterized in that: The step of updating the fusion weight coefficient by using the parameter deviation matrix to generate a dynamically updated geotechnical characteristic parameter set includes: Extract regional geostress distribution patterns from the preset cross-engineering knowledge migration database as migration optimization parameters; Performing joint optimization calculation on the migration optimization parameters and the parameter deviation matrix to generate a cross-domain corrected fusion weight coefficient; The cross-domain corrected fusion weight coefficient is used to perform weighted superposition on the corrected geotechnical implicit characteristic vector and the initial geotechnical mechanics parameter distribution to generate a dynamically updated geotechnical characteristic parameter set.

4. The method for evaluating geological characteristics of soil-rock composite strata based on multi-source data fusion according to claim 1, characterized in that: The method further comprises: Based on the dynamically updated set of geotechnical characteristic parameters, boundary identification is performed on the geotechnical layer interface to generate optimized layer interface data; Statistical distribution compensation is performed on the clay-gravel interaction layer parameters in the optimized layered interface data to generate a high-resolution rock and soil layering atlas.

5. The method for evaluating geological characteristics of soil-rock composite strata based on multi-source data fusion according to claim 4 is characterized in that: The step of performing boundary identification on the geotechnical layer interface based on the dynamically updated geotechnical characteristic parameter set and generating optimized layer interface data includes: Dynamically assign weights to the confidence levels of physical detection parameters and remote sensing image parameters to generate a multi-source attention weight matrix; The dynamically updated geotechnical characteristic parameter set is weightedly fused based on the multi-source attention weight matrix to generate optimized layered interface data.

6. The method for evaluating geological characteristics of soil-rock composite strata based on multi-source data fusion according to claim 5, characterized in that: Generating a multi-source attention weight matrix includes: generating a semantic attention score according to the rock and soil texture features of the remote sensing image parameters; generating a statistical attention score based on the signal stability of the physical detection parameter; The semantic attention score is multiplied and fused with the statistical attention score to generate a multi-source attention weight matrix.

7. The method for evaluating geological characteristics of soil-rock composite strata based on multi-source data fusion according to claim 1, characterized in that: The method further comprises: Matching the geological characteristic evaluation report with a preset construction parameter recommendation library to generate shield advancement speed and grouting pressure parameters to guide construction; The shield advancement speed and grouting pressure parameters are transmitted to the BIM collaborative management platform in real time through a preset interface protocol.

Images