Geothermal resource three-dimensional data model construction method

Abstract

The invention discloses a geothermal resource three-dimensional data model construction method, which comprises the following steps of: firstly, constructing a short-pulse LiDAR, tunable long-pulse laser, multi-frequency geological radar and geophysical exploration collaborative acquisition system, and synchronously acquiring surface topography and underground 0-100m stratum data; data preprocessing is completed through hierarchical denoising and multi-scale registration of an improved ICP (Inductively Coupled Plasma) algorithm; constructing a dynamic weight distribution model based on data spatial correlation and physical consistency to realize intelligent fusion; fusing temperature logging and seismic wave impedance data, and generating a thermal reservoir probability model containing temperature, permeability and porosity parameter layers by adopting stochastic simulation; and finally, dynamic optimization of the model is realized through an incremental updating mechanism and precision verification. Full-dimensional modeling of the earth surface, the underground and the thermal reservoir is achieved, and accurate data support is provided for geothermal resource reserve assessment and development planning.

Description

Technical Field

[0001] This invention belongs to the field of geothermal resource exploration and three-dimensional modeling technology, specifically, it relates to a method for constructing a three-dimensional data model of geothermal resources. Background Technology

[0002] Geothermal resources, as a clean and renewable energy source, play a crucial role in the transformation of the energy structure. Three-dimensional data models are the core foundation for geothermal resource reserve assessment and development planning, directly impacting the scientific and economic viability of geothermal development. LiDAR scanning technology, with its rapid and high-precision three-dimensional point cloud acquisition capabilities, has become a core technology for geothermal exploration modeling. By emitting laser pulses and recording the return time and intensity of the reflected signals, the generated point cloud data can accurately characterize the terrain surface morphology, providing fundamental data for boundary delineation and terrain analysis of geothermal exploration areas.

[0003] However, existing LiDAR technology has significant technical limitations in constructing 3D models of geothermal resources: traditional LiDAR generally uses picosecond-level short-pulse lasers, which have poor penetration, only reflecting surface information and unable to penetrate surface vegetation and shallow soil, making it difficult to obtain shallow subsurface stratigraphic data. Geothermal resource exploration requires not only knowledge of surface topography but also a clear understanding of the distribution, thickness, and temperature characteristics of underground thermal reservoirs. Single-source LiDAR data can only achieve surface modeling and cannot meet the needs of comprehensive geothermal resource exploration.

[0004] Existing improvement methods have many limitations: increasing LiDAR transmission power can improve penetration, but the effect is limited and it can easily lead to a significant increase in surface point cloud noise; some schemes collect ground-penetrating radar data separately to supplement underground information, but they lack an effective fusion mechanism with LiDAR data, resulting in problems such as spatial misalignment and feature disconnection, and they cannot form an integrated three-dimensional model of the surface and underground.

[0005] In view of this, the present invention is proposed. Summary of the Invention

[0006] To solve the above-mentioned technical problems, the basic concept of the technical solution adopted by the present invention is as follows:

[0007] A method for constructing a three-dimensional data model of geothermal resources includes the following steps:

[0008] Step S1: Establish a multi-source data collaborative acquisition system. Use a short-pulse laser LiDAR device to acquire three-dimensional point clouds of the surface topography, and dynamically adapt the scanning parameters to the exploration area; use a tunable long-pulse laser detection device to acquire shallow underground strata data; and use a multi-frequency ground-penetrating radar device to acquire underground medium structure data from 0 to 100 meters.

[0009] Step S2: Adaptive data preprocessing and hierarchical denoising: LiDAR point cloud data is denoised using improved statistical filtering, ground-penetrating radar data is denoised using wavelet transform, and geophysical data is denoised using Fourier transform; Multi-scale registration: First, coarse registration is performed using ground control points (GCPs), and then fine registration is performed using an improved ICP algorithm. The objective function of the improved ICP algorithm is:

[0010]

[0011] in, As a reference point cloud, Let R be the point cloud to be registered, R be the rotation matrix, and T be the translation vector. This represents the elevation gradient characteristic. For feature constraint weights;

[0012] Step S3: Intelligent weighted fusion to construct a dynamic weight allocation model; based on data spatial correlation and physical consistency The formula for calculating the weights is:

[0013]

[0014] in, Let be the spatial correlation coefficient between the k-th type of data and the baseline data. For physical matching degree, Number of data sources;

[0015] Step S4: Refined 3D Model Construction. For the surface terrain model: co-kriging interpolation (introducing slope constraints) is used, with an interpolation error ≤3cm. The formula is:

[0016]

[0017] For interpolation weights, For drift term, This is the slope influence coefficient. This represents the slope value.

[0018] Subsurface stratigraphic model: Based on sequence stratigraphy, an isochronous surface model is constructed by combining lithological interface reflection characteristics and integrating temperature logging data with seismic impedance inversion results. A probabilistic model of thermal reservoir is generated using stochastic simulation methods.

[0019] Step S5: Dynamic update and optimization, establishing an incremental update mechanism: The new data acquisition cycle is set to 3-12 months based on the geothermal dynamics, and the sliding window method is used to locally update the model;

[0020] Model accuracy verification: Root mean square error (RMSE) is calculated using new borehole data. When RMSE > 5cm, global optimization is triggered to adjust the interpolation parameters and weight model.

[0021] In a preferred embodiment of the present invention, the pulse width of the tunable long pulse laser detection device in step S1 is continuously adjustable from 5 to 20 ns through an FPGA driving circuit, and it has a built-in dual-channel receiving module to collect surface reflection signals and underground penetration signals respectively.

[0022] As a preferred embodiment of the present invention, the iteration termination condition of the improved ICP algorithm in step S2 is: the change in registration error between two adjacent iterations is <0.05mm, and the change in feature constraint terms is <0.02. The maximum number of iterations is set to 80 to ensure the registration accuracy in complex terrain areas.

[0023] In a preferred embodiment of the present invention, the spatial correlation coefficient mentioned in step S3 The physical degree of matching is obtained by calculating the covariance function between data points. Fuzzy reasoning is used to assign values ​​based on the geothermal geology knowledge base, with the value range being 0-1.

[0024] In a preferred embodiment of the present invention, the probabilistic model of the thermal reservoir in step S4 includes three key parameter layers: temperature distribution layer, permeability distribution layer, and porosity distribution layer, and each parameter layer is correlated through a geostatistical variation function.

[0025] In a preferred embodiment of the present invention, the incremental update mechanism in step S5 adopts the feature pyramid matching algorithm, which first matches the new data with the key feature points of the model, and then performs local re-interpolation on the 50m range around the feature points.

[0026] In a preferred embodiment of the present invention, the multi-frequency ground-penetrating radar device in step S1 has a built-in real-time quality assessment module that dynamically adjusts the antenna height and moving speed by means of the reflected wave signal-to-noise ratio and the continuity of the phase axis.

[0027] In a preferred embodiment of the present invention, the dual-mode LiDAR achieves synchronous scanning through a time-division multiplexing mechanism, with short pulse and long pulse lasers emitted alternately at an interval of ≤10μs, thus avoiding mutual interference between the two laser signals.

[0028] The slope influence coefficient of the co-kriging interpolation in step S4 Dynamic adjustment: flat areas or =0.1, steep area =0.5, achieving accurate preservation of terrain features.

[0029] When constructing the underground stratigraphic model as described in step S4, the fault zone influence factor is introduced. The corrected stratigraphic depth is determined by the following formula:

[0030]

[0031] In the formula To correct the stratigraphic depth, The initial stratigraphic depth, This is the distance from the point to the fault zone. This is the influence coefficient of the fault zone, ranging from 0.1 to 0.3. The stronger the activity of the fault zone, the greater the influence coefficient. The larger the value, the better the model reflects the disturbance characteristics of fault structures on the stratigraphic morphology.

[0032] Compared with the prior art, the present invention has the following advantages:

[0033] This invention innovatively employs multi-source data collaborative acquisition, with short-pulse LiDAR ensuring the accuracy of surface modeling, and long-pulse lasers overcoming the limitations of traditional technologies to achieve data penetration acquisition of shallow layers (0-20m underground). Multi-frequency ground-penetrating radar and geophysical exploration supplement the stratigraphic structure and tectonic information of layers (0-100m underground), constructing a full-dimensional data system of "surface-shallow layer-deep thermal reservoir," completely solving the technical bottleneck that a single LiDAR can only characterize the surface.

[0034] An improved ICP algorithm is proposed, which optimizes the objective function by introducing an elevation gradient feature constraint term, reducing the registration error from 15-20cm in the traditional method to ≤2cm. A dynamic weight allocation model is constructed, which accurately allocates the weights of multi-source data based on spatial correlation and physical consistency, avoiding spatial misalignment and feature disconnection caused by simple splicing. The consistency of the fused data is improved by more than 80%.

[0035] Enhanced scientific rigor in geothermal reservoir modeling: The probabilistic model of geothermal reservoirs integrates multi-dimensional data such as temperature logging and seismic impedance, and includes three key parameter layers: temperature, permeability, and porosity. Furthermore, it correlates each parameter through a geostatistical variation function, resulting in a geothermal reservoir boundary prediction error of ≤5cm, a temperature prediction error of ≤2℃, and a porosity prediction error of ≤2%, providing accurate parameter support for geothermal resource reserve calculation.

[0036] The specific embodiments of the present invention will now be described in further detail with reference to the accompanying drawings. Attached Figure Description

[0037] In the attached diagram:

[0038] Figure 1 This is a flowchart of a method for constructing a three-dimensional data model of geothermal resources. Detailed Implementation

[0039] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments will be clearly and completely described below with reference to the accompanying drawings. The following embodiments are used to illustrate the present invention.

[0040] A method for constructing a three-dimensional data model of geothermal resources includes the following steps:

[0041] Step S1: Establish a multi-source data collaborative acquisition system. Use a short-pulse laser LiDAR device to acquire three-dimensional point clouds of the surface topography, and dynamically adapt the scanning parameters to the exploration area; use a tunable long-pulse laser detection device to acquire shallow underground strata data; and use a multi-frequency ground-penetrating radar device to acquire underground medium structure data from 0 to 100 meters.

[0042] Step S2: Adaptive data preprocessing and hierarchical denoising: LiDAR point cloud data is denoised using improved statistical filtering, ground-penetrating radar data is denoised using wavelet transform, and geophysical data is denoised using Fourier transform; Multi-scale registration: First, coarse registration is performed using ground control points (GCPs), and then fine registration is performed using an improved ICP algorithm. The objective function of the improved ICP algorithm is:

[0043]

[0044] in, As a reference point cloud, Let R be the point cloud to be registered, R be the rotation matrix, and T be the translation vector. This represents the elevation gradient characteristic. For feature constraint weights;

[0045] Step S3: Intelligent weighted fusion to construct a dynamic weight allocation model; based on data spatial correlation and physical consistency The formula for calculating the weights is:

[0046]

[0047] in, Let be the spatial correlation coefficient between the k-th type of data and the baseline data. For physical matching degree, Number of data sources;

[0048] Step S4: Refined 3D Model Construction. For the surface terrain model: co-kriging interpolation (introducing slope constraints) is used, with an interpolation error ≤3cm. The formula is:

[0049]

[0050] For interpolation weights, For drift term, This is the slope influence coefficient. This represents the slope value.

[0051] Subsurface stratigraphic model: Based on sequence stratigraphy, an isochronous surface model is constructed by combining lithological interface reflection characteristics. Temperature logging data and seismic impedance inversion results are fused, and a probabilistic model of the thermal reservoir is generated using a stochastic simulation method. First, temperature logging data (accuracy ±0.1℃) and seismic impedance data (resolution ≤5m) are correlated to establish a mapping relationship between impedance, temperature, porosity, and permeability. Then, based on the fused data, spatial distribution characteristics of the thermal reservoir (such as boundaries and thickness) are extracted, and the simulation is set to 50 times to generate the probabilistic model of the thermal reservoir. This model includes three key parameter layers: a temperature distribution layer (accuracy ±2℃), a permeability distribution layer (resolution ≤10mD), and a porosity distribution layer (accuracy ±1%). Each parameter layer is correlated through a geostatistical variability function (horizontal range 500-1000m, vertical range 50-100m) to ensure the rationality and consistency of parameter distribution.

[0052] Step S5: Dynamic update and optimization, establishing an incremental update mechanism: The new data acquisition cycle is set to 3-12 months based on the geothermal dynamics, and the sliding window method is used to locally update the model;

[0053] Model accuracy verification: Root mean square error (RMSE) is calculated using new borehole data. When RMSE > 5cm, global optimization is triggered to adjust the interpolation parameters and weight model.

[0054] Furthermore, the pulse width of the tunable long pulse laser detection device described in step S1 is continuously adjustable from 5 to 20 ns through an FPGA driving circuit, and it has a built-in dual-channel receiving module to collect surface reflection signals and underground penetration signals respectively.

[0055] Furthermore, the iteration termination condition of the improved ICP algorithm in step S2 is: the change in registration error between two adjacent iterations is <0.05mm, and the change in feature constraint terms is <0.02. The maximum number of iterations is set to 80 to ensure the registration accuracy in complex terrain areas.

[0056] Furthermore, the spatial correlation coefficient mentioned in step S3 The physical degree of matching is obtained by calculating the covariance function between data points. Fuzzy reasoning is used to assign values ​​based on the geothermal geology knowledge base, with the value range being 0-1.

[0057] Furthermore, the probabilistic model of the thermal reservoir described in step S4 includes three key parameter layers: temperature distribution layer, permeability distribution layer, and porosity distribution layer, and each parameter layer is correlated through a geostatistical variation function.

[0058] Furthermore, the incremental update mechanism described in step S5 employs a feature pyramid matching algorithm. First, it matches the new data with the key feature points of the model, then performs local re-interpolation within a 50m radius around the feature points. The new data acquisition cycle is set to 3-12 months based on geothermal dynamics: for areas with significant dynamic changes in geothermal reservoirs (such as geothermal fields with high extraction intensity), the update cycle is set to 3 months; for areas with stable geothermal reservoirs, the update cycle is set to 12 months. By using the feature pyramid matching algorithm, first matching the newly acquired data with the key feature points of the existing model (such as stratigraphic interface inflection points, fault zone endpoints, and geothermal reservoir boundary points), and then performing local re-interpolation within a 50m radius around the feature points, the inefficiency caused by global reconstruction is avoided, and the update efficiency is improved by more than 60% compared to global reconstruction.

[0059] Furthermore, the multi-frequency ground-penetrating radar equipment described in step S1 has a built-in real-time quality assessment module that dynamically adjusts the antenna height and moving speed based on the reflected wave signal-to-noise ratio and the continuity of the phase axis.

[0060] In step S2, LiDAR point cloud denoising employs an improved statistical filtering algorithm. The filtering threshold K is dynamically set to K = 1.5 + 0.02 × terrain complexity. By calculating the mean and standard deviation of the distance between each point and its neighboring points (with a neighborhood radius of 0.3m), discrete noise points with a distance greater than K times the standard deviation are removed. Compared to traditional statistical filtering, this algorithm can dynamically adjust the threshold according to terrain complexity, retaining more details in flat areas and effectively removing noise in complex terrain areas. After denoising, the proportion of noise points is ≤3%, and the feature point retention rate is ≥98%.

[0061] Ground-penetrating radar data denoising: A wavelet transform denoising algorithm based on db4 wavelet basis is adopted, with the number of decomposition layers set to 5. The low-frequency coefficients (containing effective signals) after decomposition are retained, while the high-frequency coefficients (containing noise) are thresholded (soft threshold = 0.05 × maximum value of high-frequency coefficients), effectively removing shallow interference and electromagnetic noise. The signal-to-noise ratio is improved to over 25dB after denoising.

[0062] Geophysical data denoising: Fourier transform denoising algorithm is adopted. The time domain data is converted to the frequency domain through fast Fourier transform, the signal main frequency distribution is analyzed, and the cutoff frequency is dynamically set to 1.2 times the signal main frequency to filter high-frequency noise and retain the frequency components of the effective signal. The data fluctuation amplitude after denoising is ≤5%.

[0063] Furthermore, the dual-mode LiDAR achieves synchronous scanning through a time-division multiplexing mechanism, with short-pulse and long-pulse lasers emitting alternately at intervals of ≤10μs to avoid mutual interference between the two laser signals;

[0064] The slope influence coefficient of the co-kriging interpolation in step S4 Dynamic adjustment: flat areas or =0.1, steep area =0.5, achieving accurate preservation of terrain features.

[0065] In step S4, an isochronous surface model is constructed based on sequence stratigraphy principles and the reflection characteristics of lithological interfaces. First, reflection signal characteristics (such as reflection amplitude and phase axis continuity) of stratigraphic interfaces are extracted by fused data to identify the boundaries between different lithological strata. Then, the interface depth is calibrated based on borehole lithological data to construct the isochronous surface model, with a stratigraphic correlation error ≤0.5m, and a fault zone influence factor is introduced. The corrected stratigraphic depth is determined by the following formula:

[0066]

[0067] In the formula To correct the stratigraphic depth, The initial stratigraphic depth, This is the distance from the point to the fault zone. This is the influence coefficient of the fault zone, ranging from 0.1 to 0.3. The stronger the activity of the fault zone, the greater the influence coefficient. The larger the value, the better the model reflects the disturbance characteristics of fault structures on the stratigraphic morphology.

[0068] Experimental Example 1: 3D Modeling of a Sedimentary Basin-Type Geothermal Field in the North China Plain

[0069] The experimental area is located in a geothermal field within a sedimentary basin in the North China Plain. The terrain is predominantly plain with some shallow depressions; the surface vegetation cover is 35%. The underground geological structure, from top to bottom, consists of: a 0-18m shallow surface layer, an 18-75m sandstone geothermal reservoir (the core exploration target), and a 75-100m mudstone aquitard. Six verification boreholes, each 100m deep, have been drilled in the area, obtaining complete lithological stratification, temperature logging, and permeability test data for model construction and accuracy verification.

[0070] Experimental Procedures and Results

[0071] (1) Multi-source data collaborative acquisition

[0072] Short-pulse LiDAR acquisition: Flying along 24 parallel survey lines, a total of 138 million points of surface point cloud data were collected, with a point cloud density of 55 points / m², a single point accuracy of ±2cm, and complete coverage of the surface topography of the experimental area;

[0073] Long-pulse laser acquisition: synchronously following the LiDAR survey line, 42 million points of shallow surface point cloud were acquired from 0-20m underground, and the interface between sand and clay layers at a depth of 18m was successfully identified with a penetration accuracy of ±4cm.

[0074] Ground-penetrating radar data acquisition: 18 survey lines were laid out along the direction perpendicular to the LiDAR survey lines, with a total length of 180km. The real-time quality assessment module showed that the signal-to-noise ratio of the reflected wave was ≥20dB and the length of the continuous section of the same phase axis was ≥90m.

[0075] Temperature logging data from six boreholes showed that the temperature range of the thermal reservoir (18-75m) was 42-63℃, with the temperature in borehole Z4 at a depth of 65m reaching 63℃, making it the high-temperature center of the regional thermal reservoir.

[0076] (2) Adaptive data preprocessing

[0077] Layered denoising: LiDAR point cloud data was denoised using improved statistical filtering (K=1.5+0.02×0.4=1.508), removing 3.5% of noise points; after wavelet transform denoising, the signal-to-noise ratio of ground-penetrating radar data was improved from 14dB to 28dB.

[0078] Multi-scale registration: GCP coarse registration error is 4.2cm, improved ICP algorithm converges after 52 iterations, fine registration error is 1.8cm;

[0079] Format standardization: uniformly converted to GeoTIFF format, with complete metadata recording of collection and processing parameters.

[0080] (3) Intelligent weighted fusion

[0081] Dynamic weight calculation: Based on the geothermal geology knowledge base, the physical meaning matching degree is assigned through fuzzy reasoning. Combining spatial correlation coefficient The results of the hierarchical weight allocation are shown in Table 1 below:

[0082] Depth range Short-pulse LiDAR weights Long pulse laser weighting Ground Penetrating Radar Weight 0-5m 0.70 0.20 0.10 5-50m 0.10 0.30 0.60 50-100m - 0.20 0.80

[0083] Layered fusion: A fusion algorithm combining Kalman filtering and DS evidence theory is adopted, which improves data consistency by 92% after fusion.

[0084] (4) Detailed construction of three-dimensional model

[0085] Surface topography model: co-kriging interpolation (α=0.1), RMSE=2.1cm compared with GCP; Subsurface stratigraphy model: fault zone influence factor introduced. =0.15 (the regional fault zone has relatively weak activity), and the error after layer depth correction is ≤0.3m; Probabilistic model of thermal reservoir: Sequential Gaussian simulation generates three parameter layers. The temperature distribution layer predicts that the temperature at a depth of 65m in borehole Z4 is 62.8℃, which is 0.2℃ different from the measured value of 63℃; The porosity prediction error is ≤1.0%.

[0086] (5) Dynamic updates and optimizations

[0087] Simulated incremental update: After 3 months, 2 new ground-penetrating radar survey lines were collected. After the feature pyramid matching algorithm was locally updated, the prediction accuracy of the thermal reservoir boundary was improved to ±0.25m.

[0088] Accuracy verification: The verification results of the 6 borehole data are shown in the table below:

[0089] Validation metrics Z1 Drilling Z2 drilling Z3 Drilling Z4 Drilling Z5 Drilling Z6 Drilling average value Surface model error (cm) 1.9 2.3 2.0 2.2 2.1 2.4 2.1 Depth error of top boundary of thermal reservoir (cm) 3.2 2.8 3.5 2.6 3.0 3.3 3.1 Temperature error of thermal reservoir (°C) 1.3 1.5 1.2 0.2 1.4 1.6 1.2 Porosity error (%) 1.0 1.2 0.9 0.8 1.1 1.3 1.0 Overall RMSE of the model (cm) - - - - - - 3.3

[0090] The overall RMSE of the model is 3.3cm < 5cm, which meets the design requirements and does not require global optimization.

[0091] Experimental Example 2: Three-dimensional modeling of a geothermal field in a fault zone development area in southwestern mountainous region

[0092] The experimental area is located in a fault-zone-developed geothermal field in the southwestern mountainous region, covering an area of ​​8 km × 8 km. The terrain is mainly low and medium mountains with a vegetation coverage of 65%. Two main active fault zones (F1 and F2) are developed within the area, intersecting in an "X" shape. The underground geological structure consists of: 0-15m weathered granite (shallow surface layer), 15-65m fractured granite geothermal reservoir (permeability 100-180 mD), and 65-100m dense granite. Four verification boreholes (K1-K4) were drilled, with boreholes K2 and K3 located near the fault zone.

[0093] Core experimental results

[0094] For areas with high vegetation cover, the long-pulse laser equipment adjusted the pulse width to 15ns and increased the emission power to 120%, successfully penetrating 65% forest coverage to obtain point clouds of weathered granite 0-18m underground, with a penetration rate 32% higher than that of traditional equipment;

[0095] The improved ICP algorithm converged after 75 iterations in complex terrain areas (slope 30-40°) with a registration error of 2.3cm, effectively solving the registration problem caused by abrupt changes in mountainous terrain.

[0096] When constructing the underground stratigraphic model, the fault zone influence factor is set. =0.25 (the fault zone is highly active), and after correction, the prediction error of the top boundary of the thermal reservoir near the fault zone decreased from 5.8cm to 3.2cm;

[0097] The probabilistic model of the thermal reservoir shows that the temperature of the thermal reservoir near the fault zone reaches 72-85℃ and the permeability is ≥150mD. The error is <3% compared with the measured data of K2 borehole (temperature 78℃, permeability 162mD).

[0098] The model's dynamic update cycle was set to 3 months. Local re-interpolation was used to update the model within a 1.5km radius around the fault zone. After the update, the spatial correlation accuracy between the fault zone and the thermal reservoir was improved by 40%.

Claims

1. A method for constructing a three-dimensional data model of geothermal resources, characterized in that, Includes the following steps: Step S1: Establish a multi-source data collaborative acquisition system. Use a short-pulse laser LiDAR device to acquire three-dimensional point clouds of the surface topography, and dynamically adapt the scanning parameters to the exploration area; use a tunable long-pulse laser detection device to acquire shallow underground strata data. Multi-frequency ground-penetrating radar equipment was used to collect structural data of the underground medium from 0 to 100 meters. Step S2: Adaptive data preprocessing and hierarchical denoising: LiDAR point cloud data is denoised using improved statistical filtering, ground-penetrating radar data is denoised using wavelet transform, and geophysical data is denoised using Fourier transform; Multi-scale registration: First, coarse registration is performed using ground control points (GCPs), and then fine registration is performed using an improved ICP algorithm. The objective function of the improved ICP algorithm is: in, As a reference point cloud, Let R be the point cloud to be registered, R be the rotation matrix, and T be the translation vector. This represents the elevation gradient characteristic. For feature constraint weights; Step S3: Intelligent weighted fusion to construct a dynamic weight allocation model; based on data spatial correlation and physical consistency The formula for calculating the weights is: in, Let be the spatial correlation coefficient between the k-th type of data and the baseline data. For physical matching degree, Number of data sources; Step S4: Refined 3D model construction, surface terrain model: co-kriging interpolation (introducing slope constraints) is used, with an interpolation error ≤3cm. The formula is: For interpolation weights, For drift term, This is the slope influence coefficient. This represents the slope value. Subsurface stratigraphic model: Based on sequence stratigraphy, an isochronous surface model is constructed by combining lithological interface reflection characteristics; By integrating temperature logging data and seismic impedance inversion results, a probabilistic model of thermal reservoirs is generated using stochastic simulation methods. Step S5: Dynamic update and optimization, establishing an incremental update mechanism: The new data acquisition cycle is set to 3-12 months based on the geothermal dynamics, and the sliding window method is used to locally update the model; Model accuracy verification: Root mean square error (RMSE) is calculated using new borehole data. When RMSE > 5cm, global optimization is triggered to adjust the interpolation parameters and weight model.

2. The method for constructing a three-dimensional data model of geothermal resources according to claim 1, characterized in that, The pulse width of the tunable long pulse laser detection device described in step S1 is continuously adjustable from 5 to 20 ns through an FPGA driving circuit, and it has a built-in dual-channel receiving module to collect surface reflection signals and underground penetration signals respectively.

3. The method for constructing a three-dimensional data model of geothermal resources according to claim 1, characterized in that, The iteration termination condition of the improved ICP algorithm in step S2 is: the change in registration error between two adjacent iterations is <0.05mm, and the change in feature constraint terms is <0.

02. The maximum number of iterations is set to 80 to ensure the registration accuracy in complex terrain areas.

4. The method for constructing a three-dimensional data model of geothermal resources according to claim 1, characterized in that, The spatial correlation coefficient mentioned in step S3 The physical degree of matching is obtained by calculating the covariance function between data points. Fuzzy reasoning is used to assign values ​​based on the geothermal geology knowledge base, with the value range being 0-1.

5. The method for constructing a three-dimensional data model of geothermal resources according to claim 1, characterized in that, The probabilistic model of the thermal reservoir described in step S4 includes three key parameter layers: temperature distribution layer, permeability distribution layer, and porosity distribution layer, and each parameter layer is correlated through a geostatistical variation function.

6. The method for constructing a three-dimensional data model of geothermal resources according to claim 1, characterized in that, The incremental update mechanism described in step S5 uses a feature pyramid matching algorithm to first match the new data with the key feature points of the model, and then perform local re-interpolation on the 50m range around the feature points.

7. The method for constructing a three-dimensional data model of geothermal resources according to claim 1, characterized in that, The multi-frequency ground-penetrating radar equipment described in step S1 has a built-in real-time quality assessment module that dynamically adjusts the antenna height and moving speed based on the reflected wave signal-to-noise ratio and phase axis continuity.

8. The method for constructing a three-dimensional data model of geothermal resources according to claim 1, characterized in that, The dual-mode LiDAR achieves synchronous scanning through a time-division multiplexing mechanism, with short-pulse and long-pulse lasers emitted alternately at intervals of ≤10μs to avoid mutual interference between the two laser signals.

9. The method for constructing a three-dimensional data model of geothermal resources according to claim 1, characterized in that, The slope influence coefficient of the co-kriging interpolation in step S4 Dynamic adjustment: flat areas or =0.1, steep area =0.5, achieving accurate preservation of terrain features.

10. When constructing the underground stratigraphic model as described in step S4, the fault zone influence factor is introduced. The corrected stratigraphic depth is determined by the following formula: In the formula To correct the stratigraphic depth, The initial stratigraphic depth, This is the distance from the point to the fault zone. This is the influence coefficient of the fault zone, ranging from 0.1 to 0.

3. The stronger the activity of the fault zone, the greater the influence coefficient. The larger the value, the better the model reflects the disturbance characteristics of fault structures on the stratigraphic morphology.

Images