A data acquisition and processing method and device applied to digital engineering investigation

Abstract

This disclosure provides a data acquisition and processing method and apparatus for digital engineering exploration, relating to the field of data processing technology. The method includes: acquiring multi-source geological data of the area to be explored; determining a three-dimensional geological model corresponding to the area based on the multi-source geological data; determining multimodal features based on the multi-source geological data, geological report data of the area to be explored, and spatial structure data corresponding to the three-dimensional geological model; and determining the predicted geological state values ​​of each grid cell in the three-dimensional geological model; inputting the current geological state of the multi-source acquisition device at its current location and the control parameters of the multi-source acquisition device into a reinforcement learning-based feedback control model to generate acquisition control commands for the area to be explored, thereby controlling the multi-source acquisition device to continue acquiring multi-source data in the area based on the acquisition control commands. The method and apparatus proposed in this disclosure improve the overall accuracy of digital engineering exploration operations.

Description

Technical Field

[0001] This disclosure relates to the field of data processing technology, and in particular to a data acquisition and processing method and apparatus for digital engineering surveying. Background Technology

[0002] Modern society has an increasingly urgent need for precise and efficient management of geological information. Exploration data processing technology plays a vital role in many fields, including mineral resource development, infrastructure construction, urban underground space planning, geological disaster prevention and control, agricultural land assessment, and ecological environmental protection.

[0003] However, existing exploration systems often suffer from a disconnect between front-end data acquisition and back-end analysis and prediction. Multi-source exploration equipment typically operates blindly according to pre-set fixed rules or routes. Because the front-end physical equipment cannot obtain real-time geological predictions of the surrounding unknown areas from the back-end model, it cannot autonomously adjust its physical operation behavior. When facing complex, critical, and drastically changing geological structural zones, the overall accuracy of the exploration operation may be affected by the omission of key geological data. Summary of the Invention

[0004] To overcome the problems existing in related technologies, this disclosure provides a data acquisition and processing method and device for digital engineering exploration. By inputting the predicted values ​​of geological parameters into a reinforcement learning model, the control parameters of multi-source acquisition equipment are updated, realizing an adaptive acquisition full-link intelligent closed loop, effectively avoiding the omission of key geological data in exploration operations, and improving the accuracy of exploration operations.

[0005] According to a first aspect of the present disclosure, a data acquisition and processing method for digital engineering surveying is provided, the method comprising: Acquire multi-source geological data collected in real time in the area to be explored by multi-source acquisition devices, and determine the three-dimensional geological model corresponding to the area to be explored and the spatial structure data corresponding to the three-dimensional geological model based on the multi-source geological data; The multi-source geological data, the geological report data of the area to be explored, and the spatial structure data are input into a pre-trained prediction model. The pre-trained prediction model determines the multimodal characteristics of the area to be explored and determines the geological state prediction value of each grid cell in the three-dimensional geological model based on the multimodal characteristics. The three-dimensional geological model is then updated based on the geological state prediction value. Based on the updated 3D geological model, the current geological state of the multi-source acquisition device at its current location and the current control parameters of the multi-source acquisition device are input into a reinforcement learning-based feedback control model to generate acquisition control instructions for the area to be explored. Based on the acquisition control instructions, the multi-source acquisition device is controlled to continue to perform multi-source geological data acquisition in the area to be explored. The acquisition control instructions include acquisition frequency, acquisition trajectory, and device parameters.

[0006] In one exemplary embodiment of this disclosure, acquiring multi-source geological data collected in real time by a multi-source acquisition device in the area to be explored includes: The surface mineral spectral data and three-dimensional topographic data of the area to be explored are collected using airborne acquisition equipment. Geotechnical parameter data and geotechnical environmental data of the area to be investigated are collected using ground-based acquisition equipment; Underground geological environment data of the area to be investigated were collected using underground acquisition equipment. The surface mineral spectral data, the three-dimensional terrain data, the rock and soil parameter data, the rock and soil environmental data, and the underground geological environmental data are preprocessed using edge computing nodes; The preprocessed data is then subjected to spatial consistency processing and temporal consistency processing to obtain the multi-source geological data.

[0007] In one exemplary embodiment of this disclosure, determining the three-dimensional geological model corresponding to the area to be explored based on the multi-source geological data includes: The multi-source geological data is mapped onto a three-dimensional grid, and the geological parameters corresponding to each grid cell in the three-dimensional grid are determined. Based on the geological parameters corresponding to each grid cell in the three-dimensional grid, the geological features corresponding to each grid cell are determined, and the geological features corresponding to each grid cell are subjected to feature fusion processing to obtain the feature vector corresponding to the three-dimensional grid. The geological structural lines and stratigraphic distribution information of the three-dimensional mesh are determined based on the feature vectors, and the three-dimensional mesh is converted into a three-dimensional geological model based on the geological structural lines and stratigraphic distribution information. The geological parameters are mapped onto the three-dimensional geological model to obtain the three-dimensional geological model of the area to be explored; The geological parameters include mineral parameters, soil and rock parameters, environmental parameters, and topographic parameters; the geological features include mineral features, soil and rock features, environmental features, and topographic texture features.

[0008] In one exemplary embodiment of this disclosure, mapping the multi-source geological data onto a three-dimensional mesh and determining the geological parameters corresponding to each mesh cell in the three-dimensional mesh includes: Determine the coordinate mapping relationship between the area to be explored and the three-dimensional grid, map the multi-source geological data to the corresponding grid cells based on the coordinate mapping relationship, and determine the geological parameters corresponding to each grid cell; For each type of geological parameter in the multi-source geological data, the grid cells in the three-dimensional grid that are missing the geological parameter are determined as the first grid cell, and the grid cells in the three-dimensional grid that contain the geological parameter are determined as the second grid cell. For each first grid cell, the estimated value of the geological parameter in the first grid cell is predicted based on the value of the geological parameter in each second grid cell.

[0009] In one exemplary embodiment of this disclosure, the pre-trained prediction model includes a multi-branch feature extraction layer, a multi-modal feature fusion layer, and a joint inference output layer. The step of determining the multi-modal features of the area to be explored by the pre-trained prediction model, and determining the predicted geological state value of each grid cell in the three-dimensional geological model based on the multi-modal features, includes: Through the multi-branch feature extraction layer, the first semantic features corresponding to the surface mineral spectral data in the multi-source geological data, the second semantic features corresponding to the geological report data, the temporal features corresponding to the underground geological environment data and the soil and rock environment data in the multi-source geological data, the spatial features corresponding to the three-dimensional topographic data in the multi-source geological data, the soil and rock features corresponding to the soil and rock parameter data in the multi-source geological data, and the structural features corresponding to the spatial structure data are obtained respectively. The multimodal feature fusion layer determines multimodal features based on the global dependencies between the first semantic feature, the second semantic feature, the temporal feature, the spatial feature, the soil and rock feature, and the structural feature. The geological state prediction value of each grid cell in the three-dimensional geological model is determined based on the multimodal features through the joint inference output layer.

[0010] In one exemplary embodiment of this disclosure, determining multimodal features based on the global dependencies between the first semantic feature, the second semantic feature, the temporal feature, the spatial feature, the soil and rock feature, and the structural feature through the multimodal feature fusion layer includes: The first semantic feature and the second semantic feature are fused based on the cross-attention mechanism to obtain the fused feature. The association weights among the fused features, temporal features, spatial features, soil and rock features, and structural features are determined based on the self-attention mechanism. Based on the association weights, feature fusion processing is performed on each feature to obtain the multimodal features.

[0011] In one exemplary embodiment of this disclosure, the step of inputting the current geological state and the current control parameters of the multi-source acquisition device into a reinforcement learning-based feedback control model to generate acquisition control instructions for the area to be explored includes: Based on the travel direction and spatial coordinates of the multi-source acquisition device, local grid slices within the predicted path range are obtained from the updated three-dimensional geological model; the current geological state is determined based on the predicted geological state values ​​within the local grid slices. The current state vector is input into the feedback control model to calculate the probability distribution corresponding to each candidate acquisition action; The candidate acquisition action with the highest probability in the probability distribution is determined as the target control action, and the acquisition frequency, acquisition trajectory and device parameters are determined based on the target control action to generate the acquisition control command.

[0012] According to a second aspect of the present disclosure, a data acquisition and processing apparatus for digital engineering surveying is provided, comprising: The first determining module is used to acquire multi-source geological data collected in real time by multi-source acquisition devices in the area to be explored, and to determine the three-dimensional geological model corresponding to the area to be explored and the spatial structure data corresponding to the three-dimensional geological model based on the multi-source geological data. The second determining module is used to input the multi-source geological data, the geological report data of the area to be explored, and the spatial structure data into a pre-trained prediction model, and the pre-trained prediction model determines the multimodal features of the area to be explored, and determines the geological state prediction value of each grid cell in the three-dimensional geological model based on the multimodal features, and updates the three-dimensional geological model according to the geological state prediction value. The instruction generation module is used to input the current geological state of the multi-source acquisition device at its current location and the current control parameters of the multi-source acquisition device into a reinforcement learning-based feedback control model based on the updated three-dimensional geological model, and generate acquisition control instructions for the area to be explored, so as to control the multi-source acquisition device to continue to perform multi-source geological data acquisition in the area to be explored based on the acquisition control instructions. The acquisition control instructions include acquisition frequency, acquisition trajectory and device parameters.

[0013] According to a third aspect of the present disclosure, a computer device is provided, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the aforementioned data acquisition and processing method applied to digital engineering surveying.

[0014] According to a fourth aspect of the present disclosure, a computer-readable storage medium is provided, wherein computer program code instructions are stored thereon, and when the computer program code instructions are executed by a processor, the method of any of the above claims is implemented.

[0015] The technical solutions provided by the embodiments of this disclosure may include the following beneficial effects: This disclosure provides a data acquisition and processing method for digital engineering exploration. While constructing a three-dimensional geological model, it further acquires its spatial structure data, and inputs this spatial structure data, along with real-time acquired multi-source geological data and unstructured geological report data, into a pre-trained prediction model. Thus, this spatial structure data can serve as a spatial location code, providing the prediction model with an accurate spatial topological reference system. This ensures that the geological state prediction values ​​derived from joint inference are consistent with the grid cells of the three-dimensional geological model, enabling dynamic overwriting and updating of the three-dimensional geological model and improving the prediction accuracy of geological states. Therefore, compared to exploration methods relying on a single data source or traditional static interpolation, this method can improve the prediction accuracy of geological states in unknown areas.

[0016] Furthermore, the updated 3D geological model acquires the current geological state of the multi-source acquisition equipment at its current location in real time. This current geological state, along with the current control parameters of the multi-source acquisition equipment, is then input into a reinforcement learning-based feedback control model to generate acquisition control commands for the area to be explored. Thus, in the digital engineering exploration workflow, the barrier between front-end acquisition and back-end analysis can be broken down, achieving closed-loop adaptive feedback. This allows the front-end physical equipment to autonomously solve strategies and dynamically adjust its operational behavior based on the geological predictions of the surrounding unknown areas from the large model, without relying on manual intervention.

[0017] Furthermore, by issuing acquisition control commands that include acquisition frequency, acquisition trajectory, and equipment parameters to control multi-source equipment to continue data acquisition, differentiated physical action responses can be achieved when facing complex and ever-changing geological scenarios. By adjusting equipment parameters in advance, the overall efficiency of regional exploration operations can be improved, while effectively mitigating the defect of missing key data in abnormally changing geological structural zones caused by blind acquisition, thus enhancing the overall accuracy of digital engineering exploration operations.

[0018] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0019] The accompanying drawings, which are incorporated in and form part of this disclosure, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure.

[0020] Figure 1 This disclosure is a flowchart illustrating a data acquisition and processing method applied to digital engineering surveying according to an exemplary embodiment.

[0021] Figure 2 This disclosure is a structural diagram of a three-dimensional visualization platform for data acquisition and processing applied to digital engineering surveying, according to an exemplary embodiment.

[0022] Figure 3 This disclosure is a block diagram illustrating a data acquisition and processing apparatus applied to digital engineering surveying according to an exemplary embodiment.

[0023] Figure 4 This disclosure is a hardware structure diagram of a computer device according to an exemplary embodiment. Detailed Implementation

[0024] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0025] The terminology used in this disclosure is for the purpose of describing particular embodiments only and is not intended to be limiting of the disclosure. The singular forms “a,” “the,” and “the” as used in this disclosure and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any and all possible combinations of one or more of the associated listed items.

[0026] It should be understood that although the terms first, second, third, etc., may be used in this disclosure to describe various information, such information should not be limited to these terms. These terms are used only to distinguish information of the same type from one another. For example, without departing from the scope of this disclosure, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to determination."

[0027] The embodiments of this disclosure will now be described in detail.

[0028] like Figure 1 As shown, Figure 1This disclosure is a flowchart illustrating a data acquisition and processing method applied to digital engineering surveying according to an exemplary embodiment, including the following steps: Step 101: Obtain multi-source geological data collected in real time by multi-source acquisition equipment in the area to be explored, and determine the three-dimensional geological model corresponding to the area to be explored and the spatial structure data corresponding to the three-dimensional geological model based on the multi-source geological data.

[0029] In this embodiment, multi-source geological data acquisition of the area to be explored can be achieved through airborne acquisition equipment, ground acquisition equipment, and underground acquisition equipment. Among these, airborne acquisition equipment can be drone swarms, equipped with hyperspectral imagers and LiDAR (Light Detection and Ranging Sensor) to collect drone image data of the area to be explored. This drone image data can include surface mineral data and 3D terrain data. Ground acquisition equipment can be a six-legged exploration robot equipped with an all-terrain chassis and robotic arm, integrating ground-penetrating radar, microseismic sensors, temperature and humidity sensors, etc., to autonomously navigate to the area to be explored and collect geotechnical parameter data and geotechnical environmental data. The geotechnical parameter data can include geotechnical temperature, geotechnical humidity, geotechnical type, etc., while the geotechnical environmental data can include groundwater level, gas concentration, etc. Ground acquisition equipment can be drilling rigs, such as those that control the torque, rotation speed, and borehole pressure sensors of the drilling rig to achieve real-time acquisition. During the acquisition process, the drilling speed and mud parameters of the drilling rig can be automatically adjusted using the LSTM (Long Short-Term Memory Algorithm) algorithm to reduce the risk of stuck drill bit. Fiber optic grating sensor arrays can be deployed along the borehole to monitor changes in ground stress and temperature, and collect underground geological environment data. This underground geological environment can include data such as underground ground stress and temperature, and can be used to construct underground stress field models. Surface mineral spectral data can include the types and spatial distribution information of various minerals; three-dimensional topographic data can include elevation information and surface undulation information of the area to be explored; rock and soil parameter data can include rock and soil type, density, distribution location, and content; rock and soil environmental data can include water level and gas concentration; and underground geological environment data can include underground stress field data and temperature change data.

[0030] In this embodiment, various types of data, including airborne, ground-based, and underground data, can be integrated. These data are preprocessed, and then spatial and temporal consistency processing is performed on the preprocessed data to obtain multi-source geological data. Preprocessing of various data types can be based on edge computing nodes. Data integrity can be checked, missing or invalid values ​​can be identified and processed, and data cleaning can be performed on the multi-source geological data. For the cleaned multi-source geological data, visualization tools such as box plots and scatter plots can be used to check the distribution of the cleaned multi-source geological data, ensuring that outliers are effectively handled. A comparison report of data before and after cleaning can also be output, recording the number of outliers, processing methods, and data quality scores. By preprocessing data based on edge computing nodes, the pressure on subsequent cloud processing can be reduced. Simultaneously, spatial and temporal consistency processing eliminates coordinate and temporal deviations in the multi-source data, avoiding analytical errors caused by data misalignment, and laying a high-quality data foundation for subsequent modeling and analysis.

[0031] In this embodiment, while preprocessing multi-source geological data through edge computing nodes, the multi-source geological data can also be stored on the blockchain based on the deployed blockchain evidence storage module.

[0032] When airborne, ground-based, and underground data acquisition devices generate raw data, the blockchain storage module integrated within the edge computing node extracts the hash value of the raw data in real time and uploads it to the blockchain. Simultaneously, it also records and uploads the corresponding metadata, such as acquisition time, acquisition location coordinates, and the device identifier of the acquisition device, to the blockchain.

[0033] By introducing a blockchain evidence storage mechanism during the data acquisition and preprocessing stages, and uploading the hash value and metadata of the original data to the blockchain in real time, it is possible to ensure that the source of all underlying exploration data is absolutely traceable and tamper-proof, thus ensuring the compliance and auditability of the original geological data in the deliverables of the exploration project and effectively avoiding the risk of data being tampered with.

[0034] In this embodiment, multi-source geological data can be mapped onto a preset three-dimensional grid. Each grid cell contains mineral parameters, soil and rock parameters, environmental parameters, and topographic parameters. Mineral parameters may include mineral type and content; soil and rock parameters may include soil and rock type, density, porosity, permeability, and compressive strength; environmental parameters may include groundwater level, soil and rock temperature, soil and rock moisture, and geostress; and topographic parameters may include elevation and slope. Interpolation can be performed based on the data distribution characteristics in the three-dimensional grid to fill in missing grid cells. A pre-trained deep learning model can be used to segment geological structural lines such as faults and folds to determine the location of geological interfaces. The grid data can be converted into a three-dimensional surface to generate a continuous three-dimensional model. Geological parameters can be mapped onto the three-dimensional model to achieve visualization of the geological parameters. During the determination of the three-dimensional geological model, the spatial structure data corresponding to the three-dimensional geological model can be acquired simultaneously. For example, based on the abrupt boundary of the grid cell attributes, the traveling cube algorithm can be used to extract the three-dimensional irregular triangular meshes between different strata and fault planes. The vertex coordinate set of these triangular meshes and the triangular facet connection table are used as the spatial structure data.

[0035] In this embodiment, geological data of the area to be explored is collected from multiple dimensions, including airborne, ground, and underground sources, to comprehensively obtain the geological information of the area. After preprocessing the data, a three-dimensional geological model is generated, and the corresponding spatial structure data is obtained. Compared with the traditional single method of data collection and modeling, the obtained three-dimensional geological model can accurately reflect the geological conditions of the area to be explored.

[0036] Step 102: Input multi-source geological data, geological report data of the area to be explored, and spatial structure data into the pre-trained prediction model. The pre-trained prediction model determines the multimodal characteristics of the area to be explored and determines the geological state prediction value of each grid cell in the three-dimensional geological model based on the multimodal characteristics. Update the three-dimensional geological model according to the geological state prediction value.

[0037] In this embodiment, the multi-source geological data includes surface mineral spectral data and three-dimensional topographic data collected from the air, rock and soil parameter data and rock and soil environmental data collected from the ground, and underground geological environmental data collected from the ground; the geological report data may include historical text information such as geological survey reports, borehole columnar section descriptions, and stratigraphic age records corresponding to the area to be explored.

[0038] In this embodiment, a pre-trained prediction model can be used to extract features from multi-source geological data, geological report data, and spatial structure data, respectively. The extracted features are then fused to obtain multimodal features, and the predicted geological state values ​​of each grid cell in the 3D geological model are determined based on these multimodal features. The predicted geological state values ​​of each grid cell can include predicted values ​​of geological parameters and classification results. Specifically, the predicted geological parameters can include at least one or more numerical parameters from the categories of mineral parameters, soil and rock parameters, environmental parameters, and topographic parameters, which can be determined according to the actual situation. The classification results can include at least one or more discrete parameters from the categories of soil and rock type and structural type, which can also be determined according to the actual situation. For example, the predicted values ​​of mineral parameters may include mineral content, the predicted values ​​of soil and rock parameters may include density, porosity, permeability, compressive strength, etc., the predicted values ​​of environmental parameters may include groundwater level, soil and rock temperature, soil and rock moisture, geostress, etc., and the predicted values ​​of topographic parameters may include elevation, slope, etc. The soil and rock type in the classification results is used to mark the lithology label corresponding to the grid unit, and the structure type is used to mark whether the grid unit belongs to an abnormal structure label such as a fault zone or fold. The specific values ​​can be determined according to the actual situation.

[0039] After obtaining the predicted geological conditions, the 3D geological model constructed in the previous step can be updated. For example, the classification results can be used as new discrete attribute labels, written into and assigned to the classification attributes of the corresponding grid cells. The known numerical parameters measured and written by multi-source acquisition devices in the previous step remain unchanged. The predicted values ​​of the numerical parameters obtained by the prediction model are used to overwrite and replace the initial values ​​of the corresponding grid cells in the previous step, completing the update of the 3D geological model. This pre-trained prediction model can collect historical data from the surveyed area, divide the historical data into multiple data blocks according to spatial regions or time series, and distribute and store them in a cloud storage service for distributed parallel model training. This historical data includes historical multi-source geological data, historical grid spatial structure data, historical geological report data, and measured values ​​of geological parameters obtained through laboratory analysis and field testing, as well as historical geological classification labels. Based on the above method, multimodal features of historical data are extracted. The output multimodal feature vectors are paired with the corresponding measured values ​​of historical geological parameters and the true labels of historical geological classification to form standard training sample pairs. The composite loss function constructed by the cascaded mean square error and cross-entropy loss is used as the multi-task optimization objective. The gradient is calculated by the backpropagation algorithm and the network parameters of the entire prediction model are dynamically optimized until the prediction error and classification residual of the model converge to a minimum threshold below the preset threshold. Then the model weights are solidified to obtain the pre-trained prediction model.

[0040] By integrating the multimodal features of multi-source geological data and geological report data, and introducing spatial structure data, the geological attributes of the area to be explored can be more comprehensively characterized compared to the features of a single data source, making the geological parameters in the predicted three-dimensional geological model closer to the actual situation.

[0041] Step 103: Based on the updated 3D geological model, input the current geological state of the multi-source acquisition device at its current location and the current control parameters of the multi-source acquisition device into the reinforcement learning-based feedback control model to generate acquisition control instructions for the area to be explored. Based on the acquisition control instructions, control the multi-source acquisition device to continue to perform multi-source geological data acquisition in the area to be explored. The acquisition control instructions include acquisition frequency, acquisition trajectory and device parameters.

[0042] In this embodiment, after updating the 3D geological model, local grid slices of each multi-source acquisition device within its current position and the predicted path ahead can be obtained from the updated 3D geological model based on the current spatial coordinates and direction of travel of each multi-source acquisition device. The predicted geological state values ​​(including numerical parameter predictions and discrete classification results) corresponding to each grid cell within these local grid slices are extracted and encapsulated into corresponding local state context vectors, serving as the current geological state of each multi-source acquisition device at its current position.

[0043] In this embodiment, the current geological state and current control parameters are concatenated using feature vectors to construct a current state vector. Specifically, the current control parameters of each multi-source acquisition device during its current operation can be obtained, such as the torque, rotation speed, and borehole wall pressure of the drilling rig, and the current moving speed and heading angle of the hexapod exploration robot or drone swarm, to determine the current control parameters of each multi-source acquisition device. The current geological state of each multi-source acquisition device is then concatenated with its own current control parameters using corresponding feature vectors to construct the current state vector of each multi-source acquisition device.

[0044] In this embodiment, the current state vector is input into the feedback control model to calculate the probability distribution corresponding to each candidate acquisition action. For example, this feedback control model is built based on a reinforcement learning algorithm and may contain parallel multi-action output branches or sub-policy networks deployed for different devices. The current state vectors of each multi-source geological device are input separately according to device type, and the corresponding sub-policy networks decouple and output corresponding control commands, thereby eliminating dimensional conflicts in the action spaces of heterogeneous devices and ensuring the convergence speed of the feedback control model. After the current state vectors of each multi-source geological device are input into the feedback control model, the corresponding policy network performs forward mapping calculations and outputs the probability distribution corresponding to the candidate acquisition actions of the multi-source acquisition devices in the discrete action space under the current state.

[0045] To ensure the real-time performance and safety of construction site control, offline inference can be performed based on the previously trained feedback control model. To achieve continuous optimization of the feedback control model, during the intervals between data acquisition actions by the multi-source acquisition devices, the multi-source geological data actually acquired after the execution of the acquisition control command, which is synchronously transmitted back from the site, and the real-time equipment mechanical state data of the multi-source acquisition devices are collaboratively compared to dynamically calculate the multi-objective optimization reward value. The equipment mechanical state data includes the motor output power and kinematic attitude angle of the airborne acquisition device; the joint drive load stress and three-dimensional tilt angle of the ground acquisition device; and the motor torque and borehole wall pressure of the underground acquisition device. For example, when calculating the multi-objective optimization reward value, the acquisition frequency or acquisition trajectory density in the acquisition control command can be cross-matched with the geological state prediction value of the current grid cell determined based on multi-modal characteristics. If it is determined that the acquisition control command performed encrypted sampling in anomaly areas, a positive reward item is calculated. By extracting the motor output power of the airborne acquisition device, the joint drive load stress of the ground acquisition device, and the motor torque of the underground acquisition device, the overall electrical energy and mechanical power consumption loss are calculated, and negative rewards are determined. The system employs a power consumption penalty term. Threshold compliance checks are performed on the kinematic attitude angle of the airborne acquisition device, the three-dimensional tilt angle of the ground acquisition device, and the motor torque and borehole pressure of the underground acquisition device. If any hardware state data exceeds a preset safety boundary threshold, a negative risk penalty term is determined to quantify the risks of drone crashes, robot rollovers, and drill bit overload jamming. The positive reward term, power consumption penalty term, and risk penalty term are weighted and summed to obtain the final multi-objective optimization reward value. Finally, the gradient update algorithm of reinforcement learning is used to asynchronously train and update the policy network parameters of the feedback control model offline, so that during the actual execution inference stage, the updated model parameters are used to generate acquisition control commands for the area to be explored.

[0046] In this embodiment, the candidate acquisition action with the highest probability in the probability distribution of each multi-source acquisition device is determined as the target control action. Specifically, the candidate acquisition action with the highest probability value in the probability distribution can be directly selected as the current target control action by calling a maximum value index function (such as the argmax function).

[0047] In this embodiment, the acquisition frequency, acquisition trajectory, and equipment parameters are determined based on the target control action to generate acquisition control commands. Specifically, after obtaining the target control action value, the abstract action value is converted into the underlying drive values ​​of the corresponding sensors and physical actuators by calling the built-in action hardware physical parameter mapping and parsing dictionary. This accurately determines the acquisition frequency, acquisition trajectory, and equipment parameters, generating acquisition control commands for the area to be explored. For example, the acquisition frequency is used to dynamically fine-tune the sampling rate of geophysical sensors such as ground-penetrating radar, microseismic sensors, or spectrometers mounted on the equipment; the acquisition trajectory is used to fine-tune or regenerate the flight waypoints of the UAV swarm or the travel path of the hexapod robot online; and the equipment parameters are used to dynamically adjust the operating status of the mechanical hardware, such as the frequency converter drilling speed, motor torque, and mud pump flow parameters of the intelligent drilling system. Finally, the acquisition control commands containing these physical characteristics are sent to the actuators of the corresponding acquisition equipment to control the multi-source acquisition equipment to continue to perform real-time acquisition of multi-source geological data in the area to be explored.

[0048] This approach enables closed-loop adaptive feedback control between front-end data acquisition and back-end prediction. For example, if the current state vector indicates that the strata ahead are simple and the soil and rock compressive strength is low, the control command can adaptively reduce the sensor's acquisition frequency and increase the trajectory speed to improve exploration efficiency. Conversely, when a hidden fault is predicted ahead or the rock hardness suddenly increases, the control command adaptively adjusts the acquisition frequency to the highest level for intensive sampling and fine-tunes the equipment parameters online to control the equipment to slow down or correct the acquisition trajectory, ensuring that the multi-source acquisition equipment can completely acquire geological data. By feeding back the acquired multi-source geological data and equipment parameters to the back-end, and updating the model parameters and dynamically overwriting the 3D geological model through continuous learning technology, a continuous data control closed loop is formed, thereby ensuring the integrity of key geological data acquisition and further improving the overall accuracy of digital engineering exploration operations.

[0049] In one embodiment of this disclosure, determining a three-dimensional geological model corresponding to the area to be explored based on multi-source geological data includes: Multi-source geological data is mapped onto a 3D grid, and the geological parameters corresponding to each grid cell in the 3D grid are determined. Based on the geological parameters corresponding to each grid cell in the 3D grid, the geological features corresponding to each grid cell are determined, and the geological features corresponding to each grid cell are fused to obtain the feature vector corresponding to the 3D grid. Geological structural lines and stratigraphic distribution information of the 3D mesh are determined based on the feature vectors, and the 3D mesh is converted into a 3D geological model based on the geological structural lines and stratigraphic distribution information. Geological parameters are mapped onto a three-dimensional geological model to obtain a three-dimensional geological model of the area to be explored. Among them, geological parameters include mineral parameters, soil and rock parameters, environmental parameters, and topographic parameters; geological features include mineral features, soil and rock features, environmental features, and topographic texture features.

[0050] In this embodiment, a three-dimensional mesh can be constructed according to the scope and accuracy requirements of the area to be surveyed. As an example, the mesh can be divided into planar and vertical dimensions to form a three-dimensional mesh structure with planar units and vertical layers, and each mesh unit corresponds to a unique spatial coordinate.

[0051] In this embodiment, multi-source geological data can be matched to corresponding grid cells according to spatial coordinates. For surface mineral spectral data, mineral parameters can be calculated for each grid cell. Since surface mineral data mainly reflects shallow information, these mineral parameters can be assigned to the first distance from the surface to the subsurface, such as a vertically layered grid at a depth of 5 meters. For three-dimensional topographic data, topographic parameters can be extracted from each planar grid cell. For example, if a grid cell has an elevation of 250m, a slope of 15°, and a northeast orientation, it can be mapped to all vertically layered deep grid cells under that spatial coordinate. This allows each deep grid cell to calculate its actual burial depth from the surface based on its own Z-axis coordinate, serving as the spatial topological prior feature of the deep grid cell. For soil and rock parameters... According to reports, the soil and rock parameters collected by the robot at the sampling point can be diffused to the surrounding grid cells through a spatial interpolation algorithm. For example, the parameter values ​​can be assigned according to distance weight with the sampling point as the center, and the closer the distance, the greater the weight, to ensure that the grid cell within a preset distance range, such as 50 meters, can have the characteristics of the soil and rock parameters of the sampling point. For underground geological environment data and soil and rock environment data, the environmental parameters corresponding to different depths can be mapped to the vertically layered grid cells of the corresponding depth according to the three-dimensional coordinates of the borehole location. Then, the grid parameters of the area without boreholes can be supplemented by interpolation. For example, the in-situ stress value of the grid cell in the middle area can be calculated using the in-situ stress data of three adjacent boreholes.

[0052] When determining the geological parameters of each grid cell, data compatibility processing can be performed. If multiple data sources exist for the same grid cell, such as both a robot and a drilling rig collecting water content data at a certain depth, the source with higher data reliability should be used as the final value. For example, drilling data comes directly from deep underground layers, and its reliability is higher than that of shallow sampling data from the robot, so drilling data should be used first. If a grid cell has no directly collected data, in addition to using the interpolation method mentioned above, the parameter trends of adjacent grid cells can also be referenced for assignment. For example, if the groundwater level is known to gradually rise from west to east, a grid cell without data can be assigned a value according to this trend.

[0053] In this embodiment, the geological features corresponding to each grid cell can be determined based on the geological parameters of each grid cell. For example, mineral spectral features corresponding to mineral parameters, topographic texture features corresponding to topographic parameters, and rock and soil features corresponding to rock and soil parameters can be extracted by convolutional neural networks. Temporal variation features corresponding to environmental parameters can be extracted by recurrent neural networks. The weights of various geological features can be determined, and the geological features corresponding to each grid cell can be obtained by weighted fusion of the features.

[0054] In this embodiment, the geological features corresponding to all grid cells in the 3D mesh can be fused to obtain the feature vector corresponding to the 3D mesh. As an example, the geological features of each grid cell can be standardized first, mapping the values ​​of different types of geological features to the same data range. The weighted feature sequences of all grid cells can be sequentially concatenated according to the spatial coordinate order of the 3D mesh to form the original feature matrix of the 3D mesh. This original feature matrix can be converted into a one-dimensional vector, and all elements in the matrix can be arranged sequentially in row-major order to obtain the feature vector corresponding to the 3D mesh.

[0055] In this embodiment, geological structural lines of a 3D mesh can be determined based on a pre-trained geological structure recognition model. This geological structural model can include convolutional layers, recurrent layers, and an output layer. The convolutional layer extracts local features from the input feature vector, capturing spatial relationships between mesh cells. The recurrent layer processes the feature sequence output by the convolutional layer, mining the continuity of geological structures in 3D space. The output layer uses a softmax function to map features to a probability distribution of structural types, such as faults and folds, marking mesh cells with probabilities greater than a preset structural threshold, such as 0.8, as candidate cells. For candidate cells, continuous fault lines can be extracted using morphological operations. For example, adjacent fault candidate cells can be connected to form initial line segments, and linear fitting can be used to eliminate discrete point interference, resulting in continuous geological structural lines.

[0056] In this embodiment, a clustering algorithm can be used to group feature vectors, calculate the cosine similarity of feature vectors of different grid units, and classify grid units with similarity greater than a preset similarity threshold, such as 0.9, into the same stratigraphic group. For each stratigraphic group, the rock types in the stratigraphic group can be determined, and the spatial distribution range of the stratigraphic group in the three-dimensional grid can be counted. The depth between the top and bottom interfaces of each stratum and the boundary coordinates of each stratum on the plane can be recorded. The rock type, spatial distribution range, depth, and boundary coordinates corresponding to each stratigraphic group can be determined as the stratigraphic distribution information corresponding to each stratigraphic group.

[0057] In this embodiment, based on stratigraphic distribution information, grid cells of the same stratigraphic layer can be merged into a continuous stratigraphic volume, and a three-dimensional surface of the top and bottom surfaces of the stratigraphic layers can be generated using a triangulation algorithm. For geological structural lines that cross the stratigraphic layers, such as faults, the stratigraphic volume can be cut. For example, according to the strike and dip of the fault, the stratigraphic volume separated by the fault can be divided into two parts, causing the two parts to be displaced along the fault line by a preset distance, simulating the fault's effect on the stratigraphic layers. For fold structures, the shape of the stratigraphic surface can be adjusted so that the stratigraphic layers in the anticline axis bend upwards and gradually tilt in the limbs, while the stratigraphic layers in the syncline axis bend downwards, maintaining the relative consistency of the stratigraphic thickness.

[0058] In this embodiment, the geological parameters of each grid cell, such as mineral content, soil density, and groundwater level, are bound to their corresponding spatial locations in the 3D geological model. For a stratigraphic body, the average geological parameters of all grid cells within that stratigraphic body can be calculated as the overall attribute of that stratigraphic body. For geological structural zones, such as fault zones, the geological parameters within their respective ranges can be individually marked to clearly distinguish them from surrounding strata. Geological parameters can be visualized using color mapping techniques; for example, a gradient from red to blue can represent mineral content from high to low, and color blocks of varying transparency can represent the depth of groundwater levels. This allows the 3D geological model to not only display spatial morphology but also intuitively reflect the distribution characteristics of geological parameters.

[0059] By mapping multi-source geological data onto a 3D mesh and determining the geological parameters and features of each mesh cell, the feature vector of the resulting 3D mesh, after feature fusion processing, integrates multi-dimensional information. Compared to single features, this provides a more comprehensive characterization of geological attributes, leading to more accurate extraction of geological structural lines and stratigraphic distribution information. This improves recognition efficiency and accuracy while reducing subjective errors. Through stratigraphic modeling and structural cutting, the 3D geological model can realistically reproduce the spatial relationships of underground geological structures. Combined with the attribute mapping of geological parameters, the final 3D geological model includes both morphological information and attribute features, providing a comprehensive geological reference for exploration scenarios and effectively improving the accuracy of geological analysis and decision-making.

[0060] In one embodiment of this disclosure, multi-source geological data is mapped onto a three-dimensional mesh, and the geological parameters corresponding to each mesh cell in the three-dimensional mesh are determined, including: Determine the coordinate mapping relationship between the area to be explored and the three-dimensional grid, map the multi-source geological data to the corresponding grid cells based on the coordinate mapping relationship, and determine the geological parameters corresponding to each grid cell; For each type of geological parameter in multi-source geological data, the grid cells in the three-dimensional grid that lack geological parameters are identified as the first grid cell, and the grid cells in the three-dimensional grid that contain geological parameters are identified as the second grid cell. For each first grid cell, the estimated value of the geological parameter in the first grid cell is predicted based on the value of the geological parameter in each second grid cell.

[0061] In this embodiment, the multi-source geological data generated by airborne acquisition equipment, ground acquisition equipment and underground acquisition equipment when performing real-time exploration of the area to be explored can be mapped to the three-dimensional grid cells with corresponding spatial coordinates through this coordinate mapping relationship.

[0062] Because field survey sampling points are spatially discrete and sparse, there will be blank areas in the mapped 3D grid that are not covered by the collected data. Therefore, for each type of geological parameter in the multi-source geological data, the entire 3D grid can be scanned: the grid cells that lack this type of geological parameter are identified as the first grid cell; and the grid cells that already contain data for this type of geological parameter are identified as the second grid cell.

[0063] Subsequently, for geological parameters with numerical continuity that conform to the characteristics of continuous spatial distribution (such as mineral content, soil density, porosity, permeability, compressive strength, groundwater level, soil temperature and humidity, and geostress), the inverse distance weighted method or Kriging spatial interpolation algorithm can be used. Based on this spatial interpolation algorithm, the interpolation weight of each second grid cell corresponding to the first grid cell is calculated according to the three-dimensional spatial distance or semi-variogram between the second grid cell of each known data and the first grid cell to be interpolated. Then, the interpolation weight of each second grid cell is weighted and summed with the known observed values ​​of the corresponding geological parameters in each grid cell, and the final weighted sum is determined as the estimated value of the corresponding geological parameter in the first grid cell.

[0064] For discontinuous geological parameters (such as soil type, mineral type, and stratigraphic name) that conform to discrete spatial distribution characteristics, since these parameters are non-numerical text labels, they cannot be directly multiplied, divided, or weighted summed. A K-nearest neighbor classification interpolation algorithm or a mode decision voting mechanism can be used. Specifically, a neighborhood set is formed by searching for a predetermined number (e.g., K) of known second grid cells that are closest to the first grid cell in space. The frequency of each discrete label (e.g., granite, sandstone) in this neighborhood set is counted, and the category label with the highest frequency (i.e., the mode voting result), or the category attribute value of the second grid cell with the absolute closest spatial distance, is directly assigned to the first grid cell.

[0065] Furthermore, for terrain parameters (such as elevation and slope), since these are two-dimensional surface attributes, spatial interpolation is performed only in the surface two-dimensional grid cells of the three-dimensional grid to completely fill in the missing terrain data. Subsequently, the interpolated surface terrain parameters are projected and directly mapped to the corresponding vertically layered deep grid cells under the spatial coordinates, making them the inherent terrain prior attributes of the deep grid cells (so that each deep grid cell can combine its own...). Z The axis coordinates are used to calculate the actual burial depth of the current grid cell from the ground surface, thus avoiding the physical meaningless volume weighted interpolation in the vertical depth direction and ensuring the rigor of the physical and spatial logic within the entire 3D mesh model.

[0066] Through the heterogeneous spatial interpolation processing based on coordinate mapping and parameter type adaptation, the chaotic and discrete multi-source raw geological data from space, air, and ground can be accurately and completely incorporated into a standardized 3D grid data base. This not only eliminates the mathematical calculation errors that inevitably arise when dealing with discrete geological labels using traditional single-weighted interpolation, but also preserves to the greatest extent the local abrupt changes at geological interfaces and the smooth transition characteristics of physical continuity attributes. This provides high-quality, zero-missing-value, and structured basic data support for subsequent extraction of high-precision geological features, construction of high-fidelity 3D geological models, and large-scale model inference using deep neural networks.

[0067] In one embodiment of this disclosure, a pre-trained prediction model determines the multimodal features of the area to be explored, and based on the multimodal features, the predicted geological state values ​​of each grid cell in the three-dimensional geological model are determined. This can be achieved in the following way: In this embodiment, the pre-trained prediction model can be a hybrid Transformer model, including a multi-branch feature extraction layer, a multi-modal feature fusion layer, and a joint inference output layer. The multi-branch feature extraction layer includes a visual encoding sub-network layer, a text extraction sub-network layer, a temporal encoding sub-network layer, a 3D convolutional sub-network layer, a structure encoding sub-network layer, and a structure extraction layer.

[0068] The multi-branch feature extraction layer independently extracts features using its internal parallel sub-network layers. Specifically, for surface mineral spectral data, the visual encoding sub-network layer performs image patch segmentation and self-attention computation to extract and output the corresponding first semantic features. For geological report data, the text extraction sub-network layer (e.g., BERT model) performs word embedding and bidirectional Transformer encoding to extract geological entities and conclusive descriptions, obtaining the second semantic features. Simultaneously, underground geological environment data (e.g., continuous monitoring data of ground stress and groundwater level) and geotechnical environment data (e.g., water level, temperature, humidity, etc., stored in time series) are acquired. Temporal features with time dependence are extracted using the time encoding sub-network layer (e.g., Long Short-Term Memory network layer). Three-dimensional terrain data is processed through a 3D convolution sub-network layer to capture spatial elevation undulations and local geometric topological relationships, extracting corresponding spatial features. Geotechnical parameter data in data blocks is dimensionality-reduced and mapped to high-dimensional space using a structured encoding sub-network layer, extracting corresponding geotechnical features. Finally, a structure extraction layer uses multi-layer graph convolution or graph attention mechanisms to extract features from spatial structure data, obtaining structural features.

[0069] Then, a multimodal feature fusion layer is used to determine multimodal features based on the global dependencies between the first semantic feature, the second semantic feature, temporal features, spatial features, geotechnical features, and structural features. Specifically, the first semantic feature, the second semantic feature, the temporal feature, the spatial feature, the geotechnical features, and the structural features are input as local fusion sequences into the multimodal feature fusion layer. A cross-attention module performs feature fusion processing on the first semantic feature and the second semantic feature based on a cross-attention mechanism, outputting image-text fused features. Furthermore, a global self-attention module obtains the correlation weights between each feature, and based on these correlation weights, feature fusion processing is performed on each feature to obtain the multimodal feature vector.

[0070] Finally, through the joint inference output layer, the predicted geological state values ​​of each grid cell in the 3D geological model are determined based on multimodal features. This predicted geological state value is a composite multi-task output tensor, which includes predicted geological parameters and classification results, decoupling the output of continuous physical values ​​and discrete category labels across different feature channels. The multi-task fully connected output head within the joint inference output layer can adaptively call the corresponding activation function to output the appropriate channel based on the parameter type of the predicted geological state value. The predicted geological state value can include predicted geological parameters and classification results.

[0071] In this embodiment, the pre-trained prediction model can be packaged and deployed in a low-latency distributed stream processing cluster. After receiving the real-time collected data, the geological state prediction value and classification result can be output with low latency.

[0072] By employing a multi-branch feature extraction layer and a dual attention mechanism, the global dependencies between multi-source geological data, geological reports, and spatial structures are deeply explored, improving the inference accuracy for attributes of unknown underground blind areas. By decoupling continuous physical values ​​from discrete category labels, gradient conflicts during backpropagation of heterogeneous tasks are overcome, ensuring the stability of predictions. Simultaneously, by utilizing the encapsulation and deployment of a low-latency distributed stream processing cluster, the traditional time-consuming static geological modeling is compressed into a rapid response during field operations, eliminating the time lag in data processing and enabling dynamic rolling updates of the 3D geological model.

[0073] In one embodiment of this disclosure, multimodal features are determined based on the global dependencies among first semantic features, second semantic features, temporal features, spatial features, soil and rock features, and structural features. This can be achieved in the following ways: First, based on the cross-attention mechanism, the first semantic feature and the second semantic feature are fused to obtain the fused feature.

[0074] In this embodiment, the cross-attention module maps the acquired first semantic feature and second semantic feature to different feature spaces, which serve as the query vector matrix, key vector matrix, and value vector matrix of the cross-attention mechanism, respectively.

[0075] Specifically, the cross-attention module calculates the similarity between the query vector matrix and the key vector matrix through dot product operation to quantify the cross-modal correlation between the spatial regions of the spectral image and the vocabulary of the geological report text. Subsequently, the calculated similarity matrix is ​​input into the Softmax activation function for normalization, converting the dot product similarity values ​​into cross-modal adaptive attention weights distributed in the [0-1] interval, and ensuring that the sum of the weights of all dimensions is 1.

[0076] Next, the cross-attention module performs element-wise multiplication of the normalized attention weights with the value vector matrix, dynamically empowering the key entity descriptions in the text semantics and mapping them to the corresponding spectral image spatial features. Finally, the multiplied feature matrix is ​​residually connected with the query vector matrix, and a fully connected layer is used to reconstruct the high-dimensional features and perform nonlinear spatial mapping on the fusion intermediate results, thereby outputting the fused features.

[0077] Secondly, based on the self-attention mechanism, the correlation weights between each feature in the fusion feature, temporal feature, spatial feature, soil and rock feature and structural feature are determined. Based on the correlation weights, feature fusion processing is performed on each feature to obtain multimodal features.

[0078] In this embodiment, the global self-attention module receives the fused features and acquires the temporal features, spatial features, geotechnical features, and structural features output from the aforementioned steps in parallel. These different feature vectors—fusion features, temporal features, spatial features, geotechnical features, and structural features—are concatenated and combined into a global multimodal feature input sequence, with each feature vector serving as an independent feature mode within the sequence.

[0079] Subsequently, the global multimodal feature input sequence is transformed into high dimensions through three independent linear mapping fully connected layers, thereby constructing the query matrix, key matrix, and value matrix corresponding to each feature modality in the global view.

[0080] During the calculation process, the module calculates the dot product similarity between all pairs of feature modes in the global multimodal feature input sequence to mine and calculate the nonlinear spatial association weights between different geological feature modes; then, the obtained global similarity matrix is ​​normalized to determine the global attention association weights between each feature mode and other feature modes.

[0081] Subsequently, the global attention associated weights for each feature modality are weighted and multiplied with their corresponding value matrices to obtain the weighted feature vectors of each feature modality under global dependencies. Next, the weighted feature vectors of all feature modalities are concatenated according to a fixed spatial sequence to construct a global fusion feature matrix. Finally, a fully connected layer combined with a nonlinear activation function is used to perform dimensional integration, deep decoupling, and information enhancement processing on the global fusion feature matrix in a multidimensional feature space, thereby obtaining multimodal feature vectors.

[0082] The method achieves alignment and strong cross-modal correlation mapping between the first and second semantic features through the aforementioned cross-attention mechanism. Then, a self-attention mechanism captures the global nonlinear dependencies between the fused features and subsurface temporal, spatial topological, and inherent soil and rock characteristics. Compared to traditional static weighted fusion methods such as arithmetic averaging or simple matrix concatenation, this method dynamically and adaptively allocates feature weights based on the real-time spatial distribution and anomalous mutation points of each modality's data. This allows the final multimodal features to comprehensively and three-dimensionally reflect the inherent physical connections and causal evolution patterns between data from different geological sensing layers, thereby improving the overall accuracy of predictions in the subsequent joint inference output layer.

[0083] In one embodiment of this disclosure, when determining an exploration planning scheme, the density and depth of drilling sites can be determined based on predicted values ​​of mineral parameters, such as mineral type and estimated content. Combined with predicted values ​​of soil and rock parameters, such as soil and rock compressive strength and porosity, suitable drilling equipment can be selected; for example, a diamond drill bit can be used for high-hardness soil and rock. The patrol routes of the UAV swarm can be planned based on predicted underground geological environment parameters, such as groundwater level depth and ground stress magnitude, to avoid areas with excessively high groundwater levels causing surface water accumulation, and the inspection paths of the six-legged exploration robot can be optimized, such as traveling along terrain slopes less than 15° to reduce equipment wear. The planning results of exploration areas at all levels can be integrated to generate an exploration planning scheme that includes drilling site coordinates, equipment selection lists, and work sequence arrangements.

[0084] In one embodiment of this disclosure, a preset number of verification sampling points, such as 5-10, can be selected within the area to be explored. The actual geological parameters of the verification sampling points are collected using a drilling device. The deviation rate between the predicted values ​​of the geological parameters in the three-dimensional geological model corresponding to the verification sampling points and the actual geological parameters can be calculated. If the deviation rate is greater than a preset deviation threshold, such as 15%, the attention weight when determining multimodal features is adjusted, such as increasing the weight ratio of the first semantic feature and temporal feature associated with the verification sampling point. The steps of determining multimodal features based on multi-source geological data and geological report data and updating the predicted values ​​of geological parameters based on multimodal features are then re-executed. If the deviation rate is less than or equal to the preset deviation threshold, the current three-dimensional geological model and the predicted values ​​of geological parameters are stored as the final result and associated with the geographic information database of the area to be explored for historical data tracing and scheme iteration optimization in subsequent exploration operations.

[0085] In one embodiment of this disclosure, after determining the predicted values ​​of geological parameters in the three-dimensional geological model based on multimodal features, the method may further include verifying and correcting the predicted values: comparing the predicted values ​​with the measured values ​​obtained from borehole sampling, and calculating the deviation rate between the two; if the deviation rate is greater than a preset deviation threshold, adjusting the weights of the multimodal features based on the measured values, and re-inputting the predicted values ​​into the prediction model to obtain corrected predicted values; if the deviation rate is less than or equal to the preset deviation threshold, retaining the predicted values. This approach can further improve the accuracy of geological parameter prediction, making the three-dimensional geological model more closely reflect actual geological conditions.

[0086] In one embodiment of this disclosure, after the exploration planning process is performed on the area to be explored based on the predicted values, a planning report can be generated. The report may include a distribution map of the primary exploration area, an equipment acquisition plan table, a path diagram, and dynamic adjustment rules, such as setting deviation thresholds and adjustment procedures, and ensuring the feasibility and traceability of the planning scheme.

[0087] In one embodiment of this disclosure, the prediction results can be pushed to a mobile device or a monitoring screen in real time to guide on-site personnel to adjust the data collection strategy. Based on reinforcement learning algorithms, the flight path of the geotechnical drone, the robot path, and the drilling parameters can be dynamically adjusted according to the prediction results. Newly collected data can be fed back to the pre-trained prediction model in real time, and the model parameters can be updated through incremental learning or continuous learning techniques. The system operation status can also be monitored, including indicators such as data processing speed, model prediction accuracy, and resource utilization.

[0088] In one embodiment of this disclosure, a spatiotemporal database design can be used to uniformly manage various structured and unstructured data. A database supporting spatiotemporal data storage and querying can be selected, and a database table structure can be designed that includes spatial data fields, such as geometric body type and coordinate reference system; temporal data fields, such as timestamps and time-series indexes; and unstructured data fields, such as geological report data and image URLs. This system can convert and import spatial data such as UAV LiDAR point cloud data and geological structural lines into database-supported formats; convert sensor monitoring data and UAV-collected time-series data into timestamp formats and store them in time series; store unstructured data such as geological reports and field photos as BLOB type files with file paths, establishing associations with spatial / temporal data; use the ST_Contains and ST_Intersects spatial functions to query geological body inclusion and intersection relationships; utilize TimeScaleDB's time-series indexes for time range queries, such as water level changes over the past 24 hours; perform complex queries using SQL statements combined with spatial, temporal, and attribute conditions, such as querying changes in sandstone content in a certain area over the past week; and create indexes for frequently queried fields, such as spatial index GIST and temporal index BRIN, using partitioning table technology to divide data by time and space.

[0089] In one embodiment of this disclosure, a data dictionary can be defined to improve cross-departmental data compatibility. Key attributes of exploration data, such as geological body name, mineral content, and coordinate system, can be identified. Unified data element standards, such as name, code, data type, and value range, can be established. Extended fields can be reserved for non-core data, allowing departments to customize data elements according to their needs, such as drilling equipment model and rock weathering degree. The data element standards are organized into a tree structure or table format, including data element ID, name, definition, format, and example fields. A bilingual (Chinese and English) data dictionary is provided to meet internationalization requirements. Version control is implemented for the data dictionary, recording change history (e.g., V1.0: initial version; V1.1: added "rock weathering degree" data element). At the data acquisition layer, raw data is standardized and mapped, converting departmental customized data into standard data elements in the data dictionary. Before data is entered into the database, compliance checks are performed using the data dictionary, rejecting data that does not meet the standards (e.g., coordinate system not WGS84). Data dictionary training is provided to cross-departmental users, along with mapping tools and API interfaces, to lower the barrier to entry.

[0090] In one embodiment of this disclosure, data sharing and retrieval can be achieved through an API gateway, which integrates data resources from a spatiotemporal database. CRUD operations are defined for each resource, such as create, read, update, and delete operations, supporting conditional queries such as filtering by time or spatial range. API requests are authenticated and authorized. HTTPS is used for encrypted data transmission. A high-concurrency, scalable API gateway is deployed on a cloud server or private data center. API requests are routed to the backend spatiotemporal database service, and load balancing strategies, such as round-robin and weighted round-robin, are configured. API request rate limits are set to prevent service overload. A circuit breaker is integrated for rapid failure in the event of downstream service failure. Departments access standardized data through the API gateway without direct database connection, supporting real-time spatiotemporal queries and returning structured data and visualized results. WebSocket and Server-Sent Events interfaces are provided to support real-time data push, such as proactively notifying the monitoring center when sensor data is updated.

[0091] In one embodiment of this disclosure, a data acquisition layer consisting of drone swarms, robots, etc., a spatiotemporal database, and an API gateway can be integrated into a unified platform to automate the entire process of data acquisition and sharing. Data usage feedback is collected through the API gateway. For example, if a data element is frequently queried but its accuracy is insufficient, the data acquisition layer is optimized, such as by adding high-precision sensors. All API requests are recorded, including the requester, time, operation, and result, to support security auditing and troubleshooting. The API gateway and spatiotemporal database are scanned regularly to ensure compliance with data security regulations.

[0092] like Figure 2 As shown, Figure 2This disclosure presents a structural diagram of a 3D visualization platform for data acquisition and processing in digital engineering exploration, according to an exemplary embodiment. The 3D visualization platform may include a data acquisition layer, an intelligent analysis layer, a standardized management layer, and a scenario-based application layer. The data acquisition layer is used to: utilize drone swarm collaborative sensing, employ 5G or BeiDou for precise positioning and time synchronization, and execute the steps related to acquiring multi-source geological data as described in the aforementioned method embodiments. This layer integrates data collected in real-time by airborne, ground-based, and underground acquisition devices in the area to be explored. This data undergoes preprocessing such as cleaning and anomaly removal via edge computing nodes corresponding to each device, and spatial and temporal consistency alignment is achieved. Simultaneously, this layer integrates a blockchain notarization module, which extracts hash values ​​from the collected multi-source geological data in real-time and anchors them to the blockchain along with device metadata, constructing a trusted intelligent sensing network integrating air, space, and ground at the underlying layer. The intelligent analysis layer is used to: receive cleaned data uploaded from the data acquisition layer, remove outliers using an automated cleaning rule engine, and execute the steps related to constructing a 3D geological model and determining predicted geological parameter values ​​as described in the aforementioned method embodiments. This layer receives multi-source geological data aggregated from the data acquisition layer, fills in missing data in the 3D grid using spatial interpolation algorithms such as Kriging, and transforms the grid data into a continuous 3D geological model through feature extraction and geological structural line segmentation. Furthermore, based on a pre-trained prediction model, this layer extracts in parallel the first semantic features of surface mineral spectral data, the second semantic features of geological report text, as well as temporal, spatial, rock and soil, and structural features. Subsequently, cross-attention and self-attention mechanisms are used to capture global dependencies between multimodal features and perform joint inference, accurately outputting the geological state prediction values ​​of each grid unit in the 3D geological model, and updating the 3D geological model based on these geological state prediction values. The standardized management layer can be used to: record the entire lifecycle information of data based on blockchain notarization technology of the data acquisition layer; adopt a spatiotemporal database design to uniformly manage spatial and unstructured data; define a data dictionary to improve cross-departmental data compatibility; and achieve efficient and secure data sharing and retrieval through an API gateway, solving the problems of scattered storage and difficult sharing in traditional systems, and improving data management efficiency and utilization value. The scenario-based application layer can be used to: call dynamic data collected by the data acquisition layer in real time; provide real-time monitoring screens to display exploration progress and multi-source geological parameters for different scenarios; build a 3D visualization platform to support stratigraphic profile cutting and attribute query of 3D geological data; develop mobile applications for offline data collection and task reception by on-site personnel; and optimize collection control commands through real-time data feedback and closed-loop control mechanisms to provide fast and accurate support for decision-making.Based on the absolute spatial coordinates and current direction of travel of each multi-source acquisition device, this layer dynamically extracts local grid slices from the updated 3D geological model of the intelligent analysis layer, covering the current position of the multi-source acquisition device and the predicted path directly in front of it. It extracts the geological state prediction value corresponding to each grid cell in the local grid slice, and inputs the geological state prediction value and the current control parameters of the multi-source acquisition device into the reinforcement learning-based feedback control model. This generates acquisition control commands containing specific acquisition frequency, acquisition trajectory and equipment parameters, and sends them to the underlying execution mechanism of the data acquisition layer in real time. This drives the fine-tuning of the UAV flight path, the obstacle avoidance of the robot path and the adaptive adjustment of the drilling mud pressure of the drilling rig, thereby realizing intelligent closed-loop control of the exploration operation.

[0093] In the process of constructing a 3D geological model, this 3D visualization platform can generate a 3D mesh model through stratigraphic interface interpolation, such as the kriging method; it can also infer the distribution of geological bodies directly from voxel data based on deep learning models, such as U-Net, which is suitable for complex structures, such as fault cross zones; and it can encode mineral content and permeability attributes into the 3D model through color mapping and transparency adjustment. For example, high copper content areas are rendered as red gradients, and low copper content areas as blue gradients.

[0094] This 3D visualization platform can calculate the intersection of model voxels and planes using predefined cutting plane equations to generate 2D profile outlines. It supports users to draw multi-segment lines with the mouse to generate continuous profiles along the path, such as S-shaped profiles to display folded structures. The platform can dynamically adjust the profile detail levels according to the viewing distance to improve the smoothness of interaction. It uses WebGL shaders to accelerate profile rendering and leverages the parallel computing capabilities of GPUs to improve performance.

[0095] This 3D visualization platform also enables real-time querying of attributes, i.e., geological parameters. When a user clicks on the 3D model, the attribute value of the nearest voxel can be obtained through a ray casting algorithm, such as the granite content being 85% at this location. The user can select an area, and the platform calculates statistical values ​​of attributes within that area, such as average mineral content and maximum porosity. A dialog box pops up after clicking, displaying a detailed list of attributes, such as lithology, age, and engineering parameters. When a profile or area is selected, the sidebar automatically loads related data, such as borehole columnar sections and seismic wave velocity curves. The query results, such as mineral content, are overlaid on the 3D model in the form of a heat map to aid spatial analysis.

[0096] In developing a 3D visualization platform, a lightweight platform can be built based on CesiumJS and Three.js, supporting cross-browser access; high-performance applications can be developed using Unity and Unreal Engine, adapting to VR / AR devices; interactive controls for plane cutting, path cutting, and profile rotation / translation can be provided; search boxes and filtering conditions, such as mineral content >50%, and result export buttons, such as exporting to CSV or Shapefile format, can be integrated; the 3D model can be divided into multiple tiles, and data within the viewport can be loaded on demand to reduce memory usage; frequently used profiles and query results can be cached, such as using Redis, to accelerate the response of repetitive operations; When integrating and deploying the platform, it can interface with the data acquisition layer, such as drones and robots, to automatically update the 3D model with new data; it can interface with the intelligent analysis layer, such as prediction models, to visualize prediction results, such as ore body boundaries; the platform can be packaged as a Docker container and deployed in a Kubernetes cluster to support horizontal scaling; it can support concurrent operation by providing online access; it can be tested across browsers to improve compatibility; and it can be adapted for mobile devices to support touch operation and zoom operation.

[0097] When conducting offline data collection and task reception for field personnel based on the development of mobile applications, field personnel can collect geological data, such as lithological descriptions, photos, and videos, in environments without network access and cache them locally; receive task instructions pushed by the backend in real time, such as rock samples from collection point A, and support task status updates, such as "arrived," "collecting," and "completed"; automatically or manually upload offline data to the cloud after network recovery to improve data integrity; user roles are divided as follows: field personnel use the mobile app for data collection, task execution, and data upload; and management personnel use the web interface to assign tasks, monitor progress, and review data.

[0098] The design architecture of this 3D visualization platform can be Flutter or React Native, while also ensuring compatibility with iOS and Android, thus reducing development costs. For storing structured data, such as sampling point coordinates and lithological codes, photos and videos are saved to the application sandbox directory and unique filenames are generated, such as UUID + timestamp. Other push notification services are integrated to push tasks in real time. Unread tasks are cached in the local database, so historical tasks can still be viewed even without a network connection. During offline data collection, this 3D visualization platform can dynamically load form fields such as lithology, color, and grain size based on task type, such as soil sampling or geological mapping; verify required fields such as coordinates and sampler; and adjust data formats such as longitude range (-180~180) to reduce invalid data. It can also access device cameras for continuous shooting, such as panoramic geological profiles; automatically add watermarks such as timestamps and task numbers; integrate speech-to-text functionality to assist with complex descriptions, such as weathering degree: strong weathering, fracture development; preload map tile data to support navigation to task points without network access; and obtain real-time coordinates via GPS / BeiDou with an accuracy of ≤5 meters.

[0099] When the 3D visualization platform receives management tasks, a system notification pops up when a new task arrives. This notification can include brief task information, such as "Sampling point B: Sandstone sample". The task can be categorized by status (pending execution, in progress, or completed) and can be filtered, such as by priority or deadline. Clicking on a task will directly take you to the map navigation interface to plan the optimal route. On-site personnel can update the task status via buttons or voice commands, such as "Start sampling" or "Upload completed".

[0100] This 3D visualization platform can support uploading on-site photos / videos as evidence of mission obstruction, such as road closures or inability to reach the collection point.

[0101] This 3D visualization platform can synchronize data and resolve conflicts. When a network is detected, such as 4G / Wi-Fi, it can automatically upload unsynchronized data in the background. It provides an "Instant Synchronization" button to allow users to force uploads, such as in emergencies. It generates a unique version number for each data record, such as UUID + timestamp. When conflicts are detected between server and local data, the server data takes precedence and overwrites conflicting local data, such as when the task status has been updated by another device. It prompts users to choose to keep the local or server version, such as allowing for more detailed local notes. It uses an HTTPS / TLS encrypted data synchronization channel to prevent man-in-the-middle attacks. This 3D visualization platform can encrypt local databases and files with AES-256, making data unreadable even if the device is lost; it integrates fingerprint / facial recognition, allowing only authorized personnel to launch the application; and it restricts access to functions based on user roles, such as ordinary data collectors and administrators, for example, prohibiting the deletion of other users' tasks. During testing and deployment, this 3D visualization platform can be tested offline by turning off the network to verify whether the data acquisition, task viewing, and local storage functions are normal. Stress testing: simulates the synchronization of 1000 tasks to test the application's response speed, with the target being: synchronization time <30 seconds. During compatibility testing, the 3D visualization platform tested mainstream mobile models to ensure screen adaptation. This 3D visualization platform can be deployed via Apple Business Manager / Google Play Enterprise, avoiding app store review delays; it also supports hot updates, reducing the frequency of manual upgrades by users.

[0102] It should be noted that the three-dimensional visualization platform for data acquisition and processing in digital engineering surveying in this embodiment is used to implement the corresponding data acquisition and processing methods for digital engineering surveying in the aforementioned method embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be elaborated here.

[0103] like Figure 3 As shown, Figure 3 This disclosure is a block diagram of a data acquisition and processing device for digital engineering surveying according to an exemplary embodiment. The data acquisition and processing device for digital engineering surveying includes a first determining module 310, a second determining module 320, and an instruction generating module 330. The first determining module 310 is used to acquire multi-source geological data collected in real time by the multi-source acquisition device in the area to be explored, and determine the three-dimensional geological model corresponding to the area to be explored and the spatial structure data corresponding to the three-dimensional geological model based on the multi-source geological data. The second determining module 320 is used to input the multi-source geological data, the geological report data of the area to be explored and the spatial structure data into a pre-trained prediction model, determine the multimodal features of the area to be explored by the pre-trained prediction model, determine the geological state prediction value of each grid unit in the three-dimensional geological model based on the multimodal features, and update the three-dimensional geological model based on the geological state prediction value. The instruction generation module 330 is used to input the current geological state corresponding to the current position of the multi-source acquisition device and the current control parameters of the multi-source acquisition device into a reinforcement learning-based feedback control model based on the updated three-dimensional geological model, generate acquisition control instructions for the area to be explored, and control the multi-source acquisition device to continue to perform multi-source geological data acquisition in the area to be explored based on the acquisition control instructions. The acquisition control instructions include acquisition frequency, acquisition trajectory and device parameters.

[0104] The embodiments of the data acquisition and processing device disclosed herein for digital engineering surveying can be applied to computer equipment, such as servers or terminal devices. The device embodiments can be implemented through software, hardware, or a combination of both. Taking software implementation as an example, as a logical device, it is formed by the processor used for data acquisition and processing in digital engineering surveying loading the corresponding computer program instructions from non-volatile memory into memory for execution. From a hardware perspective, such as... Figure 4 The diagram shown is a hardware structure diagram of a computer device used for data acquisition and processing in digital engineering surveying, according to an embodiment of this disclosure. Except for... Figure 4In addition to the processor 410, memory 430, network interface 420, and non-volatile memory 440 shown, the server or electronic device where the data acquisition and processing device 431 applied to digital engineering survey is located in the embodiment may also include other hardware depending on the actual function of the computer device, which will not be described in detail here.

[0105] Accordingly, this disclosure also provides a computer device including a processor; a memory for storing processor-executable instructions; wherein the processor is configured to implement the above-described method for data acquisition and processing for digital engineering surveying.

[0106] The specific implementation process of the functions and roles of each module in the above device can be found in the implementation process of the corresponding steps in the above method, and will not be repeated here.

[0107] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical modules, that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0108] The foregoing has described specific embodiments of this disclosure. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired results. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0109] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention applied herein. This disclosure is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not claimed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the following claims.

[0110] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

[0111] The above description is merely a preferred embodiment of this disclosure and is not intended to limit this disclosure. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A data acquisition and processing method applied to digital engineering surveying, characterized in that, include: Acquire multi-source geological data collected in real time in the area to be explored by multi-source acquisition devices, and determine the three-dimensional geological model corresponding to the area to be explored and the spatial structure data corresponding to the three-dimensional geological model based on the multi-source geological data; The multi-source geological data, the geological report data of the area to be explored, and the spatial structure data are input into a pre-trained prediction model. The pre-trained prediction model determines the multimodal characteristics of the area to be explored and determines the geological state prediction value of each grid cell in the three-dimensional geological model based on the multimodal characteristics. The three-dimensional geological model is then updated based on the geological state prediction value. Based on the updated 3D geological model, the current geological state of the multi-source acquisition device at its current location and the current control parameters of the multi-source acquisition device are input into a reinforcement learning-based feedback control model to generate acquisition control instructions for the area to be explored. Based on the acquisition control instructions, the multi-source acquisition device is controlled to continue to perform multi-source geological data acquisition in the area to be explored. The acquisition control instructions include acquisition frequency, acquisition trajectory, and device parameters.

2. The method according to claim 1, characterized in that, The acquisition of multi-source geological data collected in real time by multi-source acquisition devices in the area to be explored includes: The surface mineral spectral data and three-dimensional topographic data of the area to be explored are collected using airborne acquisition equipment. Geotechnical parameter data and geotechnical environmental data of the area to be investigated are collected using ground-based acquisition equipment; Underground geological environment data of the area to be investigated were collected using underground acquisition equipment. The surface mineral spectral data, the three-dimensional terrain data, the rock and soil parameter data, the rock and soil environmental data, and the underground geological environmental data are preprocessed using edge computing nodes; The preprocessed data is then subjected to spatial consistency processing and temporal consistency processing to obtain the multi-source geological data.

3. The method according to claim 1, characterized in that, The step of determining the three-dimensional geological model corresponding to the area to be explored based on the multi-source geological data includes: The multi-source geological data is mapped onto a three-dimensional grid, and the geological parameters corresponding to each grid cell in the three-dimensional grid are determined. Based on the geological parameters corresponding to each grid cell in the three-dimensional grid, the geological features corresponding to each grid cell are determined, and the geological features corresponding to each grid cell are subjected to feature fusion processing to obtain the feature vector corresponding to the three-dimensional grid. The geological structural lines and stratigraphic distribution information of the three-dimensional mesh are determined based on the feature vectors, and the three-dimensional mesh is converted into a three-dimensional geological model based on the geological structural lines and stratigraphic distribution information. The geological parameters are mapped onto the three-dimensional geological model to obtain the three-dimensional geological model of the area to be explored; The geological parameters include mineral parameters, soil and rock parameters, environmental parameters, and topographic parameters; the geological features include mineral features, soil and rock features, environmental features, and topographic texture features.

4. The method according to claim 3, characterized in that, The step of mapping the multi-source geological data onto a three-dimensional grid and determining the geological parameters corresponding to each grid cell in the three-dimensional grid includes: Determine the coordinate mapping relationship between the area to be explored and the three-dimensional grid, map the multi-source geological data to the corresponding grid cells based on the coordinate mapping relationship, and determine the geological parameters corresponding to each grid cell; For each type of geological parameter in the multi-source geological data, the grid cells in the three-dimensional grid that are missing the geological parameter are determined as the first grid cell, and the grid cells in the three-dimensional grid that contain the geological parameter are determined as the second grid cell. For each first grid cell, the estimated value of the geological parameter in the first grid cell is predicted based on the value of the geological parameter in each second grid cell.

5. The method according to claim 1, characterized in that, The pre-trained prediction model includes a multi-branch feature extraction layer, a multi-modal feature fusion layer, and a joint inference output layer. The process of determining the multi-modal features of the area to be explored using the pre-trained prediction model, and determining the predicted geological state of each grid cell in the three-dimensional geological model based on the multi-modal features, includes: Through the multi-branch feature extraction layer, the first semantic features corresponding to the surface mineral spectral data in the multi-source geological data, the second semantic features corresponding to the geological report data, the temporal features corresponding to the underground geological environment data and the soil and rock environment data in the multi-source geological data, the spatial features corresponding to the three-dimensional topographic data in the multi-source geological data, the soil and rock features corresponding to the soil and rock parameter data in the multi-source geological data, and the structural features corresponding to the spatial structure data are obtained respectively. The multimodal feature fusion layer determines multimodal features based on the global dependencies between the first semantic feature, the second semantic feature, the temporal feature, the spatial feature, the soil and rock feature, and the structural feature. The geological state prediction value of each grid cell in the three-dimensional geological model is determined based on the multimodal features through the joint inference output layer.

6. The method according to claim 5, characterized in that, The step of determining multimodal features through the multimodal feature fusion layer based on the global dependencies between the first semantic feature, the second semantic feature, the temporal feature, the spatial feature, the soil and rock feature, and the structural feature includes: The first semantic feature and the second semantic feature are fused based on the cross-attention mechanism to obtain the fused feature. The association weights among the fused features, temporal features, spatial features, soil and rock features, and structural features are determined based on the self-attention mechanism. Based on the association weights, feature fusion processing is performed on each feature to obtain the multimodal features.

7. The method according to claim 1, characterized in that, The step of inputting the current geological state corresponding to the current location of the multi-source acquisition device and the current control parameters of the multi-source acquisition device into a reinforcement learning-based feedback control model to generate acquisition control instructions for the area to be explored includes: Based on the travel direction and spatial coordinates of the multi-source acquisition device, local grid slices within the predicted path range are obtained from the updated three-dimensional geological model; the current geological state is determined based on the predicted geological state values ​​within the local grid slices. The current geological state and the current control parameters are concatenated using feature vectors to construct the current state vector; The current state vector is input into the feedback control model to calculate the probability distribution corresponding to each candidate acquisition action; The candidate acquisition action with the highest probability in the probability distribution is determined as the target control action, and the acquisition frequency, acquisition trajectory and device parameters are determined based on the target control action to generate the acquisition control command.

8. A data acquisition and processing device for digital engineering surveying, characterized in that, The device includes: The first determining module is used to acquire multi-source geological data collected in real time by multi-source acquisition devices in the area to be explored, and to determine the three-dimensional geological model corresponding to the area to be explored and the spatial structure data corresponding to the three-dimensional geological model based on the multi-source geological data. The second determining module is used to input the multi-source geological data, the geological report data of the area to be explored, and the spatial structure data into a pre-trained prediction model, and the pre-trained prediction model determines the multimodal features of the area to be explored, and determines the geological state prediction value of each grid cell in the three-dimensional geological model based on the multimodal features, and updates the three-dimensional geological model according to the geological state prediction value. The instruction generation module is used to input the current geological state of the multi-source acquisition device at its current location and the current control parameters of the multi-source acquisition device into a reinforcement learning-based feedback control model based on the updated three-dimensional geological model, and generate acquisition control instructions for the area to be explored, so as to control the multi-source acquisition device to continue to perform multi-source geological data acquisition in the area to be explored based on the acquisition control instructions. The acquisition control instructions include acquisition frequency, acquisition trajectory and device parameters.

9. A computer device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, it implements a data acquisition and processing method for digital engineering survey as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer program code instructions, which, when executed by a processor, implement the method as described in any one of claims 1-7.

Images