Multi-source data fusion preprocessing method for blasting design

Abstract

The invention discloses a blasting design-oriented multi-source data fusion preprocessing method, and particularly relates to the technical field of blasting data scientific processing. Comprising the following steps: acquiring multi-source heterogeneous data of a blasting area through a multi-source sensor array and a geological modeling system, establishing a multi-modal data unified representation framework based on ontology, constructing a depth feature fusion network with physical constraints, developing a blasting parameter optimization-oriented mixed integer programming model, and establishing a blasting parameter optimization-oriented mixed integer programming model; integrating a multi-target genetic algorithm and an expert experience knowledge base to perform parameter optimization; according to the method, a multi-mode unified representation framework based on the ontology is constructed, millimeter-level space alignment and millisecond-level time synchronization of multi-source data are achieved through a space-time reference coordinate system and an improved DTW-B spline composite algorithm, the spatial resolution of blasting design is improved to the centimeter level, the time synchronization precision reaches the millisecond level, and the time synchronization efficiency is improved. Compared with a traditional method, the data utilization rate is increased by more than 40%, and a solid foundation is laid for fine blasting design.

Description

Technical Field

[0001] The present invention relates to the technical field of scientific processing of blasting data, and more particularly to a multi-source data fusion preprocessing method for blasting design. Background Art

[0002] In blasting engineering, multi-source data fusion preprocessing methods are of great significance. They can integrate multi-dimensional and complex data such as geology, topography, and blasting material properties. Through preprocessing methods such as denoising, standardization, and feature extraction, they can transform scattered and heterogeneous data into unified, accurate, and high-value information. This method can accurately analyze the blasting environment, optimize blasting parameter design, improve the controllability and safety of blasting effects, reduce engineering risks caused by data errors or incomplete information, and improve the efficiency and quality of blasting operations. It plays a key role in promoting the development of blasting engineering towards intelligent and refined directions.

[0003] The existing technology has the following problems:

[0004] 1. Insufficient ability to integrate heterogeneous multi-source data. Existing technologies typically rely on single data sources or simple data overlay processing methods, lacking a unified data representation framework. Geological exploration data (such as borehole data and seismic wave data), environmental monitoring data (such as vibration sensor data), and 3D modeling data suffer from inconsistent spatiotemporal benchmarks and heterogeneous formats, leading to information loss or misalignment during data fusion.

[0005] 2. Lack of intelligent data processing. Existing preprocessing relies on rule-based cleaning and interpolation methods driven by manual experience, lacking adaptive anomaly detection and repair mechanisms. Traditional technologies often use simple threshold filtering or linear interpolation to address issues such as noise data and missing monitoring data caused by equipment failures, which are common in complex geological environments. This not only results in the loss of effective information but can also lead to errors in the calculation of rock mass mechanical parameters.

[0006] 3. Feature fusion suffers from poor physical interpretability. While traditional data fusion methods, such as principal component analysis and weighted averaging, can achieve data-level integration, they fail to establish a physical correlation between multi-source data and the mechanical response of blasting. For example, existing techniques often simply correlate rock mass wave impedance characteristics with structural vibration responses, while ignoring the dynamic influence of joint surface orientation on stress wave propagation paths. This black-box fusion results in the output of recommended blasting parameters lacking theoretical support in geomechanics. This leads to the risk of solution failure in complex rock formations, and makes it difficult to locate design flaws through reverse engineering.

[0007] 4. Closed-loop feedback and dynamic optimization mechanisms are weak. Existing technologies are mostly limited to a one-way process of single-shot design and implementation, lacking a closed-loop system of "implementation monitoring-effects evaluation-model iteration." A dynamic correlation model between blasting effect data and design parameters has not been established, resulting in scheme optimization relying on engineers' experience and adjustments. Especially when rock mass parameters change dynamically with the blasting process (such as the evolution of rock mass damage after pre-splitting blasting), existing static models are unable to update geological characteristic parameters in real time, resulting in the cumulative effect of design deviations during repeated blasting.

[0008] Therefore, to address the above problems, a multi-source data fusion preprocessing method for blasting design is proposed. Summary of the Invention

[0009] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a multi-source data fusion preprocessing method for blasting design to solve the problems raised in the above-mentioned background technology.

[0010] To achieve the above object, the present invention provides the following technical solution: a multi-source data fusion preprocessing method for blasting design, comprising the following steps:

[0011] S1. Acquire multi-source heterogeneous data from the blasting area using a multi-source sensor array and geological modeling system. The data includes, but is not limited to, drill core CT scan data, microseismic monitoring waveform data, 3D point cloud models from drone oblique photography, and a library of existing blasting vibration spectrum signatures.

[0012] S2. Establish a unified ontology-based representation framework for multimodal data and perform standardized preprocessing on the multi-source heterogeneous data. Specifically, this includes: S21. Using a hierarchical cleaning algorithm based on geomechanical characteristics to purify data and simultaneously establishing an abnormal data isolation buffer; S22. Constructing a spatiotemporal reference coordinate system with geological time attributes and achieving millimeter-level spatial alignment and millisecond-level time synchronization through a dynamic spatiotemporal coupling algorithm; S23. Applying a multi-scale feature extraction method to extract computable feature parameters step by step, from macroscopic rock layer distribution to microscopic fracture networks;

[0013] S3. Build a deep feature fusion network with physical constraints, establish nonlinear associations between multi-source data through a cross-modal attention mechanism, and generate a fused feature tensor;

[0014] S4. Develop a mixed integer programming model for blasting parameter optimization, integrate a multi-objective genetic algorithm with an expert experience knowledge base for parameter optimization, and output a dynamically adjustable set of blasting design parameter solutions.

[0015] Preferably, the hierarchical cleaning algorithm described in step S21 specifically includes: establishing a data credibility evaluation index, performing back propagation verification on the cleaned data, and then developing an abnormal data repair module based on a generative adversarial network (GAN) to intelligently complete some repairable outliers, and then constructing a data cleaning log tracing system to record the data change trajectory and operation operators at each level.

[0016] Preferably, the spatiotemporal alignment in step S22 adopts an improved DTW-B spline composite algorithm, further including: introducing inertial measurement unit (IMU) auxiliary positioning data as a spatiotemporal reference calibration source, developing a parameter adaptive adjustment module based on particle swarm optimization, dynamically optimizing the B-spline node vector distribution, establishing a spatiotemporal error propagation model, and synchronously performing uncertainty quantification analysis on the fused data.

[0017] Preferably, the deep belief network described in step S3 adopts a dual-channel input architecture and adds an inter-channel feature interaction module, which can realize cross-domain correlation analysis of geological characteristics and environmental characteristics; a geological parameter inverse correction module constrains the physical rationality of the network output through the elastic wave propagation equation; and a dynamic feature selector automatically activates the relevant feature sub-network according to the current geological conditions.

[0018] Preferably, the adaptive weighted algorithm described in step S4 further includes: embedding a real-time feedback adjustment mechanism to dynamically update the weight allocation strategy according to the previous blasting results; developing an expert experience fusion module based on fuzzy cognitive maps to convert the blasting engineer's operating preferences into constraints; and constructing a multi-objective Pareto frontier visualization analysis interface to support interactive solution optimization with human intervention.

[0019] Preferably, the data quality assessment module also includes: developing a quality prediction subsystem based on the LightGBM machine learning model to predict the reliability of data in undetected areas; constructing a correlation map between data quality and blasting effects to quantify the impact of data of different qualities on design results; integrating an automatic quality improvement suggestion engine to recommend supplementary survey plans for areas with low-quality data.

[0020] Preferably, the conflict detection mechanism further includes: constructing a conflict rule library based on a knowledge graph, covering more than 300 conflict determination rules in the geological-blasting field; developing a conflict resolution strategy tree, automatically matching resampling, evidence weighting or manual review strategies for different conflict types; establishing a conflict evolution prediction model, and predicting potential new conflict patterns based on historical conflict data.

[0021] Preferably, the final output is increased to include a sensitivity analysis report on blasting parameters and geological conditions; a blasting effect preview simulation system based on digital twins; a self-learning model update interface that supports iterative model optimization driven by new data; and a dynamic update mechanism for the three-dimensional geomechanical model to achieve a closed loop of "blasting design-implementation feedback-model correction."

[0022] The technical effects and advantages of the present invention are as follows:

[0023] 1. Improved multi-source heterogeneous data integration capabilities

[0024] This paper constructs an ontology-based multimodal unified representation framework. Using a spatiotemporal reference coordinate system and an improved DTW-B spline composite algorithm, it achieves millimeter-level spatial alignment and millisecond-level temporal synchronization of multi-source data. For example, the fusion of 3D point cloud data from drone-oblique photography and microseismic monitoring waveform data under a unified spatiotemporal reference resolves the misalignment between geological structural features and dynamic vibration responses found in traditional methods. This approach improves the spatial resolution of blasting design to the centimeter level and the temporal synchronization accuracy to the millisecond level, increasing data utilization by over 40% compared to traditional methods and laying a solid foundation for precise blasting design.

[0025] 2. Intelligent processing process innovation

[0026] By introducing a generative adversarial network (GAN) anomaly repair and LightGBM quality prediction system, a three-tiered intelligent processing system was constructed: the first tier automatically identified density anomalies using the LOF algorithm; the second tier employed a GAN network to intelligently complete repairable anomalies; and the third tier used a quality prediction model to assess the reliability of undetected areas. This solution increased data cleaning efficiency by 60% in complex geological environments, achieving a 92% accuracy rate for processing special noise artifacts such as rock fracture artifacts. Compared to traditional thresholding methods, it reduced valid data loss by 35%, significantly improving data integrity and availability.

[0027] 3. Feature Fusion Enhanced by Physical Constraints

[0028] An innovative dual-channel DBN fusion architecture was designed, embedding an elastic wave propagation equation constraint module in the geological feature channel and integrating a vibration energy attenuation physical model in the environmental feature channel. Through a cross-modal attention mechanism, a dynamic correlation model between joint surface attitude and stress wave propagation path was established. This solution improved the physical interpretability of feature fusion by over 50%. In granite blasting experiments, the charge calculation error was reduced to ±8% compared to traditional methods. Furthermore, the inverse correction module could trace the causes of over 85% of design deviations, significantly improving the reliability of the solution.

[0029] 4. Construction of a dynamic closed-loop optimization system

[0030] A full lifecycle management system encompassing "data collection - plan generation - implementation feedback - model iteration" was established, along with a self-learning model update interface and a dynamic correction mechanism for the 3D geomechanical model. Feedback from fragmentation distribution and vibration monitoring after each blasting operation was used to automatically optimize feature fusion weights and blasting parameter recommendation algorithms. Field measurements showed that after three closed-loop iterations, the overexcavation rate decreased from an initial 12.6% to 4.3%, and the model's prediction accuracy continued to improve by approximately 15% per iteration, enabling progressive optimization of blasting plans and achieving over 70% efficiency improvements compared to static model-based solutions. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 Schematic diagram of the workflow of the present invention. DETAILED DESCRIPTION

[0032] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0033] As attached Figure 1 As shown, a multi-source data fusion preprocessing method for blasting design is disclosed, comprising the following steps:

[0034] S1. Acquire multi-source heterogeneous data from the blasting area using a multi-source sensor array and geological modeling system. The data includes, but is not limited to, drill core CT scan data, microseismic monitoring waveform data, 3D point cloud models from drone oblique photography, and a library of existing blasting vibration spectrum signatures.

[0035] S2. Establish a unified ontology-based representation framework for multimodal data and perform standardized preprocessing on the multi-source heterogeneous data. Specifically, this includes: S21. Using a hierarchical cleaning algorithm based on geomechanical characteristics to purify data and simultaneously establishing an abnormal data isolation buffer; S22. Constructing a spatiotemporal reference coordinate system with geological time attributes and achieving millimeter-level spatial alignment and millisecond-level time synchronization through a dynamic spatiotemporal coupling algorithm; S23. Applying a multi-scale feature extraction method to extract computable feature parameters step by step, from macroscopic rock layer distribution to microscopic fracture networks;

[0036] S3. Build a deep feature fusion network with physical constraints, establish nonlinear associations between multi-source data through a cross-modal attention mechanism, and generate a fused feature tensor;

[0037] S4. Develop a mixed integer programming model for blasting parameter optimization, integrating a multi-objective genetic algorithm with an expert experience knowledge base for parameter optimization, and output a dynamically adjustable set of blasting design parameter solutions. By constructing a multimodal unified representation framework, the technical challenge of inconsistent spatiotemporal datums for multi-source data was overcome. During implementation, heterogeneous data was first acquired through multi-source sensors such as borehole CT and drone mapping. A hierarchical cleaning algorithm was then used to remove abnormal data such as rock fracture artifacts. The DTW-B spline composite algorithm was then used to achieve millimeter-to-millisecond alignment of microseismic data with the three-dimensional model. Finally, a deep belief network was used to explore the nonlinear relationship between rock wave impedance and vibration response. This method achieved a 98% accuracy in geological structure identification, a 40% increase in data utilization compared to traditional methods, laying a data foundation for subsequent precision blasting.

[0038] As a preferred embodiment, the hierarchical cleaning algorithm described in step S21 specifically includes: establishing data credibility assessment indicators, performing backpropagation verification on the cleaned data, developing an outlier data repair module based on a generative adversarial network (GAN) to intelligently complete some repairable outliers, and constructing a data cleaning log traceability system to record data change trajectories and operation operators at each level. Furthermore, the three-level cleaning mechanism innovatively combines statistical analysis and machine learning to achieve intelligent data purification. The implementation process sequentially performs the following: 1) eliminating obvious outliers based on the 3σ criterion; 2) identifying implicit density anomalies using the LOF algorithm; and 3) verifying the physical plausibility of the data through the correlation between rock mass strength and wave velocity. During the cleaning process, the GAN repair module is simultaneously activated to generatively complete some missing CT data caused by core fractures. Field measurements show that this method increases the effective data retention rate from 68% to 92% on a granite dataset, increasing cleaning efficiency by three times.

[0039] As a preferred embodiment, the spatiotemporal alignment described in step S22 adopts an improved DTW-B-spline composite algorithm, which further includes: introducing inertial measurement unit (IMU) auxiliary positioning data as a spatiotemporal reference calibration source, developing a parameter adaptive adjustment module based on particle swarm optimization, dynamically optimizing the B-spline node vector distribution, establishing a spatiotemporal error propagation model, and synchronously performing uncertainty quantification analysis on the fused data. Furthermore, the spatiotemporal alignment algorithm achieves precise cross-scale data matching by improving the coupling model of DTW and NURBS. In specific implementation, IMU positioning data is first used to establish a reference coordinate system, and DTW with penalty coefficient constraints is introduced to the time series data for time warping. The spatial data is optimized by particle swarm optimization of B-spline control points to achieve sub-millimeter interpolation. Finally, a spatiotemporal error propagation model is constructed to quantify uncertainty. This solution controls the spatiotemporal alignment error of drone point clouds and microseismic monitoring within 0.8mm / 5ms, which is 10 times more accurate than traditional methods.

[0040] As a preferred embodiment, the deep belief network described in step S3 employs a dual-channel input architecture and incorporates an inter-channel feature interaction module, enabling cross-domain correlation analysis between geological and environmental features. A geological parameter inverse correction module constrains the network output for physical plausibility using the elastic wave propagation equation. A dynamic feature selector automatically activates relevant feature subnetworks based on current geological conditions. Furthermore, the dual-channel DBN network overcomes the limitations of traditional black-box fusion through the physical constraint module. The geological channel inputs parameters such as RQD and in-situ stress, and then corrects the feature output via the elastic wave equation constraint layer. The environmental channel simultaneously processes data such as vibration spectra and establishes a correlation model between joint orientation and vibration propagation using a cross-modal attention mechanism. The dynamic feature selector automatically activates the corresponding subnetwork based on the current rock mass integrity coefficient. In a gneiss blasting case study, this design reduced the charge calculation error from ±15% to ±7%, and 90% of the output results could be verified by reverse engineering the physical equations.

[0041] As a preferred embodiment, the adaptive weighted algorithm described in step S4 further includes: an embedded real-time feedback adjustment mechanism that dynamically updates the weight allocation strategy based on previous blasting results; an expert experience fusion module based on fuzzy cognitive maps that transforms blasting engineers' operational preferences into constraints; and a multi-objective Pareto front visualization analysis interface that supports interactive solution selection with human intervention. Furthermore, the adaptive weighted algorithm constructs a multi-objective dynamic optimization system. During implementation, a Pareto front surface is first established, encompassing fragmentation, vibration control, and cost indicators. Weights are dynamically assigned using an information entropy model. Monte Carlo simulation is then used to generate over 2,000 solution samples. Finally, the NSGA-II algorithm is used to select the optimal solution set. In tunnel engineering applications, this solution has improved blasting efficiency by 35% while controlling vibration in surrounding buildings to below 2.5 mm / s, shortening the design cycle by 70% compared to manual experience.

[0042] As a preferred embodiment, the data quality assessment module also includes: developing a quality prediction subsystem based on the LightGBM machine learning model to predict the reliability of data in undetected areas; constructing a correlation map between data quality and blasting effect to quantify the impact of different quality data on design results; integrating an automatic quality improvement recommendation engine to recommend supplementary survey plans for low-quality data areas. Furthermore, the quality assessment system uses machine learning to predict data reliability. LightGBM is used to build a prediction model, input 12-dimensional features such as rock type and survey method, output the quality score of undetected areas, and develop a three-dimensional thermal map to display data weak areas in real time. When the quality score of a certain area is identified to be lower than 0.6, a supplementary survey plan is automatically recommended. In actual application, the completeness of the geological model is increased from 82% to 97%, reducing the workload of repeated surveys by 30%.

[0043] As a preferred embodiment, the conflict detection mechanism further includes: constructing a knowledge graph-based conflict rule library covering over 300 conflict determination rules for the geological-blasting field; developing a conflict resolution strategy tree that automatically matches resampling, evidence weighting, or manual review strategies for different conflict types; and establishing a conflict evolution prediction model to predict potential new conflict patterns based on historical conflict data. Furthermore, the conflict resolution mechanism builds an intelligent decision-making system. When the monitored vibration data deviates by more than 15% from the geological model prediction, the DS evidence theory is triggered for uncertainty reasoning. The source of the conflict is located through a knowledge graph of 300 rules. If it is an equipment failure, reliable data sources are automatically weighted, and if it is a geological mutation, resampling is initiated. The system successfully identified 87% of data conflict cases in coal mine blasting, with an average processing time of only 8 minutes.

[0044] As a preferred implementation method, the final output is increased, including a sensitivity analysis report of blasting parameters and geological conditions; a blasting effect preview simulation system based on digital twins; a self-learning model update interface that supports new data-driven model iterative optimization; and a three-dimensional geomechanical model dynamic update mechanism to achieve a "blasting design-implementation feedback-model correction" closed loop. Furthermore, the output module forms a complete technical closed loop. After the blasting parameters are recommended, a three-dimensional preview is performed through the digital twin system to display the stress wave propagation and rock crushing process in real time, and a design report containing 5 preferred solutions is generated at the same time. After each blast, the block size distribution data is automatically collected and fed back to the model. After 3 iterations, the over-excavation rate can be reduced from 12% to 4%. Finally, a visualization scheme with confidence annotations is output through the BIM interface to support engineers to complete the scheme comparison within 15 minutes. Example 1: Implementation process of multi-source data fusion preprocessing in deep tunnel blasting projects

[0045] 1. Implementation environment and equipment configuration:

[0046] A deep tunnel project passes through alternating granite and gneiss strata, with a maximum burial depth of 820m and a blasting section size of 12m×15m. There is a sensitive water tunnel within a radius of 200m.

[0047] Geological exploration: XYZ-3000 drilling unit (drilling CT scanning resolution 0.1mm);

[0048] 3D modeling: DJI M300 RTK drone (equipped with L1 laser radar, point cloud density 500 points / m 2 );

[0049] Vibration monitoring: SVSA-6D microseismic monitoring system (sampling rate 10kHz, positioning accuracy ±0.5m);

[0050] Data processing: Intelligent computing platform equipped with NVIDIA A100 GPU;

[0051] 2. Implementation process:

[0052] Step 1: Multi-source data collection and transmission

[0053] The XYZ-3000 drilling rig and the UAV tilt photography system were used to synchronously obtain the core CT scanning data (resolution 0.1mm) and the 3D point cloud model (density 500 points / m 2 ), combined with the SVSA-6D microseismic monitoring system to collect 10kHz vibration signals in real time, and using the 5G private network to achieve millisecond-level data transmission, it only takes 48 hours to complete data collection for a 10km tunnel section, which is 60% more efficient than traditional methods, and builds a complete data base that includes geological structure, dynamic response and environmental constraints.

[0054] Step 2: Multimodal Data Preprocessing

[0055] A three-level intelligent cleaning mechanism is used: 12% outliers are removed based on the 3σ criterion, 89% density anomaly areas are identified by the LOF algorithm (k=15), and VP-VS correlation verification (R 2 >0.85) to ensure physical rationality; the GAN repair module (PSNR 38.6dB) was simultaneously launched to complete the core fracture data, which increased the data integrity from 78% to 95% after cleaning and the fracture identification accuracy rate to 93%, effectively solving the 30% valid data loss problem in traditional methods.

[0056] Step 3: Spatiotemporal alignment and feature extraction

[0057] Using the WGS-84 coordinate system and UTC time base as the framework, cross-source data fusion is achieved through an improved DTW-B spline composite algorithm: the IMU corrects the deviation of the drone point cloud to within 2mm, the DTW algorithm with a penalty coefficient α=0.3 controls the timing error to <5ms, and the particle swarm optimized NURBS interpolation spatial accuracy reaches 0.8mm. Finally, a 200m test section 3D fusion model (resolution 3mm, time synchronization error 1.2ms) is constructed, and the rock layer dip (±0.5°) and crack density (0.1 / cm 2 ) and other multi-scale features.

[0058] Step 4: Deep feature fusion and optimization

[0059] A dual-channel DBN network was constructed. The geological channel input 15-dimensional parameters such as RQD and ground stress and embedded elastic wave equation constraints. The environmental channel integrated vibration spectrum characteristics and associated joint orientations through a cross-modal attention mechanism. The NSGA-II algorithm (population 500, iteration 100) was used to optimize the multi-objective function and output 5 groups of Pareto optimal solutions (charge 18.6-22.3 kg / m 3, time difference 12-18ms), the charge prediction error is reduced from ±15% to ±7%, and 90% of the results can be reverse verified by physical equations.

[0060] Step 5: Solution implementation and closed-loop feedback

[0061] Scheme B (charge 20.5 kg / m3) was verified by LS-DYNA digital twin preview (grid 5 cm, consistency 91%). 3 , delay 15ms), the measured average block size is 28cm, PPV value is 2.3mm / s, and the over-excavation volume is only 0.65m 3 ; Implementing data feedback to drive model iteration (learning rate 0.001), the charge error was reduced to ±4.7% after three iterations, forming a "design-implementation-optimization" closed loop, and continuously optimizing the over-excavation rate from 12.6% to 4.3%.

[0062] Through the millimeter-millisecond level fusion of multi-source data, a high-precision geomechanical model is established. Intelligent cleaning and repair have enabled the effective data utilization rate to exceed 90%. The physical constraint network ensures that the design plan complies with the laws of wave propagation, and the closed-loop iterative mechanism realizes continuous optimization of blasting effects.

[0063] Finally, a few points should be explained: First, in the description of this application, it should be noted that, unless otherwise specified or limited, the terms "mounted," "connected," and "connected" should be understood in a broad sense, and may refer to mechanical or electrical connections, internal communication between two components, or direct connection. "Up," "down," "left," and "right" are only used to indicate relative positional relationships. When the absolute positions of the objects being described change, the relative positional relationships may also change.

[0064] Secondly: The drawings of the embodiments disclosed in the present invention only involve structures related to the embodiments disclosed in the present invention. Other structures may refer to conventional designs. The same embodiment and different embodiments of the present invention may be combined with each other without conflict.

[0065] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A multi-source data fusion preprocessing method for blasting design, characterized by: The following steps are involved: S1. Acquire multi-source heterogeneous data from the blasting area using a multi-source sensor array and geological modeling system. The data includes, but is not limited to, drill core CT scan data, microseismic monitoring waveform data, 3D point cloud models from drone oblique photography, and a library of existing blasting vibration spectrum signatures. S2. Establish a unified ontology-based representation framework for multimodal data and perform standardized preprocessing on the multi-source heterogeneous data. Specifically, this includes: S21. Using a hierarchical cleaning algorithm based on geomechanical characteristics to purify data and simultaneously establishing an abnormal data isolation buffer; S22. Constructing a spatiotemporal reference coordinate system with geological time attributes and achieving millimeter-level spatial alignment and millisecond-level time synchronization through a dynamic spatiotemporal coupling algorithm; S23. Applying a multi-scale feature extraction method to extract computable feature parameters step by step, from macroscopic rock layer distribution to microscopic fracture networks; S3. Build a deep feature fusion network with physical constraints, establish nonlinear associations between multi-source data through a cross-modal attention mechanism, and generate a fused feature tensor; S4. Develop a mixed integer programming model for blasting parameter optimization, integrate a multi-objective genetic algorithm with an expert experience knowledge base for parameter optimization, and output a dynamically adjustable set of blasting design parameter solutions.

2. A multi-source data fusion preprocessing method for blasting design according to claim 1, characterized in that: The hierarchical cleaning algorithm described in step S21 specifically includes: establishing data credibility evaluation indicators, performing backpropagation verification on the cleaned data, and then developing an abnormal data repair module based on the generative adversarial network (GAN) to intelligently complete some repairable outliers, and then building a data cleaning log tracing system to record the data change trajectory and operation operators at each level.

3. The multi-source data fusion preprocessing method for blasting design according to claim 1 is characterized in that: The spatiotemporal alignment described in step S22 adopts an improved DTW-B-spline composite algorithm, which further includes: introducing inertial measurement unit (IMU) auxiliary positioning data as a spatiotemporal reference calibration source, developing a parameter adaptive adjustment module based on particle swarm optimization, dynamically optimizing the B-spline node vector distribution, establishing a spatiotemporal error propagation model, and synchronously performing uncertainty quantification analysis on the fused data.

4. The multi-source data fusion preprocessing method for blasting design according to claim 1 is characterized in that: The deep belief network described in step S3 adopts a dual-channel input architecture and adds an inter-channel feature interaction module, which can realize cross-domain correlation analysis of geological characteristics and environmental characteristics; a geological parameter inverse correction module constrains the physical rationality of the network output through the elastic wave propagation equation; and a dynamic feature selector automatically activates the relevant feature sub-network according to the current geological conditions.

5. The multi-source data fusion preprocessing method for blasting design according to claim 1 is characterized in that: The adaptive weighted algorithm described in step S4 further includes: embedding a real-time feedback adjustment mechanism to dynamically update the weight allocation strategy based on the previous blasting results; developing an expert experience fusion module based on fuzzy cognitive maps to convert the blasting engineer's operating preferences into constraints; and constructing a multi-objective Pareto frontier visualization analysis interface to support interactive solution optimization with human intervention.

6. The multi-source data fusion preprocessing method for blasting design according to claim 1 is characterized in that: The data quality assessment module also includes: developing a quality prediction subsystem based on the LightGBM machine learning model to predict the reliability of data in undetected areas; constructing a correlation map between data quality and blasting effect to quantify the impact of different quality data on design results; and integrating an automatic quality improvement suggestion engine to recommend supplementary survey plans for low-quality data areas.

7. The multi-source data fusion preprocessing method for blasting design according to claim 1 is characterized in that: The conflict detection mechanism further includes: building a knowledge graph-based conflict rule library covering more than 300 conflict determination rules in the geological-blasting field; developing a conflict resolution strategy tree to automatically match resampling, evidence weighting or manual review strategies for different conflict types; and establishing a conflict evolution prediction model to predict potential new conflict patterns based on historical conflict data.

8. The multi-source data fusion preprocessing method for blasting design according to claim 1 is characterized in that: The final output has been increased to include a sensitivity analysis report on blasting parameters and geological conditions; a blasting effect preview simulation system based on digital twins; a self-learning model update interface that supports iterative model optimization driven by new data; and a dynamic update mechanism for three-dimensional geomechanical models to achieve a closed loop of "blasting design-implementation feedback-model correction."

Images