Tunnel blasting quality evaluation and optimization method based on multi-source data fusion

Abstract

The invention discloses a tunnel blasting quality evaluation and optimization method based on multi-source data fusion, and belongs to the field of tunnel blasting quality evaluation, and the method comprises the steps: collecting and preprocessing multi-source data of a tunnel blasting region; based on the preprocessed multi-source data, performing blasting quality evaluation according to local back break, a blasting contour line, average linear back break and point cloud extraction to obtain a blasting quality evaluation result; according to the blasting quality evaluation result, the blasting quality is graded, and a comprehensive blasting quality score is calculated and graded; and establishing a database containing geological parameters, surrounding rock response parameters and blasting process parameters, training through a convolutional neural network model to generate a blasting parameter optimization scheme, and dynamically adjusting blasting parameters of the next cycle according to the comprehensive blasting quality score. According to the method, the blasting parameter optimization and the quality evaluation process are closely combined to form a closed-loop system, the specific situation in the construction can be reflected in real time, and the accuracy of the blasting effect is ensured.

Description

Technical Field

[0001] The present invention belongs to the technical field of tunnel blasting quality evaluation, and in particular relates to a tunnel blasting quality evaluation and optimization method based on multi-source data fusion. Background Art

[0002] At present, the evaluation of tunnel blasting quality mainly relies on traditional manual experience to set blasting parameters, or an optimization model based on preset standard features. For example, Chinese patent CN202410783820 proposes a method for optimizing blasting parameters in open-pit mines. It constructs a mine model by collecting physical parameters of the ore body, and adjusts the parameters using a fragmentation prediction model and a cost model after simulating blasting. Another Chinese patent, CN202311223077, discloses a method for evaluating the quality of tunnel blasting. By establishing a tunnel model and updating the blasting area data, the blasting plan is optimized by combining indicators such as over-excavation and under-excavation, and footage. The above schemes all use data modeling methods, but the core still relies on preset features and manual experience to adjust parameters.

[0003] The existing technology still has certain shortcomings. First, the existing blasting quality evaluation methods mainly rely on manual experience or optimization models based on preset standard features. These models lack real-time and adaptability, and cannot fully consider changes in actual on-site conditions, such as changes in rock formations after actual blasting, blasting footage, cross-section unevenness and other key factors. This approach may lead to deviations in the evaluation results of blasting quality and cannot accurately reflect the blasting effects in actual construction. Secondly, although the existing methods adjust blasting parameters through model optimization, they still rely on manual experience and simulation experiments, and fail to fully achieve automation and real-time adjustment. Therefore, the existing blasting quality evaluation methods have not formed a closed-loop optimization system, and lack real-time feedback and iterative optimization of actual blasting results. Summary of the Invention

[0004] In order to solve the above technical problems, the present invention proposes a tunnel blasting quality evaluation and optimization method based on multi-source data fusion to solve the problems existing in the above-mentioned prior art.

[0005] To achieve the above objectives, in a first aspect, the present invention provides a tunnel blasting quality evaluation and optimization method based on multi-source data fusion, comprising:

[0006] Collecting and preprocessing multi-source data of the tunnel blasting area, the multi-source data including laser point cloud data and geological radar detection data;

[0007] Based on the pre-processed multi-source data, blasting quality evaluation is performed according to local over-break, blasting contour, average linear over-break and under-break, and point cloud extraction to obtain the blasting quality evaluation results.

[0008] According to the blasting quality evaluation results, the blasting quality is graded, and the comprehensive blasting quality score is calculated and graded;

[0009] A database containing geological parameters, surrounding rock response parameters and blasting process parameters is established. The blasting parameter optimization scheme is generated through convolutional neural network model training, and the blasting parameters of the next cycle are dynamically adjusted according to the comprehensive blasting quality score.

[0010] Preferably, the process of collecting and preprocessing multi-source data of the tunnel blasting area includes:

[0011] Collect laser point cloud data through 3D laser scanner;

[0012] Collect geological radar detection data through geological radar technology;

[0013] Based on the laser point cloud data, the statistical filtering method is used to remove point cloud noise, filter out abnormal points whose centroid distance is greater than the threshold, and correct the point cloud error through the iterative closest point algorithm;

[0014] Based on the geological radar detection data, the wavelet transform algorithm is used to remove noise.

[0015] Preferably, the process of evaluating blasting quality according to local over-break and under-break includes:

[0016] Project each measured point onto the theoretical cross section to obtain the local over-excavation and under-excavation ratio;

[0017] If the local over-excavation and under-excavation rate exceeds a threshold, it is automatically marked as a blasting deviation point and the density is counted.

[0018] Preferably, the process of evaluating blasting quality according to the blasting contour line includes:

[0019] The measured contour is extracted in the projection plane of each construction cycle, and the designed contour is reconstructed by spline;

[0020] The Hausdorff distance is used to measure the contour consistency, and the deviation rate of the blasting contour line is calculated.

[0021] Preferably, the process of evaluating blasting quality according to average linear overbreak and underbreak includes:

[0022] Perform Boolean operations on the closed area formed by the measured contour and the designed contour to obtain the over-excavation area and under-excavation area;

[0023] The average linear over-break and under-break rate is calculated. If the average linear over-break and under-break rate is higher than the warning threshold, it is recommended to adjust the charge amount of the surrounding hole lines or the spacing between the blastholes.

[0024] Preferably, the process of blasting quality evaluation based on point cloud extraction includes:

[0025] The 3D point cloud is partitioned into sparse and dense parts using kernel density estimation, and potential cavities and cracks are identified by combining geological radar prestack migration profiles.

[0026] The M3C2 algorithm is used to quantitatively compare the point clouds before and after blasting, and output the surrounding rock volume damage index.

[0027] Preferably, the process of grading the blasting quality and calculating and grading the comprehensive blasting quality score includes:

[0028] Normalize the local over-excavation rate, contour line deviation rate, average linear over-excavation rate and surrounding rock volume damage index to obtain the normalized value;

[0029] assigning weights by the analytic hierarchy process, and calculating a comprehensive blasting quality score based on the normalized values ​​and the weights;

[0030] The blasting quality is divided into four levels according to the range of comprehensive blasting quality scores.

[0031] Preferably, the records of the database are stored in a PostgreSQL database according to the cycle number-hole number-timestamp index.

[0032] Preferably, the process of dynamically adjusting the blasting parameters of the next cycle according to the comprehensive blasting quality score includes:

[0033] Call the convolutional neural network model to output the gradient vector;

[0034] Based on the gradient vector, a new round of optimization parameters is generated in the three-dimensional space of blasthole spacing, charge amount, and detonation interval through a Bayesian optimizer;

[0035] When the comprehensive blasting quality score increases by ≥3 points and meets the safety threshold, the trolley control script is written and executed.

[0036] In a second aspect, the present invention further discloses a computer-readable storage medium having a computer program stored thereon, which implements the steps of the method described in the first aspect when executed by a processor.

[0037] Compared with the prior art, the present invention has the following advantages and technical effects:

[0038] The application provides a tunnel blasting quality evaluation and optimization method based on multi-source data fusion, comprising the following steps: first, collecting and preprocessing multi-source data of a tunnel blasting area, wherein the multi-source data comprises laser point cloud data and geological radar detection data; second, based on the preprocessed multi-source data, performing blasting quality evaluation according to local overbreak and underbreak, blasting contour line, average linear overbreak and underbreak and point cloud extraction to obtain a blasting quality evaluation result; third, classifying the blasting quality according to the blasting quality evaluation result, and calculating a comprehensive blasting quality score and grading; and fourth, establishing a database comprising geological parameters, surrounding rock response parameters and blasting process parameters, training a convolutional neural network model to generate a blasting parameter optimization scheme, and dynamically adjusting the next cycle blasting parameters according to the comprehensive blasting quality score.

[0039] The application integrates actual data after field blasting (such as three-dimensional laser scanning data, rock stratum detection data, etc.), and through a real-time feedback mechanism, the actual blasting effect is included in the optimization model, so that the optimization parameters can be closer to the actual working conditions. The evaluation of blasting quality is no longer limited to a preset model, but can reflect the specific situation in construction in real time, ensuring the accuracy of the blasting effect. By closely combining the blasting parameter optimization and the quality evaluation process, a closed-loop system is formed.

[0040] The application provides a more accurate blasting quality evaluation method through multi-source data fusion technology, thereby improving the accuracy and persuasiveness of blasting quality evaluation. In addition, the application can achieve better blasting parameter optimization effect, ensure the stability and safety of the blasting process, and ultimately achieve the goal of improving blasting quality, optimizing resource utilization, ensuring construction progress and personnel safety. BRIEF DESCRIPTION OF DRAWINGS

[0041] The accompanying drawings, which form a part of this application, are included to provide a further understanding of the application and are incorporated in and constitute a part of this application. The embodiments of this application and their description together with the drawings serve to explain the application. In the drawings:

[0042] Figure 1 A tunnel blasting quality evaluation and optimization flowchart of the embodiments of the application. DETAILED DESCRIPTION

[0043] It should be noted that the embodiments and features in the embodiments of the application can be combined with each other without conflict. The application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0044] It should be noted that the steps shown in the flowchart of the drawings can be executed in a computer system such as a set of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described herein can be executed in an order different from that shown herein.

[0045] Example 1

[0046] like Figure 1 As shown, this embodiment provides a tunnel blasting quality evaluation and optimization method based on multi-source data fusion. The multi-source data required by this embodiment mainly include laser point cloud data and geological radar detection data, providing a blasting quality evaluation system of point (local over-excavation) → line (blasting contour line) → surface (average linear over-excavation) → volume (point cloud extraction), and then calculating the weight of the blasting quality evaluation index to finally obtain a more three-dimensional and comprehensive blasting quality evaluation method. Then, the on-site blasting data is collected to establish a database (address parameters, surrounding rock parameters, blasting parameters), and the database is handed over to the convolutional neural network for model building and feature extraction, and then the model is trained to achieve the effect of automatic optimization of blasting parameters. Specifically including:

[0047] S1. Collect and pre-process multi-source data of the tunnel blasting area, wherein the multi-source data includes laser point cloud data and geological radar detection data;

[0048] Specifically, laser point cloud data uses a laser scanner to capture three-dimensional spatial information about the tunnel blasting area. This dataset contains a large number of point cloud coordinates. The point cloud data measured using laser scanning technology accurately reflects the geometry of the surrounding rock within the tunnel. The laser point cloud data contains detailed three-dimensional information about the tunnel surface and surrounding rock, with precise measurements capturing data features such as the three-dimensional coordinates and reflection intensity of each point. This point cloud data can be used to build an accurate three-dimensional model of the tunnel interior and provide important spatial data support for subsequent blasting quality evaluation. Processing and analysis of the laser point cloud data enables comparison of the tunnel's morphology before and after blasting, providing a basis for evaluating and optimizing blasting effectiveness.

[0049] Geological radar data captures the internal structure of the tunnel's surrounding rock using ground-penetrating radar equipment. This dataset, consisting of signals reflected from radar waves, analyzes key geological features such as the layer positions, crack distribution, and cavities in the tunnel's surrounding rock. The high-frequency bands of ground-penetrating radar clearly reveal the physical properties of the surrounding rock and potential geological problems. The images provided by ground-penetrating radar data can help identify the state of the surrounding rock before blasting and provide a crucial basis for subsequent blasting effectiveness assessment. Combined with other data sources, ground-penetrating radar data can further support the accurate assessment of blasting quality and the development of optimization plans.

[0050] This embodiment collects data through laser point cloud technology and geological radar technology. These two technologies can provide high-precision data support for tunnel blasting. After the data collection is completed, preprocessing is required. The preprocessing process includes steps such as noise removal, error correction, and data format conversion. Since laser point cloud data may be affected by external factors such as environmental changes and equipment errors, these data must be filtered to ensure data accuracy. In addition, geological radar data may also contain interference signals, which need to be denoised through specific algorithms, and the resulting reflected signals must be matched with the actual geological conditions. Ultimately, through data preprocessing, it is ensured that the collected data can be used for subsequent blasting quality evaluation and the establishment of optimization models, providing accurate and reliable basic data support.

[0051] First, data from the tunnel blasting area was collected using 3D laser scanning and geological radar technology. The 3D laser scanner rotates and scans the tunnel to obtain a large amount of high-precision point cloud data P (x, y, z), which details the tunnel contours and surrounding rock geometry. The geological radar uses high-frequency electromagnetic waves to detect the internal structure of the tunnel surrounding rock, recording the amplitude A and propagation time T of the radar reflection signal.

[0052] Data preprocessing specifically includes the following detailed steps:

[0053] Point cloud filtering: Use statistical filtering to remove noise points, which is achieved by the following formula:

[0054]

[0055] Where, is the centroid coordinate of the point cloud, and outliers with values ​​greater than a certain threshold are filtered out.

[0056] Point cloud error correction: Use the Iterative Closest Point (ICP) algorithm to achieve precise alignment between scan data and design models.

[0057] Data formatting: Unify data formats to standard XYZ coordinates for easy subsequent analysis.

[0058] Radar signal processing: Wavelet transform algorithm is used for filtering and denoising. The formula is:

[0059] S ' (t) = W -1 [W(S(t))×F t ]

[0060] Where W represents the wavelet transform operator, F t is a threshold function to improve data matching accuracy.

[0061] This example combines laser point cloud technology with geological radar technology to collect multi-dimensional data on the tunnel surrounding rock and the blasting process. This data fusion method can provide more comprehensive and accurate tunnel blasting information, thus providing an accurate basis for subsequent blasting quality evaluation and optimization.

[0062] S2. Based on the pre-processed multi-source data, blasting quality evaluation is performed according to local overbreak and underbreak, blasting contour line, average linear overbreak and underbreak, and point cloud extraction to obtain blasting quality evaluation results;

[0063] This embodiment provides accurate quality assessment by performing a refined analysis of the collected data. This assessment is divided into four levels: point (local overbreak and underbreak), line (blasting contour), surface (average linear overbreak and underbreak), and volume (point cloud extraction). Each level represents a different evaluation dimension and data analysis method.

[0064] First, the point (local over-excavation and under-excavation) evaluation evaluates the accuracy of local blasting by analyzing the over-excavation and under-excavation of each blasting point. The calculation of over-excavation and under-excavation helps to find possible control problems during the blasting process, thereby ensuring the stability of each blasting point. A one-to-one correspondence is established between the fused point cloud and the design contour, and each measured point p is i Projected onto the theoretical section, the local over-excavation and under-excavation are obtained If Δd i Exceeding limit Δd lim , automatically marked as "blasting deviation point" and counted the density to trace hidden dangers such as uneven charging of single holes or deviation of drilling holes.

[0065] Then, we enter the line level analysis, and evaluate the geometry of the actual blasting contour line and the theoretical blasting contour line to assess the accuracy of the blasting process. By analyzing the contour line, we can identify the uniformity of the blasting and potential errors, and ensure that the blasting effect is consistent with the design plan. The measured contour L is extracted in the projection plane of each construction cycle through the α-shape algorithm. a , and design the outline L d Perform β-spline reconstruction. Using Hausdorff distance Measure profile consistency; simultaneously calculate blast profile deviation rate to identify asymmetric excavation caused by uneven explosive coupling or sudden changes in lithology.

[0066] Based on the line-level evaluation, the surface-level indicators focus on the overall forming accuracy of the single-cycle section. The surface (average linear overbreak and underbreak) evaluation evaluates the overall control accuracy of the blasting by calculating the overbreak and underbreak of the entire blasting area. If the overbreak and underbreak are too large, it may affect the structural stability of the tunnel. Therefore, this evaluation can effectively reflect the quality level of the blasting. a and L dThe closed area is subjected to Boolean operation to obtain the over-excavation area A. ov and undercut area A uv The average linear over-excavation is Expression, where U d is the design contour perimeter. If the value is above the warning threshold, the control logic will recommend adjusting the charge amount of the surrounding hole line or the spacing between the holes.

[0067] Finally, we enter the volume-level integrity analysis. The volume (point cloud extraction) evaluation uses laser point cloud technology to obtain three-dimensional data after blasting, and extracts the geometric shape of the tunnel after blasting by analyzing the point cloud data. This method can comprehensively evaluate the integrity of the surrounding rock after blasting, identify possible cracks or cavities, and provide a basis for optimizing blasting parameters. With the help of kernel density estimation The 3D point cloud is partitioned into sparse and dense parts, and the GPR pre-stack migration section R is combined with the mig (x,z) identifies potential cavities and cracks. M3C2 quantitatively compares point clouds before and after blasting, outputting a volumetric damage index (DI) for the surrounding rock. This DI, along with previous point, line, and surface metrics, is fed into an entropy weighting model to generate a comprehensive blasting quality score, which is used for closed-loop optimization of charge parameters.

[0068] This embodiment divides the blasting quality evaluation into four levels: point (local over-break and under-break), line (blasting contour line), surface (average linear over-break and under-break) and volume (point cloud extraction), and conducts a detailed analysis to comprehensively evaluate the blasting effect, identify potential problems, and ensure the accuracy and safety of the blasting.

[0069] S3. Classifying the blasting quality according to the blasting quality evaluation results, and calculating and grading the comprehensive blasting quality score;

[0070] Specifically, the results of blasting quality evaluations need to be graded to enable more targeted optimization and improvement of blasting performance. Blasting quality is graded based on various blasting quality assessment indicators, such as localized overbreak and underbreak, blasting profile deviation, and overbreak and underbreak volume, according to their impact on tunnel stability and safety. Each grade represents a different blasting effect, ranging from excellent to unsatisfactory. This provides an intuitive understanding of the overall level of blasting quality and provides a clear basis for subsequent optimization.

[0071] Furthermore, a weight is calculated for each blasting quality evaluation indicator based on its importance to the blasting outcome. This weighting allows for objective quantification of the relative importance of different indicators, providing an accurate reference for comprehensive evaluation. Localized overbreak and underbreak generally have a greater impact on blasting quality, so their weighting is relatively high. The weights of other indicators, such as the blasting contour and the amount of overbreak and underbreak, are adjusted based on actual project needs. By calculating and assigning these weights, we ensure that each evaluation indicator contributes appropriately to the comprehensive evaluation, thereby improving the accuracy and scientific nature of blasting quality assessment.

[0072] To achieve closed-loop control, the four core indicators (local over-excavation rate γ1, contour deviation rate γ2, average linear over-excavation γ3, and volume damage index DI) were normalized to a membership score ni on a scale of 0-100. The weights wi were assigned using the analytic hierarchy process (AHP) (typical values: γ1 = 0.35, γ3 = 0.25, γ2 = 0.20, DI = 0.20, and consistency test CR ≤ 0.10). The comprehensive score formula is as follows:

[0073] Q=100×(w1·n1+w2·n2+w3·n3+w4·n4) / ∑w i

[0074] Blasting quality is categorized into four levels, A–D, based on the Q range: 85–100 (excellent), 70–85 (good), 55–70 (fair), and 0–55 (unacceptable). Lower scores indicate a greater potential impact on tunnel stability, leading to the development of appropriate parameter optimization strategies for different surrounding rock conditions.

[0075] Table 1 lists the recommended thresholds (in cm) for each gear level on the three geometric indicators. When any indicator exceeds the corresponding threshold, it is judged as a poor gear and the poor gear is used as the final comprehensive grade.

[0076] Table 1

[0077]

[0078] S4. Establish a database containing geological parameters, surrounding rock response parameters and blasting process parameters, generate a blasting parameter optimization plan through convolutional neural network model training, and dynamically adjust the blasting parameters of the next cycle according to the comprehensive blasting quality score.

[0079] Furthermore, the blasting parameter optimization model achieves optimization by building a database containing multiple blasting scenarios. Based on constraints such as over-break, under-break, residual void ratio, and degree of fragmentation, an advanced algorithm is used to automatically select and adjust blasting parameters to achieve the optimal blasting parameters.

[0080] Specifically, on-site blasting data, including geological parameters, surrounding rock parameters, and blasting parameters, was collected. This data encompasses the tunnel's geological conditions, the physical and mechanical properties of the surrounding rock, and various parameters used in the actual blasting process, such as blasting charge, blasthole layout, and blasting time. By integrating this data, a high-quality database was established: ① geological parameters (lithology, RQD, in-situ stress, etc.), updated in real time by measurement while drilling and geological radar; ② surrounding rock response parameters (point cloud error field, GPR fracture echo energy, volume damage index DI); and ③ blasting process parameters (blasthole coordinates, hole diameter, charge structure, detonation sequence, and charge amount). All records were written to a PostgreSQL data warehouse using the "cycle number - hole number - timestamp" index, and a tensor sample X = [G, R, B] was simultaneously generated, with the label Y being the comprehensive score Q of the previous cycle.

[0081] Next, a convolutional neural network was used to build a model and perform feature extraction. In the offline phase, the training / validation set was split 8:2, and a 1D-CNN-BiLSTM combined network was used to extract temporal-spatial features: convolutional layers (kernel = 3, channel = 64) encoded local parameter associations, followed by a bidirectional LSTM layer that captured inter-loop dependencies. A two-layer fully connected network with softmax was then added on top to predict the Q level. The network used the Adam optimizer (lr = 1e-3) and a cross-entropy loss, freezing weights when F1 ≥ 0.92 on the validation set. SHAP was also used to decompose feature contributions to generate a "parameter sensitivity dictionary." Convolutional neural networks can automatically extract key features from complex blasting data and identify underlying patterns between different blasting schemes. Through deep learning training, the model can be continuously optimized, learning how to adjust blasting parameters based on geological conditions and surrounding rock characteristics to maximize blasting effectiveness.

[0082] The trained model is used to automatically optimize the blasting parameters. The trained model is deployed on the edge server. After each excavation cycle is completed, the model is called to infer the gradient vector of ΔQ / ΔB. Combined with the constraints (blasting vibration velocity, support thickness, safety factor), a new round of parameters B*=(hole spacing*, charge*, detonation interval*) is generated using the Bayesian optimizer in the 3-dim search space. If the model predicts that Q is improved by ≥3 points and meets the on-site safety threshold, the trolley control script is automatically written; if it does not meet the requirements, it will fall back to manual review. The closed loop can converge to the local optimal parameters within 2-3 cycles. The model can automatically adjust blasting parameters such as blasthole arrangement, charge distribution and blasting timing based on real-time collected field data to ensure that each blasting can achieve the best effect.

[0083] This embodiment proposes a database containing multiple blasting schemes to optimize blasting parameters. Based on constraints such as over-excavation, under-excavation, residual void ratio, and degree of fragmentation, advanced algorithms are used to automatically select and adjust blasting parameters to achieve the best blasting results.

[0084] Beneficial effects of this embodiment:

[0085] The present invention deeply integrates the centimeter-level external contour information of the laser point cloud with the sub-meter-level dielectric characteristics of the geological radar, and then superimposes the stress-strain and blasting process parameters collected in real time during the construction process to achieve multi-source information fusion of the three domains of "geometry-physical property-process". With the help of the four-level evaluation system of point-line-surface-body - local over-break (point), contour deviation (line), average linear over-break (surface) and volume damage index (body) extracted from point cloud - it is possible to distinguish the hierarchical contribution of micro-drilling deviation, overall cross-section forming error and three-dimensional surrounding rock damage distribution. The evaluation results are superior to a single indicator system in both spatial resolution and physical interpretability, thus providing a more scientific quantitative basis for risk assessment and support design.

[0086] This method establishes a database containing multiple blasting scenarios and uses machine learning algorithms to automatically optimize blasting parameters. Compared with manual selection methods, this method has a wider search space, more efficient calculations, and better parameter selection, thus achieving accurate prediction and control of blasting effects.

[0087] Example 2

[0088] This embodiment further discloses a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the method described in the first embodiment are implemented.

[0089] The above are merely preferred embodiments of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. Tunnel blasting quality evaluation and optimization method based on multi-source data fusion, characterized by: The following steps are involved: S1. Collect and pre-process multi-source data of the tunnel blasting area, wherein the multi-source data includes laser point cloud data and geological radar detection data; S2. Based on the pre-processed multi-source data, blasting quality evaluation is performed according to local overbreak and underbreak, blasting contour line, average linear overbreak and underbreak, and point cloud extraction to obtain blasting quality evaluation results; S3. Classifying the blasting quality according to the blasting quality evaluation results, and calculating and grading the comprehensive blasting quality score; S4. Establish a database containing geological parameters, surrounding rock response parameters and blasting process parameters, generate a blasting parameter optimization plan through convolutional neural network model training, and dynamically adjust the blasting parameters of the next cycle according to the comprehensive blasting quality score.

2. The tunnel blasting quality evaluation and optimization method based on multi-source data fusion according to claim 1 is characterized in that: In step S1, 3D laser scanning and geological radar technology are used to collect data from the tunnel blasting area. Using a 3D laser scanner, a large amount of high-precision point cloud data P (x, y, z) is acquired through rotational scanning within the tunnel, which details the tunnel contours and the geometry of the surrounding rock. The geological radar equipment uses high-frequency electromagnetic waves to detect the internal structure of the tunnel surrounding rock, recording the amplitude A and propagation time T of the radar reflection signal. Data preprocessing specifically includes the following detailed steps: Point cloud filtering: Use statistical filtering to remove noise points, which is achieved by the following formula: Where, is the centroid coordinate of the point cloud, and outliers with values ​​greater than a certain threshold are filtered out. Point cloud error correction: Use the Iterative Closest Point (ICP) algorithm to achieve precise alignment between scan data and design models. Data formatting: Unify data format to standard XYZ coordinates. Radar signal processing: Wavelet transform algorithm is used for filtering and denoising. The formula is: S'(t)=W -1 [W(S(t))×F t ] Where W represents the wavelet transform operator, F t is a threshold function to improve data matching accuracy.

3. The tunnel blasting quality evaluation and optimization method based on multi-source data fusion according to claim 1 is characterized in that: The process of collecting and preprocessing multi-source data in the tunnel blasting area includes: Collect laser point cloud data through 3D laser scanner; Collect geological radar detection data through geological radar technology; Based on the laser point cloud data, the statistical filtering method is used to remove point cloud noise, filter out abnormal points whose centroid distance is greater than the threshold, and correct the point cloud error through the iterative closest point algorithm; Based on the geological radar detection data, the wavelet transform algorithm is used to remove noise.

4. The tunnel blasting quality evaluation and optimization method based on multi-source data fusion according to claim 1 is characterized in that: The process of evaluating blasting quality based on local over-break and under-break includes: Project each measured point onto the theoretical cross section to obtain the local over-excavation and under-excavation ratio; If the local over-excavation and under-excavation rate exceeds a threshold, it is automatically marked as a blasting deviation point and the density is counted.

5. The tunnel blasting quality evaluation and optimization method based on multi-source data fusion according to claim 1 is characterized in that: The process of blasting quality evaluation according to blasting contour line includes: The measured contour is extracted in the projection plane of each construction cycle, and the designed contour is reconstructed by spline; The Hausdorff distance is used to measure the contour consistency, and the deviation rate of the blasting contour line is calculated.

6. The tunnel blasting quality evaluation and optimization method based on multi-source data fusion according to claim 1 is characterized in that: The process of evaluating blasting quality based on average linear over-break and under-break includes: Perform Boolean operations on the closed area formed by the measured contour and the designed contour to obtain the over-excavation area and under-excavation area; The average linear over-break and under-break rate is calculated. If the average linear over-break and under-break rate is higher than the warning threshold, it is recommended to adjust the charge amount of the surrounding hole lines or the spacing between the blastholes.

7. The tunnel blasting quality evaluation and optimization method based on multi-source data fusion according to claim 1 is characterized in that: The process of blasting quality evaluation based on point cloud extraction includes: The 3D point cloud is partitioned into sparse and dense parts using kernel density estimation, and potential cavities and cracks are identified by combining geological radar prestack migration profiles. The M3C2 algorithm is used to quantitatively compare the point clouds before and after blasting, and output the surrounding rock volume damage index.

8. The tunnel blasting quality evaluation and optimization method based on multi-source data fusion according to claim 1 is characterized in that: The process of grading blasting quality and calculating and grading the comprehensive blasting quality score includes: Normalize the local over-excavation rate, contour line deviation rate, average linear over-excavation rate and surrounding rock volume damage index to obtain the normalized value; assigning weights by the analytic hierarchy process, and calculating a comprehensive blasting quality score based on the normalized values ​​and the weights; The blasting quality is divided into four levels according to the range of comprehensive blasting quality scores.

9. The tunnel blasting quality evaluation and optimization method based on multi-source data fusion according to claim 1 is characterized in that: The records of the database are stored in the PostgreSQL database according to the cycle number-hole number-timestamp index.

10. The tunnel blasting quality evaluation and optimization method based on multi-source data fusion according to claim 1 is characterized in that: The process of dynamically adjusting the blasting parameters for the next cycle based on the comprehensive blasting quality score includes: Call the convolutional neural network model to output the gradient vector; Based on the gradient vector, a new round of optimization parameters is generated in the three-dimensional space of blasthole spacing, charge amount, and detonation interval through a Bayesian optimizer; When the comprehensive blasting quality score increases by ≥3 points and meets the safety threshold, the trolley control script is written and executed.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the tunnel blasting quality evaluation and optimization method based on multi-source data fusion according to any one of claims 1 to 10 are implemented.

Images