An intelligent optimization method for tunnel blasting monitoring point layout based on data fusion

Abstract

The application discloses an intelligent optimization method for tunnel blasting monitoring point layout based on data fusion, and relates to the technical field of tunnel engineering safety monitoring. The method comprises the following steps: collecting geological parameters, structural parameters, blasting parameters and early monitoring data to form a multi-source heterogeneous data set; pre-processing the multi-source heterogeneous data set, and fusing and feature mining the pre-processed data to identify a monitoring key area; based on the fused data and the monitoring key area, an improved deep Q network model is constructed with monitoring coverage, monitoring accuracy and monitoring cost as optimization objectives, and an initial layout scheme of the monitoring points is output through iterative optimization; the monitoring points are laid out on site according to the initial layout scheme, and a linkage feedback mechanism of the monitoring data and the layout scheme is established to collect the on-site monitoring data and construction condition changes in real time; when the monitoring data or the construction condition meets a preset triggering condition, the improved deep Q network model is called again to dynamically adjust the layout scheme.

Description

Technical Field

[0001] This invention relates to the field of tunnel engineering safety monitoring technology, and more specifically to an intelligent optimization method, device, equipment, and readable storage medium for the deployment of tunnel blasting monitoring points based on data fusion. Background Technology

[0002] With the continuous improvement of the digitalization and intelligence level of transportation infrastructure, the monitoring and safety management during engineering construction is gradually shifting from traditional experience-driven to data-driven and intelligent decision-making. In the tunnel blasting construction process, the layout of monitoring points is essentially a resource allocation and optimization decision-making problem under complex constraints. Its goal is to achieve comprehensive optimization among monitoring coverage, monitoring accuracy, and implementation cost under limited monitoring resources.

[0003] However, existing monitoring point deployment methods are mostly based on empirical rules or simple heuristic strategies, lacking systematic integration and in-depth utilization of multi-source data, making it difficult to support the intelligent decision-making needs in complex engineering environments. Specifically, existing technologies have the following problems: First, there is a lack of decision support mechanisms based on multi-source data fusion. Existing methods typically do not perform unified modeling and correlation analysis of geological information, structural information, construction parameters, and historical monitoring data, resulting in a lack of data-driven basis for the deployment of monitoring points, making it difficult to effectively identify key risk areas, and thus affecting the accuracy of deployment decisions. Secondly, there is a lack of optimization decision-making methods oriented towards multi-objective constraints. The deployment of monitoring points requires a trade-off between multiple objectives such as monitoring coverage, monitoring accuracy, and deployment cost. However, existing methods usually rely on a single indicator or empirical adjustments, lacking a unified optimization model and making it difficult to achieve coordinated optimization among multiple objectives. Secondly, there is a lack of dynamic and adaptive decision-making update mechanisms. During construction, environmental conditions and construction parameters change continuously, but the existing deployment plan usually remains unchanged once it is determined. The lack of dynamic adjustment capabilities based on real-time data feedback makes it difficult for decision-making results to respond to external changes in a timely manner, reducing the effectiveness and adaptability of the overall monitoring system.

[0004] Therefore, there is an urgent need for a monitoring point deployment method based on multi-source data fusion and intelligent optimization algorithms to achieve efficient allocation of monitoring resources and dynamic decision optimization in complex engineering scenarios. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent optimization method, device, equipment, and readable storage medium for the deployment of tunnel blasting monitoring points based on data fusion. First, by uniformly collecting, preprocessing, and fusing geological parameters, structural parameters, blasting parameters, and pre-monitoring data, a comprehensive analysis framework for multi-source heterogeneous data is constructed. Based on the correlation characteristics between data, key areas are identified, providing data-driven decision-making basis for monitoring point deployment and improving the accuracy and scientific nature of deployment decisions. Second, by constructing an improved deep Q-network model, the monitoring point deployment problem is abstracted into a sequential decision problem under multi-objective constraints. An iterative learning mechanism is used to continuously optimize the deployment strategy, enabling synergistic optimization of monitoring resources in terms of coverage, monitoring accuracy, and implementation cost, improving resource allocation efficiency and reducing redundant deployment. Third, by establishing a feedback update mechanism based on real-time monitoring data, environmental changes and construction status are dynamically perceived during construction, triggering adaptive adjustments to the deployment strategy. This achieves dynamic optimization and continuous updating of the monitoring point deployment scheme, thereby improving the responsiveness and adaptability of decision results to complex working conditions.

[0006] To achieve the above objectives, the present invention provides the following technical solution: In a first aspect, the present invention provides an intelligent optimization method for the deployment of tunnel blasting monitoring points based on data fusion, the method comprising: Geological parameters, structural parameters, blasting parameters, and previous monitoring data were collected to form a multi-source heterogeneous dataset; The multi-source heterogeneous dataset is preprocessed, and the preprocessed data is fused and feature mined to identify key monitoring areas. Based on the fused data and key monitoring areas, an improved deep Q-network model is constructed with monitoring coverage, monitoring accuracy and monitoring cost as optimization objectives. The initial deployment scheme of monitoring points is output through iterative optimization. Based on the initial deployment plan, monitoring points are set up on site, and a linkage feedback mechanism between monitoring data and deployment plan is established. On-site monitoring data and changes in construction conditions are collected in real time. When the monitoring data or construction conditions meet the preset trigger conditions, the improved deep Q-network model is called again to dynamically adjust the deployment plan.

[0007] In some embodiments, the step of collecting multi-source heterogeneous data includes: The collected geological parameters include the surrounding rock grade, rock strength, distribution of joints and fissures, and groundwater conditions. The collected structural parameters include the thickness of the old and new tunnel linings, lining stiffness, lining aging coefficient, and connection method. The collected blasting parameters include the amount of explosives, the depth of the blast holes, the spacing between the blast holes, the detonation sequence, and the number of detonators. The collected preliminary monitoring data includes peak blasting vibration, dominant frequency, attenuation pattern, settlement amount, and convergence displacement.

[0008] In some embodiments, preprocessing of multi-source heterogeneous datasets includes: Outliers were removed from geological and structural parameters using the 3σ principle and then normalized. Wavelet transform algorithm is used to filter and denoise the blasting parameters and monitoring data, and the main frequency is extracted by fast Fourier transform. The preprocessed data is formatted uniformly and associated with grid cell coordinates.

[0009] In some embodiments, the preprocessed data is fused and feature-mined, including: A multi-dimensional data fusion model of "geology-structure-blasting-monitoring" was constructed. The random forest algorithm was used to calculate the influence weight of various data on the effectiveness of monitoring points, and the multi-source data were weighted and fused according to the weight. The Pearson correlation analysis method was used to quantify the correlation coefficient between the blasting charge of the new tunnel and the vibration response of the old tunnel. Based on the fused data values ​​and correlation coefficients, key monitoring areas are identified, including weak sections of the old tunnel, the connection section between the old and new tunnels, and the section affected by the excavation of the central pilot tunnel. Areas composed of grid cells with fused data values ​​≥ 0.7 are identified as key monitoring areas.

[0010] In some embodiments, an improved deep Q-network model is constructed with monitoring coverage, monitoring accuracy, and monitoring cost as optimization objectives, including: The monitoring range is divided into 1m×1m grid units, and the state of each grid unit is composed of its corresponding fused data features; Define the action space, which includes setting up monitoring points, encrypting monitoring points, removing monitoring points, and adjusting the position of monitoring points; Construct a comprehensive reward function R = 0.4×R1 + 0.4×R2 - 0.2×R3, where R1 is the monitoring coverage reward, R2 is the data quality reward, and R3 is the cost control reward; Set the algorithm parameters for the improved deep Q-network model, including a learning rate of 0.001, a discount factor of 0.9, an experience replay buffer size of 10000, a target network update frequency of 100 steps, and an exploration rate that linearly decreases from 0.9 to 0.1. Starting with an initial plan based on the principle of "dense near and sparse far", the fused data is used as input. Through iterative optimization, the reward function is converged, and the optimal monitoring point deployment plan is output.

[0011] In some embodiments, a linkage feedback mechanism between monitoring data and deployment scheme is established to collect on-site monitoring data and changes in construction conditions in real time. When the monitoring data or construction conditions meet preset trigger conditions, the improved deep Q-network model is invoked again to dynamically adjust the deployment scheme, including: Real-time acquisition of blasting vibration data and static deformation data; When the vibration peak value in a certain area reaches more than 70% of the warning benchmark value, or the daily settlement reaches more than 0.5mm, the improved deep Q network model is automatically triggered to densify the monitoring points in the area, reduce the spacing between monitoring points to 2-3m, and adjust the sensor sampling rate to 200Hz. When the peak vibration value of a certain area is lower than 50% of the warning benchmark value for 7 consecutive days and the daily settlement is lower than 0.2mm, and there is no significant change in geological conditions and blasting conditions, the redundant monitoring points are automatically reduced, the spacing between monitoring points is adjusted to 5-6m, and the sensor sampling rate is reduced to 100Hz. The deployment plan, monitoring data and operating parameters after each dynamic adjustment are stored in the database, and the parameters of the improved deep Q network model are updated every 15 days. When blasting parameters are adjusted, geological conditions change, or procedures are switched, the system automatically triggers the improved depth Q network model to re-optimize the layout scheme, achieving real-time adaptation to the construction conditions.

[0012] In some embodiments, the method further includes: The monitoring point deployment scheme was calibrated using a process of "simulation optimization - on-site calibration - iterative improvement". Pilot deployment was carried out in typical areas. By comparing the monitoring effects of the optimized scheme and the traditional scheme, the weights of the data fusion model and the parameters of the deep Q-network model were adjusted and improved. Establish an evaluation index system for the effectiveness of monitoring points, and evaluate them from three dimensions: monitoring coverage, data accuracy, and cost control. The effectiveness evaluation indexes are: 100% coverage of key areas, 95% or more coverage of the total monitoring range, dynamic vibration data measurement error not exceeding 5%, static deformation data measurement error not exceeding 3%, effective data volume not less than 90%, and the number of monitoring points reduced by more than 20% compared with traditional schemes. If the evaluation results do not meet the preset standards, the data fusion model or the parameters of the deep Q-network model will be adjusted and the deployment plan will be re-optimized until the evaluation standards are met, and then the final monitoring point deployment plan will be determined.

[0013] Secondly, the present invention also provides an intelligent optimization device for the deployment of tunnel blasting monitoring points based on data fusion, the device comprising: The data acquisition module is used to collect geological parameters, structural parameters, blasting parameters and previous monitoring data to form a multi-source heterogeneous dataset; The data processing module is used to preprocess multi-source heterogeneous datasets, and to fuse and mine the preprocessed data to identify key monitoring areas. The model building module is used to construct an improved deep Q-network model based on the fused data and key monitoring areas, with the optimization objectives of monitoring coverage, monitoring accuracy and monitoring cost, and output the initial layout scheme of monitoring points through iterative optimization. The scheme adjustment module is used to set up monitoring points on site according to the initial deployment scheme and establish a linkage feedback mechanism between monitoring data and deployment scheme. It collects on-site monitoring data and changes in construction conditions in real time. When the monitoring data or construction conditions meet the preset trigger conditions, it calls the improved deep Q network model to dynamically adjust the deployment scheme.

[0014] Thirdly, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the intelligent optimization method for the deployment of tunnel blasting monitoring points based on data fusion provided in the first aspect.

[0015] Fourthly, the present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the intelligent optimization method for the deployment of tunnel blasting monitoring points based on data fusion provided in the first aspect.

[0016] Fifthly, the present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the intelligent optimization method for the deployment of tunnel blasting monitoring points based on data fusion provided in the first aspect.

[0017] The beneficial effects of this invention are as follows: First, by uniformly collecting, preprocessing, and fusing geological parameters, structural parameters, blasting parameters, and pre-monitoring data, a comprehensive analysis framework for multi-source heterogeneous data is constructed. Based on the correlation characteristics between data, key areas are identified, providing data-driven decision-making basis for monitoring point deployment and improving the accuracy and scientific nature of deployment decisions. Second, by constructing an improved deep Q-network model, the monitoring point deployment problem is abstracted into a sequential decision problem under multi-objective constraints. An iterative learning mechanism continuously optimizes the deployment strategy, enabling coordinated optimization of monitoring resources in terms of coverage, monitoring accuracy, and implementation cost, improving resource allocation efficiency and reducing redundant deployment. Third, by establishing a feedback update mechanism based on real-time monitoring data, environmental changes and construction status are dynamically perceived during construction, triggering adaptive adjustments to the deployment strategy. This achieves dynamic optimization and continuous updating of the monitoring point deployment scheme, thereby improving the responsiveness and adaptability of decision results to complex working conditions.

[0018] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, the preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating an intelligent optimization method for the deployment of tunnel blasting monitoring points based on data fusion, according to an embodiment of the present invention. Figure 2 This is a flowchart illustrating another intelligent optimization method for the deployment of tunnel blasting monitoring points based on data fusion, as shown in an embodiment of the present invention. Figure 3 This is a schematic diagram of the structure of a detection system according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of an intelligent optimization device for the deployment of tunnel blasting monitoring points based on data fusion, according to an embodiment of the present invention. Figure 5 This is a schematic diagram of an electronic device structure provided in an embodiment of this application. Detailed Implementation

[0020] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] It should be noted that references to "an embodiment," "embodiment," "example embodiment," etc., in this specification refer to the described embodiment including specific features, structures, or characteristics; however, not every embodiment must include these specific features, structures, or characteristics. Furthermore, such expressions do not refer to the same embodiment. Moreover, when describing specific features, structures, or characteristics in conjunction with embodiments, whether or not explicitly described, it is indicated that incorporating such features, structures, or characteristics into other embodiments is within the knowledge of those skilled in the art.

[0022] Furthermore, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0023] In some embodiments, such as Figure 1 As shown, this application provides a flowchart illustrating an intelligent optimization method for the deployment of tunnel blasting monitoring points based on data fusion. The specific method includes: S101 collects geological parameters, structural parameters, blasting parameters, and previous monitoring data to form a multi-source heterogeneous dataset.

[0024] The geological parameters collected include the surrounding rock grade, rock strength, joint and fissure distribution, and groundwater conditions; the structural parameters collected include the thickness of the new and old tunnel linings, lining stiffness, lining aging coefficient, and connection method; the blasting parameters collected include the amount of explosives, blast hole depth, blast hole spacing, detonation sequence, and number of detonators; and the preliminary monitoring data collected include the peak value of blasting vibration, dominant frequency, attenuation law, settlement, and convergence displacement.

[0025] For example, in this embodiment, multi-source heterogeneous data collection is first carried out for a "two-new-two-old" long arch tunnel (specifically, the Hainan Damao Tunnel, which includes two existing old tunnels, two newly built tunnels, and a central pilot tunnel). The "two-new-two-old" long arch tunnel refers to a long-distance arch tunnel structure composed of two existing old tunnels and two newly built tunnels. This type of tunnel is characterized by different construction dates, varying material properties, inconsistent structural stiffness, and stress concentration at connection points. It is significantly affected by blasting operations during reconstruction and expansion, posing special requirements for the placement of monitoring points.

[0026] The collected geological parameters include: surrounding rock grade (II-V), rock strength (30-150MPa), joint and fissure distribution (spacing 0.5-5m), and groundwater conditions (dry, wet, seeping). The above data were obtained using a combination of ground-penetrating radar detection and on-site sampling. For key areas (such as fractured zones and seeping sections), data were collected at a rate of one set every 0.5m.

[0027] The collected structural parameters include: the thickness of the old tunnel lining (30-50cm), lining stiffness (20-40GPa), lining aging coefficient (0.7-0.9), and connection method (rigid connection / flexible connection); the thickness of the new tunnel lining (40-60cm) and lining stiffness (35-50GPa). These data were obtained through on-site measurements, drawing review, and testing equipment. For each key structural section (such as weak sections of the old tunnel or the connection between the old and new sections), 3-5 sets of data were collected, and the average value was taken as the structural parameter value for that area.

[0028] The collected blasting parameters include: explosive charge (maximum charge per section 8-12kg), borehole depth (2.0-2.5m), borehole spacing (0.8-1.2m), detonation sequence (millisecond delay detonation), and number of detonators (10-15 per section). Parameter changes for each blast are recorded in real time and correlated with the corresponding construction procedures (tunnel excavation, main tunnel widening, etc.) and the monitoring area.

[0029] The collected preliminary monitoring data includes: peak blasting vibration, dominant frequency, attenuation pattern, settlement, and convergence displacement. The data collection period is no less than 15 days, with static data collected three times a day (8:00 AM, 2:00 PM, and 8:00 PM), increasing to once every hour during blasting operations; dynamic data is collected in real time during blasting operations and once every two hours during non-blasting periods to ensure the capture of instantaneous vibration response and long-term static deformation trends.

[0030] S102 preprocesses the multi-source heterogeneous dataset, and then fuses and mines the preprocessed data to identify key monitoring areas.

[0031] The preprocessing of the multi-source heterogeneous dataset included: removing outliers from geological and structural parameters using the 3σ principle and normalizing them; filtering and denoising blasting parameters and monitoring data using wavelet transform and extracting the dominant frequency using fast Fourier transform; and unifying the format of the preprocessed data and associating it with grid cell coordinates. The preprocessed data was then fused and its features were mined, including: constructing a multi-dimensional data fusion model of "geology-structure-blasting-monitoring"; using the random forest algorithm to calculate the influence weights of various data on the effectiveness of monitoring points; and weighting and fusing the multi-source data according to these weights; using Pearson correlation analysis to quantify the correlation coefficient between the blasting charge of the new tunnel and the vibration response of the old tunnel; identifying key monitoring areas based on the fused data values ​​and correlation coefficients, including weak sections of the old tunnel, the connection section between the new and old tunnels, and the section affected by the excavation of the pilot tunnel; and identifying areas composed of grid cells with fused data values ​​≥ 0.7 as key monitoring areas.

[0032] For example, after data acquisition is completed, the aforementioned multi-source heterogeneous dataset is preprocessed. The preprocessing employs a hierarchical processing strategy: First, outliers were removed from geological and structural parameters using the 3σ principle (i.e., data deviating more than three standard deviations from the mean). The data after outlier removal was then normalized to the [0,1] interval, calculated using the formula x'=(x-xmin) / (xmax-xmin). Missing data were supplemented using linear interpolation to ensure data integrity in each grid cell.

[0033] Secondly, wavelet transform algorithm was used to filter and denoise the blasting parameters and monitoring data. The db4 wavelet basis was selected and the number of decomposition layers was set to 5 to remove noise signals such as natural interference and mechanical vibration. The dominant frequency (5-50Hz) was extracted by fast Fourier transform (FFT) to distinguish blasting vibration from other vibration sources. Linear trend fitting was used for static monitoring data to remove the influence of temperature (temperature correction coefficient was set to 0.0015 / ℃). Outliers in the blasting parameters were corrected or removed in conjunction with the construction log.

[0034] Finally, all preprocessed data were converted to CSV format and associated with grid cell coordinates. The monitoring area was divided into 1m×1m grid cells using an equal grid division method. Each grid cell was labeled with its location coordinates and associated with its corresponding geological and structural parameters.

[0035] The preprocessed data is then fused and its features are mined, specifically including the following steps: A multi-dimensional data fusion model integrating geology, structure, blasting, and monitoring was constructed. A random forest algorithm was used to calculate the influence weights of various data types on the effectiveness of monitoring points: preprocessed data was used as input, and the accuracy of monitoring points in capturing vibration response and structural deformation was used as output to train the random forest model. The number of decision trees in the model was set to 100, the maximum depth to 10, and the influence weights of various data types were output. For example, the weight for the stiffness of the old tunnel lining was 0.32, the weight for the surrounding rock grade was 0.28, and the weight for the blasting charge amount was 0.20. The multi-source data were then weighted and fused based on the calculated weights.

[0036] Pearson correlation analysis was used to quantify the correlation coefficient between the blasting charge amount of the new tunnel and the vibration response of the old tunnel. Referring to data from the Hainan Damao Tunnel, the correlation coefficient between the blasting charge amount of the new tunnel and the vibration response of the old tunnel reached 0.83, clearly identifying the side of the pilot tunnel in the old tunnel as the key monitoring area.

[0037] Based on the fused data values ​​and the aforementioned correlation coefficients, key monitoring areas were identified. Areas comprised of grid cells with fused data values ​​≥ 0.7 (after normalization) were identified as key monitoring areas, specifically including weak sections of the old tunnel, the connection section between the old and new tunnels, and the section affected by the excavation of the central pilot tunnel (within 50m of the excavation face). Furthermore, based on Pearson correlation analysis results, the side of the old tunnel adjacent to the central pilot tunnel was further confirmed as a key area for intensive monitoring.

[0038] S103, based on the fused data and key monitoring areas, constructs an improved deep Q-network model with monitoring coverage, monitoring accuracy and monitoring cost as optimization objectives, and outputs the initial deployment scheme of monitoring points through iterative optimization.

[0039] The process involves constructing an improved deep Q-network model with monitoring coverage, accuracy, and cost as optimization objectives. This includes: dividing the monitoring area into 1m×1m grid cells, where the state of each grid cell is determined by its corresponding fused data features; defining an action space that includes deploying monitoring points, increasing the density of monitoring points, removing monitoring points, and adjusting monitoring point positions; constructing a comprehensive reward function R = 0.4×R1 + 0.4×R2 - 0.2×R3, where R1 is the monitoring coverage reward, R2 is the data quality reward, and R3 is the cost control reward; setting the algorithm parameters for the improved deep Q-network model, including a learning rate of 0.001, a discount factor of 0.9, an experience replay buffer size of 10000, a target network update frequency of 100 steps, and an exploration rate that linearly decreases from 0.9 to 0.1; starting with an initial deployment scheme based on the "near-dense, far-sparse" principle, using the fused data as input, iteratively optimizing the reward function to converge and outputting the optimal monitoring point deployment scheme.

[0040] Deep Q-Network (DQN) is an algorithm that combines deep learning and reinforcement learning. It approximates the Q-function through a neural network to achieve optimal policy learning in a high-dimensional state space. The "improvements" in this invention are specifically reflected in: designing a multi-objective comprehensive reward function that includes monitoring coverage, data quality, and cost control for the monitoring point deployment optimization problem; employing priority experience replay and target network techniques to improve the algorithm's convergence speed and stability; and establishing a dynamic optimization mechanism that links with real-time monitoring data.

[0041] For example, based on the fused data and the identified key monitoring areas, an improved deep Q-network model is constructed with monitoring coverage, monitoring accuracy and monitoring cost as optimization objectives, and the initial deployment scheme of the output monitoring points is optimized through iterative optimization.

[0042] First, the monitoring area is divided into 1m×1m grid cells, and the state of each grid cell is composed of its corresponding fused data features. The state vector contains fused feature parameters such as the geological conditions, structural characteristics, and blasting disturbance intensity of the cell.

[0043] Define the action space, which includes four actions: setting up monitoring points, encrypting monitoring points, removing monitoring points, and adjusting the position of monitoring points.

[0044] Construct a comprehensive reward function R = 0.4×R1 + 0.4×R2 - 0.2×R3, where: R1 is the monitoring coverage reward, calculated as R1 = actual monitoring coverage / target coverage, with the target coverage set to ≥95%; R2 is the data quality reward, calculated as R2 = effective monitoring data volume / total monitoring data volume, with the effective data volume requirement ≥90%; R3 is the cost control reward, calculated as R3 = actual number of monitoring points / traditional number of monitoring points, with lower costs resulting in smaller R3 and higher rewards.

[0045] Set the algorithm parameters for the improved deep Q-network model as follows: learning rate is set to 0.001, discount factor is set to 0.9, experience replay buffer size is set to 10000, target network update frequency is set to 100 steps, and exploration rate ε is initially set to 0.9, which decreases linearly to 0.1 with the number of iterations (decay rate is 0.001 / iteration step).

[0046] Starting with an initial deployment scheme based on the "dense near, sparse far" principle (16 monitoring points initially deployed), the fused data was used as input to initiate iterative optimization of the improved deep Q-network model. During each iteration, the algorithm selected the optimal deployment action and updated the Q-value based on the grid cell state and reward function. After 1200 iterations, the reward function value fluctuation was ≤0.01 for 50 consecutive steps, indicating algorithm convergence.

[0047] Output the optimal monitoring point layout scheme, with a total of 12 monitoring points, specifically meeting the following layout requirements: the monitoring point density of the old tunnel is increased by 30% compared to the new tunnel (6 points in the old tunnel and 4 points in the new tunnel); 2 additional sets of dense monitoring points are added at the connection section between the old and new tunnels, with the spacing controlled at 4m; a set of monitoring points is placed every 3m within 50m around the excavation face of the central pilot tunnel (a total of 3 points).

[0048] The optimized scheme was simulated and verified using the Sadovsky vibration propagation model. The monitoring coverage reached 98%, and the vibration peak measurement error was ≤4.2%, which met the preset requirements.

[0049] S104. Based on the initial deployment plan, monitoring points are deployed on site, and a linkage feedback mechanism between monitoring data and deployment plan is established. On-site monitoring data and changes in construction conditions are collected in real time. When the monitoring data or construction conditions meet the preset trigger conditions, the improved deep Q-network model is called again to dynamically adjust the deployment plan.

[0050] The linkage feedback mechanism refers to establishing a closed-loop system that dynamically links and mutually influences monitoring data, deployment plans, and construction conditions. Specifically, real-time monitoring data is used to evaluate the effectiveness of the current deployment plan. When monitoring data is abnormal or construction conditions change, the system automatically triggers an algorithm to re-optimize the deployment plan. The optimized plan then guides on-site adjustments, forming a closed-loop process of "monitoring-evaluation-optimization-adjustment-re-monitoring".

[0051] This includes establishing a linkage feedback mechanism between monitoring data and deployment plans, collecting real-time on-site monitoring data and changes in construction conditions, and dynamically adjusting the deployment plan by re-invoking the improved deep Q-network model when monitoring data or construction conditions meet preset trigger conditions. This includes: real-time collection of blasting vibration data and static deformation data; automatically triggering the improved deep Q-network model to densify monitoring points in the area when the vibration peak value in a certain area reaches more than 70% of the warning benchmark value, or the daily settlement reaches more than 0.5mm, reducing the spacing between monitoring points to 2-3m and adjusting the sensor sampling rate to 200Hz when a certain area... If the vibration peak value remains below 50% of the warning baseline for 7 consecutive days and the daily settlement is below 0.2 mm, and there are no significant changes in geological conditions and blasting conditions, the system automatically reduces redundant monitoring points, adjusts the spacing between monitoring points to 5-6 m, and reduces the sensor sampling rate to 100 Hz. The system stores the layout plan, monitoring data, and operating parameters after each dynamic adjustment into the database, and updates the parameters of the improved depth Q-network model every 15 days. When blasting parameters are adjusted, geological conditions change, or procedures are switched, the system automatically triggers the improved depth Q-network model to re-optimize the layout plan, achieving real-time adaptation to the construction conditions.

[0052] For example, monitoring points are deployed on-site according to the initial deployment plan described above. Specific deployment requirements are as follows: monitoring points in key areas (weak sections, connecting sections) of the old tunnel are fixed with expansion bolts; monitoring points in the new tunnel are pre-embedded; monitoring points in the pilot tunnel are temporarily fixed to facilitate adjustments to the excavation process.

[0053] Static and dynamic monitoring equipment are deployed simultaneously, with settlement gauges, convergence meters, and MEMS triaxial accelerometers installed at each monitoring point to ensure multi-parameter collaborative monitoring. The initial sampling rate of the dynamic monitoring sensors is set to 100-150Hz, adjusted to 200Hz during peak blasting periods (critical processes such as the excavation of the pilot tunnel and the widening of the main tunnel). The monitoring equipment is connected to a data acquisition terminal, which connects to the backend server via a dual-mode wired + wireless transmission module (short-distance wired, long-distance 5G + fiber optic), enabling data uploads in seconds.

[0054] Establish a linkage feedback mechanism between monitoring data and deployment plans, collect on-site monitoring data and changes in construction conditions in real time, and dynamically adjust the deployment plan by re-invoking the improved deep Q-network model when the monitoring data or construction conditions meet preset trigger conditions. Specifically, this includes: Real-time acquisition of blasting vibration data (peak value, dominant frequency, attenuation pattern) and static deformation data (settlement, convergence displacement).

[0055] The dynamic adjustment triggering conditions and actions include: when the vibration peak value in a certain area reaches more than 70% of the warning benchmark value, or the daily settlement reaches more than 0.5mm, the improved deep Q-network model is automatically triggered to densify the monitoring points in that area, reducing the spacing between monitoring points to 2-3m and adjusting the sensor sampling rate to 200Hz. For example, on August 10, 2025, during the blasting operation at Y2K232+955 in the right tunnel of the entrance (charge of 10.8kg), monitoring data showed that the vibration peak value at the measuring point Y1K233+160 in the left tunnel of the old tunnel reached 0.65cm / s (close to 70% of the warning benchmark value). The system automatically triggered the algorithm to densify one monitoring point in that area (chainage Y1K233+150), reducing the spacing to 2.5m and adjusting the sampling rate to 200Hz. When the peak vibration value in a certain area is below 50% of the warning benchmark for seven consecutive days, and the daily settlement is below 0.2 mm, with no significant changes in geological conditions and blasting conditions, the system automatically reduces redundant monitoring points, adjusts the spacing between monitoring points to 5-6 m, and lowers the sensor sampling rate to 100 Hz. For example, if seven consecutive days of monitoring data show that the peak vibration value in the area from Z2K232+890 to Z2K232+900 of the central tunnel is stable at 0.2-0.3 cm / s (<50% of the warning benchmark), and the geological conditions remain unchanged, the system automatically reduces one redundant monitoring point (station number Z2K232+895), adjusts the spacing to 5.5 m, and lowers the sampling rate to 100 Hz. When blasting parameters are adjusted, geological conditions change, or procedures are switched, the system automatically triggers an improved depth Q-network model to re-optimize the deployment scheme, achieving real-time adaptation to the construction conditions.

[0056] Each dynamically adjusted deployment plan, monitoring data, and operating parameters are stored in the database, and the parameters of the improved deep Q-network model are updated every 15 days. Combined with accumulated data from 32 blasting case studies, the optimized algorithm improves the accuracy of monitoring point deployment in key areas by 15%.

[0057] In another embodiment, such as Figure 2 As shown, the intelligent optimization method for the deployment of tunnel blasting monitoring points based on data fusion also includes the following steps: S201 employs a process of "simulation optimization - on-site calibration - iterative improvement" to calibrate the monitoring point deployment scheme. Pilot deployments are conducted in typical areas. By comparing the monitoring effects of the optimized scheme with the traditional scheme, the weights of the data fusion model and the parameters of the deep Q-network model are adjusted and improved.

[0058] For example, firstly, based on the geological parameters, structural parameters, blasting parameters and previous monitoring data of the Hainan Damao Tunnel, an initial optimized layout scheme (a total of 12 monitoring points, the specific locations of which are shown in step three) is generated through the aforementioned steps one to four (data acquisition, preprocessing, fusion mining, and optimization of the improved DQN model).

[0059] Then, pilot deployment and calibration were carried out in typical areas selected at the tunnel site. Typical areas included: weak sections of the old tunnel (such as chainage Y1K233+120~Y1K233+200); the connection section between the old and new tunnels (such as chainage Z2K232+870~Z2K232+874); and the section affected by the excavation of the central pilot tunnel (such as chainage Z2K232+873~Z2K232+886).

[0060] Monitoring points were deployed in the aforementioned areas according to both the optimized plan and the traditional experience-based plan (i.e., a plan based solely on the "closer density, farther sparser" principle), and monitoring data from 3 to 5 blasting operations were collected simultaneously. The comparison included: monitoring coverage: the coverage of key areas by the optimized plan (whether it achieved full coverage) and the overall coverage rate; data accuracy: vibration peak measurement error, dominant frequency deviation, proportion of effective data, etc.; cost indicators: number of monitoring points, equipment and maintenance costs.

[0061] Based on the comparison results, adjust the weights of the data fusion model. For example, if the field data shows that the influence of the surrounding rock grade on the vibration response is higher than the initial weight, then retrain the model using the random forest algorithm to update the influence weights of various data types (e.g., adjust the weight of the surrounding rock grade from 0.28 to 0.30); at the same time, adjust and improve the parameters of the DQN model, such as the weight coefficients in the reward function (the coefficients of R1, R2, and R3 can be fine-tuned) or the learning rate, exploration rate decay rate, etc.

[0062] After adjusting the parameters, rerun the improved DQN model to generate an optimized deployment plan, and conduct another field pilot verification. Repeat the above "calibration-optimization-verification" process until the monitoring effect meets the preset standards.

[0063] S202. Establish an evaluation index system for the effectiveness of monitoring points, and evaluate them from three dimensions: monitoring coverage, data accuracy, and cost control.

[0064] The effectiveness evaluation indicators are as follows: 100% coverage of key areas, over 95% coverage of the total monitoring range, dynamic vibration data measurement error not exceeding 5%, static deformation data measurement error not exceeding 3%, effective data volume not less than 90%, and the number of monitoring points reduced by more than 20% compared to traditional methods.

[0065] For example, an indicator system is established that comprehensively evaluates monitoring coverage, data accuracy, and cost control, with the following pre-set standards: Monitoring coverage: Key areas (weak sections of old tunnels, connecting sections between old and new tunnels, and sections affected by the excavation of the pilot tunnel) must achieve 100% coverage; the coverage of the total monitoring scope (covering the entire old tunnel, the entire new tunnel, and the entire pilot tunnel) shall not be less than 95%.

[0066] Data accuracy: Dynamic vibration data: vibration peak measurement error not exceeding 5%, main frequency measurement error not exceeding 10%; Static deformation data: settlement and convergence displacement measurement error not exceeding 3%; Valid data volume: in continuous 30-day monitoring, the proportion of valid data (no packet loss, no abnormal noise) is not less than 90%.

[0067] Cost control: The number of monitoring points is reduced by more than 20% compared to traditional experience-based solutions; the construction and maintenance costs of the monitoring system (including equipment, installation, labor, data transmission, etc.) are reduced by more than 20%.

[0068] In the Hainan Damao Tunnel project, through the aforementioned calibration process, the final optimized scheme achieved a monitoring coverage rate of 98% (100% in key areas), with dynamic vibration data measurement error ≤4.2%, static deformation data measurement error ≤2.5%, and effective data volume of 92%. The number of monitoring points was reduced from 16 in the traditional scheme to 12 (a 25% reduction), and monitoring costs were reduced by 25%, fully meeting the preset standards. The data accuracy rate was improved to 92% compared to the traditional experience-based scheme (approximately 75%), significantly enhancing the structural safety management capabilities during blasting construction.

[0069] S203 If the evaluation results do not meet the preset standards, return to adjust the data fusion model or improve the parameters of the deep Q network model, and re-optimize the deployment scheme until the evaluation standards are met, and then determine the final monitoring point deployment scheme.

[0070] For example, if the evaluation results do not meet the preset standards (e.g., insufficient monitoring coverage or failure to meet cost reduction targets), the weights of the data fusion model or the parameters of the DQN model are adjusted (e.g., adjusting the weight coefficient of the cost control item in the reward function), the data points are re-optimized, and on-site verification is conducted again until all indicators meet the requirements, at which point the final monitoring point layout plan is determined. This plan can be applied to the blasting vibration monitoring of long arch tunnels (both new and old) throughout their entire life cycle (construction and operation).

[0071] In another embodiment, such as Figure 3 As shown, Figure 3 A schematic diagram of the structure of a detection system is provided. The tunnel blasting monitoring system of the present invention mainly includes a blast source, a blasting site, sensors, a signal conditioning unit, a data acquisition unit, a SOC unit, a data storage unit, a wireless communication unit, a GPS timing unit, a USB interface, and a built-in rechargeable lithium battery. The units work together to achieve high-precision acquisition and transmission of blasting vibration data.

[0072] The blast source is the location of the blasting operation during tunnel construction, used to generate blasting vibration signals, and is the monitoring object of the entire monitoring system. The blasting site refers to the actual engineering environment where the blast source is located, including the tunnel structure, surrounding rock, and surrounding structures, and is used to transmit vibration signals generated by the blasting. The sensor, set at a predetermined monitoring point at the blasting site, is used to collect vibration signals generated by the blasting. It is preferably a triaxial accelerometer or MEMS sensor to obtain vibration data in three directions. The signal conditioning unit, electrically connected to the sensor, is used to amplify, filter, and perform anti-interference processing on the raw analog signal output by the sensor in order to improve the signal quality and match the input requirements of the subsequent acquisition unit. The data acquisition unit, connected to the signal conditioning unit, is used to perform analog-to-digital conversion on the conditioned signal, converting the analog signal into a digital signal, and to perform preliminary data processing. The SOC unit, connected to the data acquisition unit, is used to further process the acquired data, including digital filtering, feature extraction and peak detection, and to control and schedule each functional unit. The data storage unit, connected to the SOC unit, is used to store the collected raw data and processed data locally to prevent data loss and support subsequent data analysis. The wireless communication unit, connected to the SOC unit, is used to transmit monitoring data to the remote monitoring center wirelessly. LoRa, 4G or 5G communication methods are preferred to achieve long-distance low-latency transmission. The GPS timing unit, connected to the SOC unit, is used to provide the system with a high-precision time synchronization signal, so that the data of each monitoring node has a unified time reference, which facilitates the comparison and analysis of multi-point data. The USB interface connects to the SOC unit and is used to enable local data export, parameter configuration, and program upgrades for the device. The built-in rechargeable lithium battery provides power to the above units, ensuring continuous system operation when external power is unavailable, thus improving system reliability and adaptability.

[0073] In this invention, each component unit forms a collaborative working mechanism according to data flow and control logic, as detailed below: The blast source generates vibration signals during the blasting operation, which propagate outward through the medium at the blasting site. Sensors placed at different locations at the blasting site sense the vibration signals propagating to each monitoring point in real time and output corresponding analog electrical signals.

[0074] The analog signal output by the sensor is first input to the signal conditioning unit. After amplification, filtering and anti-interference processing, a stable signal that meets the sampling requirements is obtained. The conditioned signal is then input to the data acquisition unit, which performs analog-to-digital conversion to convert the analog signal into a digital signal.

[0075] The converted digital signal is transmitted to the SOC unit, which performs unified scheduling and processing of the data, including digital filtering, feature extraction and vibration peak identification, and determines abnormal vibrations according to preset rules.

[0076] During data processing, the GPS timing unit provides a unified high-precision time reference for the SOC unit, enabling data from different monitoring nodes to be synchronized in time, thereby achieving alignment analysis and propagation pattern identification of multi-point data.

[0077] The data processed by the SOC unit is written to the data storage unit for local caching and long-term storage to prevent data loss; on the other hand, it is sent to the remote monitoring center through the wireless communication unit to realize remote real-time monitoring and centralized management.

[0078] Meanwhile, the USB interface connects to the SOC unit for on-site debugging, parameter configuration, data export, and system upgrades, enhancing system maintenance and expansion capabilities.

[0079] The built-in rechargeable lithium battery provides power to the above units, maintaining the continuous operation of the system when the external power supply is interrupted, and ensuring the continuity and integrity of the monitoring data.

[0080] Through the above collaborative work, a complete closed-loop monitoring process from "vibration generation - signal acquisition - data processing - data transmission - data storage" has been realized.

[0081] Based on the same inventive concept, this application also provides an intelligent optimization device for the deployment of tunnel blasting monitoring points based on data fusion, used to implement the aforementioned intelligent optimization method for the deployment of tunnel blasting monitoring points based on data fusion. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of the intelligent optimization device for the deployment of tunnel blasting monitoring points based on data fusion provided below can be found in the limitations of the intelligent optimization method for the deployment of tunnel blasting monitoring points based on data fusion described above, and will not be repeated here.

[0082] In one embodiment, such as Figure 4 As shown, an intelligent optimization device for the deployment of tunnel blasting monitoring points based on data fusion is provided. The device includes: Data acquisition module 30 is used to collect geological parameters, structural parameters, blasting parameters and previous monitoring data to form a multi-source heterogeneous dataset; The data processing module 31 is used to preprocess multi-source heterogeneous datasets, and to fuse and mine the preprocessed data to identify key monitoring areas. The model building module 32 is used to build an improved deep Q-network model based on the fused data and key monitoring areas, with the optimization objectives of monitoring coverage, monitoring accuracy and monitoring cost, and output the initial layout scheme of monitoring points through iterative optimization. The scheme adjustment module 33 is used to set up monitoring points on site according to the initial deployment scheme and establish a linkage feedback mechanism between monitoring data and deployment scheme. It collects on-site monitoring data and changes in construction conditions in real time. When the monitoring data or construction conditions meet the preset trigger conditions, it calls the improved deep Q network model again to dynamically adjust the deployment scheme.

[0083] This application also provides an electronic device, in some embodiments, referring to... Figure 5 As shown, the electronic device 700 includes an input unit 710, a memory 720, a processor 730, and an output unit 740. The memory 720 stores program instructions that can be executed on the processor 730. The processor 730 can execute the intelligent optimization method and / or technical solution for the deployment of tunnel blasting monitoring points based on data fusion, as described in the foregoing embodiments, by calling the program instructions. The electronic device 700 can be a mobile terminal device such as a mobile phone or computer.

[0084] Furthermore, embodiments of this application also provide a computer-readable storage medium for storing a computer program that executes an intelligent optimization method for the deployment of tunnel blasting monitoring points based on data fusion. For example, computer program instructions, when executed by a computer, can invoke or provide the methods and / or technical solutions according to this application through the operation of the computer. The program instructions that invoke the methods of this application may be stored in a fixed or removable storage medium, and / or transmitted via data streams in broadcast or other signal carrying media, and / or stored in a storage medium that operates according to the program instructions.

[0085] Obviously, those skilled in the art should understand that the modules or steps of this application described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a storage device for execution by a computing device, or fabricating them separately as individual integrated circuit modules, or fabricating multiple modules or steps as a single integrated circuit module. Thus, this application is not limited to any particular combination of hardware and software.

[0086] The technical features of the above embodiments can be arbitrarily integrated. For the sake of brevity, not all possible integrations of the technical features in the above embodiments are described. However, as long as the integration of these technical features does not contradict each other, they should be considered to be within the scope of this specification.

[0087] The above embodiments merely illustrate several implementation methods of the present invention, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention patent should be determined by the appended claims.

Claims

1. An intelligent optimization method for the deployment of tunnel blasting monitoring points based on data fusion, characterized in that, The method includes: Geological parameters, structural parameters, blasting parameters, and previous monitoring data were collected to form a multi-source heterogeneous dataset; The multi-source heterogeneous dataset is preprocessed, and the preprocessed data is fused and feature mined to identify key monitoring areas. This includes: constructing a multi-dimensional data fusion model of "geology-structure-blasting-monitoring"; using the random forest algorithm to calculate the influence weight of various data on the effectiveness of monitoring points; and weighting and fusing the multi-source data according to the weights; using the Pearson correlation analysis method to quantify the correlation coefficient between the blasting charge of the new tunnel and the vibration response of the old tunnel; identifying key monitoring areas based on the fused data values ​​and the correlation coefficients, including weak sections of the old tunnel, the connection section between the new and old tunnels, and the section affected by the excavation of the pilot tunnel; and identifying the areas composed of grid cells with fused data values ​​≥ 0.7 as key monitoring areas. Based on the fused data and the key monitoring areas, an improved deep Q-network model is constructed with monitoring coverage, monitoring accuracy, and monitoring cost as optimization objectives. The initial deployment scheme of the output monitoring points is iteratively optimized, including: dividing the monitoring area into 1m×1m grid cells, where the state of each grid cell is composed of its corresponding fused data features; defining an action space, which includes deploying monitoring points, densifying monitoring points, removing monitoring points, and adjusting monitoring point positions; and constructing a comprehensive reward function R = 0.4×R1 + 0.4×R2 - 0.2×R3, where R1 is the monitoring coverage reward, R2 is the data quality reward, and R3 is the cost control reward; set the algorithm parameters of the improved deep Q network model, including learning rate of 0.001, discount factor of 0.9, experience replay buffer size of 10000, target network update frequency of 100 steps, and exploration rate linearly decaying from 0.9 to 0.1; starting with the initial scheme of "near dense and far sparse" deployment, take the fused data as input, and iteratively optimize to make the reward function converge, outputting the optimal monitoring point deployment scheme; According to the initial deployment plan, monitoring points are set up on site, and a linkage feedback mechanism between monitoring data and deployment plan is established. On-site monitoring data and changes in construction conditions are collected in real time. When the monitoring data or construction conditions meet the preset trigger conditions, the improved deep Q network model is called again to dynamically adjust the deployment plan.

2. The intelligent optimization method for the deployment of tunnel blasting monitoring points based on data fusion as described in claim 1, characterized in that, In the steps of collecting multi-source heterogeneous data: The collected geological parameters include the surrounding rock grade, rock strength, distribution of joints and fissures, and groundwater conditions. The collected structural parameters include the thickness of the old and new tunnel linings, lining stiffness, lining aging coefficient, and connection method. The collected blasting parameters include the amount of explosives, the depth of the blast holes, the spacing between the blast holes, the detonation sequence, and the number of detonators. The collected preliminary monitoring data includes peak blasting vibration, dominant frequency, attenuation pattern, settlement amount, and convergence displacement.

3. The intelligent optimization method for the deployment of tunnel blasting monitoring points based on data fusion as described in claim 1, characterized in that, Preprocessing of the multi-source heterogeneous dataset includes: Outliers were removed from geological and structural parameters using the 3σ principle and then normalized. Wavelet transform algorithm is used to filter and denoise the blasting parameters and monitoring data, and the main frequency is extracted by fast Fourier transform. The preprocessed data is formatted uniformly and associated with grid cell coordinates.

4. The intelligent optimization method for the deployment of tunnel blasting monitoring points based on data fusion as described in claim 1, characterized in that, A linkage feedback mechanism between monitoring data and deployment plan is established to collect on-site monitoring data and changes in construction conditions in real time. When the monitoring data or construction conditions meet preset trigger conditions, the improved deep Q-network model is invoked again to dynamically adjust the deployment plan, including: Real-time acquisition of blasting vibration data and static deformation data; When the vibration peak value in a certain area reaches more than 70% of the warning benchmark value, or the daily settlement reaches more than 0.5mm, the improved deep Q network model is automatically triggered to densify the monitoring points in the area, reduce the spacing between monitoring points to 2-3m, and adjust the sensor sampling rate to 200Hz. When the peak vibration value of a certain area is lower than 50% of the warning benchmark value for 7 consecutive days and the daily settlement is lower than 0.2mm, and there is no significant change in geological conditions and blasting conditions, the redundant monitoring points are automatically reduced, the spacing between monitoring points is adjusted to 5-6m, and the sensor sampling rate is reduced to 100Hz. The deployment plan, monitoring data and operating parameters after each dynamic adjustment are stored in the database, and the parameters of the improved deep Q network model are updated every 15 days. When blasting parameters are adjusted, geological conditions change, or procedures are switched, the system automatically triggers the improved depth Q network model to re-optimize the layout scheme, achieving real-time adaptation to the construction conditions.

5. The intelligent optimization method for the deployment of tunnel blasting monitoring points based on data fusion as described in claim 1, characterized in that, The method further includes: The monitoring point deployment scheme was calibrated using a process of "simulation optimization - on-site calibration - iterative improvement". Pilot deployment was carried out in typical areas. By comparing the monitoring effects of the optimized scheme and the traditional scheme, the weights of the data fusion model and the parameters of the deep Q-network model were adjusted and improved. Establish an evaluation index system for the effectiveness of monitoring points, and evaluate them from three dimensions: monitoring coverage, data accuracy, and cost control. The effectiveness evaluation indexes are: 100% coverage of key areas, 95% or more coverage of the total monitoring range, dynamic vibration data measurement error not exceeding 5%, static deformation data measurement error not exceeding 3%, effective data volume not less than 90%, and the number of monitoring points reduced by more than 20% compared with traditional schemes. If the evaluation results do not meet the preset standards, the data fusion model or the parameters of the deep Q-network model will be adjusted and the deployment plan will be re-optimized until the evaluation standards are met, and then the final monitoring point deployment plan will be determined.

6. An intelligent optimization device for the deployment of tunnel blasting monitoring points based on data fusion, characterized in that, The device includes: The data acquisition module is used to collect geological parameters, structural parameters, blasting parameters and previous monitoring data to form a multi-source heterogeneous dataset; The data processing module is used to preprocess the multi-source heterogeneous dataset, fuse the preprocessed data, and perform feature mining to identify key monitoring areas. This includes: constructing a multi-dimensional data fusion model of "geology-structure-blasting-monitoring"; using a random forest algorithm to calculate the influence weights of various data types on the effectiveness of monitoring points; and weighting and fusing the multi-source data according to these weights; using Pearson correlation analysis to quantify the correlation coefficient between the blasting charge amount of the new tunnel and the vibration response of the old tunnel; identifying key monitoring areas based on the fused data values ​​and the correlation coefficients, including weak sections of the old tunnel, the connection section between the new and old tunnels, and the section affected by the excavation of the pilot tunnel; and identifying areas composed of grid cells with fused data values ​​≥ 0.7 as key monitoring areas. The model building module is used to construct an improved deep Q-network model based on the fused data and the key monitoring areas, with the optimization objectives of monitoring coverage, monitoring accuracy, and monitoring cost. It iteratively optimizes the initial deployment scheme of the output monitoring points, including: dividing the monitoring area into 1m×1m grid cells, where the state of each grid cell is composed of its corresponding fused data features; defining an action space, which includes deploying monitoring points, densifying monitoring points, removing monitoring points, and adjusting monitoring point positions; and constructing a comprehensive reward function R = 0.4×R1 + 0.4×R2 - 0.2×R3, where R1 is the monitoring coverage reward, R2 is the data quality reward, and R3 is the cost control reward; set the algorithm parameters of the improved deep Q network model, including learning rate of 0.001, discount factor of 0.9, experience replay buffer size of 10000, target network update frequency of 100 steps, and exploration rate linearly decaying from 0.9 to 0.1; starting with the initial scheme of "near dense and far sparse" deployment, take the fused data as input, and iteratively optimize to make the reward function converge, outputting the optimal monitoring point deployment scheme; The scheme adjustment module is used to deploy monitoring points on site according to the initial deployment scheme, and establish a linkage feedback mechanism between monitoring data and deployment scheme. It collects on-site monitoring data and changes in construction conditions in real time. When the monitoring data or construction conditions meet the preset trigger conditions, it calls the improved deep Q network model again to dynamically adjust the deployment scheme.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the intelligent optimization method for the deployment of tunnel blasting monitoring points based on data fusion, as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the intelligent optimization method for the deployment of tunnel blasting monitoring points based on data fusion, as described in any one of claims 1 to 5.

Images