A remote centralized control method and system for unmanned intelligent equipment in drill-and-blast tunnels

Abstract

This invention relates to the field of tunnel blasting technology, specifically to a remote centralized control method and system for intelligent equipment used in drill-and-blast tunnel construction without manned sites. The method includes: generating a multimodal fusion feature vector by fusing multi-source heterogeneous geological data; obtaining a geological interface sequence containing confidence information by dynamically clustering similar geological feature regions; analyzing the multimodal fusion feature vector, the geological interface sequence, and network resource allocation strategies to obtain a lightweight geological situation information package; dynamically generating a blasting scheme adapted to the current geological conditions based on the surrounding rock grade and geological interface characteristics; combining the multi-equipment collaborative operation sequence and path plan with the blasting scheme to generate a final operation instruction set; and using the final operation instruction set, a comprehensive blasting quality assessment report, and the lightweight geological situation information package to achieve remote automated centralized control of the intelligent equipment. This invention is applicable to the collaborative operation and remote precise control of multiple intelligent equipment in tunnel construction.

Description

Technical Field

[0001] This invention relates to the field of tunnel blasting technology, specifically to a remote centralized control method and system for intelligent equipment for unmanned construction in drill-and-blast tunnels. Background Technology

[0002] Drill-and-blast method is one of the core methods of tunnel construction, but traditional drill-and-blast tunnel construction relies heavily on manual operation and has many technical bottlenecks: First, geological identification depends on a single data source (such as drilling data), and multi-source geological data (drilling, geophysical exploration, real-time monitoring data, etc.) lack effective integration, resulting in low accuracy in identifying the surrounding rock grade and an inability to match geological changes in real time; Second, geological interface identification mostly adopts static analysis methods, without real-time and confidence assessment, making it difficult to accurately locate areas of geological abrupt change and easily leading to mismatches between blasting parameters and geology; Third, blasting parameter design relies on manual experience and has not established a system for matching geological characteristics and parameters. The dynamic mapping relationship leads to unstable blasting effects (such as over- or under-excavation, uneven block size); fourth, when multiple equipment work together, path planning is mostly static planning for a single piece of equipment, lacking a dynamic conflict resolution mechanism, which easily leads to equipment collisions and low work efficiency; fifth, the large amount of data transmitted remotely and the lack of business priority distinction make key control data (such as equipment instructions) susceptible to interference, and service quality cannot be guaranteed; sixth, the lack of simulation verification and safety interlocking mechanisms before issuing work instructions poses safety hazards; seventh, the lack of a closed-loop optimization mechanism makes it impossible to continuously improve the strategies of each link based on the feedback of blasting effects, making it difficult to improve long-term construction accuracy.

[0003] While some existing technologies attempt to solve single problems (such as deep learning-based geological identification and single-equipment path planning), they have not formed a complete technical system encompassing "geological perception - parameter generation - equipment collaboration - remote transmission - command verification - closed-loop optimization." For example, multi-source data fusion only employs simple splicing or weighted summation, failing to consider the spatiotemporal correlation and modal differences of the data; multi-agent path planning does not incorporate tunnel operation scenarios (such as blasting area restrictions); remote transmission does not utilize network slicing technology to ensure critical business operations; and each stage is independent, lacking data feedback and strategy iteration mechanisms. Therefore, a comprehensive, hierarchical decision-making method for remote centralized control of intelligent equipment is urgently needed to overcome existing technological bottlenecks. Summary of the Invention

[0004] To address the shortcomings of existing methods and the needs of practical applications, and to solve the aforementioned problems, this invention provides a remote centralized control method for intelligent equipment used in drill-and-blast tunnels without manned ground. The method includes the following steps: generating a multimodal fusion feature vector by fusing multi-source heterogeneous geological data; obtaining a geological interface sequence containing confidence information by dynamically clustering similar geological feature regions; analyzing the multimodal fusion feature vector, the geological interface sequence, and network resource configuration strategies to obtain a lightweight geological situation information package; dynamically generating a blasting scheme adapted to the current geological conditions based on the surrounding rock grade and geological interface characteristics; combining the multi-equipment collaborative operation sequence and path plan with the blasting scheme to generate a final operation instruction set; and using the final operation instruction set, a comprehensive blasting quality assessment report, and the lightweight geological situation information package to achieve remote automated centralized control of the intelligent equipment.

[0005] This invention improves the accuracy of surrounding rock grade discrimination by fusing multi-source heterogeneous geological data to generate multimodal fusion feature vectors; dynamically clusters similar regions and obtains geological interface sequences with confidence levels to avoid misjudgments of geological abrupt changes; analyzes the above features, interface sequences, and network strategies to generate lightweight geological situation information packages, solving the problems of large data volume and poor interactivity in remote transmission; dynamically generates adaptive blasting schemes by combining surrounding rock grade and geological interfaces, and generates the final operation instruction set by combining multi-equipment collaborative timing and path planning to avoid equipment conflicts and parameter mismatch risks; finally, remote centralized control is achieved based on the instruction set, blasting assessment report, and lightweight package, effectively breaking through the bottlenecks of traditional construction such as strong reliance on manual labor, poor geological adaptability, and low transmission efficiency, significantly improving construction safety, efficiency, and accuracy, and providing key technical support for the unmanned landing of drill-and-blast tunnels.

[0006] Optionally, the step of generating a multimodal fusion feature vector by fusing multi-source heterogeneous geological data includes the following steps:

[0007] Acquire multi-source heterogeneous geological data and systematically clean the data; use deep learning and signal processing techniques to transform the heterogeneous data and obtain structured features; use an attention mechanism to fuse multimodal data and generate multimodal fusion feature vectors based on random forest.

[0008] By systematically cleaning multi-source heterogeneous geological data, outliers can be removed and missing values ​​filled in, ensuring data quality and laying a reliable foundation for subsequent processing. Deep learning and signal processing techniques are used to transform heterogeneous data into structured features, breaking down data modality barriers and resolving issues of inconsistent data formats. Attention mechanisms are employed to fuse multimodal data, and feature vectors are generated based on random forests, highlighting the weights of key data and achieving deep data fusion. This overall process significantly improves the accuracy of surrounding rock grade determination, providing precise data support for subsequent geological interface identification and automatic generation of blasting parameters, and is a crucial data guarantee for achieving safe and efficient tunnel construction without artificial sites.

[0009] Optionally, the step of dividing geologically similar regions based on dynamic clustering to obtain a geological interface sequence containing confidence information includes the following steps:

[0010] Data is collected via a sliding window; based on a density clustering algorithm, regions with similar geological features are dynamically clustered; for cluster centers at adjacent locations in the clustering results, Euclidean distance is used to calculate inter-cluster distances and construct an inter-cluster distance sequence; the confidence level of geological interfaces is calculated by combining cluster purity and the significance of abrupt changes, resulting in a geological interface sequence containing confidence information. Real-time data collection via a sliding window ensures the timeliness of geological information; density clustering dynamically divides similar regions to adapt to the irregular distribution characteristics of geological data and improve the accuracy of region division; the inter-cluster distance sequence constructed using Euclidean distance clearly locates the geological interfaces; the confidence level is calculated by combining cluster purity (ensuring the reliability of region division) and the significance of abrupt changes (confirming the authenticity of interface abrupt changes), solving the problem of unreliable assessment in traditional geological interface identification and avoiding misjudgments. Overall, real-time and accurate identification and reliable confidence assessment of geological interfaces are achieved, providing accurate geological basis for the adaptive generation of subsequent blasting parameters, ensuring the adaptation of blasting parameters to geological conditions, and supporting the safety and accuracy of unmanned construction.

[0011] Optionally, the step of analyzing the multimodal fusion feature vector, the geological interface sequence, and the network resource allocation strategy to obtain a lightweight geological situation information package includes the following steps:

[0012] By fusing key geological information through the multimodal fusion feature vector and the geological interface sequence, and performing dimensionality reduction processing, a lightweight geological situation information package is obtained by applying a curvature-based simplification algorithm for lightweight processing and adaptive compression according to the network resource allocation strategy. This invention significantly reduces the volume of geological data, adapts to network resource allocation requirements, ensures the efficiency and stability of remote transmission, and avoids data redundancy that consumes bandwidth. Simultaneously, it provides efficient and usable geological data for remote control centers, supporting real-time visualization and interactive construction decisions, and laying the data foundation for unmanned remote and precise control of drill-and-blast tunnels.

[0013] Optionally, obtaining the network resource configuration policy includes the following steps:

[0014] Service classification and slicing are performed based on service priority and performance sensitivity; an intelligent resource allocation model is constructed to dynamically allocate network resources; transmission protocols and processing strategies are determined according to service characteristics to generate network resource configuration strategies. Service classification and slicing achieve logical isolation of services with different priorities (such as control commands and evaluation data), ensuring priority for critical service resources; the intelligent model dynamically allocates resources, scheduling bandwidth and latency as needed to avoid resource idleness and overload; transmission protocols and processing strategies are determined according to service characteristics to improve data transmission adaptability. Overall, this ensures low-latency and high-reliability transmission of critical services, improves network resource utilization, lays a solid network foundation for stable data interaction in remote centralized control, and supports efficient, unmanned collaborative operations.

[0015] Optionally, the step of dynamically generating a blasting scheme adapted to the current geological conditions based on the surrounding rock grade and geological interface characteristics includes the following steps:

[0016] A geological-blasting parameter mapping library is established, and an initial blasting scheme is obtained based on this library. The initial blasting scheme is then adjusted according to the adaptability of geological interfaces to generate a blasting scheme adapted to the current geological conditions. By establishing the geological-blasting parameter mapping library, the limitations of traditional reliance on manual experience are overcome, providing data-supported initial blasting schemes for different surrounding rock grades and ensuring basic adaptability. Furthermore, by adjusting the initial scheme in conjunction with geological interface characteristics, it can specifically adapt to geological differences at geological interfaces, solving the problem that traditional fixed blasting parameters cannot cope with geological changes. This allows blasting schemes to accurately fit current geological conditions, reducing over-excavation, under-excavation, uneven block size, and other problems, lowering safety hazards, improving blasting effects and construction safety, while reducing manual adjustment costs, increasing construction efficiency, providing a reasonable blasting basis for subsequent multi-equipment collaborative operations, and facilitating precise unmanned construction.

[0017] Optionally, the step of combining the multi-equipment collaborative operation timing and path plan with the blasting scheme to generate the final operation instruction set includes the following steps:

[0018] A high-precision 3D digital twin model of the tunnel is constructed. Based on this model, safety rules are matched, and a final set of operational instructions is generated by combining the multi-equipment collaborative operation sequence and path plan with the blasting scheme. The high-precision 3D digital twin model realistically recreates the tunnel's geology, equipment, and operational area, providing an accurate virtual environment for subsequent safety verification and avoiding verification deviations caused by scene distortion. Safety rule matching based on the model incorporates the multi-equipment collaborative sequence, path, and blasting scheme into a unified safety verification framework, proactively identifying potential hazards such as equipment collisions and unauthorized entry into blasting areas, ensuring that instructions comply with safety regulations. The final generated set of operational instructions combines collaboration and safety, can be directly issued for execution, reduces manual intervention, and ensures the accurate and safe conduct of equipment collaborative operations and blasting operations during unmanned construction.

[0019] Optionally, the step of utilizing the final operation instruction set, the comprehensive blasting quality assessment report, and the lightweight geological situation information package to achieve remote automated centralized control of intelligent equipment includes the following steps:

[0020] A multi-dimensional data association system was constructed, establishing a dynamic mapping model of "operation instructions - geological conditions - blasting effects." The mapping model parameters were iteratively optimized through data-driven approaches, and updated and validated using an incremental learning strategy, enabling remote automated centralized control of intelligent equipment. By constructing this multi-dimensional data association system and dynamic mapping model, the gap in direct dynamic correlation between the three elements was filled, providing precise data logic support for remote centralized control. Incremental learning was used to iteratively optimize and validate model parameters, allowing for real-time adaptation to geological changes and avoiding control deviations caused by outdated models. Ultimately, this significantly improved the accuracy, adaptability, and reliability of remote centralized control of intelligent equipment, ensuring long-term safe, efficient, and precise construction on unmanned sites.

[0021] Optionally, analyzing the comprehensive blasting quality assessment report includes the following steps:

[0022] Based on digital twins and computer vision, a 3D model of the actual cross-section after blasting is generated. Based on this 3D model, contour deviations and debris particle size are analyzed to obtain a comprehensive blasting quality assessment report. This invention solves the problems of high subjectivity, low accuracy, and poor efficiency in manual assessment. It objectively evaluates blasting quality using 3D models and quantitative data, providing a reliable basis for subsequent optimization of blasting parameters and correction of work instructions, ensuring tunnel excavation accuracy and construction efficiency, and facilitating safe, unmanned construction.

[0023] Secondly, in order to efficiently execute the remote centralized control method for unmanned intelligent equipment in drill-and-blast tunnels provided by this invention, this invention also provides a remote centralized control system for unmanned intelligent equipment in drill-and-blast tunnels, comprising:

[0024] The multi-source heterogeneous geological data fusion module is used to fuse multi-source heterogeneous geological data and then extract multi-modal fusion feature vectors that are strongly correlated with the surrounding rock grade.

[0025] The dynamic clustering geological interface identification module is used to divide regions with similar geological features through dynamic clustering, and to locate the interfaces of different lithologies by combining abrupt change point detection, so as to realize real-time, confidence-based identification of geological interfaces and obtain a geological interface sequence containing confidence information.

[0026] The blasting parameter scheme generation module is used to dynamically generate blasting parameter schemes that are adapted to the current geological conditions based on the surrounding rock grade and geological interface characteristics.

[0027] The dynamic conflict resolution and path planning module is used to generate globally optimal multi-equipment collaborative operation timing and path plans;

[0028] The intelligent evaluation module for blasting effects is used to quantify indicators such as the outline and block size after blasting by combining digital twin models with computer vision, and generate a comprehensive evaluation report on blasting quality.

[0029] The network resource configuration module is used to dynamically allocate network resources according to the priority of services within the tunnel and generate network resource configuration strategies.

[0030] The first data processing module is used to integrate key geological information and process it in a lightweight manner to generate a lightweight geological situation information package, enabling the remote center to accurately perceive the geological condition.

[0031] The second data processing module is used to ensure the safety and feasibility of blasting design and equipment operation plans through digital twin simulation and safety rule verification, and to generate a set of final operation instructions that can be directly executed.

[0032] The autonomous learning and closed-loop optimization module is used to achieve remote automated centralized control of intelligent equipment based on the final operation instruction set, blasting quality comprehensive assessment report, and lightweight geological situation information package.

[0033] The present invention provides a remote centralized control system for unmanned intelligent equipment in drill-and-blast tunnels. This system has a compact structure and stable performance, and can reliably execute the remote centralized control method for unmanned intelligent equipment in drill-and-blast tunnels provided by the present invention, further enhancing the overall applicability and practical application capability of the present invention. Attached Figure Description

[0034] Figure 1 A flowchart illustrating a remote centralized control method for unmanned intelligent equipment in drill-and-blast tunnels, provided by an embodiment of the present invention.

[0035] Figure 2 This is a framework diagram of a remote centralized control system for unmanned intelligent equipment in a drill-and-blast tunnel, provided as an embodiment of the present invention. Detailed Implementation

[0036] Specific embodiments of the present invention will now be described in detail. It should be noted that the embodiments described herein are for illustrative purposes only and are not intended to limit the invention. In the following description, numerous specific details are set forth in order to provide a thorough understanding of the invention. However, it will be apparent to those skilled in the art that these specific details are not necessary to practice the invention. In other instances, well-known circuits, software, or methods have not been specifically described to avoid obscuring the invention.

[0037] Throughout this specification, references to "an embodiment," "an embodiment," "an example," or "an example" mean that a particular feature, structure, or characteristic described in connection with that embodiment or example is included in at least one embodiment of the invention. Therefore, the phrases "in an embodiment," "in an embodiment," "an example," or "an example" appearing in various places throughout the specification do not necessarily refer to the same embodiment or example. Furthermore, specific features, structures, or characteristics can be combined in one or more embodiments or examples in any suitable combination and / or sub-combination. Moreover, those skilled in the art will understand that the illustrations provided herein are for illustrative purposes and are not necessarily drawn to scale.

[0038] Please see Figure 1 To address the aforementioned problems, this invention provides a remote centralized control method for unmanned intelligent equipment used in drill-and-blast tunnel construction, such as... Figure 1 As shown, in one embodiment, the method includes the following steps:

[0039] S1. Generate a multimodal fusion feature vector by fusing multi-source heterogeneous geological data.

[0040] In this embodiment, the step of generating a multimodal fusion feature vector by fusing multi-source heterogeneous geological data includes the following steps:

[0041] S11. Acquire multi-source heterogeneous geological data and perform systematic data cleaning.

[0042] Multi-source heterogeneous geological data includes geological drilling data (core compressive strength, joint density), geophysical data (seismic wave velocity, electromagnetic reflection signal), real-time monitoring data (surrounding rock deformation, stress value), and image data (rock texture image of the tunnel face).

[0043] The raw data underwent systematic cleaning, and outliers such as sudden noise in seismic wave velocities were removed using statistical methods (such as the 3σ principle) to avoid data bias interfering with subsequent analysis. A minimum-maximum standardization method was employed to map data of different dimensions to the [0,1] interval, eliminating the influence of dimensions. Based on GPS timestamps, a dynamic time warping algorithm was used to achieve spatiotemporal alignment of drilling and geophysical data from the same location, ensuring temporal consistency and spatial correspondence of multi-source data.

[0044] S12. Utilize deep learning and signal processing techniques to transform heterogeneous data and obtain structured features.

[0045] By leveraging deep learning and signal processing techniques, unstructured data is transformed into structured features. For tunnel face images, a ResNet-50-based convolutional neural network model is constructed. Through multi-layer convolution and pooling operations, texture feature vectors such as edge density and gray-level mean are extracted, providing image-level quantitative information for subsequent analysis. For seismic wave signals, Discrete Wavelet Transform (DWT) is used to convert the time-domain signal into a time-frequency domain representation, thereby extracting key time-frequency features such as dominant frequency and energy spectrum distribution, capturing the signal's variation patterns at different times and frequencies.

[0046] S13. Employ an attention mechanism to fuse multimodal data and generate multimodal fusion feature vectors based on random forest.

[0047] An attention mechanism is employed to achieve deep fusion of structured features (drilling / monitoring data) and unstructured transformed features (image / signal features), satisfying the following requirements:

[0048]

[0049] in, Indicates multimodal fusion features, Indicates the number of data sources. This represents the attention weights, which are learned and optimized during training using the backpropagation algorithm. Based on the requirements for determining the surrounding rock grade, the weights of different features are automatically adjusted. For example, in soft rock environments, features such as joint density are given higher weights to highlight their importance in assessing the surrounding rock grade. This represents the feature vector of the i-th type of data source, which includes data such as rock density and porosity obtained from drilling, as well as features obtained from image and signal conversion.

[0050] Furthermore, the importance score of each feature is calculated based on the random forest algorithm. By calculating the reduction in node impurity after feature splitting, the contribution of each feature to the model prediction is quantified. The top 80% of core features with the highest contribution are dynamically retained. A recursive feature elimination (RFE) strategy is adopted to gradually eliminate redundant features that are irrelevant to the surrounding rock grade, such as local magnetic field interference signals. Finally, a concise and effective multimodal fusion feature vector is generated, improving the model training efficiency and prediction accuracy.

[0051] By fusing multi-source data such as seismic wave velocity, rock mass integrity index, rock compressive strength, and groundwater occurrence status, a feature database for classifying surrounding rock is constructed. Deep learning algorithms are used to denoise and enhance the data, effectively improving the signal-to-noise ratio and feature correlation. Simultaneously, a multi-parameter cross-validation mechanism is established to avoid misjudgments caused by single data points. For example, when seismic wave velocity data shows anomalies, comprehensive analysis combining auxiliary indicators such as rock mass structural surface spacing and groundwater flow can effectively identify wave velocity anomalies caused by fracture zones or karst cave development, ensuring the accuracy and reliability of surrounding rock classification.

[0052] S2. Based on dynamic clustering, divide regions with similar geological features to obtain a geological interface sequence containing confidence information.

[0053] In this embodiment, the step of dividing geologically similar regions based on dynamic clustering to obtain a geological interface sequence containing confidence information includes the following steps:

[0054] S21. Collect sliding window data.

[0055] Using the tunnel face advancement direction as the spatial axis, a dynamic sliding window with a length of 5 meters was constructed. Each window integrated 10 feature vectors (covering key geological parameters such as rock compressive strength, density, wave velocity, and porosity). A real-time data stream processing mechanism was employed; every 0.5 meters of tunnel face advancement, the earliest data within the window was automatically removed, and newly acquired geological data was simultaneously added, ensuring that the window always covered the latest exploration area. In this way, a continuous geological data sequence was formed, providing a dynamically updated dataset for subsequent analysis.

[0056] S22. Based on density clustering algorithm, dynamically cluster and divide regions with similar geological features.

[0057] Clustering analysis of feature vectors within a window is performed using a density-based clustering algorithm. The core parameter, the neighborhood radius ϵ, employs a density-adaptive adjustment strategy: by calculating the local density of data points within the window, when the local density exceeds a preset threshold (e.g., 50 data points per cubic meter), ϵ is reduced by 20% to prevent over-clustering in high-density areas; conversely, ϵ is appropriately increased in low-density areas. This process effectively identifies data points with similar geological characteristics, grouping them into the same category, with each category corresponding to a lithological region along the tunnel face advancement path.

[0058] S23. For cluster centers at adjacent positions in the clustering results, use Euclidean distance to calculate the inter-class distance sequence.

[0059] For cluster centers at adjacent locations in the clustering results, Euclidean distance is used to calculate the inter-cluster distance sequence, constructing an inter-cluster distance sequence. Abrupt change point detection is then performed on this sequence using an exact pruning linear-time algorithm, satisfying the following condition:

[0060]

[0061] in, This represents the inter-class distance at position t. Represents the mean of the distance sequence. This represents the standard deviation of the distance sequence. When the inter-class distance at a certain location exceeds the mean plus three times the standard deviation, that location is considered a geological interface abrupt change point. This threshold setting is based on the 3σ principle and can effectively identify significant geological interface changes.

[0062] S24. Calculate the confidence level of geological interfaces by combining cluster purity and mutation point significance, and obtain a geological interface sequence containing confidence level information.

[0063] Cluster purity is defined as the proportion of data points with the same true lithology in the same cluster. When the purity exceeds 90%, it indicates that the clustering results have high reliability. The significance of abrupt change points is measured by the magnitude of the inter-cluster distance exceeding the threshold. When the magnitude of the exceedance is greater than 5 times the standard deviation, it indicates that the interface change is significant.

[0064] Based on these two indicators, the confidence score calculation formula is constructed as follows:

[0065] When purity ≥ 90% and amplitude ≥ 5σ, the confidence level is 100%.

[0066] When purity ≥ 90% and 3σ ≤ amplitude < 5σ, the confidence level is 90%.

[0067] In other cases, linear interpolation is performed based on the actual parameter values ​​to ensure that the confidence level ranges from 0 to 100%.

[0068] This generates a sequence of geological interfaces containing confidence level information, formatted as "mileage marker, confidence percentage" (e.g., "K1+200.5m, confidence level 92%"). Simultaneously, detailed geological feature snapshots are generated at each interface, including lithological feature vectors (such as rock type and mineral composition ratio) and key physical parameter values ​​(elastic modulus, Poisson's ratio, etc.) on both sides of the interface, providing comprehensive data support for geological analysis and engineering decision-making.

[0069] By utilizing high-precision ground-penetrating radar and intelligent sensing equipment, a millimeter-level three-dimensional geological model is constructed to identify the undulations, attitude changes, and extension trends of the interface between soft and hard rock in real time, and to simultaneously generate dynamic visualization maps. Based on deep learning algorithms, the interface mechanical parameters are intelligently inverted, and combined with engineering design requirements, the spatial boundaries of blasting operations are precisely delineated. This provides a scientific basis for optimizing blasting hole network parameters and intelligently controlling the charge amount, enabling refined and intelligent management of blasting operations.

[0070] S3. Analyze the multimodal fusion feature vector, the geological interface sequence, and the network resource allocation strategy to obtain a lightweight geological situation information package.

[0071] In one embodiment, the analysis of the multimodal fusion feature vector, the geological interface sequence, and the network resource allocation strategy to obtain a lightweight geological situation information package includes the following steps:

[0072] S31. The key geological information is fused through the multimodal fusion feature vector and the geological interface sequence, and then dimensionality reduction is performed.

[0073] In this embodiment, a multi-source data fusion framework is constructed. Through multi-modal data acquisition methods such as ground-penetrating radar and drill cores, deep learning algorithms are used to cross-validate and extract features from the data. The focus is on analyzing the current rock mass grade at the tunnel face, and a comprehensive evaluation is conducted based on the "Engineering Rock Mass Classification Standard" and parameters such as the Rock Quality Degree (RQD) and rock mass integrity coefficient. Probabilistic seismic imaging technology is used to accurately identify the locations of the three nearest geological interfaces. Simultaneously, the confidence level of each interface location is quantified using Monte Carlo simulation, with the error range controlled within ±0.5 meters. At the interfaces, acoustic wave testing and indentation testing are used to accurately obtain key characteristic parameters such as compressive strength (error <3%) and joint density (resolution up to 0.1 joints / meter).

[0074] Furthermore, a deep convolutional autoencoder is employed to reduce the dimensionality of the fused feature vectors. By designing a multi-layer convolutional and deconvolutional network structure, the dimensionality is reduced to 128 dimensions while retaining over 95% of the key information. Simultaneously, a feature importance evaluation model is constructed to eliminate redundant features, ensuring that core geological information is not lost. For geological interface sequence data, an improved differential coding algorithm is used to record the three-dimensional spatial differences between adjacent interfaces, combined with run-length encoding to further compress the data.

[0075] S32. Lightweight geological situation information packets are obtained by using a curvature-based simplified algorithm for lightweight processing and adaptive compression according to the network resource configuration strategy.

[0076] A curvature-based simplification algorithm is used to convert the 3D geological model into a wireframe model, preserving the topological structure of key interface contours. A confidence-driven texture mapping mechanism is established. For areas with a confidence level ≥70%, high-resolution texture mapping technology is used to restore realistic geological textures; for areas with a confidence level <70%, a uniform grayscale gradient representation is used to highlight key areas while reducing the computational load of visualization rendering.

[0077] Furthermore, based on the aforementioned network resource configuration strategy, the compression algorithm and parameters are dynamically adjusted according to dynamic parameters such as network bandwidth and latency: when the network bandwidth is less than 1Mbps, a secondary compression mechanism is initiated, and the feature vector is further compressed to 64 dimensions using vector quantization technology; at the same time, block transmission and incremental update technology are adopted to generate lightweight geological situation information packets with a size strictly controlled within 10MB, and CRC check codes are embedded to ensure the integrity and accuracy of data transmission.

[0078] In this embodiment, network resources (bandwidth, time slots) are dynamically allocated based on the priority of services within the tunnel (e.g., control commands > geological data > video streams) to ensure the real-time performance and reliability of critical services. Obtaining the network resource allocation strategy includes the following steps:

[0079] First, business is classified and sliced ​​based on business priority and performance sensitivity.

[0080] Based on service priority and performance sensitivity, services are divided into three levels, P1-P3, and differentiated logical network slices are constructed. P1 level services cover highly sensitive scenarios such as industrial control commands and remote medical surgery, requiring end-to-end latency <100ms and packet loss rate <0.1% to meet real-time and reliability requirements. P2 level services include interactive applications such as high-definition video conferencing and cloud gaming, requiring a minimum bandwidth greater than 5Mbps to ensure a smooth experience. P3 level services are for non-real-time tasks such as file transfer and background data synchronization, allowing a maximum latency of <500ms, and can be appropriately downgraded when resources are scarce. Each slice is isolated through network function virtualization technology to avoid performance degradation caused by resource preemption.

[0081] Secondly, an intelligent resource allocation model is constructed to dynamically allocate network resources.

[0082] A deep Q-network algorithm from deep reinforcement learning is introduced to construct an intelligent resource allocation model. A 20-dimensional state space is collected in real-time with a 5-second sampling period, comprising network topology, bandwidth utilization of each slice, CPU load, etc.; the action space is defined as 1000 different combinations of slice bandwidth allocation ratios (step size 0.1%), and exploration and utilization are balanced through an ε-greedy policy.

[0083] The reward mechanism adopts a tiered design: when the QoS of P1 services meets the standard, a reward of +100 is given, and when it does not meet the standard, a reward of -500 is given; the rewards for meeting the QoS of P2 / P3 services are +20 and +10, respectively.

[0084] In network congestion scenarios (total available bandwidth < 10Mbps), a predefined three-level emergency strategy is triggered: P1 slices are allocated 60% of the bandwidth to ensure control command transmission, P2 slices are reserved 30% to maintain basic service operation, and P3 slices are allocated only 10% of the bandwidth and non-essential data transmission is suspended.

[0085] A sliding window mechanism is also introduced to dynamically adjust the allocation threshold based on the resource utilization rate over the past 10 minutes.

[0086] Furthermore, based on the characteristics of the business, the transmission protocol and processing strategy are determined, and a network resource configuration strategy is generated.

[0087] Transmission protocols and processing strategies are customized based on business characteristics: P1-level services use UDP protocol combined with forward error correction (FEC) encoding to reduce transmission latency while compensating for potential packet loss; P2 / P3-level services use the TCPBBR congestion control algorithm to optimize transmission efficiency by probing network bandwidth and minimizing latency. For non-critical historical video data, deep learning-based compression models (such as VVC encoders) are deployed at edge nodes to transcode 1080P video to 720P format in real time, and then transmitted locally using a content delivery network (CDN) to reduce pressure on the core network. For data with extremely low timeliness requirements, chunked transmission and asynchronous processing mechanisms are further adopted to complete transmission during network idle periods.

[0088] Based on real-time data acquisition, a dynamic resource configuration strategy set is generated. The core outputs include: ① a slice bandwidth allocation table updated every second, containing the current bandwidth, water level threshold, and adjustment suggestions for each slice; ② a transmission protocol intelligent decision matrix, recommending the optimal protocol configuration based on service type and network status; and ③ a QoS guarantee strategy library, providing emergency switching solutions for different fault scenarios (such as single-link interruption or equipment overload). By continuously monitoring the QoS indicators of the P1 service, when the compliance rate falls below 99%, a strategy optimization process is automatically triggered, retraining the DQN model and adjusting resource allocation parameters, forming a closed-loop optimization mechanism.

[0089] In complex tunnel network environments (multiple obstacles, signal attenuation), by constructing adaptive routing protocols and redundant transmission channels, combined with anti-interference coding techniques and dynamic bandwidth allocation strategies, the effects of multipath effects and shadow fading are effectively resisted, ultimately ensuring the stable transmission of critical services such as control commands.

[0090] S4. Based on the surrounding rock grade and geological interface characteristics, dynamically generate a blasting scheme that adapts to the current geological conditions. Combine the multi-equipment collaborative operation sequence and path plan with the blasting scheme to generate the final operation instruction set.

[0091] In one embodiment, the dynamic generation of a blasting scheme adapted to the current geological conditions based on the surrounding rock grade and geological interface characteristics includes the following steps:

[0092] S411. Establish a geological-blasting parameter mapping library, and obtain an initial blasting scheme based on the geological-blasting parameter mapping library.

[0093] Based on historical tunnel blasting engineering data, a random forest regression model was used for deep training. By analyzing the surrounding rock grade (covering different geological conditions from I to V) and interface characteristics (including multi-dimensional parameters such as the difference in compressive strength on both sides of the interface, interface dip angle, and the degree of rock joint development), a model was established that correlates with blasting parameters. The nonlinear mapping relationship between (charge amount Q, hole spacing a, and detonation time difference Δt) is established. Specifically, historical data is divided into training and testing sets in an 8:2 ratio. Model parameters are optimized through cross-validation, ultimately constructing a high-precision mapping expression that satisfies: This mapping library is continuously updated, and the model is retrained when new engineering data is accumulated to ensure the timeliness and accuracy of the mapping relationships.

[0094] In practical engineering applications, the first step is to obtain the surrounding rock grade data for the current construction section. Simultaneously, combined with the working face location information (accurately calculating its distance to the nearest geological interface), the corresponding initial blasting parameters are retrieved from the geological-parameter mapping library. Different initial parameter benchmarks are set for different surrounding rock grades. For example, for Grade V soft rock, due to its low rock strength and poor stability, the initial hole spacing is set to 0.8m, and the charge quantity is controlled at a relatively low level; while for Grade I hard rock, the initial hole spacing can be appropriately increased to 1.2m, and the charge quantity is correspondingly increased. At the same time, the distance between the working face and the geological interface is considered. If the distance is far, standard initial parameters can be used; if it is gradually approaching the interface, parameter fine-tuning is performed in advance to prepare for subsequent adaptive adjustments.

[0095] S412. Adjust the initial blasting scheme according to the geological interface adaptability to generate a blasting scheme that is suitable for the current geological conditions.

[0096] Specifically, when the distance between the tunnel face and the geological interface is ≤1m, the fine-tuning stage begins. High-definition cameras and ground-penetrating radar installed on-site capture snapshots of interface characteristics, and analyze the difference in compressive strength and interface type (fractured zone, fault, abrupt lithological change interface, etc.) on both sides of the interface in real time. If the difference in compressive strength on both sides of the interface is detected to be greater than 30MPa, to avoid over-excavation on the soft rock side due to excessive blasting impact, the charge amount is reduced by 15% on the side closer to the soft rock, and the hole spacing is reduced by 10% to decrease the concentration of blasting energy. When the interface is a fractured zone, to effectively control blasting vibration and prevent damage to the stability of the surrounding rock, the initiation time difference is appropriately increased from the conventional 25ms to 50ms. Furthermore, based on the width and degree of fracture of the fractured zone, the initiation sequence is further optimized, employing segmented micro-delay initiation to disperse blasting vibration energy.

[0097] Furthermore, based on the tunnel's designed cross-sectional dimensions, a rigorous safety verification system is established. Verification includes the minimum resistance line to ensure proper energy transfer during blasting and prevent blasting from occurring in the blast; and the maximum charge limit to control over-excavation to no more than 15cm, ensuring tunnel quality. Simultaneously, it is necessary to verify whether the blasting vibration velocity meets the stability requirements of surrounding buildings and rock, calculated and evaluated using the Sadovsky formula. If parameters do not meet safety constraints, an iterative adjustment program will automatically initiate, adjusting the hole spacing in 0.1m increments and increasing / decreasing the charge in 5% increments, recalculating all parameters until all safety indicators are met.

[0098] After the aforementioned rigorous process, a parametric blasting design scheme is automatically generated, detailing key data such as borehole coordinates (accurate to the centimeter level, including three-dimensional spatial coordinate information), charge quantity for each borehole, and detonation sequence. The blasting design scheme is highly compatible and can be directly read and analyzed by intelligent drilling rigs, enabling automated and precise blasting operations. Furthermore, the scheme document includes detailed parameter descriptions and safety precautions, facilitating on-site personnel's review and execution, providing comprehensive and reliable digital guidance for tunnel blasting construction.

[0099] Based on the surrounding rock grade and geological interface characteristics, this invention dynamically generates blasting parameters that are adapted to the current geological conditions, achieving precise matching between "geology and parameters" and dynamic adaptation between blasting parameters and geological conditions, thereby reducing over-excavation, under-excavation, and excessive disturbance to the surrounding rock.

[0100] In this embodiment, the step of combining the multi-equipment collaborative operation timing and path plan with the blasting scheme to generate the final operation instruction set includes the following steps:

[0101] S421. Construct a high-precision three-dimensional digital twin model of the tunnel.

[0102] A high-precision three-dimensional digital twin model of the tunnel was constructed using BIM+GIS fusion technology. Blasting parameters such as drilling depth, explosive equivalent, and detonator segment, as well as the preset travel paths of equipment such as loaders and rock drilling rigs, were seamlessly imported into the simulation platform through API interfaces.

[0103] Furthermore, the entire construction process, from borehole positioning and charging connection to ventilation and muck removal, is dynamically simulated with a time step of 0.1 seconds. The spatial coordinates and operating speed of each key piece of equipment are recorded in real time. At the same time, data such as the propagation of blasting stress waves and the range of rock mass fracture are visualized based on the LS-DYNA algorithm, forming a digital mirror of the construction process.

[0104] S422. Based on the digital twin model, perform safety rule matching, and combine the multi-equipment collaborative operation sequence and path plan with the blasting scheme to generate the final operation instruction set.

[0105] During the simulation, a rule base comprised of the "Blasting Safety Regulations" and enterprise internal control standards is invoked. Based on the multi-equipment collaborative operation sequence and path plan, an event-driven architecture is used for real-time verification. For example, through a preset spatial fencing algorithm, if it is detected that the loader is still within a danger zone with a radius of 30 meters centered on the blasting point when the blasting countdown is 10 minutes, a level three audible and visual interlock alarm is immediately triggered. Simultaneously, the violation equipment number, current coordinates, and remaining evacuation time are highlighted on the central control screen, and a handling reminder is pushed to the on-site safety officer via the 5G private network.

[0106] For conflict scenarios that trigger alarms, multiple optimization schemes are generated based on genetic algorithms. For example, if the blasting time is delayed, the detonation time is automatically postponed until the loader has withdrawn to the safe zone via the shortest path, and the timeline of subsequent processes is updated simultaneously; or the path planning module is invoked to calculate an alternative route for the loader to detour via the lateral connecting passage based on the real-time traffic conditions of the tunnel, while assessing the impact of the detour on the overall project schedule (with an error controlled within ±5%), and finally providing the decision-making confirmation to the dispatcher through the human-machine interface.

[0107] The optimized construction plan is transformed into equipment control instructions compliant with the ISO15765 protocol, specifying parameters such as the 3D drilling coordinates of the drilling rig, the amount of explosives per hole, and the precise triggering time of the detonator. Simultaneously, a multi-layered safety interlocking mechanism is embedded. In addition to the basic condition of "equipment evacuating to a safe distance of 50 meters," environmental verification conditions such as a methane concentration below 0.5% and normal operation of ventilation equipment are added, forming a priority-based instruction execution sequence to ensure safe and reliable closed-loop control when instructions are sent to each terminal device via industrial Ethernet.

[0108] This invention ensures the safety and feasibility of blasting design and equipment operation plans through digital twin simulation and safety rule verification, generates directly executable instruction sets, eliminates potential safety hazards during construction in advance, ensures the feasibility of operation plans, and avoids accidents.

[0109] In this embodiment, multiple intelligent equipment (drilling rigs, loaders, shotcrete machines, etc.) are regarded as cooperative intelligent agents. Through reinforcement learning, they are trained to autonomously avoid conflicts in dynamic environments and generate globally optimal operation sequences and paths.

[0110] Specifically, each piece of engineering equipment is abstracted as an independent intelligent agent unit, constructing a refined state space S. Its own position (x, y, z) is acquired in real time through a high-precision BeiDou positioning system or UWB indoor positioning technology, with errors controlled to the centimeter level. Task progress indicators are dynamically calculated based on equipment sensor data; for example, drilling equipment records drilling depth using pressure sensors and rotary encoders, which is then converted into a completion rate. Distance information to surrounding equipment and obstacle positions is obtained through the fusion of LiDAR and millimeter-wave radar to construct a three-dimensional environmental map. The motion space A is further refined, with movement directions divided into 16 discrete angles and speed settings in 5 adjustable modes (0.5m / s-2.5m / s). Operation start / stop commands are directly interfaced with the equipment's PLC control system, achieving millisecond-level response.

[0111] Furthermore, a hierarchical, multi-objective reward system is constructed to meet the following needs:

[0112]

[0113] in, Indicates the target reward. The difference between the actual completion time and the preset deadline is awarded as follows: 100 points are awarded for each hour completed ahead of schedule, and 200 points are deducted for each hour delayed. This indicates a spacing bonus: when the equipment spacing d ≥ 5m, 10 points are awarded every 10 seconds; when 3m ≤ d < 5m, a warning is triggered and no bonus is awarded; when d < 3m, 50 points are immediately deducted and emergency braking is triggered. Energy consumption rewards are given based on equipment type (e.g., drilling rig, loader) and task (e.g., drilling depth, hauling distance), with preset energy consumption thresholds for each task (e.g., 5kWh for drilling rigs per 1m hole, 5kWh for loaders per 10m hauling distance). 3 The energy consumption threshold for construction waste is 8 kWh. If the actual energy consumption is less than or equal to the preset threshold (e.g., 9 kWh ≤ 2 × 5 kWh), a positive reward will be given according to the formula "(threshold - actual energy consumption) × energy consumption reward coefficient" (e.g., if the coefficient is 20, the reward is (10-9) × 20 = 20 points). If the actual energy consumption is greater than the preset threshold (e.g., a drilling rig consumes 12 kWh to drill a 2m hole), a negative reward will be given according to the formula "(actual energy consumption - threshold) × energy consumption penalty coefficient" (e.g., if the coefficient is 30, the penalty is (12-10) × 30 = 60 points). This represents dynamic weights, which are adaptively adjusted during training using a genetic algorithm.

[0114] Furthermore, a distributed training framework is constructed based on the MADDPG algorithm. Each agent maintains an independent policy network and value network while sharing a global state encoder. The experience replay mechanism adopts Priority Experience Replay (PER) technology, which calculates sample priority through TD error to improve training efficiency. The communication protocol uses a customized industrial IoT message queue. After the drilling rig completes the drilling operation, it sends a structured message containing information such as the operation coordinates and remaining space to the loader through a publish-subscribe pattern. Upon receiving the message, the loader automatically generates the optimal entry path.

[0115] Furthermore, a dual-loop path planning system is constructed. The outer layer uses the A* algorithm for global path search, while the inner layer embeds a conflict avoidance strategy trained using deep reinforcement learning. The environmental information update frequency is synchronized with the equipment decision cycle, triggering path replanning every 10 seconds. When equipment in operation is detected, a "safe corridor" planning strategy is adopted: a 2-meter-wide buffer zone is generated on both sides of the target path, and the safe speed and turning angle are calculated using the dynamic window method (DWA) to ensure obstacle avoidance while minimizing operational delay.

[0116] Furthermore, a priority decision-making model based on the Critical Path Method (CPM) is established. In the pre-blasting stage, the dependencies between drilling, charging, and wiring processes are determined through topological sorting, and the Resource Constrained Project Scheduling (RCPSP) algorithm is used to optimize equipment allocation. In the post-blasting stage, a dynamic programming algorithm is used to solve for the shortest time path for waste removal. The generated collaborative operation sequence table includes fields such as equipment ID, operation content, start and end times, and resource usage, and is visualized through a Gantt chart. It also supports API interface integration with the on-site scheduling system to achieve dynamic plan adjustments.

[0117] By using a multi-agent distributed learning and global reward mechanism, the equipment can autonomously form "collaborative memory," enabling multiple pieces of equipment to work together efficiently without collisions, thus solving the scheduling problem in the confined space of a tunnel.

[0118] S5. Using the final operation instruction set, the blasting quality comprehensive assessment report, and the lightweight geological situation information package, remote automated centralized control of intelligent equipment is achieved.

[0119] In this embodiment, the process of achieving remote automated centralized control of intelligent equipment using the final operation instruction set, the comprehensive blasting quality assessment report, and the lightweight geological situation information package includes the following steps:

[0120] S51. Construct a multi-dimensional data association system and establish a dynamic mapping model of "operation instructions - geological conditions - blasting effects".

[0121] A multi-dimensional data association system is constructed, establishing a dynamic mapping model of "operation instructions - geological conditions - blasting effects." By combining historical construction data with real-time on-site monitoring data, a causal analysis chain is formed. For example, in tunnel excavation scenarios, when the surrounding rock grade is determined to be Class V (extremely unstable rock strata), and the distance between the tunnel face and the design interface is less than 0.5 meters, if the charge amount is not reduced by 15% compared to the standard value, statistical analysis of numerous cases shows that the over-excavation rate increases by an average of 20%. Such conditions will be marked by the system as typical negative feedback cases and included in a dedicated optimization database.

[0122] S52. By using data-driven iterative optimization of mapping model parameters, and updating and verifying them based on an incremental learning strategy, remote automated centralized control of intelligent equipment can be achieved.

[0123] For negative feedback cases, a three-level model optimization mechanism is initiated, using data-driven algorithm iteration:

[0124] Feature extraction model: A deep learning architecture based on an attention mechanism, dynamically adjusting feature weight allocation. For the key influencing factor of "interface distance," a gradient ascent algorithm is used to increase its weight in model computation, ensuring that even minute changes in distance can quickly trigger alerts.

[0125] Blasting parameter mapping library: The calculation model for the amount of explosives at the interface of Class V surrounding rock is modified and verified through historical cases.

[0126] Multi-agent reward function: In the intelligent decision-making system, the weight system of the reward function is reconstructed, and combined with the time series prediction model, the evacuation path planning algorithm is trained by reinforcement learning, thereby improving the efficiency of equipment evacuation and reducing construction safety risks.

[0127] Furthermore, an incremental learning strategy was adopted, incorporating 200 newly collected sets of construction data (including 50 negative feedback cases) into the training set to retrain the optimized model. A 1:1 simulation environment was built in the digital twin model to simulate 100 extreme working conditions for stress testing, focusing on monitoring key indicators such as over-excavation rate and under-excavation rate. When the over-excavation rate decreased by more than 15% compared to before optimization and passed 3 rounds of independent verification, the model version upgrade process was triggered, and the optimized model was deployed to the online production system.

[0128] Furthermore, the optimized intelligent decision-making model will be seamlessly integrated into the construction management system and applied to the next round of tunnel excavation. Construction data will be collected in real time via IoT devices, triggering a data review every 5 meters of excavation, automatically executing a closed-loop process of "perception-decision-execution-feedback-optimization." Blockchain technology will be used to store data from each iteration, creating a traceable construction optimization archive and providing a standardized reference template for subsequent similar projects.

[0129] This invention integrates geological parameters, equipment operation data, and expert experience collected during construction by constructing a dynamic feedback mechanism, and continuously optimizes the system using machine learning algorithms. It automatically adjusts construction parameters and processes for different geological conditions (such as soft soil foundations and rock strata). At the same time, it establishes a full life-cycle health monitoring system to provide real-time warnings of potential risks, ensuring that the system maintains efficient operation under complex working conditions, eliminating safety hazards from the source, and achieving the goal of sustainable and safe construction in engineering projects.

[0130] In this embodiment, a digital twin model is combined with computer vision to quantify indicators such as the contour and block size after blasting, generating a comprehensive evaluation report to provide a basis for subsequent optimization. Analyzing the comprehensive evaluation report of blasting quality includes the following steps:

[0131] First, a three-dimensional model of the actual cross-section after the blast is generated based on digital twins and computer vision.

[0132] Using specialized data processing software, the collected 3D scan point cloud data of the tunnel face after blasting is accurately imported into the tunnel's digital twin model. Advanced algorithms are then used to quickly generate a 3D model of the actual cross-section after blasting. To ensure the model's accuracy and reliability, rigorous calibration based on the tunnel's axis coordinates is required, precisely aligning the actual cross-section 3D model with the designed cross-section model, controlling the error to the millimeter level, for subsequent accurate analysis.

[0133] Secondly, based on the three-dimensional model, the contour deviation and slag particle size are analyzed to obtain the comprehensive evaluation report of the blasting quality.

[0134] Spatial calculation methods are used to measure the spatial distance between corresponding points in the actual model and the design model. Based on the calculation results, the over-excavation (i.e., the portion with a deviation value greater than 0) and under-excavation (the portion with a deviation value less than 0) are statistically analyzed in detail. Furthermore, professional visualization tools can be used to generate an intuitive deviation heatmap, in which severely over-excavated areas are marked in striking red, moderately over-excavated areas in orange, slightly over-excavated areas in yellow, and normal areas in green, facilitating engineers to quickly identify and analyze problems.

[0135] The YOLOv8 algorithm is used to process the debris images, efficiently identifying individual debris blocks in the images. The equivalent diameter is calculated by the pixel area of ​​each debris block. Granularity statistics are then performed on all identified debris blocks, for example, the proportion of blocks with a diameter less than 30cm is counted; according to engineering requirements, this proportion must reach 80% or higher. Simultaneously, a debris particle size distribution chart is generated, visually displaying the proportion of debris blocks of different sizes.

[0136] Based on the relative importance of each indicator to the blasting quality, three weights are assigned: contour deviation (60%), muck particle size (30%), and blasting vibration monitoring value (10%). The evaluation results of these three indicators are then weighted and integrated to generate a quality score from 0 to 100. A detailed quality assessment report is generated based on the score results. This report not only includes the score results and analysis of each indicator but also provides specific optimization suggestions for areas with severe over-excavation, such as "adjusting the charge amount in areas with severe over-excavation, reducing the explosive dosage by 5%-10%, and optimizing the borehole layout," providing scientific guidance for subsequent construction.

[0137] Please see Figure 2 In an embodiment, to efficiently execute the remote centralized control method for unmanned intelligent equipment in drill-and-blast tunnels provided by this invention, this invention also provides a remote centralized control system for unmanned intelligent equipment in drill-and-blast tunnels, comprising: a multi-source heterogeneous geological data fusion module, used to fuse multi-source heterogeneous geological data and extract multi-modal fusion feature vectors strongly correlated with the surrounding rock grade; a dynamic clustering geological interface identification module, used to divide similar geological feature regions through dynamic clustering, and locate the interfaces of different lithologies by combining abrupt change point detection, to achieve real-time, confidence-based identification of geological interfaces and obtain a geological interface sequence containing confidence information; a blasting parameter scheme generation module, used to dynamically generate blasting parameter schemes adapted to the current geological conditions based on the surrounding rock grade and geological interface characteristics; and a dynamic conflict resolution and path planning module, used to generate globally optimal multi-equipment collaborative operation... The system comprises the following modules: a sequence and path planning module; an intelligent blasting effect evaluation module, which uses a combination of digital twin models and computer vision to quantify indicators such as the contour and block size after blasting and generate a comprehensive blasting quality evaluation report; a network resource allocation module, which dynamically allocates network resources according to the priority of services within the tunnel and generates network resource allocation strategies; a first data processing module, which integrates key geological information and performs lightweight processing to generate a lightweight geological situation information package, enabling the remote center to accurately perceive the geological conditions; a second data processing module, which ensures the safety and feasibility of blasting design and equipment operation plans through digital twin simulation and safety rule verification, and generates a set of final operation instructions that can be directly executed; and an autonomous learning and closed-loop optimization module, which enables remote automated centralized control of intelligent equipment based on the final operation instruction set, the comprehensive blasting quality evaluation report, and the lightweight geological situation information package.

[0138] The present invention provides a remote centralized control system for unmanned intelligent equipment in drill-and-blast tunnels. This system has a compact structure and stable performance, and can reliably execute the remote centralized control method for unmanned intelligent equipment in drill-and-blast tunnels provided by the present invention, further enhancing the overall applicability and practical application capability of the present invention.

[0139] The embodiment also provides a storage medium storing a computer program. When the computer program is executed by a processor, it implements the steps of the above-described remote centralized control method for unmanned intelligent equipment in drill-and-blast tunnels. The storage medium may include various media capable of storing program code, such as a USB flash drive, portable hard drive, read-only memory, random access memory, magnetic disk, or optical disk.

[0140] In summary, this invention improves the accuracy of surrounding rock grade discrimination by fusing multi-source heterogeneous geological data to generate multimodal fusion feature vectors; dynamically clusters similar regions and obtains geological interface sequences with confidence levels to avoid misjudgments of geological abrupt changes; analyzes the above features, interface sequences, and network strategies to generate lightweight geological situation information packages, solving the problems of large data volume and poor interactivity in remote transmission; dynamically generates adapted blasting schemes by combining surrounding rock grade and geological interfaces, and generates the final operation instruction set by combining multi-equipment collaborative timing and path planning to avoid equipment conflicts and parameter mismatch risks; finally, it achieves remote centralized control based on the instruction set, blasting assessment report, and lightweight package, effectively breaking through the bottlenecks of traditional construction such as strong reliance on manual labor, poor geological adaptability, and low transmission efficiency, significantly improving construction safety, efficiency, and accuracy, and providing key technical support for the unmanned landing of drill-and-blast tunnels.

[0141] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the present invention.

Claims

1. A method for remote centralized control of unmanned intelligent equipment for drill-and-blast tunnel construction, characterized in that, Includes the following steps: By fusing multi-source heterogeneous geological data, a multimodal fusion feature vector is generated; Based on dynamic clustering, regions with similar geological features are divided to obtain a sequence of geological interfaces containing confidence information. By analyzing the multimodal fusion feature vector, the geological interface sequence, and the network resource allocation strategy, a lightweight geological situation information package is obtained. Based on the surrounding rock grade and geological interface characteristics, a blasting scheme adapted to the current geological conditions is dynamically generated. Combined with the multi-equipment collaborative operation sequence and path plan and the blasting scheme, a final operation instruction set is generated. By utilizing the final operation instruction set, the comprehensive blasting quality assessment report, and the lightweight geological situation information package, remote automated centralized control of intelligent equipment can be achieved. The step of dividing geologically similar regions based on dynamic clustering to obtain a geological interface sequence containing confidence information includes the following steps: Collect data from the sliding window; Based on density clustering algorithm, regions with similar geological features are dynamically clustered and divided; For cluster centers at adjacent positions in the clustering results, Euclidean distance is used to calculate the inter-class distance sequence. The confidence level of geological interfaces is calculated by combining cluster purity and mutation point significance, and a sequence of geological interfaces containing confidence information is obtained. Detailed geological feature snapshots are generated at each geological interface, including lithological feature vectors and key physical parameter values ​​on both sides of the interface. The process of combining the multi-equipment collaborative operation timing and path plan with the blasting scheme to generate the final operation instruction set includes the following steps: Construct a high-precision three-dimensional digital twin model of the tunnel; Safety rules are matched based on a digital twin model, and the final set of operation instructions is generated by combining the multi-equipment collaborative operation sequence and path plan with the blasting scheme. Construct a hierarchical, multi-objective reward system to satisfy: in, Indicates the target reward. The difference between the actual completion time and the preset deadline is awarded as follows: 100 points are awarded for each hour completed ahead of schedule, and 200 points are deducted for each hour delayed. This indicates a spacing bonus: when the equipment spacing d ≥ 5m, 10 points are awarded every 10 seconds; when 3m ≤ d < 5m, a warning is triggered and no bonus is awarded; when d < 3m, 50 points are immediately deducted and emergency braking is triggered. Indicates energy consumption reward, This represents dynamic weights, which are adaptively adjusted during training using a genetic algorithm. A dual-loop path planning system is constructed. The outer layer uses the A* algorithm to perform global path search, while the inner layer embeds conflict avoidance strategies trained by deep reinforcement learning. A priority decision-making model based on the critical path method is established. In the pre-blasting stage, the dependencies between drilling, charging, and connection processes are determined by topological sorting, and a resource-constrained project scheduling algorithm is used to optimize equipment allocation. In the post-blasting stage, a dynamic programming algorithm is used to solve the shortest time path for waste removal.

2. The remote centralized control method for unmanned intelligent equipment in drill-and-blast tunnels according to claim 1, characterized in that, The process of generating a multimodal fusion feature vector by fusing multi-source heterogeneous geological data includes the following steps: Acquire multi-source heterogeneous geological data and perform systematic data cleaning; Using deep learning and signal processing techniques to transform heterogeneous data and obtain structured features; An attention mechanism is used to fuse multimodal data, and a multimodal fusion feature vector is generated based on random forest.

3. The remote centralized control method for unmanned intelligent equipment in drill-and-blast tunnels according to claim 1, characterized in that, The analysis of the multimodal fusion feature vector, the geological interface sequence, and the network resource allocation strategy to obtain a lightweight geological situation information package includes the following steps: The key geological information is fused using the multimodal fusion feature vector and the geological interface sequence, and then dimensionality reduction is performed. A lightweight geological situation information packet is obtained by using a curvature-based simplified algorithm for lightweight processing and adaptive compression according to the network resource allocation strategy.

4. The remote centralized control method for unmanned intelligent equipment in drill-and-blast tunnels according to claim 3, characterized in that, Obtaining the network resource configuration policy includes the following steps: Classify and slice businesses based on business priority and performance sensitivity; Construct an intelligent resource allocation model to dynamically allocate network resources; Based on the characteristics of the business, determine the transmission protocol and processing strategy, and generate a network resource configuration strategy.

5. The remote centralized control method for unmanned intelligent equipment in drill-and-blast tunnels according to claim 1, characterized in that, The method of dynamically generating a blasting scheme adapted to the current geological conditions based on the surrounding rock grade and geological interface characteristics includes the following steps: Establish a geological-blasting parameter mapping library, and obtain an initial blasting scheme based on the geological-blasting parameter mapping library; The initial blasting scheme is adjusted according to the geological interface adaptability to generate a blasting scheme that is suitable for the current geological conditions.

6. The remote centralized control method for unmanned intelligent equipment in drill-and-blast tunnels according to claim 1, characterized in that, The process of utilizing the final operation instruction set, the comprehensive blasting quality assessment report, and the lightweight geological situation information package to achieve remote automated centralized control of intelligent equipment includes the following steps: Construct a multi-dimensional data association system and establish a dynamic mapping model of "operation instructions - geological conditions - blasting effects"; By using data-driven iterative optimization of mapping model parameters and updating and verifying them based on an incremental learning strategy, remote automated centralized control of intelligent equipment can be achieved.

7. The remote centralized control method for unmanned intelligent equipment in drill-and-blast tunnels according to claim 6, characterized in that, The analysis of the blasting quality comprehensive assessment report includes the following steps: Based on digital twins and computer vision, a three-dimensional model of the actual cross-section after blasting is generated. Based on the three-dimensional model, the contour deviation and slag particle size are analyzed to obtain the comprehensive evaluation report of the blasting quality.

8. A remote centralized control system for unmanned intelligent equipment in drill-and-blast tunnel construction, characterized in that: include: The multi-source heterogeneous geological data fusion module is used to fuse multi-source heterogeneous geological data and then extract multi-modal fusion feature vectors that are strongly correlated with the surrounding rock grade. The dynamic clustering geological interface identification module is used to divide regions with similar geological features through dynamic clustering, and to locate the interfaces of different lithologies by combining abrupt change point detection, so as to realize real-time, confidence-based identification of geological interfaces and obtain a geological interface sequence containing confidence information. The blasting parameter scheme generation module is used to dynamically generate blasting parameter schemes that are adapted to the current geological conditions based on the surrounding rock grade and geological interface characteristics. The dynamic conflict resolution and path planning module is used to generate globally optimal multi-equipment collaborative operation timing and path plans; The intelligent evaluation module for blasting effects is used to quantify the contour and block size indicators after blasting by combining digital twin models with computer vision, and generate a comprehensive evaluation report on blasting quality. The network resource configuration module is used to dynamically allocate network resources according to the priority of services within the tunnel and generate network resource configuration strategies. The first data processing module is used to integrate key geological information and process it in a lightweight manner to generate a lightweight geological situation information package, enabling the remote center to accurately perceive the geological condition. The second data processing module is used to ensure the safety and feasibility of blasting design and equipment operation plans through digital twin simulation and safety rule verification, and to generate a set of final operation instructions that can be directly executed. The autonomous learning and closed-loop optimization module is used to achieve remote automated centralized control of intelligent equipment based on the final operation instruction set, blasting quality comprehensive assessment report, and lightweight geological situation information package.

Images