A mine space-air-ground integrated geological disaster three-dimensional visualization monitoring system and method adapted to a creative environment

Abstract

The application discloses a mine space-air-ground integrated geological disaster three-dimensional visualization monitoring system and method suitable for a creative environment, and belongs to the technical field of mine safety monitoring. The system adopts a five-layer hierarchical architecture: a data acquisition layer constructs a space-air-ground sensing network of satellites InSAR, unmanned aerial vehicles and ground sensors and internally embeds a plug-in protocol adaptation module; a data transmission layer adopts 5G / 4G+LoRa combined transmission and an edge cache breakpoint continuation mechanism; a data processing and model layer adopts a hybrid database architecture of PostgreSQL, TDengine time series database and object storage, and internally embeds a full-process data management engine; an application service layer adopts a micro-service architecture; and a display interaction layer constructs a digital twin base based on a domestic three-dimensional GIS engine, realizes dynamic linkage of monitoring data and a three-dimensional model and deformation simulation. The system is compatible with domestic servers, operating systems, databases and GIS engines, realizes that data delay is less than 1 second, the completeness rate is 99.8%, and the emergency response time is shortened by more than 60%.

Description

Technical Field

[0001] This invention relates to the field of mine geological disaster monitoring and early warning technology, specifically to a three-dimensional visualization monitoring system for mine geological disasters that is adapted to the information technology innovation environment and integrates air, space, and ground. Background Technology

[0002] With the advancement of intelligent mine construction, geological disaster monitoring is gradually transforming from traditional manual inspections and single-point sensor monitoring to three-dimensional monitoring. Currently, mine geological disaster monitoring platforms mainly suffer from the following technical deficiencies.

[0003] The lack of adaptability to domestic IT infrastructure is a significant challenge. The localization of critical information infrastructure in mines is crucial for ensuring national energy security. However, existing mine monitoring platforms primarily use foreign hardware and software products for their core servers, operating systems, databases, and GIS engines. One key difficulty in the intelligentization of mining equipment lies in the fact that underground industrial operating systems are mainly foreign-based. A single domestically produced coal mining machine uses as many as three different operating systems, three different communication protocols, and 161 different chips, all of which are imported, posing a significant risk. As the lifeline of the national economy, the energy sector faces even stricter requirements for its localization transformation—core production systems in mining, power, and other fields bear the heavy responsibility of safe production and energy supply. During the migration process, 24-hour uninterrupted operation must be guaranteed; "zero downtime" is an inviolable red line.

[0004] Specifically, in the field of geological disaster monitoring, existing technical solutions generally do not consider compatibility with domestically developed technologies. The digital twin evolution method for intelligent mine scenarios proposed by China University of Mining and Technology (Beijing) (CN111210359A) uses Unity3D development tools and foreign commercial software such as Creo Parametric and UG NX for 3D modeling, making its technical solution dependent on the foreign software ecosystem. The geological disaster monitoring and early warning system based on digital twins proposed by Runjian Co., Ltd. (CN121236888A) also does not address compatibility with domestically produced hardware and software environments. Transplanting these technologies, which rely on foreign software ecosystems, to domestic platforms faces multiple technical obstacles, including differences in operating system interfaces, database compatibility, and cross-platform adaptation of 3D rendering engines, which cannot be solved simply by software replacement.

[0005] The efficiency of multi-source data fusion is low. Data from satellite InSAR remote sensing, UAV aerial surveys, ground sensors, and meteorological monitoring vary greatly in format, communication protocols, and acquisition frequencies. Traditional mine transparent geological data originates from various business subsystems, with different standards across subsystems, making data sharing difficult and resulting in severe data silos between mine business systems. Lagging safety construction exposes mines to numerous safety hazards, and data fusion is difficult—data from different systems and sources cannot be efficiently integrated, severely hindering the advancement of smart mine construction. Currently, the level of mine intelligence is generally low, and the ability to comprehensively analyze multi-source heterogeneous data is lacking.

[0006] In existing technologies, the multimodal land surface change monitoring method (CN120252832A) proposed by Qingdao Jielida Geographic Information Group Co., Ltd. mainly involves the simple aggregation and AI analysis of data from different sources, without establishing a standardized multi-source heterogeneous data governance system. The landslide monitoring multi-source heterogeneous data fusion algorithm based on BP neural network proposed by Wang Zhiwei et al. (Journal of Geomechanics, Vol. 26, No. 4, 2020) is an algorithm-level data fusion approach that does not involve a systematic data governance architecture. None of the above solutions address the systemic problems of multi-source data in terms of format uniformity, quality verification, and spatiotemporal alignment.

[0007] There is a disconnect between 3D visualization and monitoring operations. Existing 3D platforms mostly only provide static displays of terrain models, failing to deeply integrate monitoring data, equipment status, and early warning information with the 3D model. When digital twin technology is applied to intelligent mine disaster early warning and comprehensive prevention and control, the instantaneity of data acquisition, transmission, and computation is a core requirement—the suddenness of accidents demands that the digital twin system complete data acquisition, transmission, and simulation calculations in real time. If the transmission and computation speeds cannot keep up, the disaster status in the digital twin system will lag behind the physical world, failing to achieve the effect of disaster early warning. However, existing mine digital twin solutions mostly focus on equipment-level or working face-level twin modeling (such as the overall mining face area, single-machine mechanical equipment status monitoring, etc.), and have not yet formed a complete technical solution for geological disaster monitoring that links monitoring data with 3D models in real time.

[0008] The lightweight 3D model technology proposed by Chen Long et al. (Journal of Industrial and Mining Automation, Vol. 47, No. 5, 2021) implements lightweighting techniques such as texture merging, vertex compression, LOD, and triangle simplification based on the Cesium 3D Tiles format. However, it does not address the linkage mechanism of monitoring data driving model updates in mining geological disaster monitoring scenarios. Upgrading static 3D models to dynamic digital twin systems requires solving the technical problem of real-time vertex updates driven by monitoring data, including spatial interpolation mapping from multi-source heterogeneous data to model vertices and hierarchical rendering strategies for deformation data. These technologies are not publicly disclosed in existing solutions.

[0009] Poor coordination among business modules. The existing platform's monitoring, early warning, inspection, and maintenance modules operate independently, making it impossible to automatically link drone inspection tasks with potential disaster sites. While there are reports of drone-based automated inspection technologies in the current mine safety monitoring field—such as acquiring images and reconstructing 3D models in the cloud through drone automated inspections—the issuance of early warning signals and the distribution of drone inspection tasks still rely on manual coordination, failing to form an automated, closed-loop linkage.

[0010] In the field of open-pit mine slope monitoring and early warning, existing technologies have recognized problems such as reliance on single monitoring technologies, poor fusion of multi-source data, and difficulty in providing effective early warnings. However, the connection between early warning and emergency response is still mainly based on manual operation. After the early warning information is pushed out, managers need to manually analyze it, manually arrange inspection tasks, and manually dispatch drones. The entire process involves multiple stages of manual intervention, which significantly restricts the efficiency of emergency response.

[0011] Large-scale 3D models load slowly. Mine geological disaster monitoring involves a massive amount of data when multiple models, including terrain, buildings, slopes, tunnels, and equipment, are overlaid. Existing lightweight 3D modeling technologies are primarily geared towards general scenarios and are not specifically optimized for large-scale scenes in mine geological disaster monitoring. Loading large-scale 3D models on ordinary terminals and mobile devices results in long loading times and low rendering frame rates, failing to meet the needs of frontline operators and managers for real-time monitoring of overall disaster risks.

[0012] In summary, existing mine geological disaster monitoring platforms have significant technical shortcomings in areas such as information technology application adaptation, multi-source data fusion, 3D visualization linkage, business collaboration, and multi-terminal adaptation. Therefore, there is an urgent need in this field for a mine geological disaster monitoring platform that can simultaneously solve the aforementioned technical problems. Summary of the Invention

[0013] This invention provides a three-dimensional visualization monitoring platform for geological disasters in mines that is adapted to the information technology innovation environment and its data processing method. It aims to solve the technical problems of existing mine geological disaster monitoring platforms, such as lack of information technology innovation adaptation capability, low efficiency of multi-source data fusion, disconnect between three-dimensional visualization and monitoring business, poor coordination of business modules, and slow loading of large scene models.

[0014] To achieve the above objectives, the present invention adopts the following technical solution.

[0015] A three-dimensional visualization monitoring platform for geological hazards in mines, adapted to the domestic IT innovation environment, adopts a five-layer architecture and three supporting systems. The five layers are: data acquisition layer, data transmission layer, data processing and modeling layer, application service layer, and display and interaction layer. The three supporting systems are: a domestic IT innovation adaptation system, a security assurance system, and an operation and maintenance management system. The platform is developed based on a domestically produced microservice architecture, and all components are compatible with domestic servers, domestic operating systems, domestic time-series databases, and domestic 3D GIS engines.

[0016] The data acquisition layer constructs a three-dimensional perception network encompassing air, space, and ground. The space-based unit connects to satellite InSAR remote sensing equipment to collect macroscopic data on surface deformation across the entire mining area. The air-based unit deploys fully automated UAV airport suites, supporting preset route inspections, emergency-triggered inspections, and real-time video streaming, acquiring high-definition imagery and oblique photography data. The ground unit uniformly connects to industrial-grade sensors such as GNSS receivers, crack gauges, inclinometers, rain gauges, AI smart cameras, and ground-based radar to collect microscopic monitoring data such as surface displacement, crack width, retaining wall tilt angle, rainfall, and on-site video.

[0017] The data acquisition layer has a built-in plug-in protocol adaptation module that supports mainstream industrial protocols such as MQTT, TCP / IP, and Modbus. Device access plug-ins can be dynamically added to ensure compatibility with monitoring equipment from different brands.

[0018] The data acquisition layer supports two acquisition strategies: timed acquisition and rainfall-triggered high-frequency acquisition. Under normal circumstances, displacement and crack data are acquired at a preset frequency; when the rainfall exceeds a threshold, it automatically switches to high-frequency acquisition mode.

[0019] The data transmission layer employs a combined 5G / 4G and LoRa wireless communication transmission mode. Key equipment utilizes dual-link backup. All transmitted data is encrypted. Edge nodes support local data caching, automatically caching data during network interruptions and resuming transmission once the network is restored.

[0020] The data processing and model layer builds a domestically produced data platform, adopting a hybrid database architecture to adapt to different types of data storage: the domestic PostgreSQL is used to store structured business data such as user permissions, equipment ledgers, and early warning records; the domestic time-series database TDengine is used to store high-frequency monitoring data at the second to minute level; an object storage system is used to store unstructured multimedia data such as drone videos, images, and 3D models; and the PostGIS spatial database is integrated to support 3D GIS spatial operations.

[0021] The data platform has a built-in full-process data governance engine that sequentially performs six processes: data extraction, format conversion, outlier removal, duplicate data cleaning, multi-source data cross-validation, and missing data interpolation and completion, outputting a standardized dataset.

[0022] The data platform also features a geological disaster identification model, which uses AI algorithms to intelligently identify potential hazards from drone images and satellite remote sensing data.

[0023] The application service layer is built on a microservice cluster using domestic middleware, divided into basic services and business services. Basic services include gateway service, registration and discovery service, authentication and authorization service, and configuration center service. Business services are further divided into five independent microservices: data collection service, intelligent inspection service, monitoring and early warning service, comprehensive display service, and system operation and maintenance service. These services are deployed decoupled and communicate via RESTful interfaces.

[0024] The interactive display layer is based on a domestically developed 3D GIS engine to build a high-precision digital twin base for the mining area, integrating DEM elevation data, DOM orthophotos, detailed ground feature models, and underground roadway models, and uniformly adopting the CGCS2000 coordinate system.

[0025] The interactive layer performs lightweight processing on 3D models, employing topology simplification, texture compression, and LOD (Level of Detail) algorithms, and supports streaming loading of 3D Tiles, GLB, and GLTF format models.

[0026] The interactive display layer enables dynamic linkage and binding between monitoring data and 3D models: all sensor, drone, and camera locations are bound to the 3D model one by one, and clicking on a location can display real-time data, historical curves, and device status.

[0027] The interactive layer integrates geological spatial analysis tools such as visual field analysis, profile analysis, topographic measurement, and regional mapping.

[0028] The interactive display layer also enables dynamic deformation simulation: the monitored displacement, crack, and tilt data are dynamically rendered on the 3D model in proportion, intuitively displaying the trend of disaster evolution.

[0029] The interactive display layer is compatible with three terminals: PC Web, mobile mini-program, and large-screen visualization. Warning information is pushed through multiple channels, including platform messages, SMS, and mini-programs.

[0030] Among the three supporting systems, the domestic IT innovation adaptation system requires the use of domestically produced CPU servers and storage devices for hardware, and domestically produced operating systems, databases, middleware, and GIS engines for software. The security system implements tiered user permissions, encrypted data transmission, full recording of operation logs, regular vulnerability scanning, and redundant storage of core data with multiple copies. The operation and maintenance management system integrates full lifecycle management of equipment, system status monitoring, visualized configuration of early warning rules, and log auditing functions.

[0031] The platform's closed-loop business process is as follows: raw data is collected by various air, space, and ground devices, encrypted and uploaded to the data platform after transmission; the data platform cleans, integrates, and stores the data before sending it to the disaster identification and early warning models; after the models determine the risk level, they trigger four levels of early warning (blue, yellow, orange, and red), and push the warning information through multiple channels; the 3D visualization module automatically locates the early warning points and displays the surrounding monitoring data and terrain environment; management personnel issue drone inspection tasks, and on-site personnel receive the tasks and report the handling results via mobile devices; the operation and maintenance module monitors the equipment status, and faulty equipment automatically alarms.

[0032] Compared with the prior art, the present invention has the following beneficial effects.

[0033] First, full-stack domestic IT innovation adaptation. From hardware servers to operating systems, databases, and GIS engines, all components are domestically produced, meeting the policy requirements for domestic IT innovation construction in mines and avoiding the risks of technology supply disruptions and data leaks.

[0034] Second, efficient integration of multi-source data. Adopting a plug-in protocol adaptation and standardized data platform architecture, it is compatible with various monitoring devices, achieving unified governance of heterogeneous data from multiple sources (air, space, and ground), with data latency below 1 second and a data integrity rate of 99.8%.

[0035] Third, the 3D digital twin technology provides profound advantages. The static terrain model is upgraded to a dynamic digital twin system, with real-time linkage between monitoring data and the model. It integrates professional geological analysis tools, which can intuitively simulate the disaster deformation process. After the model is lightweighted, it is adapted to all terminals, solving the problem of loading lag in large scenes.

[0036] Fourth, fully closed-loop business collaboration. Intelligent inspection, monitoring and early warning, equipment operation and maintenance, and emergency command are deeply integrated to achieve a streamlined process of hazard identification, early warning delivery, on-site handling, and equipment operation and maintenance, reducing emergency response time by more than 60%.

[0037] Fifth, it boasts strong versatility and scalability. The plug-in device access and microservice architecture design support the addition of new monitoring devices, expansion of monitoring areas, and integration with external mine management systems, making it widely applicable to various mine geological disaster monitoring scenarios, including lead-zinc mines, coal mines, and metal and non-metal mines. Attached Figure Description

[0038] Figure 1 This is a schematic diagram of the overall architecture of a three-dimensional visualization monitoring platform for geological disasters in mines that is adapted to the information technology innovation environment.

[0039] Figure 2 This is a schematic diagram of the integrated air-space-ground sensing network structure of the data acquisition layer in this invention.

[0040] Figure 3This is a schematic diagram illustrating the working principle of the hybrid database architecture for data processing and model layers and the full-process data governance engine in this invention.

[0041] Figure 4 This is a schematic diagram illustrating the principle of constructing a three-dimensional digital twin model and data linkage binding of the interaction layer in this invention.

[0042] Figure 5 This is a schematic diagram illustrating the division and collaboration relationships of microservice modules in the application service layer of this invention.

[0043] Figure 6 This is a flowchart of the overall business closed loop in this invention.

[0044] Figure 7 Example 2 illustrates the cross-validation method for multi-source monitoring data.

[0045] Figure 8 Example 3 illustrates the dynamic mapping method for three-dimensional model deformation.

[0046] Figure 9 Example 4 illustrates the rainfall-triggered adaptive data acquisition method.

[0047] Figure 10 Example 5: Adaptive method for determining outliers in GNSS monitoring data. Detailed Implementation

[0048] The present invention will be further described below with reference to embodiments, but this is not intended to limit the present invention in any way. Any modifications or substitutions made based on the teachings of the present invention shall fall within the protection scope of the present invention.

[0049] Example 1: Overall Platform Architecture like Figure 1 As shown, the mine air-space-ground integrated geological disaster three-dimensional visualization monitoring platform adapted to the information technology innovation environment provided by this invention adopts a five-layer hierarchical architecture and a three-dimensional supporting system design.

[0050] The five-layer architecture, from bottom to top, consists of: The data acquisition layer is used to connect to various monitoring devices in the air, space, and ground and collect raw monitoring data. This layer constructs a three-dimensional perception network of "air-space-ground". The space-based unit connects to satellite InSAR remote sensing equipment, the air-based unit deploys fully automated UAV airport kits, and the ground unit connects to industrial-grade sensors such as GNSS receivers, crack gauges, inclinometers, rain gauges, AI smart cameras, and ground-based radar.

[0051] The data transmission layer transmits the raw monitoring data acquired by the data acquisition layer to the data processing and modeling layer. This layer uses a combination of 5G / 4G and LoRa wireless communication transmission modes, with key equipment employing dual-link backup. All transmitted data is encrypted, and edge nodes support local data caching and breakpoint resumption.

[0052] The data processing and modeling layer is used to manage, integrate, store, and analyze the received raw monitoring data. This layer builds a domestically produced data platform, adopts a hybrid database architecture, and incorporates a full-process data governance engine and a geological hazard identification model.

[0053] The application service layer provides service-oriented support for the platform's business functions. This layer is built on a microservice cluster based on domestic middleware and is divided into basic services and business services. The business services include data collection services, intelligent inspection services, monitoring and early warning services, comprehensive display services, and system operation and maintenance services.

[0054] The interactive layer provides users with a 3D visualization and interactive interface. This layer is built upon a domestically developed 3D GIS engine to create a high-precision digital twin of the mining area, enabling dynamic linkage and binding of monitoring data with the 3D model, dynamic deformation simulation, and multi-terminal adaptation.

[0055] The three supporting systems include the domestic IT innovation adaptation system, the security assurance system, and the operation and maintenance management system, which run through the entire five-layer architecture.

[0056] Example 2: Cross-validation method for multi-source monitoring data In mine geological disaster monitoring, GNSS displacement monitoring data and crack monitoring data are two important monitoring methods that corroborate each other. When GNSS displacement data and crack data for the same slope area show divergent trends, it usually indicates that the data from one of the monitoring methods may be abnormal. This invention provides a method for cross-validation of multi-source monitoring data, which automatically triggers manual review instructions by quantitatively calculating the degree of divergence between the two data sequences.

[0057] Obtain GNSS displacement monitoring data sequences for the same slope area within the same time period. D ={ d 1, d 2,…, d n} and crack monitoring data sequence C ={ c 1, c 2,…, c n},in n The number of sampling points. d i For the first i Displacement at each sampling time (unit: mm). c i For the first i Crack width at each sampling time (unit: mm).

[0058] Before calculating the trend divergence coefficient, if the sampling frequencies of GNSS displacement monitoring data and crack monitoring data are different, time alignment is required. Time alignment uses a linear interpolation method to unify monitoring data with different sampling frequencies to the same time reference. For the target time... t ( t i ≤ t ≤ t i+1 The interpolation calculation formula is:

[0059] in t i and t i+1 These represent the sampling times at the two endpoints of the interpolation interval. d i and d i+1 These are the corresponding displacement values.

[0060] Calculate the GNSS displacement monitoring data sequence and crack monitoring data sequence in the sliding window W respectively. W Internal trend indicators:

[0061]

[0062] Where sgn( x ) is a symbolic function:

[0063] T d ( t The value range of ) is [-1, 1]. T d ( t When )=1, it indicates that the GNSS displacement data is in the window. W The internal rate continues to rise, when T d ( t When )=-1, it indicates a continuous decline. T c ( t Similarly, this indicates the trend direction of crack data.

[0064] Calculate the trend divergence coefficient:

[0065] when R ( t )> i And the duration exceeds the preset time threshold.T hold At that time, it was determined that the monitoring data of the slope area showed a trend deviation, triggering a manual review instruction.

[0066] in i The preset trend divergence threshold has a value range of 0.5 ≤ i ≤1.0. When i The closer the value is to 1.0, the more critical it is for the trends of the two data sequences to be completely opposite in order to trigger a divergence determination; when... i The closer the value is to 0.5, the lower the tolerance for trend differences. In this embodiment... i The value is 0.8.

[0067] Sliding window W The range of values ​​is 5≤ W ≤30. A window that is too small will cause trend judgment to be too susceptible to noise, while a window that is too large will reduce the sensitivity to trend changes. In this embodiment... W The value is 10.

[0068] Preset time threshold T hold The value range is from 10 minutes to 60 minutes. The purpose of setting this threshold is to avoid frequent manual reviews due to short-term data fluctuations; a review instruction is only triggered after a trend divergence has persisted for a certain period. In this embodiment... T hold The value is 30 minutes.

[0069] In practical application at the Yiliang Chihong Maoping lead-zinc mine, this method successfully identified three instances of divergence between GNSS displacement data and crack data trends in the B9 collapse area. After manual verification, it was confirmed that two of these instances were abnormal data caused by lightning interference with the GNSS equipment, and one instance was a measurement deviation caused by local loosening of the crack gauge installation location, effectively reducing the false alarm rate.

[0070] Example 3: Dynamic Mapping Method for 3D Model Deformation To enable monitoring data to drive the updating of the 3D model, this invention provides a dynamic mapping method for 3D model deformation, which maps the displacement data of discrete monitoring points to all grid vertices of the 3D digital twin model through an inverse distance weighted interpolation algorithm.

[0071] Obtain the set of spatial coordinates of all triangular mesh vertices in the 3D digital twin model of the mining area. ,in M This represents the total number of vertices.

[0072] Obtain the data deployed on the surface of the mine slope. K Three-dimensional spatial coordinates of each monitoring point P k=( X k , Y k , Z k ) and its corresponding displacement vector Δ P k =(Δ X k ,Δ Y k ,Δ Z k ), k =1,2,…, K .

[0073] For any vertex in the three-dimensional digital twin model v j =( x j , y j , z j ), calculate the spatial distance from the vertex to each monitoring point:

[0074] The displacement vector of this vertex is calculated using inverse distance weighted interpolation:

[0075] Among them, weight w jk The calculation formula is:

[0076] In the formula p The distance exponent is 1 ≤ p ≤3. p The larger the value, the greater the contribution of the closer monitoring points to the interpolation result, and the more the interpolation result tends to local features; p The smaller the value, the smoother the interpolation result. In this embodiment... p The value is 2.

[0077] r min The minimum distance threshold ranges from 0.1m to 1.0m. When the distance between the vertex and the monitoring point is less than... r min To avoid excessive weighting that could distort the interpolation results, the weight of this monitoring point is set to 1, while the weights of other monitoring points are ignored. In this embodiment... r min The value is 0.3m.

[0078] When the minimum distance from any vertex to all monitoring points is greater than the preset maximum influence radius R max When the value is zero, the displacement vector of that vertex is zero. R max The value ranges from 50m to 200m. The purpose of setting the maximum influence radius is to avoid unreliable deformation estimation results due to extrapolation in areas far from all monitoring points. In this embodiment... R max The value is 100m.

[0079] Based on the calculated vertex displacement vectors Δ v j Update the spatial coordinates of the corresponding vertex in the 3D digital twin model:

[0080] Generate the deformed 3D model.

[0081] Displacement vector Δ P k For GNSS monitoring stations, the preset time window Δ T The cumulative displacement within. Δ T The value range is from 1 hour to 24 hours. In this embodiment, Δ T The value is set to 6 hours, meaning the deformation state of the 3D model is updated every 6 hours.

[0082] In practical application at the Yiliang Chihong Maoping lead-zinc mine, this method maps GNSS monitoring data from 12 high-risk disaster areas to a three-dimensional digital twin model in real time. Taking the coal mine pass landslide area as an example, when the cumulative displacement of GNSS monitoring points in the area reaches 15mm, the corresponding grid vertices in the three-dimensional model undergo corresponding displacement according to the interpolation results, resulting in a visually perceptible change in the slope surface morphology. Managers can intuitively judge the deformation range and degree of the landslide body through the three-dimensional model, assisting in the formulation of emergency response plans.

[0083] Example 4: Rainfall-Triggered Adaptive Acquisition Method To optimize the allocation of monitoring resources under different meteorological conditions, this invention provides a rainfall-triggered adaptive acquisition method that dynamically adjusts the acquisition frequency of monitoring data based on real-time rainfall.

[0084] Set the normal sampling frequency f 0, and at the normal sampling frequency f 0. Collect displacement monitoring data and crack monitoring data of mine slopes. Routine data collection frequency. f The value of 0 ranges from 0.5 times / minute to 2 times / minute. In this embodiment... f 0 is set to 1 time / minute.

[0085] Real-time acquisition of rainfall monitoring data to calculate the current time t cumulative rainfall R ( t and rainfall intensity I ( t ):

[0086]

[0087] in r ( t )for t Instantaneous rainfall rate at any given time (unit: mm / h). T 0 represents the preset rainfall statistics time window, ranging from 1 hour to 6 hours. In this embodiment... T 0 represents 3 hours.

[0088] When the cumulative rainfall R ( t Exceeding the first rainfall threshold R th1 or rainfall intensity I ( t Exceeding the rainfall intensity threshold I th At that time, the sampling frequency will be changed from f 0 Switch to high frequency acquisition f 1:

[0089] in α The base ratio factor has a range of 2 ≤ α ≤5; β The rainfall response coefficient has a value range of 0.5 ≤ β ≤2.0. In this embodiment α The value is 3. β The value is 1.0.

[0090] First rainfall threshold R th1 The value ranges from 5mm to 20mm. Rainfall intensity threshold. I th The value range is from 2 mm / hour to 10 mm / hour. In this embodiment... R th1 The value is 10mm. I th The value is 5mm / hour.

[0091] when R ( t When )=15mm, the high-frequency acquisition frequencyf 1 = 3 × 1 × (1 + 1.0 × 15 / 10) = 3 × 1 × 2.5 = 7.5 times / minute, that is, from the normal 1 time / minute to 7.5 times / minute.

[0092] When the cumulative rainfall R ( t (Below the second rainfall threshold) R th2 And the duration exceeds the preset recovery time. T recover At that time, the sampling frequency will be changed from f 1. Restore to normal data collection frequency f 0. Among them R th2 < R th1 , R th2 The value ranges from 2mm to 10mm. T recover The value ranges from 30 minutes to 120 minutes. In this embodiment... R th2 The value is 5mm. T recover The value is 60 minutes.

[0093] In practical application at the Yiliang Chihong Maoping lead-zinc mine, this method effectively reduced the energy consumption and data transmission load of the monitoring system during the rainy season. Taking the Damingcao subsidence area as an example, under normal weather conditions, displacement and crack data were collected at a frequency of 1 time per minute, generating approximately 1,440 monitoring data points daily. When rainfall exceeded 10 mm, the system automatically switched to a high-frequency acquisition mode, with a maximum acquisition frequency of 8 times per minute. This fully recorded the accelerated displacement changes in the subsidence area during rainfall, providing high temporal resolution data support for disaster early warning.

[0094] Example 5: Adaptive Method for Determining Outliers in GNSS Monitoring Data GNSS monitoring data may be affected by factors such as multipath effects, ionospheric delay, and equipment failure during the acquisition process, resulting in outliers. This invention provides an adaptive outlier determination method for GNSS monitoring data, which automatically identifies and removes outliers based on the statistical characteristics of a sliding window.

[0095] Obtain GNSS monitoring station continuous time window L Displacement monitoring data sequence {x1,x2,…,x} within L}{ x 1, x 2,…, x L},in LThis represents the window length, with a value ranging from 30 to 120. In this embodiment... L The value is 60.

[0096] Calculate the current time in the displacement monitoring data sequence t displacement value x t Compared to the previous m The moving average of the displacement values ​​at each time point m t and sliding standard deviation s t :

[0097]

[0098] in m The length of the sliding window, with a value ranging from 5 to... m ≤20. In this embodiment m The value is 10.

[0099] Calculate the current rate of change of displacement:

[0100] in e To prevent division by zero, the value range is 0.001 ≤ e ≤0.01. Setting e The purpose is to avoid when s t =0 (i.e., before) m (The displacement values ​​at each moment are exactly the same) A division-to-zero error occurs. In this embodiment... e The value is 0.005.

[0101] when d t > d th At that time, mark the current displacement value. x t These are suspected outliers. d th The preset anomaly detection threshold has a value range of 2.5 ≤ δth ≤4.0. d th The value is based on 3 in statistics s The criterion is determined as follows: when the data follows a normal distribution, the probability of data points exceeding three standard deviations from the mean is approximately 0.27%, which can be considered outliers. In this embodiment... d th The value is 3.0.

[0102] For displacement data marked as suspected outliers, examine the values ​​before and after them. q Are the displacement change rates at each time point all below the preset normal threshold? d normal If so, the displacement value is confirmed as a gross error and discarded; otherwise, it is judged as a normal value.

[0103] like d t-q , d t-q+1 ,…, d t-1 , d t+1 ,…, d t+q < d normal Then confirm x t For gross error in d normal The value range is 1.0≤ d normal ≤2.0, q The range of values ​​for is 2≤ q ≤5. Set before and after. q The purpose of time-series verification is to avoid misjudgment due to sudden changes in data at a single point—if the data before and after a suspected outlier are both normal, then that point is very likely a real outlier; if the data before and after it are also abnormal, then that point may belong to a real jump in the data sequence. In this embodiment... d normal The value is 1.5. q The value is 3.

[0104] After outlier removal from GNSS monitoring data, linear interpolation is used to complete the data at the removed locations.

[0105] in a and b These are the indices of the nearest valid data points before and after the removal position, x a and x bThese represent the corresponding effective displacement values. In practical application at the Yiliang Chihong Maoping lead-zinc mine, this method processed approximately 86,400 displacement monitoring data points (60 days × 24 hours × 60 minutes) at a GNSS monitoring station (sampling frequency 1 time / minute) in the B9 collapse area, identifying and removing 47 outliers, representing approximately 0.054% of the total. Manual verification revealed that 32 of the removed outliers were due to multipath effects (GNSS satellite signals reflected off the ground and entering the receiving antenna, causing measurement errors), and 15 were due to temporary equipment lock-off, verifying the effectiveness of the method.

[0106] Example 6: Deployment and System Integration in a Domestic Information Technology Innovation Environment This embodiment uses the Yiliang Chihong Maoping lead-zinc mine area (including 12 disaster-prone areas such as the B9 collapse area, the Damingcao subsidence area, and the coal pass landslide area) to further illustrate the present invention in detail.

[0107] Hardware Deployment: DJI Airport 3-pack and Matrice 4TD drones were deployed on the roof of the China Nonferrous Metals Industry 12th Metallurgical Construction Co., Ltd. office building. Two types of tasks were set up: routine periodic inspections and emergency inspections triggered by early warnings. Fixed flight routes covering 12 disaster areas were pre-set. GNSS receivers, crack gauges, inclinometers, rain gauges, and AI cameras were deployed in the collapse and landslide areas according to the monitoring design points. All equipment adopted industrial-grade waterproof and dustproof design and connected to the transmission network via LoRa. Multiple domestic Kunpeng CPU servers and domestic distributed storage devices were deployed to build a dual-machine hot standby cluster; the network adopted a dual-link architecture with 5G as the primary and 4G as the backup.

[0108] Deployment in a domestically developed IT environment: The server-side uses a domestically developed operating system, and is equipped with domestic middleware, a domestically developed PostgreSQL database, a TDengine time-series database, and a domestically developed 3D GIS engine. Terminals are uniformly adapted to domestically developed desktop operating systems and browsers, and mobile apps are adapted to domestically developed mobile terminals. All components disable foreign commercial software.

[0109] Data Access and Data Platform Configuration: MQTT is enabled as the primary communication protocol at the data acquisition layer, with dedicated access plugins configured for each type of sensor, drone, and satellite device. Acquisition frequency is set: displacement and crack data at 1 minute / data point, rainfall data at 30 seconds / data point, and drone video transmitted in real-time. Abnormal data judgment rules are set in the data platform: a single GNSS displacement exceeding 10mm is marked as abnormal, and short-term data jumps are automatically filtered. Cross-device cross-validation rules are configured: manual review is triggered when the trends of displacement and crack data on the same slope diverge. Business data is stored in PostgreSQL, high-frequency monitoring data is written to TDengine, and drone video and 3D models are stored in object storage.

[0110] 3D Digital Twin Scene Construction: Import 1:2000 DEM elevation data and 0.5-meter resolution DOM imagery from the mining area to generate a surface terrain model. Combine oblique photogrammetry and manual surveying to construct detailed models of factory buildings, slopes, and retaining walls; import CAD drawings to construct an underground tunnel model. Record the coordinates of all sensors, cameras, and drone take-off and landing points in 12 disaster areas into the system and accurately bind them to the 3D model. Perform LOD layering on the overall model, loading low-precision models at long distances and high-precision models at close distances, compressing texture file sizes to achieve smooth loading on mobile devices. Utilize profile analysis and view area analysis tools, and preset commonly used profile lines for landslide and collapse areas.

[0111] Functional debugging and closed-loop operation: Four-level early warning thresholds are set for three types of disasters: landslides, collapses, and subsidence, with online customization of thresholds supported. Early warning push rules are configured, sending different levels of warnings to corresponding management and on-site personnel. Twelve potential disaster sites are linked to drone flight paths; when a site triggers an early warning, the system automatically dispatches a drone to the site for aerial photography. A three-color status indicator for equipment (green for normal, yellow for warning, red for fault) is enabled; maintenance work orders are automatically generated when equipment is offline or data is interrupted. Full-function testing is conducted on web, mobile, and large-screen platforms to verify that data display, early warning push, model interaction, and report export functions are normal.

[0112] Through actual operation and testing, the system has achieved technical effects such as data latency of less than 1 second, data integrity rate of 99.8%, and emergency response time reduction of more than 60% in 12 disaster-prone areas of the Yiliang Chihong Maoping lead-zinc mine.

[0113] Example 7: Performance Comparison Test of Hybrid Database Architecture To verify the technical effectiveness of the hybrid database architecture of this invention, a comparative test was conducted in the deployment environment of the Yiliang Chihong Maoping lead-zinc mine area.

[0114] Test environment: Domestic Kunpeng CPU server (48 cores), 128GB memory, 4TB NVMe SSD storage. The test data consists of monitoring data from June to August 2025 for 12 disaster areas, including the B9 collapse area, Damingcao subsidence area, and Meitanyakou landslide area. It includes approximately 525,600 GNSS displacement data points (12 locations × 60 days × 24 hours × 60 minutes × 0.5 data points / minute), approximately 525,600 crack gauge data points, and approximately 1,051,200 rainfall data points (30 seconds / data point).

[0115] Comparative Solution 1 is a single PostgreSQL storage architecture (comparative example), where all data is stored in the same PostgreSQL database. Comparative Solution 2 is the hybrid database architecture of this invention (experimental example), where business data is stored in PostgreSQL and high-frequency monitoring data is stored in TDengine.

[0116] The test results are shown in the table below:

[0117] Experimental results show that after adopting a hybrid database architecture of PostgreSQL and TDengine, the write throughput of high-frequency monitoring data is increased by about 15 times, the query latency of tens of millions of data is reduced by about 86%, and the deformation update latency of the three-dimensional model is shortened from 4.2 seconds to 0.8 seconds, providing technical support for the goal of data latency of less than 1 second.

[0118] Example 8: Comparative Experiment on the Effect of Lightweight Model Processing To verify the technical effectiveness of the lightweight processing of the three-dimensional model in this invention, a comparative test was conducted in the Yiliang Chihong Maoping lead-zinc mine area.

[0119] The test model is a 3D digital twin model of the mining area, which includes DEM terrain data (approximately 1.2 million vertices of triangular mesh), DOM image data (0.5m resolution), detailed feature models of 12 disaster areas (factory buildings, slopes, retaining walls, etc.), and underground tunnel models (converted from CAD drawings). The total data size of the original model is approximately 2.8GB.

[0120] The comparison scheme is as follows: the control example uses the original model without lightweight processing; the experimental example uses the lightweight processing scheme of this invention (topology simplification + texture compression + LOD layering).

[0121] Test terminal configuration: ordinary PC terminal (Intel Core i5-10400, 16GB memory, integrated graphics card), mobile terminal (Android system, 8GB memory).

[0122] The test results are shown in the table below:

[0123] Experimental results show that after adopting a lightweight solution combining topology simplification, texture compression, and LOD, the total data size of the model was compressed from 2.8GB to 185MB, with a compression rate of 93.4%. Large-scene 3D models can run smoothly on both ordinary PC terminals and mobile terminals.

[0124] Industrial applicability This invention's method can be widely applied to the determination of grinding media filling rate in various rotary mills, including ball mills, rod mills, semi-autogenous mills, and autogenous mills. This method only requires a single measurement of the material level height H after mill shutdown, combined with structural parameters obtained from design drawings, to complete a high-precision filling rate calculation. It eliminates the need to enter the mill interior and uses expensive equipment such as laser rangefinders. This method can serve as the foundational algorithm for intelligent energy-saving control of mills, and can be installed in the mill's automatic control system to achieve real-time online measurement and closed-loop control.

Claims

1. A method for cross-validation of multi-source monitoring data of geological hazards in mines, adapted to the information technology innovation environment, characterized in that, Includes the following steps: Obtain GNSS displacement monitoring data sequences for the same slope area within the same time period. D ={ d 1, d 2,…, d n } and crack monitoring data sequence C ={ c 1, c 2,…, c n },in n The number of sampling points. d i For the first i Displacement at each sampling time, c i For the first i The crack width at each sampling time; Calculate the GNSS displacement monitoring data sequence and the crack monitoring data sequence in the sliding window respectively. W Internal trend indicators: ， Where sgn( x () is a sign function, when x>0 x When x > 0, the value is 1; when x < 0, the value is 1. x When <0, the value is -1; when x When =0, the value is 0; Calculate the trend divergence coefficient: when R ( t )> θ And the duration exceeds the preset time threshold. T hold At that time, it was determined that the monitoring data for the slope area showed a trend deviation, triggering a manual review instruction. θ The preset trend divergence threshold has a value range of 0.5 ≤ θ ≤1.

0.

2. The method according to claim 1, characterized in that, The GNSS displacement monitoring data sequence and the crack monitoring data sequence are time-aligned before calculation. The time alignment process uses a linear interpolation method to unify monitoring data with different sampling frequencies to the same time reference. in t i ≤ t ≤ t i+1 .

3. The method according to claim 1, characterized in that, The sliding window W The range of values ​​is 5≤ W ≤30, the preset time threshold T hold The value range is 10 minutes ≤ T hold ≤60 minutes 10 minutes≤ T hold ≤60 minutes.

4. A method for dynamic mapping of deformation of a three-dimensional model in a mine geological disaster monitoring platform, characterized in that, Includes the following steps: Obtain the set of spatial coordinates of all triangular mesh vertices in the 3D digital twin model of the mining area. M M The total number of vertices; Obtain the data deployed on the surface of the mine slope. K Three-dimensional spatial coordinates of each monitoring point P k =( X k , Y k , Z k ) and its corresponding displacement vector Δ P k =(Δ X k ,Δ Y k ,Δ Z k ), k=1,2,…, k =1,2,…, K For any vertex in a 3D digital twin model v j =( x j , y j , z j ), calculate the spatial distance from the vertex to each monitoring point: The displacement vector of this vertex is calculated using inverse distance weighted interpolation: Among them, weight w jk The calculation formula is: In the formula p The distance exponent is 1 ≤ p ≤3, r min The minimum distance threshold has a value range of m ≤ r min ≤1.0m; based on the calculated displacement vector Δ of each vertex v j Update the spatial coordinates of the corresponding vertex in the 3D digital twin model. v j ′= v j +Δ v j This generates a deformed 3D model.

5. The method according to claim 4, characterized in that, In the inverse distance weighted interpolation, when the minimum distance from any vertex to all monitoring points is greater than the preset maximum influence radius... R max When the displacement vector of that vertex is zero, then... R max The value range is 50m≤ R max ≤200m.

6. The method according to claim 4, characterized in that, The displacement vector Δ P k For GNSS monitoring stations, the preset time window Δ T The cumulative displacement within the preset time window Δ T The value range is 1 hour ≤ Δ T ≤24 hours.

7. A rainfall-triggered adaptive data acquisition method for a mine geological disaster monitoring platform, characterized in that, Includes the following steps: Set the normal sampling frequency f 0, and at the aforementioned normal sampling frequency f 0. Collect displacement monitoring data and crack monitoring data of mine slopes; Real-time acquisition of rainfall monitoring data to calculate the current time t cumulative rainfall R ( t and rainfall intensity I ( t ): ， in r ( τ )for τ Instantaneous rainfall rate at any given moment T 0 represents the preset time window for rainfall statistics; When the cumulative rainfall R ( t Exceeding the first rainfall threshold R th1 or rainfall intensity I ( t Exceeding the rainfall intensity threshold I th At that time, the sampling frequency will be changed from f 0 Switch to high frequency acquisition f 1: in α The base ratio factor has a range of 2 ≤ α ≤5, β The rainfall response coefficient has a value range of 0.5 ≤ β ≤2.0; when the cumulative rainfall R ( t (Below the second rainfall threshold) R th2 And the duration exceeds the preset recovery time. T recover At that time, the sampling frequency will be changed from f 1. Restore to the normal sampling frequency. f 0, where R th2 < R th1 .

8. The method according to claim 7, characterized in that, The normal sampling frequency f The value of 0 is within the range of 0.5 times / minute ≤ f 0 ≤ 2 times / minute, the rainfall statistics time window T The value of 0 is within the range of 1 hour ≤ T 0 ≤ 6 hours, the first rainfall threshold R th1 The value range is 5mm≤ R th1 ≤20mm, the rainfall intensity threshold I th The value range is 2mm / hour ≤ I th ≤10mm / hour.

9. An adaptive method for determining outliers in GNSS monitoring data of a mine geological disaster monitoring platform, characterized in that, Includes the following steps: Obtain GNSS monitoring station continuous time window L Displacement monitoring data sequence within { x 1, x 2,…, x L },in L The length of the window; Calculate the current time in the displacement monitoring data sequence t displacement value x t Compared to the previous m The moving average of the displacement values ​​at each time point μ t and sliding standard deviation σ t : ， in m The length of the sliding window, with a value ranging from 5 to... m ≤20; Calculate the current rate of change of displacement: in ε To prevent division by zero, the value range is 0.001 ≤ ε ≤0.01; when δ t > δ th At that time, mark the current displacement value. x t These are suspected outliers, among which δ th The preset anomaly detection threshold has a value range of 2.5 ≤ δ th ≤4.0; For displacement data marked as suspected outliers, check the values ​​before and after it. q Are the displacement change rates at each time point all below the preset normal threshold? δ normal If so, the displacement value is confirmed as a gross error and discarded; otherwise, it is considered a normal value. δ normal The value range is 1.0≤ δ normal ≤2.0, q The range of values ​​for is 2≤ q ≤5.

10. The method according to claim 9, characterized in that, The continuous time window L The value range is 30≤ L ≤120; After outlier removal, the GNSS monitoring data is supplemented using linear interpolation to complete the data at the removed locations: in a and b These are the indices of the nearest valid data points before and after the removal position, respectively. x a and x b These are the corresponding effective displacement values.

Images