Abstract
To solve the problem of accurate identification of coal-rock dynamic disaster precursors, a spatiotemporal data integrated monitoring, and early warning system was proposed. The system consists of a spatiotemporal data integration model, a time series and visual monitoring, and an early warning platform. It takes the comprehensive mining face of a deep coal mine as the monitoring object. It uses structured light 3D scanning and Brillouin optical time domain reflectometry to collect physical entity data in the monitoring area, reconstructs data and processes data redundancy through edge microprocessors, and decomposes spatiotemporal objects into elements to construct a data integration model for data integration. Inner relationship and space-time unity. Using the time series database as the data integration model carrier, the processed physical entity data is mapped to the visual monitoring and early warning platform for dynamic simulation display, which provides data support for accurate early warning of coal-rock dynamic disasters. Finally, a prototype system is developed to verify the generality and feasibility of the system.
1. Introduction
Coal-rock dynamic disaster is a damage process with dynamic effects and disaster in which the coal-rock mass is unstable under the action of high stress and high-pressure mass gas and moves at high speed in the mining space. Due to the characteristics of suddenness and strong destructiveness, the occurrence of disasters is difficult to predict in time, often causing casualties and damage to the coal face, which seriously threatens the safety of mine workers and coal mine production. Moreover, with the continuous increase of mining depth and the continuous expansion of mining scale, the frequency and intensity of disasters have become increasingly serious, which seriously affects the normal production of coal mines and restricts the sustainable development of coal mines. However, due to the limited understanding of the mechanism of disasters, the effective early warning of disasters has been hindered, and the realization of reliable early warning is a difficult problem to be solved urgently by coal-producing countries in the world.
At present, the monitoring equipment used for coal-rock dynamic disasters in China mainly adopts microseismic monitoring, stress, and gas monitoring equipment. All three types of equipment have problems, such as single indicator, poor accuracy, and a large number of deployments, which often generate a large amount of uncertain data, resulting in the early warning accuracy rate being less than 80% [1]. Since the reliability and compatibility of the existing sensing monitoring technology cannot meet the requirements, the identification of disaster risk mainly depends on the mathematical deduction of monitoring data. The identification results are highly subjective and uncertain, which makes many deviations in the early warning [2]. In the past, the integrity of the data was compensated by increasing the range and density of sensor arrangement and selecting more densely disaster characteristic indicators. However, the large amount of data calculation caused by this exceeded the data processing time, resulting in the ineffective application of the data. It is urgent to study data management, which is a method to improve the accuracy of early warning [3].
The occurrence mechanism of coal-rock dynamic disasters is complex, and it is difficult to accurately predict using single monitoring data, so it is necessary to integrate a variety of monitoring data [4]. Many scholars have conducted a lot of research on this. Jiang et al. decomposed the disaster precursor state into common precursors and individual precursors, combined monitoring of common precursors and selectively monitored individual precursors [5]. Liang Yuan et al. took the multiparameter physical effects of microseismic, acoustic emission, electromagnetic radiation, stress, etc., produced by disasters as the information characteristics of disaster precursors, and established a multiparameter coupled risk assessment model [6]. Pan et al. took the three production elements of underground personnel, and sensors as a set of spatiotemporal objects, together with a set of adjacent objects and a set of relationships of different spatiotemporal objects to form a triplet to construct a spatiotemporal data model for coal mine safety production [7]. Lu and Kan constructed a 4DGIS early warning model based on the change of the original factor parameters of dynamic disasters [8]. Zhu et al. used the measured environment integrated data to realize the construction of the data model of the mine shaft and the roadway system [9]. Harshita uses a multiclass support vector machine to train the sampled data to identify coal-rock dynamic disasters [10]. In view of the fact that traditional monitoring technology can only detect the stress or strain data of the shallow surrounding rock, it cannot detect the shortage of deformation data in the deep part and in front of the excavation front. Meng et al. added the Brillouin optical time domain reflectometer (BOTDR) technology to rock mechanics parameters to establish a coal-rock settlement data calculation model [11]. Tang et al. provided real-time monitoring for deformation control of surrounding rock in large-scale roadways [12]. Baumann associates physical entity data values with coordinates, extracts data structure information, and creates a general framework for multidimensional spatiotemporal data based on spatiotemporal datasets [13]. Li et al. used the multisource information fusion method to establish a disaster data model based on the temporal correlation and spatial heterogeneity of multisource data to optimize the early warning time of coal and gas outbursts in the process of tunneling [14]. Previous studies have achieved a lot of research results in disaster monitoring and data integration, but there are few studies on data integration management based on data elements and time series database technology in the process of coal-rock dynamic disaster movement, and further research is needed.
In this paper, we focus on the spatiotemporal data integration method and disaster dynamic simulation design based on data elements of spatiotemporal objects and time series database, especially the design of spatiotemporal data model. Section 2 describes the system architecture; Section 3 discusses data model building methods; Section 4 discusses edge data processing, Section 5 discusses time-series database development, Section 6 describes application examples, and finally draws conclusions in Section 7.
2. System Structure
The spatiotemporal data integration system consists of a data acquisition system, an edge data processor, a time series and a visualization platform. A 3D structured light scanner and BOTDR optical fiber sensing equipment are deployed in the monitoring area of the fully mechanized coal mining face to build a data acquisition system. Using the physical entity and stress and temperature monitoring data collected by the sensing equipment as the data source, the edge microprocessor performs data reconstruction, data standardization, and data redundancy processing. The Prometheus time series database is used as the data model carrier and the data storage management platform to standardize data management and improve data application efficiency through the construction of data models. Grafana, ECharts, and Visual Studio Code visualization tools are used to build a dynamic simulation monitoring and an early warning platform to realize real-time monitoring of disaster risk areas. The spatiotemporal data integration architecture is shown in Figure 1.
3. Spatiotemporal Data Integration Model
After the collected monitoring data is subject to feature index constraints, data reconstruction, and time element loading, a spatiotemporal data integration model is constructed based on the time series database, and the spatiotemporal data and stress-temperature data are integrated and managed. The model can be expressed aswhere, is the time element, which is the unique identification code of the space-time object; are the time series of the movement of the monitored object; is the attribute element of the space-time object; is the space (x, y, z) point object; is the temperature value of the space-time object; is the stress value of the space-time object; is the collection of all time series-based datasets.
Based on the spatiotemporal data integration model, the integrated data is decomposed into elements, and the logical relationship with the time series database (TSDB) is established: the time element of the spatiotemporal object is used as the query index, the attribute element is used as the data label, and the spatial element and catastrophe data are used as the data source. Store in a time series database. Its logical relationship is shown in Figure 2.
In the Prometheus time series database, with two hours as a time unit, all collected monitoring data are stored in a block with a separate directory. It contains the metadata file (meta.json) for that period, all sample data (chunks), and an index file (index) that indexes the metric names and labels of the time series of boilerplate data. Deleted files are stored in tombstones. Block data is first kept in memory and automatically persisted on the disk after two hours. Prometheus uses write-ahead-log (WAL) at startup to ensure that data can be recovered when Prometheus crashes or restarts. The directory format of the data stored in the Prometheus time series database is shown in Figure 3.
4. Edge Data Processing
4.1. Data Processing Flow
The traditional data processing method uploads data to the data center for processing. First, the amount of data is too large, and second, the upload time is too long, resulting in outdated data aging and affecting the data application. With the continuous acquisition of sensory data and the loading of time elements, a large amount of redundant data will also be generated. The edge microprocessor is deployed on the sensor side to perform unified data identification, data reconstruction, and redundant processing of sensor data nearby, which can reduce data processing delay, solve data redundancy, and improve data processing efficiency. The content and flow of edge data processing is shown in Figure 4.
4.2. Data Collection Method
Data acquisition equipment consists of a 3D structured light scanner deployed at the fully mechanized mining site and a Brillouin optical time domain reflectometer. The former is used to continuously scan the geographical monitoring object to obtain point cloud data, and the latter uses the reflected light to perform the monitoring of the geographical monitoring object. For stress and temperature monitoring, point cloud data and stress-temperature data obtained by sensing equipment enter the edge microprocessor for data processing.
4.2.1. Indicator Data Selection
Under the condition that the occurrence mechanism of coal-rock dynamic disasters has not been proven, to improve the success rate of disaster early warning, a large number of indicators are often selected to form an index system. Due to the different manifestations of disasters, various indicators have different changes in different disasters, and the changes of some indicators are not closely related to the occurrence of disasters, which loses the representative meaning of indicators. Moreover, due to the misleading indicators, it affects the judgment and causes the early warning rate to be low. The selection of key general indicators is an important guarantee for the success of early warning. After in-depth analysis of the manifestations and characteristics of disasters from the perspective of geotechnical mechanics, it can be determined that all coal-rock dynamic disasters will produce displacements, stress, and temperature changes of physical entities. Therefore, the key general indicators for predicting coal-rock dynamic disasters use data based on spatial-temporal and stress-temperature changes of geographic entities.
4.2.2. Monitoring Data Collection
(1) Geographic Entity Monitoring Data Collection. The traditional environmental monitoring of the underground coal mining face needs to deploy a large number of sensing equipment and the effective monitoring range is narrow. The collected data is affected by various disturbances, and the data deviation is large. 3D-structured light technology has the advantages of quickly and accurately acquiring 3D data of physical entities and dynamically measuring the deformation of physical entities that other traditional sensors do not have. 3D-structured light scanning technology is introduced to replace traditional geographic environment data collection methods.
(2) Stress-Temperature Data Acquisition. Due to the excessive pursuit of the integrity of monitoring data in traditional disaster data collection, a large number of sensors with a wide variety of functions are installed, and the exponential growth of multisource heterogeneous sensing data causes data processing to be overwhelmed. The occurrence of coal-rock catastrophe is closely related to in situ stress. The cataclysm process is the release process of various energies, and the energy release process must be accompanied by temperature changes. Therefore, monitoring in situ stress and temperature changes can achieve effective prediction purposes. BOTDR can simultaneously measure the dual parameters of stress and temperature, and has the advantages of high measurement sensitivity, long detection distance, high spatial resolution, and low cost, and can resolution many other single-function sensors for data acquisition.
4.3. Data Processing Method
4.3.1. Spatial Data Reconstruction
Coal-rock dynamic disaster is a movement process of unbalanced geographical entities. In the real world, all geographic entities exist in the form of “bodies”. In the geographic information system, spatial entities are abstracted into four geometric types: points, line, surface, and volume. Point entities are entities with location attributes, line entities are entities with length attributes, and surface entities are entities with length and width attributes. A body entity is an entity with attributes of length and height. The occurrence of disasters is the result of the coupling of constituent elements of the entity. Cataclysms occur continuously, but data can only be collected discretely. Discrete data is difficult to describe the change trajectory of the disaster state. To visualize the disaster process more realistically, it is necessary to reconstruct the 3D spatial data into a geometric form.
4.3.2. Time Element Loading
The occurrence and development of disasters are essentially the point changes of x, y, and z values loaded with time. The movement process of disasters is an extension of various spatial changes in the time dimension. The starting point and ending point of disasters need to be determined by time. Only the spatial data loaded with time elements can reflect the characteristics of abnormal disasters in real-time. The stress and temperature will also change correspondingly during the movement of the disaster. The accurate prediction of the disaster can be realized only by fully integrating the spatiotemporal data and the stress-temperature data.
4.3.3. Data Redundancy Handling
The amount of data collection loaded on time elements increases exponentially, resulting in a large amount of data redundancy. The solution is to compare the current data with the previously recorded data during the data collection process.
4.3.4. Monitoring Data Processing
(1) Point Cloud Data Processing. The point cloud data is obtained by the dynamic scanning of the geometric position of the geographic entity object by 3D-structured light. After serial processing by the postprocessing software, a three-dimensional simulation digital model of the physical entity is created. The model can accurately determine the disaster coordinates and visually display the disaster range. Catastrophic displacement state and process. The flowchart of 3D model reconstruction based on point cloud data is shown in Figure 5.
(2) Stress-Temperature Data Processing. The principle of BOTDR to achieve stress and temperature monitoring is the interaction of light waves and sound waves in the optical fiber that produces light scattering. Scattered light contains Stokes light with a frequency lower than the incident light frequency and anti-Stokes light with a frequency higher than the incident light frequency. Light will undergo a frequency shift after scattering, known as the Brillouin shift. Among them, the reverse frequency shift will occur under the influence of temperature , and the forward frequency shift will occur under the influence of stress . The relationship between the Brillouin frequency shift and temperature, stress is shown in the following equation:where, is the Brillouin frequency shift when the strain and temperature of the fiber change; represents the frequency shift of Brillouin scattering light when the strain and temperature change values are zero; is the strain factor; is the temperature factor; is the strain value; is the temperature value.
The frequency shift is linear with strain and temperature. When T = 20°C, n ˜ 1.46, VA = 5945 m/s, and is 11.2 GHz. For every 10−3 change in strain, the Brillouin frequency shift changes by about 50MHZ. For every degree change in temperature, the Brillouin frequency shift changes by about 1.2 GHZ [15]. Therefore, the temperature and strain of the monitored object can be calculated by measuring the Brillouin frequency shift variation.
4.4. Data Classification Model Based on SVM
4.4.1. Data Set
This dataset is to classify and train the data collected by coal mine sensors, including surrounding rock data, gas data concentration data, truss support pressure, and other sample data. Each sample contains 13 feature components, and the class label of each sample is given. Taking 50% of these samples as the training set and the other 50% as the test set, the classification model can be obtained by training the SVM with the training set, and then the obtained model can be used to predict the class label of the test set. The predicted results are as follows and are shown in Figures 6 and 7.
4.4.2. SVM Model Establishment
First, the training set and the test set need to be extracted from the original data, and then a certain preprocessing is performed. The SVM is then trained on the training set, and the resulting model is finally used to predict the classification labels of the test set. The SVM model is shown in Figure 8.
The data preprocessing adopts normalization processing. The normalization preprocessing is performed on the training set and the test set, and the normalization map used is as follows:
4.4.3. Optimization of SVM
Using the sparrow search algorithm [16] to optimize the penalty parameter and kernel function parameter in SVM.
4.4.4. Test Results
The parameters of the sparrow search algorithm are set as follows:
|
From the final result, the prediction accuracy of the sparrow-optimized SVM test set is 100%. While the unoptimized SVM has a correct rate of 98.78%. Therefore, the SVM optimized by the sparrow search algorithm can improve the accuracy of data classification.
5. Time Series Database
5.1. Database Structure
Traditional data models often use relational databases for data storage, but the disadvantage is that only the current state of entity objects can be stored. When the state of the entity object changes, the original state will be overwritten, so the historical data and the evolution process of the disaster cannot be recorded. The Prometheus time series database can express the real-time status of disasters and the entire evolution cycle. As the carrier and the data storage management platform of the spatiotemporal data model, it uses the timestamp as the index and the attribute data as the identifier, and stores the spatiotemporal and stress-temperature data processed by the edge microprocessor in the time series database, which realizes the disaster monitoring data.
The Prometheus time series database consists of a database, server, and a visual interface. The database stores the data model, the server provides data management and client request services, and the visual interface is used to display the data model and application interface. The structure of the Prometheus time series database is shown in Figure 9.
5.2. Spatiotemporal Data Visualization
Using the Grafana visualization tool to read Prometheus data, you can realize the graphical visualization of spatiotemporal data. It has an excellent data visualization function interface, which can mix different data sources in the same chart, and provide a variety of display methods such as line charts, heat maps, and chart annotations. Moreover, it can also define alert rules for important indicators and notify when the data reaches a threshold. The specific application method steps are shown in Figure 10.
6. Case Study
To verify the versatility and feasibility of the spatiotemporal data model proposed in this paper, an underground coal mine in Inner Mongolia was used as an application pilot to conduct demand analysis and functional design, and then a prototype system was designed and developed on this basis.
6.1. Demand Analysis
Coal-rock dynamic disasters have the characteristics of strong suddenness and serious consequences. Rapid and accurate identification of the acquired data is the premise and guarantee of effective disaster prediction. The storage of historical data is also essential for studying the evolution of catastrophes. According to the construction plan of the coal mine smart mine safety subsystem, the system construction needs to meet the following requirements:(1)The model fits the real physical world;(2)Ability to integrate and process complex spatiotemporal data;(3)To realize historical data storage;(4)It has an algorithm to reduce data redundancy;(5)Reasonable indexing and an efficient retrieval process.
6.2. Function Design
The function design of the prototype system is based on the principles of practicality and easy management.(1)To realize the unified management and integrated application of time and space for multisource heterogeneous disaster monitoring data;(2)Real-time display of catastrophe dynamics and evolution;(3)Real-time early warning of risk precursors.
6.3. Prototype System
6.3.1. Prototype System Architecture
The prototype system is constructed by integrating the aforementioned methods in this paper. After the collected physical object data, stress and temperature change data in the monitoring area are standardized in the edge processor, the Prometheus time series database is used as the data model carrier and the modeling platform for the data storage. We create dynamic visual simulation platforms with Grafana, ECharts, Visual Studio visual source code editor, and Adobe Dreamweaver web editor. With time as the index, the physical entity data is mapped to the visual dynamic platform, and the changes of the monitoring objects are dynamically displayed in real-time monitor according to the threshold set by the monitoring sensor, and automatically alarm when the threshold is exceeded. The prototype system architecture is shown in Figure 11.
6.3.2. Prototype System Building Steps
Step 1. To determine spatiotemporal data acquisition sensor equipment and spatiotemporal data acquisition content;
Step 2. To list the data processing methods and data processing contents, and the data required by the time limit are placed on the edge for processing;
Step 3. The logical structure and the storage method of spatiotemporal data elements in the time series database;
Step 4. The server provides a data index and data dynamic visualization display application for the data application.
As shown in Figure 12.
6.3.3. Dynamic Simulation Monitoring and the Early Warning Platform
The platform is based on the framework structure of the prototype system, takes the time series database as the carrier, adopts Grafana, ECharts, and Visual Studio Code visualization technology, and Adobe Dreamweaver web page code editing technology. It presents the monitoring object data model that encapsulates time and attribute information on the visual simulation platform in the form of dynamic simulation. The data management application results expressed by the data model and monitor abnormal warnings are dynamically displayed.
The platform is divided into a video simulation interface on the main interface and a chart interface on both sides. The upper part of the main interface displays the number of safe production days and the number of early warnings, and the lower part is the real-time monitoring area for underground coal mining, which is composed of live video and digital simulation entity images, which can visually check the dynamics of the monitoring area. The interfaces on both sides display the environment, stress, temperature, and other catastrophic dynamic indicators of the monitoring area, the operation status of the monitoring and emergency plans in the form of graphs. When various values are close to the warning value, the warning bar will turn yellow and a risk precursor warning will be carried out, and the risk precursor processing plan will pop up at the same time. When the value reaches the warning value, the line will appear red and a disaster warning will be carried out, and the corresponding disaster category treatment plan will pop up at the same time. It can also manually select the emergency plan in the lower left corner. If the smart coal mine adopts the cyber-physical systems (CPS) architecture, it can automatically carry out intelligent interactive processing of disasters. Click the mining area distribution in the lower right corner to change the monitoring content of different mining areas. The table shows the interface layout of the dynamic visual monitoring and early warning platform.
7. Conclusion
This paper deeply studies the unified integrated management and visualization of coal-rock dynamic disaster monitoring data in time and space. Based on spatiotemporal objects and time series database, a model suitable for integrated management of spatiotemporal data of coal-rock dynamic disasters is proposed. The model adopts a new combination of monitoring equipment, which can comprehensively improve the monitoring level, reduce the number of data collection and data processing time, and solve the problem of processing time delay due to the large amount of data processing in the past. Based on spatiotemporal objects and time series database, a model suitable for integrated management of spatiotemporal data of coal-rock dynamic disasters is proposed. It adopts a new combination of monitoring equipment, which can comprehensively improve the monitoring level, reduce the number of data collection and data processing time, and solve the problem of processing time delay caused by large data processing in the past. Aiming at the disadvantage that the application of the traditional early warning database does not include time elements, the commonly used relational database is replaced with a time series database, and time is used as the index of catastrophe motion data. Heterogeneous data is difficult to unify application problems. By filtering out the data redundancy problem after loading time elements, the minimization of the data storage, process tracking, and the historical retrospect of catastrophic evolution are realized. Finally, by developing a prototype system and applying the above research results to the simulation and application of an example, it is concluded that the model can reflect the dynamic characteristics of the monitoring object in real-time, and can effectively improve the efficiency of monitoring data management and achieve effective early warning of disaster precursors.
Data Availability
The data used to support the findings of this study are available from the corresponding author upon request.
Conflicts of Interest
The authors declare that they have no conflicts of interest.