A gas explosion accident early warning method and system based on data fusion

Abstract

The present application provides a gas explosion accident early warning method and system based on data fusion. The method includes: performing correlation analysis on different accident characteristics of coal mine gas explosion accidents based on the Apriori algorithm to construct association rules for coal mine gas explosion accidents; and, based on the accident causation 2-4 model, constructing a case library for coal mine gas explosion accidents according to the pre-acquired coal mine gas explosion accident samples; based on preset rules, warning the working conditions of accident-prone areas according to the data from multiple data sources collected in accident-prone areas, and determining the warning level of the warning working conditions; in response to the warning level reaching a preset alarm threshold, performing normalization processing on all the data from multiple data sources to obtain normalized data; based on the association rules and case library of coal mine gas explosion accidents, realizing multi-angle cooperation of data to conduct gas explosion accident early warning for accident-prone areas.

Description

Technical Field

[0001] This application relates to the technical field of coal mine safety, and particularly relates to a gas explosion accident early warning method and system based on data fusion. Background Art

[0002] Coal mine gas explosion accidents are one of the important hidden dangers in coal mine safety production. For example, blasting flames often lead to accidents in combination with gas accumulation in confined spaces, insufficient mine air supply, chaotic local ventilation management, improper gas inspection, and unreasonable ventilation system combinations in the category of gas accumulation; electric sparks are combined with insufficient mine air supply and chaotic local ventilation management to trigger accidents; no gas monitoring system is installed and non-mine-approved explosives and electric detonators are used; gas inspectors miss shifts and blasting operators work without certificates, etc.

[0003] Gas explosion is the most serious coal mine accident in terms of casualties. With the continuous expansion of coal mine production scale, due to the reduction of shallow coal seam resources and the continuous increase of coal mine mining depth, there will be a phenomenon of a large increase in gas emission volume, and the risk of coal mine gas explosion accidents is also rising continuously. Coal mine gas explosion accidents will not only cause casualties and property losses, but also have a major impact on social stability and economic development. Therefore, the prevention and control of high gas explosion accidents in coal mines are of great significance. Summary of the Invention

[0004] The purpose of this application is to provide a gas explosion accident early warning method and system based on data fusion to solve or alleviate the problems existing in the above-mentioned prior art.

[0005] To achieve the above purpose, this application provides the following technical solutions:

[0006] This application provides a gas explosion accident early warning method based on data fusion, including: Step S101, perform association analysis on different accident characteristics of coal mine gas explosion accidents based on the Apriori algorithm to construct the association rules of the coal mine gas explosion accidents; and, based on the accident causation 2-4 model, construct the case library of the coal mine gas explosion accidents according to the pre-acquired coal mine gas explosion accident samples; Step S102, based on preset rules, perform early warning on the working conditions of the accident-prone area according to the data from multiple data sources collected in the accident-prone area, and determine the early warning level of the early warning working conditions; Step S103, in response to the early warning level reaching the preset alarm threshold, perform normalization processing on all the data from the multiple data sources to obtain normalized data; Step S104, based on the normalized data, perform gas explosion accident early warning on the accident-prone area based on the association rules and case library of the coal mine gas explosion accidents.

[0007] Preferably, in step S101, based on the Apriori algorithm, the support, confidence, and lift between different accident characteristics of the coal mine gas explosion accident are calculated respectively to construct the association rules of the coal mine gas explosion accident.

[0008] Preferably, the data from the data sources includes at least multiple of: sensor monitoring data, video monitoring data, and text data; step S102 includes: comparing the sensor monitoring data of the accident-prone area collected with a preset monitoring threshold to give early warnings about gas, flame, and ventilation conditions in the accident-prone area; and / or, based on a preset video monitoring algorithm, giving early warnings about accidents and violations in the accident-prone area according to the video monitoring data of the accident-prone area collected; and / or, based on a pre-constructed keyword library, performing text recognition on the text data of the accident-prone area collected to obtain accident information to give early warnings about risk factors in the accident-prone area.

[0009] Preferably, in step S102, the video monitoring data of the accident-prone area collected is sequentially subjected to grayscale, difference, image enhancement, and binarization operations to determine the morphological changes of dynamic targets in the accident-prone area to give early warnings about accidents and violations in the accident-prone area.

[0010] Preferably, in step S102, the text data is vectorized to generate a word vector matrix of the text data; based on the keyword library, according to the word vector matrix, text event detection is performed on the text data to obtain the accident information to give early warnings about risk factors in the accident-prone area.

[0011] Preferably, in step S103, in response to the warning level corresponding to at least one of the multiple data sources reaching the preset alarm threshold, the sensor monitoring data of the multiple data sources is labeled according to the sensor type to perform time matching on the sensor monitoring data.

[0012] Preferably, step S104 includes: determining the risk accident characteristics in the coal mine gas explosion accident according to the case library of the coal mine gas explosion accident; based on the normalized data and the association rules of the coal mine gas explosion accident, determining whether the accident-prone area contains the risk accident characteristics to give early warnings about gas explosion accidents in the accident-prone area.

[0013] The embodiment of the present application also provides a gas explosion accident early warning system based on data fusion, including: a database construction unit configured to perform correlation analysis on different accident characteristics of coal mine gas explosion accidents based on the Apriori algorithm to construct the association rules of the coal mine gas explosion accidents; and, based on the accident causation 2-4 model, construct a case base of the coal mine gas explosion accidents according to the pre-acquired coal mine gas explosion accident samples; a working condition early warning unit configured to perform early warning on the working conditions of the accident-prone area based on preset rules according to the data from multiple data sources collected in the accident-prone area, and determine the early warning level of the early warning working conditions; a normalization unit configured to perform normalization processing on all the data from multiple data sources in response to the early warning level reaching a preset alarm threshold to obtain normalized data; an accident early warning unit configured to perform gas explosion accident early warning on the accident-prone area based on the normalized data, the association rules and the case base of the coal mine gas explosion accidents.

[0014] Technical effects:

[0015] The gas explosion accident early warning method based on data fusion provided by the embodiment of the present application, first, performs correlation analysis on different accident characteristics of coal mine gas explosion accidents based on the Apriori algorithm to construct the association rules of coal mine gas explosion accidents, and based on the accident causation 2-4 model, constructs a case base of coal mine gas explosion accidents according to the pre-acquired coal mine gas explosion accident samples; then, based on preset rules, performs early warning on the working conditions of the accident-prone area according to the data from multiple data sources collected in the accident-prone area, and determines the early warning level of the early warning working conditions; when the early warning level reaches the preset alarm threshold, performs normalization processing on all the data from multiple data sources, and based on the normalized data obtained from the normalization processing, performs gas explosion accident early warning on the accident-prone area based on the constructed association rules and case base of the coal mine gas explosion accidents. Thus, through data fusion, a large amount of information extracted is summarized and the same information is fused to complete the data normalization of various different data sources. Through the constructed association rules and case base, the effective processing and in-depth analysis of the monitoring data from different data sources are realized, and the accuracy and reliability of coal mine gas explosion accident early warning are improved. Description of the drawings

[0016] The specification drawings forming a part of the present application are used to provide a further understanding of the present application. The schematic embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation to the present application. Among them:

[0017] Figure 1 It is a schematic flowchart of a gas explosion accident early warning method based on data fusion provided by some embodiments of the present application;

[0018] Figure 2 Schematic diagram of the principle logic of a gas explosion accident warning method based on data fusion provided according to some embodiments of the present application;

[0019] Figure 3 Classification schematic diagram of the characteristics of risk accidents provided according to some embodiments of the present application;

[0020] Figure 4 Schematic diagram of the structure of a gas explosion accident warning system based on data fusion provided according to some embodiments of the present application. Detailed implementation manners

[0021] The present application will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments. Each example is provided by way of explanation of the present application rather than limitation of the present application. In fact, those skilled in the art will clearly understand that modifications and variations can be made to the present application without departing from the scope or spirit of the present application. For example, features shown or described as part of one embodiment can be used in another embodiment to yield yet another embodiment. Therefore, it is desirable that the present application include such modifications and variations that fall within the scope of the appended claims and their equivalents.

[0022] With the attention to coal mine safety production, the early warning of coal mine gas explosion accidents has become the focus of research in the field of coal mine safety production. The applicant's research found that traditional coal mine gas explosion accident warning methods are mainly based on sensor data, such as gas concentration, temperature, pressure parameters, etc. Although these parameters can reflect the risk of coal mine gas explosion to a certain extent, these methods have problems such as low accuracy, missed protection and false alarms, and cannot fully meet the needs of coal mine safety production. Although the coal mine gas explosion warning based on advanced technologies such as machine learning and artificial intelligence can improve the accuracy and timeliness of early warning, it requires a large amount of monitoring data as a training set. Limited by the extreme complexity of the monitoring environment and the severe safety requirements, it also requires high-level technical personnel for development and maintenance, and may be restricted by the shortage of technical personnel in the actual application process. At present, although coal mine enterprises have established corresponding gas monitoring systems, they only target local monitoring and management, lacking effective processing and in-depth analysis of monitoring data. Under the existing technology and personnel conditions, how to effectively improve the accuracy and reliability of coal mine gas explosion accident warning has become an urgent problem to be solved in the field of coal mine safety production.

[0023] Based on this, the applicant proposed a gas explosion accident warning method based on data fusion, as Figure 1 、 Figure 2 shown, the gas explosion accident warning method based on data fusion includes:

[0024] Step S101: Conduct correlation analysis on different accident characteristics of coal mine gas explosion accidents based on the Apriori algorithm to construct the association rules of coal mine gas explosion accidents; and, based on the accident causation 2-4 model, construct a case library of coal mine gas explosion accidents according to the pre-acquired samples of coal mine gas explosion accidents.

[0025] When conducting correlation analysis on different accident characteristics, it is mainly to use data mining algorithms to conduct correlation analysis on the behaviors and behaviors, and behaviors and physical states of coal mine gas explosion accidents. Different accident characteristics specifically include: gas concentration, gas content, gas pressure, mining progress, ventilation, open fire, unsafe actions, unsafe physical states, obstacles, etc. The Apriori algorithm is used to analyze the correlation between different characteristics in order to find the laws between different accident characteristics. Specifically, based on the Apriori algorithm, the support, confidence, and lift of different accident characteristics of coal mine gas explosion accidents are calculated respectively to construct the association rules of coal mine gas explosion accidents.

[0026] In this application, M and N are used to represent two different accident characteristics respectively, and M => N represents the association rule between M and N. That is to say, when the accident characteristic M appears, the probability of the accident characteristic N occurring concomitantly increases. Specifically, the association rule M => N is characterized by the support, confidence, and lift between M and N respectively.

[0027] Among them, according to the formula:

[0028]

[0029] Calculate the support Support(M => N), confidence Confidence(M => N), and lift Lift(M => N) between M and N. Among them, P(M) is the probability of the accident characteristic M occurring; P(N) is the probability of the accident characteristic N occurring; P(MN) is the probability of the accident characteristics M and N occurring simultaneously.

[0030] In this application, when the lift Lift(M => N) < 1, there is no obvious positive correlation between the accident characteristic M and the accident characteristic N, and there may even be some negative correlations; when the lift Lift(M => N) = 1, the accident characteristic M and the accident characteristic N are independent of each other, that is, there is no significant correlation between the occurrence of accident M and the occurrence of accident characteristic N; when the lift Lift(M => N) > 1, there is a strong correlation between the accident characteristic M and the accident characteristic N, and the larger the value of Lift(M => N), the stronger the correlation between the accident characteristic M and the accident characteristic N, that is, when the accident characteristic M appears, the probability of the accident characteristic N occurring concomitantly is greater.

[0031] In a specific example, in order to further improve the accuracy of the association rules, the minimum support is set to 12% and the minimum confidence is set to 50%. Taking into account the characteristics of gas explosion accidents, in order to explore the correlation characteristics of fire sources or gas accumulation, find out the probability of gas accumulation (or fire source type) appearing under the condition of a certain type of fire source (or gas accumulation type), therefore, the association rule is a result in the form of "a certain gas accumulation => a certain ignition source". From this result, it can be seen that blasting flames are often combined with gas accumulation in confined spaces, insufficient mine air supply, chaotic local ventilation management, improper gas inspection and unreasonable ventilation system in the category of gas accumulation, leading to accidents; electric sparks are combined with insufficient mine air supply and chaotic local ventilation management to cause accidents.

[0032] In another specific example, in order to further improve the correlation between unsafe actions, the minimum support is set to 8% and the minimum confidence is set to 40%. It can be seen from the analysis results that there are 7 strong correlation combinations of unsafe actions that lead to accidents, namely, failure to install a gas monitoring system and use of non-permitted explosives and electric detonators; gas inspectors' missed inspections and blasting workers without a license; insufficient gas inspectors and failure to check gas concentration before blasting and use of water cannon mud; over-boundary mining and failure to conduct safety inspections on electrical equipment; failure to check gas concentration and improper placement of explosives and electric detonators; production beyond ventilation capacity and blasting by non-blasting workers; wind-blown gas discharge and failure to install wind power and gas-electric interlocks.

[0033] According to the analysis results, it can be seen that the correlation between unsafe actions in accidents is relatively obvious, and can be identified and summarized through this method, which can provide clues for identifying and predicting employee violations, so as to help coal mines prevent gas explosion accidents.

[0034] In this application, the cause analysis of accidents is carried out by using statistical laws, based on the theory of accident causes and supported by a sufficient number of accident case reports over a continuous period of time. Specifically, the cause analysis of coal mine gas explosion accidents that have occurred is carried out through the accident cause 2-4 model, and a case library is constructed. High-frequency unsafe actions, physical states, etc. are determined through the case library, so that the data collected in real time in each accident-prone area can be matched with the occurrence of past accidents, and accident warnings can be issued for accident-prone areas.

[0035] In a specific example, the unsafe actions that caused the accident are classified according to the type of fire source and the form of gas accumulation. The classification information is as follows: Figure 3As shown in the figure. The "2-4" model of accident causation is used to analyze the causes of multiple major and extremely large gas explosion accidents, and 1041 unsafe actions are obtained. The analyzed unsafe actions are classified according to the categories leading to fire sources and gas accumulation. Combining the content of the table, the types of operating personnel corresponding to the unsafe actions are determined. During the process of matching the work type and behavior of the work, it is found that some unsafe actions have no specific work type characteristics, and any underground operating personnel have the probability of such unsafe actions, such as smoking underground. Therefore, for such unsafe actions, they are uniformly classified as common unsafe actions; the statistical details according to the types of fire sources and gas accumulation are shown in Table 1, that is, the case library established according to the types of fire sources and gas accumulation is shown in Table 1.

[0036] Table 1 Example of Case Library

[0037]

[0038]

[0039]

[0040]

[0041]

[0042] Step S102: Based on preset rules, according to the data from multiple data sources in the accident-prone area collected, give an early warning to the working conditions in the accident-prone area, and determine the early warning level of the early warning working conditions.

[0043] In this application, the accident-prone area is monitored by different technical means to obtain multi-modal data of the accident-prone area, that is, the data from multiple data sources in the accident-prone area are collected, including at least multiple of sensor monitoring data, video monitoring data, and text data.

[0044] Among them, the sensor monitoring data is mainly obtained through real-time detection by various different types of sensors arranged in the accident-prone area. For example, the methane sensor is used to collect the gas concentration data and gas content data, the pressure sensor is used to collect the gas pressure data, the mining progress management system is used to obtain the mining progress data, the wind speed sensor or contact sensor is used to collect the ventilation data, and the flame sensor is used to collect the open fire data, etc.

[0045] Video monitoring data is mainly collected through switch sensors, cameras, infrared cameras, visible light cameras, multi-line laser sensors, audio sensors, etc. for data such as unsafe actions, personnel information, unsafe physical states, and obstacles in accident-prone areas. Text data is mainly obtained by extracting text from existing accident case texts (such as text records, accident reports, etc.) to acquire information such as risk reports and operation records in the accident case texts.

[0046] When warning about the working conditions in accident-prone areas, for sensor data, the sensor monitoring data collected in the accident-prone areas is compared with the preset detection thresholds to warn about the gas, flame, and ventilation conditions in the accident-prone areas. Specifically, for the gas concentration, when the methane concentration in the heading face collected by the methane sensor is 1% ≤ T 1 < 1.2%, a yellow warning is determined to be issued. When the methane concentration T in the heading face 1 ≥ 1.2%, a red warning is determined to be issued; when the methane concentration in the return airway collected by the methane sensor is 0.7% ≤ T 2 < 0.8%, a yellow warning is determined to be issued. When the methane concentration T in the return airway 2 ≥ 0.8%, a red warning is determined to be issued; when the methane concentration in the dedicated gas exhaust airway collected by the methane sensor is T 3 ≥ 1.6%, a yellow warning is determined to be issued; when the change value of the gas concentration under the same process conditions within one shift ≥ 0.3%, a yellow warning is determined to be issued.

[0047] When the gas content collected by the methane sensor is 7.5 m 3 / t ≤ W < 8 m 3 / t, a yellow warning is determined to be issued; when the gas content collected by the methane sensor is 8 m 3 / t ≤ W, a red warning is determined to be issued. When the gas pressure collected by the pressure sensor is 0.7 MPa ≤ P < 0.74 MPa, a yellow warning is determined to be issued; when the gas pressure collected by the pressure sensor is 0.74 MPa ≤ P, a red warning is determined to be issued.

[0048] When it is known through the mining progress management system that the heading and longwall faces are within 20 meters of a geological structure zone, within 50 meters of a surface drilling or borehole, or when the roof is broken, a yellow warning is determined to be issued; when the coal bedding in the heading face is disordered, the coal seam strike and dip suddenly change sharply, or the coal body damage type changes from Class I and II coals to Class III, IV, and V coals, a yellow warning is determined to be issued.

[0049] When it is monitored by a wind speed sensor or a contact sensor that there is a stop of air supply, gentle breeze, reverse air flow area in the working face, or the ventilation cross-section is less than 2 / 3 of the designed cross-section, a yellow warning is determined to be issued. When open fire is detected by both ultraviolet sensitive components of the flame sensor at the monitoring location, a red warning is determined to be issued.

[0050] When warning about the working conditions in accident-prone areas, for video monitoring data, the types are determined according to the monitored actions and object types, including but not limited to endless-rope winch monitoring, drilling video monitoring, personnel location video monitoring, and distribution cable video monitoring, etc. The main purpose is to monitor video targets and assist in the subsequent abnormal analysis process.

[0051] When it is monitored by a digital input sensor, a camera, an infrared camera, a visible light camera, a multi-line laser sensor, an audio sensor, etc. that there are problems such as failure to report abnormal gas or geological structures or artificial damage to ventilation facilities, and when it is found that the monitoring and control system is out of operation or does not implement gas control measures, they are all determined as unsafe actions and a red warning is issued; when the face information in the video monitoring is matched with the personnel information in the database and it is found that the personnel should not be in this dangerous area or are engaged in non-assigned operations, an alarm is given; when there are significant differences between the video monitoring physical states and the recorded pictures, it is determined as an unsafe physical state or an obstacle and a yellow warning is issued.

[0052] For video monitoring data, during the warning process, based on a preset video monitoring algorithm, according to the collected video monitoring data of the accident-prone area, warnings are issued for accidents and violations in the accident-prone area. Specifically, accidents and violations include but are not limited to rib spalling accidents, open fire, and underground illegal operations, etc. In this application, taking the rib spalling accident in a coal mine as an example, warnings are used to identify the signs before the accident occurs.

[0053] That is to say, by using the probes in the existing underground video monitoring system, in the monitoring area established in the roadway, when a moving object is detected, its position is monitored. Secondly, its position is judged again in the next few frames to realize the judgment of the object's source. Since the left and right displacement of a coal block does not change much when it falls from the top to the bottom, the moving direction (the size of the left and right displacement) of the target can be judged. At the same time, the size (area) and quantity of the moving object can also be judged because as the rib spalling accident is approaching, the falling coal blocks will become larger and larger.

[0054] Regarding the problem of abnormal detection of underground rib spalling, it mainly involves how to detect moving objects and how to determine whether it is an abnormal situation. First, video input is carried out, then the detection of moving targets is carried out, followed by the detection of abnormal situations, and finally early warning is realized or continuous detection is carried out. Using the probes in the existing underground video monitoring system, monitoring areas are set up in the roadway. When the monitoring screen changes, the detection program is started to detect the position of the object and calculate its size (i.e., area) at the same time. In the next few frames, the size of the object is judged again, and whether it is a gas outburst is judged by the change in size. During a gas outburst, the coal body will push open the coal wall and gradually return to its original state. During this process, morphological changes will occur, the coal wall top will be reopened, and this process will be repeated continuously to judge whether a gas outburst has occurred.

[0055] That is, the gray scale, difference, image enhancement and binarization operations are sequentially performed on the collected video monitoring data of the accident-prone area to determine the morphological changes of the dynamic targets in the accident-prone area, so as to give early warning of accidents and violations in the accident-prone area.

[0056] Among them, according to the formula:

[0057] g′ = t(g)

[0058] The gray scale processing is performed on the video monitoring data. In the formula, g′ and g are the gray scale values after and before the transformation of the video monitoring data (image) respectively; t is the preset transformation relationship. Here, it is assumed that the gray scale display range of the video monitoring image is [a, b], and the actual gray scale g range before the transformation of the image is [g 1 , g 2 , [g 1 , g 2 is a subset of [a, b].

[0059] According to the formula:

[0060]

[0061] Performing gray scale transformation on the image can expand the gray scale range of the image to the entire range of the display, thereby improving the contrast of the image and enhancing the display effect.

[0062] For the video monitoring detection algorithm, the extracted target gray scale value is usually greater than the threshold and is located in the upper part of the gray scale histogram, while other points in the middle and lower parts belong to the background. Through the difference operation and threshold segmentation of the background, the moving target and the background can be separated. Here, through multiple threshold segmentations of the background, the middle area and the background are separated.

[0063] After multiple threshold segmentations, three levels can be obtained, namely the background, the middle region, and the target region. Then, different transformation methods are used to process the three levels to strengthen the part of interest, weaken the unconcerned part, and retain the details in the background to further enhance the display effect of the image.

[0064] Next, perform an image enhancement operation on the graph, and obtain two thresholds T 1 , T 2 (where T 1 < T 2 ), and divide the downhole video surveillance image into three levels: the target part A, the gradient part B, and the background part C. For part A, stretching mapping can be used, for part B, it remains unchanged, and for part C, compressive mapping is adopted. The specific mapping method can be obtained through the following formula: Let [a, b] be the display range of the image display device, and [g 1 , g 2 be the actual change range of the gray level before mapping of the original image, then:

[0065]

[0066] where k 0 is the mapping coefficient of the target part A, and the specific value depends on the scene; the stretching coefficient of the gradient part B is 1, and the gray level series range of the gradient part remains unchanged. The stretching coefficient of part C is k 1 (generally less than 1), which can be considered as a compressive processing of the gray level to reduce the gray level series

[0067] Finally, perform binarization processing on the image. Before image processing, find the four extreme white points A, B, C, and D in the four directions of up, down, left, and right; use the annotation to display the coordinate

[0068] values, which are (x A , y A ), (x B , y B ), (x C , y C ), (x D , y D ), and calculate the coordinate value of the center point E of the rectangle according to the formula:

[0069]

[0070] to be (x E , y E ).

[0071] Furthermore, based on the position coordinates of the features (moving objects) in the image at two time points, the vertical height change Δh of the moving object within this period (assumed to be Δt) is calculated.

[0072] Δh = y′ E - y E

[0073] y′ E represents the ordinate of the moving object at the later moment, and y E represents the ordinate of the moving object at the previous moment. If Δh is positive, it indicates that the direction of the moving object is downward; if Δh is negative, it indicates that the direction of the moving object is upward. Based on this, it is possible to effectively determine whether a rib spalling accident occurs and make a preliminary judgment.

[0074] For text data, through the method of text extraction, from the corresponding reports and records, the key sentences of the accident description are obtained from the text through a pre - constructed keyword library. Through word segmentation and a dictionary library, accident information is obtained. When the risk information in the dictionary library appears, it is recorded as a threshold to determine and issue a warning. During the warning process, based on the pre - constructed keyword library, the text data of the accident - prone area collected is text - recognized to obtain accident information for warning the risk factors in the accident - prone area. Specifically, first, the text data is vectorized into a word vector matrix of the text data; then, based on the keyword library, according to the word vector matrix, text event detection is performed on the text data to obtain the accident information and warn the risk factors in the accident - prone area.

[0075] Among them, when performing text vectorization, the text data is successively segmented, word - vectorized, and finally a word vector matrix is generated. The word - segmentation operation is performed through "JieBa", a word - vector model is generated through word2vec, and the segmented text is input into the word - vector model to obtain the word vector of each word. After word - segmentation and word - vectorization, a word vector matrix of N×m is formed, where N is the number of words after word - segmentation and m is the dimension of the word vector.

[0076] The word vector matrices generated for each text are normalized so that the sizes of the word vector matrices generated by all texts are the same. Specifically, according to the formula:

[0077] n = avgN + 2×stdN

[0078] The word vector matrix is normalized. Among them, n is the number of words in the normalized word vector matrix, avgN is the average number of words after word - segmentation of all texts; stdN is the standard deviation after word - segmentation.

[0079] Then, through the keyword library constructed for gas explosion accidents, text event detection is carried out to determine the corresponding risk factors. The constructed keyword library is shown in Table 1:

[0080] Table 1 Keyword Library for Gas Explosion Accidents

[0081]

[0082]

[0083] In addition, to cooperate with the use of the keyword library, an accident causation dictionary is also created, as shown in Table 2. Specifically, first, key sentences in the text are extracted, and the extracted sentences are segmented. The results of sentence segmentation are matched with the information in the accident causation dictionary to obtain the risk factors (such as unsafe behaviors and physical states) that cause the accident.

[0084] Table 2 Accident Causation Dictionary for Gas Explosion Accidents

[0085]

[0086]

[0087] Finally, the SWT path width algorithm is adopted and combined with the detection of the output area of interest in the Chinese printed area to obtain accident information. Compared with the traditional detection process of the entire text image, this method can detect the text area of interest faster, with the advantages of high accuracy, low recall rate, and low time consumption, and can effectively improve the recognition effect of coal mine safety work texts.

[0088] Step S103: In response to the warning level reaching the preset alarm threshold, all data from multiple data sources are normalized to obtain normalized data.

[0089] Due to the heterogeneity and massiveness of multi-source data in the mine, it is difficult to establish the correlation between data, and the steps and methods of data processing greatly affect the monitoring and warning results of accidents; in view of the real-time requirements, therefore, the data analysis algorithm applied to the monitoring and warning of gas explosion accidents not only needs to ensure accuracy but also needs to consider real-time factors. To meet this requirement, in this application, a data normalization hierarchical processing architecture for multi-type sensors is used. This architecture consists of three layers: the first layer is the sensor cluster based on the time axis, the second layer is the layer that supports multi-modal data protocol conversion and secure transmission, and the third layer is the sensor data lake, thematic library, and API interface layer. Through this architecture, sensor data can be normalized at the source, providing a basis for the real-time and efficient analysis of various heterogeneous data. Furthermore, a comprehensive intelligent monitoring algorithm for underground gas explosion accidents can be constructed, and a hierarchical feature correlation network of data can be established to achieve comprehensive monitoring and prediction of such accidents.

[0090] Regarding the problem that there are significant differences in the data sampling periods of various sensors, video monitoring of action states, and text collection in underground coal mines, the data time matching accuracy can affect the accuracy of multimodal data fusion algorithms; and for the judgment of the occurrence, development, and correlation of accidents, data with time synchronization and unified time nodes is required. Therefore, based on a shared data buffer, this scenario designs a sensor cluster layer that supports integrated data acquisition, temporary storage, time matching, and annotation functions. This layer collects, temporarily stores, and matches the time of multimodal sensor data. Each channel updates the sensor data cache queue at different rates based on independent threads and buffers, and this cache queue can support the temporary storage of data. The time matching of data is based on an independent matching thread, which periodically reads the data in the cache queue and annotates it according to the sensor category, thus solving the time matching problem of heterogeneous data of multiple types of sensors. Specifically, when the warning level corresponding to at least one of multiple data sources reaches a preset alarm threshold; the sensor monitoring data of multiple data sources is annotated according to the sensor type to perform time matching on the sensor monitoring data. Thereby, the data with a unified timeline can provide high consistency for subsequent accident analysis, so as to better perform accident prediction and handling.

[0091] Data normalization processing is carried out after the sensor cluster architecture, and there is no rigid requirement for the communication method of the sensor itself. An adaptive communication method can be adopted, and the networking and protocol conversion of multimodal sensors are realized through the use of a sensor network gateway. Specifically, when the gateway receives data of different data types such as ZigBee and LoRa, it can perform protocol conversion and integrate and upload the audio and video data, and establish data communication with the server through WIFI or NB-LoT.

[0092] In the communication layer, the gateway uploads heterogeneous sensor data with time and category information through a serial port or an Ethernet port, and uses the TCP / IP protocol to transmit the data to the sensor data lake, the thematic database, and the API interface layer.

[0093] It should be noted that the sensor data upload method is a passive upload method, that is, according to the pre-set rules, operations are performed in the arithmetic circuit inside the sensor, and data is uploaded periodically at a certain upload frequency.

[0094] In this application, the sensor database, the thematic database, and the API interface layer have the ability to access a large amount of video, sensor, and text data, and successfully realize the work of receiving, storing, filtering, and feature extraction of three types of sensor data. These functions have the structure and format of various types of sensor data standardized and uploaded to the cloud database, and the server will parse the data, thus significantly improving the processing efficiency of subsequent monitoring and warning machine learning algorithms.

[0095] Step S104: Based on the normalized data, and on the association rules and case base of coal mine gas explosion accidents, conduct early warning of gas explosion accidents in accident-prone areas.

[0096] In this application, according to the case base of coal mine gas explosion accidents, determine the risk accident characteristics in coal mine gas explosion accidents, that is, determine the high-frequency unsafe actions, physical states, etc. through the case base. Specifically, taking the authoritative data (accident cases) released within a certain period of time as samples, conduct statistics on risk accident characteristics, and determine that the simultaneous presence of gas concentration (5% - 16%), oxygen concentration greater than or equal to 12%, and ignition source (temperature greater than or equal to 650°C) is a sufficient condition for the occurrence of gas explosion accidents.

[0097] Oxygen is a necessary gas for human survival and breathing. If a person is in an environment with an oxygen concentration lower than 12% for a long time, it can affect the person's respiratory and circulatory systems and mind, threatening life safety. Except for goafs and enclosed spaces in coal mines, the oxygen content in other locations is greater than 20%. Therefore, study the classification of unsafe actions from two aspects: the behavior leading to gas accumulation and the manifestation form of the ignition source; combine the definitions of the ignition source and gas accumulation, and classify the unsafe actions that cause accidents according to the type of ignition source and the form of gas accumulation, as Figure 3 shown.

[0098] Then, based on the normalized data and on the association rules of coal mine gas explosion accidents, determine whether the accident-prone area contains risk accident characteristics, so as to conduct early warning of gas explosion accidents in the accident-prone area. Compare the data collected in real time from each accident-prone area with the past accident occurrence situation, and conduct accident early warning for the accident-prone area.

[0099] In a specific example, taking the blasting flame in the ignition source (the flame generated by illegal blasting) as an example: According to the case base, find the following unsafe actions: failure to find that the minimum burden of the blast hole is insufficient, failure to check whether the hole sealing is qualified, failure to use water stemming, using non-mining permitted explosives and electric detonators, open-fire blasting, charging in batches for one-time charging, non-blasting workers blasting, using coal lumps and rock dust to seal the hole, blasting operators working without a license, failure to implement the system for receiving and returning blasting items, and failure to keep the cartridges in close contact. Corresponding to data sources such as open-fire sensor monitoring, unsafe action monitoring, face information monitoring, and blasting item receiving and returning record texts. Combine and summarize various types of data, and compare them with the past occurrence situation and the danger threshold. When a certain data reaches the alarm threshold, use the association rules to consider the related unsafe actions or physical states, and complete the alarm.

[0100] In another specific example, taking the insufficient air supply in the mine (insufficient ventilation capacity or no mechanical ventilation) leading to gas accumulation as an example: the following unsafe actions are found according to the case base: production beyond ventilation capacity, over-layer mining, no safety measures formulated, power failure and ventilation stop, no ventilation facilities used, relying on the return air of adjacent mines, no repair of the deformed blades of the ventilator, no start of the standby power supply after power failure, using a non-coal mine special blower for air supply, no evacuation of people after power failure and ventilation stop. Corresponding sensors such as gas concentration sensors, ventilation sensors, unsafe action monitoring (detecting whether ventilation facilities are used and whether people are evacuated in time), maintenance record texts, and unsafe state monitoring (detecting whether the blades of the ventilator are repaired, whether the return air setting is reasonable, and whether a special ventilator is used) are used. Merge and summarize various types of data, and compare them with past occurrences and danger thresholds. When a certain data above reaches the alarm threshold, consider the related unsafe actions or states using association rules, and complete the alarm.

[0101] Thereby, through data fusion, a large amount of information extracted is summarized and the same information is fused to complete the data normalization of various different data sources. By matching the constructed association rules and case base with the data fusion process, the information import association rules in the data fusion process are completed, the accidents reaching the initial unsafe information level are marked, and other prone accidents coupled with this are reminded, so as to effectively process and deeply analyze the monitoring data of different data sources, and improve the accuracy and reliability of the early warning of coal mine gas explosion accidents.

[0102] As Figure 4 shown, the embodiment of the present application further provides a gas explosion accident early warning system based on data fusion, including: a database construction unit 401, a working condition early warning unit 402, a normalization unit 403, and an accident early warning unit 404.

[0103] The database construction unit 401 is configured to perform association analysis on different accident characteristics of coal mine gas explosion accidents based on the Apriori algorithm to construct association rules for coal mine gas explosion accidents; and, based on the accident causation 2-4 model, construct a case base for coal mine gas explosion accidents according to the pre-acquired coal mine gas explosion accident samples.

[0104] The working condition early warning unit 402 is configured to perform early warning on the working conditions of accident-prone areas based on preset rules according to the data from multiple data sources collected in accident-prone areas, and determine the early warning level of the early warning working conditions.

[0105] The normalization unit 403 is configured to perform normalization processing on all data from multiple data sources in response to the early warning level reaching the preset alarm threshold to obtain normalized data.

[0106] The accident warning unit 404 is configured to perform gas explosion accident warning on accident-prone areas based on the normalized data, the association rules of coal mine gas explosion accidents, and the case base.

[0107] The gas explosion accident warning system based on data fusion provided by the embodiments of the present application can implement the steps and processes of the gas explosion accident warning method based on data fusion described in any of the above embodiments, and achieve the same technical effects, which will not be elaborated herein one by one.

[0108] The above are only the preferred embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A gas explosion accident early warning method based on data fusion, characterized in that, it includes: Step S101: Based on the Apriori algorithm, calculate the support, confidence, and lift between different accident characteristics of coal mine gas explosion accidents respectively to construct the association rules of the coal mine gas explosion accidents; And, Based on the accident causation 2-4 model, construct a case library of the coal mine gas explosion accidents according to the pre-acquired coal mine gas explosion accident samples; Wherein, according to the formula: ； Calculate accident characteristics and accident characteristics between the support and confidence and lift ; where is the probability of the accident characteristic occurring; is the probability of the accident characteristic occurring; is the probability of the accident characteristics occurring simultaneously; When the lift is concerned, there is no obvious positive correlation between accident characteristics and accident characteristics ; when the lift is concerned, accident characteristics and accident characteristics are independent of each other; when the lift is concerned, there is a strong correlation between accident characteristics and accident characteristics , and the larger the value of , the stronger the correlation between accident characteristics ; Step S102: Based on preset rules, according to the data from multiple data sources in the accident-prone area collected, give early warning to the working conditions of the accident-prone area and determine the early warning level of the early warning working conditions; wherein, the data from the data sources at least includes: sensor monitoring data; Step S103: In response to the early warning level reaching the preset alarm threshold, perform normalization processing on all the data from the multiple data sources to obtain normalized data; Step S104: Based on the normalized data, based on the association rules and case library of the coal mine gas explosion accidents, give early warning to the accident-prone area for gas explosion accidents.

2. The gas explosion accident early warning method based on data fusion according to claim 1, characterized in that, The data from the data sources also includes: multiple of video monitoring data and text data; Step S102 includes: Compare the sensor monitoring data, video monitoring data, and text data of the accident-prone area collected with the preset monitoring threshold to give early warning to the gas, flame, and ventilation working conditions of the accident-prone area; And / or, Based on the preset video monitoring algorithm, according to the video monitoring data of the accident-prone area collected, give early warning to the accidents and violations in the accident-prone area; And / or, Based on the pre-constructed keyword library, perform text recognition on the text data of the accident-prone area collected, obtain accident information, and give early warning to the risk factors of the accident-prone area.

3. The gas explosion accident early warning method based on data fusion according to claim 2, characterized in that, In step S102, Perform gray scale, difference, image enhancement, and binarization operations on the video monitoring data of the accident-prone area collected in sequence to determine the morphological changes of the dynamic targets in the accident-prone area, so as to give early warning to the accidents and violations in the accident-prone area.

4. The gas explosion accident early warning method based on data fusion according to claim 2, characterized in that, In step S102, Perform text vectorization on the text data to generate a word vector matrix of the text data; Based on the keyword library, according to the word vector matrix, perform text event detection on the text data, obtain the accident information, and give early warning to the risk factors of the accident-prone area.

5. The gas explosion accident early warning method based on data fusion according to claim 1, characterized in that, Step S104 includes: Determine multiple risk accident characteristics in the coal mine gas explosion accident according to the case base of the coal mine gas explosion accident; Based on the normalized data and the association rules of the coal mine gas explosion accident, determine whether the accident-prone area contains the risk accident characteristics to give a warning of the gas explosion accident in the accident-prone area.

6. A gas explosion accident warning system based on data fusion Characterized in that It includes: A database construction unit configured to calculate the support degree, confidence degree and lift degree between different accident characteristics of the coal mine gas explosion accident respectively based on the Apriori algorithm to construct the association rules of the coal mine gas explosion accident; and, based on the accident causation 2-4 model, construct the case base of the coal mine gas explosion accident according to the pre-acquired coal mine gas explosion accident samples; where, according to the formula: ； Calculate accident characteristics and accident characteristics between the support confidence and lift ; where is the probability of the accident characteristic occurring; is the probability of the accident characteristic occurring; is the probability of the accident characteristics occurring simultaneously; When the lift is concerned, there is no obvious positive correlation between accident feature and accident feature ; when the lift is concerned, accident feature and accident feature are independent of each other; when the lift is concerned, there is a strong correlation between accident feature and accident feature , and the larger the value of , the stronger the correlation between accident feature ; A working condition warning unit configured to give a warning of the working condition of the accident-prone area based on preset rules according to the data from multiple data sources collected in the accident-prone area, and determine the warning level of the warning working condition; where the data from the data source at least includes: sensor monitoring data; A normalization unit configured to perform normalization processing on all the data from multiple data sources in response to the warning level reaching a preset alarm threshold to obtain normalized data; An accident warning unit configured to give a warning of the gas explosion accident in the accident-prone area based on the normalized data, the association rules and the case base of the coal mine gas explosion accident.

Images