Coal mine fire simulation calibration method, device and storage medium based on machine learning

Abstract

The present invention provides a coal mine fire simulation calibration method, device and storage medium based on machine learning, which relates to the field of coal mine safety, including obtaining historical data of fire simulation and fire perception; then calculating the difference between the fire perception historical data and the fire simulation historical data; and then constructing a scene classifier based on the fire simulation historical data. Using the scene classifier, historical data are collected to produce a calibration data set to form a calibrator; the calibrator is then dynamically trained using real-time data of fire simulation and fire perception; finally, the dynamically trained calibrator is used to calibrate the fire advance simulation data to obtain the prediction results of the fire advance perception data. By calibrating the fire simulation through the above-mentioned method steps, a more accurate deduction of coal mine fires is achieved, which provides decision-making guidance for coal mine disaster prevention and disaster prevention, and is expected to bring important technical support to coal mine fire safety management.

Description

Technical Field

[0001] The present invention relates to the field of coal mine safety technology, and in particular to a coal mine fire simulation calibration method based on machine learning, a coal mine fire simulation calibration device, and a readable storage medium. Background Art

[0002] Coal mine fire prevention and control is a top priority for safe mining operations. Coal mine fires cause serious casualties and property losses, making them a top priority for safe mining. Existing coal mine safety monitoring systems, while equipped with sensors such as smoke and temperature, struggle to capture data during the catastrophic period due to the destructive power of coal mine fires. Consequently, it's difficult to predict fire conditions and the spread of high-temperature smoke within tunnels.

[0003] Therefore, technological breakthroughs are needed in coal mine fire simulation. However, due to the lack of authenticity verification criteria, the results of simulation cannot be directly applied in the field. How to combine simulation and safety monitoring data to effectively calibrate coal mine fire simulation results, realize coal mine fire prediction, and provide more professional decision-making support for mine fire prevention and control is an urgent problem that the industry needs to solve. Summary of the Invention

[0004] In view of the above problems, the present invention proposes a coal mine fire simulation calibration method based on machine learning, a coal mine fire simulation calibration device and a readable storage medium.

[0005] An embodiment of the present invention provides a coal mine fire simulation calibration method based on machine learning, the coal mine fire simulation calibration method comprising:

[0006] Obtain historical data of fire simulation and fire perception;

[0007] Calculate the difference between fire perception historical data and fire simulation historical data;

[0008] Building a scene classifier based on the fire simulation historical data;

[0009] Using the scene classifier and combining it with the historical data to create a calibration data set to form a calibrator;

[0010] Dynamically training the calibrator using real-time data from fire simulation and fire perception;

[0011] The fire advance simulation data is calibrated using a dynamic trained calibrator to obtain the prediction results of the fire advance perception data.

[0012] Optionally, obtain fire simulation and fire perception historical data, including:

[0013] Acquire the fire simulation historical data and the fire perception historical data from a database;

[0014] The fire simulation history data includes: simulation time and space history data and simulation working condition parameter history data;

[0015] The fire perception historical data includes: historical perception of smoke concentration, historical perception of carbon monoxide concentration, historical perception of oxygen concentration, historical perception of temperature, historical perception of visibility, and historical perception of wind speed;

[0016] The simulation time-space historical data includes: historical simulation smoke concentration, historical simulation carbon monoxide concentration, historical simulation oxygen concentration, historical simulation temperature, historical simulation visibility, and historical simulation wind speed;

[0017] The historical data of simulation working condition parameters include: historical fire source location, historical fire source heat release power, historical combustion material, historical tunnel type, and historical ventilation working condition.

[0018] Optionally, calculating the difference between the fire perception historical data and the fire simulation historical data includes:

[0019] Calculating the difference between the fire perception historical data and the fire simulation historical data based on the evolution time and space corresponding to the historical data;

[0020] Among them, the smoke concentration difference is the difference between the historical perceived smoke concentration and the historical simulated smoke concentration; the carbon monoxide concentration difference is the difference between the historical perceived carbon monoxide concentration and the historical simulated carbon monoxide concentration; the oxygen concentration difference is the difference between the historical perceived oxygen concentration and the historical simulated oxygen concentration; the temperature difference is the difference between the historical perceived temperature and the historical simulated temperature; the visibility difference is the difference between the historical perceived visibility and the historical simulated visibility; the wind speed difference is the difference between the historical perceived wind speed and the historical simulated wind speed.

[0021] Optionally, constructing a scene classifier based on the fire simulation historical data includes:

[0022] The scenario classifier is constructed using a data combination of demand parameters, wherein the demand parameters include: the historical fire source location, the historical fire source heat release power, the historical combustion material, the historical roadway type, and the historical ventilation condition;

[0023] Wherein, the scene classifier includes a decision tree and a clustering algorithm;

[0024] The decision tree is constructed by ranking the demand parameters using expert experience;

[0025] The clustering algorithm utilizes K-means clustering to automatically classify the demand parameters.

[0026] Optionally, using the scene classifier and combining the historical data to produce a calibration data set to form a calibrator includes:

[0027] constructing sample sets respectively using the classification results of the scene classifier;

[0028] The sample set includes an input part and an output part;

[0029] The input part includes: the historical simulated smoke concentration, the historical simulated carbon monoxide concentration, the historical simulated oxygen concentration, the historical simulated temperature, the historical simulated visibility, and the historical simulated wind speed;

[0030] The output part includes: the perceived smoke concentration difference, the perceived carbon monoxide concentration difference, the perceived oxygen concentration difference, the perceived temperature difference, the perceived visibility difference, and the perceived wind speed difference;

[0031] The input part and the output part are Z-score normalized according to items corresponding to the respective data to form the calibration data set, and then form the calibrator.

[0032] Optionally, dynamically training the calibrator using real-time data of fire simulation and fire perception includes:

[0033] Training a calibrator using the calibration data set formed after Z-score normalization;

[0034] Dynamically training the calibrator after accessing the real-time data, thereby updating the calibrator;

[0035] The dynamic training includes: updating the calibration data set and updating the training calibrator;

[0036] The updating of the calibration data set includes: performing Z-score normalization on the real-time data using the same mean and standard deviation as the historical sample set formed by the historical data, and then adding the real-time data to the calibration data set, and uniformly normalizing the real-time data with the historical sample set to update the calibration data set;

[0037] After updating the calibration dataset, updating the training calibrator includes: training the calibrator from scratch and pre-training the calibrator;

[0038] Among them, the calibrator is a deep neural network, including but not limited to a fully connected neural network, a convolutional neural network, a recurrent neural network, and an attention mechanism neural network.

[0039] Optionally, the fire advance simulation data is calibrated using a dynamically trained calibrator to obtain prediction results of the fire advance perception data, including:

[0040] Using the fire advance simulation data to input the scene classifier to obtain a calibrator category, the fire advance simulation data refers to spatiotemporal data that simulates beyond the current moment of fire evolution;

[0041] Based on the calibrator category, load the dynamic trained calibrator;

[0042] Using the fire advance simulation data as input to the loaded dynamically trained calibrator, the difference between the fire advance perception data and the fire advance simulation data is predicted in advance;

[0043] A sum operation is performed on the fire advance simulation data and the predicted difference to obtain a prediction result of the fire advance perception data.

[0044] An embodiment of the present invention provides a coal mine fire simulation calibration device based on machine learning, the coal mine fire simulation calibration device comprising:

[0045] A historical data acquisition module is used to acquire historical data of fire simulation and fire perception;

[0046] A historical data difference calculation module is used to calculate the difference between fire perception historical data and fire simulation historical data;

[0047] Constructing a classifier module, for constructing a scene classifier based on the fire simulation historical data;

[0048] Constructing a calibrator module for utilizing the scene classifier and collecting the historical data to produce a calibration data set to form a calibrator;

[0049] A training calibrator module, configured to dynamically train the calibrator using real-time data of fire simulation and fire perception;

[0050] The prediction module is used to calibrate the fire advance simulation data using the dynamic trained calibrator to obtain the prediction results of the fire advance perception data.

[0051] Optionally, the historical data acquisition module is specifically used to:

[0052] Acquire the fire simulation historical data and the fire perception historical data from a database;

[0053] The fire simulation history data includes: simulation time and space history data and simulation working condition parameter history data;

[0054] The fire perception historical data includes: historical perception of smoke concentration, historical perception of carbon monoxide concentration, historical perception of oxygen concentration, historical perception of temperature, historical perception of visibility, and historical perception of wind speed;

[0055] The simulation time-space historical data includes: historical simulation smoke concentration, historical simulation carbon monoxide concentration, historical simulation oxygen concentration, historical simulation temperature, historical simulation visibility, and historical simulation wind speed;

[0056] The historical data of simulation working condition parameters include: historical fire source location, historical fire source heat release power, historical combustion material, historical tunnel type, and historical ventilation working condition.

[0057] Optionally, the classifier building module is specifically used for:

[0058] The scenario classifier is constructed using a data combination of demand parameters, wherein the demand parameters include: the historical fire source location, the historical fire source heat release power, the historical combustion material, the historical roadway type, and the historical ventilation condition;

[0059] Wherein, the scene classifier includes a decision tree and a clustering algorithm;

[0060] The decision tree is constructed by ranking the demand parameters using expert experience;

[0061] The clustering algorithm utilizes K-means clustering to automatically classify the demand parameters.

[0062] Optionally, the building calibrator module is specifically used to:

[0063] constructing sample sets respectively using the classification results of the scene classifier;

[0064] The sample set includes an input part and an output part;

[0065] The input part includes: the historical simulated smoke concentration, the historical simulated carbon monoxide concentration, the historical simulated oxygen concentration, the historical simulated temperature, the historical simulated visibility, and the historical simulated wind speed;

[0066] The output part includes: the perceived smoke concentration difference, the perceived carbon monoxide concentration difference, the perceived oxygen concentration difference, the perceived temperature difference, the perceived visibility difference, and the perceived wind speed difference;

[0067] The input part and the output part are Z-score normalized according to items corresponding to the respective data to form the calibration data set, and then form the calibrator.

[0068] Optionally, the training calibrator module is specifically used to:

[0069] Training a calibrator using the calibration data set formed after Z-score normalization;

[0070] Dynamically training the calibrator after accessing the real-time data, thereby updating the calibrator;

[0071] The dynamic training includes: updating the calibration data set and updating the training calibrator;

[0072] The updating of the calibration data set includes: performing Z-score normalization on the real-time data using the same mean and standard deviation as the historical sample set formed by the historical data, and then adding the real-time data to the calibration data set, and uniformly normalizing the real-time data with the historical sample set to update the calibration data set;

[0073] After updating the calibration dataset, updating the training calibrator includes: training the calibrator from scratch and pre-training the calibrator;

[0074] Among them, the calibrator is a deep neural network, including but not limited to a fully connected neural network, a convolutional neural network, a recurrent neural network, and an attention mechanism neural network.

[0075] Optionally, the prediction module is specifically configured to:

[0076] Using the fire advance simulation data to input the scene classifier to obtain a calibrator category, the fire advance simulation data refers to spatiotemporal data that simulates beyond the current moment of fire evolution;

[0077] Based on the calibrator category, load the dynamic trained calibrator;

[0078] Using the fire advance simulation data as input to the loaded dynamically trained calibrator, the difference between the fire advance perception data and the fire advance simulation data is predicted in advance;

[0079] A sum operation is performed on the fire advance simulation data and the predicted difference to obtain a prediction result of the fire advance perception data.

[0080] An embodiment of the present invention provides a readable storage medium having a program stored thereon, which, when executed by a processor, implements the steps in any of the above-described coal mine fire simulation calibration methods based on machine learning.

[0081] The coal mine fire simulation calibration method based on machine learning provided by the present invention first obtains historical data of fire simulation and fire perception; then calculates the difference between the fire perception historical data and the fire simulation historical data; and then constructs a scene classifier based on the fire simulation historical data.

[0082] Using the scene classifier, historical data is collected to create a calibration data set to form a calibrator; then the calibrator is dynamically trained using real-time data from fire simulation and fire perception; finally, the dynamically trained calibrator is used to calibrate the fire advance simulation data to obtain the prediction results of the fire advance perception data.

[0083] This invention addresses the problem that traditional coal mine fire identification technology lacks disaster data and cannot deduce fire spread in real time. It creatively proposes a method for obtaining a large amount of fire prior data by combining fire simulation. Based on pre-set possible disaster scenarios, comprehensive fire simulation and real-life fire experiments on scale models are carried out. This method can provide a large amount of prior knowledge and data on fire spread patterns. Combined with this data, a classifier based on a machine learning algorithm can distinguish disaster scenarios to adapt to mines with different tunnel conditions. At the same time, a calibrator based on a machine learning algorithm is used to effectively calibrate the coal mine fire simulation results, achieving real-time fire simulation and deduction. The simulation results, combined with three-dimensional visualization, can provide more professional decision-making support for mine fire prevention and control. It has high practicality. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] Various other advantages and benefits will become apparent to those skilled in the art upon reading the detailed description of the preferred embodiment below. The accompanying drawings are for illustration purposes only and are not to be considered as limiting the present invention. The same reference symbols are used throughout the drawings to represent the same components. In the drawings:

[0085] Figure 1 is a flow chart of a coal mine fire simulation calibration method based on machine learning according to an embodiment of the present invention;

[0086] Figure 2 This is a block diagram of a coal mine fire simulation calibration and correction based on machine learning according to an embodiment of the present invention. DETAILED DESCRIPTION

[0087] In order to make the above-mentioned objects, features and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain the present invention, are only part of the embodiments of the present invention, not all of the embodiments, and are not intended to limit the present invention.

[0088] Reference Figure 1 , shows a flow chart of a coal mine fire simulation calibration method based on machine learning according to an embodiment of the present invention, the coal mine fire simulation calibration method comprising:

[0089] Step 101: Obtain historical data of fire simulation and fire perception.

[0090] The machine learning-based coal mine fire simulation calibration method proposed in this invention first requires acquiring historical data for both fire simulation and fire perception. This acquisition can be done in a variety of ways, with a preferred method including:

[0091] Obtain historical fire simulation data and fire perception data from the database. Fire simulation data includes historical simulation time and space data and historical simulation operating condition parameter data. Fire perception data includes historical smoke concentration (M1), historical carbon monoxide concentration (M2), historical oxygen concentration (M3), historical temperature (M4), historical visibility (M5), and historical wind speed (M6).

[0092] The simulation time-space historical data in the fire simulation historical data include: historical simulation smoke concentration S1, historical simulation carbon monoxide concentration S2, historical simulation oxygen concentration S3, historical simulation temperature S4, historical simulation visibility S5, and historical simulation wind speed S6.

[0093] The historical data of simulation working condition parameters in the fire simulation historical data include: historical fire source location P1, historical fire source heat release power P2, historical combustion material P3, historical tunnel category P4, and historical ventilation working condition P5.

[0094] Step 102: Calculate the difference between the fire perception historical data and the fire simulation historical data.

[0095] After obtaining the fire perception historical data and the fire simulation historical data, it is necessary to calculate the difference between the fire perception historical data and the fire simulation historical data. A better method for calculating the difference between the fire perception historical data and the fire simulation historical data includes:

[0096] Based on the evolution time and space corresponding to the historical data, the difference between the historical fire perception data and the historical fire simulation data is calculated. The smoke concentration difference is the difference between the historical perceived smoke concentration and the historical simulated smoke concentration; the carbon monoxide concentration difference is the difference between the historical perceived carbon monoxide concentration and the historical simulated carbon monoxide concentration; the oxygen concentration difference is the difference between the historical perceived oxygen concentration and the historical simulated oxygen concentration; the temperature difference is the difference between the historical perceived temperature and the historical simulated temperature; the visibility difference is the difference between the historical perceived visibility and the historical simulated visibility; and the wind speed difference is the difference between the historical perceived wind speed and the historical simulated wind speed.

[0097] That is: the difference in smoke concentration is: M1-S1, the difference in carbon monoxide concentration is: M2-S2, the difference in oxygen concentration is: M3-S3, the difference in temperature is: M4-S4, the difference in visibility is: M5-S5, and the difference in wind speed is: M6-S6.

[0098] Step 103: Construct a scene classifier based on the fire simulation historical data.

[0099] The scenario classifier needs to be constructed based on historical fire simulation data. Specifically, the scenario classifier is constructed using a combination of required parameters, including historical fire source location P1, historical fire source heat release power P2, historical combustion material P3, historical tunnel type P4, and historical ventilation conditions P5. In other words, the scenario classifier is constructed using a combination of data from historical fire source location P1, historical fire source heat release power P2, historical combustion material P3, historical tunnel type P4, and historical ventilation conditions P5.

[0100] Among them, the scenario classifier can specifically include: decision tree and clustering algorithm; the decision tree uses expert experience to sort the demand parameters to construct a decision tree; the clustering algorithm uses K-means clustering to automatically classify the demand parameters (each demand parameter can be regarded as a parameter group).

[0101] Step 104: Use the scene classifier and combine it with historical data to create a calibration data set to form a calibrator.

[0102] After the classifier is built, the scene classifier is used in combination with historical data to create a calibration data set, thereby forming a calibrator. A preferred method of using the scene classifier in combination with historical data to create a calibration data set to form a calibrator includes:

[0103] The classification results of the scene classifier are used to construct sample sets respectively; the sample sets include an input part and an output part.

[0104] The input part includes: historical perceived smoke concentration M1, historical perceived carbon monoxide concentration M2, historical perceived oxygen concentration M3, historical perceived temperature M4, historical perceived visibility M5, and historical perceived wind speed M6.

[0105] The output part includes: smoke concentration difference: M1-S1, carbon monoxide concentration difference: M2-S2, oxygen concentration difference: M3-S3, temperature difference: M4-S4, visibility difference: M5-S5, wind speed difference: M6-S6.

[0106] The input part: historical perceived smoke concentration M1, historical perceived carbon monoxide concentration M2, historical perceived oxygen concentration M3, historical perceived temperature M4, historical perceived visibility M5, historical perceived wind speed M6, and the output part: smoke concentration difference: M1-S1, carbon monoxide concentration difference: M2-S2, oxygen concentration difference: M3-S3, temperature difference: M4-S4, visibility difference: M5-S5, wind speed difference: M6-S6, are Z-score normalized according to the corresponding items of each data (i.e.: M1 item, M2 item, M3 item, M4 item, M5 item, M6 item, M1-S1 item, M2-S2 item, M3-S3 item, M4-S4 item, M5-S5 item, M6-S6 item) to form a calibration data set, and then form a calibrator.

[0107] Step 105: Dynamically train the calibrator using real-time data of fire simulation and fire perception.

[0108] After the calibrator is built, in order to accurately predict coal mine fires, it is necessary to dynamically train the calibrator using real-time data from fire simulation and fire perception. The specific method of dynamically training the calibrator using real-time data from fire simulation and fire perception includes:

[0109] First, the calibration data set formed after Z-score normalization in step 104 is used to train the calibrator. Then, the calibrator is dynamically trained after accessing real-time data, thereby updating the calibrator. This makes the calibration of the calibrator more accurate, laying a good foundation for the subsequent accurate prediction of coal mine fires.

[0110] Dynamic training includes updating the calibration dataset and updating the training calibrator. Specifically:

[0111] Updating the calibration dataset includes: using the same mean and standard deviation as the historical data to form the historical sample set, performing Z-score normalization on the real-time data and then adding it to the calibration dataset. After the real-time data is added to the historical sample set, the Z-score normalization is unified to update the calibration dataset. After updating the calibration dataset:

[0112] Updating the training calibrator includes: training the calibrator from scratch and pre-training the calibrator; among them, the calibrator is a deep neural network, including but not limited to fully connected neural networks, convolutional neural networks, recurrent neural networks, and attention mechanism neural networks.

[0113] Step 106: Use the dynamically trained calibrator to calibrate the fire advance simulation data to obtain the prediction results of the fire advance perception data.

[0114] After the dynamic training calibrator is used, the fire advance simulation data can be calibrated using the dynamically trained calibrator to obtain the prediction results of the fire advance perception data. The better methods include:

[0115] The fire advance simulation data is input into the scene classifier to obtain the calibrator category. The fire advance simulation data refers to the spatiotemporal data that simulates beyond the current moment of fire evolution. For example, if the current moment of fire evolution is 9:00 AM, then the fire advance simulation data can be the spatiotemporal data after 9:00 AM, such as the spatiotemporal data at 9:10 AM.

[0116] After obtaining the calibrator category, the dynamically trained calibrator is loaded based on the calibrator category; the fire advance simulation data is then used as the input for loading the dynamically trained calibrator to predict the difference between the fire advance perception data and the fire advance simulation data.

[0117] Finally, the difference between the fire advance simulation data and the prediction is summed to obtain the prediction result of the fire advance perception data.

[0118] Based on the above coal mine fire simulation calibration method, the embodiment of the present invention also proposes a coal mine fire simulation calibration device based on machine learning, referring to Figure 2 The block diagram of the coal mine fire simulation calibration device shown in FIG. 1 includes:

[0119] A historical data acquisition module 210 is used to acquire historical data of fire simulation and fire perception;

[0120] A historical data difference calculation module 220 is used to calculate the difference between the fire perception historical data and the fire simulation historical data;

[0121] A classifier construction module 230 is used to construct a scene classifier based on the fire simulation historical data;

[0122] Constructing a calibrator module 240 for utilizing the scene classifier and collecting the historical data to produce a calibration data set to form a calibrator;

[0123] A training calibrator module 250 is used to dynamically train the calibrator using real-time data of fire simulation and fire perception;

[0124] The prediction module 260 is used to calibrate the fire advance simulation data using the dynamic trained calibrator to obtain the prediction result of the fire advance perception data.

[0125] Optionally, the historical data acquisition module 210 is specifically configured to:

[0126] Acquire the fire simulation historical data and the fire perception historical data from a database;

[0127] The fire simulation history data includes: simulation time and space history data and simulation working condition parameter history data;

[0128] The fire perception historical data includes: historical perception of smoke concentration, historical perception of carbon monoxide concentration, historical perception of oxygen concentration, historical perception of temperature, historical perception of visibility, and historical perception of wind speed;

[0129] The simulation time-space historical data includes: historical simulation smoke concentration, historical simulation carbon monoxide concentration, historical simulation oxygen concentration, historical simulation temperature, historical simulation visibility, and historical simulation wind speed;

[0130] The historical data of simulation working condition parameters include: historical fire source location, historical fire source heat release power, historical combustion material, historical tunnel type, and historical ventilation working condition.

[0131] Optionally, the historical data difference calculation module 220 is specifically configured to:

[0132] Calculating the difference between the fire perception historical data and the fire simulation historical data based on the evolution time and space corresponding to the historical data;

[0133] Among them, the smoke concentration difference is the difference between the historical perceived smoke concentration and the historical simulated smoke concentration; the carbon monoxide concentration difference is the difference between the historical perceived carbon monoxide concentration and the historical simulated carbon monoxide concentration; the oxygen concentration difference is the difference between the historical perceived oxygen concentration and the historical simulated oxygen concentration; the temperature difference is the difference between the historical perceived temperature and the historical simulated temperature; the visibility difference is the difference between the historical perceived visibility and the historical simulated visibility; the wind speed difference is the difference between the historical perceived wind speed and the historical simulated wind speed.

[0134] Optionally, the classifier building module 230 is specifically configured to:

[0135] The scenario classifier is constructed using a data combination of demand parameters, wherein the demand parameters include: the historical fire source location, the historical fire source heat release power, the historical combustion material, the historical roadway type, and the historical ventilation condition;

[0136] Wherein, the scene classifier includes a decision tree and a clustering algorithm;

[0137] The decision tree is constructed by ranking the demand parameters using expert experience;

[0138] The clustering algorithm utilizes K-means clustering to automatically classify the demand parameters.

[0139] Optionally, the constructing calibrator module 240 is specifically configured to:

[0140] constructing sample sets respectively using the classification results of the scene classifier;

[0141] The sample set includes an input part and an output part;

[0142] The input part includes: the historical simulated smoke concentration, the historical simulated carbon monoxide concentration, the historical simulated oxygen concentration, the historical simulated temperature, the historical simulated visibility, and the historical simulated wind speed;

[0143] The output part includes: the perceived smoke concentration difference, the perceived carbon monoxide concentration difference, the perceived oxygen concentration difference, the perceived temperature difference, the perceived visibility difference, and the perceived wind speed difference;

[0144] The input part and the output part are Z-score normalized according to items corresponding to the respective data to form the calibration data set, and then form the calibrator.

[0145] Optionally, the training calibrator module 250 is specifically configured to:

[0146] Training a calibrator using the calibration data set formed after Z-score normalization;

[0147] Dynamically training the calibrator after accessing the real-time data, thereby updating the calibrator;

[0148] The dynamic training includes: updating the calibration data set and updating the training calibrator;

[0149] The updating of the calibration data set includes: performing Z-score normalization on the real-time data using the same mean and standard deviation as the historical sample set formed by the historical data, and then adding the real-time data to the calibration data set, and uniformly normalizing the real-time data with the historical sample set to update the calibration data set;

[0150] After updating the calibration dataset, updating the training calibrator includes: training the calibrator from scratch and pre-training the calibrator;

[0151] Among them, the calibrator is a deep neural network, including but not limited to a fully connected neural network, a convolutional neural network, a recurrent neural network, and an attention mechanism neural network.

[0152] Optionally, the prediction module 260 is specifically configured to:

[0153] Using the fire advance simulation data to input the scene classifier to obtain a calibrator category, the fire advance simulation data refers to spatiotemporal data that simulates beyond the current moment of fire evolution;

[0154] Based on the calibrator category, load the dynamic trained calibrator;

[0155] Using the fire advance simulation data as input to the loaded dynamically trained calibrator, the difference between the fire advance perception data and the fire advance simulation data is predicted in advance;

[0156] A sum operation is performed on the fire advance simulation data and the predicted difference to obtain a prediction result of the fire advance perception data.

[0157] Based on the above-mentioned coal mine fire simulation calibration method, an embodiment of the present invention further proposes a readable storage medium on which a program is stored. When the program is executed by a processor, the steps in the coal mine fire simulation calibration method based on machine learning as described in any one of steps 101 to 106 above are implemented.

[0158] In summary, the coal mine fire simulation calibration method based on machine learning of the present invention first obtains the historical data of fire simulation and fire perception; then calculates the difference between the fire perception historical data and the fire simulation historical data; and then constructs a scene classifier based on the fire simulation historical data.

[0159] Using the scene classifier, historical data is collected to create a calibration data set to form a calibrator; then the calibrator is dynamically trained using real-time data from fire simulation and fire perception; finally, the dynamically trained calibrator is used to calibrate the fire advance simulation data to obtain the prediction results of the fire advance perception data.

[0160] This invention addresses the problem that traditional coal mine fire identification technology lacks disaster data and cannot deduce fire spread in real time. It creatively proposes a method for obtaining a large amount of fire prior data by combining fire simulation. Based on pre-set possible disaster scenarios, comprehensive fire simulation and real-life fire experiments on scale models are carried out. This method can provide a large amount of prior knowledge and data on fire spread patterns. Combined with this data, a classifier based on a machine learning algorithm can distinguish disaster scenarios to adapt to mines with different tunnel conditions. At the same time, a calibrator based on a machine learning algorithm is used to effectively calibrate the coal mine fire simulation results, achieving real-time fire simulation and deduction. The simulation results, combined with three-dimensional visualization, can provide more professional decision-making support for mine fire prevention and control. It has high practicality.

[0161] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they become aware of the basic creative concepts. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the embodiments of the present invention.

[0162] Finally, it should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that includes a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or terminal device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or terminal device that includes the element.

[0163] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the present invention and the claims, all of which are protected by the present invention.

Claims

1. A coal mine fire simulation calibration method based on machine learning, characterized in that: The coal mine fire simulation calibration method comprises: Obtain historical data of fire simulation and fire perception; Calculate the difference between fire perception historical data and fire simulation historical data; Building a scene classifier based on the fire simulation historical data; Using the scene classifier and combining it with the historical data to create a calibration data set to form a calibrator; Dynamically training the calibrator using real-time data from fire simulation and fire perception; The fire advance simulation data is calibrated using a dynamic trained calibrator to obtain prediction results for the fire advance perception data. The method of calibrating the fire advance simulation data using the dynamically trained calibrator to obtain the prediction results of the fire advance perception data includes: Using the fire advance simulation data to input the scene classifier to obtain a calibrator category, the fire advance simulation data refers to spatiotemporal data that simulates beyond the current moment of fire evolution; Based on the calibrator category, load the dynamic trained calibrator; Using the fire advance simulation data as input to the loaded dynamically trained calibrator, the difference between the fire advance perception data and the fire advance simulation data is predicted in advance; performing a sum operation on the fire advance simulation data and the predicted difference to obtain a prediction result of the fire advance perception data; Wherein, the scene classifier is used to distinguish disaster scenes; The method of using the scene classifier and combining the historical data to produce a calibration data set to form a calibrator includes: A sample set is constructed using the classification results of the scene classifier; the sample set includes an input part and an output part; the input part includes the fire simulation history data; the output part includes the difference between the fire perception history data and the fire simulation history data; The input part and the output part are Z-score normalized according to items corresponding to the respective data to form the calibration data set, and then form the calibrator.

2. The coal mine fire simulation calibration method according to claim 1, characterized in that: Access fire simulation and fire perception historical data, including: Acquire the fire simulation historical data and the fire perception historical data from a database; The fire simulation history data includes: simulation time and space history data and simulation working condition parameter history data; The fire perception historical data includes: historical perception of smoke concentration, historical perception of carbon monoxide concentration, historical perception of oxygen concentration, historical perception of temperature, historical perception of visibility, and historical perception of wind speed; The simulation time-space historical data includes: historical simulation smoke concentration, historical simulation carbon monoxide concentration, historical simulation oxygen concentration, historical simulation temperature, historical simulation visibility, and historical simulation wind speed; The historical data of simulation working condition parameters include: historical fire source location, historical fire source heat release power, historical combustion material, historical tunnel type, and historical ventilation working condition.

3. The coal mine fire simulation calibration method according to claim 2, characterized in that: Calculate the difference between historical fire perception data and historical fire simulation data, including: Calculating the difference between the fire perception historical data and the fire simulation historical data based on the evolution time and space corresponding to the historical data; Among them, the smoke concentration difference is the difference between the historical perceived smoke concentration and the historical simulated smoke concentration; the carbon monoxide concentration difference is the difference between the historical perceived carbon monoxide concentration and the historical simulated carbon monoxide concentration; the oxygen concentration difference is the difference between the historical perceived oxygen concentration and the historical simulated oxygen concentration; the temperature difference is the difference between the historical perceived temperature and the historical simulated temperature; the visibility difference is the difference between the historical perceived visibility and the historical simulated visibility; the wind speed difference is the difference between the historical perceived wind speed and the historical simulated wind speed.

4. The coal mine fire simulation calibration method according to claim 2, characterized in that: Building a scene classifier based on the fire simulation historical data includes: The scenario classifier is constructed using a data combination of demand parameters, wherein the demand parameters include: the historical fire source location, the historical fire source heat release power, the historical combustion material, the historical roadway type, and the historical ventilation condition; Wherein, the scene classifier includes a decision tree and a clustering algorithm; The decision tree is constructed by ranking the demand parameters using expert experience; The clustering algorithm utilizes K-means clustering to automatically classify the demand parameters.

5. The coal mine fire simulation calibration method according to claim 3, characterized in that: The scene classifier is used in combination with the historical data to produce a calibration data set to form a calibrator, including: The input part includes: the historical simulated smoke concentration, the historical simulated carbon monoxide concentration, the historical simulated oxygen concentration, the historical simulated temperature, the historical simulated visibility, and the historical simulated wind speed; The output part includes: the smoke concentration difference, the carbon monoxide concentration difference, the oxygen concentration difference, the temperature difference, the visibility difference, and the wind speed difference.

6. The coal mine fire simulation calibration method according to claim 5, characterized in that: The calibrator is dynamically trained using real-time data of fire simulation and fire perception, including: Training a calibrator using the calibration data set formed after Z-score normalization; Dynamically training the calibrator after accessing the real-time data, thereby updating the calibrator; The dynamic training includes: updating the calibration data set and updating the training calibrator; The updating of the calibration data set includes: performing Z-score normalization on the real-time data using the same mean and standard deviation as the historical sample set formed by the historical data, and then adding the real-time data to the calibration data set, and uniformly normalizing the real-time data with the historical sample set to update the calibration data set; After updating the calibration dataset, updating the training calibrator includes: training the calibrator from scratch and pre-training the calibrator; Among them, the calibrator is a deep neural network, including but not limited to a fully connected neural network, a convolutional neural network, a recurrent neural network, and an attention mechanism neural network.

7. A coal mine fire simulation calibration device based on machine learning, characterized in that: The coal mine fire simulation calibration device comprises: A historical data acquisition module is used to acquire historical data of fire simulation and fire perception; A historical data difference calculation module is used to calculate the difference between fire perception historical data and fire simulation historical data; Constructing a classifier module, for constructing a scene classifier based on the fire simulation historical data; Constructing a calibrator module for utilizing the scene classifier and collecting the historical data to produce a calibration data set to form a calibrator; A training calibrator module, configured to dynamically train the calibrator using real-time data of fire simulation and fire perception; A prediction module is used to calibrate the fire advance simulation data using a dynamically trained calibrator to obtain prediction results of the fire advance perception data; The prediction module is specifically used for: Using the fire advance simulation data to input the scene classifier to obtain a calibrator category, the fire advance simulation data refers to spatiotemporal data that simulates beyond the current moment of fire evolution; Based on the calibrator category, load the dynamic trained calibrator; Using the fire advance simulation data as input to the loaded dynamically trained calibrator, the difference between the fire advance perception data and the fire advance simulation data is predicted in advance; performing a sum operation on the fire advance simulation data and the predicted difference to obtain a prediction result of the fire advance perception data; Wherein, the scene classifier is used to distinguish disaster scenes; The construction calibrator module is specifically used for: A sample set is constructed using the classification results of the scene classifier; the sample set includes an input part and an output part; the input part includes the fire simulation history data; the output part includes the difference between the fire perception history data and the fire simulation history data; The input part and the output part are Z-score normalized according to items corresponding to the respective data to form the calibration data set, and then form the calibrator.

8. The coal mine fire simulation calibration device according to claim 7, characterized in that: The module for obtaining historical data is specifically used for: Acquire the fire simulation historical data and the fire perception historical data from a database; The fire simulation history data includes: simulation time and space history data and simulation working condition parameter history data; The fire perception historical data includes: historical perception of smoke concentration, historical perception of carbon monoxide concentration, historical perception of oxygen concentration, historical perception of temperature, historical perception of visibility, and historical perception of wind speed; The simulation time-space historical data includes: historical simulation smoke concentration, historical simulation carbon monoxide concentration, historical simulation oxygen concentration, historical simulation temperature, historical simulation visibility, and historical simulation wind speed; The historical data of simulation working condition parameters include: historical fire source location, historical fire source heat release power, historical combustion material, historical tunnel type, and historical ventilation working condition.

9. A readable storage medium having a program stored thereon, characterized in that: When the program is executed by a processor, the steps in the coal mine fire simulation calibration method based on machine learning are implemented as described in any one of claims 1 to 6.

Images