A method and system for testing for the odor of coal self-ignition

Abstract

A coal spontaneous combustion odor test method and system, steps include: S1. Pretreatment of raw coal sample, the target coal sample after treatment is filled in a sealed container; S2. Dry air is continuously introduced into the container, and the coal spontaneous combustion process is simulated by using a segmented heating mode to generate a coal spontaneous combustion odor; S3. The odor samples released by the coal spontaneous combustion are collected and stored in a preset temperature range; S4. The collected odor samples are detected by an electronic nose analysis, and the response signal of the gas sensitive sensor to the odor samples is obtained; S5. The characteristic value of the response signal is extracted, and the mapping relationship between the characteristic value and the coal spontaneous combustion temperature stage is established; S6. A learning model is trained based on the characteristic value to construct a prediction model, and automatic classification and prediction of the coal spontaneous combustion stage are realized. It is suitable for coal spontaneous combustion low temperature oxidation odor detection, by focusing on two specific marker gases, acetaldehyde and benzene, and combining segmented heating and machine learning model, the early stage of coal spontaneous combustion is quickly, automatically identified and warned.

Description

Technical Field

[0001] This application belongs to the field of coal mine odor detection technology, and in particular relates to a method and system for testing the odor of spontaneous combustion of coal, which is used to achieve early monitoring and warning of spontaneous combustion of coal through electronic nose and machine learning technology. Background Technology

[0002] Coal plays a vital role in the global energy structure and is a primary fuel source for industry and power generation in many regions. Safety is always a top priority in coal mining and utilization. Mine fires are one of the main types of disasters threatening coal mine safety, with spontaneous combustion of coal being a common cause. Statistics show that most mine fires are related to spontaneous combustion of coal, posing a serious challenge to normal mine production and personnel safety. Therefore, effectively predicting and preventing spontaneous combustion of coal is of great significance for ensuring the safe operation of coal mines.

[0003] Currently, the main methods for predicting coal spontaneous combustion include indicator gas analysis and temperature monitoring. Indicator gas analysis determines the spontaneous combustion trend by detecting changes in the concentration of specific gases (such as carbon monoxide and ethylene) released during the combustion process; this method is widely used both domestically and internationally. Temperature monitoring directly judges the trend based on the heat generated during the oxidation and heating process of the coal. However, due to the poor thermal conductivity of coal and the fact that spontaneous combustion often occurs in hidden areas such as goafs, accurate temperature measurement is often difficult to achieve in practice.

[0004] Furthermore, odor analysis is increasingly being applied to coal spontaneous combustion prediction. Its basic principle is that coal releases various odors during the low-temperature oxidation stage, and the odor components and concentrations produced at different spontaneous combustion stages exhibit characteristic differences. By detecting this odor information, the spontaneous combustion state of coal can be identified and judged. However, traditional odor detection methods have limitations in terms of testing speed and cost. In recent years, with the development of sensor technology, odor detection methods based on electronic noses have gradually begun to be applied in fields such as environmental protection and food testing. This technology provides a new technical path for the field of coal spontaneous combustion prediction.

[0005] Against this backdrop, it is necessary to construct an experimental testing system for coal spontaneous combustion odor based on an electronic nose. This system would enable accurate collection and analysis of odor components released by different types of coal at different temperature stages. Furthermore, by integrating machine learning methods, an intelligent identification and early warning mechanism for the coal spontaneous combustion stage could be established. This would provide a faster, simpler, and more practical technical means for monitoring and early warning of coal mine spontaneous combustion. Summary of the Invention

[0006] This application provides a method and system for testing the smell of spontaneous combustion of coal, and proposes to use an odor analysis method based on electronic nose technology to provide rapid early warning of early spontaneous combustion of coal in coal mines.

[0007] To solve at least one of the above-mentioned technical problems, the technical solution adopted in this application is:

[0008] A method for testing the spontaneous combustion odor of coal, comprising the following steps:

[0009] S1. Pre-treat the original coal sample, and then fill the treated target coal sample into a sealed container;

[0010] S2. Dry air is continuously introduced into the container, and a staged heating method is used to simulate the spontaneous combustion process of coal to generate the smell of spontaneous combustion of coal;

[0011] S3. Collect and store the odor samples released by spontaneous combustion of coal within a preset temperature range;

[0012] S4. Analyze the collected odor samples using an electronic nose to obtain the response signal of the gas sensor to the odor samples;

[0013] S5. Extract the feature values ​​of the response signal and establish the mapping relationship between the feature values ​​and the coal spontaneous combustion temperature stage;

[0014] S6. Based on feature value training, a learning model is constructed to build a prediction model, thereby achieving automatic classification and prediction of the coal spontaneous combustion stage.

[0015] Furthermore, S1 specifically includes:

[0016] The original coal sample was crushed into a target coal sample with a particle size of 0.2-4.5 mm;

[0017] The target coal sample was placed in a drying oven at 40-50℃ to dehydrate until constant weight.

[0018] Weigh a quantitative amount of target coal sample and fill it into a sealed container, then cover the target coal sample with an air-permeable asbestos layer.

[0019] Furthermore, in step S2, dry air at a flow rate of 50-150 ml / min is continuously introduced into the container; the segmented heating includes:

[0020] Within the range of [30℃, 100℃], the heating rate is 0.5℃ / min;

[0021] Within the range of (100℃, 200℃), the heating rate is 1℃ / min.

[0022] Furthermore, in S3, the preset temperature range is based on three characteristic temperature ranges divided into the coal spontaneous combustion stage, namely the early, middle and late stages; and odor samples released by coal spontaneous combustion are sampled in each of these three characteristic temperature ranges; when collecting odor samples, gas sampling bags made of inert material are used to store the collected odor samples, and the sampling time interval is once every 5 minutes, while the temperature is recorded.

[0023] The early temperature ranges from 30 to 60°C, the middle temperature ranges from 60 to 130°C, and the late temperature ranges from >130°C.

[0024] Furthermore, step S4 includes:

[0025] Turn on the electronic nose to preheat, introduce air to wash the electronic nose, and stop washing the air after the resistance values ​​of all the gas-sensitive sensors in the electronic nose have stabilized.

[0026] Odor samples at different temperatures extracted from a gas sampling bag using a syringe are injected into the sampling port of the electronic nose;

[0027] Record the odor response signal of the odor sample measured in the electronic nose;

[0028] Repeat the above steps, clean the samples, and then perform sampling and testing to complete the detection of odor samples at different temperatures;

[0029] The waste gas must be vented after each test; the sampling time is set to 70-80 seconds, and the gas washing time is set to 90-100 seconds.

[0030] Furthermore, S5 includes:

[0031] Feature extraction was performed on the odor response signals of odor samples at different temperatures, and the stable resistance value in the odor response signal was used as the feature value R of the odor response intensity.

[0032] Obtain the characteristic value R of the response intensity of the gas sensor to the odor sample at different temperatures;

[0033] The characteristic value R captured by the acetaldehyde sensor and the benzene sensor in the gas sensor is curve-fitted with temperature to obtain the response intensity-temperature fitting curve.

[0034] Furthermore, the process of extracting the odor signal response intensity feature value R is as follows:

[0035] The response characteristics of gas sensors to odor samples at different temperatures were analyzed.

[0036] Take the resistance value of the sensor in a stable state before it comes into contact with the odor sample, and use the average resistance in the 30-35s time period during its detection cycle as the baseline resistance Ra.

[0037] The resistance value of the sensor when the response reaches a stable state after contact with the odor sample is taken, and the average resistance during the 130-135s period of its detection cycle is taken as the response resistance Rg.

[0038] The response intensity characteristic value is R = Ra - Rg, ​​with the unit being Ω.

[0039] Furthermore, step S6 includes:

[0040] Using the response intensity R as a feature value, a predictive model is constructed by training the data through a computer model learning algorithm.

[0041] Based on the prediction model, the early, middle and late stages of coal spontaneous combustion are classified and predicted.

[0042] Automatic early warning processing is performed after classification and prediction;

[0043] The prediction model is constructed by selecting at least one learning algorithm from principal component analysis-support vector machine, support vector machine, random forest, and artificial neural network; 80% of the feature values ​​are used as the training set and 20% of the feature values ​​are used as the test set.

[0044] Furthermore, the early warning process includes:

[0045] If the prediction indicates that the coal is in the early stage of spontaneous combustion, it is determined that the risk of spontaneous combustion is low or that it is in the initial oxidation stage, and further monitoring and preventive measures need to be taken.

[0046] If the prediction indicates that the spontaneous combustion of coal is in its middle stage, then it is determined that spontaneous combustion has occurred and entered the development stage, and corresponding prevention and control measures need to be taken.

[0047] If the prediction indicates that the fire is in the later stages of spontaneous combustion of coal, then the spontaneous combustion phenomenon is considered severe, the fire risk is extremely high, and comprehensive measures need to be taken to prevent spontaneous combustion fires.

[0048] A coal spontaneous combustion odor testing system, used to implement the method described above, includes:

[0049] Air supply unit: used to introduce dry air into the container, comprising a dry air cylinder connected to the container via a pipe;

[0050] Simulation unit: used to simulate spontaneous combustion of coal, including simulation device, which contains the container, air heating pipe, heater for heating the container, and operation interface;

[0051] Acquisition unit: Acquires and stores the odor samples released by spontaneous combustion of coal, including a gas sample bag, which is connected to the container and the electronic nose respectively;

[0052] Analysis unit: Analyzes and detects the collected odor samples, including the electronic nose and air pump, wherein the electronic nose is communicatively connected to a computer terminal.

[0053] The method and system for detecting the odor of spontaneous combustion of coal designed in this application are particularly suitable for odor detection during the low-temperature oxidation process of spontaneous combustion of coal. By using an electronic nose to focus on two specific characteristic odors, acetaldehyde and benzene, and combining segmented heating and machine learning models, rapid and automatic identification and early warning of the early stage of spontaneous combustion of coal can be achieved.

[0054] This application can detect the odor components of different coal types released at different temperature stages during spontaneous combustion and obtain their concentration changes; and make early judgments on coal spontaneous combustion based on odor characteristics.

[0055] The entire testing system has the advantages of compact structure, high detection efficiency and low cost. By simulating the real coal spontaneous combustion process through programmed temperature rise experiments, combined with classification models such as PCA-SVM, it effectively improves the accuracy and reliability of stage identification, providing a new technical means for coal spontaneous combustion early warning in the field of coal mine safety. Attached Figure Description

[0056] Figure 1 This is a flowchart of a method for testing the spontaneous combustion odor of coal according to this application;

[0057] Figure 2 This is a simplified structural diagram of the test system in this application;

[0058] Figure 3 This is a simplified structural diagram of the electronic nose in this application;

[0059] Figure 4 This is a graph showing the response intensity of acetaldehyde and benzene compounds measured in lignite in this application, along with temperature variation curves.

[0060] Figure 5 This is a 2D diagram showing the results of principal component analysis (PCA) modeling the odor characteristics of coal during the spontaneous combustion stage in this application.

[0061] Figure 6 This is a 3D diagram showing the results of principal component analysis (PCA) modeling the odor characteristics of coal during the spontaneous combustion stage in this application.

[0062] Figure 7 This is the confusion matrix obtained using the Principal Component Analysis-Support Vector Machine (PCA-SVM) model in this application;

[0063] Figure 8 This is the confusion matrix diagram obtained using the Support Vector Machine (SVM) model in this application;

[0064] Figure 9 This is a confusion matrix diagram obtained using the Random Forest (RF) model in this application;

[0065] Figure 10 This is a confusion matrix diagram obtained using an artificial neural network (ANN) model in this application.

[0066] In the picture:

[0067] 10. Gas supply unit 11. Dry air cylinder 12. Pressure reducing valve 13. Pressure regulating valve 14. Flow control valve 20. Simulation Unit 21. Container 22. Asbestos layer 23. Target coal sample 24. Temperature sensor 25. Air heating element 26. Heater 27. User Interface 28. Temperature Parameter Table 30. Acquisition Unit 31. Gas sample bag 32. Syringe 40. Analysis Unit 41. Electronic nose 411. Air chamber 412. Air Inlet 413. Air outlet 414. Gas sensor 415. Circuit board 416. Microcontroller 417. Circuit Interface 418. Cable Outlet 42. Shut-off valve one 43. Air pump 44. Stop Valve II 50. Computer terminal Detailed Implementation

[0068] The present application will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0069] This embodiment proposes a method for testing the spontaneous combustion smell of coal, such as... Figure 1 As shown, it is applicable to, for example Figure 2 In the test system shown, the steps include:

[0070] S1. Pre-treat the original coal sample and fill the treated target coal sample into a sealed container.

[0071] Specifically, it includes:

[0072] The original coal sample was crushed and then sieved into powdered target coal sample 23 with a particle size of 0.2-4.5 mm. Target coal sample 23 was then divided into several portions according to the experimental purpose; specifically, it was divided into no fewer than four experimental coal samples, each with a mass of no less than 150 g. The coal separation standard referred to the cone-pile quartering method in GB474-2008, section 7.4.4. Target coal sample 23 was weighed using an electronic scale, and then placed in a drying oven at 40-50℃ to dehydrate until constant weight.

[0073] Container 21 is a dedicated coal sample container. A measured amount of target coal sample 23 is weighed and distributed onto the asbestos layer 22 within the sealed container 21. A layer of highly permeable asbestos 22 is also placed over the target coal sample 23 to ensure even odor distribution. The asbestos layer 22 facilitates uniform airflow through the target coal sample, better simulating the underground coal spontaneous combustion environment and providing a reliable data basis for early warning. Simultaneously, a temperature sensor 24 is installed at the location of the target coal sample 23 to monitor the coal spontaneous combustion temperature. The temperature sensor 24 is directly connected to the controller in the simulation unit 20. The real-time data monitored by the temperature sensor 24 is displayed on the operating interface 27 and can also be transmitted to the computer terminal 50 via electrical signals.

[0074] The operation interface 27 serves as an input and display window, allowing personnel to call and execute standardized heating programs stored in the temperature parameter table 28, thereby achieving high-precision, repeatable, and automated simulation of the coal spontaneous combustion process to ensure the scientific rigor, accuracy, and ease of operation of the test.

[0075] This step of pre-treating the original coal sample ensures the homogeneity and representativeness of the target coal sample, reduces interference from moisture and other impurities on the detection of odors released during coal spontaneous combustion, and makes the odor signal clearer. Simultaneously, particle size control and drying of the target coal sample make the experimental results more reliable and facilitate comparisons between different experiments to simulate real-world conditions.

[0076] S2. Dry air is continuously introduced into the container, and a staged heating method is used to simulate the spontaneous combustion process of coal to generate the smell of spontaneous combustion of coal.

[0077] S2.1 Continuously introduce dry air into the container.

[0078] The dry air cylinder 11 in the gas supply unit 10 is connected to the sealed container 21 containing the target coal sample 23. From one side of the dry air cylinder 11, the flow rate of dry air is precisely controlled and stabilized within the range of 50-150 ml / min through the pressure reducing valve 12, the pressure regulating valve 13, and the flow regulating valve 14 in sequence, so as to continuously supply dry air to the container 21 to ensure that the airflow is continuous and stable throughout the entire programmed heating process.

[0079] The preferred flow rate of the dry air is 100 ml / min, which includes, but is not limited to, 50 ml / min, 55 ml / min, 60 ml / min, 65 ml / min, 70 ml / min, 75 ml / min, 80 ml / min, 85 ml / min, 90 ml / min, 95 ml / min, 105 ml / min, 110 ml / min, 115 ml / min, 120 ml / min, 125 ml / min, 130 ml / min, 135 ml / min, 140 ml / min, 145 ml / min, and 150 ml / min.

[0080] Spontaneous combustion of coal is essentially a low-temperature oxidation reaction between coal and oxygen, and oxygen in dry air is an indispensable reactant in this chemical reaction. By simulating the oxidation environment of coal in contact with air underground in container 21, a prerequisite for spontaneous combustion of coal can be achieved. The continuous airflow can promptly carry the odor molecules generated by the target coal sample 23 during the oxidation process out of container 21, creating conditions for subsequent odor collection and analysis.

[0081] Using dry air can avoid the influence of water vapor in the air on the humidity of the coal sample, and also prevent water vapor from interfering with or damaging the gas sensor in the electronic nose 41, so that the odor produced by the reaction can be effectively and continuously delivered to the sampling unit, thereby ensuring the accuracy and repeatability of the experimental results.

[0082] S2.2 adopts a segmented heating method for programmed heating.

[0083] The simulation unit 20 includes an integrated simulation device, which includes a container 21, an air heating pipe 25 connected to a dry air bottle 11, a heater 26 for heating the container 21, and an operating interface 27. The heater 26 directly heats the container 21 containing the target coal sample 23 to simulate the spontaneous combustion process of coal.

[0084] By connecting the pre-set heating program in the temperature parameter table 28 through the operation interface 27, the container 21 is heated in stages, which includes two phases:

[0085] The first stage is the low-temperature oxidation stage, with a temperature range of [30℃, 100℃], that is, heating from 30℃ to 100℃. The heating rate in this stage is relatively slow, and it is set to 0.5℃ / min.

[0086] The second stage is the accelerated oxidation stage, with a temperature range of (100℃, 200℃), i.e., the temperature is increased from 100℃ to 200℃. In this stage, the heating rate is accelerated, and the heating rate is set to 1℃ / min.

[0087] During the segmented heating process, odor samples are collected and temperature data is recorded simultaneously when each preset specific temperature point is reached.

[0088] The spontaneous combustion of coal underground is a process of slow heat accumulation and gradually accelerating oxidation. By staged heating of the target coal sample 23 in container 21 within simulation unit 20, especially a stepped heating mode with slow initial heating followed by rapid heating, the spontaneous combustion process of coal can be recreated, avoiding the distortion of the reaction caused by excessively rapid heating under laboratory conditions. The purpose is to collect odor samples at different temperatures during the spontaneous combustion process of coal, namely acetaldehyde and benzene compounds, which begin to be released or undergo abrupt concentration changes at specific temperature thresholds. Slow, staged heating allows the acquisition unit 30 to collect odor samples at precise temperature points, thereby establishing a correspondence between temperature and odor characteristics.

[0089] Odor samples are collected at specific temperature points to obtain the reaction products at that temperature, making the data more representative. Programmed temperature control avoids the risk of temperature runaway, allowing the entire dangerous coal spontaneous combustion simulation process to be conducted in a safe experimental environment.

[0090] The procedure for this step is as follows:

[0091] The operator selects a heating program from a stored list of programs via the user interface 27. In this embodiment, a lignite sample is selected, along with the preset standard heating program from the temperature parameter table 28. The start button on the user interface 27 is then pressed. The heating control system reads the selected parameter table and begins operating according to its instructions.

[0092] First, the heater 26 is controlled to heat the container 21 from 30℃ at a rate of 0.5℃ / min. The temperature sensor 24 monitors the internal temperature of the container 21 in real time and feeds the data back to the control system. Simultaneously, the control system displays real-time data, such as the current temperature and heating rate, on the operation interface 27 and compares it with the preset values ​​in the temperature parameter table 28. The system then dynamically adjusts the power of the heater 26 to ensure that the actual heating curve matches the preset curve. When 100℃ is reached, the system automatically switches the heating rate to 1.0℃ / min according to the instructions in the parameter table and continues heating until the experiment is completed.

[0093] By simulating the complete ignition process of underground coal samples from low-temperature oxidation to violent oxidation in simulation unit 20, and also simulating the collection of target coal samples 23 at different stages of spontaneous combustion, including early, middle and late stages, corresponding characteristic acetaldehyde and benzene samples, a solid foundation is laid for the next step of accurately collecting odor samples at different temperature points.

[0094] In this step, the coal spontaneous combustion process is reproduced in the laboratory through two key steps: continuous and stable dry air flow and segmented programmed heating. The aim is to generate a series of characteristic odor samples that correspond to each stage of real coal spontaneous combustion and can be used for electronic nose analysis, thereby providing high-quality and highly correlated training and validation data for the odor recognition-based early warning model of coal spontaneous combustion.

[0095] S3. Collect and store the odor samples released by spontaneous combustion of coal within a preset temperature range.

[0096] This step aims to simulate the spontaneous combustion of coal by capturing and preserving representative odor samples at key temperature points, providing an input source for subsequent analysis using the electronic nose 41 in analysis unit 40. Specifically, it includes the following steps:

[0097] S3.1 Preset temperature range and sampling temperature point

[0098] Based on the spontaneous combustion process of coal, the preset temperature range is divided into three stages: early, middle, and late. Correspondingly, the early temperature is 30-60℃, that is, greater than 30℃ and less than or equal to 60℃; the middle temperature is 60-130℃, that is, greater than 60℃ and less than or equal to 130℃; and the late temperature is >130℃.

[0099] Within a preset temperature range, specific sampling temperature points are set, such as every 10℃ or 20℃ as a sampling point, and the sampling time interval is determined to be once every 5 minutes, while recording the temperature. This ensures that the collected odor samples cover the entire process of coal spontaneous combustion, especially key transition temperature points, such as 40℃ when acetaldehyde begins to appear and 110℃ when benzene compounds begin to appear. Systematic sampling at different stages can capture the complete trajectory of odor characteristics as they evolve with temperature, avoiding the omission of important information.

[0100] S3.2 Setting the Data Acquisition Timing

[0101] When the temperature rises to a preset sampling temperature point, the system monitors the reading of the temperature sensor 24 to confirm that the temperature inside the container 21 has reached the preset temperature value of the sampling point. The system then immediately collects the odor sample released at this time. The composition and concentration of the odor sample can truly reflect the characteristics of the temperature point and are the most reliable and representative reaction products at that temperature point, greatly improving the accuracy and reliability of the experimental data.

[0102] S3.3 Collect and store odor samples released from spontaneous combustion of coal.

[0103] Odor samples were collected from spontaneous combustion of coal within these three characteristic temperature ranges. The collected odor samples were stored in a gas sampling bag 31 made of inert material. The inlet of the gas sampling bag 31 was connected to the outlet of the container 21 via a flexible tube. The valve of the gas sampling bag 31 was opened, and a continuous flow of dry air was used to fill the bag with the odor sample from the outlet of the container 21. After filling to a certain volume (100ml), the valve was quickly closed, the connection was disconnected, and the gas sampling bag 31 was ensured to be well sealed. The volume of the gas sampling bag 31 (100ml) was larger than the volume of the odor sample injected into the electronic nose 41 each time (50ml).

[0104] Simultaneously, the sampling number, corresponding collection temperature, and collection time are clearly marked on the gas sample bag 31. These labels ensure that each odor sample accurately corresponds to the temperature and time information during the experiment. One experiment can collect 25 odor samples, requiring four repetitions to obtain a total of 100 odor samples and 100 feature values. These 100 feature values ​​are used for model training, with 80% of the feature values ​​serving as the training set and 20% as the test set.

[0105] In this embodiment, the use of an inert gas sample bag 31 effectively prevents the odor sample from adsorbing or chemically reacting with the bag wall, which is especially crucial for low-concentration characteristic odor samples, ensuring the integrity of the sample before analysis. Furthermore, the gas sample bag 31 is easily connected to the sampling port of the syringe 32 and the electronic nose 41, enabling offline and flexible analysis of odor samples.

[0106] S3.4 Synchronous Data Recording

[0107] While collecting each odor sample, the actual temperature value displayed by the temperature sensor 24 and the precise collection time are recorded. The purpose is to establish an accurate mapping between the sensor response intensity and temperature, that is, to establish a curve of sensor response intensity-temperature, and ultimately to determine the spontaneous combustion stage of coal by smell.

[0108] S4. Analyze the collected odor samples using an electronic nose to obtain the response signals of acetaldehyde and benzene in the odor samples.

[0109] S4.1 Electronic Nose Initialization

[0110] Turn on the electronic nose 41 and allow its internal gas sensors and circuitry to warm up for a period of time. When the resistance of all gas sensors reaches a stable operating state, all gas sensors will be in their optimal and most sensitive operating temperature and environment.

[0111] Figure 3 As shown, this is a simplified structural diagram of the electronic nose 41. The electronic nose 41 includes a cylindrical air chamber 411, which is a sealed structure. The air chamber 411 has an air inlet 412 and an air outlet 413 at its two ends, respectively. One end of the air inlet 412 is equipped with a shut-off valve 42 connected to it, and directly connected to the syringe 32 through the shut-off valve 42. The other end of the air outlet 413 is equipped with a shut-off valve 44 connected to it, and directly connected to the air pump 43 through the shut-off valve 44.

[0112] The air chamber 411 contains a circuit board 415 for mounting all gas sensors 414, a microcontroller 416 for controlling the gas sensors 414, a circuit interface 417 for electrical connection to the computer terminal 50, and a cable outlet 418 for wires to pass through. The wires connecting the circuit interface 417 to the computer terminal 50 and the wires connecting the circuit board 415 to the external power supply are both routed through the cable outlet 418 and are sealed.

[0113] The gas sensor 414 is a sensor array consisting of 10 different types of sensors. All sensors, the microcontroller 416, and the circuit interface are mounted on a circuit board. The circuit board is fixed inside the air chamber 411 with four screws, and its fixing position is only required to not affect the airflow transmission.

[0114] The specific models of these 10 gas sensors 414 are: MOS-C1, MOS-C3, MOS-C4, MOS-D1, MOS-D2, MOS-D3, MOS-D4, MOS-D5, MOS-D6, and MOS-1; their corresponding functions and accuracy ranges are shown in Table 1.

[0115] Table 1. Models, functions, and accuracy range of gas sensors

[0116] Serial Number Sensor model Function Accuracy range / ppm 1 MOS-C1 Response to organic alcohols 1-100ppm 2 MOS-C3 Response to benzene 1-500ppm 3 MOS-C4 Response to ketones 1-200ppm 4 MOS-D1 Response to alkanes 1-50ppm 5 MOS-D2 Response to amines 1-30ppm 6 MOS-D3 VOC response 10-1000ppm 7 MOS-D4 <![CDATA[Response to NH3]]> 0.01-20ppm 8 MOS-D5 <![CDATA[Response to SO2]]> 0.1-1000ppm 9 MOS-D6 <![CDATA[Response to H2S]]> 20-10000ppm 10 MOS-1 Response to acetaldehyde 1-500ppm

[0117] For the gas sensor 414, since the acetaldehyde sensor and benzene sensor have high correlation and sensitivity to changes in coal spontaneous combustion temperature, this embodiment mainly focuses on monitoring the response intensity of the acetaldehyde sensor and benzene sensor to reflect the coal spontaneous combustion odor test.

[0118] During preheating, open the shut-off valve 44 connected to the air pump 43 to introduce clean, dry air into the detection chamber of the electronic nose 41 to purge the electronic nose 41. Continue purging with dry air until the resistance readings of all gas sensors in the electronic nose 41 stabilize, then stop purging. The purpose is to remove any residual odor molecules from the previous test from the detection chamber and the surfaces of each gas sensor 414 within the electronic nose 41, preventing cross-contamination. The electronic nose 41 is a commonly used analytical device in this field; its structure and operating principle are omitted here.

[0119] In clean air, the resistance value of the gas sensor 41 will stabilize at a reference level, which is recorded as the baseline resistance Ra, and serves as the reference point for all subsequent calculations.

[0120] S4.2 Sample Injection and Detection

[0121] Using syringe 32, extract a certain volume, such as 50 ml, of odor sample from the gas sample bag 31 prepared in step S3 at a specific temperature; open the shut-off valve 42, then insert the needle of syringe 32 into the sampling port of the electronic nose 41, and slowly and evenly inject the odor sample into the detection gas chamber of the electronic nose 41. After injection, immediately close the sampling port shut-off valve 42 to seal the odor sample in the detection gas chamber, ensuring full contact with the gas-sensitive sensor array.

[0122] The enclosed environment ensures that odor molecules have enough time to diffuse to the sensor surface, enabling the gas sensor to respond characteristically to the target odor and thus generate a reliable odor response signal.

[0123] S4.3 Signal Recording and Exhaust Gas Cleaning

[0124] Within the preset sampling time of 70-80 seconds, the data acquisition system of the electronic nose 41 will continuously record the resistance changes of all sensors, especially the acetaldehyde sensor and the benzene sensor, at a fixed frequency of 1 time / second, forming a response curve of the odor response signal.

[0125] After each test is completed, the shut-off valve 44 is opened simultaneously, and the air pump 43 is started. A large amount of clean air is used to thoroughly purge the waste gas from the test chamber and discharge it into the air through the air pump 43. The air pump 43 can either extract waste gas from the test chamber in the electronic nose 41 or fill the test chamber with clean air.

[0126] Continue to purge with clean air for 90-100 seconds until the resistance values ​​of all sensors return to near the initial baseline Ra. Purge is performed to restore the gas sensors' response to the baseline state, preparing them for the next sample analysis and preventing cross-contamination, ensuring that each sample's data is independent and accurate.

[0127] By recording the complete response process of the sensor from contact to separation from the target gas, its response resistance Rg is usually the stable value when the resistance change in the curve reaches the extreme value. This yields a complete fitting curve of the electronic nose 41 response intensity that includes information from acetaldehyde and benzene sensors.

[0128] S4.4 Cyclic Detection and Data Recording

[0129] Repeat steps S4.1 to S4.3. In subsequent loops, the long preheating time can be omitted; only standard gas washing is needed. Each detection corresponds to an odor sample at a specific temperature point. All data, including sample number, corresponding temperature, and Ra and Rg values ​​of each sensor, are saved in computer terminal 50. This results in a dataset containing multiple temperature points and multiple response curves, providing a data foundation for feature extraction and model training. This allows for the systematic acquisition of response signals from odor samples at all different temperature stages, constructing a complete dataset.

[0130] S5. Extract the feature values ​​of the response signal and establish the mapping relationship between the feature values ​​and the coal spontaneous combustion temperature stage.

[0131] The aim is to extract stable and representative mathematical features from the complex dynamic signals acquired by the electronic nose 41, identify the patterns between these features and the spontaneous combustion temperature of coal, and establish their mapping relationship. This includes the following steps:

[0132] S5.1 Extracting Feature Values ​​R

[0133] Ten key resistance values ​​were extracted from each sensor response curve for each odor sample:

[0134] The resistance value of the sensor in a stable state before contact with the odor sample is taken, and the average resistance during the 30-35s period of its detection cycle is taken as the baseline resistance Ra; it represents the zero point of the sensor.

[0135] The resistance value of the sensor when it reaches a stable response after contact with the odor sample is taken, and the average resistance during the 130-135s period of its detection cycle is taken as the response resistance Rg, which represents the maximum response of the sensor to the target odor.

[0136] The response intensity characteristic value R is calculated based on the formula R=Ra-Rg.

[0137] Since the response intensity R is positively correlated with odor concentration, this feature value R directly reflects the relative magnitude of odor concentration. Feature extraction eliminates the influence of individual sensor differences and baseline drift, providing a unified response intensity index. This results in a clean and well-organized feature value dataset, where each data point corresponds to the response intensity of a specific sensor at a given temperature.

[0138] S5.2 Curve Fitting

[0139] From the entire sensor array of the electronic nose 41, the sensor data most indicative of the coal spontaneous combustion process were selected, typically from acetaldehyde and benzene sensors. Therefore, the release risk of acetaldehyde and benzene should be a primary focus during the heating process. Odor response signals from odor samples at different temperatures were feature-extracted, and the stable resistance value in the odor response signal was used as the characteristic value R of the odor response intensity. The characteristic values ​​R of the gas sensors' response intensity to acetaldehyde and benzene in odor samples at different temperatures were obtained. Curve fitting was performed on the change of characteristic value R with temperature to identify the key characteristic odors, acetaldehyde and benzene. Acetaldehyde is a marker gas for the early stages of low-temperature coal oxidation, while benzene is a marker gas for the middle and later stages. The numerical tables were then converted into intuitive graphs to facilitate observation of the sensor response's variation with temperature, especially identifying inflection points—the temperature points where the response value begins to rise significantly.

[0140] Scatter plots were created using coal sample temperature as the x-axis and the corresponding sensor response intensity R as the y-axis to show the temperature variation of acetaldehyde and benzene sensors. Mathematical methods, such as polynomial fitting, were then used to fit these scatter plots into a smooth curve, yielding the response intensity-temperature fitted curve.

[0141] In this embodiment, Figure 4 The image shows the fitted curves of the electronic nose 41's response intensity to acetaldehyde and benzene sensor information released by spontaneous combustion of lignite. The X-axis represents coal temperature (°C), ranging from 40°C to 200°C; the left Y-axis represents the acetaldehyde sensor's response intensity (Ω), ranging from 0 to 350Ω; and the right Y-axis represents the benzene sensor's response intensity (Ω), ranging from 1 to 9Ω. The red hollow circles represent experimental data points for the acetaldehyde sensor's response intensity, and the blue hollow circles represent experimental data points for the benzene sensor's response intensity. Fitted curve 1 (the solid red line) is the fitted curve for the acetaldehyde sensor's response intensity; fitted curve 2 (the dashed blue line) is the fitted curve for the benzene sensor's response intensity.

[0142] from Figure 4As can be seen, the nonlinear relationship between coal temperature and the response intensity of acetaldehyde and benzene sensors is clearly demonstrated. For the acetaldehyde sensor, the response intensity is 2.13Ω when the coal temperature reaches 40℃; when the coal temperature reaches 80℃, the response intensity of the acetaldehyde sensor begins to increase, reaching its limit of 330Ω after 160℃, after which the response intensity tends to stabilize. For the benzene sensor, its response intensity begins to appear when the coal temperature reaches 110℃, and then increases with the increase of coal temperature. The highest response intensity of this sensor is 8.54Ω, that is, it reaches its peak at 180℃. Acetaldehyde can be detected at about 40℃ and increases significantly after 80℃; benzene begins to appear at about 110℃; acetaldehyde release starts early and is of high intensity; benzene release is delayed but equally significant, and both reach their release peak at 160-180℃; and after the temperature is above 180℃, the release rate of acetaldehyde and benzene may decrease slightly due to the completion of combustion or decomposition.

[0143] Figure 4 The patterns observed can provide important data support for understanding the odor release patterns during spontaneous combustion of coal underground, and have practical guiding value for fields such as environmental monitoring and safe production; a multi-sensor collaborative early warning model should be constructed based on the temperature characteristics of different odor responses.

[0144] S5.3 Establish the mapping relationship between eigenvalues ​​and temperature stages

[0145] Based on the inflection point of the curve obtained in step S5.2, and combined with the spontaneous combustion law of coal, the continuous heating process is divided into three discrete stages:

[0146] Early stage: 30-60℃, the main characteristic is that the acetaldehyde sensor begins to have a weak response, while benzene sensors usually have no response or a negligible response.

[0147] Mid-term: 60-130℃, the main characteristic is that the response intensity of acetaldehyde sensors increases sharply, while benzene sensors usually begin to show response intensity at about 110℃ and the response intensity increases with temperature.

[0148] Later stage: >130℃, the main characteristics are that the responses of both acetaldehyde and benzene sensors are at a high level, and the acetaldehyde sensor may tend to saturate.

[0149] A mapping relationship between characteristic values ​​and temperature stages is established: when the acetaldehyde sensor response intensity R < threshold 1 = 2.13Ω and the benzene sensor R ≈ 0Ω, the coal is in the early stage of spontaneous combustion; when the acetaldehyde sensor response intensity threshold 1 ≤ R < threshold 2 = 280Ω and the benzene sensor R < 3Ω, the coal is in the middle stage of spontaneous combustion; when the acetaldehyde sensor response intensity threshold 2 ≤ R ≤ threshold 3 = 330Ω and the benzene sensor R ≤ 8.54Ω, the coal is in the late stage of spontaneous combustion; the acetaldehyde sensor response intensity of 330Ω represents the maximum value; and the benzene sensor response intensity of 8.54Ω represents the maximum value. Based on this mapping, a rule-based discrimination criterion can be formed to determine the stage of coal spontaneous combustion based on a set of acetaldehyde or benzene characteristic values ​​R.

[0150] S5.4 Validation and Visualization Using Principal Component Analysis

[0151] The response intensity characteristics R of the entire sensor array (not just acetaldehyde and benzene) are input into the principal component analysis (PCA) algorithm.

[0152] Using the PCA algorithm, firstly, the multidimensional features can be reduced in dimensionality, allowing the multidimensional characteristics of coal spontaneous combustion odor to be represented more intuitively in a lower dimension. The original odor features are 10-dimensional (10 gas sensors), which are then... Figure 5-6 It can be seen that after dimensionality reduction using principal component analysis (PCA), the two-dimensional features of PC1 and PC2 can reflect 85.6% of the signal distribution characteristics of the original features, while the three-dimensional features of PC1, PC2, and PC3 can reflect 92.4% of the signal distribution characteristics of the original features. Secondly, regarding data visualization, the distribution of multi-dimensional features of spontaneous combustion odor cannot be directly displayed, but PCA dimensionality reduction can obtain the sample distribution while preserving the effective information of the original odor characteristics. Finally, irrelevant features are removed; irrelevant features are generated during the collection of spontaneous combustion odor samples, and PCA dimensionality reduction can more accurately restore the characteristics of spontaneous combustion odor samples.

[0153] The PCA algorithm is used to find several new coordinate axes, PC1, PC2, and PC3, that can best distinguish all data points and generate a two-dimensional ( Figure 5 ) Scatter plot and 3D ( Figure 6 A scatter plot; where PC1, PC2, and PC3 refer to the three dimensions under the new coordinate axes after principal component dimensionality reduction. Figure 5 and Figure 6 In the diagram, purple dots represent early-stage coal spontaneous combustion samples, green dots represent mid-stage samples, and pink dots represent late-stage samples. Observe these two PCA plots and assess the clustering effect, noting whether the early, mid, and late-stage samples cluster into different groups.

[0154] If the sample points at different stages can be clearly separated in the PCA plot, it indicates that the partitioning is effective. Figure 5-6 As can be seen from the results, the first, second, and third principal components can represent the ability of odor feature signals to distinguish the stages of coal spontaneous combustion. Although there is some overlap in the feature signals of the early and middle stages of coal spontaneous combustion, they can still be distinguished. The odor feature signals of the later stages of coal spontaneous combustion show a good distinguishing effect from the early and middle stages. This is because the oxidation process in the early and middle stages of coal spontaneous combustion is relatively slow, producing fewer types of odor gases with corresponding low concentrations. In contrast, the oxidation reaction in the later stages of coal spontaneous combustion is relatively intense, producing a large amount of odor with corresponding high concentrations. Both 2D and 3D principal component analysis plots can distinguish between the early, middle, and late stages of coal spontaneous combustion, indicating that the odor data from the electronic nose can be used as a basis for classifying the stages of coal spontaneous combustion.

[0155] The above feature analysis indicates that there are indeed essential differences in the odor characteristics of different stages of spontaneous combustion, thus enabling the establishment of a more scientific and credible response intensity-temperature mapping relationship.

[0156] The original signal is digitized through feature extraction, its changing patterns are identified through curve analysis, preliminary judgment rules are established through stage division, and finally, multi-dimensional verification is performed using PCA. The final output is a verified and reliable mapping model. This mapping model serves as the foundation and guiding framework for training machine learning classifiers in the next step.

[0157] S6. Based on feature value training, a learning model is constructed to build a prediction model, thereby achieving automatic classification and prediction of the coal spontaneous combustion stage.

[0158] Based on the response intensity feature value R extracted in step S5, the data is trained by a computer model learning algorithm to identify the odors of different coal spontaneous combustion stages and to build an intelligent prediction model that can automatically and quickly classify the odor samples to be tested.

[0159] S6.1 preprocesses the data and partitions the dataset.

[0160] The feature values ​​R from all sensors are standardized so that their mean is 0 and their variance is 1. This eliminates the differences in dimensions and orders of magnitude between different sensors.

[0161] One hundred labeled datasets, including the feature values ​​of each sample and their corresponding early, middle, and late stage labels, were randomly split proportionally. 80% of the feature values ​​were used as the training set, and 20% as the test set. This resulted in a clean, well-defined, and pre-segmented dataset, ready for model training.

[0162] S6.2 Model Selection and Training

[0163] Multiple machine learning algorithms can be selected for model training, and the algorithm with the best performance can be selected based on the evaluation results to build the final prediction model. Among them, at least one learning algorithm can be selected from Principal Component Analysis-Support Vector Machine (PCA-SVM), Support Vector Machine (SVM), Random Forest (RF), and Artificial Neural Network (ANN) when building the prediction model.

[0164] For prediction models obtained based on the Principal Component Analysis-Support Vector Machine (PCA-SVM) learning algorithm, the key feature is that PCA is first used to reduce the dimensionality of the data, retaining the most important information (e.g., more than 90%), and then SVM is used for classification. This simplifies the problem and removes noise.

[0165] For prediction models obtained based on the Support Vector Machine (SVM) learning algorithm, which are good at finding the optimal classification boundary in small sample and high-dimensional data.

[0166] For prediction models obtained based on the Random Forest (RF) learning algorithm, the characteristic is that they are classified by constructing multiple decision trees and combining their voting results, which has a strong resistance to overfitting.

[0167] Predictive models based on artificial neural network (ANN) learning algorithms are characterized by their ability to simulate complex nonlinear relationships and their powerful learning capabilities.

[0168] The feature values ​​R of the training set are used as input, and their corresponding coal spontaneous combustion stage labels are used as target outputs. These are then fed into each of the algorithms described above for training. The model automatically learns a complex mapping function from feature combinations to stage categories.

[0169] Different models have their own advantages and disadvantages. Training multiple models simultaneously allows us to compare and determine the one that best suits the current data characteristics, thus obtaining several well-trained candidate prediction models with preliminary classification capabilities. This allows the machine to summarize the inherent patterns that distinguish between early, middle, and late stages from a large number of labeled samples, rather than relying on fixed threshold rules set by humans.

[0170] S6.3 Model Evaluation and Optimization Selection

[0171] The feature values ​​of the test set are input into each trained model, allowing the model to provide a predicted classification, thus obtaining... Figures 7-10 The confusion matrix diagram is shown below. Wherein, Figure 7 The confusion matrix obtained using the Principal Component Analysis-Support Vector Machine (PCA-SVM) model is shown in the figure. Figure 8 The confusion matrix obtained using the Support Vector Machine (SVM) model, Figure 9 The confusion matrix obtained using the Random Forest (RF) model, Figure 10 This is a confusion matrix diagram obtained using an artificial neural network (ANN) model.

[0172] In a confusion matrix, the horizontal axis represents the predicted values, indicating the model's classification results (early, mid, and late stages); the vertical axis represents the true values, indicating the actual classification labels. The numbers in the matrix represent the number of samples that the model predicted as belonging to a particular true class. The values ​​on the diagonal represent the number of correctly predicted samples (higher is better); the values ​​off-diagonally represent the number of incorrectly predicted samples (lower is better). The confusion matrix clearly shows which classes the model is prone to misclassifying, providing direction for further optimization.

[0173] By analyzing and comparing the performance of these four candidate models, and comparing the model's prediction results with the true labels of the test set, we can fairly evaluate which model is the most reliable and accurate based on the independent data of the test set, and select the model with the highest computational accuracy as the optimal model.

[0174] Figure 7 Compared to Figure 8-10 The experimental results showed that the PCA-SVM model had the highest computational accuracy, reaching 95%, and it achieved 100% early recognition rate with only a very small number of misjudgments in the middle and late stages. Therefore, it was selected as the final best prediction model.

[0175] from Figure 8-10 As can be seen, the computational accuracy of the RF model is 85%; the computational accuracy of the ANN model is 90%; and the computational accuracy of the SVM model is 89%. Although the computational accuracy of these three models is greater than 85%, which meets the requirements of the prediction model for actual early warning scenarios, the RF model is relatively poor in identifying the mid-stage of spontaneous combustion of coal compared to other models, the ANN model has slight overfitting, and the accuracy of the SVM model can be further improved after principal component dimensionality reduction.

[0176] S6.4 Model Deployment and Automatic Classification Early Warning

[0177] The trained optimal prediction model (PCA-SVM) is integrated into the system's computer terminal, and then the application process is built.

[0178] When a new unknown odor sample is collected from the well, it is detected by an electronic nose 41 according to step S4 to obtain a response signal; then, data preprocessing and feature extraction are performed according to step S5; finally, the extracted feature values ​​are input into the deployed PCA-SVM prediction model; the prediction model will immediately output a classification result of "early", "mid" or "late".

[0179] The system automatically triggers corresponding early warning commands based on the classification results. Early warning processing includes:

[0180] If the prediction indicates early-stage spontaneous combustion of coal, the risk is considered low or in the initial oxidation stage. In this case, enhanced monitoring and preventative measures are necessary. The control system on computer terminal 50 will then operate in power-saving mode in the background. Further measures, such as enhanced temperature monitoring and CO gas testing, will be implemented to comprehensively assess the spontaneous combustion trend of coal.

[0181] If the prediction indicates the middle stage of spontaneous combustion of coal, it is determined that spontaneous combustion has occurred and entered the development stage; at this time, corresponding prevention and control measures need to be taken. The warning indicator light in the control system of computer terminal 50 will remain on, and an SMS notification will be sent to the on-duty person and the person in charge; it will also display that the monitored area is in a potentially abnormally high temperature, and the buzzer will sound continuously until the test returns to normal and the alarm stops. The corresponding prevention and control measures are: promptly locate the high-temperature area, use inert gas injection, water injection, or other measures to cool it down, and continuously monitor changes in temperature and gas composition in the high-temperature area.

[0182] If the prediction indicates the late stage of spontaneous combustion of coal, it is considered a severe case of spontaneous combustion with an extremely high fire risk. In this case, a combination of measures is required to prevent spontaneous combustion fires. The warning indicator lights in the computer terminal 50 control system will remain constantly lit, indicating the presence of a direct hazardous heat source; a buzzer will sound continuously; and a text message will be sent to notify the on-duty personnel and contact the responsible person. The alarm will stop only after the test returns to normal. Corresponding preventative measures include: evacuating workers; constructing explosion-proof airtight walls and injecting mortar to temporarily seal the spontaneous combustion area; injecting inert gas and chemical inhibitors to control the spontaneous combustion area; and immediately extinguishing any open flames that appear.

[0183] The laboratory-developed model was transformed into a field-use decision-making tool, enabling real-time and rapid early warning of coal spontaneous combustion. Ultimately, the system's goal was achieved: to automatically and quickly determine the stage of coal spontaneous combustion based on odor without manual analysis, providing intuitive and actionable early warning information for safe coal mine production.

[0184] Through a machine learning process, the experimental data obtained in previous steps were trained into a reusable and efficient predictive model that can replace expert experience. This model transforms the coal spontaneous combustion early warning system based on the electronic nose 41 from a complex laboratory analysis technique into an automated monitoring tool applicable to mine sites.

[0185] A coal spontaneous combustion odor testing system, such as Figure 2 As shown, the method described above includes:

[0186] Air supply unit 10: used to introduce dry air into container 21, including dry air cylinder 11, which is connected to container 21 through a pipe. A pressure reducing valve 12, a pressure regulating valve 13, and a flow regulating valve 14 are sequentially installed on the pipe from one side of dry air cylinder 11. After entering simulation unit 20, the dry air is heated by air heating pipe 25, allowing the dry air to enter container 21.

[0187] Simulation unit 20: used to simulate coal spontaneous combustion, including a simulation device, which contains a container 21, an air heating pipe 25, a heater 26 for heating the container 21, and an operating interface 27; it also includes a temperature parameter table 28 with a preset coal spontaneous combustion heating program. Inside the container 21, the target coal sample 23 is placed on an asbestos layer 22, and another asbestos layer 22 is laid on top of the target coal sample 23.

[0188] The acquisition unit 30 collects and stores the odor samples released by spontaneous combustion of coal. It includes a gas sampling bag 31, which is directly connected to the container 21 via a pipe and indirectly connected to the electronic nose 41 via a syringe 32. In use, first, the shut-off valve 42 is closed, and the syringe 32 extracts the odor sample from the gas sampling bag 31; then, the shut-off valve 42 is opened, and the syringe 32 is connected to the sampling port of the electronic nose 41, thus completing the transfer of the odor sample.

[0189] Analysis unit 40: Analyzes and detects the collected odor samples, including electronic nose 41 and air pump 43. Electronic nose 41 is communicatively connected to computer terminal 50; and temperature parameter table 28 is also communicatively connected to computer terminal 50. An example of the communication connection is shown below. Figure 2 As shown by the dashed line.

[0190] The method and system for detecting the odor of spontaneous combustion of coal designed in this application are particularly suitable for odor detection during the low-temperature oxidation process of spontaneous combustion of coal. By using an electronic nose to focus on two specific characteristic odor substances, acetaldehyde and benzene, and combining segmented heating and machine learning models, the method achieves rapid and automatic identification and early warning of the spontaneous combustion stage of coal.

[0191] This application can detect the odor components of different coal types released at different temperature stages during spontaneous combustion and obtain their concentration changes; it can make an early judgment on coal spontaneous combustion based on odor characteristics.

[0192] The entire testing system has the advantages of compact structure, high detection efficiency and low cost. By simulating the real coal spontaneous combustion process through programmed temperature rise experiments, combined with classification models such as PCA-SVM, it effectively improves the accuracy and reliability of stage identification, providing a new technical means for coal spontaneous combustion early warning in the field of coal mine safety.

[0193] The embodiments of this application have been described in detail above. These descriptions are merely preferred embodiments and should not be construed as limiting the scope of this application. All equivalent variations and modifications made within the scope of this application should still fall within the patent coverage of this application.

Claims

1. A method for testing the spontaneous combustion odor of coal, characterized in that the steps include... include: S1. Pre-treat the original coal sample, and then fill the treated target coal sample into a sealed container; S2. Dry air is continuously introduced into the container, and a staged heating method is used to simulate the spontaneous combustion process of coal to generate the smell of spontaneous combustion of coal; dry air with a flow rate of 50-150 ml / min is continuously introduced into the container; The segmented heating includes: Within the range of [30℃, 100℃], the heating rate is 0.5℃ / min; Within the range of (100℃, 200℃), the heating rate is 1℃ / min; S3. Collect and store the odor samples released by spontaneous combustion of coal within a preset temperature range; S4. Analyze and detect the collected odor samples using an electronic nose to obtain the response signals of the gas-sensitive sensors to the odor samples; the gas-sensitive sensors include acetaldehyde sensors and benzene sensors; the steps include: turning on the electronic nose for preheating, purging the electronic nose with air, and stopping the purging after the resistance values ​​of all gas-sensitive sensors in the electronic nose stabilize; injecting odor samples at different temperatures drawn from the gas sample bag into the sampling port of the electronic nose using a syringe; recording the odor response signals of the odor samples measured in the electronic nose; repeating the above steps, purging, and then sampling and detecting again to complete the detection of odor samples at different temperatures; wherein, the exhaust gas must be vented after each detection; the sampling time is set to 70-80s, and the purging time is set to 90-100s; S5. Extract the feature values ​​of the response signal and establish a mapping relationship between the feature values ​​and the coal spontaneous combustion temperature stage; the steps include: extracting features from the odor response signals of odor samples at different temperatures, and using the stable resistance value in the odor response signal as the feature value R of the odor response intensity; obtaining the feature value R of the gas sensor's response intensity to the odor sample at different temperatures; performing curve fitting on the feature value R captured by the acetaldehyde sensor and benzene sensor in the gas sensor as a function of temperature to obtain a fitting curve of response intensity-temperature; the process of extracting the odor response signal of the odor sample at different temperatures is as follows: for the response characteristics of the gas sensor to the odor sample at different temperatures, take the resistance value of the sensor before contacting the odor sample and in a stable state, and take the average resistance in the 30-35s time period of its detection cycle as the baseline resistance Ra; take the resistance value of the sensor after contacting the odor sample and when the response reaches a stable state, and take the average resistance in the 130-135s time period of its detection cycle as the response resistance Rg; the response intensity feature value R = Ra - Rg, ​​in Ω. S6. Based on feature value training, a learning model is constructed to build a prediction model, thereby achieving automatic classification and prediction of the coal spontaneous combustion stage.

2. The method according to claim 1, characterized in that, S1 specifically includes: The original coal sample was crushed into a target coal sample with a particle size of 0.2-4.5 mm; The target coal sample was placed in a drying oven at 40-50℃ to dehydrate until constant weight. Weigh a quantitative amount of target coal sample and fill it into a sealed container, then cover the target coal sample with an air-permeable asbestos layer.

3. The method according to claim 1, characterized in that, In S3, the preset temperature range is based on three characteristic temperature ranges divided into the coal spontaneous combustion stage, namely the early, middle and late stages; and odor samples released by coal spontaneous combustion are sampled in each of these three characteristic temperature ranges; when collecting odor samples, gas sampling bags made of inert material are used to store the collected odor samples, and the sampling time interval is once every 5 minutes, while the temperature is recorded. The early temperature ranges from 30 to 60°C, the middle temperature ranges from 60 to 130°C, and the late temperature ranges from >130°C.

4. The method according to any one of claims 1-3, characterized in that, Step S6 includes: Using the response intensity R as a feature value, a predictive model is constructed by training the data through a computer model learning algorithm. Based on the prediction model, the early, middle and late stages of coal spontaneous combustion are classified and predicted. Automatic early warning processing is performed after classification and prediction; The prediction model is constructed by selecting at least one learning algorithm from principal component analysis-support vector machine, support vector machine, random forest, and artificial neural network; 80% of the feature values ​​are used as the training set and 20% of the feature values ​​are used as the test set.

5. The method according to claim 4, characterized in that, The early warning processing includes: If the prediction indicates that the coal is in the early stage of spontaneous combustion, it is determined to be in the initial oxidation stage, and further monitoring and preventive measures need to be taken. If the prediction indicates that the spontaneous combustion of coal is in its middle stage, then it is determined that spontaneous combustion has occurred and entered the development stage, and corresponding prevention and control measures need to be taken. If the prediction indicates that the fire is in the later stages of spontaneous combustion of coal, then the spontaneous combustion phenomenon is considered severe, the fire risk is extremely high, and comprehensive measures need to be taken to prevent spontaneous combustion fires.

6. A coal spontaneous combustion odor testing system, used to implement the method described in any one of claims 1-5, characterized in that, include: Air supply unit: used to introduce dry air into the container, comprising a dry air cylinder connected to the container via a pipe; Simulation unit: used to simulate spontaneous combustion of coal, including simulation device, which contains the container, air heating pipe, heater for heating the container, and operation interface; Acquisition unit: Acquires and stores the odor samples released by spontaneous combustion of coal, including a gas sample bag, which is connected to the container and the electronic nose respectively; Analysis unit: Analyzes and detects the collected odor samples, including the electronic nose and air pump, wherein the electronic nose is communicatively connected to a computer terminal.

Images