A method and device for detecting concealed fire sources based on multimodal fusion identification

Abstract

This invention provides a method and apparatus for detecting concealed fire sources based on multimodal fusion identification, belonging to the field of concealed fire source detection technology. The method includes: S1: transmitting multi-frequency radar wave signals; S2: collecting borehole bottom temperature; S3: performing preprocessing; S4: constructing a radar-temperature fusion feature vector; S5: inputting a multimodal classification model and outputting category probabilities; S6: determining a preliminary geological category based on the maximum category probability; S7: correcting the confidence level of the concealed fire source anomaly zone identification results and determining whether to output drilling cessation information; S8: generating a spatial distribution estimation model and generating visualization results and a detection report based on the spatial distribution estimation model. This method can provide a multi-source information fusion discrimination basis for concealed fire source identification, which helps reduce the risk of misjudgment caused by relying solely on a single radar echo or single-point temperature data.

Description

Technical Field

[0001] This invention relates to the field of concealed fire source detection technology, specifically to a concealed fire source detection method and device based on multimodal fusion identification. Background Technology

[0002] Hidden underground ignition sources in coal mines are often concealed in goaf areas, fracture zones, broken coal and rock masses, or areas with complex geological structures. They are characterized by high concealment, difficulty in determining their spatial location, susceptibility to repeated reignition, and great difficulty in control. Borehole ground-penetrating radar (GPR) utilizes the characteristic of electromagnetic waves reflecting at the interface of coal and rock media with different dielectric properties to detect the coal and rock structure around the borehole. Drilling-while-drilling (DWD) methods can acquire geological information simultaneously during drilling, reducing the need for separate detection procedures and demonstrating significant engineering application value.

[0003] However, the following problems exist when using drilling-while-drilling radar for concealed fire source detection in underground coal mines:

[0004] First, conventional drilling often uses standard metal drill pipes, which shield and attenuate radar electromagnetic waves, affecting the emission of electromagnetic waves to the surrounding coal and rock mass and the reception of reflected echo signals. Second, the heat from the drill bit cutting and the high temperature in the combustion zone of the concealed fire source may overlap, causing electronic components such as radar antenna, PCB control board, and power module to be affected by high temperature, resulting in insufficient stability of the detection device. Third, traditional borehole temperature measurement methods mainly rely on single-point temperature data, making it difficult to distinguish different geological anomalies such as hidden fire source combustion zones, goaf zones, and fracture zones; when relying solely on radar echo interpretation, they are easily affected by differences in coal and rock media, drilling noise, reflection interference, and human interpretation experience. Fourth, during deep hole drilling, the signal transmission distance is relatively long, and radar wave echo signals and temperature signals are prone to transmission attenuation or delay, requiring a signal forwarding structure suitable for drilling conditions. Fifth, existing radar signal processing methods mostly focus on interpreting single signal sources, which is not conducive to the comprehensive identification and early warning of hidden disaster-causing bodies such as concealed fire sources and burning areas.

[0005] Therefore, we propose a method and device for detecting concealed fire sources based on multimodal fusion identification. Summary of the Invention

[0006] The purpose of this invention is to provide a method and apparatus for detecting concealed fire sources based on multimodal fusion identification, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: This invention provides a method for detecting concealed fire sources based on multimodal fusion identification, comprising the following steps: S1: Drill along the drilling direction and simultaneously transmit multi-frequency radar wave signals to the coal and rock mass around the borehole; S2: Use a temperature sensor to collect the temperature at the bottom of the borehole; S3: Preprocess the radar echo signal; S4: The radar echo signal features preprocessed in step S3 are fused with the borehole bottom temperature features obtained in step S2 to construct a radar-temperature fusion feature vector. S5: Input the radar-temperature fusion feature vector from step S4 into the multimodal classification model. The multimodal classification model outputs the class probabilities of various geological categories distributed within a preset radial range around the borehole along the borehole depth direction. S6: Determine the preliminary geological category based on the maximum category probability; S7: Based on the bottom hole temperature or bottom hole temperature change rate from step S2, adjust the confidence level of the identification results of the hidden fire source abnormal area and determine whether to output drilling stop information. S8: Generate a spatial distribution estimation model, and generate visualization results and detection reports based on the spatial distribution estimation model.

[0008] Furthermore, in step S3, the preprocessing includes basic preprocessing and advanced preprocessing. The basic preprocessing includes pre-amplification, A / D conversion, moving average filtering, and zero-mean normalization. The advanced preprocessing includes median filtering, spectral feature extraction, and outlier removal.

[0009] Furthermore, step S4 specifically includes: S41: After preprocessing, the radar echo signal characteristics and bottom hole temperature characteristics are aligned by time or drilling distance according to the sampling time, drilling distance or sampling point number. S42: For the same sampling point or the same borehole depth, construct a radar-temperature fusion feature vector that satisfies:

[0010] in, , , , The first The sampling point or the first Radar echo amplitude characteristics at different frequency bands at various borehole depths; For the first The sampling point or the first Spectral characteristics at each borehole depth location; For the first The sampling point or the first Bottom hole temperature characteristics at each borehole depth location; Characterized by the rate of temperature change; For the first The sampling point or the first The radar-temperature fusion feature vector corresponding to each borehole depth location.

[0011] Furthermore, in step S5, the multimodal classification model includes a feature projection module, an attention mechanism module, and a classification output module. The attention mechanism module includes a query linear transformation layer, a key linear transformation layer, and a value linear transformation layer. The feature projection module is used to map radar echo features and borehole bottom temperature features to the same feature dimension. The attention mechanism module is used to assign weights to radar echo features and borehole bottom temperature features of different frequency bands to obtain weighted fusion features. The classification output module is used to output the category probabilities of various geological categories distributed within a preset radial range around the borehole along the borehole depth direction.

[0012] Furthermore, step S5 specifically includes: S51: Combine the radar-temperature fusion feature vector from step S42 The input feature projection module obtains the hidden feature vector through linear transformation, satisfying the following:

[0013] in, This represents the hidden feature vector after mapping; Represents the feature projection weight matrix; Represents the bias vector; S52: The query linear transformation layer, key linear transformation layer, and value linear transformation layer perform linear transformations on the hidden feature vectors respectively to obtain the query vector, key vector, and value vector, satisfying:

[0014]

[0015]

[0016] in, For query vector; Represents the key vector; Represents a value vector; This indicates a query for the weight matrix of the linear transformation layer; This represents the weight matrix of the key-linear transformation layer; This represents the weight matrix of the linear transformation layer; S53: Calculate the attention weights, satisfying:

[0017] in, Indicates attention weight; Indicates feature dimension; Indicates the scaling factor; S54: Calculate the weighted fusion features, satisfying:

[0018] in, For weighted fusion features; S55: Based on weighted fusion characteristics The classification output module obtains the original category scores; S56: Based on the original category scores, the classification output module outputs the category probabilities of various geological categories distributed within a preset radial range around the borehole along the borehole depth direction using a probability normalization function, satisfying:

[0019] in, This is the normalized class probability vector; The original category score.

[0020] Furthermore, in step S56, the geological categories include intact rock strata, concealed fire source anomalous areas, goaf areas, and geological anomalies, with geological anomalies including faults, fissures, and aquifers.

[0021] Furthermore, in step S6, a preliminary geological category is determined based on the maximum category probability, satisfying:

[0022] in, For the first The sampling point or the first The preliminary geological category corresponding to the pre-defined radial range around the borehole at each borehole depth location.

[0023] Furthermore, step S7 specifically includes: Set a preset probability threshold, a preset temperature threshold, a preset rate of change threshold, and a recovery temperature threshold. Based on the bottom temperature or the rate of change of the bottom temperature in step S2, if the probability of the hidden fire source abnormal area category is greater than the preset probability threshold, the bottom temperature is greater than or equal to the preset temperature threshold, or the rate of change of the bottom temperature is greater than or equal to the preset rate of change threshold, increase the confidence level of the hidden fire source abnormal area identification result, and identify the hidden fire source abnormal area as a high-confidence hidden fire source abnormal area. If the probability of a hidden fire source abnormal zone is greater than a preset probability threshold, the temperature at the bottom of the hole is less than a preset temperature threshold, and the rate of change of the temperature at the bottom of the hole is less than a preset rate of change threshold, then the hidden fire source abnormal zone will be identified as a coal fire abnormal zone to be reviewed. In other cases, areas with concealed fire sources are identified as low-confidence areas with concealed fire sources. If the bottom hole temperature is greater than or equal to the preset temperature threshold or the bottom hole temperature change rate is greater than or equal to the preset change rate threshold, a drilling stop message is output. The drilling stop message is released when the bottom hole temperature drops below the recovery temperature threshold and remains stable for a preset time.

[0024] This invention also provides a concealed fire source detection device based on multimodal fusion recognition, used to implement the concealed fire source detection method based on multimodal fusion recognition described above. The device includes a drill bit, a first drill rod section, a detection connector, a main unit, and a ground host. The main unit includes several standard metal drill rods and several repeater connectors, which are alternately connected. Each repeater connector contains a signal relay repeater and a repeater power supply battery, which is electrically connected to the signal relay repeater. A temperature sensor is located at the end of the first drill rod near the drill bit. The drill bit is installed at one end of the first drill rod. The detection connector is installed at the end of the first drill rod away from the drill bit. The end of the detection connector away from the first drill rod is connected to the standard metal drill rod of the main unit. Both the drill bit and the temperature sensor are electrically connected to the detection connector. The detection connector is electrically connected to the signal relay repeater, and the signal relay repeater is electrically connected to the ground host.

[0025] Furthermore, the detection connector includes an outer load-bearing cylinder with threaded connection interfaces at both ends. The inner wall of the outer load-bearing cylinder is provided with a vibration damping layer, a heat insulation layer, and a sealing layer sequentially from the outside to the inside. The outer load-bearing cylinder has a through-flow channel with an hourglass-shaped structure. The diameters at both ends of the through-flow channel are larger than the diameter at the middle. The space between the outer wall of the through-flow channel and the inner wall of the outer load-bearing cylinder forms a sandwich cavity. The sandwich cavity houses a radar transmitting antenna, a radar receiving antenna, a PCB control board, and a power supply... The system includes a battery, and the radar transmitting antenna, radar receiving antenna, and PCB control board are all electrically connected to the battery. The radar transmitting antenna and radar receiving antenna are both electrically connected to the PCB control board. The outer load-bearing cylinder has an electromagnetic wave transmitting window and an electromagnetic wave receiving window made of non-metallic wave-transparent material. The radar transmitting antenna is oriented towards the electromagnetic wave transmitting window, and the radar receiving antenna is oriented towards the electromagnetic wave receiving window. The temperature sensor is electrically connected to the PCB control board, and the PCB control board is electrically connected to the signal relay and the ground host.

[0026] Compared with the prior art, the present invention has the following technical effects: 1. In this invention, the method processes multi-frequency radar echo signals through a two-layer preprocessing architecture of basic preprocessing and advanced preprocessing. It fuses radar echo signal features with borehole bottom temperature features to construct a radar-temperature fusion feature vector. A multimodal classification model outputs the probability of different geological categories distributed within a preset radial range around the borehole along the borehole depth direction, providing a multi-source information fusion basis for identifying concealed fire sources. Combining the probability of concealed fire source anomaly zone categories with borehole bottom temperature and borehole bottom temperature change rate allows for confidence level correction of concealed fire source anomaly zone identification results, reducing the risk of misjudgment caused by relying solely on a single radar echo or single-point temperature data. Through geological profiles and spatial distribution estimation models, visualized results and detection reports are generated, facilitating identification and early warning decisions by on-site personnel.

[0027] 2. In this invention, the radar transmitting antenna, radar receiving antenna, PCB control board, and power supply battery are housed within the interlayer cavity of the probe connector. An electromagnetic wave transmitting window and an electromagnetic wave receiving window made of non-metallic transparent material are provided on the probe connector, which reduces the shielding effect of the standard metal drill pipe on radar electromagnetic wave transmission and reception. Positioning the probe connector between the first drill pipe section and the standard metal drill pipe maintains a certain distance between the probe connector and the high-temperature cutting zone of the drill bit. Simultaneously, a temperature sensor is installed at the end of the first drill pipe section near the drill bit, which helps reduce the impact of drill bit cutting heat and coal fire high temperatures on the probe connector. Signal forwarding via a repeater connector facilitates data transmission requirements during deep hole drilling. Attached Figure Description

[0028] Figure 1 This is a flowchart of a concealed fire source detection method according to an embodiment of the present invention; Figure 2 This is a flowchart illustrating the process of identifying concealed fire source abnormal zones according to an embodiment of the present invention. Figure 3 This is a diagram illustrating the radar signal denoising effect of an embodiment of the present invention. Figure 4 This is a comparison chart of geological profile prediction results from an embodiment of the present invention; Figure 5 This is a schematic diagram of the detection device according to an embodiment of the present invention; Figure 6 This is a schematic diagram of the probe connector structure according to an embodiment of the present invention; Figure 7 This is a schematic diagram of the sealing layer, heat insulation layer, and vibration damping layer according to an embodiment of the present invention; Figure 8 This is a schematic diagram of the high-temperature interlocking control process according to an embodiment of the present invention.

[0029] In the diagram: 1. Ground host, 2. Repeater connector, 3. Standard metal drill rod, 4. Detector connector, 41. External load-bearing cylinder, 42. Threaded connection interface, 43. Through internal flow channel, 44. Sandwich cavity, 45. Radar transmitting antenna, 46. Radar receiving antenna, 47. PCB control board, 48. Power supply battery, 49. Sealing layer, 50. Heat insulation layer, 51. Vibration damping layer, 52. Electromagnetic wave transmitting window, 53. Electromagnetic wave receiving window, 5. First section of drill rod, 6. Drill bit. Detailed Implementation

[0030] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the protection scope of the present invention.

[0031] In this article, terms such as "left," "right," "up," "down," "front," and "back" are established based on the positional relationships shown in the attached drawings. Depending on the attached drawings, the corresponding positional relationships may also change. Therefore, they should not be interpreted as an absolute limitation on the scope of protection.

[0032] Please see Figures 1 to 2 This embodiment provides a method for detecting concealed fire sources based on multimodal fusion recognition, including the following steps: S1: Drill along the drilling direction while simultaneously transmitting multi-frequency radar wave signals to the coal and rock mass surrounding the borehole.

[0033] Specifically, in this embodiment, the multi-frequency radar wave signal includes three electromagnetic wave signals of different frequencies: 50MHz, 100MHz, and 200MHz. These different frequency signals are used to acquire the reflected echo characteristics of the coal and rock mass at different electromagnetic response scales.

[0034] S2: Use a temperature sensor to collect the temperature at the bottom of the borehole.

[0035] S3: Preprocess the radar echo signal.

[0036] Specifically, in step S3, preprocessing includes basic preprocessing and advanced preprocessing. Basic preprocessing includes pre-amplification, A / D conversion, moving average filtering, and zero-mean normalization. Advanced preprocessing includes median filtering, spectral feature extraction, and outlier removal. A two-layer preprocessing architecture is adopted, consisting of a basic preprocessing layer and an advanced preprocessing layer. The basic preprocessing layer sequentially performs pre-amplification, A / D conversion, moving average filtering, and zero-mean normalization on the radar echo signal. The advanced preprocessing layer performs median filtering, spectral feature extraction, and outlier removal on the radar echo signal processed by the basic preprocessing layer, thereby improving signal robustness.

[0037] Specifically, Figure 3 The image shows the denoising effect of the radar signal. In this embodiment, the moving average filtering uses a five-point moving average filter to denoise the signal of each channel individually. The filtering formula is:

[0038] in, The filtered first The sampling point Radar signal amplitude of each channel; This is the radar signal after pre-amplification; The filter window offset, when When the sampling range is exceeded, boundary values ​​are used for padding.

[0039] Specifically, the process of zero-mean standardization is as follows: Let the first Radar echo sequences of each frequency band for:

[0040] in, This represents the number of sampling points; For the first The first frequency band The echo amplitude of each sampling point.

[0041] Calculate the mean of this frequency band. and standard deviation The standardized result is obtained according to the following formula:

[0042] in, Indicates the standardization result; To prevent constants with a denominator of zero.

[0043] Specifically, spectral features are extracted using Welch power spectrum estimation. The formulas for calculating peak frequency, average frequency, and spectral width are as follows:

[0044]

[0045]

[0046] in, Peak frequency; Average frequency; The bandwidth is the frequency spectrum. For frequency; This represents the power spectral density.

[0047] S4: The radar echo signal features preprocessed in step S3 are fused with the borehole bottom temperature features obtained in step S2 to construct a radar-temperature fusion feature vector.

[0048] Specifically, step S4 includes: S41: After preprocessing, the radar echo signal characteristics and bottom hole temperature characteristics are aligned by time or drilling distance according to the sampling time, drilling distance or sampling point number. S42: For the same sampling point or the same borehole depth, construct a radar-temperature fusion feature vector that satisfies:

[0049] in, , , , The first The sampling point or the first Radar echo amplitude characteristics at different frequency bands at various borehole depths; For the first The sampling point or the first Spectral characteristics at each borehole depth location; For the first The sampling point or the first Bottom hole temperature characteristics at each borehole depth location; Characterized by the rate of temperature change; For the first The sampling point or the first The radar-temperature fusion feature vector corresponding to each borehole depth location.

[0050] Specifically, when only the radar echo amplitude characteristics of three frequency bands and the bottom temperature characteristic of the borehole are used, the radar-temperature fusion feature vector can be expressed as: .

[0051] S5: Input the radar-temperature fusion feature vector from step S4 into the multimodal classification model. The multimodal classification model outputs the class probabilities of various geological categories distributed within a preset radial range around the borehole along the borehole depth direction.

[0052] Specifically, in step S5, the multimodal classification model includes a feature projection module, an attention mechanism module, and a classification output module. The attention mechanism module includes a query linear transformation layer, a key linear transformation layer, and a value linear transformation layer. The classification output module includes a fully connected neural network, which includes an input layer, at least two hidden layers, and an output layer. The number of nodes in the output layer corresponds to the number of geological categories. Geological categories include intact rock strata, concealed fire source anomaly zones (i.e., coal fire combustion zones), goaf areas, and geological anomalies. Geological anomalies include faults, fissures, and aquifers.

[0053] Specifically, the feature projection module is used to map radar wave echo features and borehole bottom temperature features to the same feature dimension; the attention mechanism module is used to assign weights to radar wave echo features and borehole bottom temperature features of different frequency bands to obtain weighted fusion features; and the classification output module is used to output the class probabilities of various geological categories distributed within a preset radial range around the borehole along the borehole depth direction.

[0054] Specifically, the multimodal classification model is trained using labeled multi-frequency radar echo samples and temperature samples. The training samples include multi-frequency radar echo features, borehole bottom temperature features, and category labels corresponding to different geological categories. During training, the difference between the model's output category probability and the true category label can be calculated using the cross-entropy loss function, and the model parameters are updated using the gradient descent class optimization algorithm. The trained multimodal classification model is then used for forward inference on the radar echo and temperature signals acquired during drilling. In this embodiment, the forward inference process is as follows: the preprocessed signal is used as the input data for model training, the Adam optimizer is used, the learning rate is set to 0.001, the cross-entropy loss function is used, the training epochs are 150, and the training loss value is output every 10 epochs. The hidden layers use the ReLU activation function, and batch normalization layers and dropout layers are sequentially set after each hidden layer to prevent overfitting. The dropout probabilities are 0.3, 0.3, and 0.2, respectively.

[0055] Specifically, in this embodiment, labeled forward simulation samples are used as training samples. Each sample includes 1024 sampling points, and each sampling point includes radar echo features in three frequency bands and a bottom temperature feature. The multimodal classification model can use a fully connected neural network with a 4→128→64→4 structure; when the geological categories are set to six, a fully connected neural network with a 4→128→64→6 structure can be used. The above network structure is only an example, and in actual applications, it can be adjusted according to the number of input features, the number of geological categories, and computing power.

[0056] Specifically, step S5 includes: S51: Combine the radar-temperature fusion feature vector from step S42 The input feature projection module obtains the hidden feature vector through linear transformation, satisfying the following:

[0057] in, This represents the hidden feature vector after mapping; Represents the feature projection weight matrix; This represents the bias vector.

[0058] S52: The query linear transformation layer, key linear transformation layer, and value linear transformation layer perform linear transformations on the hidden feature vectors respectively to obtain the query vector, key vector, and value vector, satisfying:

[0059]

[0060]

[0061] in, For query vector; Represents the key vector; Represents a value vector; This indicates a query for the weight matrix of the linear transformation layer; This represents the weight matrix of the key-linear transformation layer; The weight matrix represents the linear transformation layer.

[0062] S53: Calculate the attention weights, satisfying:

[0063] in, Indicates attention weight; Indicates feature dimension; This represents the scaling factor.

[0064] S54: Calculate the weighted fusion features, satisfying:

[0065] in, This is a weighted fusion feature.

[0066] S55: Based on weighted fusion characteristics The classification output module obtains the original category scores.

[0067] Specifically, the weighted fusion features After inputting into the fully connected neural network, the output layer outputs the original category score. The specific process is as follows: First layer fully connected: Linear transformation:

[0068] in, This represents the output of the first fully connected layer; This represents the weight matrix of the first layer. ; This represents the bias term of the first layer. .

[0069] Normalization:

[0070] in, or This represents the final output value after batch normalization. Indicates the scaling parameter; This represents the mean of the input to the first neuron in the current mini-batch of data; This represents the variance of the input to the first neuron in the current mini-batch of data; Indicates the translation parameter; This represents a constant to prevent the denominator from being zero. RELU activation:

[0071] in, This is the output of the first hidden layer; This represents the modified linear unit activation function; Second layer fully connected: Linear transformation:

[0072] in, This indicates the output of the second fully connected layer; This represents the weight matrix of the second layer. ; This represents the bias term of the second layer. .

[0073] Normalization:

[0074] in, or This represents the final output value after batch normalization. Indicates the scaling parameter; Indicates input The mean over a batch; Indicates input Variance over a batch; Indicates the translation parameter; This represents a constant to prevent the denominator from being zero. RELU activation:

[0075] in, This is the output of the second hidden layer.

[0076] Third layer fully connected: Linear transformation:

[0077] in, For the original category score, ; This represents the weight matrix of the third layer. ; This represents the bias term of the third layer. .

[0078] S56: Based on the original category scores, the classification output module outputs the category probabilities of various geological categories (including intact rock strata, concealed ignition source anomalous areas, goaf areas, and geological anomalies, including faults, fissures, and aquifers) distributed within a preset radial range around the borehole along the borehole depth direction using a probability normalization function, satisfying the following:

[0079] in, This is the normalized class probability vector; The original category score.

[0080] S6: Determine the preliminary geological category based on the maximum category probability.

[0081] Specifically, in step S6, a preliminary geological category is determined based on the maximum category probability, satisfying:

[0082] in, For the first The sampling point or the first The preliminary geological category corresponding to the pre-defined radial range around the borehole at each borehole depth location.

[0083] S7: Based on the bottom hole temperature or bottom hole temperature change rate from step S2, adjust the confidence level of the concealed fire source abnormal area identification results and determine whether to output drilling stop information.

[0084] Specifically, step S7 includes: Set preset probability thresholds, preset temperature thresholds, preset change rate thresholds, and recovery temperature thresholds. Based on the bottom hole temperature or bottom hole temperature change rate from step S2, if the probability of the concealed fire source anomalous area category is greater than the preset probability threshold, the bottom hole temperature is greater than or equal to the preset temperature threshold, or the bottom hole temperature change rate is greater than or equal to the preset change rate threshold, increase the confidence level of the concealed fire source anomalous area identification result and identify the concealed fire source anomalous area as a high-confidence concealed fire source anomalous area; if the probability of the concealed fire source anomalous area category is greater than the preset probability threshold, the bottom hole temperature is less than the preset temperature threshold, or the bottom hole temperature change rate is less than the preset change rate threshold, identify the concealed fire source anomalous area as a coal fire anomalous area to be verified; otherwise, identify the concealed fire source anomalous area as a low-confidence concealed fire source anomalous area; if the bottom hole temperature is greater than or equal to the preset temperature threshold or the bottom hole temperature change rate is greater than or equal to the preset change rate threshold, output drilling stop information; when the bottom hole temperature drops below the recovery temperature threshold and remains stable for a preset time, release the drilling stop information.

[0085] Specifically, the temperature of normal rock strata (shallow, <30m) is the local annual average temperature, while in deeper strata (>30m), the temperature increases by approximately 3°C for every 100m of depth, with higher temperatures at greater depths. In this embodiment, the preset temperature threshold can be 100°C~150°C, and the preset rate of change threshold can be 3°C / s~8°C / s. Preferably, the preset temperature threshold is 120°C, the preset rate of change threshold is 5°C / s, the recovery temperature threshold is 100°C, and the preset time is 5 minutes. When the probability of a concealed fire source anomaly at a certain depth is greater than the preset probability threshold, and the bottom temperature is greater than or equal to 120°C or the bottom temperature change rate is greater than or equal to 5°C / s, the concealed fire source anomaly is identified as a high-confidence concealed fire source anomaly, and a high-temperature warning is generated. When the probability of a concealed fire source anomaly is greater than the preset probability threshold, and the bottom temperature change rate is less than 120°C or less than 5°C / s, the concealed fire source anomaly is identified as a coal fire anomaly to be verified. If the bottom hole temperature is greater than or equal to 120℃ or the bottom hole temperature change rate is greater than or equal to 5℃ / s, a drilling stop message will be output and an audible and visual alarm will be triggered. The drilling stop message will be released when the bottom hole temperature drops below 100℃ and remains below 100℃ for 5 minutes.

[0086] S8: Generate a spatial distribution estimation model, and generate visualization results and detection reports based on the spatial distribution estimation model.

[0087] Specifically, step S8 includes: S81: Based on the drilling depth, sampling point number, and maximum borehole detection depth, establish the correspondence between sampling points and actual borehole depths. The formula is as follows:

[0088] in, For the first The actual borehole depth corresponding to each sampling point; In this embodiment, the initial drilling depth is... ; This represents the maximum drilling depth.

[0089] S82: Based on the correspondence between sampling points and the actual borehole depth, and according to the preliminary geological category in step S6, generate a comparison map of geological profile prediction results distributed within a preset radial distance range around the borehole along the borehole depth direction, such as... Figure 4 As shown.

[0090] S83: Based on the geological profile and the category probabilities of each geological type, a cylindrical coordinate system is established with the borehole axis as the Z-axis to generate spatial distribution estimation models for concealed ignition source anomaly zones, goaf areas, intact rock strata, faults, fissures, and aquifers. The spatial distribution estimation model is a visualized estimation result generated based on the borehole axis, depth coordinates, and category probabilities of geological types. It is used to assist field personnel in geological type identification and early warning decisions, and is not limited to absolute three-dimensional quantitative measurements of geological category boundaries.

[0091] S84: Generate visualization results and detection reports based on the spatial distribution estimation model. The detection report should include at least the borehole depth, geological category and category probability, geological category location, confidence level of concealed ignition source anomaly zone, bottom hole temperature change curve, and drilling stoppage warning information.

[0092] Specifically, based on the spatial distribution estimation model, the display density, radius range, or marker size of scatter points can be determined according to the category probability of geological types to form a spatial estimation map of geological anomalies distributed along the borehole axis. Hidden ignition source anomaly areas can be highlighted based on their category probability and temperature changes. Goaf areas can be distinguished and displayed based on their radar echo amplitude abrupt changes and spectral characteristics.

[0093] Specifically, in this invention, the method processes multi-frequency radar echo signals through a two-layer preprocessing architecture of basic preprocessing and advanced preprocessing. It then fuses radar echo signal features with borehole bottom temperature features to construct a radar-temperature fusion feature vector. A multimodal classification model outputs the probability of different geological categories distributed within a preset radial range around the borehole along the borehole depth direction, providing a multi-source information fusion basis for identifying concealed fire sources. Combining the probability of concealed fire source anomaly zone categories with borehole bottom temperature and borehole bottom temperature change rate allows for confidence level correction of concealed fire source anomaly zone identification results, reducing the risk of misjudgment caused by relying solely on a single radar echo or single-point temperature data. Finally, through geological profiles and spatial distribution estimation models, visualized results and detection reports are generated, facilitating identification and early warning decisions by on-site personnel.

[0094] Please see Figure 5 This embodiment also provides a concealed fire source detection device based on multimodal fusion recognition, used to implement the concealed fire source detection method described above. The device includes a drill bit 6, a first drill rod section 5, a detection connector 4, a main unit, and a ground host 1. The main unit includes several standard metal drill rods 3 and several repeater connectors 2, which are alternately connected. The spacing between adjacent repeater connectors 2 is 30m to 80m, preferably 50m. Each repeater connector 2 contains a signal relay repeater and a repeater power supply battery. The signal relay repeater is used to relay radar echo signals and temperature signals during deep hole drilling. The repeater power supply battery is electrically connected to the signal relay repeater, providing power to the signal relay repeater.

[0095] Specifically, a temperature sensor is installed at the end of the first drill rod 5 near the drill bit 6. This sensor collects the bottom hole temperature near the drill bit 6, and the temperature collection cycle can be set according to drilling speed and control requirements, for example, once per second. The drill bit 6 is installed at one end of the first drill rod 5, and a probe connector 4 is installed at the end of the first drill rod 5 furthest from the drill bit 6. The end of the probe connector 4 furthest from the first drill rod 5 is connected to the standard metal drill rod 3 of the main unit. The probe connector 4 and the standard metal drill rod 3 are of equal diameter. Both the drill bit 6 and the temperature sensor are electrically connected to the probe connector 4, which is electrically connected to a signal relay repeater. The signal relay repeater is electrically connected to the ground host unit 1. The ground host unit 1 receives radar echo signals and temperature signals, fuses the radar echo signals and temperature signals, and outputs the geological category probability, concealed fire source identification results, and drilling stop control signals.

[0096] For details, please refer to Figure 6 and Figure 7The probe connector 4 includes an outer load-bearing cylinder 41, which is used to withstand tensile, torque, and bending loads during drilling. The outer load-bearing cylinder 41 has threaded connection interfaces 42 at both ends. The probe connector 4 connects to the first drill rod section 5 through one threaded connection interface 42, and to the standard metal drill rod 3 through the other threaded connection interface 42. The inner wall of the outer load-bearing cylinder 41 is provided with a vibration damping layer 51, a heat insulation layer 50, and a sealing layer 49 from the outside to the inside. The sealing layer 49 reduces the risk of downhole water vapor and coal dust entering the interlayer cavity 44. The heat insulation layer 50 reduces the impact of external high temperatures on the PCB control board 47 and the power supply battery 48. The vibration damping layer 51 reduces the impact of drilling vibration on the radar transmitting antenna 45, the radar receiving antenna 46, and the PCB control board 47. The sealing layer 49, heat insulation layer 50, and vibration damping layer 51 can be made of rubber seals, heat-resistant insulating materials, elastic vibration damping materials, or combinations thereof, depending on the actual working conditions.

[0097] Specifically, the outer load-bearing cylinder 41 has a through-flow channel 43 inside, which is used to ensure the passage of fluid or air inside the drill pipe. The through-flow channel 43 has an hourglass-shaped structure, and the diameters at both ends of the through-flow channel 43 are larger than the diameter at the middle. This ensures the continuity of the internal flow channel while providing installation space for the interlayer cavity 44, allowing the radar element to be encapsulated inside the detection connector 4 without significantly increasing the outer diameter of the drill bit.

[0098] Specifically, the space between the outer wall of the inner flow channel 43 and the inner wall of the outer support cylinder 41 forms a sandwich cavity 44. The sandwich cavity 44 houses a radar transmitting antenna 45, a radar receiving antenna 46, a PCB control board 47, and a power supply battery 48. The radar transmitting antenna 45, radar receiving antenna 46, and PCB control board 47 are all electrically connected to the power supply battery 48, which supplies power to the radar transmitting antenna 45, radar receiving antenna 46, and PCB control board 47. In this embodiment, the power supply battery 48 can be a heat-resistant sealed power supply battery, an intrinsically safe power supply battery, or other power supply batteries suitable for downhole environments. The outer support cylinder 41 has an electromagnetic wave transmitting window 52 and an electromagnetic wave receiving window 53 made of a non-metallic wave-transparent material. In this embodiment, the non-metallic wave-transparent material is fiberglass. The electromagnetic wave transmitting window 52 and electromagnetic wave receiving window 53 are arranged radially opposite to each other. The radar transmitting antenna 45 is positioned facing the electromagnetic wave transmitting window 52, ​​and the radar receiving antenna 46 is positioned facing the electromagnetic wave receiving window 53. The multi-frequency electromagnetic wave signal emitted by the radar transmitting antenna 45 is radiated to the coal and rock mass around the borehole through the electromagnetic wave transmitting window 52. The reflected echo generated by the coal and rock mass interface or anomaly is received by the radar receiving antenna 46 through the electromagnetic wave receiving window 53.

[0099] Specifically, the PCB control board 47 includes a radar signal transmission control module, a radar echo receiving module, a temperature signal acquisition module, a power management module, and a communication module. The radar signal transmission control module controls the radar transmitting antenna 45 to output multi-frequency electromagnetic wave signals; the radar echo receiving module receives the reflected echo signals acquired by the radar receiving antenna 46; the temperature signal acquisition module receives the bottom hole temperature signals acquired by the temperature sensor; the power management module distributes and protects the power supply battery 48; and the communication module uploads the radar echo signals and bottom hole temperature signals to the ground host 1. When the drilling depth is large, the radar echo signals and temperature signals are forwarded via repeater connector 2. The radar transmitting antenna 45 is electrically connected to the radar signal transmission control module, the radar receiving antenna 46 is electrically connected to the radar echo receiving module, the temperature sensor is electrically connected to the temperature signal acquisition module, and the communication module is electrically connected to the signal relay repeater and the ground host 1.

[0100] Specifically, such as Figure 8 The high-temperature interlocking control process shown involves the ground host 1 receiving the bottom hole temperature collected by the temperature sensor in real time and calculating the temperature change rate. When the temperature is greater than or equal to the preset temperature threshold or the temperature change rate is greater than or equal to the preset change rate threshold, the ground host 1 outputs a stop drilling signal and triggers an audible and visual alarm. The drilling rig executes the stop drilling control, and the temperature sensor continuously monitors the temperature recovery. When the temperature is less than the recovery temperature threshold and remains stable for a preset time, the ground host 1 releases the stop drilling control and resumes drilling.

[0101] Specifically, the workflow of this device is as follows: Drilling and detection are performed at the downhole face or in front of the working face. The drill bit 6, the first drill rod section 5, the detection connector 4, the standard metal drill rod 3, and the repeater connector 2 are connected sequentially according to the drilling direction. The detection connector 4 is installed between the first drill rod section 5 and the standard metal drill rod 3, and the temperature sensor is located at the end of the first drill rod section 5 closest to the drill bit 6. Before starting drilling or in the initial stage of drilling, the ground host 1 sends a start or self-test command to the PCB control board 47, sequentially checking the working status of the radar signal transmission control module, radar echo receiving module, temperature signal acquisition module, power management module, communication module, and repeater connector 2. After the self-test is completed, the device enters the drilling detection state.

[0102] While the drilling rig starts drilling, the PCB control board 47 controls the radar transmitting antenna 45 to transmit multi-frequency electromagnetic wave signals to the surrounding coal and rock mass through the electromagnetic wave transmitting window 52. Different frequency signals are used to obtain the reflected echo characteristics of the coal and rock mass at different electromagnetic response scales. When the electromagnetic wave propagates in the coal and rock mass and encounters abnormal structures near interfaces with different dielectric properties, goafs, fracture zones, water-bearing anomalies, or hidden ignition source anomalies, reflected echoes are generated. The reflected echoes are received by the radar receiving antenna 46 through the electromagnetic wave receiving window 53 and transmitted to the PCB control board 47. The temperature sensor synchronously collects the bottom temperature near the drill bit 6. The PCB control board 47 receives the temperature signal and the radar wave echo signal and uploads them to the ground host 1 through the communication module. When the drilling depth is large, the radar wave echo signal and the temperature signal are forwarded through the repeater connector 2.

[0103] Because of the installation distance between the probe connector 4 and the drill bit 6, the ground host 1 can perform depth correction on the radar echo sampling position and the bottom hole temperature sampling position based on the length of the first drill rod 5, the installation position of the probe connector 4, and the real-time drilling depth, so that the radar echo signal and the bottom hole temperature signal correspond in the time or depth dimension. The ground host 1 preprocesses the received radar echo signal. After preprocessing, the ground host 1 performs multimodal fusion identification on the radar echo signal and the temperature signal, outputs the category probability of each geological category, and corrects the confidence level of the identification results of the concealed fire source anomaly area and whether to output drilling stop information. Then, the ground host 1 maps the classification results to the actual drilling depth according to the sampling point sequence number and the maximum drilling depth, generates a geological profile along the drilling depth direction, and further generates a spatial distribution estimation model. Based on the spatial distribution estimation model, it generates visualization results and a detection report.

[0104] Specifically, in this invention, the radar transmitting antenna 45, radar receiving antenna 46, PCB control board 47, and power supply battery 48 are disposed within the interlayer cavity 44 of the probe connector 4. An electromagnetic wave transmitting window 52 and an electromagnetic wave receiving window 53 made of non-metallic wave-transparent material are provided on the probe connector 4, which can reduce the shielding effect of the standard metal drill rod 3 on radar electromagnetic wave transmission and reception. The probe connector 4 is positioned between the first section of drill rod 5 and the standard metal drill rod 3, maintaining a certain distance between the probe connector 4 and the high-temperature cutting zone of the drill bit 6. Simultaneously, a temperature sensor is provided at the end of the first section of drill rod 5 near the drill bit 6, which helps reduce the impact of the cutting heat of the drill bit 6 and the high temperature of the coal fire on the probe connector 4. The signal is forwarded through the repeater connector 2, which is beneficial for adapting to the data transmission requirements during deep hole drilling.

[0105] The above embodiments merely illustrate the basic principles and characteristics of the present invention, but are not limited to the above implementation schemes. It should be understood that those skilled in the art can make various changes and modifications to the present invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for detecting concealed fire sources based on multimodal fusion recognition, characterized in that, Includes the following steps: S1: Drill along the drilling direction and simultaneously transmit multi-frequency radar wave signals to the coal and rock mass around the borehole; S2: Use a temperature sensor to collect the temperature at the bottom of the borehole; S3: Preprocess the radar echo signal; S4: The radar echo signal features preprocessed in step S3 are fused with the borehole bottom temperature features obtained in step S2 to construct a radar-temperature fusion feature vector. S5: Input the radar-temperature fusion feature vector from step S4 into the multimodal classification model. The multimodal classification model outputs the class probabilities of various geological categories distributed within a preset radial range around the borehole along the borehole depth direction. S6: Determine the preliminary geological category based on the maximum category probability; S7: Based on the bottom hole temperature or bottom hole temperature change rate from step S2, adjust the confidence level of the identification results of the hidden fire source abnormal area and determine whether to output drilling stop information. S8: Generate a spatial distribution estimation model, and generate visualization results and detection reports based on the spatial distribution estimation model.

2. The concealed fire source detection method based on multimodal fusion recognition according to claim 1, characterized in that, In step S3, the preprocessing includes basic preprocessing and advanced preprocessing. The basic preprocessing includes pre-amplification, A / D conversion, moving average filtering, and zero-mean normalization. The advanced preprocessing includes median filtering, spectral feature extraction, and outlier removal.

3. The concealed fire source detection method based on multimodal fusion recognition according to claim 1, characterized in that, Step S4 specifically includes: S41: After preprocessing, the radar echo signal characteristics and bottom hole temperature characteristics are aligned by time or drilling distance according to the sampling time, drilling distance or sampling point number. S42: For the same sampling point or the same borehole depth, construct a radar-temperature fusion feature vector that satisfies: in, , , , The first The sampling point or the first Radar echo amplitude characteristics at different frequency bands at various borehole depths; For the first The sampling point or the first Spectral characteristics at each borehole depth location; For the first The sampling point or the first Bottom hole temperature characteristics at each borehole depth location; Characterized by the rate of temperature change; For the first The sampling point or the first The radar-temperature fusion feature vector corresponding to each borehole depth location.

4. The concealed fire source detection method based on multimodal fusion recognition according to claim 3, characterized in that, In step S5, the multimodal classification model includes a feature projection module, an attention mechanism module, and a classification output module. The attention mechanism module includes a query linear transformation layer, a key linear transformation layer, and a value linear transformation layer. The feature projection module is used to map radar echo features and borehole bottom temperature features to the same feature dimension. The attention mechanism module is used to assign weights to radar echo features and borehole bottom temperature features of different frequency bands to obtain weighted fusion features. The classification output module is used to output the category probabilities of various geological categories distributed within a preset radial range around the borehole along the borehole depth direction.

5. The concealed fire source detection method based on multimodal fusion recognition according to claim 4, characterized in that, Step S5 specifically includes: S51: Combine the radar-temperature fusion feature vector from step S42 The input feature projection module obtains the hidden feature vector through linear transformation, satisfying the following: in, This represents the hidden feature vector after mapping; Represents the feature projection weight matrix; Represents the bias vector; S52: The query linear transformation layer, key linear transformation layer, and value linear transformation layer perform linear transformations on the hidden feature vectors respectively to obtain the query vector, key vector, and value vector, satisfying: in, For query vector; Represents the key vector; Represents a value vector; This indicates a query for the weight matrix of the linear transformation layer; This represents the weight matrix of the key-linear transformation layer; This represents the weight matrix of the linear transformation layer; S53: Calculate the attention weights, satisfying: in, Indicates attention weight; Indicates feature dimension; Indicates the scaling factor; S54: Calculate the weighted fusion features, satisfying: in, For weighted fusion features; S55: Based on weighted fusion characteristics The classification output module obtains the original category scores; S56: Based on the original category scores, the classification output module outputs the category probabilities of various geological categories distributed within a preset radial range around the borehole along the borehole depth direction using a probability normalization function, satisfying: in, This is the normalized class probability vector; The original category score.

6. The concealed fire source detection method based on multimodal fusion recognition according to claim 5, characterized in that, In step S56, the geological categories include intact rock strata, hidden fire source abnormal areas, goaf areas, and geological anomalies. Geological anomalies include faults, fissures, and aquifers.

7. The concealed fire source detection method based on multimodal fusion recognition according to claim 6, characterized in that, In step S6, the preliminary geological category is determined based on the maximum category probability, satisfying the following: in, For the first The sampling point or the first The preliminary geological category corresponding to the pre-defined radial range around the borehole at each borehole depth location.

8. The concealed fire source detection method based on multimodal fusion recognition according to claim 7, characterized in that, Step S7 specifically includes: Set a preset probability threshold, a preset temperature threshold, a preset rate of change threshold, and a recovery temperature threshold. Based on the bottom temperature or the rate of change of the bottom temperature in step S2, if the probability of the hidden fire source abnormal area category is greater than the preset probability threshold, the bottom temperature is greater than or equal to the preset temperature threshold, or the rate of change of the bottom temperature is greater than or equal to the preset rate of change threshold, increase the confidence level of the hidden fire source abnormal area identification result, and identify the hidden fire source abnormal area as a high-confidence hidden fire source abnormal area. If the probability of a hidden fire source abnormal zone is greater than a preset probability threshold, the temperature at the bottom of the hole is less than a preset temperature threshold, and the rate of change of the temperature at the bottom of the hole is less than a preset rate of change threshold, then the hidden fire source abnormal zone will be identified as a coal fire abnormal zone to be reviewed. In other cases, areas with concealed fire sources are identified as low-confidence areas with concealed fire sources. If the bottom hole temperature is greater than or equal to the preset temperature threshold or the bottom hole temperature change rate is greater than or equal to the preset change rate threshold, a drilling stop message is output. The drilling stop message is released when the bottom hole temperature drops below the recovery temperature threshold and remains stable for a preset time.

9. A concealed fire source detection device based on multimodal fusion recognition, used to implement the concealed fire source detection method based on multimodal fusion recognition as described in any one of claims 1-8, characterized in that, The system includes a drill bit (6), a first drill rod section (5), a probe connector (4), a main unit, and a ground host (1). The main unit includes several standard metal drill rods (3) and several repeater connectors (2). The standard metal drill rods (3) and repeater connectors (2) are alternately connected. Each repeater connector (2) contains a signal relay repeater and a repeater power supply battery. The repeater power supply battery is electrically connected to the signal relay repeater. The first drill rod section (5) has a [missing information - likely a reference to a drill bit (6)] end near the drill bit (6). A temperature sensor is installed at one end of the first drill rod (5), and a probe connector (4) is installed at the end of the first drill rod (5) away from the drill bit (6). The end of the probe connector (4) away from the first drill rod (5) is connected to the standard metal drill rod (3) of the main unit. The drill bit (6) and the temperature sensor are both electrically connected to the probe connector (4). The probe connector (4) is electrically connected to the signal relay, and the signal relay is electrically connected to the ground host (1).

10. The concealed fire source detection device based on multimodal fusion recognition according to claim 9, characterized in that, The detection connector (4) includes an outer load-bearing cylinder (41), with threaded connection interfaces (42) at both ends. A vibration damping layer (51), a heat insulation layer (50), and a sealing layer (49) are sequentially arranged on the inner wall of the outer load-bearing cylinder (41) from the outside to the inside. A through-flow channel (43) is provided inside the outer load-bearing cylinder (41). The through-flow channel (43) has an hourglass-shaped structure, with the diameters at both ends of the through-flow channel (43) larger than the diameter at the middle. The space between the outer wall of the through-flow channel (43) and the inner wall of the outer load-bearing cylinder (41) forms a sandwich cavity (44). The sandwich cavity (44) is equipped with a radar transmitting antenna (45), a radar receiving antenna (46), a PCB control board (47), and a power supply. The battery (48), the radar transmitting antenna (45), the radar receiving antenna (46), and the PCB control board (47) are all electrically connected to the power supply battery (48). The radar transmitting antenna (45) and the radar receiving antenna (46) are both electrically connected to the PCB control board (47). The outer load-bearing cylinder (41) is provided with an electromagnetic wave transmitting window (52) and an electromagnetic wave receiving window (53) made of non-metallic wave-transparent material. The radar transmitting antenna (45) is set facing the electromagnetic wave transmitting window (52), and the radar receiving antenna (46) is set facing the electromagnetic wave receiving window (53). The temperature sensor is electrically connected to the PCB control board (47), and the PCB control board (47) is electrically connected to the signal relay and the ground host (1).

Images