Safety monitoring method and system based on multi-source data fusion

Abstract

This invention discloses a safety monitoring method and system based on multi-source data fusion, relating to the field of data fusion technology. After collecting gas data from each sensor, multi-channel spatiotemporal convolution is performed to determine the predicted gas concentration points and confidence level intervals for each gas collection sensor's corresponding monitoring point. CT images of the target metal ore area are acquired and substituted into a preset metal crack feature extraction model to obtain target metal crack features. Based on the target metal crack features and the predicted gas concentration points and confidence level intervals for each monitoring point, a gas concentration distribution field for the target metal ore area is constructed. The gas concentration distribution field is then input into a preset safety monitoring model to obtain safety monitoring results. This method deeply couples and fuses gas monitoring information with metal crack structure information, and performs safety monitoring based on the gas concentration distribution field, significantly improving the accuracy, stability, and intelligence of metal mine safety monitoring, effectively enhancing early warning reliability and overall monitoring efficiency.

Description

Technical Field

[0001] This invention belongs to the field of data fusion technology, specifically relating to a security monitoring method and system based on multi-source data fusion. Background Technology

[0002] Metal mining is a core pillar industry for energy and mineral resource supply. Its safe production is not only related to the life and property safety of underground workers, but also directly affects the stability of mineral resource supply and the safety of the ecological environment. As metal mining extends to deeper areas, the underground geological structure becomes increasingly complex, and safety risks such as rock bursts, roof falls, and the accumulation and leakage of toxic and harmful gases become increasingly prominent. This places stringent demands on the accuracy, comprehensiveness, and intelligence of metal mine safety monitoring technology. Traditional metal mine safety monitoring methods can no longer meet the safety production assurance needs of deep metal mines.

[0003] Currently, the field of metal mine safety monitoring has mainly formed two core technology paths: one is downhole environmental monitoring with gas parameter monitoring as the core, and the other is geological condition monitoring with metal ore rock mass structure monitoring as the core. However, in existing technologies, these two monitoring paths are independent and disconnected, failing to achieve effective data fusion and collaborative analysis, resulting in significant technical bottlenecks and making it difficult to comprehensively and accurately assess the downhole safety status.

[0004] With the rapid development of intelligent mining technology for metal mines, multi-source data fusion has become a core development direction for improving the level of safety monitoring in metal mines. However, most existing multi-source data fusion technologies are limited to simple data splicing or shallow fusion, failing to achieve deep fusion of gas sensor data and CT metal crack image data, and lacking a collaborative processing mechanism for these two types of heterogeneous data. On the one hand, there is a lack of efficient feature extraction and enhancement methods, making it impossible to accurately extract metal crack features; on the other hand, it fails to construct a global gas concentration distribution field that combines metal crack features, gas concentration prediction values, and confidence levels, making it impossible to transition from point monitoring to surface monitoring. This results in monitoring results that cannot comprehensively and accurately reflect the underground safety status, leading to low efficiency in underground safety monitoring of metal mines. Summary of the Invention

[0005] The purpose of this invention is to solve the problem that existing multi-source data fusion is only a shallow stitching, which cannot achieve deep fusion of gas data and CT metal crack data, resulting in low efficiency of metal mine safety monitoring. Therefore, this invention proposes a safety monitoring method and system based on multi-source data fusion.

[0006] In a first aspect of this invention, a security monitoring method based on multi-source data fusion is first proposed, wherein a gas collection sensor array is set up in the target metal mining area, the method comprising: Gas data is collected synchronously from each sensor to obtain a gas data array; Multi-channel spatiotemporal convolution is performed on the gas data array to determine the predicted gas concentration point value and confidence level interval prediction value for each gas acquisition sensor in the gas acquisition sensor array at the corresponding monitoring point. CT images of the target metal mining area are acquired, and the CT images are substituted into a preset metal fracture feature extraction model to obtain the target metal fracture features. Based on the target metal fracture characteristics and the predicted values ​​of gas concentration points and confidence level intervals at the corresponding monitoring points of each gas acquisition sensor, a gas concentration distribution field of the target metal ore area is constructed. The gas concentration distribution field is input into a preset safety monitoring model to obtain safety monitoring results.

[0007] By employing multi-channel spatiotemporal convolution to achieve deep spatiotemporal feature mining of gas data arrays, the predicted values ​​and confidence level intervals of gas concentration at each monitoring point are accurately output. Simultaneously, the core features of the geological structure of metal mines are obtained by combining the CT metal fracture feature extraction model. This breaks through the limitations of traditional multi-source data that are only shallowly stitched together, realizing deep coupling and feature fusion of gas monitoring data and metal fracture structure data. Based on this, a high-precision gas concentration distribution field of metal mine areas is constructed. Finally, reliable results are output through the safety monitoring model, effectively improving the accuracy, stability and intelligence level of metal mine safety monitoring, and significantly improving the efficiency and reliability of metal mine safety monitoring and early warning.

[0008] Optionally, performing multi-channel spatiotemporal convolution on the gas data array to determine the predicted gas concentration point value and confidence level interval value for each gas acquisition sensor in the gas acquisition sensor array includes: Acquire gas data sets collected by the sensor array, and preprocess the gas data sets to obtain effective gas data sets; Spatiotemporal correlation analysis was performed on the effective gas data set to obtain a spatiotemporal data set; IMF decomposition is performed on each set of data in the spatiotemporal data group to obtain the intrinsic mode function set corresponding to each set of data; Clustering the intrinsic modulus functions in all intrinsic modulus function sets by a preset frequency magnitude yields low-frequency trend component sets, mid-frequency trend component sets, and high-frequency trend component sets. Spatiotemporal features are obtained by inputting independent prediction channels for each trend component set; the independent prediction channels include cascaded spatiotemporal convolutional modules; Based on the attention mechanism of modal channels, the attention weight of each channel is calculated and the spatiotemporal features of the three channels are weighted and fused to obtain the predicted gas concentration point value and confidence level interval value of each monitoring point in the sensor array.

[0009] By decomposing the raw gas data from the sensor array into trend components of different frequencies, accurately extracting spatiotemporal features in independent channels and adaptively weighting and fusing them, the accuracy and stability of gas concentration point prediction are greatly improved, and the confidence level range is output simultaneously, enabling reliable, robust, and interpretable accurate quantitative prediction of gas concentration change trends.

[0010] Optionally, the spatiotemporal convolution module consists of a graph convolutional network and a temporal convolutional network connected in series.

[0011] Optionally, substituting the CT image into a preset metal crack feature extraction model to obtain the target metal crack features includes: The input CT image is acquired, and a 1×1 convolution operation is performed on the CT image to obtain the first metallic mineral convolution feature; Substituting the first metal ore convolution feature into the improved multi-channel convolution feature module yields the first improved metal ore convolution feature. The first improved metal ore convolution feature is sequentially substituted into the max pooling module and the improved multi-channel convolution feature module to obtain the second improved metal ore convolution feature. The second improved metal ore convolution feature is successively substituted into the max pooling module and the improved multi-channel convolution feature module to obtain the third improved metal ore convolution feature; Substituting the second improved metal ore convolution feature into the ECA module yields the enhanced metal ore convolution feature; After upsampling the enhanced metal ore convolutional features, they are fused with the first improved metal ore convolutional features and substituted into the ECA module to obtain the first fused enhanced metal ore convolutional features; After downsampling the enhanced metallic ore convolutional features, they are fused with the third improved metallic ore convolutional features and substituted into the ECA module to obtain the second fused enhanced metallic ore convolutional features; The target metal fracture feature is obtained by fusing the first fused enhanced metal ore convolution feature and the second fused enhanced metal ore convolution feature.

[0012] By improving the multi-channel convolutional feature module to extract multi-scale metal crack features hierarchically, using max pooling to achieve feature downsampling and semantic enhancement, and using the ECA module to accurately enhance the response of low-contrast metal crack features, and combining the progressive design of cross-scale feature fusion through upsampling and downsampling with secondary enhancement through ECA, we have achieved hierarchical capture of metal crack features from shallow edges to deep semantics. The ECA module has also specifically solved the problems of gray-scale overlap and feature blurring between cracks and the matrix in CT images of metal ore rock masses. At the same time, cross-scale fusion effectively makes up for the information loss of single-scale feature extraction, so that the fused target metal crack features retain fine-grained details such as microcracks and crack edges, while strengthening global semantic features such as crack connectivity and overall morphology, which greatly improves the completeness, accuracy and recognizability of metal crack features.

[0013] Optionally, the working principle of the improved multi-channel convolutional feature module is as follows: The input features are obtained, and after multi-directional single-channel convolution, 1×1 pointwise convolution is performed to obtain the first fused channel convolution feature; the multi-directional features include the horizontal direction, the vertical direction, the left tilt direction, and the right tilt direction; Global average pooling is applied to the convolutional features of the first fused channel to obtain global average pooling features; The target convolution kernel is determined based on the number of channels of the global average pooling feature, and a second fused channel convolution feature is obtained by performing a 1D convolution operation on the global average pooling feature according to the target convolution kernel. Substitute the convolutional features of the second fusion channel into the Sigmoid activation function to obtain the target weights; The output feature is obtained by fusing the second fusion channel convolutional feature and the first fusion channel convolutional feature according to the target weight.

[0014] Optionally, determining the target convolutional kernel based on the number of channels of the global average pooling feature includes: Through formula Determine the target convolutional kernel size, where γ=2, b=1, and C is the number of channels. , indicates that the calculation result is rounded down.

[0015] By combining multi-directional single-channel convolution and point-by-point convolution, fine-grained spatial features of cracks in CT images of metal ore rock masses in different directions (horizontal, vertical, oblique) are captured from all angles, solving the problem that single-directional convolution easily misses multi-dimensional crack morphology. At the same time, by combining global average pooling, adaptive one-dimensional convolution, and channel target weights generated by Sigmoid activation, dynamic weighted enhancement of crack feature channels is achieved. This can specifically enhance the response of low-contrast crack-related channels and suppress background noise channel interference. Furthermore, the weight fusion stage combines the attention-enhanced features with the initial fused channel features, which not only preserves the complete spatial features extracted from multiple directions but also improves the recognizability and effectiveness of features through channel attention. This significantly improves the ability of convolutional features to represent metal cracks and lays a high-quality feature foundation for further fusion and segmentation of metal crack features.

[0016] Optionally, based on the target metal fracture characteristics and the predicted gas concentration points and confidence level intervals of each gas acquisition sensor at the corresponding monitoring points, the gas concentration distribution field of the target metal ore area is constructed as follows: The target metal fracture features are mapped to a preset three-dimensional spatial grid corresponding to the target metal ore region, and the gas diffusion coefficient and permeability coefficient of each grid cell are calibrated according to preset calibration rules; The predicted values ​​of gas concentration points and confidence level intervals are bound to the corresponding sensor units of the preset three-dimensional spatial grid, and the weight coefficients of the corresponding units are obtained by weighting according to the predicted values ​​of confidence level intervals. A physical field model for gas diffusion in a two-pore medium of a metal ore rock mass is constructed based on the calibrated gas diffusion coefficient and permeability coefficient. The predicted values ​​of the bound gas concentration points are used as sample points. By combining the physical field model and the confidence weighting coefficient through Kriging interpolation, the gas concentration values ​​of all grid cells are completed to obtain the initial gas concentration distribution field. Verify the matching degree between the initial gas concentration distribution field and the target metal crack characteristics. For grid cells with matching degrees exceeding the preset threshold, re-call the physical field model to correct the gas diffusion coefficient and perform interpolation calculation again until the matching degree of all grid cells falls within the threshold range to obtain the gas concentration distribution field.

[0017] By mapping metal fracture features to a three-dimensional spatial grid and calibrating the gas diffusion and permeability coefficients, and combining the predicted gas concentration points with their confidence level intervals for weighting, the entire concentration distribution field is completed based on the physical field model of gas diffusion in the dual-pore medium of the metal ore rock mass and confidence-weighted kriging interpolation. Finally, the accuracy of the distribution field is optimized through matching degree verification and iterative correction. This approach can fully utilize the physical constraint of the metal rock fracture structure on gas migration, effectively improving the authenticity and reliability of the gas concentration distribution field. At the same time, the confidence level is used to achieve adaptive weighting of the reliability of sensor data, reducing the interference of low-quality monitoring data on the construction of the distribution field. Ultimately, a more accurate and robust global gas concentration distribution field that better reflects the actual geological environment of the metal ore is obtained.

[0018] Optionally, the weighting coefficients for the corresponding units obtained by weighting based on the predicted values ​​of the confidence level intervals include: Let the predicted confidence level interval value of the monitoring point corresponding to the i-th sensor be denoted as . Define the confidence interval width for this monitoring point. for Weighting coefficients based on confidence interval width Divide , The first preset width threshold, The second preset width threshold is satisfied. .

[0019] Optionally, the target metal crack features include crack density and permeability coefficient; verifying the matching degree between the initial gas concentration distribution field and the target metal crack features includes: Through formula Determine the matching degree corresponding to the i-th grid cell; in, Let be the initial gas concentration value of the i-th grid cell. Let be the crack density of the i-th grid cell. Let be the permeability coefficient of the i-th grid cell, and α be the constant correlation coefficient of metal crack concentration.

[0020] In a second aspect of this invention, a security monitoring system based on multi-source data fusion is proposed, comprising: A gas data acquisition module is used to synchronously acquire gas data from each sensor to obtain a gas data array; a gas acquisition sensor array is set up in the target metal mining area; The gas prediction value generation and acquisition module is used to perform multi-channel spatiotemporal convolution on the gas data array to determine the gas concentration point prediction value and confidence level interval prediction value of each gas acquisition sensor in the gas acquisition sensor array corresponding to the monitoring point. The target metal fracture feature acquisition module is used to acquire CT images of the target metal ore area and substitute the CT images into a preset metal fracture feature extraction model to obtain the target metal fracture features. The gas concentration distribution field construction module is used to construct the gas concentration distribution field of the target metal mine area based on the target metal fracture characteristics and the predicted values ​​of gas concentration points and confidence level intervals of the corresponding monitoring points of each gas acquisition sensor. The safety monitoring module is used to input the gas concentration distribution field into a preset safety monitoring model to obtain safety monitoring results.

[0021] Optionally, the gas prediction value generation and acquisition module includes: The gas data preprocessing module is used to acquire gas data sets collected by the sensor array and preprocess the gas data sets to obtain effective gas data sets. The spatiotemporal data group determination module is used to perform spatiotemporal correlation analysis on the effective gas data group to obtain the spatiotemporal data group; The mode decomposition module is used to perform IMF decomposition on each group of data in the spatiotemporal data group to obtain the intrinsic mode function set corresponding to each group of data. The intrinsic modulus function clustering module is used to cluster the intrinsic modulus functions in all intrinsic modulus function sets according to a preset frequency to obtain low-frequency trend component sets, mid-frequency trend component sets, and high-frequency trend component sets; The spatiotemporal feature extraction module is used to obtain spatiotemporal features by inputting independent prediction channels for each trend component set; the independent prediction channels include cascaded spatiotemporal convolution modules; The spatiotemporal feature weighted fusion module is used to calculate the attention weight of each channel based on the attention mechanism of the modal channels and perform weighted fusion of the spatiotemporal features of the three channels to obtain the predicted gas concentration point value and confidence level interval value of each monitoring point in the sensor array.

[0022] Optionally, the target metal crack feature acquisition module includes: The first metal ore convolution feature generation module is used to acquire the input CT image and perform a 1×1 convolution operation on the CT image to obtain the first metal ore convolution feature. The first improved metal ore convolution feature generation module is used to substitute the first metal ore convolution feature into the improved multi-channel convolution feature module to obtain the first improved metal ore convolution feature. The second improved metal ore convolution feature generation module is used to sequentially substitute the first improved metal ore convolution feature into the max pooling module and the improved multi-channel convolution feature module to obtain the second improved metal ore convolution feature. The third improved metal ore convolution feature generation module is used to sequentially substitute the second improved metal ore convolution feature into the max pooling module and the improved multi-channel convolution feature module to obtain the third improved metal ore convolution feature. An enhanced metal ore convolution feature generation module is used to substitute the second improved metal ore convolution feature into the ECA module to obtain the enhanced metal ore convolution feature; The first fusion-enhanced metal ore convolution feature generation module is used to upsample the enhanced metal ore convolution feature and fuse it with the first improved metal ore convolution feature, and then substitute it into the ECA module to obtain the first fusion-enhanced metal ore convolution feature. The second fusion-enhanced metal ore convolution feature generation module is used to downsample the enhanced metal ore convolution feature and fuse it with the third improved metal ore convolution feature, and then substitute it into the ECA module to obtain the second fusion-enhanced metal ore convolution feature. The target metal crack feature generation module is used to fuse the first fused enhanced metal ore convolution feature and the second fused enhanced metal ore convolution feature to obtain the target metal crack feature.

[0023] Optionally, the working principle of the improved multi-channel convolutional feature module is as follows: The input features are obtained, and after multi-directional single-channel convolution, 1×1 pointwise convolution is performed to obtain the first fused channel convolution feature; the multi-directional features include the horizontal direction, the vertical direction, the left tilt direction, and the right tilt direction; Global average pooling is applied to the convolutional features of the first fused channel to obtain global average pooling features; The target convolution kernel is determined based on the number of channels of the global average pooling feature, and a second fused channel convolution feature is obtained by performing a 1D convolution operation on the global average pooling feature according to the target convolution kernel. Substitute the convolutional features of the second fusion channel into the Sigmoid activation function to obtain the target weights; The output feature is obtained by fusing the second fusion channel convolutional feature and the first fusion channel convolutional feature according to the target weight.

[0024] Optionally, the gas concentration distribution field construction module includes: The grid unitization module is used to map the crack features of the target metal to a preset three-dimensional spatial grid corresponding to the target metal ore region, and to calibrate the gas diffusion coefficient and permeability coefficient of each grid unit according to preset calibration rules; The unit weight coefficient determination module is used to bind the gas concentration point prediction value and the confidence level interval prediction value to the corresponding sensor unit of the preset three-dimensional spatial grid, and to obtain the weight coefficient of the corresponding unit by weight division according to the confidence level interval prediction value. The initial gas concentration distribution field generation module is used to construct a physical field model of gas diffusion in a two-pore medium of a metal ore rock mass based on the calibrated gas diffusion coefficient and permeability coefficient. Using the bound gas concentration point prediction value as the sample point, the module completes the gas concentration values ​​of all grid cells by combining the physical field model and the confidence weighting coefficient through Kriging interpolation to obtain the initial gas concentration distribution field. The gas concentration distribution field update module is used to verify the matching degree between the initial gas concentration distribution field and the target metal crack features. For grid cells whose matching degree exceeds the preset threshold, the physical field model is called again to correct the gas diffusion coefficient, and interpolation calculation is performed again until the matching degree of all grid cells falls within the threshold range to obtain the gas concentration distribution field.

[0025] The beneficial effects of this invention are: This invention proposes a safety monitoring method based on multi-source data fusion. By performing deep spatiotemporal feature extraction on a gas data array through multi-channel spatiotemporal convolution, it can accurately output the predicted values ​​and confidence intervals of gas concentration points at each monitoring point. At the same time, relying on the CT metal fracture feature extraction model, it obtains key features of the geological structure of metal mines, overcoming the shortcomings of simply splicing traditional multi-source data. This achieves deep coupling and feature fusion of gas monitoring information and metal fracture structure information, and constructs a high-precision gas concentration distribution field in the metal mine area. Then, through the safety monitoring model, it outputs stable and reliable monitoring conclusions, thereby significantly improving the accuracy, stability, and intelligence of metal mine safety monitoring, and effectively improving the reliability of early warning and the overall monitoring efficiency. Attached Figure Description

[0026] The invention will now be further described with reference to the accompanying drawings.

[0027] Figure 1 A flowchart illustrating a security monitoring method based on multi-source data fusion provided in this embodiment of the invention; Figure 2This invention provides a data flow diagram of a preset metal crack feature extraction model. Figure 3 This is a framework diagram of a security monitoring system based on multi-source data fusion, provided as an embodiment of the present invention. Detailed Implementation

[0028] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0029] Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] This invention provides a security monitoring method based on multi-source data fusion. See also... Figure 1 , Figure 1 The flowchart illustrates a security monitoring method based on multi-source data fusion, provided as an embodiment of the present invention. A gas sampling sensor array is installed in the target metal mining area. The method includes the following steps: S101, synchronously acquire gas data from each sensor to obtain a gas data array; S102, Perform multi-channel spatiotemporal convolution on the gas data array to determine the predicted gas concentration point value and confidence level interval prediction value of each gas acquisition sensor in the gas acquisition sensor array for the corresponding monitoring point. S103: Acquire CT images of the target metal mining area and substitute the CT images into a preset metal fracture feature extraction model to obtain the target metal fracture features; S104. Based on the crack characteristics of the target metal and the predicted values ​​of gas concentration points and confidence level intervals of the corresponding monitoring points of each gas acquisition sensor, a gas concentration distribution field of the target metal mine area is constructed. S105, input the gas concentration distribution field into the preset safety monitoring model to obtain the safety monitoring results.

[0031] In one implementation, metal fractures are represented as rock fissures within a metal ore deposit. CT images cover different geological strata, mining-affected areas, and potential fracture development zones within the target metal ore deposit, clearly showcasing the distribution characteristics of micro and macro fractures within the metal ore rock mass. The acquired CT images undergo sequential noise reduction, enhancement, and segmentation preprocessing to remove imaging noise and irrelevant interference information. The characteristics of the target metal fractures include fracture density, fracture connectivity, permeability coefficient, pore distribution, fracture orientation, and fracture width.

[0032] In one implementation, the preset safety monitoring model is the M1 spatiotemporal prediction model GCN-GRU-M1 based on graph convolutional networks and gated recurrent units. The safety monitoring results include safety level, concern level, and danger level. The safety level is when the gas concentration is within the normal range (determined by technical personnel) and there is no risk of abnormal accumulation. The concern level is when the gas concentration is in the normal-to-high range (above the normal value but below the safety threshold, which is determined by technical personnel) and requires continuous monitoring. The danger level is when the gas concentration exceeds the safety threshold and there is a risk of exceeding the limit of toxic and harmful gases.

[0033] In one embodiment, performing multi-channel spatiotemporal convolution on the gas data array to determine the predicted gas concentration point value and confidence level interval prediction value for each gas acquisition sensor in the gas acquisition sensor array includes: Acquire gas data sets collected by the sensor array, and preprocess the gas data sets to obtain effective gas data sets; Spatiotemporal correlation analysis was performed on the effective gas data set to obtain the spatiotemporal data set; IMF decomposition is performed on each set of data in the spatiotemporal data set to obtain the intrinsic mode function set corresponding to each set of data. Clustering the intrinsic modulus functions in all intrinsic modulus function sets by a preset frequency magnitude yields low-frequency trend component sets, mid-frequency trend component sets, and high-frequency trend component sets. Spatiotemporal features are obtained by inputting independent prediction channels for each trend component set; the independent prediction channels include cascaded spatiotemporal convolutional modules; Based on the attention mechanism of modal channels, the attention weight of each channel is calculated and the spatiotemporal features of the three channels are weighted and fused to obtain the predicted gas concentration point value and confidence level interval value of each monitoring point in the sensor array.

[0034] In one implementation, after the sensor collects gas data, it performs gas data preprocessing, which includes outlier removal, missing value interpolation, and standardization. The gas data array also contains the time-series acquisition information of each gas acquisition sensor and its spatial location information within the target metal mining area. The confidence level interval prediction value is used to characterize the reliability of the prediction value of the corresponding gas concentration point, providing a reliable basis for subsequent adaptive spatial mapping and weighted fusion.

[0035] In one implementation, based on a dynamic programming strategy, the spatiotemporal data set is obtained by extracting the results of time correlation analysis (Dynamic Time Distortion (DTW) value, periodicity) and spatial correlation analysis (Mutual Information Method (MI) value, correlation coefficient, adjacency matrix) from the time series of any two monitoring points. The extraction process is based on existing technology and will not be described in detail.

[0036] In one implementation, based on a variational optimization framework, the M-dimensional time series (M being the number of monitoring points) is decomposed into k intrinsic mode functions (IMFs), ensuring that the center frequencies of IMFs of the same order at different monitoring points are aligned, maintaining spatiotemporal consistency. Since the traditional ADMM algorithm is serial, serial updates lead to low efficiency and high computational complexity in multi-sensor data decomposition. Therefore, it is modified to be parallel. The parallel optimization operation is as follows: Modal component updates are performed in parallel along the sensor dimension, with the following formula: ,in This represents the update value of the k-th modal component at the m-th monitoring point in the (j+1)-th iteration. For the j-th iteration, the k-th modal component of the m-th monitoring point is... Let λ be the mean of the k-th modal component of all monitoring points in the j-th iteration, λ be the Lagrange multiplier, and M be the number of monitoring points. Let $k$ be the global mean of the k-th modal component across all monitoring points. For all possible Among the possible values, find the one that minimizes the objective function within the curly braces, and use it as the update result for the next iteration.

[0037] In one implementation, each set of data is decomposed using an improved multivariate variational model based on a preset frequency (determined by technical personnel) to obtain low-frequency trend components, mid-frequency trend components, and high-frequency trend components. Then, all low-frequency trend components are aggregated to obtain a low-frequency trend component set, and similarly, mid-frequency trend component sets and high-frequency trend component sets are obtained. The spatiotemporal convolution module consists of a graph convolutional network (GCN) and a temporal convolutional network (TCN) connected in series, with three STGCM modules connected in series for each channel.

[0038] In one implementation, for each monitoring point m, the spatiotemporal features of the three channels are multiplied by their corresponding weights and then summed to obtain a fused feature vector. The fused features are then input into a fully connected layer, and the output is the predicted gas concentration value for each monitoring point at a future preset time step. Based on the residual statistical characteristics of the point prediction, a 90% confidence interval is constructed. The residual between the predicted point value and the actual concentration value is calculated on the validation set. The upper and lower limits of the interval are determined by the residual following a normal distribution to obtain the predicted value of the confidence level interval.

[0039] In one embodiment, substituting CT images into a preset metal crack feature extraction model to obtain target metal crack features includes: The input CT image is obtained, and a 1×1 convolution operation is performed on the CT image to obtain the first metallic mineral convolution feature; Substituting the first metal ore convolution feature into the improved multi-channel convolution feature module yields the first improved metal ore convolution feature. The first improved metal ore convolution feature is successively substituted into the max pooling module and the improved multi-channel convolution feature module to obtain the second improved metal ore convolution feature; The second improved metal ore convolution feature is successively substituted into the max pooling module and the improved multi-channel convolution feature module to obtain the third improved metal ore convolution feature; Substituting the second improved metal ore convolution feature into the ECA module yields the enhanced metal ore convolution feature; After upsampling the enhanced metallic ore convolutional features, they are fused with the first improved metallic ore convolutional features and substituted into the ECA module to obtain the first fused enhanced metallic ore convolutional features; After downsampling the enhanced metallic ore convolutional features, they are fused with the third improved metallic ore convolutional features and substituted into the ECA module to obtain the second fused enhanced metallic ore convolutional features; The target metal fracture feature is obtained by fusing the first and second fusion-enhanced metal ore convolution features.

[0040] In one implementation, see [link to implementation details]. Figure 2 , Figure 2 This is a data flow diagram of a preset metal crack feature extraction model provided in an embodiment of the present invention.

[0041] In one implementation, basic convolutional features are sequentially substituted into an improved multi-channel convolutional module, and max pooling is used to achieve two downsampling layers, generating three improved convolutional features at different scales. This operation fits the multi-scale spatial distribution characteristics of metal cracks. The first improved feature retains fine-grained features such as shallow high-resolution crack edges and micro-cracks (a few pixels wide). The second and third improved features deepen the semantics through downsampling, capturing deep features such as crack connectivity, local morphology and crack network, and overall distribution, respectively, solving the problem that single-scale extraction cannot take into account both fine-grained details and global semantics. The improved multi-channel convolutional module itself can capture multi-directional features of cracks, such as horizontal, vertical, and oblique directions, further making up for the defect of traditional convolution that easily misses the multi-dimensional morphology of cracks.

[0042] In one implementation, the second improved metal ore convolutional feature is selected and substituted into the ECA module to generate enhanced features. This scale feature is the core intersection of fine-grained details and deep semantics, which not only retains sufficient crack spatial information but also has certain semantic correlation features, making it the core carrier of metal crack features. Through the channel attention mechanism of the ECA module, the response of low-contrast crack-related channels can be specifically enhanced, the interference of background noise channels such as metal matrix and scanning artifacts can be suppressed, and the feature submersion problem caused by the gray-scale overlap between metal cracks and matrix can be solved, making the core metal crack features more prominent.

[0043] In one implementation, the target metal fracture feature is obtained by fusing the first fusion-enhanced metal ore convolution feature and the second fusion-enhanced metal ore convolution feature by averaging the first fusion-enhanced metal ore convolution feature and the second fusion-enhanced metal ore convolution feature.

[0044] In one implementation, the enhanced core features are upsampled and downsampled respectively, and then fused with the shallow first improved feature and the deep third improved feature before being enhanced again by the ECA module. The core of this operation is to achieve the complementarity and purification of cross-scale metal crack information: upsampling and fusing shallow features can combine deep semantic features with shallow fine-grained crack edges and micro-crack details, avoiding the loss of details caused by downsampling of deep features; downsampling and fusing deep features can integrate shallow features into the global semantics of the crack network, avoiding the bias of shallow features in judging the overall crack morphology; and after fusion, the ECA module can perform secondary weight calibration on the cross-scale fused features, remove redundant information introduced during the fusion process, strengthen the effective metal crack features after cross-scale fusion, and make the fused features more accurate.

[0045] In one embodiment, the improved multi-channel convolutional feature module works as follows: The input features are obtained, and after multi-directional single-channel convolution, 1×1 pointwise convolution is performed to obtain the first fusion channel convolution feature; the multi-directional features include the horizontal direction, vertical direction, left tilt direction, and right tilt direction; Global average pooling is applied to the convolutional features of the first fusion channel to obtain global average pooling features; The target convolution kernel is determined based on the number of channels of the global average pooling feature. The second fusion channel convolution feature is obtained by performing a 1D convolution operation on the global average pooling feature according to the target convolution kernel. Substitute the convolutional features of the second fusion channel into the Sigmoid activation function to obtain the target weights; The output feature is obtained by fusing the convolutional features of the second fusion channel and the convolutional features of the first fusion channel according to the target weight.

[0046] In one implementation, single-channel convolutions are performed on the input features in horizontal, vertical, left-tilted, and right-tilted directions respectively. This allows for the targeted extraction of spatial contour features of metal ore cracks with different orientations, avoiding the loss of oblique and narrow crack features caused by the lack of orientation targeting in traditional general convolutions. This ensures that the spatial features of different types of metal cracks, such as horizontal bedding cracks, vertical tension cracks, and oblique shear cracks, can be accurately captured. The single-channel convolution design also reduces information interference between channels, allowing crack features in each direction to be completely preserved in an independent channel, which is suitable for the characteristics of weak metal crack features that are easily obscured by the background.

[0047] In one implementation, CT images of metallic ore rock masses are prone to artifacts and grayscale fluctuations during scanning. These noises can distract the convolutional model from focusing on crack features. Furthermore, metallic ore cracks occupy only a small number of pixels in the image, and a large amount of redundant background information in the high-dimensional feature map can obscure the channel feature trends of the cracks. By performing global average pooling on the convolutional features of the first fusion channel, the high-dimensional spatial feature map can be compressed into low-dimensional channel statistics. This discards the spatial details of local noise and redundant background, while retaining the core trends of metallic ore crack features in each channel. This allows the model to focus on the channel features related to cracks and lays a lightweight foundation for the subsequent channel attention modeling. It avoids computational redundancy caused by high-dimensional features and adapts to the pain points of scenarios where the proportion of metallic crack pixels is low and effective features are easily obscured by noise.

[0048] In one implementation, the response of metal ore fracture features differs significantly across different channels. High-frequency fracture edge features are concentrated in some channels, while low-frequency fracture connectivity features are distributed in other channels. Furthermore, changes in the number of channels directly affect the distribution range of metal fracture features. If a 1D convolution with a fixed kernel size is used, it either fails to capture the correlation of metal fracture features between channels or introduces noise from irrelevant background channels. The 1D convolution kernel size is dynamically adapted based on the number of channels using global average pooling features, allowing the coverage of the convolution kernel to match the distribution pattern of metal fracture features in the channels. When the number of channels is small, a small kernel is used to focus on the fine-grained feature correlation of metal fractures in adjacent channels; when the number of channels is large, a large kernel is used to capture the semantic feature correlation of metal fractures in a wider range of channels.

[0049] In one implementation method, through formula The size of the one-dimensional convolution kernel is dynamically determined, with γ=2, b=1, and C being the number of channels. This means rounding the calculation result down to the nearest odd number, ensuring that the convolution kernel size is odd.

[0050] In one embodiment, constructing the gas concentration distribution field of the target metal ore area based on the target metal fracture characteristics and the predicted gas concentration points and confidence level intervals of the corresponding monitoring points of each gas acquisition sensor includes: The target metal fracture features are mapped to a preset three-dimensional spatial grid corresponding to the target metal ore area, and the gas diffusion coefficient and permeability coefficient of each grid cell are calibrated according to the preset calibration rules. The predicted values ​​of gas concentration points and confidence level intervals are bound to the corresponding sensor units of a preset three-dimensional spatial grid, and the weight coefficients of the corresponding units are obtained by weighting the predicted values ​​of confidence level intervals. A physical field model of gas diffusion in a two-pore medium of a metal ore rock mass is constructed based on the calibrated gas diffusion coefficient and permeability coefficient. The predicted values ​​of the bound gas concentration points are used as sample points. By combining the physical field model and the confidence weighting coefficient with Kriging interpolation, the gas concentration values ​​of all grid cells are completed to obtain the initial gas concentration distribution field. Verify the matching degree between the initial gas concentration distribution field and the target metal crack characteristics. For grid cells with matching degrees exceeding the preset threshold, re-call the physical field model to correct the gas diffusion coefficient and perform interpolation calculation again until the matching degree of all grid cells falls within the threshold range to obtain the gas concentration distribution field.

[0051] In one implementation, the target metal ore area is divided into a regular, preset three-dimensional spatial grid according to a preset size, thus discretizing the entire monitoring area into several uniform three-dimensional grid units. The target metal crack features extracted from CT images are mapped one-to-one to the aforementioned preset three-dimensional spatial grid according to spatial coordinates, ensuring that each grid unit possesses corresponding metal crack structure parameters. The preset calibration rule is that when the crack connectivity of a grid unit is greater than or equal to 80%, it is determined to be a highly connected region, and the gas diffusion coefficient is taken as 0.015–0.025 m. 2 / s; When the fracture connectivity is between 30% and 80%, it is considered a moderately connected region, and the gas diffusion coefficient is between 0.005 and 0.015m. 2 / s; when the fracture connectivity is less than 30%, it is judged as a low connectivity region, and the gas diffusion coefficient is taken as 0.001~0.005m. 2 / s; the permeability coefficient is determined by multiplying the pore correction coefficient by the gas diffusion coefficient, where the pore correction coefficient ranges from 0.6 to 0.9, and the value increases as the pore density increases (by normalizing the porosity of all units to 0.6 to 0.9).

[0052] In one implementation, the predicted confidence level interval for the monitoring point corresponding to the i-th sensor is denoted as... Define the confidence interval width for this monitoring point. for Weighting coefficients based on confidence interval width Divide , The first preset width threshold, The second preset width threshold is satisfied. .

[0053] In one implementation, each three-dimensional mesh unit is abstracted as a dual-pore medium unit consisting of a fracture system and a pore system. The fracture system serves as a channel for rapid gas seepage and diffusion, while the pore system acts as a storage space for slow gas diffusion, closely mirroring the actual structure of the metal ore rock mass. The diffusion coefficient D and permeability coefficient K of each mesh unit are assigned one-to-one to the dual-pore medium unit in the model. D controls the gas diffusion rate in the pores and fractures, and K controls the gas seepage capacity in the fractures. The parameters of mesh units differ depending on the fracture characteristics of the metal, realistically reflecting geological differences. Based on the law of conservation of gas mass, a gas transport control equation for the dual-pore medium is established, clarifying the gas exchange relationship and gas concentration variation law between the fracture and pore systems. The seepage control equation for the fracture system is: The diffusion control equation for the porous system is: where t is time, Here, Hamiltonian operator represents spatial gradient and spatial rate of change. Let be the partial derivative with respect to time, representing the rate of change of a physical quantity with respect to time. The gas density within the fracture system. Fracture porosity represents the proportion of fracture volume to the total volume in a metallic ore body. Let P be the gas concentration within the fracture system, K be the permeability coefficient of the metallic ore rock mass (obtained from the previous characterization of metallic fractures), and P be the gas pressure within the fracture system. This refers to the amount of gas exchanged from the pore system to the fracture system within a unit volume of metallic ore rock per unit time. The gas density within the porous system. Matrix porosity refers to the percentage of pore volume within the matrix of metallic rocks. The gas concentration within the pore system is given by denoted as ...

[0054] In one implementation, the constructed model is imported into conventional numerical solving software (such as MATLAB, COMSOL, FLUENT) to solve the governing equations globally, obtaining the gas diffusion constraint parameters for each grid cell. These parameters serve as physical constraints for subsequent interpolation calculations. The specific solution results include: the gas diffusion direction of each grid cell, the gas diffusion rate of each grid cell, and the gas influence weights between grid cells. Grid cells with predicted gas concentration points are used as known sample points for interpolation. The coordinates, concentration values, and confidence weighting coefficients of each sample point are specified. Based on the obtained gas diffusion rate and diffusion direction, a variogram for Kriging interpolation is set to ensure that the variogram conforms to the physical laws of gas diffusion. The gas influence weights between grid cells are used as interpolation constraints to ensure that the spatial variation trend of gas concentration is consistent with the diffusion trend predicted by the physical field model during the interpolation process.

[0055] In one implementation, to address the issue of low-confidence sensor data interfering with interpolation accuracy, the confidence weighting coefficient is incorporated into the weight calculation of Kriging interpolation to achieve adaptive correction. The implementation steps include: using the Kriging interpolation algorithm, calculating the basic interpolation weight of each known sample point to the surrounding unknown grid cells based on the spatial distance and variogram between sample points; multiplying the basic interpolation weight of each sample point by the confidence weighting coefficient of that sample point to obtain the corrected final interpolation weight; and normalizing the corrected interpolation weights of all sample points (ensuring that the sum of the weights of all sample points to a certain unknown grid cell is 1) to avoid weight imbalance leading to interpolation result deviation.

[0056] In one implementation, for each unknown grid cell, known sample points within a preset range around it are selected; the preset range is determined by technicians. The corrected interpolation weights are called, and combined with the predicted gas concentration values ​​of the sample points, the predicted gas concentration value of the unknown grid cell is calculated using the Kriging interpolation formula. For each calculated concentration value, it is checked whether it conforms to the diffusion law of the physical field model. If it does not conform, the interpolation weights are readjusted and the calculation is repeated. The above operation is repeated to traverse all unknown grid cells one by one to complete the concentration value completion of all grid cells. The gas concentration values ​​of all grid cells are then checked.

[0057] In one implementation, the matching degree between the initial gas concentration distribution field and the crack characteristics of the target metal is specifically verified by formula... accomplish, Let be the initial gas concentration value of the i-th grid cell. Let be the crack density of the i-th grid cell. α is the permeability coefficient of the i-th grid cell, and α is the metal crack concentration correlation coefficient with a value of 1.2~1.5, which shall be determined by the technical personnel. The preset threshold shall be determined by the technical personnel.

[0058] Based on the same inventive concept, this invention also provides a security monitoring system based on multi-source data fusion. See also Figure 3 , Figure 3 A framework diagram of a security monitoring system based on multi-source data fusion provided in this embodiment of the invention includes: A gas data acquisition module is used to synchronously acquire gas data from each sensor to obtain a gas data array; a gas acquisition sensor array is set up in the target metal mining area; The gas prediction value generation and acquisition module is used to perform multi-channel spatiotemporal convolution on the gas data array to determine the gas concentration point prediction value and confidence level interval prediction value of each gas acquisition sensor in the gas acquisition sensor array at the corresponding monitoring point. The target metal crack feature acquisition module is used to acquire CT images of the target metal mining area and input the CT images into a preset metal crack feature extraction model to obtain the target metal crack features. The gas concentration distribution field construction module is used to construct the gas concentration distribution field of the target metal mine area based on the target metal crack characteristics and the predicted values ​​of gas concentration points and confidence level intervals of the corresponding monitoring points of each gas acquisition sensor. The safety monitoring module is used to input the gas concentration distribution field into a preset safety monitoring model to obtain safety monitoring results.

[0059] The safety monitoring system based on multi-source data fusion provided by this invention extracts deep spatiotemporal features from a gas data array through multi-channel spatiotemporal convolution, accurately outputting predicted gas concentration values ​​and confidence intervals for each monitoring point. Simultaneously, it leverages a CT metal fracture feature extraction model to obtain key features of the geological structure of metal mines, overcoming the shortcomings of simply stitching together traditional multi-source data. This achieves deep coupling and feature fusion of gas monitoring information and metal fracture structure information, constructing a high-precision gas concentration distribution field for the metal mine area. The system then outputs stable and reliable monitoring conclusions through a safety monitoring model, significantly improving the accuracy, stability, and intelligence of metal mine safety monitoring, effectively enhancing early warning reliability and overall monitoring efficiency.

[0060] The foregoing has described one embodiment of the present invention in detail, but this content is merely a preferred embodiment and should not be considered as limiting the scope of the present invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the scope of the claims of this invention.

Claims

1. A security monitoring method based on multi-source data fusion, characterized in that, A gas sampling sensor array is installed in the target metal mining area, and the method includes: Gas data is collected synchronously from each sensor to obtain a gas data array; Multi-channel spatiotemporal convolution is performed on the gas data array to determine the predicted gas concentration point value and confidence level interval prediction value for each gas acquisition sensor in the gas acquisition sensor array at the corresponding monitoring point. CT images of the target metal mining area are acquired, and the CT images are substituted into a preset metal fracture feature extraction model to obtain the target metal fracture features. Based on the target metal fracture characteristics and the predicted values ​​of gas concentration points and confidence level intervals at the corresponding monitoring points of each gas acquisition sensor, a gas concentration distribution field of the target metal ore area is constructed. The gas concentration distribution field is input into a preset safety monitoring model to obtain safety monitoring results.

2. The security monitoring method based on multi-source data fusion according to claim 1, characterized in that, Performing multi-channel spatiotemporal convolution on the gas data array to determine the predicted gas concentration point value and confidence level interval value for each gas acquisition sensor in the gas acquisition sensor array includes: Acquire gas data sets collected by the sensor array, and preprocess the gas data sets to obtain effective gas data sets; Spatiotemporal correlation analysis was performed on the effective gas data set to obtain a spatiotemporal data set; IMF decomposition is performed on each set of data in the spatiotemporal data group to obtain the intrinsic mode function set corresponding to each set of data; Clustering the intrinsic modulus functions in all intrinsic modulus function sets by a preset frequency magnitude yields low-frequency trend component sets, mid-frequency trend component sets, and high-frequency trend component sets. Spatiotemporal features are obtained by inputting independent prediction channels for each trend component set; the independent prediction channels include cascaded spatiotemporal convolutional modules; Based on the attention mechanism of modal channels, the attention weight of each channel is calculated and the spatiotemporal features of the three channels are weighted and fused to obtain the predicted gas concentration point value and confidence level interval value of each monitoring point in the sensor array.

3. The security monitoring method based on multi-source data fusion according to claim 2, characterized in that, The spatiotemporal convolution module consists of a graph convolutional network and a temporal convolutional network connected in series.

4. The security monitoring method based on multi-source data fusion according to claim 1, characterized in that, Substituting the CT image into a preset metal crack feature extraction model yields the target metal crack features, including: The input CT image is acquired, and a 1×1 convolution operation is performed on the CT image to obtain the first metallic mineral convolution feature; Substituting the first metal ore convolution feature into the improved multi-channel convolution feature module yields the first improved metal ore convolution feature. The first improved metal ore convolution feature is sequentially substituted into the max pooling module and the improved multi-channel convolution feature module to obtain the second improved metal ore convolution feature. The second improved metal ore convolution feature is successively substituted into the max pooling module and the improved multi-channel convolution feature module to obtain the third improved metal ore convolution feature; Substituting the second improved metal ore convolution feature into the ECA module yields the enhanced metal ore convolution feature; After upsampling the enhanced metal ore convolutional features, they are fused with the first improved metal ore convolutional features and substituted into the ECA module to obtain the first fused enhanced metal ore convolutional features; After downsampling the enhanced metallic ore convolutional features, they are fused with the third improved metallic ore convolutional features and substituted into the ECA module to obtain the second fused enhanced metallic ore convolutional features; The target metal fracture feature is obtained by fusing the first fused enhanced metal ore convolution feature and the second fused enhanced metal ore convolution feature.

5. A security monitoring method based on multi-source data fusion according to claim 4, characterized in that, The working principle of the improved multi-channel convolutional feature module is as follows: The input features are obtained, and after multi-directional single-channel convolution, 1×1 pointwise convolution is performed to obtain the first fused channel convolution feature; the multi-directional features include the horizontal direction, the vertical direction, the left tilt direction, and the right tilt direction; Global average pooling is applied to the convolutional features of the first fused channel to obtain global average pooling features; The target convolution kernel is determined based on the number of channels of the global average pooling feature, and a second fused channel convolution feature is obtained by performing a 1D convolution operation on the global average pooling feature according to the target convolution kernel. Substitute the convolutional features of the second fusion channel into the Sigmoid activation function to obtain the target weights; The output feature is obtained by fusing the second fusion channel convolutional feature and the first fusion channel convolutional feature according to the target weight.

6. A security monitoring method based on multi-source data fusion according to claim 5, characterized in that, Determining the target convolutional kernel based on the number of channels of the global average pooling feature includes: Through formula Determine the target convolutional kernel size, where γ=2, b=1, and C is the number of channels. , indicates that the calculation result is rounded down.

7. The security monitoring method based on multi-source data fusion according to claim 1, characterized in that, Based on the target metal fracture characteristics and the predicted gas concentration points and confidence level intervals of each gas acquisition sensor at the corresponding monitoring points, the gas concentration distribution field of the target metal ore area is constructed as follows: The target metal fracture features are mapped to a preset three-dimensional spatial grid corresponding to the target metal ore region, and the gas diffusion coefficient and permeability coefficient of each grid cell are calibrated according to preset calibration rules; The predicted values ​​of gas concentration points and confidence level intervals are bound to the corresponding sensor units of the preset three-dimensional spatial grid, and the weight coefficients of the corresponding units are obtained by weighting according to the predicted values ​​of confidence level intervals. A physical field model for gas diffusion in a two-pore medium of a metal ore rock mass is constructed based on the calibrated gas diffusion coefficient and permeability coefficient. The predicted values ​​of the bound gas concentration points are used as sample points. By combining the physical field model and the confidence weighting coefficient through Kriging interpolation, the gas concentration values ​​of all grid cells are completed to obtain the initial gas concentration distribution field. Verify the matching degree between the initial gas concentration distribution field and the target metal crack characteristics. For grid cells with matching degrees exceeding the preset threshold, re-call the physical field model to correct the gas diffusion coefficient and perform interpolation calculation again until the matching degree of all grid cells falls within the threshold range to obtain the gas concentration distribution field.

8. A security monitoring method based on multi-source data fusion according to claim 7, characterized in that, The weighting coefficients for the corresponding units are obtained by weighting the predicted values ​​within the confidence level intervals, including: Let the predicted confidence level interval value of the monitoring point corresponding to the i-th sensor be denoted as . Define the confidence interval width for this monitoring point. for Weighting coefficients based on confidence interval width Divide , The first preset width threshold, The second preset width threshold is satisfied. .

9. A security monitoring method based on multi-source data fusion according to claim 7, characterized in that, The target metal fracture characteristics include fracture density and permeability coefficient; Verifying the matching degree between the initial gas concentration distribution field and the target metal crack characteristics includes: Through formula Determine the matching degree corresponding to the i-th grid cell; in, Let be the initial gas concentration value of the i-th grid cell. Let be the crack density of the i-th grid cell. Let be the permeability coefficient of the i-th grid cell, and α be the constant correlation coefficient of metal crack concentration.

10. A security monitoring system based on multi-source data fusion, characterized in that, The system includes: A gas data acquisition module is used to synchronously acquire gas data from each sensor to obtain a gas data array; a gas acquisition sensor array is set up in the target metal mining area; The gas prediction value generation and acquisition module is used to perform multi-channel spatiotemporal convolution on the gas data array to determine the gas concentration point prediction value and confidence level interval prediction value of each gas acquisition sensor in the gas acquisition sensor array corresponding to the monitoring point. The target metal fracture feature acquisition module is used to acquire CT images of the target metal ore area and substitute the CT images into a preset metal fracture feature extraction model to obtain the target metal fracture features. The gas concentration distribution field construction module is used to construct the gas concentration distribution field of the target metal mine area based on the target metal fracture characteristics and the predicted values ​​of gas concentration points and confidence level intervals of the corresponding monitoring points of each gas acquisition sensor. The safety monitoring module is used to input the gas concentration distribution field into a preset safety monitoring model to obtain safety monitoring results.

Images